Industry Automation:

The Technologies, Platforms

and Use Cases
RIVER PUBLISHERS SERIES IN AUTOMATION,
CONTROL AND ROBOTICS

Series Editors:
ISHWAR K. SETHI
Oakland University, USA

TAREK SOBH
University of Bridgeport, USA

FENG QIAO
Shenyang JianZhu University, China

The “River Publishers Series in Automation, Control and Robotics” is a

series of comprehensive academic and professional books which focus on
the theory and applications of automation, control and robotics. The series
focuses on topics ranging from the theory and use of control systems,
automation engineering, robotics and intelligent machines.
Books published in the series include research monographs, edited
volumes, handbooks and textbooks. The books provide professionals,
researchers, educators, and advanced students in the field with an invaluable
insight into the latest research and developments.
Topics covered in the series include, but are by no means restricted to the
following:
• Robots and Intelligent Machines
• Robotics
• Control Systems
• Control Theory
• Automation Engineering

For a list of other books in this series, visit [Link]

 Industry Automation:
The Technologies, Platforms
and Use Cases

Editors

Pethuru Raj
Edge AI Division, Reliance Jio Platforms Ltd, Bangalore, India

Abhishek Kumar
Chandigarh University, Punjab, India

Ananth Kumar
IFET College of Engineering, Tamil Nadu, India

Neha Singhal
Christ University, Central Campus, Bangalore, India

River Publishers
Published, sold and distributed by:
River Publishers
Alsbjergvej 10
9260 Gistrup
Denmark

[Link]

ISBN: 978-87-7004-039-6 (Hardback)

978-87-7004-038-9 (E book)

©2024 River Publishers

All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted in any form or by any means, mechanical,
photocopying, recording or otherwise, without prior written permission of
the publishers.
 Contents

Preface xvii

List of Figures xxi

List of Tables xxvii

List of Contributors xxix

List of Abbreviations xxxiii

1 An Analytical Framework for the Industrial Internet of Things

(IIoT): Importance, Recent Challenges, and Enabling
Technologies 1
Gouse Baig Mohammed, Shitharth Selvarajan, A. Ravi Kumar,
K. Sangeetha, Alaa O. Khadidos, and S. Vatchala
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Industrial automation with IoT . . . . . . . . . . . 4
1.1.2 Objective . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Industry 4.0 . . . . . . . . . . . . . . . . . . . . . 10
1.3 Enabling Technologies for IIoT . . . . . . . . . . . . . . . 13
1.3.1 Blockchain technology . . . . . . . . . . . . . . . 13
1.3.2 Cloud computing . . . . . . . . . . . . . . . . . . 13
1.3.3 Big data analytics . . . . . . . . . . . . . . . . . . 14
1.3.4 Artificial intelligence and cyber-physical systems . 14
1.3.5 Augmented and virtual reality . . . . . . . . . . . 15
1.4 Framework and Case Studies . . . . . . . . . . . . . . . . . 15
1.4.1 SnappyData . . . . . . . . . . . . . . . . . . . . . 16
1.4.2 Fault detection classification . . . . . . . . . . . . 16
1.5 Challenges in IIoT . . . . . . . . . . . . . . . . . . . . . . 16
1.5.1 Schemes for efficient data storage . . . . . . . . . 16

v
vi Contents

1.5.2 IoT systems from different vendors working

together . . . . . . . . . . . . . . . . . . . . . . . 17
1.5.3 Adaptable and resilient technologies for analyzing
large datasets . . . . . . . . . . . . . . . . . . . . 17
1.5.4 Trust in IIoT systems . . . . . . . . . . . . . . . . 18
1.5.5 Integration of wireless technologies and protocols
in the Internet of Things (IIoT) . . . . . . . . . . . 18
1.5.6 The edge of decentralization . . . . . . . . . . . . 18
1.5.7 New operating systems for the Internet of Things . 19
1.5.8 Public safety in IIoT . . . . . . . . . . . . . . . . . 19
1.6 Application for IIoT Framework . . . . . . . . . . . . . . . 20
1.7 Conclusion and Future Scope . . . . . . . . . . . . . . . . 20

2 Industry Automation: The Contributions of Artificial

Intelligence (AI) 25
M. Nalini
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Automation Systems Potential . . . . . . . . . . . . . . . . 29
2.3 Application Landscape and Production-related Scenarios . . 30
2.3.1 Autonomy-level classification of industrial AI
applications . . . . . . . . . . . . . . . . . . . . . 31
2.4 Impact of Artificial Intelligence in Industry4.0 (I4.0) . . . . 33
2.4.1 Order-controlled production (OCP) . . . . . . . . . 33
2.4.2 Smart production (SP2) . . . . . . . . . . . . . . . 34
2.4.3 Innovative product development (IPD) . . . . . . . 35
2.4.4 Seamless and dynamic engineering of plants (SDP) 36
2.4.5 Circular economy (CRE) . . . . . . . . . . . . . . 37
2.4.6 5G for digital factories – mobile controlled produc-
tion (MCP) . . . . . . . . . . . . . . . . . . . . . 39
2.5 Industry Use Cases for AI-enabled Collaboration . . . . . . 40
2.5.1 Artificial intelligence in healthcare industry . . . . 40
[Link] The use of predictive analytics to confirm
the need for surgery . . . . . . . . . . . . 40
[Link] Intelligent surgical robots . . . . . . . . . 42
2.5.2 Artificial intelligence in manufacturing and factories 42
[Link] Analytical services for advanced data . . . 42
[Link] Predictive maintenance . . . . . . . . . . 43
[Link] Automation of robotic processes . . . . . 44
2.5.3 Artificial intelligence in automobile . . . . . . . . 45
 Contents vii

[Link] The use of artificial intelligence to improve

design . . . . . . . . . . . . . . . . . . . 45
[Link] AI application in manufacturing . . . . . . 45
[Link] Examples of AI in manufacturing –
inspiring changes . . . . . . . . . . . . . 46
2.5.4 Application of artificial intelligence in quality
control . . . . . . . . . . . . . . . . . . . . . . . . 47
2.5.5 Manufacturing industry trends with emerging AI . 49
2.5.6 The Internet of Things is emerging as Industry 4.0’s
future . . . . . . . . . . . . . . . . . . . . . . . . 51
2.5.7 Future scope of research . . . . . . . . . . . . . . 52
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3 Industry Automation: The Contributions of Artificial

Intelligence 57
B. Akoramurthy, K. Dhivya, and Kassian T. T. Amesho
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 Industry 5.0 and AI . . . . . . . . . . . . . . . . . . . . . 60
3.4 Problems with Human–Robot Collaboration . . . . . . . . . 64
3.4.1 Issues with law and regulation . . . . . . . . . . . 65
3.4.2 Subjective opinion for using robots at work . . . . 66
3.4.3 Psychosocial problems caused by human–robot col-
laboration . . . . . . . . . . . . . . . . . . . . . . 66
3.4.4 Changes that result from human–robot collaboration 66
3.4.5 The shifting functions of human resources divisions 67
3.5 Wafer Fabrication Automation . . . . . . . . . . . . . . . . 68
3.6 AI as a Vital Technology in Industry 5.0 . . . . . . . . . . . 69
3.6.1 Impact of AI on different industries . . . . . . . . . 69
3.7 Artificial General Intelligence (AGI) . . . . . . . . . . . . . 71
3.8 The Scenario of AI in the Focus of Manufacturing . . . . . 71
3.9 Automation based on AI (ABAI) . . . . . . . . . . . . . . . 72
3.9.1 Computerized root cause analysis using AI . . . . 73
3.9.2 Intelligent computing in product matching . . . . . 75
3.10 Robotic Process Automation . . . . . . . . . . . . . . . . . 76
3.11 The Digital Solutions Entangled in Industry 5.0 . . . . . . . 77
3.12 AMS for Industry 5.0: Advanced Manufacturing System . . 78
3.13 Methods, Data, and Results . . . . . . . . . . . . . . . . . 78
3.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 80
viii Contents

4 Artificial Intelligence (AI) Driven Industrial Automation 85

S. S. Blessy Trencia Lincy
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.2 Evolution of Artificial Intelligence . . . . . . . . . . . . . . 87
4.3 Industry 4.0 Technologies . . . . . . . . . . . . . . . . . . 87
4.4 Development in AI . . . . . . . . . . . . . . . . . . . . . . 88
4.5 AI Future Perception . . . . . . . . . . . . . . . . . . . . . 91
4.6 Digital Transformation . . . . . . . . . . . . . . . . . . . . 91
4.7 Components of AI in Automation . . . . . . . . . . . . . . 92
4.8 Artificial Intelligence Applications in Automation . . . . . . 93
4.9 Automation and AI . . . . . . . . . . . . . . . . . . . . . . 97
4.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 98

5 Quantum Machine and Deep Learning Models for Industry

Automation 101
Pawan Whig, Arun Velu, Rahul Reddy Nadikattu,
and Yusuf Jibrin Alkali
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.2 Difference Between Classical and Quantum Data . . . . . . 104
5.3 Quantum Computing . . . . . . . . . . . . . . . . . . . . . 104
5.3.1 Qubit . . . . . . . . . . . . . . . . . . . . . . . . 105
5.3.2 Superposition . . . . . . . . . . . . . . . . . . . . 105
5.3.3 Entanglement . . . . . . . . . . . . . . . . . . . . 105
5.4 Quantum Machine Learning (QML) . . . . . . . . . . . . . 106
5.5 Classical Machine Learning vs. Quantum Computing . . . . 107
5.5.1 Linear algebra problems have been solved via quan-
tum machine learning . . . . . . . . . . . . . . . . 107
5.6 Quantum Thinking in Depth . . . . . . . . . . . . . . . . . 108
5.6.1 Principal component analysis in quantum . . . . . 109
5.6.2 Support vector quantum machines . . . . . . . . . 110
5.6.3 Optimization . . . . . . . . . . . . . . . . . . . . 110
5.7 Quantum Learning in Depth . . . . . . . . . . . . . . . . . 111
5.7.1 Why is quantum machine learning so exciting? . . 111
5.8 The Essence of Quantum Computing . . . . . . . . . . . . 113
5.8.1 Taking the initiative to manage uncertainty . . . . . 113
5.8.2 Welcoming a new AI era . . . . . . . . . . . . . . 113
5.8.3 Cybersecurity advancement . . . . . . . . . . . . . 113
5.8.4 Accuracy of weather predictions . . . . . . . . . . 114
5.8.5 A signal to develop better life-saving drugs . . . . 114
 Contents ix

5.9 A Portal to Exciting Future Technology . . . . . . . . . . . 114

5.9.1 How AI will change thanks to quantum
computing . . . . . . . . . . . . . . . . . . . . . . 114
5.9.2 Processes for making better business decisions . . . 115
5.9.3 Quantum security and artificial intelligence . . . . 115
5.9.4 AI and quantum computing complement DevOps . 116
5.9.5 Where are our IT systems vulnerable? . . . . . . . 116
5.9.6 Limitation of quantum machine learning . . . . . . 116
5.9.7 Hardware constraints . . . . . . . . . . . . . . . . 117
5.9.8 Program restrictions . . . . . . . . . . . . . . . . . 117
5.10 More on Quantum Computing and Machine Learning
Connections . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.10.1 Wavefunction . . . . . . . . . . . . . . . . . . . . 118
5.10.2 The significance of accuracy . . . . . . . . . . . . 119
5.10.3 Data power and quantum machine learning . . . . . 121
5.11 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.11.1 Q-SVM (quantum support vector machine
algorithm) . . . . . . . . . . . . . . . . . . . . . . 123
5.11.2 Why did they need Q-SVM? . . . . . . . . . . . . 124
5.11.3 Import the library . . . . . . . . . . . . . . . . . . 124
5.11.4 Install the dataset . . . . . . . . . . . . . . . . . . 125
5.12 Quantum Computing and Machine Learning for Industry
Automation . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.12.1 Discover . . . . . . . . . . . . . . . . . . . . . . . 128
5.12.2 Design . . . . . . . . . . . . . . . . . . . . . . . . 128
5.12.3 Control . . . . . . . . . . . . . . . . . . . . . . . 128
5.12.4 Supply chains . . . . . . . . . . . . . . . . . . . . 129
5.12.5 How does manufacturing begin? . . . . . . . . . . 129
5.13 Conclusion and Future Scope . . . . . . . . . . . . . . . . 130

6 The Contribution of Computer Vision in the Manufacturing

Industries and the Scope for Further Excellence 135
P. Dharanyadevi, C. Krishnakoumar, K. Revathy, J. Karkavelraja,
K. Venkatalakshmi, and G. Zayaraz
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.2 Components of a Machine Vision Systems . . . . . . . . . 138
6.3 Image Formation . . . . . . . . . . . . . . . . . . . . . . . 141
6.4 Computer vision algorithms . . . . . . . . . . . . . . . . . 141
6.5 Use Case of the Computer Vision in Industries . . . . . . . 141
x Contents

6.5.1 Product assembly . . . . . . . . . . . . . . . . . . 141

6.5.2 Defect detection . . . . . . . . . . . . . . . . . . . 142
6.5.3 3D Vision system . . . . . . . . . . . . . . . . . . 144
6.5.4 Vision-guided robots . . . . . . . . . . . . . . . . 144
6.5.5 Predictive maintenance . . . . . . . . . . . . . . . 145
6.6 Safety and Security Standards . . . . . . . . . . . . . . . . 145
6.7 Packaging Standards . . . . . . . . . . . . . . . . . . . . . 147
6.8 Barcode Analysis . . . . . . . . . . . . . . . . . . . . . . . 148
6.9 Inventory Management . . . . . . . . . . . . . . . . . . . . 148
6.10 Optimizing Supply Chains . . . . . . . . . . . . . . . . . . 149
6.11 Quality Inspection with Computer Vision . . . . . . . . . . 149
6.12 Computer Vision during the Covid-19 Pandemic . . . . . . 150
6.13 Computer Vision in the Automotive Industry . . . . . . . . 151
6.13.1 Press shop . . . . . . . . . . . . . . . . . . . . . . 152
6.13.2 Body shop . . . . . . . . . . . . . . . . . . . . . . 153
6.13.3 Paint shop . . . . . . . . . . . . . . . . . . . . . . 154
6.13.4 Final assembly shop . . . . . . . . . . . . . . . . . 154
6.14 Computer Vision Performance Metrics . . . . . . . . . . . 155
6.14.1 Intersection over union (IoU) . . . . . . . . . . . . 155
6.14.2 Precision . . . . . . . . . . . . . . . . . . . . . . . 155
6.14.3 Recall . . . . . . . . . . . . . . . . . . . . . . . . 156
6.14.4 F1 score . . . . . . . . . . . . . . . . . . . . . . . 156
6.15 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 158

7 Waste Management 4.0: An Industry Automation Approach to

the Future Waste Management System 163
M. Julie Therese, P. Dharanyadevi, A. Devi, and Christo Ananth
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.2 Exploring CPS . . . . . . . . . . . . . . . . . . . . . . . . 166
7.2.1 CPS . . . . . . . . . . . . . . . . . . . . . . . . . 166
7.2.2 Drawbacks of CPS . . . . . . . . . . . . . . . . . 167
7.3 Industry 4.0 Environment . . . . . . . . . . . . . . . . . . 168
7.4 Challenges in the Waste Management Industry . . . . . . . 169
7.5 Applications of CPS in the Waste Management Industry . . 171
7.6 Influence of Industry 4.0 on the Waste Management
Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.7 Barriers to Implementing Industry 4.0 in the Waste Manage-
ment Industry . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.8 Case Study: Machine Learning for Waste Management . . . 175
 Contents xi

7.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 177

8 Industrial Internet of Things (IIoT) for E-waste Recycling

System 181
A. P. Jyothi and P. Padma Priya Dharishini
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 181
8.2 Background Study . . . . . . . . . . . . . . . . . . . . . . 182
8.3 IIoT Working . . . . . . . . . . . . . . . . . . . . . . . . . 182
8.4 IIoT Security . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.4.1 Risks and challenges of IIoT . . . . . . . . . . . . 183
8.4.2 Difference between IoT and IIoT . . . . . . . . . . 184
8.4.3 IIoT applications and examples . . . . . . . . . . . 184
8.5 Industries using IIoT . . . . . . . . . . . . . . . . . . . . . 185
8.6 Advantages and Disadvantages of IIoT . . . . . . . . . . . 186
8.6.1 Hindrances of IIoT . . . . . . . . . . . . . . . . . 188
8.7 Case Study - IoT for E-waste Recycling System . . . . . . . 189
8.8 Future Trends of IIoT . . . . . . . . . . . . . . . . . . . . . 192
8.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 195

9 A Multi-hazard Industry Assessment System

Based on Unmanned Aerial Vehicles (UAVs) for Bridges
Crossing Seasonal Rivers 199
Allan J. Wilson, A. Pon Bharathi, M. Leeban Moses,
D. Vedha Vinodha, K. Kalaiselvi, and Kannan Pauliah Nadar
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 200
9.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . 203
9.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 206
9.3.1 UAV-derived DEM generation by 3D style . . . . . 206
9.3.2 Hydrodynamic analyses . . . . . . . . . . . . . . . 207
9.3.3 3D FEM generation . . . . . . . . . . . . . . . . . 208
9.3.4 Tectonic evaluation . . . . . . . . . . . . . . . . . 208
9.3.5 Soil modeling . . . . . . . . . . . . . . . . . . . . 208
9.3.6 Bridge modeling . . . . . . . . . . . . . . . . . . 211
9.4 Result and Discussion . . . . . . . . . . . . . . . . . . . . 213
9.4.1 Scour depth and flood load calculations by
hydraulic modeling . . . . . . . . . . . . . . . . . 213
9.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 219
xii Contents

10 Air Quality Prediction using Machine Learning Techniques for

Intelligent Monitoring Systems 223
Marimuthu Rajendran Ezhilkumar, Singaram Karthikeyan,
Dimplekumar N. Chalishajar, and Rajappa Ramesh
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 224
10.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . 226
10.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . 233
10.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 233

11 Facial Emotion Classification for Industry Automation using

Convolutional Neural Networks 237
R. Bharath Raam, Balaji Srinivasan, Prithiviraj Rajalingam,
and Dinesh Jackson Samuel
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 237
11.2 Related Works . . . . . . . . . . . . . . . . . . . . . . . . 238
11.3 Dataset Description . . . . . . . . . . . . . . . . . . . . . . 242
11.4 Model Architecture . . . . . . . . . . . . . . . . . . . . . . 242
11.5 Model Training . . . . . . . . . . . . . . . . . . . . . . . . 245
11.6 Model Metrics . . . . . . . . . . . . . . . . . . . . . . . . 247
11.7 Activation Maps . . . . . . . . . . . . . . . . . . . . . . . 248
11.8 Implementation Workflow . . . . . . . . . . . . . . . . . . 249
11.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 251

12 Automatic Hand Sanitizer Dispenser by Industrial Automation

using Arduino and Photodiode 255
R. Kiruthika, T. Prabhu, and Naveenbalaji Gowthaman
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 256
12.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . 257
12.2.1 The goal . . . . . . . . . . . . . . . . . . . . . . . 258
12.2.2 The extent of the sanitizer dispenser with
photodiode . . . . . . . . . . . . . . . . . . . . . 258
12.3 Design of Extensive Hand Sanitizer . . . . . . . . . . . . . 258
12.3.1 Requirements of the design . . . . . . . . . . . . . 259
12.3.2 Development of the hand sanitizer dispenser
framework . . . . . . . . . . . . . . . . . . . . . . 259
12.3.3 Default pins and conditions associated
with the sensor . . . . . . . . . . . . . . . . . . . 259
12.3.4 Adjusting procedure of the potentiometer . . . . . 260
12.4 Theoretical and Mathematical Proofs . . . . . . . . . . . . 261
 Contents xiii

12.5 Schematic Layout of the Proposed Work . . . . . . . . . . . 262

12.6 Examination of the Proposed Work . . . . . . . . . . . . . 262
12.7 A Comparative Study with the Market Products . . . . . . . 264
12.8 Drawbacks of the Conventional System of Dispensers . . . 265
12.9 Performance Comparison of Automatic Hand Sanitizer
Dispensers . . . . . . . . . . . . . . . . . . . . . . . . . . 266
12.10 The Desirable Properties of the Proposed Idea . . . . . . . . 266
12.11 Additional Perks of the Industrial Automated System . . . . 266
12.12 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 267

13 Energy Efficiency Investigation in Massive MIMO

Communication for Industry Automation using Precoding
Schemes 273
R. Surender, S. Sudharsan, S. Sundaresan,
and Sunday Adeola Ajagbe
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 274
13.2 Evolution of Wireless Communication System . . . . . . . 275
13.3 Challenges in the 5G Communication System . . . . . . . . 276
13.4 Massive Multiple Input Multiple Output . . . . . . . . . . . 277
13.5 Precoding Schemes . . . . . . . . . . . . . . . . . . . . . . 285
13.5.1 Maximal ratio combining . . . . . . . . . . . . . . 285
13.5.2 Zero forcing . . . . . . . . . . . . . . . . . . . . . 286
13.5.3 Pilot zero-forcing (P-ZF) precoding scheme . . . . 286
13.6 Superimposed Pilots . . . . . . . . . . . . . . . . . . . . . 287
13.7 System Model . . . . . . . . . . . . . . . . . . . . . . . . 288
13.8 Regular Pilots . . . . . . . . . . . . . . . . . . . . . . . . . 288
13.9 Superimposed Pilot . . . . . . . . . . . . . . . . . . . . . . 289
13.10 Energy Efficiency for Different Precoding Schemes . . . . . 290
13.11 Average Rate per Unit Area . . . . . . . . . . . . . . . . . 290
13.12 Average Power Consumption . . . . . . . . . . . . . . . . . 291
13.13 Results and Discussion . . . . . . . . . . . . . . . . . . . . 292
13.13.1 Performance of various precoding schemes . . . . . 292
13.14 Power Amplifier Power of Different Precoding Schemes . . 294
13.14.1 Average sum rate of various precoding schemes
under different BS antennas . . . . . . . . . . . . . 295
13.14.2 Performance of various precoding schemes
for different UEs . . . . . . . . . . . . . . . . . . 296
13.14.3 Average rate per UEs using different precoding
schemes for various coherence block length . . . . 297
xiv Contents

13.14.4 Energy efficiency for various levels of SNR . . . . 298

13.15 Conclusion and Future Scope . . . . . . . . . . . . . . . . 299

14 Internet of Things and Cybersecurity Mechanism for

Industrial Automation Systems 305
Shitharth Selvarajan, Gouse Baig Mohammed,
Mohamed Sirajudeen Yoosuf, D. Shivaprasad, Mustapha Hedabou,
and C. K. Yogesh
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 306
14.1.1 IoT security . . . . . . . . . . . . . . . . . . . . . 309
14.1.2 IoT and cloud computing . . . . . . . . . . . . . . 309
14.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . 311
14.3 Security Requirements . . . . . . . . . . . . . . . . . . . . 314
14.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 316
14.4.1 Privacy and technology knowledge in IoT . . . . . 317
14.5 Cyberthreats on IoT . . . . . . . . . . . . . . . . . . . . . 318
14.5.1 Device attack . . . . . . . . . . . . . . . . . . . . 319
14.5.2 Network attack . . . . . . . . . . . . . . . . . . . 319
14.6 Data Security Attacks . . . . . . . . . . . . . . . . . . . . 320
14.6.1 SQL injection attack . . . . . . . . . . . . . . . . 321
14.6.2 Attack by cross-site scripting . . . . . . . . . . . . 321
14.7 Results and Discussion . . . . . . . . . . . . . . . . . . . . 322
14.7.1 Cyberattack in Industrial IoT devices . . . . . . . . 322
14.8 Cybersecurity Challenges in IIoT . . . . . . . . . . . . . . 326
14.8.1 IT/OT (Data security issues) . . . . . . . . . . . . 326
14.8.2 Cybersecurity threats to the Industrial Internet of
Things . . . . . . . . . . . . . . . . . . . . . . . . 326
14.9 The Roadmap for Unifying IIoT Standards . . . . . . . . . 328
14.10 Conclusion and Future Scope . . . . . . . . . . . . . . . . 330

15 Illustrating the Mesmerizing Edge AI Journey for Industry

Automation 335
Pethuru Raj and Chellammal Surianarayanan
15.1 Delineating the Paradigm of Cloud Computing . . . . . . . 336
15.2 Enterprises Embrace the Cloud-first Strategy . . . . . . . . 336
15.3 Tending Toward Cloud-native Computing . . . . . . . . . . 337
15.4 The Edge AI Enablers . . . . . . . . . . . . . . . . . . . . 339
15.5 Machine and Deep Learning (ML/DL) Algorithms on Edge
Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
15.6 The Emergence of TinyML . . . . . . . . . . . . . . . . . 342
 Contents xv

15.7 The Significance of Edge Security . . . . . . . . . . . . . . 343

15.8 Cloud AI vs. Edge AI . . . . . . . . . . . . . . . . . . . . 343
15.9 The Key Motivations of Edge AI . . . . . . . . . . . . . . . 345
15.10 The Edge AI Implications . . . . . . . . . . . . . . . . . . 346
15.11 The edge AI advantages . . . . . . . . . . . . . . . . . . . 347
15.11.1 Enabling real-time data capture and analytics . . . 347
15.11.2 Edge AI supplements with cloud AI . . . . . . . . 347
15.11.3 Transitioning to intelligent machinery and devices . 348
15.12 Edge AI – Individual and Industrial Use Cases . . . . . . . 349
15.12.1 Speech recognition . . . . . . . . . . . . . . . . . 349
15.12.2 Smart healthcare . . . . . . . . . . . . . . . . . . . 350
15.12.3 Ambient assisted living (AAL) . . . . . . . . . . . 350
15.12.4 Intelligent robots and drones . . . . . . . . . . . . 350
15.12.5 Predictive maintenance . . . . . . . . . . . . . . . 351
15.12.6 Edge AI in retail industry . . . . . . . . . . . . . . 351
15.12.7 Edge AI for video surveillance system . . . . . . . 351
15.12.8 Edge AI in smart homes . . . . . . . . . . . . . . . 352
15.12.9 Edge AI for banking and financial services . . . . . 352
15.12.10 Edge AI in automobile: Enhancing real-time
capabilities . . . . . . . . . . . . . . . . . . . . . 353
15.12.11 Edge AI is transforming Enterprises . . . . . . . . 353
15.12.12 Edge AI impacts the IoT . . . . . . . . . . . . . . 353
15.12.13 Edge AI for smarter homes . . . . . . . . . . . . . 354
15.12.14 Edge AI for smart cities . . . . . . . . . . . . . . . 354
15.12.15 Edge AI for smart manufacturing . . . . . . . . . . 354
15.12.16 Surveillance . . . . . . . . . . . . . . . . . . . . . 355
15.12.17 Mining, oil, and gas and industrial automation . . . 355
15.13 The Distinctions of Edge AI . . . . . . . . . . . . . . . . . 355
15.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 356
15.15 The Glossary of Terms . . . . . . . . . . . . . . . . . . . . 357

16 Edge Computing and Analytics: A New Computing Paradigm

for Better User Experience 361
Santosh Das and Pethuru Raj
16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 362
16.1.1 Edge devices . . . . . . . . . . . . . . . . . . . . 362
16.1.2 Network edge . . . . . . . . . . . . . . . . . . . . 362
16.1.3 On-premises infrastructure . . . . . . . . . . . . . 363
16.2 Importance of Edge Computing . . . . . . . . . . . . . . . 363
xvi Contents

16.3 The Recent Technological Fusion . . . . . . . . . . . . . . 363

16.4 Edge Computing Benefits and Application . . . . . . . . . 365
16.5 Edge Computing Examples . . . . . . . . . . . . . . . . . 366
16.6 Case Studies in Edge Computing . . . . . . . . . . . . . . . 368
16.7 Challenges and Opportunities in Edge Computing . . . . . . 368
16.8 Next-generation Analytics on the Cutting Edge . . . . . . . 369
16.9 The Urgent Requirement for Edge Analytics . . . . . . . . 370
16.10 Definitions of Edge Analytics . . . . . . . . . . . . . . . . 372
16.11 ML and Near-real-time Analytics on the Periphery . . . . . 373
16.12 Edge and Big Data Analytics . . . . . . . . . . . . . . . . . 375
16.13 Applications . . . . . . . . . . . . . . . . . . . . . . . . . 376
16.14 Transportation . . . . . . . . . . . . . . . . . . . . . . . . 378
16.15 Hopes and Plans for the Future . . . . . . . . . . . . . . . . 382
16.16 Extreme 5G Networking . . . . . . . . . . . . . . . . . . . 384
16.16.1 A vocabulary for the 5G frontier . . . . . . . . . . 384
16.17 Edge Data Classification and Storage Analysis . . . . . . . 385
16.17.1 Classifying information for a cutting-edge approach 386
16.18 Edge-clustering Phenomenon . . . . . . . . . . . . . . . . 386
16.18.1 The primary benefits of edge clusters are as follows 386
16.19 Safety on the Fringe . . . . . . . . . . . . . . . . . . . . . 387
16.20 Alternatives in Edge Architecture . . . . . . . . . . . . . . 388
16.20.1 A boundary architectural choice . . . . . . . . . . 388
16.21 A Peripheral Approach to Automation . . . . . . . . . . . . 388
16.22 The Edge and Data Sovereignty . . . . . . . . . . . . . . . 389
16.23 Problems and Obstacles . . . . . . . . . . . . . . . . . . . 389
16.24 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 390

Index 393

About the Editors 395

 Preface

With the faster maturity and stability of digital technologies and tools, all
kinds of industrial houses across the globe are journeying through delectable
and decisive advancements. Industry verticals are experiencing a dazzling
array of evolutions and revolutions in their offerings, operations, and outputs.
Industrial processes are continuously optimized with the strategic aim of
accelerating and automating industrial activities. Business behemoths and
start-ups meticulously leverage all noteworthy improvisations occurring in
the digital technology ecosystem. There are plentiful digitization and edge
technologies such as sensors, actuators, stickers, codes, chips and controllers,
beacons, RFID tags, LED lights, etc., for empowering all kinds of phys-
ical, mechanical, and electrical systems in our midst to be methodically
digitized. With the adoption of powerful communication technologies and
network topologies, digitized entities in our everyday environments (homes,
hotels, hospitals, industrial floors, airports, etc.) become connected. These
are being termed as networked embedded systems. Further on, digitized
elements increasingly are getting integrated with the web-based software
applications and data sources in order to be smartly enabled in their actions
and reactions. The dynamic network of Internet-enabled digitized entities is
being pronounced as the Internet of Things (IoT). When the Internet-attached
sensors and devices interact with one another in the vicinity and with the
software libraries and packages running in the Internet servers, there is a
possibility for the realization of a massive amount of multi-structured digital
data. Thus, the aspect of digitization is becoming penetrative and pervasive
these days.
Another concept capturing the imagination of technocrats and academic
professors is the cyber−physical system (CPS) paradigm. All kinds of phys-
ical systems on the ground are formally hooked into web-based software
services and datastores (cyber systems) and thereby empowered physical
systems can exhibit an adaptive behavior in their assignments and obliga-
tions. Thus, software-enabled devices, equipment, appliances, instruments,
robots, and drones are all set to enthuse the total human society. There will

xvii
xviii Preface

be people-centric, event-driven, service-oriented, knowledge-filled, process-

aware, business-critical, context-sensitive, and real-world cyber applications
that can be elegantly designed and deployed in order to succulently delight
employees, end-users, and executives.
The digital twins is another popular subject of study and research
across information and communication technologies (ICT) organizations and
academic institutions. With the Internet-enabled physical systems, a wider
variety of automation and acceleration activities are being unearthed and
unleased. For designing, developing, and deploying mission-critical systems
such as satellites, rockets, humanoid robots, drones, manufacturing machiner-
ies, defense equipment, medical instruments, etc., the role and responsibility
of AI-inspired and blockchain-supported digital twins are on the higher
side. We have explained these technological paradigms in the book chapters
for enlightening our esteemed readers. Highly reliable and high-bandwidth
communication technologies such as 5G are being widely put in place to
ensure low-latency applications. Cloud-native, edge, and server-less comput-
ing paradigms are also acquiring a surging popularity as the next-generation
phenomenon for enabling the much-talked digital transformation targets.
Next in line is the arrival of competent digitalization technologies. The
tremendous amount of digital data has to be subjected to a variety of deeper
investigations to extract actionable insights. That is, the challenges of transi-
tioning raw data into information and into knowledge are heavily minimized
through the careful application of digitalization technologies such as the data
analytics methods and platforms, artificial intelligence (AI) algorithms and
models, etc. Thus, digital data and making sense out of digital data are the
twin requirements for bringing forth a galaxy of state-of-the-art systems for
real business transformation. The blending of the IoT, artificial intelligence,
and blockchain is being proclaimed as the technology cluster for the future.
With the continued advancements in the digital technology space, there
are mesmerizing innovations and disruptions happening in the industry front.
Today we extensively talk about Industry 4.0 and 5.0 applications. Smart
manufacturing will see the light, whereas factory automation initiatives and
implementations will be speeded up so that the industry of the future will
be agile, adaptive, affordable, and adroit. This book is designed and devel-
oped with the serious goal of putting forward the details of the significance
of industry automation and how various digital technologies individually
and collaboratively contribute for simplifying and speeding up the complex
process of industry automation. Also, the book chapters will illustrate how
digital technologies assist and aid in envisaging and implementing intelligent
 Preface xix

industry applications. This book explains how people, industrial assets and
machineries, ICT infrastructures, and optimized processes gel well in order
to establish and sustain digitally transformed factories, which are rewarding,
risk-free, resilient, and versatile.
 List of Figures

Figure 1.1 Industrial Internet of Things framework. . . . . . . 4

Figure 1.2 A general IIoT framework system. . . . . . . . . . 7
Figure 1.3 Critical elements of industrial internet. . . . . . . . 9
Figure 2.1 Market value of AI. . . . . . . . . . . . . . . . . . 27
Figure 2.2 Impacts of AI on the industry. . . . . . . . . . . . . 28
Figure 2.3 Percentages of AI technology breakdown. . . . . . 28
Figure 2.4 Customer-driven production. . . . . . . . . . . . . 29
Figure 2.5 RAMI4.0 model. . . . . . . . . . . . . . . . . . . 30
Figure 2.6 SAE standard level of automation for
autonomous car. . . . . . . . . . . . . . . . . . . . 32
Figure 2.7 The scenario of order-controlled production. . . . . 34
Figure 2.8 The scenario of smart production. . . . . . . . . . 35
Figure 2.9 The scenario of innovative product development. . 36
Figure 2.10 The scenario of seamless and dynamic engineering
of plants usage. . . . . . . . . . . . . . . . . . . . 37
Figure 2.11 The scenario of circular economy. . . . . . . . . . 38
Figure 2.12 The scenario of different actors in the value chain. . 39
Figure 2.13 Predictive analytics in healthcare: use cases. . . . . 41
Figure 2.14 Example of surgical robot. . . . . . . . . . . . . . 42
Figure 2.15 Graphical representation – effect of predictive
maintenance. . . . . . . . . . . . . . . . . . . . . 43
Figure 2.16 The scenario of robotic process automation. . . . . 44
Figure 2.17 Benefits of AI in manufacturing industries. . . . . . 46
Figure 2.18 Features of AI-powered quality assurance. . . . . . 48
Figure 2.19 AI effect on industrial growth. . . . . . . . . . . . 50
Figure 2.20 The effect of AI on the growth of different industries. 50
Figure 2.21 Scenario of AIoT. . . . . . . . . . . . . . . . . . . 51
Figure 2.22 The trend of next industry automation. . . . . . . . 53
Figure 3.1 Bibliographic analysis of Industry 5.0. . . . . . . . 61
Figure 3.2 From Industry 1.0 to Future Industry 5.0. . . . . . . 62
Figure 3.3 Wafer fabric automation. . . . . . . . . . . . . . . 68

xxi
xxii List of Figures

Figure 3.4 Impact of AI on different use cases (industries). . . 70

Figure 3.5 Percentage of technology adopted for deploying AI
in Industry. . . . . . . . . . . . . . . . . . . . . . 72
Figure 3.6 Automated trace analysis. . . . . . . . . . . . . . . 74
Figure 3.7 Aggregated product matching and trace analysis. . 76
Figure 3.8 AI spending on various sectors of India. . . . . . . 79
Figure 4.1 ML and data scientists. . . . . . . . . . . . . . . . 90
Figure 4.2 AI automation applications. . . . . . . . . . . . . . 94
Figure 5.1 Relation between quantum computing and machine
learning. . . . . . . . . . . . . . . . . . . . . . . . 103
Figure 5.2 Classical data vs. quantum data. . . . . . . . . . . 104
Figure 5.3 Classical bit and qubit. . . . . . . . . . . . . . . . 105
Figure 5.4 Quantum computing vs. machine learning. . . . . . 106
Figure 5.5 Classical machine learning vs. quantum machine
learning. . . . . . . . . . . . . . . . . . . . . . . . 108
Figure 5.6 Example to show principal component analysis in
quantum. . . . . . . . . . . . . . . . . . . . . . . 109
Figure 5.7 Support vector quantum machines. . . . . . . . . . 110
Figure 5.8 Quantum machine learning model. . . . . . . . . . 112
Figure 5.9 Threats to some cutting-edge technologies. . . . . . 115
Figure 5.10 Wavefunction for the electrons. . . . . . . . . . . . 118
Figure 5.11 Variational quantum eigensolver. . . . . . . . . . . 119
Figure 5.12 Quantum neural network. . . . . . . . . . . . . . . 120
Figure 5.13 Quantum machine learning model. . . . . . . . . . 122
Figure 5.14 Separation of data in input and feature space. . . . 122
Figure 5.15 Quantum property space. . . . . . . . . . . . . . . 123
Figure 6.1 Machine vision. . . . . . . . . . . . . . . . . . . . 138
Figure 6.2 Image sensing pipeline. . . . . . . . . . . . . . . . 140
Figure 6.3 Timeline of the latest advancement in computer
vision algorithms. . . . . . . . . . . . . . . . . . . 142
Figure 6.4 Overview of salami placement in pizza. . . . . . . 143
Figure 6.5 Fundamental steps in digital image processing in
machine vision. . . . . . . . . . . . . . . . . . . . 146
Figure 6.6 Computer vision in automobile industries. . . . . . 152
Figure 6.7 Intersection over union (IoU). . . . . . . . . . . . . 156
Figure 6.8 Precision. . . . . . . . . . . . . . . . . . . . . . . 157
Figure 6.9 Recall. . . . . . . . . . . . . . . . . . . . . . . . . 157
Figure 7.1 Evolution of Industry 4.0. . . . . . . . . . . . . . . 165
Figure 7.2 Industry 4.0 environment. . . . . . . . . . . . . . . 168
 List of Figures xxiii

Figure 7.3 Critical elements of Industry 4.0. . . . . . . . . . . 168

Figure 7.4 Challenges in waste management system. . . . . . 171
Figure 7.5 Visualization of the dataset. . . . . . . . . . . . . . 176
Figure 7.6 Accuracy evaluation of machine learning
algorithms. . . . . . . . . . . . . . . . . . . . . . 177
Figure 8.1 Categorization of e-waste. . . . . . . . . . . . . . 190
Figure 8.2 Process in e-waste recycling. . . . . . . . . . . . . 191
Figure 8.3 Hardware setup in dock door of the storage center. . 191
Figure 8.4 RFID tag information. . . . . . . . . . . . . . . . . 192
Figure 8.5 E-waste dispatched for recycling. . . . . . . . . . . 192
Figure 9.1 Structure of warning system. . . . . . . . . . . . . 201
Figure 9.2 Proposed system. . . . . . . . . . . . . . . . . . . 206
Figure 9.3 Modeling procedures to access flood consequences
from meterological effects. . . . . . . . . . . . . . 210
Figure 9.4 Model data visualization: (a) precipitation (input),
(b) runoff, (c) mizuRoute river flow, and (d) flood
depth. . . . . . . . . . . . . . . . . . . . . . . . . 210
Figure 9.5 (a) Parameters that have been set; (b) the pressures
and moments acting on the flooded bridge deck. . . 212
Figure 9.6 UAV images taken between the seasons (a)–(k)
November 2016 and January 2021. . . . . . . . . . 213
Figure 9.7 (a) Digital elevation data from TanDEM-X;
(b) drainage lines with sub-basins. . . . . . . . . . 214
Figure 9.8 2D hydraulic model of the research area. . . . . . . 215
Figure 9.9 The scaled suite’s RotD100 spectrum, the selected
record, and the desired spectrum. . . . . . . . . . . 216
Figure 9.10 For current and multi-hazard circumstances, the
maximum pier column displacement at the higher
ends. . . . . . . . . . . . . . . . . . . . . . . . . . 216
Figure 9.11 For current and multi-hazard circumstances, the
maximum pier column internal forces. . . . . . . . 217
Figure 9.12 For current and multi-hazard scenarios, the maxi-
mum pile of internal forces. . . . . . . . . . . . . . 218
Figure 10.1 Map showing the sampling locations. . . . . . . . . 227
Figure 10.2 Neutral network architecture was developed for this
study. . . . . . . . . . . . . . . . . . . . . . . . . 228
Figure 10.3 Mass concentration profile of NSC. . . . . . . . . . 228
Figure 10.4 Mass concentration profile of ASC. . . . . . . . . . 229
Figure 10.5 Mass concentration profile of SSCV. . . . . . . . . 229
xxiv List of Figures

Figure 10.6 PM concentration prediction by ANN model

for NSC. . . . . . . . . . . . . . . . . . . . . . . . 230
Figure 10.7 PM concentration prediction by ANN model
for ASC. . . . . . . . . . . . . . . . . . . . . . . . 231
Figure 10.8 PM concentration prediction by ANN model
for SSCV. . . . . . . . . . . . . . . . . . . . . . . 232
Figure 11.1 Samples from FER-2013 dataset [12]. . . . . . . . 243
Figure 11.2 Samples from JAFFE dataset [13]. . . . . . . . . . 243
Figure 11.3 Model visualization. . . . . . . . . . . . . . . . . . 244
Figure 11.4 Model training and validation: (a) accuracy
and (b) loss. . . . . . . . . . . . . . . . . . . . . . 246
Figure 11.5 Confusion matrix. . . . . . . . . . . . . . . . . . . 248
Figure 11.6 Activation maps. . . . . . . . . . . . . . . . . . . 249
Figure 11.7 Implementation workflow. . . . . . . . . . . . . . 250
Figure 11.8 MTCNN architecture [22]: P-NET, R-NET,
and O-NET. . . . . . . . . . . . . . . . . . . . . . 250
Figure 12.1 Schematic layout of the automatic hand sanitizer
dispenser. . . . . . . . . . . . . . . . . . . . . . . 262
Figure 13.1 Illustration of transmission in an mMIMO. . . . . . 278
Figure 13.2 Illustration of precoding scheme at the transmitter. . 279
Figure 13.3 Regular and superimposed pilot protocol. . . . . . 287
Figure 13.4 Cell under investigation surrounded by a cluster. . . 292
Figure 13.5 Energy efficiency of various precoding schemes
under different BS antennas. . . . . . . . . . . . . 293
Figure 13.6 Total power amplifier power of various precoding
schemes under different BS antennas. . . . . . . . 294
Figure 13.7 Average sum rate for various precoding schemes
under different BS antennas. . . . . . . . . . . . . 295
Figure 13.8 Energy efficiency of UEs in the network. . . . . . . 296
Figure 13.9 Energy efficiency of the network for different coher-
ence block length. . . . . . . . . . . . . . . . . . . 297
Figure 13.10 Average sum rate per UE for different SNRs in the
network. . . . . . . . . . . . . . . . . . . . . . . . 298
Figure 14.1 The study’s scope, structure, and objectives. . . . . 308
Figure 14.2 Hierarchy of ransomware families. . . . . . . . . . 310
Figure 14.3 Futuristic Internet of Everything (IoE)
ecosystem [10]. . . . . . . . . . . . . . . . . . . . 313
Figure 14.4 Characteristics of a future IoE [21]. . . . . . . . . . 313
Figure 14.5 IERC definition of IoT [22]. . . . . . . . . . . . . 314
 List of Figures xxv

Figure 14.6 AutoBot Catcher system flow. . . . . . . . . . . . 317

Figure 14.7 Proposed research model. . . . . . . . . . . . . . . 317
Figure 14.8 The benign and malware detection clustering. . . . 324
Figure 14.9 Accuracy of classifiers based on machine learning. 325
Figure 14.10 Comparing false-positive and true-positive rates. . . 325
Figure 14.11 Cybersecurity taxonomy for IIoT. . . . . . . . . . . 327
Figure 14.12 Unified IIoT standards roadmap [23]. . . . . . . . . 329
Figure 16.1 Venn diagram showing the overlap between edge
devices, edge computing, and edge analytics. . . . 370
 List of Tables

Table 3.1 A comparison of different industrial revolutions’

visions. . . . . . . . . . . . . . . . . . . . . . . . . . 63
Table 3.2 Impacts associated with robot integration in
industries. . . . . . . . . . . . . . . . . . . . . . . . 64
Table 3.3a Amount invested in Indian start-ups in 2022. . . . . . 79
Table 3.3b Amount invested in the last seven years in Indian
start-ups. . . . . . . . . . . . . . . . . . . . . . . . . 79
Table 4.1 Industry revolution. . . . . . . . . . . . . . . . . . . 88
Table 4.2 ML and AI techniques. . . . . . . . . . . . . . . . . 89
Table 6.1 Comparison between human vision and computer
vision. . . . . . . . . . . . . . . . . . . . . . . . . . 137
Table 6.2 Components of vision systems. . . . . . . . . . . . . 139
Table 6.3 Object defect detection using computer vision. . . . . 150
Table 7.1 Performance evaluation of machine learning
algorithms. . . . . . . . . . . . . . . . . . . . . . . . 177
Table 8.1 Advantages and disadvantages of IIoT. . . . . . . . . 186
Table 9.1 Ground motion collection. . . . . . . . . . . . . . . . 214
Table 10.1 List of parameters measured and monitored
in this study. . . . . . . . . . . . . . . . . . . . . . . 227
Table 11.1 Pseudocode for model training. . . . . . . . . . . . . 246
Table 11.2 Model metrics. . . . . . . . . . . . . . . . . . . . . . 247
Table 12.1 A comparative study of the parameters of the designed
prototype with the available dispensers in the market. 264
Table 13.1 Simulation parameters. . . . . . . . . . . . . . . . . 293
Table 14.1 WSN vulnerabilities (main threats and
countermeasures). . . . . . . . . . . . . . . . . . . . 320
Table 14.2 RFID vulnerabilities (main threats and
countermeasures). . . . . . . . . . . . . . . . . . . . 320
Table 14.3 Wi-Fi vulnerabilities (main threats and
countermeasures). . . . . . . . . . . . . . . . . . . . 320
Table 14.4 Methods of prevention. . . . . . . . . . . . . . . . . 323

xxvii
 List of Contributors

Ajagbe, Sunday Adeola, Department of Computer Engineering, Ladoke

Akintola University of Technology, LAUTECH, Nigeria
Akoramurthy, B., School of Computing & Information Technology, REVA
University, India
Alkali, Yusuf Jibrin, Federal Inland Revenue Service, Nigeria
Amesho, Kassian T. T., Tshwane School for Business and Society, Tshwane
University of Technology, South Africa
Ananth, Christo, Samarkand State University, Uzbekistan
Chalishajar, Dimplekumar N., Department of Applied Mathematics, Vir-
ginia Military Institute (VMI), USA
Das, Santosh, SWAN Lab, Indian Institute of Technology, Kharagpur;
Department of Computer Science & Engineering, OmDayal Group of Insti-
tutions, India
Devi, A., Department of ECE, IFET College of Engineering, India
Dharanyadevi, P., Department of CSE, Puducherry Technological Univer-
sity, India
Dharishini, P. Padma Priya, Department of CSE, Faculty of Engineering
and Technology, Ramaiah University of Applied Sciences, India
Dhivya, K., Department of Computer Science & Engineering, India
Ezhilkumar, Marimuthu Rajendran, Department of Civil Engineering, Sri
Krishna College of Engineering and Technology, India
Gowthaman, Naveenbalaji, Electronic Engineering, University of KwaZulu-
Natal, South Africa
Hedabou, Mustapha, School of Computer Science, University Mohammed
VI Polytechnic, Morocco

xxix
xxx List of Contributors

Jyothi, A. P., Department of CSE, Faculty of Engineering and Technology,

Ramaiah University of Applied Sciences, India
Kalaiselvi, K., Department of ECE, Hindusthan College of Engineering and
Technology, India
Karkavelraja, J., Department of CSE, Puducherry Technological University,
India
Karthikeyan, Singaram, Centre for Environmental Studies, Anna Univer-
sity, India
Khadidos, Alaa O., Department of Information Technology, Faculty of
Computing and Information Technology, King Abdulaziz University, Saudi
Arabia
Kiruthika, R., Electronics and Communication Engineering, SNS College of
Technology, India
Krishnakoumar, C., Department of CSE, Puducherry Technological Univer-
sity, India
Lincy, S. S. Blessy Trencia, School of Mathematics and Data Science,
Emirates Aviation University, Dubai, The United Arab Emirates
Mohammed, Gouse Baig, Department of Computer Science & Engineering,
Vardhaman College of Engineering, India
Moses, M. Leeban, Department of ECE, Bannari Amman Institute of
Technology, India
Nadikattu, Rahul Reddy, Department of IT, University of Cumbersome,
USA
Nalini, M., Department of Electronics and Instrumentation Engineering, Sri
Sairam Engineering
Pauliah Nadar, Kannan, Department of Electrical and Computer Engineer-
ing, Institute of Technology, Jigjiga University, Ethiopia
Pon Bharathi, A., Department of ECE, Amrita College of Engineering and
Technology, India
Prabhu, T., Electronics and Communication Engineering, Presidency Uni-
versity, India
Raam, R. Bharath, Loyola-ICAM College of Engineering and Technology,
India
 List of Contributors xxxi

Raj, Pethuru, Edge AI Division, Reliance Jio Platforms Ltd., India

Rajalingam, Prithiviraj, SRM University, India
Ramesh, Rajappa, Department of Science & Humanities, Sri Krishna
College of Engineering and Technology, India
Ravi Kumar, A., Department of Computer Science & Engineering, Sridevi
Women’s Engineering College, India
Revathy, K., Department of CSE, Puducherry Technological University,
India
Samuel, Dinesh Jackson, Biomedical Engineering, University of California,
USA
Sangeetha, K., Department of Computer Science & Engineering, Kebri
Dehar University, Ethiopia
Senthil, P., Department of ECE, SRM Institute of Science & Technology
Ramapuram Campus, India
Shitharth, S., Department of Computer Science & Engineering, Kebri Dehar
University, Ethiopia
Shivaprasad, D., School of Science and Technology, International University
of East Africa, Uganda
Sirajudeen, School of Computer Science and Engineering, VIT University,
India
Srinivasan, Balaji, Loyola-ICAM College of Engineering and Technology,
India
Sudharsan, S., Department of ECE, Rajalakshmi Engineering College, India
Sundaresan, S., Department of Electronics & Communication, NIT
Karaikal, India
Surender, R., Department of ECE, SRM Institute of Science & Technology
Ramapuram Campus, India
Surianarayanan, Chellammal, Centre for Distance and Online Education,
Bharathidasan University, India
Therese, M. Julie, Department of ECE, Sri Manakula Vinayagar Engineer-
ing College, India
xxxii List of Contributors

Vatchala, S., School of Computer Science and Engineering, VIT Chennai

Campus, India
Velu, Arun, Equifax, USA
Venkatalakshmi, K., Department of ECE, University College of Engineering
Tindivanam, India
Vinodha, D. Vedha, Department of ECE, JCT College of Engineering and
Technology, India
Whig, Pawan, Vivekananda Institute of Professional Studies, India
Wilson, Allan J., Department of ECE, Amrita College of Engineering and
Technology, India
Yogesh, C. K., School of Computer Science and Engineering, VIT Chennai
Campus
Zayaraz, G., Department of CSE, Puducherry Technological University,
India
 List of Abbreviations

3GPP 3G partnership project

ABAI Automation based on AI
ABHS Alcohol-based hand sanitizer
ABS Anti-lock braking system/anti-skid breaking
system
AC Automated coding
ADAS Advanced driver aid system
ADC Analog to digital conversion
ADS Automated driving system
AFDS Ambient fine dust samplers
AGC Automatic gain control
AGI Artificial general intelligence
AI Artificial intelligence
AIoT Artificial intelligence of things
AL Algorithmic learning
ANFIS Adopted neuro-fuzzy inference system
ANN Artificial neural network
ANPR Automatic number plate recognition
AP Automatic hypothesis proving
API Application programming interface
AR Augmented reality
ASC Asymmetrical street canyon
AWS Amazon web services
BDA Big data analytics
BDMA Beam division multiple access
BPNN Back-propagation neural network
BS Base station
CAGR Compound annual growth rate
CAP Constrained application protocol
CH Cluster head
CHF Congestive heart failure
CMMS Computerized maintenance management system

xxxiii
xxxiv List of Abbreviations

CNN Convolutional neural network

CoAP Constrained application protocol
CP Collapse prevention
CPS Cyber−physical system
CPU Central processing unit
CRE Circular economy
CRM Customer relationship management
CSI Channel state information
CV Computer vision
CVS Concurrent versions system
DBM Deep Boltzmann machine
DBMS Database management system
DCNN Deep convolutional neural network
DDoS Distributed denial-of-service
DEM Digital elevation model
DL Deep learning
DPC Dirty paper coding
DR-PSLTE Dual-resolution polarimetric synthetic aperture
radar time-frequency domain data
DSP Digital signal processing
DTC Decision tree classifier
DYN Dynamic
ECC Elliptic curve cryptography
ECG Electrocardiography
ED-CNN Edge-deployed convolution neural network
ECNN Enhanced convolutional neural network
EC-OU Energy consumption optimization using
EDA Event-driven architecture
EDRLEACH Enhanced distributed energy-efficient LEACH
EEG Electroencephalography
EFFRA European factories of the future research
association
eMBB Enhanced mobile broadband
EMG Electromyogram
ENISA European union agency for cybersecurity
ERP Enterprise resource planning
ES-MAC Energy-saving medium access control
ESODR Earth system observing data and information
retrieval
 List of Abbreviations xxxv

EWS Early warning system

FDC Fault detection and classification
FEM Finite element model
FERA 2015 Facial expression recognition and analysis
FFNN Feedforward neural network
FL Federated learning
FN False negative
FP False positive
FUSE Framework for understanding structural errors
GCP Ground control point
GDP Gross domestic product
GIS Geographic information system
GPU Graphics processing unit
GSD Ground sampling distance
GSI Global standards initiative
GSR Galvanic skin response
HCI Human-computer interaction
HEC-RAS Hydrologic engineering center’s river analysis
system
HHL Harrow, Hassidim, and Lloyd
HOG Histogram of oriented gradient
HRM Human resource management
I4.0 Fourth industrial revolution
I4.0 Industry 4.0
I-A Industrial automation and artificial intelligence
ICS Industrial control systems
ICT Information and communication technologies
ITS Intelligent transportation systems
IDS Intrusion detection system
IDT Intrusion detection technique
IEC International electrotechnical commission
IERC Industrial internet consortium
IHSS Intelligent healthcare systems
IIC Industrial internet consortium
IIoT Industrial internet of things
IIRA Industrial internet reference architecture
IIS Internet information services
InTraSafEd5G INcreasing TRAffic SAFety with EDge and 5G
IO Immediate occupancy
xxxvi List of Abbreviations

IoD Internet of data

IoE Internet of everything
IoP Internet of people
IoS Internet of services
IoT Internet of things
IoU Intersection over union
IP Internet protocol
IPA Intelligent process automation
IPD Innovative product development
IPv6 Internet protocol version 6
IQ Intelligence quotient
IR Infrared
ISP Intelligent power system
ISWA International solid waste association
IT Information technology
ITS Intelligent mobility system
JAFFE Japanese female facial expression
KF-MAC Kalman filter-based medium access control
KNN K-neighbor classifier
KPI Key performance indicator
LDA Linear discriminant analysis
LEACH Low energy adaptive clustering hierarchy
LIVE Layered information viewing environment
LR Logistic regression
LS Life safety
LSTM Long short-term memory
LTE Long-term evolution
LTE-A Long-term evolution advanced
LWC Lightweight cipher
M2M Machine-to-machine
MaaS Manufacturing as a service
MAE Mean absolute error
MANET Mobile Ad hoc network
MCP Mobile controlled production
MCS Monte Carlo simulation
MEC Multi-access edge computing
MF Matched filter
MIMO Multiple input multiple output
ML Machine learning
 List of Abbreviations xxxvii

MLP Multilayer perceptron

MMF Max-min fairness
mMIMO Massive multiple input multiple output
MMSE Minimum mean square error
mMTC Massive machine type communication
MQTT Messaging queue telemetry transport
MRC Maximal ratio combining
MRF Materials recovery facility
MRT Multiple-input multiple-output radio
transmission
MSA Microservices architecture
MTCNN Multi-task cascaded convolutional network
NAN Neighborhood area network
NB Naive Bayes
NFC Near field communication
NFV Network function virtualization
NLP Natural language processing
NPU Neural processing unit
NSC Non-street canyon
OCAP Out of control action plan
OCP Order-controlled production
OCR Optical character recognition
OEE Overall equipment effectiveness
ORLEACH Optimized Re-clustering LEACH
OS Operating system
OT Operational technology
PaaS Platform as a service
PC Pre-stressed concrete
PCA Principal component analysis
PdM Predictive maintenance
PI4.0 Industry 4.0 platform
PLC Power line communication
PM Particulate matter
POS Point of sales
P-ZF Pilot zero-forcing
QE Quantitative easing
QHCR Qualified and heterogeneous cluster routing
QML Quantum machine learning
QoE Quality of everything
xxxviii List of Abbreviations

QoS Quality of service

QRAM Quantum random access memory
Q-SVM Quantum support vector
RAMI4.0 Reference architectural model industry 4.0
RAN Radio access network
RBAC Role-based access control
RBF Radial basis function
RC Remote control
RES Renewable energy sources
RFID Radio-frequency identification
RMSE Root mean square error
ROI Region of interest
RP Regular pilot
RPA Robotic process automation
RR Respiration rate
RS Robotic systems
RSA Rivest-Shamir-Adleman (encryption algorithm)
RSU Road side unit
RTK Real-time kinematic
RTU Remote telemetry unit
RZF Regularized zero-forcing
SaaS Software as a service
SAE Society of automotive engineers
SC Street canyon
SCA Small cell access
SCADA Supervisory control and data acquisition
SDMA Space division multiple access
SDN Software-defined networking
SDP Seamless and dynamic engineering of plants
SHM Structural health monitoring
SI Self-interference
SINR Signal-to-interference noise ratio
SLR Single lens reflex
SNR Signal-to-noise ratio
SoF State of function
SoH State of health
SRE Site reliability engineering
SSCV Symmetrical street canyon with viaduct
SSI Soil structure interaction
 List of Abbreviations xxxix

SSP Social signal processing

SVM Support vector machine
TanDEM-X TerraSAR-X add-on for digital elevation
measurement
TCP Transmission control protocol
TCS Traction control system
TN True negative
TP True positive
TPU Tensor processing unit
TSN Time-sensitive networking
UAV Unmanned aerial vehicle
UI Urban intelligence
UMTS Universal mobile telecommunication system
UPF User plane function
URLLC Ultra-reliable low latency communication
VM Virtual machine
VPU Vision processing unit
VQE Variational quantum eigensolver
VSM Value stream management
WEVA Wireless evolution for automation
WoT Web of things
WSN Wireless sensor network
ZF Zero-forcing
 1
An Analytical Framework for the Industrial
Internet of Things (IIoT): Importance, Recent
Challenges, and Enabling Technologies

Gouse Baig Mohammed1 , Shitharth Selvarajan2 , A. Ravi Kumar3 ,

K. Sangeetha4 , Alaa O. Khadidos5 , and S. Vatchala6
1 Department of Computer Science & Engineering, Vardhaman College of
Engineering, India
2 School of Built Environment, Engineering and Computing, Leeds Beckett

University, LS1 3HE Leeds, U.K.

3 Department of Computer Science & Engineering, Sridevi Women’s

Engineering College, India

4 Department of Computer Science & Engineering, Kebri Dehar University,

Ethiopia
5 Department of Information Technology, Faculty of Computing and

Information Technology, King Abdulaziz University, Saudi Arabia

6 School of Computer Science and Engineering, VIT Chennai Campus, India

E-mail: gousebaig@[Link]; [Link]@[Link];

aravikumar007@[Link]; sangeethak@[Link];
aokhadidos@[Link]; vatchala.s@[Link]

Abstract
The Internet of Things (IoT) is a novel idea that can benefit any manufacturing
company that adopts it. IoT is still in its early stages in industrial operations,
leading to higher prices, slower development in data management, and fewer
deployments. The proliferation of Internet of Things (IoT) applications and
the adoption of cutting-edge technology trends in industrial systems are
driving the development of industrial IoT (IIoT). A novel vision of the
Internet of Things applied to the manufacturing sector is realized when smart
things are used to automatically detect, gather, process, and communicate

1
2 An Analytical Framework for the Industrial Internet of Things (IIoT)

real-time events in industrial processes. By creating smart monitoring of

production floor shops and machine health applications and for predictive
and preventative maintenance of industrial equipment, the industrial Internet
of Things (IIoT) seeks to enhance operational efficiency, productivity, and
the management of industrial assets. Due to the proliferation of IoT (Internet
of Things) applications that gather information from real and virtual sen-
sors, massive amounts of digital data are becoming increasingly important.
However, without the right tools, such information is useless. This research
provides a novel and concise definition of IIoT that can aid readers in their
understanding of this emerging field. Current research trends in the II T have
been outlined. We conclude by outlining the current issues and enabling
technologies for the IIoT.

Keywords: Industrial Internet of Things, cyber-physical systems, cloud

computing, machine to machine (M2M) communication

1.1 Introduction
The Internet of Things (IoT) continues to be one of the currently avail-
able options that are the most comprehensive, productive, and cost-effective
choice. Physical data is augmented with computer systems, which are then
deployed to give sensors the impression that they are more powerful, trust-
worthy, and in control of their environments [1]. The objective of the subset
of the IoT known as the industrial Intemet of Things (IIoT) is to improve the
safety, security, and reliability of industrial processes while simultaneously
maintaining the capability for real-time data exchange. The industrial Internet
of Things (IIoT) allows businesses to improve their asset and operation
management and reduce downtime through predictive maintenance. The
Internet of Things (IoT) primarily seeks to improve workplace security and
operational effectiveness in manufacturing (IIoT). In the next generation of
industrial networks, it is anticipated that large-scale applications of the indus-
trial Internet of Things (IIoT) will be implemented. The industrial Intemet of
Things (IIOT) will help break down barriers between humans and robots,
which is necessary to realize the vast personalization potential of Industry
6.0 systems. As a result, we will discuss the industrial Intemet of Things and
how it relates to the final objective of the Industry 4.0 initiative. According to
the most recent forecasts [2], seventy billion devices will be connected to the
Internet by 2025 . The global market share of the IIoT is anticipated to reach
$14.2 trillion by 2023.
 1.1 Introduction 3

As a result, the Intemet of Things will be able to provide interconnectivity

between a variety of sensors and PLCs (in addition to other smart devices)
for the more advanced cyber systems that surround the significant tech-
nologies, “such as a smart grid and intelligent vehicle systems,” in addition
to city intelligence and other advanced cyber systems. Machine-to-machine
communications, also known as M2M communications, are on the verge of
becoming a reality, thanks to the widespread adoption of real-time protocols,
networks, and system software and hardware based on M2M. The Intemet of
Things (IoT) is a network that connects and manages various devices, such as
automobiles, buildings, and other physical infrastructure [2]. The Intemet of
Things Global Standards Initiative (IoT-GSI) referred to it as “infrastructure
for the information society by the Global Standards Initiative on the Intemet
of Things” in its official description in 2013. This description was published
in 2013. By utilizing the Intemet of Things, the physical integration of
computer-based applications can be improved, and monitoring and managing
artifacts can be carried out by utilizing pre-existing network infrastructures.
The Intemet of Things, which includes sensors and drives, is only a subset of
the more comprehensive cyber-physical networks.
Every object incorporates a computer system capable of connecting to the
pre-existing network infrastructure of the Internet and transmitting data. It is
anticipated that almost 50 billion devices will be connected to the Internet of
Things by 2020 [3, 4]. The Internet of Things (IoT) refers to a diverse col-
lection of devices connected to the web through various sensors and network
connections, including wired and wireless. A device that can do additional
research will accept data from embedded sensors connected to the Internet
of Things, humans, and any inanimate item (IoT). Thanks to the Internet
of Things, we can soon link industrial networks, smart medical telephone
monitoring systems, smart transportation systems, and information-sharing
equipment. With the rapid growth of IoT devices and networks in indus-
trial manufacturing, several sensors result in a large volume of data. The
inspection of production, which detects defects in the goods, is one of the
most common examples. In this chapter, we suggest a profound learning-
based classification model and introduce a rigorous inspection scheme with
greater precision to find potential faulty goods. As a factory may be equipped
with several assembly lines, the real-time processing of such big data is
a significant challenge in this case [4]. That is why we build our device
using the fog computing principle. The machine can handle extensive data by
downloading the calculation workload from the central server to the fog node.
In our scheme, there are two clear benefits. The first is that the CNN paradigm
is adapted to the fog computing environment and increases its computing
4 An Analytical Framework for the Industrial Internet of Things (IIoT)

Figure 1.1 Industrial Internet of Things framework.

performance considerably. The other is that we develop an inspection model

to denote the form and degree of the defect simultaneously. The trials have
shown that the approach proposed is stable and efficient. The analytical
framework of IIoT is given in Fig 1.1

1.1.1 Industrial automation with IoT

Manufacturing is critical to economic growth in the globalization period [1]
and continues to play a vital role in economic development. It has a good
effect on both industrialized and developing countries growth. The manufac-
turing industry is using emerging technology to boost individual enterprises’
economic competitiveness and the sustainability of the whole industrial field.
In the manufacture of ICTs, a transfer from traditional to modern produc-
tion techniques has been possible [2]. As a part of the ICT application,
monitoring systems play a significant role in the supervision and control of
manufacturing processes. Recent improvements in IT allow the integration
of one complicated system for the supply chain of multiple monitoring
applications. Usually, a monitoring system is essential in illness predic-
tion, improved production, cost reduction, and early warning system [4].
Applying and integrating contemporary technology monitoring systems, such
as the Internet of Things (IoT) based sensors, is feasible. Industrial studies
 1.1 Introduction 5

showed significant benefits from IoT-based monitors, including improved

working conditions, defect avoidance, quality prediction, and support for
better decision-making by managers [6].
The Internet of Things (IoT) will be able to link increasingly sophisticated
cyber systems based on essential technologies such as intelligent power sys-
tems (ISPs), intelligent transportation systems (ICTs), intelligent healthcare
systems (IHSS), urban intelligence (UI), and other intelligent systems (IIS).
It is essential for the Internet of Things to be able to provide advanced
and intelligent communications to future electronic and intelligent machines
and computers that various standard and actual protocol implementations,
network realms, and machine-to-machine IoT software and hardware sys-
tems be implemented. The Internet of Things (IoT) refers to interconnected
electronic devices that collect data from other devices, such as automo-
biles. These devices include sensors, actuators, and communication systems
(IoT). Infrastructure for the information society was how the Internet of
Things Global Standards Initiative (IoT-GSI) referred to it back in 2013.
Implementing Internet of Things based monitoring and control of artifacts
over pre-existing-existing network infrastructure can increase efficiencies,
precision, and cost-effectiveness [6]. This can be accomplished by connecting
the artifacts in question to the Internet. Examples include the more extensive
cyber-physical networks that employ IoT enhancements such as sensors and
drives. These networks are also an example. Each is outfitted with a computer
system capable of communicating with the Internet [56]. By 2020, it is
anticipated that there will be approximately 50 billion devices connected to
the Internet of Things.
Sensing devices can now create large volumes of data, thanks to the
Internet of Things (IoT), raising the challenge of making this data useful. A
networked system can be optimized or interacted more often via the Internet
of Things to improve people’s quality of life [7]. While your ability to
acquire, analyze, and act on data depends on your background knowledge,
this does not guarantee your success. Big data analytics’ current state is
therefore clouded with doubt. Analyses can range from basic correlations
between outdoor temperature and grid loading to more complicated causative
relationships, such as roads causing a gradual delay in energy use owing to
electric vehicles or warmer days leading to more air conditioner usage and
higher system loads.
The industry is a term reflecting the latest phase in the production rev-
olution or, in other words, the smart industry. Automation and data are two
major topics in this era of the smart industry. The first is one of our key goals,
6 An Analytical Framework for the Industrial Internet of Things (IIoT)

and the other is one of our most valuable instruments. Some techniques of
artificial intelligence (AI), deep learning, and machine learning, for example,
will process and learn data and make computers and manipulators capable
of becoming human-like. IoT-enabled devices are increasingly used in smart
factories to gather more data, which is essential for AI approaches. With the
help of these data, several new methods have been found for automating and
efficient manufacturing processes.

1.1.2 Objective
a. To make the smart industrial environment, which allows users to monitor
and control industrial parameters on a mobile device in real-time.
b. To preserve the workstation’s energy by completing the cutter tools
blade aging system on the website.
For this research, the term “industrial IoT” refers to a framework for
investigating how Internet of Things (IoT) technologies are implemented in
industrial settings. During this study, our goal was to develop a framework for
analyzing the nature and applications of IoT devices and their vulnerabilities
and threats, which would be used as part of this process. For example, we
hope to discover cross-cutting risks and vulnerabilities by methodically char-
acterizing the devices rather than focusing on technological or sector-specific
difficulties.
The following is how the chapter is organized. Literature survey is
detailed in Section 1.3, which is framed within the context of Industry 4.0.
It is in Section 1.5 that we provide enabling technologies for IIoT. We lay
the framework and case studies in Section 1.6. Section 1.7 gives details about
challenges in IIoT, Section 1.8 details the application for IIoT framework,
and Section 1.9 concludes by pointing out areas of research that still need to
be carried out.

1.2 Literature Survey

Data from IoT devices and industrial data sources flows through layers 1 and
2 of the stack. Edge servers and cloud computing systems are used at layer
3 to run IIoT applications. The Industrial Internet Consortium [8] examined
this general design for IIoT systems. Enterprise applications are illustrated in
Layer-4. As shown in Figure 1.2, data and information are being orchestrated
for resource management, operational purposes, and the flow of data and
information between the various layers.
 1.2 Literature Survey 7

Figure 1.2 A general IIoT framework system.

Researchers have a new perspective on these designs because of the

numerous changes in location awareness and communication paradigms,
computational assignments and execution paradigms, and resource manage-
ment systems. These architectures are described in depth in the following
paragraphs: It was built by [9] and used open-source software and IoT
communication protocols to create the wireless evolution for automation
(WEVA) system. Everything from operating systems to protocols to access
gateways to services is part of the system’s infrastructure. In addition, the
network is monitored and controlled by WEVA using easy WSN, a graphical
administration application. According to the authors, IPv6 is vital for the
IIoT in terms of adaptability. However, achieving a high-performance IIoT
in terms of networks is not an easy process (latency, security, etc.). Many
academics, on the other hand, have failed to take into account the critical role
played by WSNs in industrial applications.
8 An Analytical Framework for the Industrial Internet of Things (IIoT)

Katsikeas et al. [10] have proposed an IoT suite that addresses several
issues, including identifying real-time object locations throughout manu-
facturing processes, establishing a network system that allows objects to
communicate between the network and other objects in real time, and so forth.
A cloud platform and a smart hub are essential for the IIoT suite design.
Managing IoT devices is done through the smart hub, which acts as both a
gateway and a management platform.
In [11], the phrase “Internet of Things” was coined to represent the
growing number of smart and connected objects and highlight the new
opportunities they can create. Because of their dynamic character, smart and
connected products stand out from the crowd. I believe it is their increased
skills and the data they can collect. For example, “Internet” and “Thing”
are derived from two words and two conceptions. Internet is defined as
TCP/IP, a standard communication protocol that connects computer networks
worldwide. An “unidentifiable” entity is implied by it.
After being coined in 1999 by Kevin Ashton of Auto-ID Center, the
phrase “Intemet of Things” has been used. Computers can interpret, iden-
tify, and monitor the world, thanks to RFID (radio frequency identification)
and sensor technologies presented by Ashton, who argued that we should
empower computers to acquire information independently without limits of
human-centered data to keep an accurate count and track of everything. This
would allow us to cut costs, eliminate wasteful spending, and determine
whether anything is needed to be repaired or replaced (Ashton 2009). As
well as being quoted in Forbes Magazine in 2002, Ashton prophesied that
RFID would lead to a 100% automation of data collection. According to his
statement, “We need an intemet for things so that computers can grasp the
real world,” which may be the first reported use of this word [13].
The Intemet of Things (IoT) envisions a scenario in which the Intemet
extends into the real world, connecting everyday things and physical products
to the virtual world. Access points for Intemet services are built-in, and
remote control and monitoring are possible [14]. People, media, and content
will not be the only things on the Intemet in the following years. It will
be a comprehensive system if it includes all real-world assets that develop
knowledge, share information, interact with people, and support business
activities.
We believe that the developments in information technology, communi-
cations, and microelectronics will continue for a long time in the Internet of
Things (IoT). As the size of communication modules, processors, and other
electronic components shrinks, energy usage decreases, and the price of these
 1.2 Literature Survey 9

components continues to fall; they can already be integrated into our everyday
products. As a result of embedded ICT, smart things play a vital part in this
IoT vision. They can see their surroundings by using sensors. They can also
communicate with each other and engage with people by using networking
systems that are built into their devices [15].
Most importantly, the IoT vision will have a significant impact on a
variety of daily activities. As shown in Figure 1.3, the Internet of Things
is a network of interconnected gadgets affecting many aspects of our daily
lives. For private users, IoT’s impact will be seen in the home and at work.
As for business users, the most noticeable consequences will be in industrial
manufacturing and automation, business and process management, logistics,
and smart transportation of products and people [16].
According to the company, the industrial internet is a network of sensors
and software that integrates complicated physical machines. Machine data
may be collected and analyzed with machine-to-machine connectivity and
the Intemet of Things (IoT). By connecting smart devices, machines, and
people at work, the Industrial Intemet Consortium claims that it will improve
decision-making through advanced analytics, leading to revolutionary eco-
nomic results. IoT in industrial contexts is a non-consumer application of
IoT [17].
Figure 1.3 shows the three major components of the industrial intemet.
First and foremost, there are intelligent machines. To do this, modern con-
trols, sensors, and software applications must be used to connect the world’s
machines as well as fleets, facilities, and networks. The strength of physics-
based analytics is combined with domain expertise, automation, and pre-
diction algorithms when doing advanced analytics to better understand how
machines and systems operate. Linked personnel at all times are included as a

Figure 1.3 Critical elements of industrial internet.

10 An Analytical Framework for the Industrial Internet of Things (IIoT)

third component, which encourages smarter operational design, maintenance,

and excellent service quality.

1.2.1 Industry 4.0

Factory automation, electricity, and mass labor fueled the first three industrial
revolutions (water and steam power, mass labor, and electrical energy). In
2011, the German government adopted “Industry 4.0” as a strategy to increase
the country’s gross domestic product [18]. With the ability to communi-
cate and make autonomous, decentralized decisions, CPS is crucial to this
revolution, which aims to increase industrial efficiency, productivity, safety,
and transparency. The American Industrial Internet and the German idea of
Industry 4.0 have a lot in common. To put it another way, the term “industrial
intemet” refers to a network of interconnected devices, computers, and people
that enables advanced data analytics to revolutionize industrial operations
and, in tum, the landscape for enterprises and individuals alike [5]. The rise
of the industrial intemet impacts businesses and individuals alike.
Value chain management is the context in which the various technologies
and ideas that make up “Industry 4.0 ” are discussed. In the modular smart
factories of Industry 4.0, CPS keeps tabs on the workings of the factory
floor, builds a digital twin of the real world, and makes decisions without
a command center. Through the IoT, CPS can have two-way conversations in
real time with one another and with humans (IoT). Intemet-based access is
provided for the provision of collaborative as well as collaboration services
(IoS). Research demonstrates that the standards for the industry are still being
developed, even though both the Intemet of Things and the industrial Internet
of Things are relatively new concepts. The market is fragmented, and the
range of applications is vast. Even within a single application, the number of
distinct technical solutions is staggering.
Perhaps 5-10 years will pass before standards are adequately developed.
Research in the manufacturing industry has shown that IoT sensors can
improve working conditions, avoid erroneous designs, provide fault diagnosis
and quality prediction, and assist managers in making better decisions. To
measure the air quality inside a facility, Moon et al. created an Intemet
of Things-based sensor [19]. Infrared sensors were used to capture data
such as temperature, humidity, CO2 level, dust, and odors. Since this tech-
nology is incredibly reliable, it should be able to help managers maintain
a healthy working environment for everyone in production. A low-cost
IoT sensor system was developed by [20] in order to avoid errors during
 1.2 Literature Survey 11

the design process of additive manufacturing. Researchers were able to

obtain data on temperature and humidity using the sensors. According to
the study, knowing the environmental factors during the design phase of
additive manufacturing could help reduce errors in the production process.
According to the study’s findings, IoT sensors can help give complete diag-
nostic data and improve diagnostic results. IoT and machine leaming were
combined in a framework [14]. Real-world implementation of the technol-
ogy was done using metal casting. We were able to forecast the quality
of castings using this technology and increase the operation control with
it. Securing the fourth industrial revolution framework was developed by
[15] by integrating sensors and SCADA systems. It was found that the
proposed system is feasible and will aid managers in migrating old systems to
Industry 4.0.
The introduction of IoT has enabled this shift from traditional to modern
digitalized production possible. It is possible to manage huge volumes of
sensor data input using big data technology since the number of sensors in
production is being multiplied. Ge et al. have used big data technologies
in IoT to build a conceptual paradigm for critical decision-making [21].
Managers can make better judgments when provided with a wealth of data
gathered from a number of sources (sensor devices).
A framework for simultaneous temperature monitoring on at least two
fishing vessels was examined experimentally by [22]. The fishing vessel
became a remote sensor arrangement when distant temperature data were
circulated over an associated platform and base station. As a result of the
presence of this remote sensor arrangement, the temperature was constantly
monitored, and the temperature data was sent over the Intemet.
One of the components of emotional support network was an overview
of a remote monitoring structure. The remote monitoring framework could
aid the unmanned evening shifts of extended manufacturing, which are com-
mon in medium-sized Japanese enterprises. 3D layered information viewing
environment (LIVE) and an array adjusting framework based on remote
monitoring were developed using open-source technologies, tested, and
debated [23].
Centralized database management system (DBMS) collected real-time
meteorological data as well as numerous recurrent functional measurements
in intemational RES factories (DBMS). In order to calculate each plant
and govern database activities, a properly created graphic representation of
statistical and functional parameters was used.
12 An Analytical Framework for the Industrial Internet of Things (IIoT)

According to Shitharth et al. [24], wireless sensor networks (WSNs)

needed to have stronger signals in order to get near-ideal answers for the
location of sensors in terms of limiting the number of jumps between each
hub of the objective district and the sink hub while covering as much of
the intrigue zone as could reasonably be anticipated (2005). By reducing the
number of jumps, the energy saving was also advanced. In the end, this work
demonstrated the adaptability of this system to focus on the impact of many
aspects of enthusiasm at the planning phase of the system, prior to the actual
organization of wireless sensor network hubs.
In the first phase, this comprehensive research process encompasses var-
ious stages, including the formulation of inquiries, the development of study
designs, the collection of data, and the interpretation of findings [25]. During
the second phase, the decline continued. By selecting the working conditions
between segments, it was possible to construct minuscule decreased subsets
of parts, which provided comprehensive information about the status of the
overall framework. To maintain the framework’s regular by and large graphic
structure, the second step decided how to present the links between sections
in a reduced context.
Sensors, processors (MSP430), wireless radio interfaces (CC2650), and
multisource power support are the fundamental components of I3Mote, an
open industrial hardware platform suggested by Rehman et al. [16] (battery,
solar, thermoelectric, etc.). I3Mote supplies all of the sensors and connectivity
components necessary for the IIoT, ultimately leading to the finished product.
The I3Mote platform also provides a set of software packages for the quick
creation of industrial applications.
The use of software tools can help rapid application development.
Because it has two distinct CPUs for communication and apps, I3Mote
provides a unique potential for simple app creation (CC2650; MSP432).
If you are looking for an industrial automation solution, I3Mote has you
covered.
Mitchell and Jones [26] have presented an LTE-based DR-PSLTE
(a threelayer architecture). Unmanned aerial vehicles such as cloudlets,
software-defined networks, and radio access networks are only a few of
the new technologies included in the proposed design to improve disaster
resilience and minimize communication delays. The three strata of the pro-
posed architecture are network synchronization and control signals, as well as
resources, which are managed by the SDN controller in the first layer. Using a
UAV, the second layer serves as a cloudlet. In disaster or emergency scenarios,
 1.3 Enabling Technologies for IIoT 13

these unmanned aerial vehicles (UAVs) perform two essential services: data
processing and data communication.

1.3 Enabling Technologies for IIoT

The Internet of Things (IoT) facilitates real-time data collection and the
initiation of actions in the context of a connected factory. These gadgets
are the backbone of IIoT’s global factory asset tracking. Intemet of Things
devices monitor the entire production process, from the raw materials to the
finished goods, to reduce labor costs and increase the efficiency of manual
system management. An industrial Internet of Things (IIoT) system that uses
Intemet of Things devices is integrated into every facet of a manufacturing
facility, from warehouses to assembly lines to distribution centers. In order to
set up, deploy, monitor, and maintain these devices, you will need a team of
highly skilled technicians.

1.3.1 Blockchain technology

When realizing the IIoT vision, one of the most critical technologies is
the blockchain [2]. Blockchain technology is the subject of extensive aca-
demic and industrial investigation in various domains, including banking,
healthcare, supply chain management, and automobile insurance. The smart
industry’s use of IoTenabled devices generates massive amounts of data.
These Intemet of Things devices have a wide variety of applications,
including but not limited to performance monitoring, anomaly detection
and diagnosis, predictive maintenance, asset monitoring, and tracking the
entire product lifecycle from raw materials to finished goods. Other possi-
ble applications include: It is a challenging task, however, to communicate
this information securely with all of the participants in the IIoT. For IIoT,
blockchain technology has many advantages, including distributed nature,
data provenance, tamper-resistance, trust, and trustworthiness. IoT device
firmware upgrades and access control have recently been implemented using
blockchain technology.

1.3.2 Cloud computing

Massive amounts of data generated by IIoT require distributed high-
performance computing systems across a wide area for management, pro-
cessing, analysis, and storage. These systems must be able to handle high
14 An Analytical Framework for the Industrial Internet of Things (IIoT)

volumes of data. The capabilities of storage, computing, and networking are

made available to all participating facilities in an IIOT system through the uti-
lization of cloud computing technologies. Backend clouds and linked devices
communicate directly with one another without the use of any intermediaries.
Depending on the organization’s requirements, the Intemet of Things can be
implemented using private, public, or hybrid cloud service models (whereby
a mix of both service models is used). Private cloud service models are out
of reach for new entrants and small- and medium-sized businesses due to the
significant costs associated with setting up data centers and hiring technical
employees. As a result, major and well-established intemational corporations
choose the construction of a private cloud in order to protect their data from
industrial espionage.

1.3.3 Big data analytics

Data streams generated by IoT devices and systems are so large that high-
end, high-performance computing platforms are needed to handle them. Due
to latency and real-time considerations, big data processing and analysis in
IIoT systems may be troublesome. Big data collection, storage, management,
and processing, as well as the capacity to analyze and act on that data, are
all supported by IIoT systems. Onboard data collectors, web-enabled data
sources, and humane machine movements are ways data can be collected
in an industrial Internet of Things (IIoT) system. Similar to how big data
storage can benefit from on-premises, networked, and remote cloud storage,
a cloud environment can reap benefits from all three types of cloud storage.
New data management and processing methods have been developed to cope
with the massive amounts of data that must be dealt with. At various levels
within an IIoT system, various data mining, machine leaming, deep leaming,
and statistical data analysis methods can be utilized. Actuation technology
allows devices that are part of an IIoT network to communicate with the
environments in which they are located. Despite the complexity of these
systems, the next generation of IIoT systems places a significant emphasis
on the processing and analysis of large amounts of data.

1.3.4 Artificial intelligence and cyber-physical systems

Humane interventions are minimized, and system efficiency is increased by
using AI technology to ensure that IIoT systems operate autonomously and
intelligently on their own. The IIoT is self-aware, thanks to AI technologies
 1.4 Framework and Case Studies 15

like multi-agent systems and conversational artificial intelligence. IoT sys-

tems use a wide range of search and optimization algorithms, from sensors
to devices to edge servers and cloud data centers. Various industrial robots
and manufacturing systems can benefit from IoT technology by reducing the
need for human intervention and effort. In order for sensors and actuators
to be used in industrial settings, CPS relies on embedded Internet of Things
devices. These onboard integrated IoT devices also allow autonomous oper-
ations and better efficiency in IIoT systems. For example, in CPS and IIoT,
efficiencies might range from operational to overall system efficiency.

1.3.5 Augmented and virtual reality

Workers in the industrial sector can benefit from augmented reality (AR) tech-
nology, which assists them in doing complex tasks, including assembling and
dismantling machinery, sophisticated industrial products, and mission-critical
systems. While working, people and machines can be monitored using AR
technology to make changes or notifications quickly to avoid mishaps. Visual-
izing an industrial function and module configurations and re-configurations
using VR technologies is possible before they are implemented in IIoT
systems. Virtual reality can shorten the time it takes to reconfigure and shut
down industrial machinery and plants. Open standards are used to construct
VR simulations for CPS and IoT systems with heterogeneity.

1.4 Framework and Case Studies

BDA processes increasingly serve as the primary engine for delivering value
for customers and businesses [8]. Similarities can be drawn between this new
technology and the Internet of Things (IoT). BDA processes make combining
consumer and company data feasible for highly customized production with
zero faults. The Internet of Things is all about integrating historical and real-
time data at various levels (IoT). This multi-source data integration can help
design new business models more successfully. In addition to developing
product offerings and value capture methods, value networks, and value
communication strategies for intemal and extemal stakeholders, businesses
address a wide range of industrywide value creation processes [18]. In an
ideal world, BDA processes would make it easier for companies to create
value at the corporate level by enhancing internal operations. Outbound
intelligence, on the other hand, leads to consumer value development. Despite
16 An Analytical Framework for the Industrial Internet of Things (IIoT)

these possibilities, realizing the full potential of BDA technologies is difficult.

Few frameworks and use cases like this can be found in the existing literature.

1.4.1 SnappyData
Apache Spark and GemFire are integrated with SnappyData, an open-source
BDA framework. Data processing with Apache Spark is used for large
datasets, while GemFire handles transactional data storage. SnappyData’s
unified BDA engine is unmatched in the industry when it comes to online
transaction processing, analytical processing, and operational data streaming
analytics. Because of this, SnappyData’s real-time interactive visualization
speed suffers when dealing with large amounts of streaming data.

1.4.2 Fault detection classification

Industrial cyber manufacturing systems are built on a foundation of numerous
geographically dispersed yet interconnected production facilities [6, 7]. The
framework for fault detection and classification, also known as FDC, is what
is responsible for finding flaws in the product during the manufacturing
process. The foundation of the FDC architecture is the integration of IoT
devices into CPS and cloud computing. Io T devices in manufacturing plants
continuously gather and analyze data streams to detect various signals. This
information is then transmitted to cloud servers. These cloud servers use
deep belief networks and deep leaming algorithms to discover and classify
defective products [87, 88]. An automobile headlight manufacturing facility
tested FDC, and the results were trustworthy.

1.5 Challenges in IIoT

Interoperability, privacy, scalability, heterogeneity, reliability, and resource
management are just some of the technical concerns that have been brought to
light as a result of the varied and complex structure of the IIoT system. How-
ever, there are a number of significant challenges that need to be addressed.
In the following chapter, we will talk about the challenges that we faced.

1.5.1 Schemes for efficient data storage

Data volume has skyrocketed as a result of the widespread adoption of het-
erogeneous IoT devices. There is a high transmission rate of data generated
by sensors and actuators built into industrial devices. A wide variety of IIoT
 1.5 Challenges in IIoT 17

devices, gateway/edge servers, and cloud servers are used to collect and store
the data that is generated by these heterogeneous devices. Processing and
storing sensor data is a time-consuming and challenging task. Data manage-
ment models that are effective in dealing with these problems are required.
These models should be able to deal effectively with the vast volume of raw
data created by heterogeneous IoT devices. These models should be able to
provide fast, dependable, and secure data processing, storage, retrieval, and
flow for data management services.

1.5.2 IoT systems from different vendors working together

The IIoT system suite is made up of a variety of components, including
industrial machines, robotics, Intemet of Things devices, sensors, actuators,
gateways, edge nodes, edge/cloud data servers (data centers), wired/wireless
communication networks, and cellular networks (Wi-Fi, 5G, etc.). One of the
most significant challenges posed by IIoT is figuring out how to effectively
integrate a diverse array of technologies and vendors and collaborate with
them. Integration and collaboration are made more challenging by problems
such as synchronization, sharing of resources and data, interoperability, and
data privacy. Research and development efforts still need to be expanded to
produce methods of interoperability and collaboration that are both flexible
and effective.

1.5.3 Adaptable and resilient technologies for analyzing large

datasets
To achieve the goal of the IIoT and get the most out of the massive amount
of data produced by the devices that make up the IIoT, there is a significant
need for dependable and flexible big data analytics solutions. Unfortunately,
massive datasets cannot be processed and analyzed using state-of-the-art
database management tools. Real-time processing of this data is required
because it is pivotal for a wide range of real-time industrial automation activ-
ities, including fault prediction, predictive maintenance, output improvement,
downtime reduction, and anomaly detection. Because of the volume of data
produced by IIoT devices, real-time big data analytics solutions are essential
for satisfying the various needs of IIoT applications (in terms of, for example,
data rates, latency, dependability, and so on). An enterprise-wide perspective
can be gained from processing and visualizing data gathered throughout the
entire product lifecycle (including the stages of development and production,
testing, customer feedback, and post-sale services, for example).
18 An Analytical Framework for the Industrial Internet of Things (IIoT)

1.5.4 Trust in IIoT systems

The success of any technology is strongly linked to consumer acceptance
and adaptation, which is heavily impacted by consumer trust in these tech-
nologies. The commercial clients’ (e.g., industry owners) trust in these IIoT
systems has an impact on the effective adoption of these systems. Security
and privacy have been a prominent focus of recent research on IIOT systems,
which are still in their infancy. Poor IIoT security and privacy will deter
customers from using these systems, harming IIoT adoptions in the long
run. As a result, for IIoT systems to be successfully deployed and adapted
in the industry, client trust must be appropriately dealt with. As a result,
further study into customer trust models is needed if IIoT solutions are to
gain widespread acceptance.

1.5.5 Integration of wireless technologies and protocols in the

Internet of Things (IIoT)
The Intemet of Things (IoT) is receiving an ever-increasing amount of atten-
tion from businesses and academic institutions. Communication is necessary
for the functioning of the industrial Intemet of Things. Consequently, commu-
nication in the IIoT needs to be able to link a large number of different types
of devices, have an adequate capacity for the transfer of data, and behave in a
deterministic manner while having a low amount of delay. On the other hand,
the requirements for reliability, availability, and security in specific industrial
applications are significantly more stringent than those in others. IloT uses
a wide range of communication technologies, protocols, and standards. With
wireless systems (WLAN, IEEE 802.15) and wireless devices (WSN) becom-
ing more popular, cable communication has been losing its luster. There are
also other difficulties associated with using wireless communication. Some
numerous wireless systems and protocols must work together in the IIoT.
There are a plethora of communication protocols and technologies available.
There is no single wireless technology or protocol that can meet all of the
needs of the IIoT. As a result, deciding on communication technology and
protocol might be difficult.

1.5.6 The edge of decentralization

Network, storage, and storage services must be computed at the Intemet’s
edge to accommodate the variety of data sources and the vast output of con-
tinuous data streams. Centralized cloud controllers fully choreograph edge
 1.5 Challenges in IIoT 19

services and data management at their edges, allowing end nodes to benefit
from edge computing. Not only does this raise the need for highly accessible
communication lines, it also leads to industrial systems having a solitary
point of failure. This can be addressed by utilizing edge-cloud services that
are decentralized. Using blockchain technology, end-point devices, and app
resource requirements can be permanently recorded on edge servers. Servers
can also enable decentralized resource provisioning and service orchestration
without central control mechanisms being in place first.

1.5.7 New operating systems for the Internet of Things

An operating system (OS) for the Internet of Things is developed to function
properly despite the constraints imposed by IoT devices, which can include
restrictions on memory, size, power consumption, and processor capability. In
[3], popular Intemet of Things operating systems are broken down regarding
memory use, programming language support, scheduler, and architectural
design. Researchers utilize TinvOS and Contiki as the two primary Intemet
of Things operating systems because they satisfy most of their requirements.
It is essential for an operating system designed for the Intemet of Things
to have a low memory footprint, the ability to perform real-time operations,
high energy efficiency, and independence from the underlying hardware. The
quality of service (Q0 S) standards for IIoT applications are noticeably more
stringent than those for IoT applications. Some examples of IIoT applica-
tions include the “Smart Home,” “Smart Grid,” “Smart Traffic,” and “Smart
Health.” The most critical concerns regarding the Intemet of Things are data
protection and security, dependable network connections (such as cellular and
Wi-Fi), low power consumption, interoperability, support for various devices,
and adequate bandwidth (IoT). All of the IoT OS implementations, issues,
and case studies brought up during their conversation took place within the
context of the communication technologies they had already implemented.

1.5.8 Public safety in IIoT

When dealing with an incident involving Internet of Things technology, the
general public’s safety must be the top priority. In an emergency, the fire
department, ambulance service, disaster management team, traffic police, and
other law enforcement agencies are crucial to the safety of industrial workers
and equipment. Nevertheless, one of the most significant challenges in disas-
ter zones is the inability of communication systems to function correctly or
entirely. Interoperability between the IIRI’s disparate components and older
20 An Analytical Framework for the Industrial Internet of Things (IIoT)

forms of electronic networking is also a significant issue. Unmanned aerial

vehicles (UAVs), software-defined networks (SDNs), edge computing, and
other cuttingedge communication technologies like LTE, 4G/5G, etc., were
all recommended in a recent study on the Intemet of Things and possible
future intelligent cities that prioritized public safety communications. Even
yet, disaster-resilient, selfsufficient architectures that enable high-efficiency
communication under normal circumstances would necessitate development.
As a result, these architectures are viewed as providing public safety with
disaster recovery methods in an emergency.

1.6 Application for IIoT Framework

Industrial businesses may take full advantage of the IIoT’s potential by
utilizing the framework, which outlines how the IIoT can be used to inform
product, technical, and business choices. To get a better idea of what is needed
in terms of technical and business decisions on different tiers of the IIoT, we
can follow these steps:
1. First, it is essential to establish a clear business case or need for
the benefits that the IIoT might bring to industrial processes before
implementing it.
2. Second, an evaluation of the necessary infrastructure (hardware and
software) for any IIoT implementation follows the business.
3. As soon as an IIoT solution has been assessed and understood, the user
would consider the before, during, and post-implementation considera-
tions in the industrial contexts.

1.7 Conclusion and Future Scope

With the help of the IIoT, businesses are able to collect and analyze vast
amounts of data, which can then be used, monetized, and improved upon
to offer customers brand-new services. This chapter discusses new findings
in the study of IIoT in industry. The primary concerns of this study are the
frameworks and architecture of the IIoT, the communication protocols and
data management mechanisms that are currently in use, and their poten-
tial future development. For instance, the Internet of Things can use many
enabling technologies (IoT). Problems related to IIoT implementation, such
as an extensive list of research questions, have also been uncovered. Let us
assume that the IoT in the industry will catch on and become commonplace.
 References 21

If this is the case, the IIoT will have to overcome several obstacles, such as
the need for trustworthy systems and the coexistence of wireless technology
and protocol standards. Other difficulties include the requirement for effec-
tive data management schemes and collaborations between heterogeneous
IIoT systems and robust and flexible technologies for big data analytics.
The correct responses to these issues are not out of reach at all. Some of
IIoT’s outstanding issues can be addressed with the help of this research.
Personalized manufacturing in IIoT systems will be a focus of our future
study.

References
[1] Muntjir, M., Rahul, M., Alhumyani, H. A.: An Internet of Things (IoT)
analysis: novel architectures, modern applications, security aspects and
future scope with latest case studies. Int. J. Eng. Res. Technol. 2017;
6(2), 422-427.
[2] C. K. M. Lee, S. Z. Zhang, K. K. H. Ng, Development of an industrial
internet of things suite for smart factory towards re-industrialization,
Advances in Manufacturing. 2017; 5 (4):335-343. doi:10.1007/s40436-
017-0197-2.
[3] W. Z. Khan, M. Y. Aalsalem, M. K. Khan, M. S. Hossain, M. Atiquz-
zaman, A reliable internet of things-based architecture for oil and gas
industry, in Advanced Communication Technology (ICACT), 2017 19th
International Conference on, IEEE, 2017;705-710.
[4] Kumar, Abhishek, SwarnAvinash Kumar, Vishal Dutt, Ashutosh Kumar
Dubey, and Vicente García-Díaz. “IoT-based ECG monitoring for
arrhythmia classification using Coyote Grey Wolf optimization-based
deep learning CNN classifier.” Biomedical Signal Processing and Con-
trol 76 (2022): 103638.
[5] B. Martinez, X. Vilajosana, I. Kim, J. Zhou, P. Tuset-Peiro, A.
Xhafa, D. Poissonnier, X. Lu, I3mote: An open development plat-
form for the intelligent industrial internet, Sensors. 2017; 17 (5): 986.
doi:10.3390/s17050986.
[6] Pavithra, M., R. Rajmohan, T. Ananth Kumar, S. Usharani, and P.
Manju Bala. “Investigation of Energy Optimization for Spectrum Sens-
ing in Distributed Cooperative IoT Network Using Deep Learning
Techniques.” Hybrid Intelligent Approaches for Smart Energy: Practical
Applications (2022): 107-127.
22 An Analytical Framework for the Industrial Internet of Things (IIoT)

[7] Z. Meng, [Link], C. Muvianto, J. Gray, A data-oriented m2 m messaging

mechanism for industrial IoT applications, IEEE Internet of Things
Journal 4 (1) (2017) 236{246 (2017).
[8] M. Padmaja, S. Shithart, K. Prasuna, Abhay Chaturvedi, Pravin R.
Kshirsagar, A. Vani, ’Growth of Artificial Intelligence to Challenge
Security in IoT Application,’ Wireless Personal Communications,
Springer, 2021, DOI: [Link]
[9] T. Qiu, Y. Zhang, D. Qiao, X. Zhang, M. L. Wymore, A. K. Sangaiah,
A robust time synchronization scheme for industrial internet of things,
IEEE Transactions on Industrial Informatics. 2017; 14 (8): 3570-3580.
[10] S. Katsikeas, K. Fysarakis, A. Miaoudakis, A. Van Bemten, I. Askoxy-
lakis, I. Papaefstathiou, A. Plemenos, Lightweight & secure industrial
iot communications via the mq telemetry transport protocol, in Com-
puters and Communications (ISCC), 2017 IEEE Symposium on, IEEE.
2017;1193-1200.
[11] P. Ferrari, E. Sisinni, D. Brand -ao, M. Rocha, Evaluation of com-
munication latency in industrial IoT applications, Measurement and
Networking (M&N), 2017 IEEE International Workshop on, IEEE,
2017;1-6.
[12] Shitharth, [Link] Winston, “A New Probabilistic Relevancy Classi-
fication (PRC) based Intrusion Detection System (IDS) for SCADA
network,” Journal of Electrical Engineering, Vol.16, No.3, 2016, pp.
278-288.
[13] T. P. Raptis, A. Passarella, A distributed data management scheme
for industrial iot environments, in Wireless and Mobile Computing,
Networking and Communications (WiMob), IEEE, 2017;196-203.
[14] S. Rao, R. Shorey, Efficient device-to-device association and data
aggregation in industrial IoT systems, in Communication Systems and
Networks (COMSNETS), 2017 9th International Conference on, IEEE,
2017;314-321.
[15] T. P. Raptis, A. Passarella, M. Conti, Maximizing industrial iot network
lifetime under latency constraints through edge data distribution, in 1st
IEEE International Conference on Industrial Cyber-Physical Systems,
(ICPS) (May 2018), available at [Link] iit. cnr. it/traptis/2018-raptis-
icps. pdf.
[16] M. H. ur Rehman, I. Yagoob, K. Salah, M. Imran, P. P. Jayara-
man, C. Perera, The role of big data analytics in industrial internet
of things, Future Generation Computer Systems. 2019; 99:247-259.
Doi:10.1016/[Link].2019.04.020.
 References 23

[17] D. Miller, Blockchain and the Internet of Things in the industrial sector,
IT Professional. 2018; 20 (3):15-18.
[18] F. Javed, M. K. Afzal, M. Sharif, B.-S. Kim, internet of things (IoT)
operating systems support, networking technologies, applications, and
challenges: A comparative review, IEEE Communications Surveys &
Tutorials. 2018; 20 (3): 2062-2100.
[19] M. Padmaja, S. Shitharth, K. Prasuna, Abhay Chaturvedi, Pravin R.
Kshirsagar, A. Vani, ’Grow of Artificial Intelligence to Challenge Secu-
rity in IoT Application,’ wireless personal communication, Springer,
[Link]
[20] Dawson, C.: Introduction to Research Methods: A Practical Guide for
Anyone Undertaking a Research Project, 4th ed. How to Books, London
(2009).
[21] S. Shitharth, N. Satheesh, B. Praveen Kumar, K. Sangeetha, “IDS
Detection Based on Optimization Based on WI-CS and GNN Algorithm
in SCADA Network’, Architectural Wireless Networks Solutions and
Security Issues, Lecture notes in network and systems, Springer, vol.
196, Issue 1, 2021, pp. 247-266.
[22] Schultze, U., Orlikowski, W. J.: A practice perspective on technol-
ogymediated network relations: the use of internet-based self-serve
technologies. Inf. Syst. Res. 2004 ; 15(1): 89-106.
[23] Yin, R. K.: Case Study Research: Design and Methods. Applied Social
Research Methods, 5th ed. Sage, Beverly Hills (1984).
[24] S. Shitharth, Masood Shaik, Sirajudeen, Sangeetha, ’Mining of intrusion
attack in SCADA network using clustering and genetically seeded flora
based optimal classification algorithm’, Information Security, IET, vol.
14, Issue 1, 2019, pp. 1-11
[25] Creswell, J., Plano Clark, V. L.: Designing and Conducting Mixed
Method Research, 2nd edition. Sage, Thousand Oaks (2011).
[26] Mitchell, T., Jones, S.: Leading and coordinating a multi-nurse
researcher project. Nurse Res. 2004; 12(2): 42-55.
 2
Industry Automation: The Contributions
of Artificial Intelligence (AI)

M. Nalini

Department of Electronics and Instrumentation Engineering,

Sri Sairam Engineering College, India
E-mail: [Link]@[Link].

Abstract
Artificial intelligence (AI) refers to a computer’s or a robot’s capability to
execute jobs that intelligent people ordinarily achieve. Artificial intelligence
permits machines to learn from their mistakes, adapt and react to new inputs,
and accomplish human-like tasks. It is used in a wide range of industries,
including gambling, banking, retail, commercial, and government, to name
a few. AI is used in industrial automation since it is extensively used and
is steadily becoming more common in manufacturing. AI-driven robots are
leading the road to the future by offering a host of advantages, such as new
opportunities, high manufacturing efficiency, and a good fit between machine
and human engagement. Knowledge-based, automated labor is a hallmark of
the Fourth Industrial Revolution. Many internal industrial difficulties, ranging
from a lack of expertise to sophistication in decision-making, integration of
problems, and information overload, may be solved using AI. By incorpo-
rating AI into manufacturing plants, businesses can radically transform their
operations. The importance of artificial intelligence in industrial automation
is explained in this chapter, which begins with a brief introduction, then
application landscape and production-related scenarios, and then moves on
to a classification scheme with real-time examples. This chapter also covers
the impact of artificial intelligence in industrial automation, followed by

25
26 Industry Automation: The Contributions of Artificial Intelligence (AI)

applications of artificial intelligence examples in different industry sectors,

and ends with future scope.

Keywords: artificial intelligence, industrial automation, Industry 4.0, smart

production, robotic process automation

2.1 Introduction
“Industry 4.0” has been called “a name for the current trend of automation
and data exchange in manufacturing technologies.” For this trend to reach
its full potential, it needs cyber−physical systems, the Internet of Things,
cloud computing, cognitive computing, and the creation of “smart factories.”
When people talk about Industry 4.0, they usually mean “the current trend of
automation and data exchange in manufacturing technologies,” which is what
this page says it is. By developing new ways using the latest technologies to
automate jobs, we can renovate how people and machines live, engage, and
cooperate, resulting in a superior, healthier, and more robust digital economy.
For a long time, computer visualizing has been used to uncover product flaws
in real time for quality assurance. On the other hand, manufacturing now
demands more info than ever before, as well as the fact that management does
not want to pay employees to enter data. Data collection may be rationalized
using AI and computer vision. A factory employee would be able to remove
raw materials from a rack and then have the stock transaction conducted
automatically, depending on the activity being monitored with a camera. It
will be the primary user-friendly interface, with the user concentrating on
the job instead of typing or scanning data into a computer. Second, artificial
intelligence will impact the Internet of Things (IoT). The IoT will allow
businesses to distribute products and services to customers who may not
realize they require them. IoT may also provide extensive data to producers
and retailers, allowing them to assess quality and identify variables that
could create difficulties [1]. In a word, the IoT is an incoming data stream
that artificial intelligence (AI) may use to reason and grow. This will make
the enhanced generative design process much easier, allowing for the more
evolutionary reimagining of products.
From 2021 to 2028, the global market for AI is projected to expand at
a CAGR of 40.2%, from USD 62.35 billion in 2021 to USD 62.35 billion
in 2028 (Figure 2.1). Industry after industry, from healthcare to automotive
to retail to finance to manufacturing, is adopting cutting-edge technology,
thanks to the relentless efforts of the tech giants in this field. On the other
 2.1 Introduction 27

Figure 2.1 Market value of AI.

hand, technology has continuously played a vital role in these organizations,

although AI technology has transported it to the fore. Artificial intelligence
is now being integrated into nearly every tool and program, from self-driving
cars to medical devices that save lives. Artificial intelligence has proven to be
a massive game-changer in the past, and it will only get bigger and better in
the coming digital age.
Artificial intelligence technology usage is increasing in the manufacturing
sector to provide new possibilities and improve functional competencies by
utilizing innovative technologies, speeding up methods, and creating busi-
nesses more adaptive to future changes. The impact of AI in industries will
make the following changes [2].
• Formal assertions are inferred from logical statements, and systems to
verify the correctness of hardware and software are built using deductive
methods and machine-based proofs.
• Knowledge-based systems include approaches for modeling and gath-
ering knowledge and software that can mimic human competency and
assist specialists.
• Inductive computational methods, in general, and machine learning, in
particular, are used to recognize and analyze patterns.
• Robotics: the uncontrolled operation of robotic equipment, such as self-
driving cars.
• Intelligent multimodal human−machine interaction: words (in combi-
nation with linguistics), visuals, gestures, and other human contact are
analyzed and “understood.”
28 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.2 Impacts of AI on the industry.

Machine learning is the most widely utilized AI technique across sectors.

This is unquestionably owing to its broad application, which makes it applica-
ble to many use cases across the value chain represented in Figure 2.2. While
companies have typically depended on internal information for supervised
machine learning, many more are increasingly looking at the possibility of
combining internal and external datasets, as well as internal and external data
science skills, to provide even more insightful findings.

Figure 2.3 Percentages of AI technology breakdown.

 2.2 Automation Systems Potential 29

Knowledge-based representations, communication via natural languages,

the deployment of tactics, the appraisal of irregularities, and action are all
characteristics of artificial intelligence. AI systems can detect and react to
threats, as well as support decision-making, math-oriented intelligence, and
cognitive research. The most popular sort of machine learning is supervised
learning, which involves feeding software with organized data to detect
patterns and comprehend and understand new findings. Machine learning
and smart robotics have been determined to be the most beneficial in using
AI in businesses. The McKinsey Global Institute surveyed 3000 executives
of various firms in 2017; based on this survey, the technological breakdown
between various industries illustrates in Figure 2.3.

2.2 Automation Systems Potential

Bulk production is the way of producing many identical things using man-
ufacturing lines or automation technologies. Mass manufacturing allows for
the efficient fabrication of many comparable things. Client-centric produc-
tion, in which the customer is a co-designer of the product, has supplanted the
traditional mass production approach [4]. This can be enabled using artificial
intelligence technologies, as shown in Figure 2.4.
Figure 2.4, left side, shows the automation and manufacturing plan
designed to service mass manufacturing at the lowest possible cost. The
production process is stiff after it has been created and does not allow for flex-
ibility. A paradigm shift towards personalization is shown in the middle, with
the consumer becoming a co-designer of his product. Digital twins, machines,
and services are illustrated on the right side, allowing new service-oriented
cost models. Artificial intelligence, which integrates current infrastructures,
digital twins, and more data resources, enables computers to independently

Figure 2.4 Customer-driven production.

30 Industry Automation: The Contributions of Artificial Intelligence (AI)

create sophisticated goods, manufacturing processes, and lines. Digital twins

refer to both existing and non-existing (historical) materials. This strat-
egy defies existing assumptions and creates incredible opportunities for
innovative ideas to be accepted in industries around the world [4].

2.3 Application Landscape and Production-related

Scenarios
The AI Map (Platform Learned System, 2021), created by the Federal Min-
istries of Education and Research (BMBF) and Economic Affairs and Energy
(BMWi), provides a snapshot of AI application areas, research institutes, and
transfer initiatives in Germany. The properties of AI applications may be
mapped into the “orange dotted box,” as shown in Figure 2.5; it covers the
functioning of machines and systems if we use the Reference Architectural
Model Industry4.0 (RAMI4.0) model, which is the reference shown in Fig-
ure 2.5. RAMI4.0 is a three-dimensional plan that depicts the key elements
of Industry 4.0 (I4.0). Processes are optimized using data. This comes after
the initial wave of digitizing industries by integrating sensors into existing
machinery, often known as “brownfield” optimization [5]. The function of AI
in the RAMI4.0 model’s left half has yet to be researched, but the concurrent
growth of digital duplication and wireless communication will power it. The
“blue dotted box” in Figure 2.5 concentrates on the technical side of AI to
allow new use cases in networked manufacturing and the association of AI
services based on trustworthy infrastructures. RAMI4.0 model consists of
three axes [7], which are explained here.

Figure 2.5 RAMI4.0 model.

 2.3 Application Landscape and Production-related Scenarios 31

“Hierarchy levels” axis

The hierarchical levels from IEC 62264, the worldwide regulations on inte-
grating enterprise IT, and control systems are shown on the right horizontal
axis. These hierarchy stages represent the different functionalities inside the
factory or plant.
“Life cycle and value stream” axis
The lifespan of plants and goods is represented on a left horizontal axis. The
IEC 62890 standard on monitoring and management serves as the foundation
for this. There is also a difference made between type and instance. Whenever
the development and prototype production phases are completed, and the
actual product is built in the production department, a “type” becomes an
“instance.”
“Layers” axis
The IT representation, i.e., the digital picture of a machine, is defined in an
organized method layer by layer using the six layers on the y-axis of the
model. The characterization in layers arrives from information and commu-
nication technology. It is typical practice in this industry to stack sophisticated
items.
The three axes reflect all of the critical elements of Industry 4.0. They
categorize an item in the model, such as a machine. RAMI 4.0 may therefore
be used to express and implement highly flexible Industry 4.0 principles. The
Reference Architecture Model enables a gradual transition from the current
state to the world of Industry 4.0 [6].

2.3.1 Autonomy-level classification of industrial AI applications

In 2018, the Society of Automotive Engineers (SAE) investigated and issued
an autonomy for self-driving cars. According to that standard, the acceptable
phrases are driving automation or driving automation systems. The employ-
ment of electrical or mechanical equipment to replace human work is known
as automation. The amount of human effort required is significant. As a result,
automation is being pushed. The SAE standard is particularly important since
it defines the SAE levels or levels of driving automation. Figure 2.6 illustrates
these five stages and shows the role of artificial intelligence in achieving
self-driving cars.
32 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.6 SAE standard level of automation for autonomous car.

Level 1: Assistance to the driver

A cruise controller, anti-lock braking system or anti-skid breaking system
(ABS), traction control system (TCS), and electronic stability control are all
parts of the ADAS (advanced driver aid system). The driver must maintain
complete control when receiving help from the ADAS.
Level 2: Partially automated
The driver must remain engaged while driving, but the ADAS provides even
more help, such as simultaneously regulating accelerating, decelerating, and
turning. A backup camera is an example of this. Tesla’s complete self-driving
is also an example of this level.
Level 3: Automation based on condition
The ADS (automated driving system) includes environmental detecting fea-
tures such as pedestrian safety, lane-changing aid, and junction assistance,
allowing the automobile to respond to its surroundings independently. On the
other hand, the human driver must be alert and ready to interfere and take
control if necessary. Autopilot is yet another instance of a Level 3 automation
categorization.
Level 4: High-level automation
In some conditions, such as within a geo-fenced region, the ADS can operate
and perform all driving operations on its own, including monitoring its
environment. The human driver may turn their focus away from the functions,
but they retain the power to override if necessary.
 2.4 Impact of Artificial Intelligence in Industry4.0 (I4.0) 33

Level 5: Totally automated

Full automation drives the future and has been promised for decades in
futuristic blockbusters like Minority Report, Knight Rider, and Time Cop.
These cars will not require a driver; drivers may become passengers since
they can do business, work on their newest hobby, or snooze while being
transported using a Level 5 autonomous vehicle.

2.4 Impact of Artificial Intelligence in Industry4.0 (I4.0)

The German Industry 4.0 Platform (PI4.0) has released 11+ key industry
4.0 implementation and technology scenarios for several industrial value
streams. Expert groups from DIN/DKE-SVCI4.0 and IEC-TC65 examined
over 50 AI use cases in about various parts of the industrial internet reference
architecture (IIRA). According to the data, there is a shortage in engineering.
The AI capability in particular situations is presented below, using auton-
omy levels as a reference (AL). Six different application situations and a
novel technological scenario will be covered.

2.4.1 Order-controlled production (OCP)

At its core, the “order-controlled production” scenario is concerned with
how a company could automatically order the integration of other firms’
assets and capabilities into its own manufacturing to generate customized
items across manufacturing lines. This scenario is depicted in Figure 2.7. The
contributions of artificial intelligence, in this case, are as follows:
• The investigation and selection of possible suppliers as well as their roles
in the supply chain to determine the much more cost-effective offer that
fits the requirements.
• Automated discovery and negotiation with potential manufacturing-
skilled suppliers.
Manufacturers can include external manufacturing modules mostly auto-
mated with order-controlled production. They extend their manufacturing
capabilities and capacities in response to demand on a case-by-case basis.
This procedure does not necessitate any financial investment. This allows
businesses to respond quickly to changing market conditions and client needs.
On the other hand, companies that put their skills and capacities on the market
can maximize the use of their production capacity. When adopting flexible
order-controlled manufacturing, new business models are also feasible. This
34 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.7 The scenario of order-controlled production.

is strongly tied to the transition to the so-called platform economy, where

services are supplied via digital platforms in the B2B sector, as stated in the
“value-based services” scenario.

2.4.2 Smart production (SP2)

Using smart production, smart products can be developed. Future product
digital twins, also with sensor-equipped mechanisms, offer innovative types
of engineering cooperation as well as the automation of engineering tasks.
During the production process, digital twins, as well as sensor information,
are given. The digital twin will be customized for each process stage. Dif-
ferent digital clones may exist for the same product, each reacting to client
needs. This scenario is depicted in Figure 2.8. Smart goods are one of the
most important parts of the next industry transformation, which will impact
both products and processes. This section examines the recent developments
in the industrial environment as well as the improvements in product creation
procedures. Furthermore, with the introduction of novel technology that
includes powerful computer platforms, Industry 4.0 (I4.0) is creating new
options for SP2 [7].
The contributions of artificial intelligence, in this case, are as follows:
• Artificial intelligence provides a complete platform for its full lifespan.
• Simulation and modeling tools with artificial intelligence.
 2.4 Impact of Artificial Intelligence in Industry4.0 (I4.0) 35

Figure 2.8 The scenario of smart production.

2.4.3 Innovative product development (IPD)

Methods for creative product creation based on intelligent networking and
collaboration amongst various players are at the heart of this application
scenario. This is made feasible through Internet-based cooperation, which
allows people to exchange tools, services, and information. Figure 2.9 depicts
this situation, in which innovative product development, by speeding up
the digital product development process, reduces the product development
lifespan from discovery to launch. The product development process allows
for speedy prototyping and regular feedback when joined with perfectly
matched digital tools. Customer data and feedback are regularly examined
as part of the innovation generation process. Rather than segregating devel-
opment and user experience design, a multi-discipline team works to address
problems with the current products depending on customer input or problem
statements.
IPD has several advantages:
1. Increase the speed with which digital things are created.
2. Rapid prototyping helps you to test your business concepts quickly.
3. Use APIs and low-code environments to try out different features or
services easily.
4. Examine existing products for ways to improve them.
5. Product development innovation could also be used to enhance an
existing product line.
6. To fulfill the expectations of its consumers, a successful product should
evolve over time.
36 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.9 The scenario of innovative product development.

7. With such a shorter development cycle, such as that utilized in the

innovative product development process, it is feasible to adapt quickly
to client requests.
Streamline product/service development to save money. Reduce service
and product development costs by merging interdisciplinary collaboration
into a more practical approach to creative product creation.
The contributions of artificial intelligence, in this case, are as follows:
• In the value chain, the entire life cycle is covered.
• In the virtual world, customer centric product design is covered.

2.4.4 Seamless and dynamic engineering of plants (SDP)

This setup explains in what way such an integrating plant model is formed and
utilized, starting with an early engineering procedure for establishing a sys-
tem, and how it is sustained and maintained constantly during the entire life
cycle of the realized plant interconnected processes, starting from engineer-
ing, process, and service. Boundary circumstances, background information,
feasible plant variations, believable and actual engineering conclusions, as
well as the prospective and real impacts of such decisions are all included in
this model [8] (Technical Committee 6.12, VDI-GMA, 2018). IPD and SP2
are concerned with the product, whereas SDP is concerned with the plant,
 2.4 Impact of Artificial Intelligence in Industry4.0 (I4.0) 37

Figure 2.10 The scenario of seamless and dynamic engineering of plants usage.

which is usually associated with a specific location. This scenario is shown in

Figure 2.10. The following positions generally stand out.
• The model object template provider is in charge of building a model
object template library.
• The model object user is in charge of the entire collection of model
objects.
• A physical object maker can make a physical thing from one or more
model objects.
The contributions of artificial intelligence, in this case, are:
• assisting with the system engineering and model generation processes;
• monitoring the data with additional context data to make the perfect
engineering choices throughout the entire life cycle of the plant.

2.4.5 Circular economy (CRE)

One of the three pillars upon which the circular economy rests is that process
design should be prioritized. Waste and effluent must be eliminated, and
recycling products and materials must be a top priority (at their highest
value). The world’s natural environments would be revitalized. The “circular
economy” application scenario shows how industrial procedures can take
38 Industry Automation: The Contributions of Artificial Intelligence (AI)

cues from biological processes to make better use of a given resource’s

capacity and produce less waste, thereby enabling the “cradle-to-grave”
reusing, repurposing, and recycling of items. Figure 2.11 illustrates this
situation. They are reducing environmental impact while considering life
cycle analysis. Instead of discarding things and relying on the discovery of
new resources, a circular economy will have marketplaces that encourage
people to repurpose existing items. Used items such as clothing, metal scraps,
and outdated gadgets can all find new homes in this system or at least find
new uses. It is possible that this will aid in environmental protection as
well as the management of natural resources, the development of innovative
economic ventures, the fostering of fresh talent, and the provision of new
job opportunities. To facilitate this, we must shift to using renewable sources
of power and building supplies. In a circular economy, the production and
consumption of goods and services are no longer intrinsically linked. It is a
trusted method that helps businesses, people, and the planet.
The contribution of artificial intelligence, in this case, is:
• Life cycle assessment preparation and validation: Designing goods and
systems such that components may be reused, hence reducing the
production’s “green footprint.”
• Assisting in the categorization procedure: AI aids in the prediction of
the state of function (SoF) and state of health (SoH).

Figure 2.11 The scenario of circular economy.

 2.4 Impact of Artificial Intelligence in Industry4.0 (I4.0) 39

2.4.6 5G for digital factories − mobile controlled production

(MCP)
It was the first technological scenario of 2019 that has been released. By
preventing the need for cables using smart technologies like cloud-based
services and wireless communication, innovative novel design solutions are
developed for the manufacturing and logistics industry. Five use case fam-
ilies have been identified by the EFFRA (European Factories of the Future
Research Association) research and white paper [1] on 5G, each representing
a different subset of rigorous needs within the supply chain and industrial
networks, as shown in Figure 2.12.
UC1: Optimizing time-critical processes inside the factory to achieve
zero-defect production, increased efficiency, worker satisfaction, and safety
through 3D scanning, wearables, and collaborative robots in closed-loop
control systems. This set of use cases necessitates a latency of less than 1 ms.
Different sorts of network characteristics (quality of service criteria) and
solutions may be required by various players in the value chain.
The contributions of artificial intelligence, in this case, are:
• Finding the best network-slicing technique to assist the manufacturing
process and system engineering.
• Private data connection with high processing speed.
• Home and business broadband.

Figure 2.12 The scenario of different actors in the value chain.

40 Industry Automation: The Contributions of Artificial Intelligence (AI)

• Smart antenna design and activation for internal application optimiza-

tion (ETSI, 2020).
• Self-healing as well as forecasting.
• Network resource allocation and management.
• Internet Protocol version 6 (IPv6) has intelligent security, and so on.

2.5 Industry Use Cases for AI-enabled Collaboration

Artificial intelligence can take industrial automation to the next new level
of automation by allowing for automated combined engineering with many
stakeholders across the value chain in various sectors. New service-based
business models, including the value chain and manufacturing ecosystem
partners, are emerging. Even though a significant number of companies have
implemented AI to automate mundane tasks, those companies that use AI to
supplant human workers will, at best, see only temporary increases in output.
Humans and artificial intelligence can improve upon each other’s strengths by
combining their respective intelligence. These strengths include leadership,
teamwork, creativity, social skills, scalability, and quantitative abilities. What
comes naturally to humans, such as telling a story, may be impossible for
machines to imitate. On the other hand, what machines do automatically, such
as analyzing gigabytes of data, may be impossible for humans. Both of these
abilities are necessary for sustained success in the business world.
Artificial intelligence’s effect is becoming more prominent with each
advancement. Almost all the industries, like banking and finance, retail,
healthcare, automobiles, telecommunications, manufacturing, defense and
military, entertainment and media, and education, are being transformed by
technology. Machine learning, natural language processing, data analytics,
and image analytics are just a few of the artificial intelligence sub-domains
that are proving to be successful in various industries. Furthermore, by
exploiting end-to-end automated processes, artificial intelligence serves a
commercial function. AI may help us improve our analytical and decision-
making skills, as well as our creativity.

2.5.1 Artificial intelligence in healthcare industry

[Link] The use of predictive analytics to confirm the need for
surgery
Healthcare has been given a gift in the form of predictive analytics. We
occasionally encounter patients who claim they had unneeded surgery due
 2.5 Industry Use Cases for AI-enabled Collaboration 41

Figure 2.13 Predictive analytics in healthcare: use cases.

to a lack of forewarning about what was to follow. Artificial intelligence,

thankfully, is altering the fate of such hefty risks and unnecessary operations.
Healthcare providers may determine whether or not a patient requires surgery
using artificial intelligence and predictive analytics. The technology will aid
doctors in determining whether the procedure is indeed essential or whether
there is a less risky option [9].
In the ICU and general ward, predictive analytics may assist in noticing
early indicators of patient weakening and identifying at-risk patients at home
to lessen hospital readmission rates and avoid downtime for healthcare equip-
ment. Some of the notable areas of healthcare predictive analytics at different
stages of the patient journey utilize AI, as shown in Figure 2.13.
Diagnosis: Patients who are diagnosed early can begin treatment immedi-
ately, boosting their chances of survival; this prediction is a valuable tool.
For example, diagnoses of malignant mesothelioma were predicted using a
patient group’s predictive analytics.
Prognosis: The researchers employed predictive analytics to find which
patients, for example, suffering from congestive heart failure (CHF), had
the maximum chances of being readmitted after a hospital stay based on
physiological data. Doctors may utilize this knowledge to undertake therapies
early to avoid the expected readmissions.
Therapy: To identify the most successful treatment method for patients with
prolonged pain, clinicians used machine-learning-based predictive analytic
models.
42 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.14 Example of surgical robot.

[Link] Intelligent surgical robots

Robot surgeons are necessary in the twenty-first century due to the growing
worldwide population and need for doctors. Aside from standing in for human
surgeons, robot surgeons may often outperform them. Surgery operations
need extreme patience and accuracy, and medical surgeons’ abilities do not
decrease even when they operate nonstop for hours on end [8]. As a result,
robotic assistance in operations can assist surgeons in achieving a new degree
of precision, even for the tiniest movements.
Surgical robots can regulate their movements’ direction, depth, and speed
with extreme precision. Because they do not grow weary, they are especially
well-suited to activities that need the same repetitive movements. Robots can
also stay in one place for a long time and go places while traditional tools
cannot do this.

2.5.2 Artificial intelligence in manufacturing and factories

[Link] Analytical services for advanced data
The industrial industry has had great success using AI for sophisticated
data analytics. Digital transformation has resulted in an abundance of large-
scale real-time datasets that may be leveraged to get in-depth insights and
 2.5 Industry Use Cases for AI-enabled Collaboration 43

forecast current market trends. The mix of data and sophisticated analytics
has aided risk management, data visualization, supply chain management,
and the efficient and effective decision-making.

[Link] Predictive maintenance

Maintenance planned ahead of time is called predictive maintenance, or
PdM, and uses real-time data to identify critical faults in the manufacturing
process so that remedial measures may be taken quickly. It assists in data
analysis by examining differences in nature and frequency, and it notifies
the system to lessen the chance of failure. Predictive maintenance extends
a product’s life, as shown in the graph in Figure 2.15. While unplanned
maintenance and preventive maintenance have a tradeoff, predictive main-
tenance (PdM) is an auspicious method that can break the exchange by
optimizing component useful life and uptime simultaneously. Its purpose is
to keep track of the status of in-service equipment and anticipate when it may
break [10].
It is a sign that the plant’s equipment’s future behavior and condition can
be predicted, allowing for more effective maintenance (e.g., predictive health
monitoring). This allows for the minimum downtime and maintenance costs
for machines possible [15].

Figure 2.15 Graphical representation – effect of predictive maintenance.

44 Industry Automation: The Contributions of Artificial Intelligence (AI)

[Link] Automation of robotic processes

RPA software, or robotic process automation, can successfully handle the
organization’s backend functions without the need for human interaction.
It allows employees to concentrate on other tasks, resulting in increased
productivity. RPA automates high-volume, repetitive processes requiring
complicated computations and maintains correct records. RPA software may
be implemented in various production processes to assist in cutting time and
enhance workflow compared to rivals, as shown in Figure 2.16. By adopt-
ing RPA, businesses will increase their profits, agility, and responsiveness.
The software can remove time-consuming tasks from their daily routine to
improve employee happiness, engagement, and productivity. Unlike other
digital transformation methods, RPA does not have to slow down day-to-day
operations to speed up the process. Working with antiquated systems that
lack modern conveniences like application programming interfaces, virtual
desktop infrastructures, and database access is also helpful in automating
tasks.

The advantages of RPA

• RPA does not always require developer setup; less coding is required.
Drag-and-drop features in user interfaces make it easier for people
without technical knowledge to get started.
• RPA’s ability to reduce team effort results in immediate cost savings.
Because of this, personnel can be reallocated to perform other, more
critical tasks that call for human judgment, boosting output and ROI.
• Increased happiness among the company’s clientele: Customers are
more likely to be satisfied with a company after using a bot or chatbot
because they do not have to wait much long for help to become available.
• Increased happiness among workers: By automating routine, high-
volume tasks, robots like RPA allow workers to put their minds to more

Figure 2.16 The scenario of robotic process automation.

 2.5 Industry Use Cases for AI-enabled Collaboration 45

strategic, creative problems. As a result, morale among workers rises.

Workers are happier now due to the shift in the workplace.
• Robotic process automation (RPA) bots can be set up to follow a set of
predetermined rules and procedures. Therefore, the possibility of human
error is eliminated entirely, which is especially helpful in areas like
regulatory compliance, where accuracy and conformity are fundamental.
An audit trail can be generated by RPA software, making it easier to
track development and easing the way for quicker problem-solving.
• The regular operation of preexisting systems will be preserved, as the
software used in robotic process automation has no effect on the under-
lying infrastructure. This is because bots function at the presentation
layer of the currently deployed applications. If you lack an API or the
personnel to create complicated integrations, you may want to think
about using bots instead.

2.5.3 Artificial intelligence in automobile

[Link] The use of artificial intelligence to improve design
Artificial intelligence (AI), programmable shading, and real-time ray tracing
are all employed extensively to modify the product’s traditional design pro-
cess. The disruption of advanced AI has resulted in an advanced ecosystem
that speeds up new design workflows and increases team communication. It
is stated that AI algorithms will be the future of automotive design, as they
can develop infinite viable techniques by specifying product concepts and the
problem [15].

[Link] AI application in manufacturing

Companies employ AI-based robots and a human workforce to complete
industrial and supply chain operations. In manufacturing, AI-powered robots
have demonstrated outcomes in material handling, test performance, and
packaging final goods. Machine learning, automation, advanced and pre-
dictive analytics, and the Internet of Things (IoT) are all important Fourth
Industrial Revolution (I4.0) technologies that manufacturers may use to
monitor their facilities in real time (Internet of Things).
This supports to accumulate enormous quantities of operational data to:
• preserve track of vital performance measures like overall equipment
effectiveness (OEE), production, and scrap rates;
• predict date of delivery and pickup accurately to prevent missing
deadlines;
46 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.17 Benefits of AI in manufacturing industries.

• predict future supply chain interruptions;

• troubleshoot any bottlenecks in the production process;
• identify and regulate equipment inefficiencies whenever they arise.

[Link] Examples of AI in manufacturing − inspiring changes

AI is revolutionizing the production process in ground-breaking ways.
Machine learning is being utilized by the Danone Group, a French food
company, to improve its demand forecasting accuracy. As a result, we now
have the following:
• Errors in forecasting have decreased by 20%.
• Lost sales have decreased by 30%.
• Demand planners’ workload has been reduced by 50%.
 2.5 Industry Use Cases for AI-enabled Collaboration 47

• Predictive maintenance (PdM) has decreased unplanned downtime of

equipment.
• Using advanced manufacturing technologies (3D printers, robots, etc.)
has reduced labor costs and flexibility in the face of supply chain
inferences.
• The finest, AI-enabled generative design has been created to ensure
efficiency and reduce waste.
Fanuc, a Japanese company specializing in automation, uses robot labor to
keep its operations going (24 × 7) around the clock every day of the week.
Robots can construct risky components for CNCs and motors, run all factory
floor gear continuously, and monitor all processes nonstop. BMW Group uses
automatic image recognition for quality control, inspections, and eliminating
false problems (deviations from target despite no actual faults). As a result,
they have attained exceptional stages of production accuracy [11].

2.5.4 Application of artificial intelligence in quality control

Quality control, which includes checking painted automobile bodywork, is
aided by artificial intelligence. When done by humans, such delicate detec-
tions are prone to mistakes. Artificial-intelligence-enabled computers can
spot flaws more quickly and precisely than people. Machine learning (ML)
quality inspection is expected to replace present optical crack detection.
Manufacturers are benefitting from the arrangement of smart cameras with
AI-enabled software to achieve higher quality checks at speeds, latency, and
prices that are beyond the capability of human supervisors. And, considering
an example of COVID-19’s social distance criteria, the emergence of such AI-
enabled smart camera technology is fortunate. Machine vision has, of course,
been used in quality governor applications for many years. However, incor-
porating quality control software powered by deep learning is a change from
earlier machine vision technology [12]. The adoption of AI in conjunction
with test automation allows for a better understanding of the current status of
QE and aids in the actions shown in Figure 2.18.
Test cases can be used again
AI-based test automation solutions simplify the complex process of creating
and reusing well-written test cases in the future. An AI-based testing tool
inspects the application and captures critical data through images to evaluate
load time, assess fundamental UI aspects, and create appealing test case
suites.
48 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.18 Features of AI-powered quality assurance.

Gathering intelligent requirements

Several manual operations, including reporting, controlling the effect of
change, and tracking approvals, influence project deliverables during require-
ments collecting. Artificial-intelligence-assisted requirement collecting rec-
ognizes the constraints of complex systems and maintains critical indicators
and KPIs that may be used to identify quality and performance concerns
ahead of time.
Exploratory testing made simple
Using an AI-engineered technique, testing may be made more successful
with fewer resources and codes while maximizing device coverage in the
shortest amount of time. Through intelligent assistants that gather test data
and user performances while navigating through an application or system and
recording default test cases, AI readily discovers possibilities.
Error detection assisted by artificial intelligence
Utilize AI-driven QA to uncover problems ahead of time, improve testing,
and forecast failure spots, lowering total costs and increasing customer satis-
faction. Al-enabled performance analysis ensures an application’s or system’s
security by detecting any possible attempts at unwanted access.
 2.5 Industry Use Cases for AI-enabled Collaboration 49

Visual UI testing and monitoring with AI

Artificial intelligence enhances the test platform’s visual testing capabilities
to help development teams deliver aesthetically flawless web apps at unbeat-
able speed and quality. Analyzing regression testing results is straightforward,
effective, and painless using AI-powered monitoring tools.
Maximum code coverage in a short amount of time
Maximum test coverage may be achieved when the test plan, test strategy,
test cases, and other elements are well-planned. Real-time engagements, user
flows, keyword interactions, and data-driven techniques might all be used
in the tests. Businesses may attain 100% code coverage using AI-based
solutions to measure test coverage.
AI-assisted decision-making
AI can go deep into data to find the source of problems affecting corporate
performance. It allows for analyzing consumer data for maximum efficiency,
more excellent inventive capabilities, fine-tuning of products and services,
or better possibilities. It so contributes considerably to the streamlining of
decisions for better processes.

2.5.5 Manufacturing industry trends with emerging AI

AI will impact manufacturing in ways we have not considered. Nonetheless,
we may look at some more notable cases right now.
For a long time, computer visualization has been used for quality man-
agement, discovering product flaws in real time. However, because manufac-
turing now demands more data than ever, and plant managers are unwilling
to pay humans to input data, AI and machine vision can help rationalize
data collection [14]. An employee in a manufacturing facility may take
raw materials from a shelf and then have the stock operation carried out
automatically based on the activity being monitored with the assistance of
a camera. It will evolve into a natural user interface, allowing users to focus
on the task rather than manually entering data into a computer by typing it in
or scanning it in.
Second, artificial intelligence will impact the Internet of Things (IoT).
The IoT will allow businesses to distribute products and services to customers
who may not realize they require them. IoT may also provide extensive data to
manufacturers and distributors, allowing them to assess quality and identify
variables that could create difficulties.
Some key AI statistics in different industries are shown in Figure 2.19.
50 Industry Automation: The Contributions of Artificial Intelligence (AI)

Figure 2.19 AI effect on industrial growth.

• By 2035, AI technology growth is estimated to increase production by

40%.
• By 2035, AI technologies will improve economic growth by an average
1.7% across 16 industries, as shown in Figure 2.19.
The bar chart shown in Figure 2.20 depicts the expected expansion of AI as a
new component of production.

Figure 2.20 The effect of AI on the growth of different industries.

Survey Sources: MarketWatch Accenture, Accenture, Markets and Markets, Gartner, Servion,
Microsoft, and Demandbase.
 2.5 Industry Use Cases for AI-enabled Collaboration 51

2.5.6 The Internet of Things is emerging as Industry 4.0’s future

Two of the most significant recent advancements in information technology
are the Internet of Things (IoT) and artificial intelligence (AI). Nevertheless,
when placed in the context of the industrial revolution, these two technologies
are much more than just buzzwords or current debate topics. The convergence
of AI and IoT has reimagined what the future of industrial automation will
look like. It is anticipated to play a leading role in the Fourth Industrial Rev-
olution. These are two different technologies that have a significant impact
on a wide range of other business sectors. While the Internet of Things (IoT)
is the digital nervous system, AI is the brain that makes choices that manage
the whole system. AIoT (artificial intelligence of things) is a deadly mix of
AI and IoT that produces an intelligent and linked system that is capable of
self-correcting and self-healing; this scenario is shown in Figure 2.21.
By acting on the telemetry data’s patterns and correlations, AI goes
beyond visualizations. It closes the critical gap by adopting suitable data-
driven actions. Instead of simply delivering data to people so that they may
act, artificial intelligence finishes the loop by taking action on its own. It
effectively serves as the central nervous system for all related systems.
Artificial intelligence (AI) will boost industrial IoT on two levels. To
begin with, it influences telemetry data by adding intelligence to the sensors.
Second, in real-time or batch mode, AI will be employed to evaluate the
inbound telemetry data stream. It connects the beginning to the end, i.e.,
devices, to the analysis of the IoT spectrum.
For example, a camera, which functions as an image sensor, will send
a transmission to the IoT system for each frame it captures to perform an

Figure 2.21 Scenario of AIoT.

52 Industry Automation: The Contributions of Artificial Intelligence (AI)

analysis of the feed looking for specific things. The camera accessory uses
artificial intelligence to hold off on delivering the frame until a particular
object has been identified. This expedites the process by reducing the time
the central processing unit (CPU) spends processing each frame.
Other forms of telemetry data, such as speech or natural language process-
ing, can also be processed in a manner analogous to that described above. In
the not-too-distant future, sensors enabled by AI technology will dominate
IoT systems. Soon, intelligent cameras equipped with AI accelerators from
Intel, NVIDIA, and Qualcomm will be the standard image sensors used
across industries.
Deep learning models based on neural networks will be applied to the
incoming sensor telemetry data by advanced Internet of Things (IoT) systems,
allowing these systems to detect anomalies in real time. When the neural
network detects a potentially catastrophic error, the faulty machinery might
be shut down to avoid a potentially fatal accident or incident. The primary
difference between engines based on traditional Internet of Things (IoT) rules
and engines based on artificial intelligence Internet of Things (IoT) rules is
that AIoT rules are proactive rather than reactive.
AIoT systems can identify faults and events in advance, unlike IoT
systems, which are designed to react after an event. Incorporating AI into
the Internet of Things systems holds the promise of predictive maintenance,
which will save businesses millions of dollars in support and maintenance
costs for their equipment.
The integration of AI and IoT is the key to the future of industrial
automation. Artificial intelligence of things (IoT) will influence nearly every
business sector, for example, automobile, aviation, economics, healthcare,
manufacturing, and supply chain management.

2.5.7 Future scope of research

Processes in industries will become more effective and ultimately more
straightforward as a result of AI or AGI, but they will require considerably
more complicated systems to support them. To reach the automated-AI
future-designed phase, a lot of design, analysis, testing, and implementation
is required, as well as a study on top of deep industrial process and AI
expertise. AGI will manage the matching, bidding, and execution, as shown
in Figure 2.22, on a distributed network in the future.
The question of whether consumers and society will embrace the solu-
tions is dependent on proof that the keys have shown to be safe, trustworthy,
 References 53

Figure 2.22 The trend of next industry automation.

and valuable to humans. There is still more study to be done, both in terms of
technological and sociological factors, particularly regarding the importance
of acceptability. The impact of AGI is heavily influenced by the aim utilized
during the creation process. AGI research is also looking into how intention
may be managed. We must begin designing this future today, using our shared
European principles of human-centered AI.

2.6 Conclusion
This chapter starts with the role of artificial intelligence in industrial automa-
tion and then explains the potential of automated systems. Application
landscape and production-related scenarios were explained with a real-
time model. Autonomy-level classification of industrial AI applications was
explained using one particular example. The impact of artificial intelli-
gence in Industry 4.0 was explained for different production scenarios and
finally ended with industry use cases for AI-enabled collaboration in various
industry sectors. It also explained the future scope in the area of industrial
automation. Artificial intelligence blends communication, computation, and
control processes to satisfy industrial objectives. The emergence of arti-
ficial intelligence unquestionably made significant advances in all human
endeavors.

References
[1] Tomohiko Sakao, Peter Funk, Johannes Matschewsky, Marcus Bengts-
son, Mobyen Uddin Ahmed, AI-LCE: Adaptive and Intelligent Life
Cycle Engineering by applying digitalization and A.I. methods – An
emerging paradigm shift in Life Cycle Engineering, Procedia CIRP,
Volume 98, 2021, Pages 571-576, ISSN 2212-8271.
54 Industry Automation: The Contributions of Artificial Intelligence (AI)

[2] Baicun Wang, Fei Tao, Xudong Fang, Chao Liu, Yufei Liu, Theodor
Freiheit, Smart Manufacturing and Intelligent Manufacturing: A Com-
parative Review, Engineering, Volume 7, Issue 6, 2021, Pages 738-757,
ISSN 2095-8099.
[3] Kumar, T. Ananth, S. Arunmozhi Selvi, R. S. Rajesh, and G. Glorindal.
“Safety wing for industry (SWI 2020)–an advanced unmanned aerial
vehicle design for safety and security facility management in industries.”
In Industry 4.0 Interoperability, Analytics, Security, and Case Studies,
pp. 181-198. CRC Press, 2021.
[4] Berry, W. L., Hill, T. J. and Klompmaker, J. E. (1995), “Customer-driven
manufacturing,” International Journal of Operations & Production Man-
agement, Vol. 15 No. 3, pp. 4-15.
[5] Raj, Pethuru, Jyotir Moy Chatterjee, Abhishek Kumar, and B. Balamu-
rugan, eds. Internet of Things uses cases for the healthcare industry.
Springer, 2020.
[6] Erwin Rauch, Patrick Dallasega, Dominik T. Matt, The Way from Lean
Product Development (LPD) to Smart Product Development (SPD),
Procedia CIRP, Volume 50, 2016, Pages 26-31, ISSN 2212-8271.
[7] M. Lopes Nunes, A. C. Pereira, A. C. Alves, Smart products develop-
ment approaches for Industry 4.0, Procedia Manufacturing, Volume 13,
2017, Pages 1215-1222, ISSN 2351-9789.
[8] Kumar, T. Ananth, S. Arunmozhi Selvi, R. S. Rajesh, and G. Glorindal.
“Safety wing for industry (SWI 2020)–an advanced unmanned aerial
vehicle design for safety and security facility management in industries.”
In Industry 4.0 Interoperability, Analytics, Security, and Case Studies,
pp. 181-198. CRC Press, 2021.
[9] Gudin J, Mavroudi S, Korfiati A, Theofilatos K, Dietze D, Hurwitz P.
Reducing Opioid Prescriptions by Identifying Responders on Topical
Analgesic Treatment Using an Individualized Medicine and Predictive
Analytics Approach. J Pain Res. 2020;13:1255-1266.
[10] Wo Jae Lee, Haiyue Wu, Huitaek Yun, Hanjun Kim, Martin B. G.
Jun, John W. Sutherland, Predictive Maintenance of Machine Tool
Systems Using Artificial Intelligence Techniques Applied to Machine
Condition Data, Procedia CIRP, Volume 80, 2019, Pages 506-511, ISSN
2212-8271.
[11] Jay Lee, Hossein Davari, Jaskaran Singh, Vibhor Pandhare, Industrial
Artificial Intelligence for industry 4.0-based manufacturing systems,
Manufacturing Letters, Volume 18, 2018, Pages 20-23, ISSN 2213-
8463.
 References 55

[12] B. Chen, J. Wan, L. Shu, P. Li, M. Mukherjee, and B. Yin, “Smart

Factory of Industry 4.0: Key Technologies, Application Case, and
Challenges,” in IEEE Access, vol. 6, pp. 6505-6519, 2018, doi:
10.1109/ACCESS.2017.2783682.
[13] Alejandro Germán Frank, Lucas Santos Dalenogare, Néstor Fabián
Ayala, Industry 4.0 technologies: Implementation patterns in manu-
facturing companies, International Journal of Production Economics,
Volume 210, 2019, Pages 15-26, ISSN 0925-5273.
[14] ozdić, E. (2015). Smart factory for industry 4.0: A review. International
Journal of Modern Manufacturing Technologies, 7(1), 28-35.
[15] orge Ribeiro, Rui Lima, Tiago Eckhardt, Sara Paiva, Robotic Process
Automation and Artificial Intelligence in Industry 4.0 – A Literature
review, Procedia Computer Science, Volume 181, 2021, Pages 51-58,
ISSN 1877-0509.
[16] onepudi, P. K. (2018). Application of Artificial Intelligence in the
automation industry. Asian Journal of Applied Science and Engineering,
7(1), 7-20.
 3
Industry Automation: The Contributions
of Artificial Intelligence

B. Akoramurthy1 , K. Dhivya2 , and Kassian T. T. Amesho3

1 School of Computing & Information Technology, REVA University, India
2 Department of Computer Science & Engineering, India
3 Tshwane School for Business and Society, Tshwane University of

Technology, South Africa

E-mail: [Link]@[Link]; dhivyaakoramurthy@[Link];
[Link]@[Link]

Abstract
In the last decade, there has been a tremendous technological breakthrough
in almost any industry, like healthcare, the military, education, and manufac-
turing industries, which includes new gadgets, discoveries, and refinements
in the existing system. The current implementation of 5G technology tells
us that Industry 5.0 has stormed into the market, where people are working
with smart machines and robots. This also addresses the automated infor-
mation exchange mechanism and other various technologies. So, in order to
create smart factories and industries, it connects the ideas of the Internet of
Things, cloud technology, and cognitive computing. Industry 5.0 ties together
physical and digital advancements to build such intelligent businesses. Fur-
thermore, it simplifies and maintains the actual needs as well as the supply
chain in businesses. Companies and organizations profit from it because of
its efficiency and capacity for quick judgments. Artificial intelligence is the
core of this technology and a crucial factor in Industry 5.0. AI manages every
technological item we use nowadays. The environment significantly impacts
AI since every piece of data is linked to the Internet. A range of enterprises,
such as those associated with manufacturing, e-commerce, sports, security,
and defense, have shown the efficacy of AI due to its malleability. Thanks to

57
58 Industry Automation: The Contributions of Artificial Intelligence

the combination of Industry 5.0 and AI, almost all applications can access
various options and services. Even though Industry 5.0 has demonstrated
great success across all domains, it is still in the early stages of its extension,
adoption, and hiring. We researched AI and Industry 5.0 in this work and
explored their difficulties, significance, methods, and results. The relevance
of innovation in the context of industry and everyday life has also been
examined, along with its impacts and advantages.

Keywords: Industry 5.0, artificial intelligence, energy efficient metrics,

datacenter carbon footprint computation

3.1 Introduction
In order to increase production, businesses and initiatives are investing
more money daily. The Internet of Things, artificial intelligence, cloud
computing, and other cutting-edge technologies are currently driving the
production world forward. Adopting these technologies significantly impacts
the advancement of physical resources and robots.
Businesses are motivated by their advancement, that is, by incorporating
technologies that boost our country’s economic growth to a better level [1].
The data is exchanged among these intelligent robots with this technology,
and these computers use artificial intelligence to act like humans and make
choices about every issue. Currently, artificial intelligence is a growing field
of study. According to the market assessment, a 170 billion-dollar growth is
anticipated by year’s end. The manufacturing rate will increase with this new
technology, and labor costs will drop by 80 billion dollars. On the other hand,
the Internet of Things, which connects all electronic gadgets and has a target
of 95 billion US dollars to expand to around 200+ ZB bytes by the year 2030,
has had a significant influence in addition to AI. Likewise, AI systems can
be anticipated to develop into a gadget with swift and grassed-up outcomes
in the foreseeable future in order to acquire this large volume of facts and
advance significant ideas from them. The business environment is shifting due
to rapid breakthroughs in AI technology, and its use is replacing human-held
tasks. Also, it fosters various application-based concepts and boosts company
efficiency. Some of the most advanced nations are embracing this innovation
in all its application-based situations and enhancing their economies. Natural
language processing and perceptual computing have found new ways and are
being altered by the rapid development of deep neural networks and machine
learning technologies. Sales, human resource management, personal finance,
 3.2 Literature Review 59

ads, corporate plans, source and supply chains, commerce, retailing, systems
engineering, etc., are only a few characteristics that make up artificial intel-
ligence. The deployment-oriented categories of AI are depicted in Figure 3.1
precisely. AI aims to bring technology and machines together so that they
can brilliantly use human intelligence. One of the fastest-growing industries
today is AI, and jobs in this industry are also increasing.
The use of AI in numerous industries, including education, transportation,
industry, banking, construction, recreation, governance, and healthcare, is
explained in this section. The administration, individuals, or any organization
can meet the changing market and industry demands for technical solutions
to economic problems. Various forums for administration, financial plans,
and company endeavors must be built using an agglomeration of data from
various sources, including digital networking, channels, adverts, etc. Every
company’s economic progress is negatively impacted by the barriers that
arise in business [6]. Businesses of any size may use artificial intelligence in
emerging nations to aid with supply chain and delivery system problems and
provide a variety of service availability. The use of automated corporate pro-
cesses increased productivity, investment firms’ and governments’ support,
and other technologies connected to AI, advanced analytics, and IoT may all
contribute to the growth of possibilities in various fields [7]. AI systems are
a technology that has been trained to replicate human behavior. Its various
areas set it apart from other software packages, thanks to its unique attributes
of quick efficiency, numerical simulations, and methods, and it is the most
widely used data source. AI researchers use various algorithms to link highly
relevant data and deliver high-quality software competence services. Our
study will also offer a framework for using AI in business to address diverse
technical problems.

Objectives
1. To review research and analysis on the role of AI in the development of
Industry 5.0.
2. To determine the relationship between AI and Industry 5.0 and to design
necessary technologies’ future.

3.2 Literature Review

AI and its many tools are essential in modern production plants. Many
organizations are heavily exploiting it and changing their manufacturing and
60 Industry Automation: The Contributions of Artificial Intelligence

functioning methods. Most gadgets and equipment use a variety of sensors

and a digital storage system and are intelligent enough to carry out tasks.
Along with operations, they are also preserving real-time data. Due to this,
productivity and healthy competition have also improved. Real-time data
collection has increased the analytical capacity of the system, and since the
system makes decisions depending on the analysis of the information, the
industry as a whole is more efficient. Industries use various AI- and ML-
related technologies to expand operations, find new growth opportunities,
build high-quality architecture, and more. There is the participation of For-
tune 500 corporations in it as well. The majority of offices now have AI
capabilities, which boosts productivity. Industry 5.0 is closely tied to IT-based
solutions, software, and applications like AI, IoT, deep learning, etc. These
cutting-edge technical advancements are providing tremendous assistance to
organizations making progress to keep up with the times [10]. Virtual and
practical perspectives are integrated into Industry 5.0, and big data analysis
is a critical component of this. 3D printing and robotic systems are excellent
supporters of Industry 5.0.

3.3 Industry 5.0 and AI

In the past, the concept of AI was compartmentalized into various business
sectors. Natural language processing (NLP), automated coding (AC), robotic
systems (RS), computer vision (CV), automatic hypothesis proving (AP), and
intelligent data extraction are a few examples of these types of technologies
(IDT). Because these fields of application have grown to such a large extent,
each one can stand on its own as an independent academic specialization.
Currently, the most accurate definition of artificial intelligence (AI) is a set of
guiding principles underpinning many of these applications [9].
Robots’ application of artificial intelligence (AI) to complete previously
impossible tasks, reduce costs, and improve product quality is essential to
smart manufacturing and Industry 5.0 [10]. Cyber−physical systems make
incorporating artificial intelligence technologies into the manufacturing sec-
tor easier. This contributes to the ongoing convergence of the physical and
digital worlds. The application of AI has a revolutionary impact on the
manufacturing sector, allowing it to meet modern demands such as increased
reliance on sensors, increased personalization of products, and faster product
development cycles [11]. Various goods can be produced thanks to artificial
intelligence (AI) and flexible robot technology efficiently. When sorting
 3.3 Industry 5.0 and AI 61

Figure 3.1 Bibliographic analysis of Industry 5.0.

through mountains of raw datasets [12] collected by various sensors, infor-

mation retrieval and other forms of artificial intelligence can be utilized as
valuable tools.
There have been three significant shifts in the approaches taken by techno-
logical practitioners. According to the vast majority of available data sources,
we are currently in the fourth one, or at least we should be. Figure 3.2 depicts
the times that these uprisings took place in the form of a clock face. Take
note of the shorter gaps in time that now exist between rotations. There was a
100-year gap between the three earliest revolutions in history. Approximately
40 years was all it took to get from the third to the fourth one. There is a
possibility that we will reach the fifth one in fewer than 40 years from now.
The German government recently initiated a brand new initiative that they
are referring to as “Industry 4.0.” The overarching goal of Industry 4.0 is to
facilitate “Smart Manufacturing for the Future.” Its purpose is straightforward
and in line with previous revolutions: to use cutting-edge technology to
increase output and make it possible to produce on a large scale. Industry
4.0 is making progress, thanks to the contributions of many well-established
technologies. The Internet of Things, robots, artificial intelligence (AI), big
data, and cloud computing are some of the most prominent technologies
currently trending. Industry 4.0 is made possible by many other technological
62 Industry Automation: The Contributions of Artificial Intelligence

Figure 3.2 From Industry 1.0 to Future Industry 5.0.

advancements as well. This encompasses ambient intelligence and 3D print-

ing technologies, augmented and virtual reality, smart factories, intelligent
supply chains [6], and other similar innovations. It is essential to keep in
mind that none of these advancements were designed with Industry 4.0 in
mind when they were developed. These recent technological developments
have been brought together under the umbrella term of Industry 4.0 to achieve
“smart production” [7, 13]. This is because it is a government project being
carried out from the highest level down, with the idea already being decided.
Regarding Industry 5.0, there are currently two different schools of
thought. One is the idea of “humans and robots working together.” In the
future, humans and robots will work together to accomplish shared goals
whenever and wherever feasible. The remainder of the work will be com-
pleted by machines, freeing up more time for people to devote to creative
endeavors. The blue economy is not the same thing as Industry 5.0; it is its
own distinct concept [7]. It is possible that the requirements of industry, the
environment, and the economy can all be satisfied through the strategic use
of biological resources in industrial settings. This is a distinct possibility. The
European Commission refers to developing sustainable natural resources and
transforming these consumption and waste processes into value-added goods
as a “low-carbon economy.” The term “low-carbon economy” was coined
by the European Commission. Products with additional value include, but
are not limited to, food, vitamins, biofuels, and biogas. The same holds for
other examples. Biogas is yet another example of this. Agriculture, forestry,
 3.4 Problems with Human–Robot Collaboration 63

fishing, farming, pulp, and paper production are all included as segments of
the chemical, biotechnology, and energy sectors.
Microbiology, nanotechnology, information and communication tech-
nologies (ICT), engineering, and local and tacit expertise all play essential
roles in their respective fields, making them fertile ground for innovation
[14]. The concept of colonization is the driving force behind the economy,
and it has the potential to cause revolutionary shifts in many different sectors
[15]. Because of this, the blue economy may play a role in the subsequent
industrialization, or it may even be its primary focus. The perspectives of
Industry 4.0 and Industry 5.0 are contrasted and compared in Table 3.1. If
you want a thorough evaluation of these predictions, refer to [7]. Remember
that the blue economy may not be the sole focus of Industry 5.0 but rather the
collaboration of humans and robots.
Further, topics like the space industry, space farming, and space habitation
may either usher in the next generation or be a part of it. Space is a finite
resource, and scientists have warned us to use it sparingly [16]. As with
the “mining boom” of the past, space farming could be the next economic
upswing.

Table 3.1 A comparison of different industrial revolutions’ visions.

Industry 4.0 Industry 5(a) Industry 5(b)
Modus Smart Robots working Blue economy
operandi manufacturing along with humans
Inspiratio Mass manufacturing A connected Resilience
n society
Energy Electric and solar Renewable sources Sustainable ad
source bionics
Related AI, IoT, cloud Robot−human Cognitive
technolog computing, robotics, collaboration computing,
ies and big data quantum
computing, and
robot−human
interactions
Research Organizational Smart cities Agriculture, smart
areas research, process environment,
Innovation and
improvement, and business biology, and
product finance sectors
management
manufacturing
64 Industry Automation: The Contributions of Artificial Intelligence

3.4 Problems with Human–Robot Collaboration

Incorporating robots into companies was covered in a previous paper [2].
The problems found point to the potential for considerable organizational
adjustments. Changes in organizational behavior, structure, workflow, ethics,
and work environment are among the problems. Other significant chal-
lenges include the adoption of machines in the workplace, bias against
humans or robots, confidentiality and confidence in a human−robot work-
place, redesigning environments for cyborgs, teaching, and orientation. The
concerns are listed in Table 3.2.
Bagdasarov and her associates looked at the organizational aspects
of using robots at work [11]. Three variables were discussed: interper-
sonal, organizational, and robotic agents. The experts advise taking into
account numerous criteria for each collection. Individual aspects include
the employee’s age, gender, educational degree, technological knowledge
and experience, expectations, social perception, and social competence. The
organizational aspects include the procedure, external conditions, psycho-
logical and social context, orientation, and synchronization of worker and
machine goals. The robotic agent elements include the robot’s look, behavior,
interactivity, and safety.
In the remaining section, we concentrate on human−robot co-working
challenges not covered in [2]. Be aware that some of the problems affect
the organization and its stakeholders. It might be challenging to distinguish
between a business and an emotional problem.

Table 3.2 Impacts associated with robot integration in industries.

Impacts of Robot Integration on Organizations

Genesis of organizational behavior
Employment acceptance of robots
Changes in workflows and organizational structures
Changes in ethical behavior
Discriminatory policies directed at humans or machines
Confidentiality and trust are essential in a human−robot workplace
Constructing and disseminating information
Reorganizing the factory floor with robots
 3.4 Problems with Human–Robot Collaboration 65

3.4.1 Issues with law and regulation

According to one definition [5], robots are machines that have the ability to
detect, reason, and act. This definition is straightforward yet troublesome.
What do a robot’s thoughts and behavior mean? The Oxford Dictionary
defines a robot as a device capable of automatically performing a complicated
sequence of tasks, particularly one that a computer can program. The robot is
a device that resembles a person and can carry out various complicated human
functions, including walking and talking, according to the Merriam-Webster
dictionary. A robot is a device that resembles a person. These definitions
contain their own set of errors as well. One resorts to an abstract concept, such
as acting out a complicated series of behaviors in a habitual manner. The other
individual thinks that a robot can pass for a human. These hazy explanations
demonstrate that we have not yet developed a successful solution.
The many complications that would arise without the right legislation
make collaborative labor with robots impossible. An official legal definition
of a robot is needed first. Although “robot” has a scientific meaning, only
the legal meaning is enforceable by businesses and other organizations. The
range of possible automated machines is quite broad. A robot is an example
of an automated machine. Different degrees of machine automation have
been identified [4]. Any regulation governing the collaboration of humans
and robots must make it abundantly clear that an automatic system is not the
same as a robot. There is also a strong connection between drones and robots.
The legislation should spell out the similarities between robots, drones, and
cobots. It is essential for the legislation to clearly define the categories of
robots, aircraft, and cobots.
Whether or not robot software is required to adhere to stringent and
complex rules or is allowed some leeway in the event of system failures
should all be addressed in the legislation and any other relevant regulations
governing the use of robots in the workplace. The rules governing robot
creation, production, and licensing will be the subject of several controver-
sies. Are we going to permit any individual, business, or organization to
create a robot and integrate it into our residences, businesses, and everyday
lives? Most nations are presently ineffective in controlling the use of drones
in the air. With the accomplished things of autonomous robots, we must
be prepared. A large number of research investigations will concentrate on
the legal and ethical challenges related to the usage of bots, aircraft, and
cobots.
66 Industry Automation: The Contributions of Artificial Intelligence

3.4.2 Subjective opinion for using robots at work

Everyone has a distinct taste for utilizing or refraining from a specific tech-
nology. Some people are keen to work with machines and will be vehemently
opposed to the concept.
Companies interested in utilizing human−robot collaboration in the
workplace should be conscious of these individual inclinations. It will not be
simple or even practicable for firms to convert to human−robot co-working
settings if the majority of their workers have a hostile perception of robots.
Although human−robot co-working is not a reality, representations of col-
laborating with robots rather than actual experiences significantly influence
people’s preferences. Numerous well-known films and television programs
have portrayed the futuristic idea of robots destroying humanity or ruling
over the earth. These unfavorable media reports have an impact on people’s
preferences for dealing with robots.

3.4.3 Psychosocial problems caused by human–robot

collaboration
Modern technology has a particular psychological impact on people. Video
game addiction is seen as a psychological condition that is receiving more
attention. Some people are really devoted to their cell phones. Even a phobia
known as “nomophobia” or “fear of not being able to use mobile phones”
exists. Nomophobia, as described by the Collins dictionary, is a stressful
condition brought on by not having access to or being unable to use a mobile
phone. At this time, whether there will be a “robophobia” or a “nomophobia”
is unknown. Nevertheless, it is very possible that a robot will have a number
of psychological consequences for many workers that have never been seen
before. Corporate robotics will be a study area for the mental disorders
associated with using robots at work and their impact on organizational
environments.

3.4.4 Changes that result from human−robot collaboration

Humans are hardwired to interact with others. They engage with people
like them in all aspects of their lives, including their careers. Many work-
places engage in the practice of organizing social activities to increase
productivity. As the number of robots working alongside humans rises, the
number of humans may fall. A consequence of this could be diminished
 3.5 Wafer Fabrication Automation 67

social interaction between individuals. The introduction of robots into the

workplace is likely to have unintended consequences for human interaction,
even if the total number of employees remains unchanged. Some humans
may even prefer interacting with sociable robots. Some people may assume
that robots’ social behavior is merely a result of clever programming, but this
is not necessarily the case. Workers will have different ideas about interacting
with robots in the workplace. The situation deteriorates rapidly when robots
are elevated to managerial or executive roles. Employees are appreciative of
their superiors and employers in general. Treating others with deference is
a hallmark of sociability. People might wonder if they would like a robot
manager or not. This sort of issue may seem strange, confusing, or even
frustrating to some people because a robot will not have any concept of
respect. Findings from the study [18] show that young people can develop
strong emotional attachments to robots in the same way they do to real pets.
In the not-too-distant future, humans may interact with robots in a way that
is similar to how today’s youth interacts with and views their cell phones.
The study of how humans can work together with robots appears to require a
substantial contribution from the social sciences.

3.4.5 The shifting functions of human resources divisions

One of the key duties of HR departments is to carry out employment assess-
ments, create position descriptions, and fill open positions with qualified
candidates. Except for those sectors that use industrial robots, most busi-
nesses now employ people. Human resource departments will encounter new
difficulties as robots become integral to enterprises. Workers will be expected
to determine the jobs that robots will take over in addition to their existing
duties. In essence, they will choose which occupations to assign to robots.
Divisions in charge of HR will become more significant and have more duties.
HRM divisions will inevitably change, and they could even have new names.
Many organizations and enterprises nowadays promote themselves as
“eco-friendly.” They assert that they are environmentally conscious. Whether
these activities increase prices, “eco-friendly” firms and organizations aim to
behave appropriately toward the ecosystem. There may even be companies in
the future calling themselves “human.” Even if it costs more to be a “human”
company, some businesses will only hire people, even if using robots would
be less expensive. These “human” organizations will assert that they treat
people with social responsibility and provide jobs.
68 Industry Automation: The Contributions of Artificial Intelligence

3.5 Wafer Fabrication Automation

The wafer fabrication sector has changed tremendously in recent times. An
assortment of technological improvements, including both equipment and
software, are fuelling this growth, permitting the manufacturing of electronic
chipsets to automate their processes and work at their full potential. Several
aspects of production are automated, ranging from supply monitoring to
regulatory oversight, supply chain management, and sophisticated control
systems. Regardless of the process, automation has been assumed to pro-
vide a critical platform for increasing production efficiency and improving
quality for the customer industry. The method in which the items are kept,
handled, and distributed throughout manufacturing is an evident example
of how automation has enhanced chip production processes, as shown in
Figure 3.3. Previously, clean room professionals performed all of these tasks
mechanically. Thanks to automated technology, processing and delivering
clean room chips hardly require extensive human interaction. This in itself
improves pollution control, and that also enables much better product changes
across the fabrication using path optimization systems.
Another case in point is process control automation. Experts were once
obliged to physically enter input variables and make assessments to check for
operational irregularities. Once errors are found, instrument measures must

Figure 3.3 Wafer fabric automation.

 3.6 AI as a Vital Technology in Industry 5.0 69

be explicitly performed to prevent further yield losses. Such a long-time pro-

cess and error-prone procedure is tedious. Automation has gradually replaced
this traditional approach with contemporary practices (i.e., automatic error
detection systems, profit/output management systems, etc.). In some circum-
stances, OCAP (out of control action plan) may respond proactively to reduce
the loss drastically. Data generated can now be automatically gathered and
evaluated in real time. These robotic architectural solutions give businesses
unprecedented insight into their operations, allowing managers to discover
and resolve issues with minimal human interaction.

3.6 AI as a Vital Technology in Industry 5.0

Around 2032, artificial intelligence (AI) technologies are expected to have
increased the global GDP by US$17 trillion.
Between 2019 and 2024, the market for AI will increase by $79 billion.
Due to the consistent acceleration of month-over-month growth, the market’s
expansion pace will pick up throughout the projection period.
Businesses are quickly implementing AI-based products and solutions to
transform their processes and boost revenue. In order to take advantage of
technological advances, streamline operations, and make companies more
flexible for planned increases, AI-powered technologies are increasingly
being adopted in the industrial sector.

3.6.1 Impact of AI on different industries

The following are the areas where the AI technology has been mostly used,
as shown in Figure 3.4.
1. Algorithms for deducing formalized assertions from logical formula-
tions and providing machine-based proofs, as well as for demonstrating
the accuracy of hardware and software.
2. Knowledge-based systems include techniques for simulating and gath-
ering expertise; software that supports experts and simulates human
competence (formerly referred to as “expert systems”).
3. Pattern recognition and analysis: general intuitive methodological
approaches, and ML in specific.
4. Robotics: unsupervised networks, or the control of robots without
human intervention.
70 Industry Automation: The Contributions of Artificial Intelligence

5. Intelligent multimodal human−machine interaction involves linguistic

analysis and “understanding” of words, pictures, actions, and other
social interactional modalities.
The application of the AI technique known as machine learning can be found
across many different industries. The fact that it has such a wide range of
applications, making it suitable for various use cases along the value stream
depicted in Figure 3.4, is indisputable evidence.
Although in the past, most companies have relied on their internal data for
supervised machine learning, recently, several companies have begun inves-
tigating the possibility of combining their own internal and external datasets
as well as their own internal and external data-science-related capabilities in
order to provide even more in-depth insights.

Figure 3.4 Impact of AI on different use cases (industries).

 3.8 The Scenario of AI in the Focus of Manufacturing 71

3.7 Artificial General Intelligence (AGI)

However, concentrating on current actions involving poor or narrow artificial
intelligence, this chapter will also outline a future scenario in which AI no
longer merely reacts but acts proactively, smartly, and adaptable. Along with
offering the answer, it is important to comprehend the overall structure, see
connections on your own, and make new deductions. This type of AI, often
known as artificial general intelligence, is equivalent to human cognitive
abilities.
Knowledge-based presentations, linguistic interactions in natural settings,
the application of tactics, the detection of anomalies, and acting are only a few
examples of the characteristics of artificial general intelligence. AGI systems
assist cognitive research, math-focused intelligence, and decision-making, as
well as notice and respond to risks. There are currently no commercially
available solutions except ROSS Intelligence, Expert System, Self-driving
Cars, IBM’s Project Debater, etc.
Research is being done on systems with cognitive synergies that enable
the generation of AGI by combining several types of memory and their
associated learning techniques. CogPrime is a promising initiative. Work on
self-reflecting systems is still ongoing in the ability to form original and
unique judgments, which is discussed in the context of AGI. Kurt Gödel’s
self-referential formulations served as inspiration for several useful examples,
including the Gödel machine, which, besides resolving issues, searches for
self-rewrites for which it can provide evidence of their benefits.

3.8 The Scenario of AI in the Focus of Manufacturing

The most popular kind of machine learning is supervised machine learn-
ing, in which software applications are given organized data in order to
identify trends and comprehend and analyze breakthroughs. The applications
of Machine Learning and Intelligent Robots that have been a scenario of AI
Implementation in businesses. In Figure 3.5, a technological breakdown is
shown.

Possibilities for Information systems and automation throughout its life

cycle
Within the organization, services are powered by data challenge traditional
value chains. The new “manufacturing as a service (MaaS)” model makes
the consumer a stakeholder of any new product developed, which is made
72 Industry Automation: The Contributions of Artificial Intelligence

Figure 3.5 Percentage of technology adopted for deploying AI in Industry.

possible via interconnected production. Figure 3.4 shows that this results in
the design, manufacture, and consumption of items in local areas.

3.9 Automation based on AI (ABAI)

Automation benefits companies by streamlining tasks that were previously
handled by humans while adhering to a set of criteria. This saves time
and effort, ensures reproducible outcomes, and considerably improves pro-
ductivity. When mankind reaches the epoch of Industrial Automation 5.0,
powered by technology such as artificial intelligence, automated processes
are poised to grow further. Industries anticipate technology that can give some
operational flexibility rather than merely regulated automation. AI-powered
automation allows machines to make decisions like people, responding to
new things in real time. This enables jobs to be completed more wisely and
efficiently, allowing firms to enhance output quality, minimize response time
to challenges, and boost productivity and sustainability.
In this chapter, we shall review the major industrial automation applica-
tion cases to demonstrate the value of AI-driven “smart automation” in device
fabrication.
 3.9 Automation based on AI (ABAI) 73

3.9.1 Computerized root cause analysis using AI

Another vital engineering activity in the semiconductor industry is root
cause analysis. Conventional RCA methods rely significantly on manual
coordination between tools and software solutions. For instance, when a
productivity problem arises in fabrication, a product engineer may perform
the first analysis stage by reviewing silicon output maps to hunt for any
failure anomalies. The corresponding process expert will then be contacted,
and they will work together to collect and analyze pertinent data input and
search for connections. An instrument specialist will also be required if the
malfunction indicator is specific to a particular processing instrument in order
to help determine the underlying reason. The RCA procedure requires many
resources, is laborious, and often takes days to weeks to complete. Latency to
the root cause may take longer for complicated problems, delaying the effect
on productivity. This procedure can be considerably enhanced by automation.
Sophisticated analytic solutions offer a variety of mechanized capabilities,
from data collection to thorough trace analysis, to assist in reducing the
work required for RCA and accelerating time-to-RCA. Like CM in the
semiconductor industry, modern analytical tools give professionals a basis for
automated data collection that enables them to combine quality and process
data from many data sources into a unified, integrated solution. An instrument
designer, for instance, may swiftly perform drill-down analytics and quickly
investigate a problem from several viewpoints when all pertinent information
is immediately accessible.
Additionally, cutting-edge technologies like complete trace insights,
information retrieval, and semiconductor mapping predictive analysis are
combined into one coherent system, enabling a more straightforward and
automated process. This makes having different domain assets and software
packages unnecessary, considerably streamlining the RCA procedure. Let us
examine the automated components of this streamlined approach in more
detail. The system can evaluate a collection of silicon yield maps to track,
identify, and sort typical failure trends, providing a technician with unique
strategic priorities for further assessment. This is possible because all per-
tinent quality and process data have been collected into the platform of the
solution.
The next step is to specify the decision-making process, which may be
done either explicitly by an expert, depending on the domain, or automatically
by employing the system’s AI-driven anomaly identification functionality.
With the help of this automated characteristic, the option can accurately set
74 Industry Automation: The Contributions of Artificial Intelligence

the assessment cutoff point for data comparison and intelligently identify
data that falls outside of the typically performing groups. For illustration,
in Figure 3.6, chips with yield levels above a certain percentage can be
equated to chips with yield levels below, or chips with a critical failure
sequence can be, especially compared to chips without a typical failure trend.
The same analysis mentioned above in RCA platform includes similarity
functions and data harvesting, which are carried out automatically on all
relevant information in order to learn and locate the alleged process tool
(study of commonalities). Once more, utilize AI to gather data from the
suspicious tool statistics and smartly examine them to rapidly and precisely
identify the sources of problems to the ingredient and ingredient stage level.
An architect could use this autonomous data analysis platform to determine
the origin of any output or hiccup using a single technology platform without
involving other subject matter experts. A streamlined methodology like this
could reduce RCA time from days to hours.
Consequently, an architect may conduct appropriate measures more
rapidly and reduce the effect of problems. As indicated in the preceding
section, another illustration of automation that might boost training and
knowledge is the capacity to launch data analysis via a project timescale
automatically. Recurrent studies might be performed automatically, and sum-
maries containing the root of the issue with its results can be delivered to
experts, allowing them to become more successful in controlling process per-
formance and productivity. Daily productivity reports, for instance, including

Figure 3.6 Automated trace analysis.

 3.9 Automation based on AI (ABAI) 75

RC results, enable an analyst to fix output effect concerns while physically

conducting a study, thus eliminating operational effects. Likewise, a model
of data containing linear measuring data, including RCA results, enables an
expert to track failure events and take remedial steps as needed.

3.9.2 Intelligent computing in product matching

In production, product matching is a critical technical activity. Engineers may
more efficiently maintain high customer satisfaction and production output
by maintaining quality tool effectiveness across the device fleet within a
processing region. The conventional way of product matching involves an
expert collecting device information and contrasting the overall stats of the
information on each sensor, two compartments at a time. In some instances,
an expert may need to visually analyze the data collected from the suspi-
cious sensor to search for anomalies. This compartment analysis method is
laborious and might take a long time if completed over a whole array of
devices. Product matching has been dramatically enhanced by mechanization
in product comparison analysis. The very first benefit is the removal of
the requirement for experts to constantly collect information by providing
an interface to combine sensor readings from all operational instruments
automatically. Asset/inventory assessment may be completed much more
quickly since the information from all pieces of equipment is compiled onto
a unified platform, allowing for real-time data comparison between several
compartments, as shown in Figure 3.7.
A system can plan product assessment events using automation as well.
Now, periodic tank condition data may be created electronically and sent to
an engineer daily, monthly, or on other schedules. This makes it possible for
them to efficiently monitor device operation trends and quickly respond to
any emerging problems. Technical performance has significantly increased
due to the elimination of mechanical examinations.
The analysis of product data is automated through the use of computer
intelligence. An AI-driven platform may easily execute trail analysis by
automatically and frequently aligning, segmenting, and comparing all trail
data, even with thousands of variables in each cell to be studied. This
enables the model to identify every anomaly in great detail, regardless of
how obscure the problem may be. With AI, the system can automatically do
trail analysis on all track data, unlike product analysis, which uses aggregate
data and might cause delicate signals to be missed in the computations of
the overall statistics. Designers benefit from the greatest insight into possible
76 Industry Automation: The Contributions of Artificial Intelligence

Figure 3.7 Aggregated product matching and trace analysis.

process problems as well as the most precise outcomes. Additionally, AI-

driven automation enables a system to correctly and intelligently select the
compartment with the best reliable behavior and designate it as the bench-
mark product or compartment. While choosing the benchmark product, an
expert can additionally make modifications to test against relevant quality
data to strengthen the benchmark’s validity. It is represented in Figure 3.7.
A time-consuming and tedious task for an expert to physically choose the
benchmark data for each study is eliminated by this automated capability.
The functionality guarantees that the optimal cell is consistently used as
the benchmark and significantly lowers the work needed to build up newer
analyses.

3.10 Robotic Process Automation

Robotic process automation (RPA) is when computers do jobs that were
used to be done by people in the service industry. Software robots or AI
workers that can do simple tasks are used to automate the process. The
designer will probably take a screenshot and use variables to set up the task
parameters. These tasks include opening emails, filling out forms, copying
and pasting data, signing into programs, and so on [4]. Logging into programs
and copying and pasting data are also tasks that need to be done. Van der Aalst
et al. [3] say that RPA is an umbrella term for software that interacts with the
user interface of different computer systems. This is what researchers said
about RPA. RPA’s main function is based on component identifiers instead of
 3.11 The Digital Solutions Entangled in Industry 5.0 77

screen coordinates or XPath options. This is different from traditional models

of process automation, such as screen capturing, scraping, and macros, which
rely heavily on the graphical interface of the workstation. Robotic Process
Automation refers to the automation of service that mimic human labor[3].
Software robots or AI employees who can efficiently do basic activities are
used to automate the process. The designer sets the job instructions using
some type of screen capture and create variables. These operations range
from signing into programmes,copying data, pasting it , opening emails and
filling out forms among others [4]. According to Vander der Aalst [Link] [3]
RPA is an umbrella term encompassing This makes it more likely that the
user will remember the GUI for a long time. Since 2015, sales at RPA
software companies have increased [3, 4]. These technologies have also been
used to automate business processes, inspections, and digital forensics [4, 5].
The fifth industrial revolution, also called Industry 5.0, opens up new ways
to automate mundane, standard business processes using robotic process
automation (RPA) technologies and data from connected devices [6]. RPA
is the automation of low-level business processes that are done repeatedly
(in which the operations are done swiftly and very financially beneficial).
The plan is to slowly replace human labor with systems that do the work for
us. RPA differs from other methods because it does not depend on digital
information. Because of this, there is not much violence [7] and a chance to
save money. It is expected that using RPA technology will cut the costs of
running transactional tasks in shared services by up to 60% [8].

3.11 The Digital Solutions Entangled in Industry 5.0

The economic and industrial outlook has rapidly changed in recent years. In
the past 10 years, this technology has improved extraordinarily. Applications
for Industry 5.0 are diverse and numerous [13]. It needs linked gadgets
that are actively functioning. One may describe Industry 5.0 as a highly
automated, united, and effective transformation. Industry 5.0 was character-
ized as “the digitalization of the industrial sector, with embedded sensors in
almost all individual elements and production equipment, omnipresent cyber-
physical systems, and analysis of all pertinent data” [14] by Wee et al. in
2015. Additionally, Industry 5.0 incorporates HCI, data, networking, data
and analytics, and business intelligence. Schmidt outlines a blend of tradi-
tional industrial methods and contemporary innovations that leads to clever
commercial product creation. Irrespective of geography, this method fosters
collaboration across many areas [15]. According to researchers, Industry
5.0 aspires to integrate all current industrial technology and digitalize data
78 Industry Automation: The Contributions of Artificial Intelligence

generation and processing methods. The definition of the data varies. Assim-
ilation is used in both the vertical and horizontal aspects of the product
lifecycle [16]. The results of modern techniques can also be divided into
product and service categories. The business model includes both the services
and advanced applications offered by all of these sectors, which power
Industry 5.0 [17].

3.12 AMS for Industry 5.0: Advanced Manufacturing

System
The economics and well-being of a nation are significantly impacted by the
industrial sector [18]. It also possesses connections to the industry’s numer-
ous other divisions, including transportation, management, IT operations, etc.
The Internet of Things (IoT) offers several services, including real-time data
monitoring and processing and various others [19]. Industrial IoT is now
considered to be the most revolutionary technique for Industry 5.0. Therefore,
the Internet of Things (IoT) offers a solution to many of the maintenance and
growth problems in today’s businesses [20]. Consequently, smart technology
in production is seen as a new paradigm that makes use of all the advantages
of contemporary technology and methodologies to address manufacturing
concerns that are largely addressed by Industry 5.0. It turns core businesses
into intelligent industries that are able to meet consumer expectations [21].

3.13 Methods, Data, and Results

By adding up the intelligent machine motions, the suggested work in this
chapter has provided the outcome of progress and frequency change. There
exists a significant need for AI across a wide range of applications, and these
needs are met by expanding to international functional hubs [22]. There are
approximately more than 300 new AI job openings from various firms that are
implementing AI algorithms in nearly 15–25 start-ups. The Mosaic algorithm
(6), which is based on factors like forensic cases, quality of information,
and technological considerations, is used to choose the list of various AI
casts from a pool of over 1500 enterprises. The annual growth rate and the
investment sum for various AI castings throughout India are displayed in
Table 3.3b [23]. According to a survey by AIM Research, start-ups in the
fields of artificial intelligence and data analytics earned $1108 million in
investment in 2021, the most in seven years and with a 32.5-yearly growth
rate. Three months into 2023, there have already been significant investments
 3.14 Conclusion 79

Figure 3.8 AI spending on various sectors of India.

Table 3.3a Amount invested in Indian start-ups in 2022.

S. No Indian start-ups Amount invested
1 Fractal $360 M
2 DataSutram $2.07 M
3 Atlan $50 M
4 [Link] $5 M
5 [Link] $3 M
6 Scribble data $2.2 M

Table 3.3b Amount invested in the last seven years in Indian start-ups.
S. No Year Amount invested
1 2016 $260 M
2 2017 $3.89 B
3 2018 $112.8 M
4 2019 $ 529.52 M
5 2020 $762.5 M
6 2021 $ 283 M
7 2022 $ 654 M***
∗∗∗
Expecting to increase

made in the intelligence and AI fields. Let us examine some of the significant
investments for 2022 in the form of Table 3.3a, and Table 3.3b exhibits the
last seven years of AI expenditure. Figure 3.8 displays the AI spending on
various sectors in India with respect to CAGR 2020−25.
80 Industry Automation: The Contributions of Artificial Intelligence

3.14 Conclusion
The advancement of numerous mathematical equations bolstered with tech-
nology that link up AI with actual data interpretation in Industry 5.0 systems
is urgently needed because AI is becoming a mandate and aiding sectors
in growing in accordance with Industry 5.0 parameters with the help of
their cutting-edge technology. In this chapter, we provide a summary of the
manufacturing industry’s AI ecological system, which is the foundation for
industry progress. In our effort, we have aimed to develop certain tactics
that will function with industry-based AI systems. Research articles high-
light how AI has dramatically impacted the industrial landscape in crucial
domains such as production loss reduction, process automation, and security
enhancement.
Industry automation is now the largest issue, yet the application of AI
has enabled companies to develop exponentially. We have observed a six-
fold increase in investment in AI and other related technologies over the
past several years, and we are certain that this trend will continue. Many AI-
based sectors have recently developed in startup companies as well. AI has
demonstrated a vital role in many different industries. Our upcoming study
will focus on the impact of AI on Industry 5.0 and potential development
opportunities.
Regardless of whether or not Industry 5.0 places a greater emphasis on
the collaboration of humans and robots, it will still have a significant impact
on the business world. In fact, it is highly probable that significant changes
for humanity will occur as a consequence of the integration of robots into
normal human activities. We are doing everything in our power to develop
machines that are remarkably similar to us in important respects. This will
invigorate and motivate certain individuals to a greater degree. Some people
will find it humorous, while others will find it annoying or even potentially
hazardous. The media actively disseminate this anti-robot sentiment across a
variety of different platforms.
The responses to our polls and surveys about how people feel about robots
in society will be skewed as a result of this criticism. A pessimistic point
of view was considered when developing one of the very first surveys that
looked into how people feel about robots. We cannot speculate on how the
general public will react to robots until they become a standard fixture in the
homes and workplaces of everyday people. People’s perceptions of robots
may shift as they gain more experience interacting with them. Children of
today will be exposed to this technology or one very similar to it at an
 References 81

earlier age than we were, and their reactions may be different from those
of us who grew up with it. The descendants of these robots will one day
become members of a society that is highly technologically advanced. We
must consider the differences between the generations if we are going to
build a society in which people can enjoy the benefits of technology while
minimizing the potential drawbacks it may bring and if we are going to avoid
becoming a society dominated by robots. In this chapter, we took a look at
some of the difficulties that might arise from the collaboration of humans
and robots in the workplace. Concerns pertaining to law and order, social
psychology, and morality are among the most urgent of these issues.

References
[1] 5G-ACIA. (2021, 03 04). 5G-ACIA. Retrieved from 5G Alliance for
Connected Industries and Automation: [Link]
[2] Aguirre, Santiago & Rodriguez, Alejandro. (2017). Automation of a
Business Process Using Robotic Process Automation (RPA): A Case
Study. 65-71. DOI: 10.1007/978-3-319-66963-2_7. Available from:
[3] Enríquez, J. G., Jiménez-Ramírez, A., Domínguez-Mayo, F. J., & García
García, J. A. (2020). Robotic Process Automation: A Scientific and
Industrial Systematic Mapping Study. IEEE Access, 8, 39113-39129.
[4] Bahrin, M. A. K., Othman, M. F., Azli, N. N., & Talib, M. F. (2016).
Industry 4.0: A review on industrial automation and robotics. Jurnal
Teknologi, 78(6-13), pp:137-143.
[5] Tripathi, A. (2018). Learning robotic process automation: Create soft-
ware robots and automate business processes with the leading RPA tool,
UiPath. Packt Publishing Book Series.
[6] E. Gökalp, U. S, ener, and P. E. Eren, “Development of an assessment
model for industry 4.0: industry 4.0-mm,;” in International Confer-
ence on Software Process Improvement and Capability Determination.
Springer, 2017, pp. 128-142.
[7] Akoramurthy, B., and T. Ananth Kumar. “10 Digital Linked.”
Blockchain Technology: Fundamentals, Applications, and Case Studies
(2020): 155.
[8] Delloite (2019). Automation with intelligence Reimagining the organi-
zation in the ’Age of With.’ Available from: [Link]
content/dam/Deloitte/tw/Documents/strategy/tw-Automation-with-inte
[Link]
82 Industry Automation: The Contributions of Artificial Intelligence

[9] Y. Qi, Z. Mao, M. Zhang, and H. Guo, “Manufacturing practices and

servitization: The role of mass customization and product innovation
capabilities,” International Journal of Production Economics, vol. 228,
p. 107747, 2020.
[10] Chatterjee, Jyotir Moy, P. Srinivas Kumar, Abhishek Kumar, and B. Bal-
amurugan. “Blockchain, Bitcoin, and the Internet of Things: Overview.”
Blockchain Technology and the Internet of Things (2020): 47-67.
[11] P. Chaturvedi, S. Dahiya and S. Agrawal, “Technological innovation:
A necessity for sustainable MSME sector in India,” 2015 Interna-
tional Conference on Futuristic Trends on Computational Analysis
and Knowledge Management (ABLAZE), 2015, pp. 206-211, doi:
10.1109/ABLAZE.2015.7154993. Authorized
[12] Demir, Kadir Alpaslan, Ebru Caymaz, and Meral Elci (2017) “Issues in
Integrating Robots into Organizations.”, The 12th International Scien-
tific Conference on Defense Resources Management in the 21st Century,
November 09-10, 2017, Brasov, Romania.
[13] G Bhardwaj, SV Singh, V Kumar . “An empirical study of artificial
intelligence and its impact on human resource functions,” International
Conference on Computation . . . , 2020
[14] Kadir Alpaslan Demir, Halil Cicibaş “Industry 5.0 and a Critique
of Industry 4.0.” 4th International Management Information Systems
Conference Istanbul, Turkey, 2017 (2017), pp. 17-20
[15] Haenlein, Michael & Kaplan, Andreas. (2019). A Brief History of
Artificial Intelligence: On the Past, Present, and Future of Artificial
Intelligence. California Management Review.
[16] Cormier, Derek, Gem Newman, Nakane, Masayuki, James E. Young,
and Stephane Durocher (2013) “Would you do as a robot commands? An
obedience study for human–robot interaction” International Conference
on Human-Agent Interaction.
[17] Moffitt, K. C., Rozario, A. M., & Vasarhelyi, M. A. (2018). Robotic
process automation for auditing. Journal of Emerging Technologies in
Accounting, 15(1), 1-10.
[18] Peter Sachsenmeier “Industry 5.0—The Relevance and Implications of
Bionics and Synthetic Biology.” Engineering., 2 (2016), pp. 225-229
[19] E. Marilungo, A. Papetti, M. Germani, M. Peruzzini, From PSS to CPS
design: a real industrial use case toward Industry 4.0, The 9th CIRP
IPSS Conference: Circular Perspectives on Product/Service-Systems,
Procedia CIRP 64 (2017) 357 – 362.
 References 83

[20] S. Hjorth, D. Chrysostomou Human–robot collaboration in indus-

trial environments: a literature review on non-destructive disas-
sembly Robot. Comput. Integrated Manuf., 73 (2022), p. 102208,
10.1016/[Link].2021.102208
[21] Haidegger, T. (2019). Autonomy for surgical robots: Concepts and
paradigms. IEEE Transactions on Medical Robotics and Bionics, 1(2),
65-76.
[22] A. Al-Yacoub, Y. C. Zhao, W. Eaton, Y. M. Goh, N. Lohse Improving
human–robot collaboration through Force/Torque based learning for
object manipulation Robot. Comput. Integrated Manuf., 69 (Jun. 2021),
p. 102111, 10.1016/[Link].2020.102111
[23] Dmytro Romanov, Olga Korostynska, Odd Ivar Lekang, Alex Mason,
Towards human–robot collaboration in meat processing: Challenges and
possibilities, Journal of Food Engineering, Volume 331,2022,111117,
ISSN 0260-8774, [Link]
[24] Laurent, H., Boer, E. D., “Fourth Industrial Revolution Beacons
of Technology and Innovation in Manufacturing” White Paper,
January 10, 2019. [Link]
industrialrevolution- beacons-of-technology-and innovation-in-
manufacturing
[25] R. Cioffi, M. Travaglioni, G. Piscitelli, A. Petrillo, and F. De Felice,
“Artificial intelligence and machine learning applications in smart pro-
duction: Progress, trends, and directions,” Sustainability, vol. 12, no. 2,
p. 492, 2020.
[26] Gordon Briggs, Matthias Scheutz “How Robots Can Affect Human
Behavior: Investigating The Effects of Robotic Displays of Protest and
Distress.” International Journal of Social Robotics., 6 (3) (2014), pp.
343-355 2014.
[27] Kadir Alpaslan Demir, Halil Cicibaş “The Next Industrial Revolution:
Industry 5.0 and Discussions on Industry 4.0.” Industry 4.0 From the
Management Information Systems Perspectives, Peter Lang Publishing
House (2018)
[28] Rupa CD, Hasan MK, Alhumyani H, Saeed RA (2021) Industry 5.0:
Ethereum blockchain technology-based DApp smart contract. Math
Biosci Eng 18(5):7010–7027
[29] Chowdhury MZ, Shahjalal M, Ahmed S, Jang YM (2020) 6G wireless
communication systems: applications, requirements, technologies, chal-
lenges, and research directions. IEEE Open J Commun Soc 1:957–975
84 Industry Automation: The Contributions of Artificial Intelligence

[30] Phuyal S, Bista D, Bista R (2020) Challenges, opportunities and future

directions of smart manufacturing: a state of the art review. Sustain
Futures 2:100023.
[31] Briggs, Gordon, and Matthias Scheutz.(2014) “How Robots Can Affect
Human Behavior: Investigating The Effects of Robotic Displays of
Protest and Distress.” International Journal of Social Robotics 6 (3),
2014: 343-355.
[32] Demir, Kadir Alpaslan, Ebru Caymaz, and Meral Elci (2017) “Issues in
Integrating Robots into Organizations.”, The 12th International Scien-
tific Conference on Defense Resources Management in the 21st Century,
November 09-10, 2017, Brasov, Romania.
[33] Bartlett, Brendan, Vladimir Estivill-Castro, and Stuart Seymon. (2004)
“Dogs or robots: why do children see them as robotic pets rather
than canine machines?” In Proceedings of the fifth conference on
Australasian user interface-Volume 28, pp. 7-14. Australian Computer
Society.
[34] Cormier, Derek, Gem Newman, Nakane, Masayuki, James E. Young,
and Stephane Durocher (2013) “Would you do as a robot commands? An
obedience study for human–robot interaction” International Conference
on Human-Agent Interaction.
[35] Briggs, Gordon, and Matthias Scheutz.(2014) “How Robots Can Affect
Human Behavior: Investigating The Effects of Robotic Displays of
Protest and Distress.” International Journal of Social Robotics 6 (3),
2014: 343-355.
[36] Demir, Kadir Alpaslan, and Halil Cicibaş (2018) “The Next Industrial
Revolution: Industry 5.0 and Discussions on Industry 4.0.” Industry 4.0
From the Management Information Systems Perspectives. Peter Lang
Publishing House.
[37] Cormier, Derek, Gem Newman, Nakane, Masayuki, James E. Young,
and Stephane Durocher (2013) “Would you do as a robot commands? An
obedience study for human–robot interaction” International Conference
on Human-Agent Interaction.
 4
Artificial Intelligence (AI) Driven
Industrial Automation

S. S. Blessy Trencia Lincy

Faculty of Mathematics and Data Science, Emirates Aviation University,

Dubai, The United Arab Emirates
E-mail: [Link]@[Link]

Abstract
Artificial Intelligence (AI) driven industrial automation outspreads the con-
temporary use of control technologies and robotics at much greater levels.
Industrial automation is the concept of using control systems, which include
computers, robots, and other information technologies, to manage diverse
processes and machinery in the industrial world to replace human beings.
This process involves integrating smart tools and computers into numerous
operations. Artificial intelligence is the capability of the machine to learn
from practice and experience, tune and acclimate to new ideas or inputs
without manual human intervention, or perform human-like tasks. AI has
played a vital role in research for over three decades. Hence, industrial
automation and AI are transmuting businesses and will subsidize economic
growth by contributing to productivity. AI-driven automation helps overcome
the inherent challenges and demands that have been afflicting industries.
This might include the scarcity of domain expertise, complexity in decision-
making, integration issues, and the overloaded information that needs to be
processed. The progress and advancement in machine learning technologies,
the ever-growing computing power, and advances in sensors have contributed
to creating a new generation of robots. AI aids and assists the machines
in collecting and extracting knowledge, identifying or recognizing patterns,
learning, and familiarizing with new situations through machine intelligence,

85
86 Artificial Intelligence (AI) Driven Industrial Automation

adaptive learning, and speech recognition. Thus, AI helps manufacturers

make faster and data-driven decision support, enhance the production out-
come, increase the efficiency of the processes involved, reduce the costs
involved in the operations, and support the innovation of the products.

Keywords: artificial intelligence, computer vision, industry automation

4.1 Introduction
In today’s ever-growing digitized economy, researchers and experts should
concentrate on creating and adapting the right solutions and platforms that
revise and enhance business outcomes and strive toward success. The influ-
ence of AI might be remarkable, with businesses extending from retail,
healthcare, product sales involving supply chain, finance, engineering and
manufacturing, and services, all projects potentially varying and radically
changing by the inception of AI technologies [1].
Industrial automation with artificial intelligence can be shown based
on seven dimensions: objects, specific domain area, stages of application,
requirements of the application, intelligent technologies, intelligent func-
tions, roles, and solutions of industrial automation and artificial intelligence
(I-A). When integrated with the business aspects, these dimensions are
considered a breakthrough in artificial intelligence technologies [2].
Nevertheless, deploying AI strategies and solutions for industrial automa-
tion has been difficult and is a more significant challenge. While comparing
the traditional computer vision (CV), AI and ML practices are still new in
industrial automation. The automation engineers and experts in the manu-
facturing field do not yet have the skills or expertise to develop effective
algorithms for AI. Numerous AI technology establishments are eradicating
these hurdles by providing a complete inference strategy in a small form,
i.e., high-performing and low-power hardware with trained, ready-to-deploy,
tuned AI algorithms to improve efficiency and workplace safety.
Developments in the-AI processing have highlighted the means for
today’s AI technologies and will open new opportunities and prospects for
techniques in the future. The automated systems have to process an implau-
sible amount of information and will make decisions in real time; so it
is much more competent for these machines to process information at the
individual end instead of sending them to the cloud and so forth. The need for
automation grows drastically daily, and as a result, industries continue to cul-
tivate AI-powered techniques to enhance the productivity and performance of
 4.3 Industry 4.0 Technologies 87

business activity. This has led to even more innovative techniques and appli-
cations in today’s industrial world, from smart anomaly detection systems
to autonomous robotics. This chapter discusses and analyzes various tech-
nologies and platforms that contribute to industrial automation and artificial
intelligence’s impact on industrial automation.

4.2 Evolution of Artificial Intelligence

Artificial intelligence can be characterized into three elementary stages based
on the growth and progress of the technologies [3].
• Basic AI: The basic AI involves a system with only one functional area.
AlphaGo, a board game computer program, is an example.
• Advanced AI: It involves a system with additional fields, such as the
influence of abstract rational thinking, reasoning, or solving problems
with human beings.
• Autonomous AI: It involves systems across all fields, which is the last
stage of AI. This phase of AI is not likely to be established entirely for
several decades.

4.3 Industry 4.0 Technologies

Industrial development utilizing stability, reliability, growth, dynamic
response, technical advancement, selection, and strengthening, irrespective
of the diverse organizations, is indispensable. Various organizations and
governmental bodies are investing billions of dollars in the technologies that
contribute to these factors to harvest, to meet the changing demand and
supply patterns, rising demand, efficiency, optimization, and planning of their
resources.
Table 4.1 shows the evolution of the industry technologies [4]: (1) the first
revolution with the technologies for waterpower, steam power, mechanical
manufacturing, etc., evolved in the year 1784, referred to as Industry 1.0; (2)
the second revolution with the assembly line manufacturing, electrical energy,
mass production technologies, etc., progressed around the year 1870, called
Industry 2.0;
(3) the third revolution evolved in 1969 with technologies that include
computers, electronics, robotics, industrial automation, etc.; (4) the fourth
revolution refers to the current and emerging technologies, including infor-
mation exchange, cloud computing, IoT, digitization, etc.
88 Artificial Intelligence (AI) Driven Industrial Automation

Table 4.1 Industry revolution.

Industry Year Revolution Description
Industry 1.0 1784 First revolution Waterpower, steam power, and mechan-
ical manufacturing
Industry 2.0 1870 Second Assembly line manufacturing, electrical
revolution energy, and mass production
Industry 3.0 1969 Third Computers, electronics, robotics, and
revolution industrial automation
Industry 4.0 Present Fourth Information exchange, cloud comput-
revolution ing, IoT, and digitization

In this situation, AI is thriving to play a significant role in the current

market. Distinguishing and identifying the importance of AI, with diverse
technologies, schemes, and their different applications, the chief drivers
are the four required methods used in contemporary AI methodologies,
including (a) artificial neural networks, (b) fuzzy logic systems, (c) expert
systems, and (d) genetic algorithms. Various industries in developed countries
have begun using AI to connect with smart grids, meters, and IoT devices.
Therefore, these techniques in AI will prime the practical usage and man-
agement of resources, improvement of efficiency, and many more factors. In
recent years, emerging technologies in AI have advanced system devices and
communications mean within the systems in unprecedented ways [4].

4.4 Development in AI
The unceasing evolution of intelligent smart cities, medical systems care,
intelligent transport carriage units, intelligent robots, self-driving automated
vehicles, intelligent toys, intelligent logistics, smart economies, smartphones,
and smart communities, to tag but a few, affords a total market demand and
lashing force in terms of innovative development of both AI technologies and
applications [5].
Intelligent industrial manufacturing is a novel manufacturing model and
the methodological means where many technologies are being integrated,
which includes intelligent science and technology, systems engineering tech-
niques, information and communication technology, massive manufacturing
technologies (from the various phases of design, production/development,
maintenance/management, testing, and integration), and associated product
technologies throughout the entire lifecycle of the product development. The
process involved in the manufacturing/industrial lifecycle uses self-directed
 4.4 Development in AI 89

sensing devices, autonomous collaboration, interconnection, perception,

investigation, knowledge, control, and decision-making based on the learn-
ing and the implementation of human, material, machines mechanism, and
ecological information to support and aid the optimization and integration of
diverse characteristics of industrial, manufacturing or engineering organiza-
tions, other groups, and enterprises. This includes the people, management
involving the operations, equipment, or technologies, and the various flows
concerning the information, capital, logistics, knowledge, and service. This
shortens the manufacture and provides a high-quality, high-efficiency, and
profitable rewarding system, and better setting approachable facility and
provision for operators. It consequently progresses the market affordability
and competitiveness of the industrial initiative or group, irrespective of the
product involved. The role of AI tools and technologies in smart manufactur-
ing is an emerging topic. The revolution of AI is in its infancy, and numerous
companies have noteworthy activities on-going. Currently, both small and
big devices deployed in the industries are equipped/provided with sensors
that can collect and share huge volumes of data and capture a multitude of
activities. Manufacturers have started identifying the tactical importance of
big data analytics, and, therefore, to enhance manufacturing competitiveness,
the data is becoming a key enabler [6].
Visualizing automation as a batch of equipment/machines/robots imple-
menting predefined processes and operations can be rendered to a fixed set
of constraints and rules in a limited number of situations. Nevertheless, AI-
based techniques and machines do not just follow the rules; they perceptively
and intelligently comprehend and identify patterns within the data; they learn
from previous experiences and advance imminent future performance.
Table 4.2 describes the different machine learning and artificial intel-
ligence techniques that can be applied to different tasks like classifica-
tion, regression, clustering, dimensionality reduction, decision-making, and
other miscellaneous tasks with respect to the application [6]. Consequently,

Table 4.2 ML and AI techniques.

ML and AI techniques Tasks Approaches
Supervised learning Classification, Regression Support vector machine,
logistic regression
Unsupervised learning Clustering, k-Means
dimensionality reduction
Reinforcement learning Decision-making Q-learning
Miscellaneous Miscellaneous Backpropagation
90 Artificial Intelligence (AI) Driven Industrial Automation

AI empowers the same machinery, equipment, and techniques to solve

multifaceted issues within a specified solution space.
Figure 4.1 involves the most frequently used algorithms in machine
learning and artificial intelligence in industrial applications.
While this idea is all-encompassing, business leaders, scientists, and engi-
neers worldwide are facing a few issues deterring the real-time deployment
of AI, especially in the manufacturing industries. This may be due to the
below-mentioned reasons [6].
1) there is no adequate evidence that the industry will succeed in embracing
technologies in AI;
2) deficiency of a systematic approach for its deployment for a variety of
industrial applications;

Figure 4.1 ML and data scientists.

 4.6 Digital Transformation 91

3) absence of consistent and structured data from equipment and machines

as the data is collected and loaded in different formats;
4) absence of data related to machine failure since the industries hardly
tolerate running their failed machines;
5) involvement and intervention of the humans in working the dynamic
application contexts to augment common sense to validate and verify
the results.

4.5 AI Future Perception

Until recently, automation has primarily affected low-skilled routine work-
ers and tasks. Furthermore, AI may automate non-routine, intellectual, and
cognitive tasks involving highly skilled workers. An advantage of this per-
ception is that it lets us use previous historical knowledge and outcome to
enlighten us about the probable future effects of AI. The several benefits of
AI, including direct automation, improved forecasting of the demand [7], a
better workplace, smarter workforce, sophisticated root cause event analysis,
easier 24/7 delivery of service/manufacture, enhanced profit, amplified work-
place safety, reduced costs involved in operations, enhanced identification
of defects, automation of quality control, advanced and efficient supply
chain, improved product design or equipment design, rapid decision-making,
predictive maintenance, quality optimization, testing, etc.

4.6 Digital Transformation

To successfully adapt to digital transformation, businesses must recognize
the abilities and competencies of these AI methods and technologies and
how they can influence the existing environment and ecosystem. Neverthe-
less, many establishments do not have an inclusive expertise roadmap and
background about these technologies and how they could entrench AI in their
tactics and businesses. Henceforth, they are yet to see evocative bottom-line
profits from its implementation, and thereby most of its economic impression
is yet to come. They appear to be trapped in “pilot purgatory” and cannot
thrive in developing systematic capabilities. To speculate how AI can impact
digital transformation based on economic growth, the following research
queries discuss these aspects [8].
• How will AI impact economic growth if they enhance automation in
producing goods and services?
92 Artificial Intelligence (AI) Driven Industrial Automation

• Can we relate, narrate, and understand AI’s progression and the observed
business growth over the twentieth century?
• Can we assume this growth to be persistent in the twentieth century?
• Whether applying AI and automation to production contribute to new
ideas?
• Whether AI leads to a rapid increase in economic growth as data scien-
tists and researchers predicted, and under what circumstances? Are these
circumstances reasonable?
• How is the interconnection between the organization’s growth and AI
modulated?
• Whether AI affects internal industries or organizations, and with what
inferences?
These queries proximately suggest essential economic issues. For example,
what will happen if AI technologies allow a forever-growing number of tasks
formerly performed by human labor to become automated? The AI may be
organized and deployed in making or manufacturing goods and services,
possibly affecting the financial development and the organization’s income
stocks. Nonetheless, AI may also revolutionize the process and procedure
by which we produce new novel ideas, designs, notions, and technologies,
unraveling complex, intricate problems and scaling creative, inventive efforts.
This also suggests that the definition of AI, i.e., the machine’s ability to
perform tasks in a wide range of environments [8].

4.7 Components of AI in Automation

• Machine vision: Machine vision denotes any program’s competence
to understand the pictorial input. The machine exploits the training
data as a type of groundwork for the classification or the mechanism
for identification. For example, iPhone’s face recognition system uses
machine vision technology.
• Natural language processing (NLP): As the machine language depends
on the illustrations and visuals, the NLP does the same to understand the
input texts and the human voice. It is now probable for technologies to
understand the background behind the message being passed out and
then take action depending on the kind of data prebuilt and related
variables that are vital. A few examples of this include Amazon’s Alexa,
Apple’s Siri, etc.
• Machine learning: It refers to the ability of a machine to learn
using the data provided to it. This encompasses the consequences of
 4.8 Artificial Intelligence Applications in Automation 93

environmental attributes/variables and decisions to advance itself. It will

be possible to progress the total competence of existing solutions using
machine learning approaches and techniques. Consider the example of
an intelligent automation system to resolve an issue without human
intervention. The intelligent system will vary automatically and follow
the set of procedures or measures that a human uses. Consequently, over
time, the effort of humans will be reduced, and the system’s competence
will improve [9].

4.8 Artificial Intelligence Applications in Automation

Figure 4.2 shows the various applications or domains where artificial intel-
ligence technologies can be adapted in numerous ways in automation.
From drones to self-driving automated cars, inventory management, sales
and marketing, planning and decision-making services, asset forecast and
maintenance, digital development, quality control, customer relationship
management (CRM), optimization of machines, and software tools, all are
using intelligent automation.
Smart retail business: AI empowers the retail and industrial/manufacturing
industries in making appropriate decisions/choices with precise and real-
time prediction, refining management of the supply chain, defining impactful
upgrades, and improving the pricing strategy. AI is also making the operation
of the organization and processing more competent because of the automated
robotics technologies and optimization of processes, which advances produc-
tion and decreases the labor-intensive physical costs. The usage of interacting
custom robots in the warehouse and stores are well recognized. More pre-
vailing computers, novel or hybrid algorithmic models, and hefty training
datasets empower the development of heightened vision. Privileged within-
object recognition, computer vision, and semantic segmentation, which is the
capability to classify object types, such as differentiating a device from an
element or module, have lately progressed suggestively in their performance.
Enhanced satisfactory customer service: A good example of this is chat-
bots, which became prevalent and widespread in a very short period. It
started with Apple’s Siri and then has been implemented by most brands
now. The chatbots can understand the context-based user’s input and then
retort to those queries appropriately. Automating sales, customer-friendly
services, and marketing message communications make use of these bots.
The customer help desk services can be replaced with bots, which can feel
human and reduce manual physical burdens.
94 Artificial Intelligence (AI) Driven Industrial Automation

Figure 4.2 AI automation applications.

Market brand management: The brand management process and tasks

can be made a lot easier with the aid of automation. In order to completely
understand the consumer’s opinion, brand marketers struggle a lot, even with
their brand. Consequently, they will be capable of automating the analysis
of the content across the Internet with the aid of automation. In addition,
they can help in recognizing critical issues daily. Watson Analytics for Social
Media is an appropriate example of such automation. The user will define a
set of words and their context and try to understand what the online users
relate to in a specific duration.
Software development and testing: Automated software testing is an
emerging field in this current era. With a wide variety of tools that are
evolving today, the testing process can likely be fully automated in the future.
SauceLabs and ReTest are examples of a few prevalent testing tools available.
By using these tools, testers and developers can concentrate on performing
 4.8 Artificial Intelligence Applications in Automation 95

core testing and ignore the worries of fixing bugs in intelligent systems.
Development automation is a long way to go, but these tools can surely aid
developers in performing tedious tasks.
Thwarting fraud: The face of the perpetrator can be connected directly
to the theft using AI. The POS system with a camera attached will record
and save all types of transactions and then can be directly linked to the
face with the base details already saved within the system. For example, if
someone commits credit card fraud, it will be much easier to put them behind
bars. Furthermore, using an intelligent system will be capable of averting
cyberattacks by quickly recognizing abnormal or irregular behavior from the
user. In such particular situations, the system can automatically cease taking
any further requests and fire off an alert to the administrator.
Improving efficiency: People make mistakes irrespective of their experi-
ence level or job type. However, with automation, errors and mistakes can
be reduced greatly. Furthermore, with time, the machines can learn from the
outcome of the processes and improve the system’s efficiency in the future.
Human resource management: The recruitment team struggles a lot in
arranging and sorting the candidate resumes they receive. With the advance-
ment in automation, it will be easy to shortlist potential candidates based on
the data previously collected. Usually, automated applicant tracking systems
are used to receive the CVS. The automation solution is provided using these
collected materials when the users apply for different roles, which will be
stored in the database.
Reducing cost: The cost involved in training a human for a particular task
will be recurring. This includes the employee turnover rate, the time needed
for the individual to learn the task gradually, experience, and develop the
skill may vary greatly. However, a machine can be trained only once and,
over time, might improve its performance without incurring more cost and
repeated training.
Smart manufacturing: The usage of AI technologies has altered the
manufacturing business sectors, from employing virtual assistants to uncon-
ventional automated robotics, which has empowered engineering businesses
to produce more products or services with reduced faults to meet claims.
Using AI, which facilitated swift progress as they can condense the growth
processes and cycles, the efficiency of progress engineering, thwart faults,
upsurge security by systematizing hazardous actions and events, decrease
inventory expenses with improved supply and forecasting of demand,
and upsurge the revenue with improved identification of sales lead and
optimization of price, etc. This new concept is intelligent manufacturing,
96 Artificial Intelligence (AI) Driven Industrial Automation

a clever method of manufacturing where machines are associated with

humans, i.e., both humans and machines work together with negligible
supervision.
Moreover, we are starting to see autonomous systems that can accomplish
tasks without human participation, as the system can train itself and fine-tune
to new training data. Consider the automated financial trading application.
Since it depends utterly on algorithms, businesses can complete business
transactions much quicker with autonomous technologies than with systems
relying on humans. Similarly, robots are accomplishing narrow tasks and
responsibilities autonomously in industrial settings [10].
Some organizations/businesses, such as Google and Amazon, have
endeavored to produce highly determined applications of AI, including unat-
tended checkout counters, drone delivery, autonomous vehicles, etc. Some of
these applications have been effective, but some highly determined projects,
like cancer treatment, have been mostly unsuccessful despite substantial
expenditures and outlays. Less determined projects have been more success-
ful in most organizations and are perhaps more reliable and dependable with
the narrow intelligence influenced by AI systems at the moment. Similarly,
most autonomous AI applications endure limited to low-risk extents where
the cost of catastrophe is limited. Even though many AI techniques can
do few things better than humans, the workers’ faith in AI technology is
still inadequate due to the matters involving such technology, which might
increase the algorithmic biases, unexplainable consequences, conquered pri-
vacy discretion, and/or lack of credibility or accountability. Customers are
also skeptical about AI, and reviews suggest that utmost or many would
not need autonomous vehicles, do not involve dealing with chatbots, and so
forth.
There is a growing need for high-quality and trustworthy AI schemes
since they work meticulously with human beings. It is consequently vital
to deliver a vibrant path for understanding and handling the complications
intrinsic to highly eminent AI systems [11]. The datasets gathered from
heterogeneous resources and information technology (IT) systems can be
used in the data-driven analytics actions to provision additional well-versed
business intelligence support and decisions. Nevertheless, these outcomes
are presently used in remote and discrete portions of the manufacturing or
manufacture process. Meanwhile, at the same time, a complete combina-
tion of artificial intelligence in all facets of industrial systems is presently
deficient [12].
 4.9 Automation and AI 97

4.9 Automation and AI

When robotic method automation relates to elements of AI like machine
learning, the outcome is recognized as intelligent process automation (IPA).
An IPA technique is powerful since it consensuses us to get both the benefits
of automation, which include amplified speed, time savings, efficiency, and
capability to measure along with the AI’s elasticity, visions, and processing
power. Vendors who use IPA can expand their support and skills while off-
loading uninteresting operating management errands to the machine. This is
unlike pure robotic automation in that the AI techniques can begin, halt, or
even transform whatever it undertakes depending on the situation in which
it functions. Moreover, since the finest AI systems permit the vendors to set
railings, there is no chance of unanticipated actions taking consequences too
far awry. For vendors, this means preferably sooner, more improved imple-
mentation and processes, greater use and exactness in data, and enhancements
in overall client experience and satisfaction. Marketers move from unsettling
over-bid alterations or changes and budget deliveries to sophisticated higher
value-added, human-centric contributions like, “How do we raise our value
intent to initiate more business and production?” As of these clear profits,
Forrester stated that in 2021, 25% of Fortune 510 establishments reported
hundreds of illustrations of IPA use cases. Conceivably up to 2050, we
entirely hinge on automation and AI since it makes life easier and more
precise [13].
Automation and artificial intelligence is the invention of science. The
impression that the diverse types of machinery could anticipate and accom-
plish tasks and processes just as humans or individuals do is thousands
of years old. The knowledgeable truths articulated in AI and automation
systems and technologies are not new moreover. It may be better to interpret
these skills as the operation or procedures of powerful and long-recognized
knowledgeable values through engineering. The practice of automation and
intelligence upsurges day by day almost all over the world. Currently, people
rely entirely upon them. These technologies and methods are time-saving and
can perform numerous tasks with more correctness at a time that humans
cannot perform or do [13].
Innovation and adaptation are enormously imperative to the engineering
industry. This advancement should prime to supportable engineering using
novel technologies. To endorse smart delivery, sustainability, and production
necessitates global perceptions of smart application manufacturing technol-
ogy. In this respect, thanks to rigorous study and analysis efforts in the
98 Artificial Intelligence (AI) Driven Industrial Automation

arena of AI, a wide range of AI-based methods and techniques, including

machine learning, have already been familiar in the business industry to
accomplish sustainable engineering. Thus, AI subfields, including image
processing, machine learning, data mining, and natural language processing,
have also become a significant matter for today’s tech titans. The focus on
AI creates substantial attention and curiosity in the scientific community,
featuring assets of the unceasing progression of the technologies presented
today. The variety of AI techniques has grown immensely; meanwhile, the
intelligence of machines with machine learning competencies has fashioned
deep impacts on corporate businesses, civilization, and governments. They
also impact the greater trends in worldwide sustainability. AI can be ben-
eficial in unraveling perilous issues for sustainable business (e.g., logistics,
waste management, optimization of resources involving energy, management
of supply chain, etc.). With this background, in smart manufacturing, there is
an inclination to integrate AI into green industrial manufacturing processes
for firmer conservational policies [14]. The hasty progress and fusion of new
AI methods/technologies with the Internet, energy, materials, new-generation
information technologies, and biotechnology is a crucial part of this new
epoch, which in turn will empower the game-changing revolution of models
and environments in relation to their application.

4.10 Conclusion
AI technology is presently empowering substantial digital transformations
that are redefining what an organization does and even obscuring the indus-
try’s limitations and boundaries. Various traditional manufacturing and indus-
trial organizations are taking advantage of machine learning technologies to
transform their emphasis from manufacturing goods to delivering services.
Different human skills will be rendered obsolete with the AI significantly
influencing diverse occupations by automating mundane tasks. The effect of
AI-enabled automated technologies varies from the prior technologies, given
that they can accomplish tasks that previously mandated human judgment,
as they get to distress the knowledge workers for the first time. Most of
the professionals, like architects, consultants, lawyers, and doctors, whose
creativity, judgment, and expertise have thus been exceedingly valued and
considered inimitable, appear threatened now. The changing nature and way
of their work is already a realism.
 References 99

Acknowledgments
I thank the Emirates Aviation University for supporting me in performing this
analysis and study.

References
[1] Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C.,
Crick, T., & Williams, M. D. (2021). Artificial Intelligence (AI):
Multidisciplinary perspectives on emerging challenges, opportunities,
and agenda for research, practice and policy. International Journal of
Information Management, 57, 101994.
[2] Zhang, X., Ming, X., Liu, Z., Yin, D., Chen, Z., & Chang, Y. (2019).
A reference framework and overall industrial artificial intelligence (I-
AI) planning for new application scenarios. The International Journal
of Advanced Manufacturing Technology, 101(9), 2367-2389.
[3] Mou, X. (2019). Artificial intelligence: investment trends and selected
industry uses. International Finance Corporation, 8.
[4] Ahmad, T., Zhu, H., Zhang, D., Tariq, R., Bassam, A., Ullah, F., &
Alshamrani, S. S. (2022). Energetics Systems and artificial intelligence:
Applications of Industry 4.0. Energy Reports, 8, 334-361.
[5] Li, B. H., Hou, B. C., Yu, W. T., Lu, X. B., & Yang, C. W. (2017).
Applications of artificial intelligence in intelligent manufacturing: a
review. Frontiers of Information Technology & Electronic Engineer-
ing, 18(1), 86-96.
[6] Lee, J., Singh, J., & Azamfar, M. (2019). Industrial artificial intelli-
gence. arXiv preprint arXiv:1908.02150.
[7] Bedi, J., & Toshniwal, D. (2019). Deep learning framework to forecast
electricity demand. Applied energy, 238, 1312-1326.
[8] Aghion, P., Jones, B. F., & Jones, C. I. (2018). Artificial intelligence
and economic growth. In The economics of artificial intelligence: An
agenda (pp. 237-282). University of Chicago Press.
[9] Shekhar, S. S. (2019). Artificial intelligence in automation. Artificial
Intelligence, 3085(06), 14-17.
[10] Benbya, H., Davenport, T. H., & Pachidi, S. (2020). Artificial intel-
ligence in organizations: current state and future opportunities. MIS
Quarterly Executive, 19(4).
100 Artificial Intelligence (AI) Driven Industrial Automation

[11] Hamada, K., Ishikawa, F., Masuda, S., Myojin, T., Nishi, Y., Ogawa,
H., & Matsuya, M. (2020, July). Guidelines for Quality Assurance of
Machine Learning-based Artificial Intelligence. In SEKE (pp. 335-341).
[12] Trakadas, P., Simoens, P., Gkonis, P., Sarakis, L., Angelopoulos, A.,
Ramallo-González, A. P., ... & Karkazis, P. (2020). An artificial
intelligence-based collaboration approach in industrial iot manufac-
turing: Key concepts, architectural extensions and potential applica-
tions. Sensors, 20(19), 5480.
[13] Donepudi, P. K. (2018). Application of artificial intelligence in the
automation industry. Asian Journal of Applied Science and Engineer-
ing, 7(1), 7-20.
[14] Cioffi, R., Travaglioni, M., Piscitelli, G., Petrillo, A., & De Felice,
F. (2020). Artificial intelligence and machine learning applications in
smart production: Progress, trends, and directions. Sustainability, 12(2),
492.
 5
Quantum Machine and Deep Learning
Models for Industry Automation

Pawan Whig1 , Arun Velu2 , Rahul Reddy Nadikattu3 ,

and Yusuf Jibrin Alkali4
1 Vivekananda Institute of Professional Studies-TC, India
2 Equifax, USA
3 Department of IT, University of Cumbersome, USA
4 Federal Inland Revenue Service, Nigeria

Abstract
Quantum machine learning unites two of the most contentious subfields
in science: quantum computing and classical machine learning. Quantum
machine learning is the study of how to solve machine learning problems
using insights from quantum mechanics. As more and more information accu-
mulates, the capacity of existing computing machines to successfully train
it onto a conventional computation model is stretched to its limits. Quantum
computing could be helpful for continuous training with large datasets. When
developing learning algorithms, quantum machine learning aims to do so
much more quickly than traditional methods. Finding patterns in data and
using those patterns to predict the future is at the heart of traditional machine
learning. The data can be mined for insights like these. Quantum systems, on
either extreme, generate unusual patterns those conventional systems cannot,
implying that quantum computers might be able to outpace traditional CPUs
in ML tasks. This chapter discusses the past research on quantum machine
learning and an update on its current state.

Keywords: Quantum, machine learning, conventional system

101
102 Quantum Machine and Deep Learning Models for Industry Automation

5.1 Introduction
Understanding which uses for these technologies may use their potential is
crucial as quantum technologies develop quickly. Likewise, machine learning
on traditional computers has advanced significantly, revolutionizing tasks like
picture identification, language translation, and even physics applications
(Whig et al., 2022). This is because additional processing power enables
ever-improving performance. Therefore, the potential for effect is immense
if quantum computers could speed up machine learning (Hussein, 2019).
The quantum improvement of machine learning has at least two poten-
tial directions. Quantum applications first inspired them in optimization;
quantum computing’s capabilities might theoretically be utilized to speed
up the training of current classical models or improve inference in graphical
models, as shown in Figure 5.1. This can entail locating better optima in a
training environment or locating optima using fewer queries. The benefit of
these appearances may only be limited to modest multifaceted accelerations
without a deeper understanding of the issue’s underlying structure (Anand
et al., 2022).
Quantum computing employs devices that use quantum physical science
features to stock data and execute calculations. This may benefit some jobs
where they can considerably beat even our most powerful supercomputers.
In Montiel-Ross (2020), the authors analyzed which subatomic inhabitants
metaheuristics may be converted to be employed in current quantum comput-
ers based on the circuit model programming paradigm. Montiel-Ross et al.,
(2019) also proposed the quantum-inspired acronymic evolutionary algorithm
as an optimal optimization algorithm for complicated systems.
Employing quantum replicas to produce associations between variable
quantities impractical to describe using traditional computing is of particular
interest. The recent theoretical and practical advancements in achieving quan-
tum multiplications beyond classical controllability could be construed as
evidence that quantum computers are capable of transitioning from delivering
examples from increasingly difficult prospects to examples from more typical
ones. If these provisions matched real-world distributions, it would indicate
the possibility of a significant benefit (Jiwani et al., 2021).
In current research on quantum neural networks, which aim to charac-
terize an allocation through a set of tunable parameters, as well as quantum
kernel methods, which employ quantum mechanics to describe extracted fea-
tures that map conventional information into the quantum Hilbert space, this
type of advantage is frequently sought after (Alkali et al., 2022). The rationale
 5.2 Difference Between Classical and Quantum Data 103

Figure 5.1 Relation between quantum computing and machine learning.

for these techniques’ ability to outperform classical models frequently uses

the same arguments as or outcomes from quantum simulations. For instance,
there may be a quantum benefit if the model uses a quantum circuitry that is
challenging to sample data conventionally (George et al., 2021).
The distinction between a conventional computer and a quantum com-
puter was explained at a conference in 2017 by Microsoft CEO Satya Nadella
using the example of a corn maze. A traditional computer would start along
a path, run into an obstacle, and then turn around. It would then start again,
run into another obstacle, and so on until it ran out of alternatives. Although
a solution can be reached, this method could take a long time (Bhargav &
Whig, 2021).
Quantum computers, however, unlock remarkable parallelism. They
simultaneously travel every route through the cornfield. Consequently, the
number of steps needed to solve an issue is exponentially reduced (Khera
et al., 2021).
104 Quantum Machine and Deep Learning Models for Industry Automation

5.2 Difference Between Classical and Quantum Data

Figure 5.2 Classical data vs. quantum data.

5.3 Quantum Computing

The tiniest conceivable unit of any physical substance, such as energy or
mass, is called a quantum. Max Planck hypothesized in 1900 that a body’s
energy is stored in discrete packets known as quanta at the atomic and
subatomic levels. Quantic atoms have pulse dualism, which causes them to
behave occasionally like waves and occasionally like particles, depending on
the circumstances (Velu & Whig, 2021a). A key aspect of quantum theory is
finding a particle’s likelihood at a particular point x in space rather than its
precise position.
 5.3 Quantum Computing 105

5.3.1 Qubit
A computer conducts operations using traditional “bits,” which are either 0 or
1. Nevertheless, a quantum computer operates on bits, often called “qubits.”
Qubits can be modeled as follows:
While |1 and |0 are the exciton and grounded phase of electrons circling
a nuclei, correspondingly.
A photon: The polarizing filters are |1 and |0, as shown in Figure 5.3.

5.3.2 Superposition
Qubits are simultaneously both 0 AND 1. The term “superposition” refers to
this phenomenon.
However, particles can exist in numerous quantum states; when their
energies or location are measured, the combination is gone, and the atom
only exists inside one state.

5.3.3 Entanglement
The phenomenon known as “quantum entanglement” occurs when quantum
particles interact and are characterized in terms of one another rather than
separately, even though the particles are spaced far apart (Whig & Ahmad,
2019).
During the measurement of a pair of entangled particles, if one of the
particles is set to the lowest energy state (the spin state of “down” when the
electron is aligned with its magnetic field), the other correlated particles will

Figure 5.3 Classical bit and qubit.

106 Quantum Machine and Deep Learning Models for Industry Automation

take on the opposite spin state (the spin state of “up”) in response. This will
happen if the other particle is set to the lowest energy state. When two or
more particles interact with one another in a way that causes them to become
entangled, the properties of those particles become inextricably intertwined,
making it impossible to measure each of them separately. Through the pro-
cess of quantum entanglement, it is possible to have instant communication
between physically separated qubits.

5.4 Quantum Machine Learning (QML)

Recently, a theoretical field known as quantum machine learning has started
to show signs of taking shape. It can be found at the intersection of deep
learning and particle physics. Quantum machine learning’s primary goal is
to accelerate processes by integrating. Come again; we distinguish quantum
computation from ML. Aspects of standard ML theory are included in the
theory of quantum machine learning, which approaches quantum computing
through that framework.
Suppose the straightforward problem of identifying whether a quantity
is uniform or abnormal in order to contrast traditional program design,
traditional ML, and quantum ML, as shown in Figure 5.4.

Figure 5.4 Quantum computing vs. machine learning.

 5.5 Classical Machine Learning vs. Quantum Computing 107

The answer is straightforward: you need to ask the user for a number, and
then split it in half. The number is odd if you receive a remainder. The number
is even if there is not any leftover (Rupani & Sujediya, 2016).
A similar concept underlies quantum computing. Data is processed at
the bit level in a traditional computer. Quantum physics, in the context of
quantum computers, is the specific behavior that controls the system. The
interaction between several atoms may be described using several methods
from the field of quantum physics. Such atoms are known as “qubits” in
the context of quantum computers. A qubit functions as both a wave and a
particle. A wave dispersion may hold a lot more information (or bit) relative
to a particle.
Algorithms are used to monitor a machine learning optimizer’s accuracy.
We frequently notice that not all of the predictions by a machine learning
model are accurate when building it. The output of the mathematical expres-
sion used to describe the loss function reveals how much the program has
exceeded the desired outcome (Nadikattu et al., 2020).
The loss function is another goal of a quantum computer. It contains a
technology called quantum tunneling that scans the whole loss function space
in search of the value with the smallest loss, which determines where the
method will operate most effectively and quickly.

5.5 Classical Machine Learning vs. Quantum Computing

Quantum computers employ several approaches to handle machine learning
challenges now that you are familiar with the fundamental ideas of quantum
computing, as shown in Figure 5.5. The following is a list of the approaches
we will examine:
• Algebraic problems solved via quantum computer vision
• Confirmatory factor analysis in quantum
• Kernel techniques and quantum support vector machines
• Precision optimization
• Quantum thinking in depth

5.5.1 Linear algebra problems have been solved via quantum

machine learning
Matrix operations on variables in an ample vector space can be used to solve
a wide range of issues arising in data analysis and machine learning. An
interplanetary trajectory within a complex sequence of dimensions, denoted
108 Quantum Machine and Deep Learning Models for Industry Automation

Figure 5.5 Classical machine learning vs. quantum machine learning.

as 2a, symbolizes the classical state of a quantum system in a quantum

computer. A planetary orbit trajectory describes this pattern. Several matrices
undergo different transformations inside this domain. The Fourier transform,
eigenvector and eigenvalue identification, and the solution of linear systems
of equations over two-dimensional trajectory places are all examples of
problems that could be solved in polynomial time by quantum systems. One
example is the Harrow, Hassidim, and Lloyd (HHL) method.

5.6 Quantum Thinking in Depth

The dimensions of large datasets can be simplified with the help of a tech-
nique known as principal component analysis. Whenever we process data,
we have to decide which factors to keep and which to eliminate, and in doing
so, we sacrifice accuracy. Working with a dataset can become noticeably less
complicated when correctly carried out, making the machine learning process
significantly less taxing.
 5.6 Quantum Thinking in Depth 109

5.6.1 Principal component analysis in quantum

For instance, when given a dataset with 10 input values, a standard system
can perform principal component analysis (PCA) in a reasonable amount of
time. However, if the input dataset has a million characteristics, traditional
principal component analysis methods will not work because of the difficulty
in displaying the relative value of each feature. Principal component anal-
ysis’s tried-and-true methods break down when dealing with input datasets
with more than a million characteristics.
Computing eigenvectors and eigenvalues presents another difficulty when
using classical computers. The larger the input, the larger the set of eigen-
vectors and associated eigenvalues. The use of QRAM (quantum random
access memory), which allows quantum computers to pick a path through
the data in a completely random fashion, allows quantum computers to solve
this problem quickly and efficiently. As shown in Figure 5.6, the vector is
transformed into a quantum state by employing qubits.
When performing quantum principal component analysis, the resulting
summary vector contains logarithmic qubits. A dense matrix is produced
when the provided random vector is used as the generator. Designers can
generate the basic version of any random course and deconstruct it into its
primary mechanisms by repeatedly sampling the data, using a density matrix

Figure 5.6 Example to show principal component analysis in quantum.

110 Quantum Machine and Deep Learning Models for Industry Automation

exponentiation method, and employing the quantum phase estimation algo-

rithm (Velu & Whig, n.d.). Consequently, both the computational complexity
and the time complexity are drastically reduced by an exponential factor.

5.6.2 Support vector quantum machines

When it comes to machine learning, one of the most well-known methods is
called the support vector machine, and we use that to carry out both regression
and classification. Its primary function in classification tasks is to organize
low-dimensional datasets into the appropriate categories. Imagine that the
data sizes are increased to the point where they can be linearly separated, as
demonstrated in Figure 5.7.
On desktop computers, the SVM will only make use of a selected number
of dimensions. It will become challenging after a certain point because these
machines do not possess adequate computing control (Asopa et al., 2021)..
The support vector algorithm can be executed exponentially more quickly
on quantum systems. It operates effectively and produces results more
quickly thanks to the overlap and entangled principles.

5.6.3 Optimization
Optimizing means getting something done with as little work and as few
resources as possible. Machine learning models can be made more capable
of making good and accurate predictions by optimizing the learning process.

Figure 5.7 Support vector quantum machines.

 5.7 Quantum Learning in Depth 111

Optimization focuses on reducing the loss function as much as possi-

ble. However, increasing the loss function can be costly and result in less
precise estimates because it reduces the outputs’ accuracy and consistency.
The success of the vast majority of machine learning strategies requires
iterative performance optimization. Quantum algorithms may be the key to
solving the optimization problems plaguing machine learning. Because of
the quantum entanglement property, a quantum government can pre-program
infinite copies of the currently applicable explanation. In order to improve the
final results of machine learning, they are incorporated at every stage of the
process (Velu & Whig, 2021b).

5.7 Quantum Learning in Depth

Teaching neural networks can go faster when deep learning and quantum
computing are used together. Using this method, we might be able to make a
new framework for deep learning and optimize the core. Scientists can copy
the results of traditional deep learning methods if they have access to a real,
physical quantum computer.
For multi-layer perceptron topologies, the amount of work that needs to
be done on the computer is proportional to the number of network neurons.
By using dedicated GPU clusters to improve performance, training time
may be able to be cut down by a lot. Unlike what was thought, it will
grow compared to quantum computers (Pawan Whig et al., 2017). Instead
of the traditional software used by regular computers, quantum computers
use neural networks to model their hardware. In particular, a qubit replaces a
neuron, the basic building block of any neural network. So, a quantum system
with qubits might be able to imitate a neuronic net and be used for deep
learning applications much faster than more traditional machine learning
techniques.

5.7.1 Why is quantum machine learning so exciting?

Leading businesses worldwide are vying for the most practical quantum
computer. By doing so, it will be possible to handle highly complicated
problems, such as cracking encryptions, advancing the development of new
medicines, revolutionizing communications networks, or advancing artificial
intelligence (AI) solutions.
112 Quantum Machine and Deep Learning Models for Industry Automation

Figure 5.8 Quantum machine learning model.

More than three billion online users are feeding massive data banks,
testing the capabilities of contemporary PCs. When Moore’s law was initially
proposed in 1965, it foretold that the number of integrated circuits would
double every two years. At the time, however, conditions were far from
stable. Quantum machine learning model is shown in Figure 5.8. Today’s top
businesses are vying for the most practical quantum machine available; so let
us fast forward to today. By doing so, it will be possible to access a tenfold
quicker machine and use it to tackle complex problems like cracking encryp-
tions, advancing drug discovery, transforming communications networks, or
advancing artificial intelligence (AI) solutions.
Consider Google as an example. With a 54-qubit Sycamore processor that
can compute in under 200 seconds, as opposed to the best supercomputer’s
10,000 years, it claims to have attained quantum supremacy today. That is
important because, when processing a huge quantity of data to solve complex
issues, things are sure to change dramatically. Therefore, it is crucial to
understand all real-world problems exclusively regarding quantum theory.
Even though Google’s assertion drew criticism, it nevertheless represents a
turning point for technology when looking into the future.
In research, Google examined the quantum advantage theory for machine
learning to better understand its applicability. The study showed how a
moderately complicated problem may alter when the appropriate datasets are
available, and it also showed how quantum algorithms can provide you with
 5.8 The Essence of Quantum Computing 113

a competitive advantage. It also showed how to create a valuable technique

for screening for data embeddings, notably by applying kernel approaches.
Thus, a relatively innovative approach to a few properties of a quantum
machine was offered thanks to the learning constraints and the insights from
the screening process. In essence, it aids in illustrating the greatest possible
empirical separation − a key benefit of quantum computing − and helps to
support quantum techniques using ideas drawn from classical learning.

5.8 The Essence of Quantum Computing

Some very important concepts which everyone should know about quantum
computing are as follows.

5.8.1 Taking the initiative to manage uncertainty

You have two choices when trying to solve a puzzle or navigate a maze: you
may call a quantum computer for assistance or a contemporary PC.
What distinguishes quantum computers is the difference in how the issue
is solved. A quantum machine will evaluate all feasible pathways at once
instead of a typical computer, which will consider all options and rule out
each one until it finds the best one.
In other words, quantum computers are well ahead of the game when
coping with ambiguity. One can only speculate about how quantum machines
will be able to address the most severe challenges ever when you apply that
viewpoint to more significant topics like AI or healthcare.

5.8.2 Welcoming a new AI era

There is no denying that AI has profoundly influenced our lives. Everything
that makes life simpler, from smartphones to smart homes, to education,
healthcare, and transportation, is made possible by AI.
While this is all wonderful, AI is still regarded as a relatively new
technology, particularly when it comes to thinking like humans. Things might
take a fascinating turn with quantum computing, turning AI systems into
more robots by giving them what is commonly called “thinking power.” As a
result, we are striving to improve goods’ durability and dependability.

5.8.3 Cybersecurity advancement

Cybersecurity derives its guiding concept by factoring a considerable number
into its primes. This is one area in which quantum computers are naturally
114 Quantum Machine and Deep Learning Models for Industry Automation

advantageous, and as they become more widely used, things like encryption
will soon be history. Having said that, if quantum computers get into the
wrong hands, our data might be utilized against us in unfathomable ways.

5.8.4 Accuracy of weather predictions

Weather forecasting is far more complicated than just telling you to bring an
umbrella because it will rain today. It directly affects a country’s economics,
development, and general functionality, especially considering food crops.

5.8.5 A signal to develop better life-saving drugs

In order to determine the effectiveness of a medical condition, available drug
research examines several interactions between proteins, molecules, and other
substances. The exercise is supported by several combinations that demand
careful thought and investigation.
Chemists can significantly benefit from quantum computing by using it
to find the most practical medication choices. It can also drastically reduce
the time it takes for medications to be discovered before they are used for a
variety of individualized therapies. Threats to some cutting-edge technologies
are shown in Figure 5.9.

5.9 A Portal to Exciting Future Technology

In the future, quantum computing will be the hub of operations for the
research and development division of the world’s top universities. For
instance, businesses are currently investing billions in enhancing the software
used in autonomous vehicles or in enhancing chemistry.

5.9.1 How AI will change thanks to quantum computing

When the terms “quantum” and “computer” are used, it is simple to picture
science fiction programs like Star Trek. When you discover that quantum
computing uses the combined features of superposition, interference, and
entanglement to run computations exceedingly rapidly, it does not make the
idea seem any less complicated. Fortunately, most do not need to worry about
the minute details. We must be aware that quicker data access and more secure
networking are two benefits of quantum computing.
This vast data collection is the foundation for machine learning tech-
niques employed by AI; the more data an algorithm can take in, the better
its predictions or choices. Sadly, the speed and stability provided by quantum
 5.9 A Portal to Exciting Future Technology 115

Figure 5.9 Threats to some cutting-edge technologies.

computing are necessary due to exponential development and the more

complex nature of inquiries.
Big data is the foundation of AI, a multipurpose technology. AI can find
trends and forecast events by studying databases − the price of gathering
and storing data formerly acted as a barrier to AI advancement. Nowa-
days, the difficulty is in promptly consuming, finding, and presenting useful
information.

5.9.2 Processes for making better business decisions

Productivity gains and quicker decision-making will be key as we move
toward a quantum computing future. Reaching specific audiences, analyzing
data, and forecasting trends are significant benefits.

5.9.3 Quantum security and artificial intelligence

It has always been difficult to keep up with the development of security
risks and assaults. Enterprises may more accurately forecast potential secu-
rity concerns and fend off prospective assaults by combining the speed of
quantum computing with the data analytics capabilities of AI.
116 Quantum Machine and Deep Learning Models for Industry Automation

Understanding that verifying data is just as crucial as interpreting it is

crucial as quantum computing and AI advance. A growing kind of cyberter-
rorism that should not be disregarded involves weaponizing data, falsifying
analytics, and disrupting AI systems’ learning experience.

5.9.4 AI and quantum computing complement DevOps

As DevOps teams struggle to establish business targets and objectives, design
and create new software solutions, and manage the continuing maintenance
and testing of current systems, quantum computing, and AI are formidable
partners.
Regression testing, functional testing, and user acceptability testing may
all be aided by DevOps teams looking at the data that AI provides. Testing
may be consistent and thorough because quantum computing allows AI to
swiftly and effectively handle data from different sources.

5.9.5 Where are our IT systems vulnerable?

Using quantum computing and AI to assist ITOps, when should our hardware
or software be upgraded? How can we handle situations more quickly? How
much time is devoted to managing activities that might be done automati-
cally? Big data analysis is the most effective method for addressing these
ITOps problems. These AI queries can provide comprehensive visibility into
operational data and real-time information thanks to the speed provided by
quantum computing.
It is most thrilling to think about how advancing these technologies
might truly benefit society by assisting in developing solutions for diseases,
untangling traffic, or securing sensitive data as corporations use quantum
computing and AI.

5.9.6 Limitation of quantum machine learning

Users may physically alter factors in quantum computing, such as the inten-
sity of an electromagnetic field or the frequency of a laser pulse, to address
issues. Quantum computers can therefore be trained similarly to neural net-
works. The main benefit of quantum computers is their capacity to generate
patterns that classical systems find challenging. Therefore, it is logical to
anticipate that quantum computers may perform better on machine learning
tasks than conventional computers. As a result, the discipline of quantum
machine learning has emerged. Learning algorithms can be improved by
 5.9 A Portal to Exciting Future Technology 117

quantum technology. This is known as machine learning with a quantum

enhancement. Machine learning techniques for analyzing data that cannot
be processed by classical computing are the most widely used quantum
computers in the area.
Quantum machine learning accelerates computing and can control how
algorithms in software store data. Running machine learning algorithms
on cutting-edge computing hardware like quantum computers expands the
evidence of learning. Compared to computer models, information processing
is based on quantum mechanics and its laws.
However, it is fair to say that the discipline of quantum machine learning
continues to work in more fantastical areas of physics.

5.9.7 Hardware constraints

Researchers frequently find themselves cut off from others in their industry
when conducting their work. Due to quantum decoherence, qubits may lose
quantum characteristics such as entanglement, decreasing the amount of data
that qubits can store. Quantum decoherence can be caused by exposure to
heat or light. Second, changes in the states of qubits require rotations in the
logic gates of quantum computers, which are well-known for being particu-
larly prone to errors. Any misalignment can lead to issues with the output.
Quantum machine learning also requires longer computer circuit lengths and
error correction (with redundancy for every qubit).

5.9.8 Program restrictions

It is crucial to consider the physics of quantum computers when designing
algorithms for them. However, while a classical algorithm can be designed in
the Turing machine style, a quantum algorithm must be designed in the style
of pure physics without using any simple formulae that would link it to logic.
Scalability is always a top priority in a system like this one. We are
creating software with more power to process data. Unfortunately, only a
tiny fraction of the information required to create such quantum computing
algorithms is currently available. Consequently, most progress is made via
untrained intuition.
Due to the limitations of particular simulations, creating models that can
significantly influence machine learning is complex. Due to these restrictions,
most well-known quantum algorithms are not applicable in the real world.
As for the third rule of quantum computing, there is a cap on how many
qubits can be kept on a quantum circle. Although these restrictions apply to
118 Quantum Machine and Deep Learning Models for Industry Automation

all quantum computing, attracting more attention and guiding research in the
right direction is possible by incorporating disciplines like machine learning.

5.10 More on Quantum Computing and Machine Learning

Connections
Future advancements in quantum computing should increase our capacity to
handle important computational jobs. The way we now utilize computers in
research and daily life is changing due to machine learning. It is only logical
to look for linkages between these two new computing paradigms to gain
several advantages.

5.10.1 Wavefunction
Today, one of the fastest routes to quantum advantage is expected to be the
prediction of the electronic structural features of molecules and materials.
The wavefunction for electrons is shown in Figure 5.10.
On the other hand, the most cutting-edge machine learning technique,
neural networks, has a brand-new application: simulating quantum physics.
Neural networks have only recently been utilized to categorize quantum
phase transitions or as a variational approach for interacting with many-body
systems.
The representation of quantum wavefunctions is a goal shared by quan-
tum computers and neural networks. This area of agreement might serve as

Figure 5.10 Wavefunction for the electrons.

 5.10 More on Quantum Computing and Machine Learning Connections 119

a jumping-off point for investigating potential linkages. Every strategy has

strengths and weaknesses.

5.10.2 The significance of accuracy

Along with other low-depth algorithms for electronic structure, the varia-
tional quantum eigensolver (VQE), as shown in Figure 5.11, uses quantum
state storage and manipulation to recover ground and excited state features of
quantum target systems. To achieve that, we must calculate the expected value
of Hamiltonian operators, which stand in for molecule energies in the case of
molecular systems. Additionally, we must do the measurement exactly since
any significant random fluctuations will render the entire quantum procedure
useless. It turns out that quantum computers are not particularly effective at
this. More peculiarly, it is impractical for existing technology to do enough
measurements to obtain sufficient accuracy for applications demonstrating
quantum advantage.
We employed neural network methods, as shown in Figure 5.12, on
quantum computing for more precise chemical simulations in PRR’s “Precise
measurement of quantum observables with neural-network estimators,” in
cooperation with two researchers from the Flatiron Institute, Giacomo Torlai
and Giuseppe Carleo. The method relies on measurement data obtained by a

Figure 5.11 Variational quantum eigensolver.

120 Quantum Machine and Deep Learning Models for Industry Automation

Figure 5.12 Quantum neural network.

quantum computer to train a neural network. After being trained, the neural
network encodes a portion of the quantum state, which is accurate enough to
retrieve molecule energies.
By integrating a quantum computer with our novel neural-network
estimator, we can combine the benefits of the two methods. We use quan-
tum computers’ ability to interfere with states throughout an exponentially
expanding Hilbert space while a chosen quantum circuit is being run. We
get a limited set of measurements once the quantum interference process has
finished. The neural network, a classical tool, may then take advantage of
this sparse input to effectively represent some aspects of a quantum state,
including its simulated energy.
The main query that remains after this data transfer from a quantum
processor to a conventional network is:
How well can neural networks capture the quantum correlations
of a dataset of limited measurements produced by sampling molecular
wavefunctions?
 5.10 More on Quantum Computing and Machine Learning Connections 121

For computer scientists, a toolset for quantum computing encompasses

a suite of software, algorithms, hardware frameworks, and specialized lan-
guages designed to manipulate and harness the power of quantum systems.
We must consider how neural networks could replicate fermionic matter
to respond to this challenge. Spin lattice and continuous-space issue sim-
ulation has previously been carried out using neural networks. It is still
difficult to solve fermionic models with neural networks. We investigated
how molecules are simulated on quantum computers to try to identify a
workaround for that.
We employed encodings of fermionic degrees of freedom to qubit ones,
precisely the same encodings employed in variational algorithms and molec-
ular simulations on quantum computers. We have created fermionic neural
network states using these mappings accessible on Qiskit Aqua. On a tra-
ditional computer, we put them to the test against molecular ground states,
which are quantum objects.
5.10.3 Data power and quantum machine learning
In recent years, quantum computing has grown quickly in both theory and
practice, raising hopes that it may have an influence on actual applications.
How quantum computers could impact machine learning is an important
topic of research. The ability of quantum computers to naturally handle some
problems with intricate input correlations that may be highly challenging to
conventional or “classical” computers was recently experimentally shown.
In light of this, it is possible that learning models developed on quantum
computers, as shown in Figure 5.13, would be significantly more powerful
for specific applications, possibly offering quicker processing, greater gener-
alization on fewer datasets, or both. Therefore, it is crucial to comprehend the
circumstances in which a “quantum advantage” may be obtained.
In most cases, the notion of quantum advantage is expressed in terms of
computing gains. Can a quantum computer, given a task with well-specified
inputs and outputs, provide a more accurate result than a conventional
machine in the same amount of time? Quantum computers are thought to
provide a significant performance advantage for various algorithms, including
Shor’s factoring method for factoring huge prime products (important for
RSA encryption) and the quantum modeling of quantum systems. However,
the availability of data can significantly influence a problem’s difficulty and,
thus, the potential benefit for a quantum computer.
We analyze the issue of the quantum advantage in machine learning
to understand better when it would apply in “power of data in quantum
122 Quantum Machine and Deep Learning Models for Industry Automation

Figure 5.13 Quantum machine learning model.

Figure 5.14 Separation of data in input and feature space.

machine learning,” which was published in Nature Communications. We

demonstrate how a problem’s formal complexity changes when more data
become available and how this might occasionally make classical learn-
ing models competitive with quantum algorithms. We then create a useful
technique for determining if a certain collection of data embeddings may
 5.11 Case Study 123

have a quantum advantage when used with kernel approaches. We offer a

unique approach that projects certain features of feature maps from a quantum
computer back into classical space using the learning bounds and insights
from the screening method.

5.11 Case Study

5.11.1 Q-SVM (quantum support vector machine algorithm)
Although quantum computers and quantum programming are very new fields,
they both arouse curiosity and stage experimental studies to understand
what can be done. One of these experimental areas of study is the study
of machine learning and testing of deep learning algorithms, even artificial
intelligence, on quantum computers. However, even deep learning methods
and artificial intelligence studies are at the beginning, while their application
in quantum computers continues to be academic and theoretical studies rather
than practical uses.
As far as in search, implementing the QSVM algorithm can be accom-
plished in two different ways. The first of these methods is to run the kernel
function only on a quantum computer and to get the data from a conventional
computer and conventional feature space by reducing it with conventional
methods. The second is to do both the reduction of data in the property space
and the labels of the data and the kernel function on quantum computers.
We first need to know how the classic version of SVM classifies the data
to understand the difference between these two methods. Assuming we know
this, let us remember with an image that shows how data is separated from
each other in the property space by increasing size, as shown in Figure 5.10.

Figure 5.15 Quantum property space.

124 Quantum Machine and Deep Learning Models for Industry Automation

The difference between the two methods mentioned above is that our
feature space is also calculated in quantum circuits in the first one. The
quantum property space can also be shown in Figure 5.15.
The common point in both methods is that the kernel function is obtained
by quantum calculations rather than the traditional way. However, to increase
the success rate, the data labels must be provided in the superposition state.
In order to do this, the data must be read from a quantum circuit, but we do
not have the opportunity to apply this method of data provision here. How-
ever, the results obtained in experimental studies show that 100% predictive
success can be achieved with this method. (Of course, this was done by using
regular data that can achieve 100% success.)
Although we cannot provide the data from the quantum computer, a
function is provided inside the Qiskit library that allows the data to be
separated using the quantum property space. We will use this library named
feature_map below.

5.11.2 Why did they need Q-SVM?

There is no compelling reason to cling to antiquated strategies for machine
learning in an era when quantum computers and algorithms are readily
available. The use of properties of quantum physics, such as entanglement
and the computational power that is made available by quantum computers,
are two approaches that can be taken to answer this question. It is utilized
in the SVM algorithm to classify the data by forging support points from
clusters of data points that are very similar to one another. The images of
feature spaces above demonstrate that a data classification in two or three
dimensions is impossible. Images of this caliber contribute significantly to the
argument’s weight. For this, it may be necessary to increase the size towards
an n-dimensional space and separate our data from the points representing it.
However, as we work in multi-dimensional spaces, our computing power is
insufficient in classical computers.

5.11.3 Import the library

First, we will use the SVM algorithm in [Link]. But other libraries have
an alternative to SVM algorithms. The dataset we will use is the iris dataset,
which is one of the most frequently used datasets in machine learning studies.
In [2]:
import numpy as np
from [Link] import load_iris
 5.11 Case Study 125

from sklearn.model_selection import train_test_split

import qiskit
from qiskit import BasicAer, Aer
from [Link] import QuantumInstance
from [Link] import QSVM
from [Link].multiclass_extensions import one_against
_rest, all_pairs
from [Link].feature_maps import SecondOrderExpan-
sion
from [Link] import ClassificationInput
from [Link] import run_algorithm
backend = Aer.get_backend(’qasm_simulator’)
iris = load_iris()
Since it takes too long to run the algorithm on IBM-Q, I run it on
qasm_simulator. But nevertheless, it will take time.

5.11.4 Install the dataset

Iris, which is the dataset we will use, consists of three separate classes as the
target variable and four feature columns.
In [3]:
X, Y = [Link], [Link]
print([Link])
print(len(set(Y)))
train_x, test_x, train_y, test_y = train_test_split(X, Y, test_size=0.2)
num_features = 4
training_size = 120
test_size = 30
feature_map = SecondOrderExpansion(feature_dimension=num_features,
depth=3)
# With feature_map, we have defined the function we will use in the
abovementioned quantum property space.
(150, 4)
3
We defined the feature map above but did not need to define it here since we
will give it as a parameter below. For the sake of example, both remain for
now.
In [4]:
params = {
126 Quantum Machine and Deep Learning Models for Industry Automation

’problem’: {’name’: ’classification’, ’random_seed’: 794},

’algorithm’: {
’name’: ’QSVM’,
},
’backend’: {’shots’: 1},
’multiclass_extension’: {’name’: ’OneAgainstRest’},
’feature_map’: {’name’: ’SecondOrderExpansion’, ’depth’: 3, ’entan-
gler_map’: [[1, 0]]}
#’feature_map’: {’name’: ’SecondOrderExpansion,’ ’depth’: 3}
}
training_dataset={’A’:train_x[train_y==0],
’B’:train_x[train_y==1],
’C’:train_x[train_y==2]}
test_dataset={’A’:test_x[test_y==0],
’B’:test_x[test_y==1],
’C’:test_x[test_y==2]} total_arr = [Link]((test_dataset[’A’],test
_dataset[’B’],test_dataset[’C’]))
alg_input = ClassificationInput(training_dataset, test_dataset, total_arr)
In [5]:
result = run_algorithm(params, algo_input=alg_input, backend=backend)
Let us print the results and see how we got the output.
In [6]:
result
Out[6]:
{’testing_accuracy’: 0.7,
’test_success_ratio’: 0.7,
’predicted_labels’: array([0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 2, 0, 1, 1, 1, 0, 1,
1, 0, 2,
1, 2, 1, 2, 0, 2, 2, 1]),
’predicted_classes’: [’A’,
’A’,
’A’,
’A’,
’A’,
’A’,
’A’,
’A’,
’A’,
’B’,
 5.12 Quantum Computing and Machine Learning for Industry Automation 127

’A’,
’B’,
’C’,
’A’,
’B’,
’B’,
’B’,
’A’,
’B’,
’B’,
’A’,
’C’,
’B’,
’C’,
’B’,
’C’,
’A’,
’C’,
’C’,
’B’]}
Our results are relatively good. The fact that we have not run a full QSVM
technically in these results also has an impact. Nevertheless, you can see that
it is possible to obtain more precise results by examining the case study. With
the development of technology and hardware obstacles disappearing one by
one in the next 10 years, we will be able to run more advanced algorithms in
quantum computers with more precise results.

5.12 Quantum Computing and Machine Learning for

Industry Automation
It is expected that quantum computing will aid in developing revolutionary
products and services, which will shake up and reshape the manufacturing
sector. Benefits in business dealings: Quantum computing is expected to have
a significant impact on many areas of manufacturing, including chemical
discovery, product development, and process optimization, to name a few.
Significant influence: Early adopters will be able to secure advantages that
will be extremely tough to overcome.
In this section, we look at prospective manufacturing use cases in four
categories: find, design, control, and supply:
128 Quantum Machine and Deep Learning Models for Industry Automation

5.12.1 Discover
It is a shared secret that even modestly large molecules are difficult to
represent with perfect chemical precision using today’s traditional com-
puters. With its exponentially huge state space, quantum computing could
offer detailed modeling of incredibly complicated molecules, which can
significantly improve materials and medication research.
To the best of our knowledge, there are around 15 million defined
chemical compounds and around 300,000 materials today. More and more
practical chemicals will undoubtedly be uncovered in the future. Nature
creates materials with extraordinary properties that conventional industrial
manufacturing methods cannot duplicate. For example, spider silk is stronger
than steel when measured by its specific weight and can be produced at body
temperature rather than in a furnace. Only wastewater from production is
discarded. Given that spider silk is a protein derived from DNA, one day, we
may be able to create a substance that is comparable to or superior to spider
silk that is produced in a manner that is equally as good for the environment
thanks to the increased capacity of quantum computing to simulate at the
subatomic level.

5.12.2 Design
Many items today are developed and validated using numerical simula-
tions. Individual engineering safety margins are used to 3D-model auto
and aerospace components and sub-assemblies. These margins can add up,
resulting in goods that are overweight or cost more than required, limiting
their profitability.
However, in the future, quantum mechanics will be able to model com-
ponent interactions within complex hardware systems, allowing for more
accurate and comprehensive calculations of system loads, load pathways,
noise, and vibration. The potential long-term impact of multiple individual
safety margins can be reduced, costs can be reduced, and system performance
can be maintained or improved if the system as a whole is analyzed in detail.

5.12.3 Control
Cutting-edge analytics is tested to its limits when new production control
methods are implemented, especially when machine learning and many
variables are in play. String theory has the potential to enhance traditional
computing by aiding in the development of pattern recognition and the
 5.12 Quantum Computing and Machine Learning for Industry Automation 129

advancement of categorization. The following business areas are expected to

be significantly impacted by the integration of quantum computers and algo-
rithms and by the optimization opportunities that this integration presents.
Machine learning and standard multi-variate analysis are currently used
in semiconductor chip production. However, classical computing has hit a
technological wall and cannot accommodate further increases in the param-
eters required for more complex analysis. Quantum computing’s end goal
is to improve manufacturing output by allowing for the evaluation of more
interactive elements and processes.
Modeling and optimizing the production processes and robotics schedul-
ing for complicated items like automobiles requires much computational
power. These products, like automobiles, are well-known for their notoriously
high levels of complexity. Quantum computing has the potential to speed up
refinement processes and allow for more versatile optimization in practice.

5.12.4 Supply chains

The structure of supply chains is shifting from one that is linear and event-
driven, with discontinuous, sequential operations, to one that is more organic
and flexible and is based on the evolving needs of the true market and the
currently available components. In the context of Industry 4.0, the addition
of quantum computing to the toolkit of the digital supply chain has the
potential to hasten decision-making, improve risk management, and bring
about a reduction in operating costs and sales that were lost as a result of items
being discontinued or out of stock. The supply chain may undergo a complete
revolution due to quantum computing in the long run. This could involve
adapting the supply chain to optimize vendor orders and associated logis-
tics by utilizing dynamic near-real-time decision-making based on shifting
market requirements, increasing competitive flexibility.
Quantum computing is poised to become a critical transformational tool
for manufacturing. With the predicted influence on product design and devel-
opment production methods, and supply chain operations, early buyers who
embraced the quantum future today may gain a significant advantage.

5.12.5 How does manufacturing begin?

1. Delegate the task of conducting experiments with actual supercomputers
and researching the potential applications of quantum mechanics in your
sector to individuals within your company who have been designated as
130 Quantum Machine and Deep Learning Models for Industry Automation

“quantum champions.” Allow your quantum champions to report their

findings to a quantum steering committee comprising line-of-business
leaders and market analysts to assist in concentrating on the most
significant challenges.
2. Determine which potential applications of quantum computing should
be prioritized based on the extent to which they can deliver a competitive
advantage, taking into account the business strategy of your company,
the associated consumer value propositions, and the long-term devel-
opment objectives. Maintain a close eye on developments in quantum
application development so that you can stay one step ahead of any
potential use cases that could be commercialized sooner rather than later.
3. Contemplate the possibility of working together with a new quantum
ecosystem that is currently in the process of developing. This ecosystem
is made up of like-minded research laboratories and academic institu-
tions, quantum technology suppliers, quantum application developers
and coders, and start-ups with supporting technologies. Involve other
companies experiencing the same problems to gain rapid access to an all-
encompassing quantum computing stack that can construct and execute
quantum algorithms tailored to your specific company’s needs. Keep an
eye out for developments in quantum technology that might call for a
reorganization of partnerships.

5.13 Conclusion and Future Scope

Although the potential of using quantum computing for machine learning is
still intriguing, certain nuanced aspects must be carefully considered when
measuring quantum benefit. Here, we built a framework for comprehending
quantum benefit chances in a learning environment. We demonstrated in the
case study how data-driven classical ML algorithms can become exponen-
tially more effective, and we demonstrated that even if the data originate from
a dynamic process that is difficult to reproduce separately, a forecast benefit
for classical modeling is not assured. This chapter is highly beneficial for
researchers in the same field.

References
Alkali, Y., Routray, I., & Whig, P. (2022). Study of various methods for reli-
able, efficient, and Secured IoT using Artificial Intelligence. Available
at SSRN 4020364.
 References 131

Anand, M., Velu, A., & Whig, P. (2022). Prediction of Loan Behaviour with
Machine Learning Models for Secure Banking. Journal of Computer
Science and Engineering (JCSE), 3(1), 1–13.
Asopa, P., Purohit, P., Nadikattu, R. R., & Whig, P. (2021). Reducing car-
bon footprint for sustainable development of smart cities using IoT.
2021 Third International Conference on Intelligent Communication
Technologies and Virtual Mobile Networks (ICICV), 361–367.
Bhargav, R., & Whig, P. (2021). More Insight on Data Analysis of Titanic
Data Set. International Journal of Sustainable Development in Comput-
ing Science, 3(4), 1–10.
Chouhan, S. (2019). Using an Arduino and a temperature, humidity sensor,
Automate the fan speed. International Journal of Sustainable Develop-
ment in Computing Science, 1(2).
George, N., Muiz, K., Whig, P., & Velu, A. (2021). Framework of Percep-
tive Artificial Intelligence using Natural Language Processing (PAIN).
Artificial & Computational Intelligence/Published Online: July.
Hussein, A. H. (2019). Internet of Things (IoT): Research Chal-
lenges and Future Applications. In IJACSA) International Journal of
Advanced Computer Science and Applications (Vol. 10, Issue 6).
[Link]
Jiwani, N., Gupta, K., & Whig, P. (2021). Novel HealthCare Framework for
Cardiac Arrest With the Application of AI Using ANN. 2021 5th Inter-
national Conference on Information Systems and Computer Networks
(ISCON), 1–5.
Khera, Y., Whig, P., & Velu, A. (2021). efficient, effective, and secured
electronic billing system using AI. Vivekananda Journal of Research,
10, 53–60.
Nadikattu, R. R., Mohammad, S. M., & Whig, P. (2020). Novel economical
social distancing smart Device for COVID-19 (SSRN Scholarly Paper
ID 3640230). Social Science Research Network. Https://Papers. Ssrn.
Com/Abstract, 3640230.
Anupam Priyam et al. (2018). Simulation & performance analysis of various
R2R D/A converter using various topologies. International Robotics &
Automation Journal, 4(2), 128–131.
Rupani, A., & Sujediya, G. (2016). A Review of FPGA implementation
of Internet of Things. International Journal of Innovative Research in
Computer and Communication Engineering, 4(9).
Shrivastav, P., Whig, P., & Gupta, K. (n.d.). Bandwidth Enhancement by
Slotted Stacked Arrangement and its Comparative Analysis with Con-
ventional Single and Stacked Patch Antenna.
132 Quantum Machine and Deep Learning Models for Industry Automation

Velu, A., & Whig, P. (2022). Studying the Impact of the COVID Vaccination
on the World Using Data Analytics.
Velu, A., & Whig, P. (2021a). Protect Personal Privacy And Wasting Time
Using Nlp: A Comparative Approach Using Ai. Vivekananda Journal of
Research, 10, 42–52.
Whig, P., & Ahmad, S. N. (2017). Controlling the Output Error for Photo
Catalytic Sensor (PCS) Using Fuzzy Logic. Journal of Earth Science
and Climate Change, 8(4), 1–6.
Whig, P., & Ahmad, S. N. (2019). Methodology for Calibrating Photocat-
alytic Sensor Output. International Journal of Sustainable Development
in Computing Science, 1(1), 1–10.
Whig, P., Nadikattu, R. R., & Velu, A. (2022). COVID-19 pandemic analysis
using application of AI. Healthcare Monitoring and Data Analysis Using
IoT: Technologies and Applications, 1.
Jupalle, H., Kouser, S., Bhatia, A. B. et al. Automation of human behaviors
and their prediction using machine learning. Microsyst Technol (2022).
[Link]
P Whig, A Velu, P Sharma (2022) Demystifying Federated Learning for
Blockchain: A Case Study , pp1-22
Alkali, Yusuf, et al. (2022) Strategy for Reliable, Efficient and Secure IoT
Using Artificial Intelligence. IUP Journal of Computer Sciences . Apr
2022, Vol. 16 Issue 2, p16-25.
Shama et al. (2022), Fog-IoT-Assisted-Based Smart Agriculture Application,
Demystifying Federated Learning for Blockchain: A Case Study, PP-1-
20
Schuld, M.; Petruccione, F. Supervised Learning with Quantum Computers;
Springer: Berlin/Heidelberg, Germany, 2018; Volume 17.
Montanaro, A. Quantum algorithms: An overview. Npj Quantum Inf. 2016,
2, 15023.
Jordan, S. The Quantum Algorithm Zoo. 2021.
Marais, A.; Adams, B.; Ringsmuth, A. K.; Ferretti, M.; Gruber, J. M.;
Hendrikx, R.; Schuld, M.; Smith, S. L.; Sinayskiy, I.; Krüger,
T. P.; et al. The future of quantum biology. J. R. Soc. Interface 2018, 15,
20180640.
Biamonte, J.; Faccin, M.; De Domenico, M. Complex networks from classical
to quantum. Commun. Phys. 2019, 2, 53.
McMahon, D. Quantum Mechanics Demystified; McGraw-Hill Education:
New York, NY, USA, 2013.
 References 133

Samuel, A. L. Some studies in machine learning using the game of checkers.

IBM J. Res. Dev. 1959, 3, 210–229.
Alzubi, J.; Nayyar, A.; Kumar, A. Machine Learning from Theory to
Algorithms: An Overview. J. Phys. Conf. Ser. 2018,
Rivas, P. Deep Learning for Beginners: A Beginner’s Guide to Getting Up
and Running with Deep Learning from Scratch Using Python; Packt
Publishing Ltd.: Birmingham, UK, 2020.
Mehta, P.; Bukov, M.; Wang, C. H.; Day, A. G.; Richardson, C.; Fisher, C.
K.; Schwab, D. J. A high-bias, low-variance introduction to machine
learning for physicists. Phys. Rep. 2019, 810, 1–124.
Mahesh, B. Machine Learning Algorithms—A Review. Int. J. Sci. Res.
(IJSR) 2020, 9, 381–386.
Bonaccorso, G. Machine Learning Algorithms; Packt Publishing Ltd.: Birm-
ingham, UK, 2017.
Wittek, P. Quantum Machine Learning: What Quantum Computing Means to
Data Mining; Academic Press: Cambridge, MA, USA, 2014.
Nielsen, M. A.; Chuang, I. L. Quantum computation and quantum informa-
tion. Phys. Today 2001, 54, 60.
McMahon, D. Quantum Computing Explained; John Wiley & Sons: Hobo-
ken, NJ, USA, 2007.
 6
The Contribution of Computer Vision in the
Manufacturing Industries and the Scope for
Further Excellence

P. Dharanyadevi1 , C. Krishnakoumar1 , K. Revathy1 , J. Karkavelraja1 ,

K. Venkatalakshmi2 , and G. Zayaraz1
1 Department of CSE, Puducherry Technological University, India
2 Department of ECE, University College of Engineering Tindivanam, India
E-mail: dharanyadevi@[Link]; krishnakoumar.c@[Link];
karkavelraja.j@[Link]; revathy.k@[Link]; venkata_krish@[Link];
gzayaraz@[Link]

Abstract
The most important mission of manufacturing industries worldwide is to
achieve high-quality product throughput with their ongoing mass production
agenda. “Quality is inversely proportional to Quantity” is a widely well-
known notion among business leaders and consumers. Computer vision is
revolutionizing industries with its ineffable contribution to the mass produc-
tion of products and services. Computer vision is a sub-field of artificial
intelligence that deals with image recognition and automation. Computer
vision is used for automatic inspection and assessment in most industries,
but to leverage AI’s maximum potential, we must implement computer vision
in the appropriate place. Apart from manufacturing, computer vision can be
used in other industrial processes like logistics, human resource management,
security components, analysis of machinery for its proper functionality, and
more. Effectively implementing this technology has drastically reduced run-
ning costs in many manufacturing sectors. Unlike other fields of AI, computer

135
136 The Contribution of Computer Vision in the Manufacturing Industries

vision or machine vision requires minimal hardware initialization cost and

a more coherent system layout. AI is a technology, and most of the part
is used in industries to increase overall efficiency and to reduce running
costs. So, the cruder meaning of “AI” is “increase efficiency” for the most
part. If diverse industries incorporate computer vision effectively within their
systems, it will lead to better organization of raw materials, supply chain,
quality control, human resource management, maintenance of machinery,
and so on.
This chapter discusses the workflow of computer vision and its different
domains implemented in various industries, such as automation, agriculture,
and so on. This chapter also discusses how computer vision can help in
achieving sustainable industries. This paper demonstrates the various possible
implications of computer vision in manufacturing industries and also dis-
cusses how computer vision has been effectively implemented. This chapter
will furnish a new dimension for researchers and gives an idea to alleviate the
overall production cost and complexity.

Keywords: artificial intelligence, computer vision, automation, use cases,

performance metrics

6.1 Introduction
Computers are an integral part of present-day industries. They automate most
of the tiring jobs that humans performed in the past. Human laborers have
their major drawbacks in performing low-end jobs. In the past, industries
have adopted various methods to do specific tasks efficiently through vari-
ous revolutions. The fifth industrial revolution focuses on connecting every
machinery within each industry to be connected and work seamlessly. With
the advent of connectivity technologies, we have more information about the
machine than ever. We can access all the data we gathered throughout history
and feed it into the machines [1].
Humans process information from the external environment through the
sensory organs. The sensory organ is a biological system used to gather
information from the environment or respond to stimuli [2]. Vision is one
of the six senses that humans possess. It is one of the most critical sensory
systems that is closely bound to the brain. Computer vision is an area with
a huge scope for development and improvement that can replace the places
where these systems are where human interventions are least required.
 6.1 Introduction 137

Table 6.1 Comparison between human vision and computer vision.

Human vision Computer vision
Can identify people and objects It needs to be explicitly programmed to
identify people and objects
Does not care about the accuracy of the Differentiates the objects around the world
world based on probabilistic and statistical inter-
ventions
Can understand the mood in the scene It can understand the mood in the scene
when it is explicitly programmed
Ignores many details Stores and processes every tiny detail
Can imagine stories I cannot imagine
Suffers from illusions Illusions are not possible
Ambiguous description of the world It gives an accurate description of the world
Can navigate through obstacles Can navigate through the world when it is
explicitly programmed

The main objective of computer vision is to automate the information

extracted from the image. Here the information can be anything from 3D
models, object detection and recognition, searching image content, and
camera position [1]. Some tasks involved in computer vision are image
processing, cognitive processing, statistical pattern identification, and geo-
metric modeling. Computer vision requires both high-level and low-level
capabilities to perform well [3].
Computer vision contribution toward manufacturing industries is already
making remarkable progress. Human vision can identify minute nuances
within a fraction of a second. We need the best-in-class cameras and
algorithms to achieve computers that mimic human intelligence. Computer
vision algorithms need to be trained well to achieve high-precision image
recognition capabilities.
Table 6.1 illustrates the comparison between human vision and computer
vision. At present, computer vision systems are way beyond detecting objects.
They are entering the arena competing with human vision. Figure 6.1 depicts
the pictorial representation of computer vision in a different milieu. Computer
vision systems consist of optics and high-resolution cameras integrated with
algorithms to perform a specific task [4]. The number of tasks these computer
vision systems handle in today’s manufacturing industries is impeccable since
it takes care of most often considered organized and repetitive tasks.
Computer vision is hard because we are trying to actively look into
the unknown with very sparse data to conclude a solution. To tackle this
problem, we use probabilistic models and classical physics for a firm
138 The Contribution of Computer Vision in the Manufacturing Industries

Figure 6.1 Machine vision.

understanding. Researchers are currently developing mathematical models

to produce/recover the 3D shape and appearance of objects in images. For
example, let us consider optical character recognition (OCR), which is used
for recognizing handwritten postal codes on letters and recognizing numbers
on a car’s number plate using the technology of automatic number plate
recognition (ANPR) [5].

6.2 Components of a Machine Vision Systems

Table 6.2 illustrates the various components of a machine vision system. Each
component holds a specific task in computer vision.
Figure 6.2 illustrates the image-sensing pipeline. The accuracy of com-
puter vision depends on the configuration of these components along the
 6.2 Components of a Machine Vision Systems 139

Table 6.2 Components of vision systems.

Camera A camera is an optical instrument that can capture a visual image.
Its components can be digital or analog. The shutter speed, sam-
pling pitch, fill factor, chip size, analog gain, sensor noise, and
resolution are the factors that cause a delay in the performance of
the digital image sensor [6].
Shutter speed The amount of time the image sensor is exposed to light can either
be controlled by the shutter speed or the exposure time, resulting
in either an overexposed or underexposed photograph.
Sampling pitch The amount of space between neighboring cells on the sensor
of the imaging chip is referred to as the sampling pitch. The
sampling pitch has a relationship that is inversely proportional to
the sampling density. A sensor with a small sampling pitch can
produce images with a high resolution. A sensor with a more
extensive sampling pitch has a smaller sensor area since there
is a larger space between the cells. Thus, the sensor can only
accumulate smaller photons, thus making it less light sensitive
and more toward producing noise in the image.
Fill factor The fill factor is the ratio between the pixel’s light-sensitive area to
the pixel’s total area on the sensor. A higher fill factor is preferred
for a quality image. The high fill factor results in more light
capture that, in turn, increases the quality of the image [7].
Chip size Point-and-shoot cameras and camcorders often use smaller sen-
sors ranging between 0.25 inches and 0.5 inches in size, while
single lens reflex (SLR) cameras try to come closer to 1 inch
to 1.9 inches. A larger sensor size is directly proportional to a
photosensitive image. Hence a larger chip size is always advisable
for a better vision system.
Analog gain A sense amplifier boosts the signals received by the image sensor.
In cameras, this is controlled by automatic gain control (AGC)
logic. AGC logic automatically adjusts the incoming signal to
obtain a good image exposure.
Sensor noise Noise can be defined as unwanted signals. In images, noises are
always in the form of film grains. Senor noise is the random
variation noises added from various sources, including amplified
noise, fixed pattern noises, quantization noises, shot noises, and
dark current noises.
ADC resolution It is defined as analog to digital conversion resolution, a final step
procedure that can elevate the information gain of the image from
the sensor. ADC resolution works by a minute incremental voltage
that can be recognized by the camera, which changes the digital
output.
140 The Contribution of Computer Vision in the Manufacturing Industries

Table 6.2 (Continued.)

Digital This is the final step inside a camera. In this step, the output from
post-processing the sensor is converted into digital bits. Based on the camera’s
utility, it performs various digital signal processing (DSP) opera-
tions to enhance the images for extracting the most out of every
pixel.
Lighting Lighting is an essential part of image processing. There are dif-
ferent lighting techniques; each serves a particular purpose for
computer vision operations.
Directional This technique is used for picking out surface effects. It is mostly
similar to the sun.
Diffused This is used to better position the objects, providing fewer shad-
ows and reflections.

Polarized light Polarization is achieved using polarized lenses on the camera.

Polarizing backlighting can even render surface tension visibility.

Infrared or They are most commonly used for observing heat formation and
ultraviolet dissipation, fluorescence/phosphorescence, and ionization. And
the lighting can be manually arranged for front, back, and ring
lighting.

Figure 6.2 Image sensing pipeline.

 6.5 Use Case of the Computer Vision in Industries 141

pipeline. The two key computer vision concepts are image formation and
computer vision algorithms. The description is stated below:

6.3 Image Formation

An image is formed when some electromagnetic waves bounce off an
object, reaching the observer’s eyes or electromagnetic sensing instrument.
Computer vision is based on input from electromagnetic radiation. Electro-
magnetic radiation can be categorized into different ranges of frequencies.
Many electromagnetic spectrums can create a visual of an object. Electromag-
netic frequencies can be classified into different types within the spectrum
range. The following spectrum ranges are not helpful to machine vision; they
are
• Radio waves (3 Hz–30 MHz)
• Microwave waves (300 MHz–30 GHz)
• Ionizing radiation (3 PHz–300 EHz)
Infrared, visible light, and ultraviolet spectrums are used based on their spe-
cific use cases. While all spectrums have specific use cases, like gamma rays,
imaging is most commonly used in space industries to capture images from
the gamma rays emitting from distant galaxies. For example, the infrared
spectrum is most commonly used for creating a thermal image that uses
infrared radiation (IR), similar to a normal camera that uses visible light for
taking images.

6.4 Computer vision algorithms

As shown in Figure 6.3, computer vision algorithms have improved tremen-
dously in recent decades. The availability of cheaper computation systems
and improved hardware performance leads to the development of more
complex vision systems that are accurate in identifying objects.

6.5 Use Case of the Computer Vision in Industries

6.5.1 Product assembly
Machine vision’s contributions can be seen throughout product assembly
lines of various manufacturing companies across various domains around the
world. The most important part of a machine vision’s ability is its efficiency
142 The Contribution of Computer Vision in the Manufacturing Industries

Figure 6.3 Timeline of the latest advancement in computer vision algorithms.

and accuracy in looking out for defects. Incorporating computer vision in

industries reduces the overall cost of producing a product. The manufac-
turing industries are seeing a tremendous increase in the implementation
of automated systems into their workflow pipeline to cut down on running
costs. Nearly all the manufacturing giants we hear of around the planet have
implemented automation in assembly lines to ramp up their production output
[8]. Computer vision can provide inspection services like 3D modeling of
the products, surface anomaly detection, deformed product elimination, and
rupture identification, to name a few. All this happened in a fraction of a
second with the best accuracy across various other options. Computer vision’s
efficacy is not limited to just within the assembly line; they are assisting
machines and workers in manual assembly units. Additionally, computer
vision can help avoid purchases after returns and product recall situations,
thus ensuring customer satisfaction.
Industrial Revolution 4.0 is happening worldwide, and computer vision
has become a significant part of this revolution. Almost every industry is
trying to incorporate complete automation into its manufacturing pipeline.
3D models designed for the assembly process are designed with the help of
computer-aided design software, based on which the machine vision systems
guide the assembly process precisely as defined. It also helps employees and
various other robot systems work parallelly without interruptions. As shown
in Figure 6.4, for example, computer vision is implemented in analyzing the
distribution of salami over pizza crust [9].

6.5.2 Defect detection

Manufacturing products with 100 percent accuracy in defect detection is the
priority of many manufacturing companies, and they struggle to achieve that
at any cost. This is mainly because detecting the defects after delivery or at
the end of the production results in a hike in the overall cost of production
 6.5 Use Case of the Computer Vision in Industries 143

Figure 6.4 Overview of salami placement in pizza.

and decreased customer satisfaction and trust. This often leads to increased
customer dissatisfaction and decreased brand value. Ultimately this results in
loss to both manufacturers and customers. Moreover, implementing computer
vision and automation into the production line decreases the net production
cost that can incur when these upgrades are not implemented.
Detecting defects much later than sorting them out in the beginning
stage is costlier. Funding the computer vision and automation-based systems
for defect detection systems can be the most accessible and cost-effective
solution. The computer vision system generally gathers real-time data from
the image sensors. With the help of artificial neural networks or machine
learning models, it analyzes the data and, based on the predefined quality
standards, detects the defects or provides the percentage of deviation of the
defects based on the employees can solve the problem [10]. This can help
detect the flaw within the production line and approach the problem at a much
earlier stage [2].
Another application of computer vision is the detection of surface defor-
mations [11]. Stereo vision technology can be used to evaluate the precise
measurement of surface deviation. This method is efficient compared with
other methods because it provides high precision output, is rapid to develop,
and is easy to operate.
Another method to effectively segment defective products in a lot can
be performed using graph-based image segmentation. This method can be
effectively used when the background of the products is chaotic and contains
several objects of the same type [6].
144 The Contribution of Computer Vision in the Manufacturing Industries

Achieving 100 percent accuracy in detecting defects is a distant howl, but

detection at a much early stage goes a long way compared with detection
at the end of the production line. Machine vision deployed in manufacturing
units can help prevent this flaw by detecting defects at macro and micro levels
of defections in the production line.

6.5.3 3D Vision system

Humans can identify anomalies with any product in a fraction of a second.
At a faster pace, humans struggle to identify any anomalies [12]. Computer
vision will take this place in these spots where human efficiency decreases.
To handle this situation, the system takes high-resolution images to build the
3D model of the components and their connections.
This system works by analyzing the components which run through
the manufacturing pipeline; the computer vision system captures the high-
resolution image from various angles across various spots to build a 3D
model of the product, which is then fed into an artificial intelligence sys-
tem to find out any defects or anomalies [15]. This technology is used
in various industries, particularly automobiles, energy, electronic circuits,
and so on.
Advantages in manufacturing industries: 3D vision systems can identify
the object’s location, environment, and orientation in space. The cameras
inside this system take three-dimensional images of the objects, just like
how the human eye perceives the environment. Hence, they open up many
implications in the manufacturing sector. The cost of setting up 3D cameras
and effective installation of cameras at appropriate places depending on the
industry domain is still a challenging task. Another well-known complication
of setting up 3D vision is that it requires the constant intervention of experts
to look after these systems.

6.5.4 Vision-guided robots

Companies adapting to Industrial Revolution 4.0 have already implemented
robots guided by computer vision. These robots have the essential ability of
a human, such as navigation within the warehouse or mobility of products
within the industry [13]. Some applications of these robots currently perform-
ing in these industries are to pick and place the objects around the industry,
performing complex rotary die and laser cutting with accurate measurement
to sculpt into any design.
 6.6 Safety and Security Standards 145

6.5.5 Predictive maintenance

Most of the time, industries process various manufacturing processes in
harsh temperatures and environments. This may often lead to corrosion and
degradation of manufacturing equipment [5]. If not maintained properly, this
may lead to hefty maintenance costs or even halting production, leading to
severe losses. To handle this situation, these industries employ corrosion
engineers to tackle this situation. These engineers perform a constant inspec-
tion to ensure the equipment’s working condition and avoid corrosion to the
machine’s movable parts [14]. As we can see, the entire investigation process
is performed manually; hence it is restricted to human errors and the number
of metrics that can be performed. Computer vision can constantly perform
equipment maintenance or corrosion monitoring throughout the day. If the
computer vision notices any slightest depletion in the provided safety metrics,
it will immediately alert the respective departments of employees to perform
maintenance activities.
It is critical to predict maintenance emergencies across the organization,
which may incorporate heavy machines in the production line, which is
crucial for smooth operation. If the industry fails to identify these problems
earlier, it may lead to production delays or even fatal accidents. Computer
vision can help to prevent these errors because it intimates the respective
departments within the industry before it gets too late. Computer vision
performs the alert with the provided metrics set by the maintenance engineers.
The computer vision system performs these operations by comparing the
current high-resolution image with the pre-existing data in the database for
any disruption provided by the safety metrics. Machine vision also reduces
the manual intervention of maintenance checks. Limiting manual intervention
will reduce the wear and tear that may occur during the regular maintenance
process.

6.6 Safety and Security Standards

The risk of injury is much higher in manufacturing industries since some
spots within the manufacturing pipeline are prone to dangerous conditions.
Employees not adhering to safety standards and protocols can be seriously
injured or have a fatal accident. This problem majorly arises with the
employee’s failure to follow protocols.
The risk of getting injured is much higher in the manufacturing industry.
Combined with noncompliant working standards will cause serious problems
146 The Contribution of Computer Vision in the Manufacturing Industries

[16]. The manufacturing industries must follow the safety protocols given
by government bodies. Government bodies or safety agencies must adopt
penalization for those who fail to meet these standards.
Currently, most manufacturing industries perform this monitoring process
with the help of cameras installed at specific locations. Humans manually
complete this task, and they also have to keep their gaze on the video stream.
This manual process is liable to human error and also the error of negligence.
This can be automated using AI-powered systems that can monitor the entire
industry from the entrance to the exit. Based on the provided criteria, the
system will monitor constantly to check for any violations. If any employee
commits a minor violation of the company’s safety protocol, the system will
inform the concerned manager about the infringement, and they can take the
necessary action immediately.
As shown in Figure 6.5. The process of digital image processing con-
sists of various components that work in sequence to recognize the object.
Throughout the process, the knowledge base will be enumerated with data

Figure 6.5 Fundamental steps in digital image processing in machine vision.

 6.8 Barcode Analysis 147

that will have fewer errors in the recognition of images than in the previous
epoch. This process is called learning.
In this way, the manufacturing process becomes secure, and fewer fatal
accidents occur, thereby increasing the overall throughput of the company in
the long run.
In case of emergencies like an accident, the computer vision technology
will immediately alert the managers and staff of the company for their
intervention. Also, it can automatically turn off the required machinery to
stop further escalation of the situation. This will ensure the safety of other
employees, thereby reducing the overall risk of this situation escalating.
The safety of the employees is the most important concern for industries,
especially for manufacturing industries worldwide. Vision-based systems
powered by AI can aid in reducing the overall chances of any unwanted
accidents in the high-risk prone area within the industries [17]. As discussed
earlier, implementing computer vision in monitoring events can effectively
detect the employees who do not comply with the safety protocol rather than
making some employees sit and monitor the employees who are prone to
errors.

6.7 Packaging Standards

The most common procedure in almost all manufacturing industries is to tally
the items before packing them into a lot. Usually, counting the number of
products manufactured is daunting since industries produce the products at a
higher rate. Manually counting these products most often leads to errors. This
is a severe problem in retailers or pharmaceutical industries. Deployment of
computer vision in the packing process helps in making near 100 percent
accuracy in counting and packing.
Another use case of vision-based systems in the same product manufac-
turing line is that after the packages are packed, the system can check for
any damage present in the packaging. Damaged packages are diverted into a
particular product pipeline, where the items are repacked, thereby increasing
the probability of the package reaching the customer safely. They can also be
set to check the packing standards of old goods before leaving the warehouse.
Maintaining the package standards is vital in ensuring the product’s
longevity in the market. Computer vision will ensure that all the products
reaching out to the industry will be of top quality. Apart from the stated
benefits, it can also constantly collect data related to the level of problems
involved with a particular spot and keep track of the colors, measurements,
or other tools to effectively keep track of the products over time.
148 The Contribution of Computer Vision in the Manufacturing Industries

6.8 Barcode Analysis

Another critical application of computer vision is barcode recognition. Bar-
code provides a visual representation of data which is in the form of a
machine-readable configuration. Barcodes represent data by differing the
parallel lines’ shapes, widths, sizes, and spacing between them. In man-
ufacturing industries, barcodes are used to verify the various attributes of
the end product. These may include the date of manufacturing, various
information related to models, sorting out spare parts for a particular product,
etc. Verifying each bar code by hand takes a lot of man hours, is prone to
mistakes due to wrong entries, and is expensive. With the help of computer
vision, we can easily verify the barcodes along the product manufacturing
pipeline with fewer errors and a convenient operation cost over a long
period [18].
The growth of vision-based systems used to read barcodes and QR codes
has been increasing in recent years. They are handy in many instances,
especially in warehouse management, where the products are quickly sorted
out using the barcode. The computer vision system is also deployed at the exit
of production lines to check for faulty barcodes, texts, and QR codes present
on their products. This may be due to technical faults such as faulty printers,
scratches, and smudges that may confuse shop owners or consumers during
the billing process. Sorting out these defective products within the production
line makes it easier for the manufacturers to sort out the defective products
and, simultaneously have more satisfied customers.

6.9 Inventory Management

Currently, computer vision is mainly used in inventory management. Indus-
tries that employ computer vision for inventory management need around
5 percent of the original workforce that was required previously. Vision-
based systems can aid in keeping track of warehouse stocks [6]. This can
also inform the manufacturers about the raw material deficit in advance. The
computer vision system also avoids the human errors that can happen when
counting the stock [19].
In a large warehouse, it is often difficult to locate the stock of any
particular product. This is where the computer vision system shines. We
can easily locate the stock with the help of bar codes provided for each
product [23].
 6.12 Computer Vision during the Covid-19 Pandemic 149

There are many massive warehouses around the world. These warehouses
consist of millions of items in stock and are constantly bombarded with
incoming and outgoing stockpiles. With computer vision, companies can do
inventory assessments proactively. Moreover, the managers can be intimate
in advance of any depletion of product quantity.

6.10 Optimizing Supply Chains

Optimizing the supply chain process periodically is beneficial for any manu-
facturing industry to maintain the upcoming demands while simultaneously
satisfying the consumers by making the product available year-round. While
human intervention is required at the exact critical times for most of the
process, the computer vision system can handle the situation effectively [21].
Several industries have implemented computer vision in various depart-
ments within their manufacturing pipeline. Some of these are managing
inventories, warehouse management, and improving the efficiency of overall
organizational operations [22].

6.11 Quality Inspection with Computer Vision

Another important implication of machine vision systems in the field man-
ufacturing industry is to inspect the quality of end products. When indus-
tries produce hundreds of products within a minute, each product must be
inspected for defects and malfunctions. When this process is done man-
ually, it takes a lot of time with high chances of possible human error
[16]. As a result, many manufacturers are developing their deep-learning
computer vision systems specifically designed for their products to handle
their product quality controls [20]. Besides this, computer vision systems are
also cost-effective in handling the quality inspection process with much better
performance outcomes.
This makes it possible to check all the products for quality verification
when produced in large quantities rather than only a small sample size from
a large lot in the manual verification process.
For example, a hot sauce maker checks the placement of labels at 1000
per minute. This is impossible for any human being to perform these tasks at
that speed and accuracy. Therefore, the computer vision system provides the
best performance out of the money spent. Table 6.3 summarizes the object
defect detection using computer vision techniques, defect, and accuracy.
150 The Contribution of Computer Vision in the Manufacturing Industries

Table 6.3 Object defect detection using computer vision.

Author Country Object Technique Defect Accuracy
(year)
Lerones Spain Brake 3D Circularity, defective 80
et al. disk structured grinding, hard masses,
(2005) light feather edges
Blasco Spain Citrus Region Peel 95
et al. oriented
(2007) segmenta-
tion
Peng China Float Texture Bubbles, lards, optical 89
et al. glass reverse distortion, wholesome,
(2008) subtraction diseased
Yang USA Electric Blob analy- Deckle edge, back 96.7
et al. con- sis and PSO cracks, side cracks,
(2009) tacts eccentricity
Mery Chile Corn SVM Size, hedonic scale, pro- 96
et al. tortillas duction level
(2010)
Gadelmawl Saudi Spur Tolerances Outer diameter, diame- 99
et al. Arabia gears ter pitch, circular pitch
(2011)
Perng China LED Thresholding Missing components, 95
et al. incorrect orientations
(2011)
Razmjooy Iran Potato SVM and Size, color 96.86
et al. ANN
(2012)
Možina Slovenia Pharmac PCA Debossing shallow 84
et al. eutical
(2013) tablets
Lin and China Tile Corner Alignment 90.5
Fang conver-
(2013) gence and
clustering

6.12 Computer Vision during the Covid-19 Pandemic

One of the current implications of computer vision is during the Covid-19
pandemic. The pandemic caused many manufacturers to halt production and
resume their activities based on the restrictions imposed by the government
on their location. Most government authorities provide the general guidelines
that have to be followed by all the employees within the company. If the cases
start to rise among the employees of a particular company, then government
 6.13 Computer Vision in the Automotive Industry 151

officials can halt the company until the spread of the virus slows down within
their employee circle. This can cause disruption in the production process and
serious loss.
So, many companies have incorporated computer vision systems in their
operational areas, which detect whether the employees are putting the mask
on throughout their stay and maintaining a social distance of 6 feet. This
system intimates the manager when any of their employees are not following
the rule, and they can take further action to prevent such an incident from
happening.
Many government and private organizations used optical character recog-
nition (OCR) technology to process and verify applications remotely during
the lockdown period.
Apart from mask detection and social distancing monitoring applications,
they have also implemented a computer vision system that can identify people
with high body temperature with the help of an infrared camera. This system
can detect temperatures more accurately and is much more efficient than a
manual temperature monitoring system, which is time-consuming.

6.13 Computer Vision in the Automotive Industry

The automotive industry is the de facto example of how efficient computer
vision and automation can work together to produce many products with
fewer human interventions. For example, it is known that Tesla, an electric
car manufacturer, incorporated automation in their production process, which
accounts for over 70 percent. The fourth Industrial Revolution, currently
in progress, has revolutionized technology such as sensing, processing, and
integration technologies between different systems along the production line,
making it the best to interrupt less production.
For the most part, the automation process is involved in building the
products, while the machine vision is incorporated with quality control,
employee management, and warehouse management systems.
Besides these various advancements, most industries have not designed
computer vision for horizontal or vertical integration into their existing design
as a top priority. Even today, many automotive companies have deployed
computer vision to work on individual tasks. To have a near-zero defect in
the manufacturing pipeline, computer vision systems can be integrated with
the existing systems that work in parallel.
The complexity of the automotive industry comes from its manufacturing
procedures and supply chain network around the globe. Some of the activities
152 The Contribution of Computer Vision in the Manufacturing Industries

Figure 6.6 Computer vision in automobile industries.

that are required to perform regularly are manufacturing various spare parts
for all the car models in use, special components arranged based on the
current stock, material shaping, welding operations, and procedures related
to final assembly are carried out before the distribution of automobiles into
the market all around the world [27].
Machine vision systems can be incorporated into entire automotive man-
ufacturing procedures, including grading quality, welding, positioning and
handling of materials, and various other activities. They can reduce comple-
tion times, reduce faults at delivery and enhance productivity by approaching
zero-fault production.
The automotive industries are the ones that extensively use computer
vision in their manufacturing process. Due to the demanding procedures
involved in the manufacturing of an automobile, they have implemented
computer vision across various shops within their manufacturing pipeline,
depicted in Figure 6.6. In the automobile industry, the factory is classified
into four stages of production. They are the press shop, the body shop, the
paint shop, and the final assembly shop.

6.13.1 Press shop

For each vehicle, the metal components are produced in this phase. The
number of metal components produced is about 300. Detecting the defect
 6.13 Computer Vision in the Automotive Industry 153

in these metal components is difficult and consumes huge human effort. But
computer vision helps us to automate this difficult defect detection on the
metal components. It can also detect various defects. Ferguson et al. proposed
faster recurrent CNN (convolutional neural network) or faster R-CNN to
detect the defect in the aluminum casting surfaces [28]. Herger et al. proposed
the decision tree algorithm (DT-algorithm) to detect door parts’ cracks during
production time [27]. In detecting cracks in the doors, he used six stages to
identify and remove the cracks. He obtained an accuracy of 94 percent.

6.13.2 Body shop

Materials from the press shop are used for the production of chassis. The
body shop uses raw materials from various industries or press shops as a
source. In this phase, the welding of materials and assembling these materials
together. The real success of this welding operation is directly proportional to
the quality of the car body. This process of welding and assembling is time-
consuming and consumes more manual power. Nevertheless, computer vision
technology makes this welding and assembling process more accessible.
Novakovic et al. suggested a 3D vision-based system for inspecting, weld-
ing, and assembling the raw materials [27]. He used the 3D vision concept in
the autonomous robotics inspection to inspect the weld and automate robotic
arms. He obtained an accuracy of approximately 0.15 mm.
For the door panels, the machine vision concept is used in the form of
an edge detector to identify the spots to be welded and thereby increasing
the accuracy of the welding. The welding spots defect by calculating the
distance between the panels and the end-effectors (part of computer vision
technology). The entire body part of the car body is tested by considering
the accuracy, distance, and collision factor. The hard or rough surface pro-
duced during welding is converted into a smooth surface using the grinding
end-effectors together with robotic arms.
The replacement of multi-hole abrasives is done manually and requires
more accurate alignment. This drawback can be overcome by using com-
puter vision technology by calculating the latent offset of the multi-hole
abrasive.
The vision-based defect detection system is used to monitor the welding
defect. It uses three layers of CNN (convolutional neural network) to apply
filters, LSTM (long short-term memory), and FC (fully connected), which is
further connected onto the dense layer with activation functions like SoftMax,
Sigmoid, and so on, based on the problem. The CNN layer is used to extract
154 The Contribution of Computer Vision in the Manufacturing Industries

spatial features, the FC layer is used to perform classification, and the LSTM
layer is used to perform feature retrieval.

6.13.3 Paint shop

This paint shop workstation produces attractive finesse and body armor,
which protects the car body from external weather conditions. Performing
this task of painting is very time-consuming and requires more manpower.
But, with computer vision, we can automate these processes. In the painting
phase, robotic arms are used to deposit colors and perform curing operations
of the car body. In some other robotic arms are used to perform painting
operations. When we have a small size defect (0.2 mm) in the car body and
also for monitoring the painting defects on the lines, edges, and corners, we
require more manpower, cost, and time. To overcome this challenge, we use a
vision-based system that uses light patterns to detect the defects. QEyeTunnel
is a machine vision system that identifies the defect in less than 0.15 s.
Sometimes, uncertain vision conditions can affect the lighting. To tackle this
problem, SVM (support vector machine) is used to predict the quality of
illuminations and thereby ensure suitable conditions for recognizing lighting
patterns. In the latter stages, multi-camera vision systems are used to detect
the defects on the painting surface and the car body. Irregular lighting is
corrected using the TinyDefectRNet method, which in addition identifies
abnormalities on the body surfaces and controls the overall painting quality.

6.13.4 Final assembly shop

The final phase of the automation industry is assembling the painted body
with the internal and external parts. This phase is also referred to as the final
assembling phase. This phase’s main activity is verifying the activities done
in the entire lifecycle of the automation industry. This verification process
by trained and experienced persons takes more time and cost. However, using
computer vision technology, we can automate this process and make verifying
the correctness of assembled products easier.
A vision-based statistical model is used to find the rare faults in the
assembly process. In this method, the rare fault is identified at first, and
the recovery steps are carried out to eliminate this fault. But this method
cannot identify a fault less than 0.02 mm. To overcome this problem, we use
random forest classifier to identify rare faults, provide better accuracy and
performance, and improve the classification quality.
 6.14 Computer Vision Performance Metrics 155

To get personalized products, we need to arrange the products before it

arrives at the final assembly stage. This process of sorting is done manually by
placing wheels. Nevertheless, this manual method takes more time and also
more human power. To overcome this problem, we use automated wheels to
sort the products with minimum time and manpower consumption.
The glass windows to the car body are one of the external parts that must
be fitted. An in-line vision-based system inspects the quality and quantity of
glue that fits the glass windows to the car body. Computer vision has made
the daunting task of quality checking much more efficient and faster [24].

6.14 Computer Vision Performance Metrics

The accuracy of identifying the object is a crucial task in computer vision
systems. Performance metrics serve as a measure of assessment to evaluate
the machine’s correctness [26]. To observe the metrics, we need to look into
the following factors: image object segmentation, object tracking, and object
detection. They are intersection over union (IoU), precision, recall, and F1
score. The definition is defined below:

6.14.1 Intersection over union (IoU)

As shown in eqn (6.1) and Figure 6.7, it is a principle metric whose value is
calculated by dividing the union of the areas of the predicted and ground
truth bounding boxes by the area of the overlap between them. The data
demonstrate this in practice. It reports both the total number of correctly
identified objects and the total number of false positives. When the proportion
of objects detected via IoU is more significant than 0.5, we call that true
positive (TP) result. When the number of objects detected has an IoU less
than 0.5 or has been detected more than once, the result is considered a false
positive (FP). A false negative result is obtained if no objects are found or if
objects are found but their IoU is less than 0.5. (FN).
area(Bp ∩ Bgt )
J (Bp , Bgt ) = IoU = (6.1)
area(Bp ∪ Bgt )

6.14.2 Precision
As shown in eqn (6.2) and Figure 6.8, precision is a performance metric that
measures how accurately the outputs are predicted. It returns the percentage
156 The Contribution of Computer Vision in the Manufacturing Industries

Figure 6.7 Intersection over union (IoU).

of the correct prediction.

TP TP
Precision = = . (6.2)
TP + FP all detections

6.14.3 Recall
As shown in eqn (6.3) and Figure 6.9, recall returns the percentage of how
well the model predicts all the positives.
TP TP
Recall = = . (6.3)
TP + FN all ground truths

6.14.4 F1 score
As shown in eqn (6.4), the F1 score returns the harmonic mean (HM) of
precision and recall.
Precision × Recall
F 1 score = 2 × . (6.4)
Precision + Recall
 6.14 Computer Vision Performance Metrics 157

Figure 6.8 Precision.

The object perforation of a computer vision model can also be evaluated

by calculating erroneous object detection metrics, erroneous object matching
metrics, and object correction [27].

Figure 6.9 Recall.

158 The Contribution of Computer Vision in the Manufacturing Industries

6.15 Conclusion
Industries are an essential part of the development of human civilization.
Sustainability in industries can increase our longevity and efficiency in
developing a product. Computer vision has a long way to go to contribute
to different industry domains for maximizing production and reducing the
overall carbon footprint. It also transformed the human labor force to focus on
imaginative aspects of industries while leaving the intuitive part to computers.
Machine vision has enormous implications in various industries. The current
implementation of computer vision in manufacturing sectors already provides
unparalleled convenience in industrial processes. In the future, the most
profitable industry would have leveraged computer vision in their production
pipeline to the best of their capability. Automation of industrial processes will
also reduce the overall greenhouse gas emission per product, thus guiding
the overall technological development sustainably without damaging the
environment.

References
[1] S. Dey, A. K. Singh, D. K. Prasad, and K. D. McDonald-Maier,
“SoCodeCNN: Program Source Code for Visual CNN Classifica-
tion Using Computer Vision Methodology,” IEEE Access, vol. 7, pp.
157158–157172, 2019, doi: 10.1109/ACCESS.2019.2949483.
[2] Y. H. Chen, “Computer vision for general purpose visual inspection: a
fuzzy logic approach,” Optics and Lasers in Engineering, vol. 22, no. 3,
pp. 181–192, Jan. 1995, doi: 10.1016/0143-8166(94)00045-C.
[3] W. T. Freeman, K. Tanaka, J. Ohta, and K. Kyuma, “Computer vision
for computer games,” in Proceedings of the Second International Con-
ference on Automatic Face and Gesture Recognition, pp. 100–105. doi:
10.1109/AFGR.1996.557250.
[4] M. Christie, P. Olivier, and J.-M. Normand, “Camera Control in
Computer Graphics,” Computer Graphics Forum, vol. 27, no. 8, pp.
2197–2218, Dec. 2008, doi: 10.1111/j.1467-8659.2008.01181.x.
[5] Shanghai hai yang da Xue and Institute of Electrical and Electronics
Engineers, Proceedings of 2020 IEEE International Conference on Arti-
ficial Intelligence and Computer Applicationsă: ICAICA 2020ă: Dalian,
China, June 27-29, 2020.
[6] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only
look once: Unified, real-time object detection,” in Proceedings of the
 References 159

IEEE Computer Society Conference on Computer Vision and Pat-

tern Recognition, Dec. 2016, vol. 2016-December, pp. 779–788. doi:
10.1109/CVPR.2016.91.
[7] J. Wright, Y. Ma, J. Mairal, G. Sapiro, T. S. Huang, and S. Yan,
“Sparse Representation for Computer Vision and Pattern Recognition,”
Proceedings of the IEEE, vol. 98, no. 6, pp. 1031–1044, Jun. 2010, doi:
10.1109/JPROC.2010.2044470.
[8] Devi, A., Therese, M. J., Devi, P. D., & Kumar, T. A. (2021). IoT-Based
Smart Pipeline Leakage Detecting System for Petroleum Industries. In
Industry 4.0 Interoperability, Analytics, Security, and Case Studies (pp.
149-168). CRC Press.
[9] Rajmohan, R., Kumar, T. A., Pavithra, M., & Sandhya, S. G. (2020).
Blockchain: Next-generation technology for industry 4.0. In Blockchain
Technology (pp. 177-198). CRC Press.
[10] X. Zhang and S. Xu, “Research on Image Processing Technology of
Computer Vision Algorithm,” in Proceedings - 2020 International Con-
ference on Computer Vision, Image and Deep Learning, CVIDL 2020,
Jul. 2020, pp. 122–124. doi: 10.1109/CVIDL51233.2020.00030.
[11] G. Wang, H. Gao, Q. Li, and C. Fang, “Surface deformations mea-
surement based on computer vision technique,” in 2010 International
Conference on Machine Vision and Human-Machine Interface, MVHI
2010, 2010, pp. 191–194. doi: 10.1109/MVHI.2010.130.
[12] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image
Recognition,” Dec. 2015, [Online]. Available: [Link]
.03385
[13] C. S. Sharp, O. Shakernia, and S. S. Sastry, “A vision system for landing
an unmanned aerial vehicle,” in Proceedings 2001 ICRA. IEEE Interna-
tional Conference on Robotics and Automation (Cat. No.01CH37164),
pp. 1720–1727. doi: 10.1109/ROBOT.2001.932859.
[14] S. Xu, J. Wang, W. Shou, T. Ngo, A.-M. Sadick, and X. Wang,
“Computer Vision Techniques in Construction: A Critical Review,”
Archives of Computational Methods in Engineering, vol. 28, no. 5, pp.
3383–3397, Aug. 2021, doi: 10.1007/s11831-020-09504-3.
[15] Bo Zhang, “computer_vision_vs_human_vision”, 9th IEEE Int. Conf.
on Cognitive Informatics.
[16] Agin, “Computer Vision Systems for Industrial Inspection and Assem-
bly,” Computer (Long Beach Calif), vol. 13, no. 5, pp. 11–20, May 1980,
doi: 10.1109/MC.1980.1653613.
160 The Contribution of Computer Vision in the Manufacturing Industries

[17] Tai-Hoon Cho and R. W. Conners, “A neural network approach to

machine vision systems for automated industrial inspection,” in IJCNN-
91-Seattle International Joint Conference on Neural Networks, pp.
205–210. doi: 10.1109/IJCNN.1991.155177.
[18] Kumar, K. Suresh, T. Ananth Kumar, A. S. Radhamani, and S. Sun-
daresan. “Blockchain technology: an insight into architecture, use cases,
and its application with industrial IoT and big data.” In Blockchain
Technology, pp. 23-42. CRC Press, 2020.
[19] R. P. Kruger and W. B. Thompson, “A technical and economic assess-
ment of computer vision for industrial inspection and robotic assembly,”
Proceedings of the IEEE, vol. 69, no. 12, pp. 1524–1538, 1981, doi:
10.1109/PROC.1981.12199.
[20] D. Geronimo, J. Serrat, A. M. Lopez, and R. Baldrich, “Traffic
sign recognition for computer vision project-based learning,” IEEE
Transactions on Education, vol. 56, no. 3, pp. 364–371, 2013, doi:
10.1109/TE.2013.2239997.
[21] Gonzalez and Safabakhsh, “Computer Vision Techniques for Industrial
Applications and Robot Control,” Computer (Long Beach Calif), vol.
15, no. 12, pp. 17–32, Dec. 1982, doi: 10.1109/MC.1982.1653913.
[22] E. Murphy-Chutorian and M. M. Trivedi, “Head Pose Estimation in
Computer Vision: A Survey,” IEEE Transactions on Pattern Analysis
and Machine Intelligence, vol. 31, no. 4, pp. 607–626, Apr. 2009, doi:
10.1109/TPAMI.2008.106.
[23] J. F. S. Gomes and F. R. Leta, “Applications of computer vision tech-
niques in the agriculture and food industry: a review,” European Food
Research and Technology, vol. 235, no. 6, pp. 989–1000, Dec. 2012,
doi: 10.1007/s00217-012-1844-2.
[24] C. Haccius and T. Herfet, “Computer Vision Performance and Image
Quality Metricsă: A Reciprocal Relation,” Jan. 2017, pp. 27–37. doi:
10.5121/csit.2017.70104.
[25] L. Tychsen-Smith and L. Petersson, “Improving Object Localization
with Fitness NMS and Bounded IoU Loss,” Oct. 2017, [Online].
Available: [Link]
[26] R. Padilla, S. L. Netto, E. A. B. da Silva, and S. L. Netto, “A Survey on
Performance Metrics for Object-Detection Algorithms Compression of
Power Systems Signals View project A Survey on Performance Metrics
for Object-Detection Algorithms”, doi: 10.1109/IWSSIP48289.2020.
[27] Konstantinidis, Fotios K., Spyridon G. Mouroutsos, and Antonios-
Gasteratos. “The Role of Machine Vision in Industry 4.0: an automotive
 References 161

manufacturing perspective.” In 2021 IEEE International Conference on

Imaging Systems and Techniques (IST), pp. 1-6. IEEE, 2021.
[28] Longsheng Fu, Yali Feng, Yaqoob Majeed, Xin Zhang, Jing Zhang,
Manoj Karkee, Qin Zhang,Kiwifruit detection in field images using
Faster R-CNN with ZFNet, IFAC-PapersOnLine, Volume 51, Issue 17,
2018, Pages 45-50, ISSN 2405-8963, [Link]
018.08.059.
 7
Waste Management 4.0: An Industry
Automation Approach to the Future Waste
Management System

M. Julie Therese1 , P. Dharanyadevi2 , A. Devi3 , and Christo Ananth4

1 Department of ECE, Sri Manakula Vinayagar Engineering College, India
2 Department of CSE, Puducherry Technological University, India
3 Department of ECE, IFET College of Engineering, India
4 Samarkand State University, Uzbekistan

E-mail: julietherese88@[Link]; dharanyadevi@[Link];

deviarumugam02@[Link]; [Link]@[Link]

Abstract
Industry 4.0 is a trending revolution in the present society, especially in
the circular economy system. The latest advances in the domain of data
science, artificial intelligence (AI), robotics, the Internet of Things (IoT),
big data, augmented reality, and virtual reality have opened the door for
smart waste treatment systems. In India, municipal waste generation has
increased dramatically during the past several decades, partly due to the
country’s rapid population expansion. The amount of waste generated in
cities across India has risen from 6 million tons in 1947 to 48 million tons
by 1997, and it is anticipated to reach 300 million tons per year by 2047.
As a result, several substantial efforts must be taken to resolve the waste
management problem. The current system arbitrarily collects waste. As a
result, some places are occasionally left unattended, resulting in an unpleasant
odor and a threat to public health, as the smell of garbage can potentially
be harmful to little children and old-aged people. This chapter focuses on
overcoming the hazards by adopting a smart waste management system that
includes a cyber−physical system (CPS), which makes use of computers

163
164 Waste Management 4.0: An Industry Automation Approach to the Future Waste

to monitor the waste management process, and the Internet of Everything

(IoE), which is a type of intelligent computing that connects things, people,
services, data, etc., by making use of machine-to-machine communication
protocols. These approaches make the waste treatment mechanism smart and
could easily enable the waste-the-energy approach, converting energy into
electricity, heat, or fuel.

Keywords: circular economy, cyber−physical system, Industry 4.0, Inter-

net of Everything, machine-to-machine communication, waste-to-energy
approach

7.1 Introduction
Smart factory is the current trend in Industry 4.0, wherein CPS makes decen-
tralized decisions by monitoring the physical processes of the factory. These
cyber−physical systems comprise production facilities, smart machines, and
storage systems, which possess the capability of independency in information
exchange, triggering activity, and managing each other without human inter-
ference. As of this, the physical systems in the Internet of Things interact and
coordinate with each other and humans simultaneously through the wireless
web. Industry 4.0 is prominently known for its sustainability to support the
circular economy.
In order to analyze how the Fourth Industrial Revolution arose, the idea
about the previous three industrial revolutions and how evolution has changed
the lives of humans and the world should be studied. A brief note on previous
industrial revolutions is shown in Figure 7.1.
The First Industrial Revolution occurred in 1784 when a watershed was
the first mechanized loom. It also introduced steam power and the factory
system, which increased production notably in the inclusion of machines.
In 1870, the Second Industrial Revolution laid the foundation for the huge
production of electricity [1]. This paved the way for the assembly line;
the industrial sector grew exponentially over this century. In 1969, the
Third Industrial Revolution came into existence, where the enhancements
in computing paved the way for software and programming development,
and progressive automation was introduced in this era. Then came the Fourth
Industrial Revolution, with a much wider scope for future trends. The com-
bination of the digital, biological, and physical domain interactions makes
Industry 4.0 vitally different from previous revolutions. The evolution of
Industry 4.0 is visualized above.
 7.1 Introduction 165

Figure 7.1 Evolution of Industry 4.0.

According to International Solid Waste Association (ISWA), Industry

4.0 comprises AI, automation and big data, IoT, etc., which significantly
transfigures waste management [2]. Further, combining technologies creates
new chances to accomplish high treatment standards, automate recycling
tasks, reduce or prevent and even cut off waste from particular streams
and sectors. The automated recycling tasks quoted above are the robotic
recycling that will become mainstream in the markets in the next 10 years
because of its materials recovery facilities (MRFs), accuracy, flexibility, and
market adaptation [3]. More innovations are expected in waste management,
which involves the technologies mentioned that ease the development of
Industry 4.0.
The first section of this chapter gives an introduction to Industry 4.0.
The subsequent section explores the cyber−physical system (CPS) and its
advantages and disadvantages. In the third section of this chapter, we will
discover the Industry 4.0 environment that is IoE, which is an amalgamation
of IoT, industrial Internet of Things (IIoT), Internet of Services (IoS), Internet
of Data (IoD), and Internet of People (IoP). Having discussed the Industry
166 Waste Management 4.0: An Industry Automation Approach to the Future Waste

4.0 environment, the next section will highlight the challenges faced by the
waste management industry. Applications of CPS in the waste management
industry will be dealt with in the next section. Section 6 of this chapter will
discuss the influence of Industry 4.0 on the waste management industry.
Although the studies prove that there would be a significant development
in the waste management industry by implementing Industry 4.0, there are
barriers that remain challenging in the execution phase; this is discussed in the
next section. And finally, the chapter concludes in the last section discussing
the future scope of Industry 4.0 in the waste management industry.

7.2 Exploring CPS

CPS is a critical component in the implementation of Industry 4.0 concepts
in supporting the growth of smart manufacturing. The primary purpose of
the CPS is to generate continuous feedback in real time, self-adapt and self-
optimize themselves, and control the system’s entire physical process [4].
CPSs are systems that link the physical world with the digital world using
controllers and sensors. The integration of networking, computing power, and
physical processes in the CPS creates the digital representation of a physical
entity. CPS has its benefits and drawbacks and is discussed below.

7.2.1 CPS
• Integration: Integration occurs in a system when the CPS is connected
via sensor networks to the cloud. The information delivered to the CPS
is tremendous, which is passed to the cloud, a data center for processing
[5]. Also, some physical systems are self-decision-making, making them
more intelligent.
• Interaction between human and CPS: Decentralized logistic choices
are formed, which modifies the traditional production process in the
plant. This brings forth improvement in the manufacturing process by
combining human resources and the accurate information from the plant
with a decision-making system that relies on innovative scenarios.
• Managing uncertainty: Some physical systems are intelligent. They
were created with the ability to self-adapt, self-optimize, and self-
configure. Moreover, most importantly, they can grasp how to fit in and
adjust to a given circumstance. They can reduce uncertainty to a large
amount as the sensor always works, evolves, and learns the environment.
Big data analytics is a term used to describe the analysis of large amounts
 7.3 Industry 4.0 Environment 167

of data [6]. Thousands of sensors feed that data board, and the physical
system comprehends based on this data; this is usually executed by
comparing the circumstance to the past or historical data and responding
accordingly to the condition.
• Better system performance: The CPS is real-time, and they can self-
adapt in real time, which gives manufacturers a significant advantage
in ensuring business continuity across all manufacturing processes and
related activities. Physical systems can gain a comprehensive under-
standing of how to change themselves in every condition since they
interact closely with sensors and other related data.
• Scalability: The cloud is connected to the CPS. The cloud provides
tremendous stability, and CPSs can scale themselves based on demand
or requirement by leveraging the cloud’s capacity.
• Flexibility: The CPS’s cloud marketing and wireless sensor networks
add much value to deciding the flexibility in the manufacturing process.
Because the CPSs are self-configurable and adaptable, if the equipment
fails, the physical system can figure out what went wrong and adjust
itself to perform a particular task successfully.
• Faster response time: CPS permits proactive breakdown (i.e., predict-
ing when a machine will break down and take the required procedures
to prevent it from failure). System failure is possible as the machine’s
components are erratic or sometimes overloaded. As a result, CPSs
can enable early failure prediction and understanding and optimizing
resource consumption in production.

7.2.2 Drawbacks of CPS

While the CPS offers numerous advantages, they also have disadvantages.
• Creating CPSs or incorporating CPSs into the manufacturing process is
too expensive.
• Small- and medium-sized firms could not afford the considerable finan-
cial commitment required. Even huge corporations must audit the
production process [7].
• Numerous underlying infrastructures, processes, and business opera-
tions must be optimized for CPSs to work as intended.
• Furthermore, in order to provide seamless communication between
CPSs, network features like real-time and low-latency networking must
be enabled at the factory.
168 Waste Management 4.0: An Industry Automation Approach to the Future Waste

7.3 Industry 4.0 Environment

The concept of a smart factory is essential to the Industry 4.0 environment.
Smart buildings, intelligent tools, social web, business web, and smart logis-
tics such as autonomous vehicles, smart grids, and innovative mobility are all
part of the Industry 4.0 environment, as shown in Figure 7.2.
As shown in Figure 7.3, the critical elements of the Industry 4.0 environ-
ment are the Internet of Things (IoT), Industrial Internet of Things (IIoT),
Internet of Services (IoS), Internet of Data (IoD), Internet of People (IoP),
and Internet of Everything (IoE). The explanations are given below.

Industrial Internet
of Things (IIoT)
Internet of
Everything (IoE)

Figure 7.2 Industry 4.0 environment.

KEY ELEMENTS OF INDUSTRY 4.0

Internet of Industrial Internet of Internet of Internet of Internet of

Things (IoT) Internet of Services (IoS) Data (IoD) People (IoP) Everything
Things (IIoT) (IoE)

Figure 7.3 Critical elements of Industry 4.0.

 7.4 Challenges in the Waste Management Industry 169

IoT: IoT is a computer concept that entails linking everyday physical prod-
ucts, such as television, washing machine, refrigerators, toasters, etc., to
the Internet and allowing them to speak with one another [9]. To provide
a flawless experience, elements such as high-speed internet connectivity,
network infrastructure, and shared communication protocols are taken into
account. Machine-to-machine (M2M) communication is the term used in this
area to describe the process of ensuring that all devices can understand and
communicate with one another.
IIoT: The IIoT refers to a network of interconnected devices or instruments
for industrial computing applications to improve industrial and manufactur-
ing processes, such as advanced robotics and machines that can develop and
implement a wide range of products and other related processes [10].
IoS: IoS is a global marketplace for web-based software provided as a
service. IoS covers benefits such as blockchain, which is regarded as a game
changer in supply chain management. IoT enabled by IoS results in disruptive
innovation.
IoD: In Industry 4.0, billions of sensors will be connected online, generating
vast amounts of data such as distance, wetness, temperature, position, etc. All
these data must be processed using a big data analytics tool to obtain useful
insights and factors. As a result, IoD is based on the massive amounts of data
generated by billions of IoT devices. When big data analytics methods are
used to find trends in the data stream, these data streams offer vast potential
for industrial applications.
IoP: IoP is a revolutionary new internet concept in which persons and their
personal gadgets are treated as active internet elements rather than as end
users of applications. IoP is based on a decentralized network like blockchain,
where a single body does not hold or control data and information. It
aspires to be an internet “for the people, of the people” rather than one that
unknowingly compromises people’s privacy.
IoE: IoE is the unification and complexity of people, processes, data, and
things in a single network system [11]. The IoT, IIoT, IoS, IoP, and IoD are
combined to form the IoE. This is the most promising technology for Waste
Management 4.0.

7.4 Challenges in the Waste Management Industry

For waste management to be successful, several challenges must first be
surmounted. These challenges include but are not limited to a deficiency in
170 Waste Management 4.0: An Industry Automation Approach to the Future Waste

sufficient data, the requirement to guarantee the quality of recycled materials,

and the recuperation of waste energy. Figure 7.4 presents a graphical repre-
sentation of the difficulties that must be conquered in the waste management
sector.
• Reusing and recycling waste materials
The first and most crucial challenge in waste management is determining
whether resources may be reduced, recycled, or reused. The procedure
becomes more efficient due to the exact identification of the process.
• Trash gathering and transportation
Biodegradable and non-biodegradable wastes are the two most common
types of waste. Food waste, such as banana peels, rotten fruits, and veg-
etables, is considered biodegradable. Plastics and glasses are examples
of non-biodegradable materials. These are the standard categorizations;
however, in certain countries, it extends beyond that and separates
paper, glass, metal, and plastics that originate from consumer waste
from households. However, industrial garbage is a different scenario,
including hazardous materials and radioactive waste [13]. Gathering the
garbage to be stored, transported, and disposed of needs a great deal of
effort, and on a massive scale, people are hesitant to organize; this is one
of the most challenging problems in trash management.
• Garbage removal
There are two garbage removal methods; the first is burning it, which
releases hazardous gases, ash, and particulates into the atmosphere.
The most effective way is to bury the garbage. Each method comes
with its own set of difficulties. For example, the atmospheric approach
emits many hazardous gases such as carbon monoxide, carbon dioxide,
sulfur dioxide, etc. When waste is dumped into the landfill or the ocean,
dangerous metals, such as cadmium, barium, etc., are released, harming
the ecosystem and affecting human health.
• Dumping rubbish has a negative influence on the environment
Dumping garbage into the environment produces more devastating prob-
lems, such as rapid differences in humidity and chemical composition,
or alters the number of gases present in a given ecosystem, all of which
significantly impact the lifespan of all living species [14]. For example,
algae thrives in trash dumps because all the rich nitrogen sources are
present.
 7.6 Influence of Industry 4.0 on the Waste Management Industry 171

Figure 7.4 Challenges in waste management system.

• Waste dumping harms human health

Many people have had real-life experiences; people who reside near a
waste-dumped landfill are exposed to high levels of microorganisms,
such as flies, insects, rats, and bugs that survive on trash. These insects
come to feed on the garbage and then land on people’s bodies, spreading
infections quickly. This is a significant problem that Industry 4.0 experts
must overcome.

7.5 Applications of CPS in the Waste Management

Industry
Industry 4.0 and CPS are two new buzzwords that have earned much interest
from researchers and manufacturers. Advanced technologies are made avail-
able due to these terms that provide potential solutions and enhancements for
future smart waste management systems [15].
• Monitoring the level of toxicity during collection and transport
• Determination of the most environmentally friendly garbage degradation
methods
• Recognizing functional components in garbage
• Improving the effectiveness of waste management operations
172 Waste Management 4.0: An Industry Automation Approach to the Future Waste

7.6 Influence of Industry 4.0 on the Waste Management

Industry
The goal of Industry 4.0 in the waste management system is to empower
its influence to reduce waste, boost resource utilization, and build a more
sustainable, eco-friendly, and cleaner planet [16]. The following are some of
the impacts of Industry 4.0 on the waste management industry.
• Consumer products with an eco-first perspective The primary aim of
this impact is to create an eco-friendly environment. When a consumer
uses and discards a product, it can be reused, recycled, or used as a raw
material for another product. As a result, Industry 4.0 focuses on making
all discarded goods much more eco-sustainable.
• Adoption of the circular economy at a benefit
The circular economy will benefit the planet, people, and busi-
nesses alike. It gives new commercial practices, recycling and reusing
resources, and new business models. As a result of the vast volume of
data generated by Industry 4.0, the adoption of the circular economy
will be accelerated [17]. It can be created by repurposing materials and
extracting valuable components or elements from waste, resulting in a
more environmentally friendly and long-lasting product for consumer
demand.
• Greenhouse gas emissions from the waste management industry are
being reduced
Big data analytics and CPS will monitor every unit or entity in real time.
It can also self-adjust, self-optimize, and self-regulate its whole man-
ageable workflow (for example, if CPSs are in charge of a factory, they
will constantly monitor and enhance the factory’s efficiency, as well as
control the waste management sector’s greenhouse gas emissions [18]).
• Large-scale garbage dumping sites are being phased out
Many trash heaps are noticeable, especially in underdeveloped countries,
where hundreds of trucks arrive and dump their waste, resulting in
trash mountains. This is incredibly dangerous because once the trash
mountain is formed, all of the highly combustible gases build up, and
all it takes is one spark to blow up the entire mountain. This will
be overcome by Industry 4.0 and CPS because these waste manage-
ment systems will autonomously handle waste, as well as self-driving
garbage trucks and smart garbage cans; this leads to an antibacterial
environment.
 7.6 Influence of Industry 4.0 on the Waste Management Industry 173

• E-waste is being reduced significantly

The CPS will be able to accurately handle and manage e-waste in terms
of determining what resources can be extracted, reused, and recycled,
and this will undoubtedly result in a far more sustainable e-waste
management ecosystem. As a result, people can limit their e-waste,
and businesses can focus on incorporating more recycled materials into
their products to create an even more sustainable and environmentally
friendly supply chain.
• Increase the use of waste-to-energy technologies
In some countries, trash is used to generate energy, which is then used to
generate electricity. This drastically reduces the amount of landfill space
used for rubbish disposal, and with the advancement of new technolo-
gies, one can expect to see things like filters that prevent ashes, dust, and
extremely poisonous vapors and gases from reaching the atmosphere.
This will also be incredibly sustainable because when waste is burnt, it
turns into ashes that can be recycled and turned into bricks, which can
then be used to construct houses [19]. As an outcome, a great deal of
value-added thinking is required to ensure that waste management is far
more sustainable than it is presently.
• Significant reduction in marine waste
CPS can help scientists determine what contaminants are present in the
water and how they affect the entire marine ecosystem, from fish to
marine mammals and corals. They will also be able to track in real time
where these pollutants are leaking into the oceans, and penalties or taxes
may be levied to ensure that all of those who contribute to waste are
causing offense and that their funds can be used to construct sustainable
systems [20]. Ocean cleanup is one of the new business models that can
be developed to ensure that the ocean is a healthy ecosystem for marine
life.
• Food waste reduction
Food waste is a significant issue, with millions of tons of food thrown
away each year. A far more efficient operation in logistics, distribution,
storage, and production can be established using Industry 4.0 and devel-
oping more imaginative and creative solutions to food waste problems
[21].
• Waste management and treatment facilities on a small scale
Industry 4.0 is all about democratizing things; it is about taking some-
thing that is used to be done by massive government operations and
174 Waste Management 4.0: An Industry Automation Approach to the Future Waste

breaking it down so that even the average person can do it on a smaller

scale; the home may have a micro waste management facility. Perhaps
an intelligent robot will constantly be around to clear up the waste
and ensure that every garbage is collected and categorized and decides
whether the garbage should be recycled, reused, or used as a raw material
to manufacture another product.

7.7 Barriers to Implementing Industry 4.0 in the Waste

Management Industry
There are insurmountable impediments that have forced us to leave Industry
4.0, some of which are discussed below.
• Implementation costs are expensive
Industry 4.0 appears magnificent and offers many advantages, but it
costs a lot of money [22]. A few elements that influence the high
implementation costs are a robust internet connection required to build
Industry 4.0, a large internet connectivity capacity required, and a large
amount of data that will be transferred simultaneously [23]. Billions of
gadgets are connected to the Internet worldwide, and many household
items such as televisions, refrigerators, computers, and other electronic
equipment will be Wi-Fi enabled and capable of connecting online.
Also, every workflow must be recorded in real time and fed into an
analysis system. As a result, implementing only a small percentage of
IoT will not give the actual benefits that it promises.
• Security problems and concerns
Massive amounts of data are generated by Industry 4.0. As a result,
this data must be kept secure, recorded, processed, analyzed, and
transformed. Consumers would be highly concerned about sharing
information with underhanded businesses [24]. To tackle these issues,
organizations must establish a strong legal framework for data pri-
vacy that ensures consumer data security while holding corporations
accountable for the information they collect and share with third parties.
• A shortage of qualified personnel
Industry 4.0 is a relatively new concept; to cope with it, one must ensure
that employers are up-skilled to meet the technology’s problems. As a
result, putting Industry 4.0 solutions in place necessitates specialized
knowledge and abilities, and there is a demand for people who are highly
experienced and qualified in information technology, as well as people
 7.8 Case Study: Machine Learning for Waste Management 175

who have a thorough awareness of how each sector operates, particularly

the one to which they cater. As in Industry 4.0, humans and robots
work together; humans will focus on creative work while robots perform
repetitive tasks. Hence, there are a few roadblocks regarding a lack of
qualified personnel.
• Integration of technology is a problem
Innovative technologies such as cloud computing, which includes
Google Cloud, Amazon Web Services (AWS), etc., are deployed [25]. To
run an Industry 4.0-related solution investigation platform is required.
Migrating a large amount of data from a local system to the cloud is
time-consuming, complicated, and demands huge security protocols.

7.8 Case Study: Machine Learning for Waste Management

Machine learning models can be applied to the present waste management
statistical data in order to predict the waste that can be generated in the future;
with this knowledge, practical steps can be carried out to monitor, segregate,
and thereby significantly reduce the amount of waste.
For this study, the waste dataset was generated from one of Australia’s
most populated cities. The dataset contains four years of waste management
data; the above pie chart shows the visualization of the dataset, i.e., the col-
lection category and subcategory of waste [26]. Figure 7.5 show the types of
waste under several collection subcategories; the tons collected subcategory
has six types of waste that are split and shown as three categories in Figure 7.5
and three categories in Figure 7.5; Figure 7.5(d) shows the types of waste
under bins collected category.
Anaconda prompt software is used for applying machine learning algo-
rithms to the dataset; as a first step, the dataset is imported along with library
functions for analyzing and visualizing data [27]. The next step is cleaning
the data, removing redundancy, and training and testing the data with machine
learning algorithms. In order to determine which machine learning algorithm
provides the highest level of accuracy, various machine learning algorithms
such as logistic regression (LR), linear discriminant analysis (LDA), K-
neighbor classifier (KNN), decision tree classifier (DTC), Gaussian Naive
Bayes (NB), and support vector machine (SVM) [28] are applied to the
dataset. The algorithms are imported onto the platform using sklearn library
files [29]. Accuracy is measured by the ratio of correct predictions to the total
number of predictions [30]. Figure 7.5 shows the comparison for accuracy;
from the figure, it is clear that the logistic regression algorithm shows an
176 Waste Management 4.0: An Industry Automation Approach to the Future Waste

Figure 7.5(a). Collections under Figure 7.5(b). Collections under tons

number of collections subcategory. collected subcategory.

Figure 7.5(c). Collections under Figure 7.5(d). Collections under bins

tons collected subcategory. collected subcategory.
Figure 7.5 Visualization of the dataset.

accuracy of 80% for this dataset, and hence this algorithm can be used to
predict the waste generated in the next few years.
Figure 7.6 and Table 7.1 show the performance evaluation of the algo-
rithms based on accuracy; from the table and figure, it is clear that the logistic
regression algorithm works well with the dataset and shows 80% accuracy.
Hence, this algorithm can be used to predict future waste management
data.
 7.9 Conclusion 177

Figure 7.6 Accuracy evaluation of machine learning algorithms.

Table 7.1 Performance evaluation of machine learning algorithms.

Machine Learning Models

LR LDA KNN DTC NB SVM

Accuracy 80% 70% 60% 50% 60% 60%

7.9 Conclusion
To summarize, the core of Industry 4.0 is the CPS, which is composed of com-
puters connected over networks to monitor physical processes. This system
can exchange data automatically without much human intervention, leading
to the exponential increase of profit in engineering sectors, supply chain, and
product life cycle management. IoE plays a major role in Industry 4.0 and
will also be a key for future industrial revolutions; this is an amalgamation of
IoT, IIoT, IoS, IoD, and IoP. This chapter focuses on Waste Management 4.0;
as in many countries, door-to-door manual trash collection level is attained
only now. The challenges faced by the waste management industry are waste
collection, transporting waste safely, disposal of the collected waste, dealing
178 Waste Management 4.0: An Industry Automation Approach to the Future Waste

with the environmental impact, health impact, and, most importantly, segre-
gating waste for reuse and recycling; all these require technology in addition
to human resources, which leads to the necessity of Industry 4.0. Applying
CPS to identify and segregate reusable and unusable waste increases the
effectiveness of the waste management process in all its stages and paves the
way for sustainable and suitable ways for garbage decomposition. Industry
4.0 has positively impacted the waste processing and treatment system; it
has created an eco-friendly approach, thereby significantly reducing e-waste
and food waste and increasing in practicing waste-to-energy approach. It has
also set a basement for developing reliable small-scale waste management
infrastructure, successfully monitors the toxicity of the waste, and protects
against health hazards. However, some barriers have to be overcome, and
as the system uses sensors and computing resources to automate the waste
management system, implementation cost is high since physical devices are
connected over the Internet, and privacy issues may also arise.
Additionally, there is a shortage of skilled workers to coordinate the
process. In the future, with high-performance computer networks, smart
sensors, and professionals with intricate problem-solving and creative skills,
human−machine collaboration will pave the way for improving business
strategies, boosting the manufacturing domain. It will offer a better environ-
ment for work.

References
[1] Yin, Yong, Kathryn E. Stecke, and Dongni Li. “The evolution of pro-
duction systems from Industry 2.0 through Industry 4.0.” International
Journal of Production Research 56.1-2 (2018): 848-861.
[2] Xu, Li Da, Eric L. Xu, and Ling Li. “Industry 4.0: state of the art and
future trends.” International journal of production research 56.8 (2018):
2941-2962.
[3] Devi, A., et al. “IoT-Based Smart Pipeline Leakage Detecting System for
Petroleum Industries.” Industry 4.0 Interoperability, Analytics, Security,
and Case Studies. CRC Press, 2021. 149-168.
[4] Lee, Jay, Behrad Bagheri, and Hung-An Kao. “A cyber-physical systems
architecture for industry 4.0-based manufacturing systems.” Manufac-
turing Letters 3 (2015): 18-23.0
[5] Pivoto, Diego GS, et al. “Cyber-physical systems architectures for
industrial Internet of things applications in Industry 4.0: A literature
review.” Journal of manufacturing systems 58 (2021): 176-192.
 References 179

[6] Zhou, Keliang, Taigang Liu, and Ling Liang. “From cyber-physical
systems to Industry 4.0: make future manufacturing become possi-
ble.” International Journal of Manufacturing Research 11.2 (2016):
167-188.
[7] Usharani, S., et al. “Smart Energy Management Techniques in Indus-
tries 5.0.” Hybrid Intelligent Approaches for Smart Energy: Practical
Applications (2022): 225-252.
[8] Trstenjak, Maja, and Predrag Cosic. “Process planning in Industry 4.0
environment.” Procedia Manufacturing 11 (2017): 1744-1750.
[9] Devi, A., et al. “IoT Based Food Grain Wastage Monitoring and Control-
ling System for Warehouse.” 2021 International Conference on System,
Computation, Automation and Networking (ICSCAN). IEEE, 2021.
[10] Bagheri, Behrad, et al. “Cyber-physical systems architecture for self-
aware machines in industry 4.0 environment.” IFAC-PapersOnLine 48.3
(2015): 1622-1627.
[11] Hiriyannaiah, Srinidhi, et al. “A Multi-layered Framework for Internet
of Everything (IoE) via Wireless Communication and Distributed Com-
puting in Industry 4.0.” Recent Patents on Engineering 14.4 (2020):
521-529.
[12] Kumar, Sunil, et al. “Challenges and opportunities associated with waste
management in India.” Royal Society open science 4.3 (2017): 160764.
[13] Vaverková, Magdalena Daria, et al. “Municipal solid waste management
under COVID-19: challenges and recommendations.” Environmental
Geotechnics 8.3 (2020): 217-232.
[14] Joshi, Rajkumar, and Sirajuddin Ahmed. “Status and challenges of
municipal solid waste management in India: A review.” Cogent Envi-
ronmental Science 2.1 (2016): 1139434.
[15] Thada, Ayush, et al. “Custom blockchain-based cyber-physical system
for solid waste management.” Procedia computer science 165 (2019):
41-49.
[16] Oláh, Judit, et al. “Impact of Industry 4.0 on environmental sustainabil-
ity.” Sustainability 12.11 (2020): 4674.
[17] Rosa, Paolo, et al. “Assessing Relations between Circular Economy and
Industry 4.0: a systematic literature review.” International Journal of
Production Research 58.6 (2020): 1662-1687.
[18] de la Barrera, Belen, and Peter S. Hooda. “Greenhouse gas emissions
of waste management processes and options: A case study.” Waste
Management & Research 34.7 (2016): 658-665.
180 Waste Management 4.0: An Industry Automation Approach to the Future Waste

[19] Tabasová, Andrea, et al. “Waste-to-energy technologies: Impact on

environment.” Energy 44.1 (2012): 146-155.
[20] Sivadas, Sanitha K., et al. “Litter and plastic monitoring in the Indian
marine environment: A review of current research, policies, waste man-
agement, and a roadmap for multidisciplinary action.” Marine Pollution
Bulletin 176 (2022): 113424.
[21] Devi, A., et al. “IoT Based Food Grain Wastage Monitoring and Control-
ling System for Warehouse.” 2021 International Conference on System,
Computation, Automation and Networking (ICSCAN). IEEE, 2021.
[22] Kumar, K. Suresh, et al. “Blockchain technology: an insight into archi-
tecture, use cases, and its application with industrial IoT and big data.”
Blockchain Technology. CRC Press, 2020. 23-42.
[23] Kamble, Sachin S., Angappa Gunasekaran, and Rohit Sharma. “Analy-
sis of the driving and dependence power of barriers to adopt industry 4.0
in Indian manufacturing industry.” Computers in Industry 101 (2018):
107-119.
[24] Dawson, Maurice. “Cyber security in industry 4.0: The pitfalls of
having hyperconnected systems.” Journal of Strategic Management
Studies 10.1 (2018): 19-28.
[25] Therese, M. Julie, P. Dharanyadevi, and K. Harshithaa. “Integrat-
ing IoT and Cloud Computing for Wireless Sensor Network Appli-
cations.” Cloud and IoT-Based Vehicular Ad Hoc Networks (2021):
125-143.
[26] Dataset:[Link]
Waste_Management.csv
[27] Dalzochio, Jovani, et al. “Machine learning and reasoning for predictive
maintenance in Industry 4.0: Current status and challenges.” Computers
in Industry 123 (2020): 103298.
[28] Candanedo, Inés Sittón, et al. “Machine learning predictive model for
industry 4.0.” International Conference on Knowledge Management in
Organizations. Springer, Cham, 2018.
[29] Brik, Bouziane, et al. “Towards predicting system disruption in industry
4.0: machine learning-based approach.” Procedia computer science 151
(2019): 667-674.
[30] Rai, Rahul, et al. “Machine learning in manufacturing and industry
4.0 applications.” International Journal of Production Research 59.16
(2021): 4773-4778.
 8
Industrial Internet of Things (IIoT) for
E-waste Recycling System

A. P. Jyothi and P. Padma Priya Dharishini

Department of CSE, Faculty of Engineering and Technology, Ramaiah

University of Applied Sciences, India
E-mail: jyothiarcotprashant@[Link]; [Link]@[Link]

Abstract
Industrial Internet of Things (IIoT) is a word used to portray the framework’s
organization of genuine things and contraptions with embedded sensors, pro-
gramming, and correspondence limits. These IoT contraptions can assemble,
exchange, and cycle data with each other and with outside structures to addi-
tionally foster capability, proficiency, and prosperity, customer relationship
management (CRM), or enterprise resource planning (ERP). IIoT is driving
one more surge of mechanized change, changing endeavors like collecting,
transportation, and energy. By partner contraptions and structures to the web,
IIoT engages associations to assemble and analyze data logically, making it
possible to smooth out undertakings and further foster execution. A logical
examination of circulation place organization using the industrial Internet
of Things (IIoT) is inspected comprehensively. As a case study, an E-waste
recycling system using IoT is discussed to show the negative impact of e-
waste on the environment due to harmful chemicals present in e-waste.

Keywords: IIoT, IoT, CRM, ERP, real-time

8.1 Introduction
The industrial Internet of Things (IIoT) uses technologically advanced sen-
sors and actuators to improve manufacturing and other modern processes.

181
182 Industrial Internet of Things (IIoT) for E-waste Recycling System

The industrial Internet of Things (IIoT) is the modern web or Indus-

try 4.0. It uses the power of brilliant machines and regular examination
to capitalize on the data that “idiotic machines” have been producing in
contemporary environments for quite some time. The essential reasoning
behind the industrial Internet of Things (IIoT) is that clever machines
are not only superior to people in terms of gathering and analyzing data
in a more timely manner, but they are also superior in terms of con-
veying significant data that can be utilized to drive business decisions
more expediently precisely [2]. This is the fundamental reasoning behind
the IIoT.
Connected sensors and actuators allow businesses to address problems
and failures more quickly, thereby saving both time and money while sup-
porting efforts to improve business knowledge. Regarding assembly, more
specifically, the IIoT holds enormous potential for quality control, manage-
able and environmentally friendly practices, production network detectability,
and, in general, inventory network proficiency.

8.2 Background Study

The utilization of IoT innovations has expanded quickly inside numerous
businesses. Utilizing IoT innovations is likewise expected to work on the data
for sharing inside and across different ventures [4]. View of IoT innovations
has been practical toward different applications going from home computer-
ization to modern IoT and having the option to impart essential things from
any spot through an organization [18].
It tends to be noticed that the number of undertakings that connect
with the utilization of the IoT in modern regions has expanded, for exam-
ple, farming, security reconnaissance, natural observing, food handling, and
numerous others [8]. Besides, distributed papers in the IoT regions have
extended significantly, crossing various utilizations. Furthermore, numerous
different advancements and techniques have been utilized to help the IoT, like
Wi-Fi [7].

8.3 IIoT Working

IIoT is an organization of shrewd gadgets associated with structure frame-
works that screen, gather, trade, and dissect information. Each modern IoT
environment comprises:
 8.4 IIoT Security 183

• Associated gadgets that can detect, impart, and store data about them-
selves.
• Public and additionally confidential information correspondences foun-
dation.
• Investigation and applications that create business data from crude
information.
• Capacity for the information produced by the IoT gadgets, and indi-
viduals. These edge gadgets and canny resources send data straight-
forwardly to the information correspondence foundation, changing it
into important data on how a specific piece of hardware is working.
This data can be utilized for prescient support and to enhance business
processes.

8.4 IIoT Security

Manufacturers built IoT devices with little consideration for security, which
led to the assumption that IoT devices are inherently unsafe. Given the
similarities between IoT and IIoT devices, it is important to consider whether
using IoT devices is safe.
The device-by-device analysis is required for IIoT devices, just like for
any other linked device. It is completely feasible that a device from one
manufacturer is secure but not from another. However, device manufacturers
now place a higher priority than ever before on security [10].
The Industrial Internet Consortium (IIC) was founded in 2014 by various
technological corporations, including AT&T, Cisco, General Electric, IBM,
and Intel. Even though this group’s primary goal is to hasten the adoption of
IIoT and associated technologies, it has prioritized security and even gone so
far as to do so.

8.4.1 Risks and challenges of IIoT

Security-related hazards are the ones that concern IIoT use the most [5]. Even
after they have been put into production, IIoT devices frequently retain their
default passwords. Similarly to this, a lot of IoT devices send data in clear
text. Due to these circumstances, it would be pretty simple for a hacker to
intercept data flowing from an IoT device. Similar to this, an attacker might
seize control of an unsecured IIoT device and use it as a base to attack other
network resources [15].
184 Industrial Internet of Things (IIoT) for E-waste Recycling System

Security is a major concern for people in charge of an organization’s

IoT devices, but so is device management. Adopting an efficient device
management strategy will be crucial as a business deploys more and more
IoT devices [11].

8.4.2 Difference between IoT and IIoT

Cloud stages, sensors, networks, machine-to-machine communications, and
information examination are only a few of the advancements IoT and IIoT
share for all intents and purposes, even though they are used for vari-
ous things. IoT applications interface gadgets from a few industry areas,
for example, rural, medical care, venture, buyer, utilities, government, and
metropolitan regions. Brilliant apparatuses, wellness trackers, and other
IoT applications commonly do not bring about crisis conditions, assuming
something turns out badly.
Then again, IIoT applications connect gear and contraptions in the
assembling, oil, gas, and utility areas. In IIoT establishments, framework dis-
appointments and margin time can prompt high-endanger or even hazardous
circumstances. Also, contrasted with IoT applications, IoT applications are
more centered on upgrading productivity and well-being or security.

8.4.3 IIoT applications and examples

ABB, a power and robotics company, uses linked sensors to monitor its
robots’ maintenance requirements to suggest repairs before parts break in a
real-world IIoT deployment of intelligent robotics [14]. Similarly, Airbus, a
manufacturer of commercial jetliners, has started what it calls the factory of
the future, a digital manufacturing drive to improve operations and increase
output. Airbus has incorporated sensors into equipment and machines on the
shop floor to reduce errors and improve worker safety and provided personnel
with wearable technology, such as industrial intelligent glasses.
Fanuc, a different robotics manufacturer, uses sensors and cloud-based
data analytics to foresee when parts in its robots will break. This allows the
plant manager to plan maintenance at reasonable periods, cutting expenses
and avoiding unnecessary downtime.
Magna Steyr, an Austrian automaker, uses the IIoT to track its assets,
including tools and car parts [6], and purchases more stock automatically
when it runs low. In order to track components in its warehouses, the
corporation is also experimenting with “smart packaging” supplemented with
Bluetooth.
 8.5 Industries using IIoT 185

8.5 Industries using IIoT

Processes of industrial automation have many applications. Some of the most
important are manufacturing, oil and gas extraction, paper mills, and steel
mills. Below are just a few instances where these companies have used this
technology.
a. Automated production in the manufacturing industry:
The manufacturing industry extensively uses industrial automation in
many different contexts and purposes. Technology may be used in
the manufacturing process, from the design phase through the final
inspection, as well as in monitoring upkeep duties and stock levels [17].
b. Oil and gas exploration:
Since oil and gas drilling often occurs at offshore stations and other
distant locations, industrial automation greatly benefits the oil and
gas drilling industry. Sensors and other monitoring equipment will
reduce the number of complex and potentially dangerous site visits that
maintenance workers must make.
c. The paper industry:
It can be used by paper mills to control instrumentation, plant devices,
and equipment and to manage batches of production. Use in this way is
also possible. This allows for greater overall system transparency for the
production system’s operators.
d. Automation in steel mills:
Hierarchical control is used in other types of industrial automation,
such as in steel mills. Because of this technological development, a
unified system is now at hand for managing and controlling the entire
steel mill [16]. Autopilot controls on commercial jets are one example
of the long-standing use of industrial automation in the transportation
sector. There is a shift toward autonomous vehicles that can be used for
commercial and private purposes, incorporating advances in industrial
automation [9].
e. Distribution:
Once a product has been manufactured and is prepared for shipment,
distribution takes over. Worldwide, deadlines are getting shorter and
shorter. Because of this, it intends to continue introducing solutions in
the distribution field to regulate the shipping and delivery of products at
an even quicker pace.
f. Industrial controls and automation for manufacturing process:
It will be impossible to successfully implement industrial automation
controls on your production line without a computerized maintenance
186 Industrial Internet of Things (IIoT) for E-waste Recycling System

management system (CMMS). This centralized system can gather much

data, including stock levels, work order schedules, maintenance records,
and data from remote sensors. By combining vital resources, high-
quality data, historical data, and real-time alerts or sensor readings,
you will be in a solid position to select, refine, and use your industrial
automation tools to boost efficiency within your organization. The rea-
son being you will have everything you need in one convenient location.

8.6 Advantages and Disadvantages of IIoT

There are Advantages and disadvantages of IIoT as listed in Table 8.1.
Benefits of IIoT
The first way to see the advantage of robotization is that it removes the
requirement for any human intercession. For example, suppose you will
robotize your deals pipe. Instead of checking telephone discussions and
updating your site physically, you can now have a chatbot accomplish the
work for you. This will save time and cash over the long haul − something
priceless to many organizations [19]. Different benefits of computerization
are as per the following:
i. Greater rates of production: Robotization is a significant part of the
manufacturing industry, making it possible for more excellent rates of
production to be achieved while incurring fewer costs and utilizing fewer
resources. Because of these benefits, mechanization has become a more
common practice in various settings.
ii. Increased productivity: Mechanization has played a significant role in
the labor force and will continue to do so as the industry develops.

Table 8.1 Advantages and disadvantages of IIoT.

Advantages Disadvantages
Higher production rates Worker displacement
Expanded productivity It needs enormous capital consumption
More productive utilization of materials Can become redundant
Better item quality Could present new security dangers
Improved safety It still requires human mediation
Shorter workweeks for labor
Decreased manufacturing plant lead times
Consistency
Saves time
No labor issues
 8.6 Advantages and Disadvantages of IIoT 187

Because of this increase in productivity, businesses can now concentrate

their efforts on other important tasks, such as product development,
while still completing their projects on schedule. These endeavors
provide benefits to the organization and pave the way for further
accomplishment on their part.
iii. A more effective utilization of the materials: New technology affects
how products are manufactured. Mechanization leads to better products
because it makes more efficient use of available resources and produces
more of them. Because of these benefits, manufacturers can produce bet-
ter products, resulting in decreased production costs, increased customer
satisfaction, and a greater return on investment.
iv. Better item quality: Mechanization is an expense-saving system that
organizations have been involved in for a long while. Lately, com-
puterization has been embraced all the more frequently in promoting
business since it empowers organizations to deliver their items quicker,
less expensive, and with a lot greater. The utilization of mechanization
will likely continue developing as advertisers hope to profit from the
benefits it offers to increment efficiency.
v. Further developed security: Robotization enjoys many benefits. It can
assist with further developing security by restricting human mistakes,
lessening expenses, and making the creation cycle more proficient. This
is particularly significant in enterprises with a high gamble of injury, like
horticulture and mining.
vi. More limited work-filled weeks for work: Mechanization can empower
more limited work-filled weeks for work. By involving computerization
instead of people, organizations can decrease the number of workers
expected to work at a most extreme limit. This will give additional
opportunities to workers to zero-in on different undertakings, for exam-
ple, the executives, preparation, or advancement that require a human
connection.
vii. Decreased production line lead times: The advantages of computeriza-
tion should be visible for the most part in processing plants. Mechaniza-
tion decreases beginning-to-end manufacturing plant lead times by 60%
or more. This not just advantages the assembling result of a plant, yet
additionally the quality and proficiency of an item.
viii. Consistency: Mechanization will carry consistency and solidness to the
business. It will likewise allow the business an opportunity to adjust to
changes in what their clients need. Organizations will believe a more
188 Industrial Internet of Things (IIoT) for E-waste Recycling System

significant amount of this proficiency as they proceed should confront

the present changing economic situations and buyers’ requests.
xi. Saves time: Making work more accessible by automating it is a huge
step forward. The time spent on tasks that a computer program or
machine should be able to handle is saved. Tasks like these include:
Mechanization has come a long way, and with each passing day comes
more discoveries and developments that make the widespread use of
computers a reality.
x. No work issues: Mechanization is the way into a work emergency-free
future. It will not just cost as much as individuals, yet it will offer quality
types of assistance that individuals will pay for.

8.6.1 Hindrances of IIoT

Computerization is a piece of society today. Computerization can be incred-
ible for people and associations when appropriately utilized and executed. It
can save time and cash and assist you with finishing work all the more pre-
cisely [20]. The cons of mechanization are the expense of supplanting human
work with robotized frameworks. Different weaknesses of mechanization are
as per the following:
i. Specialist dislodging: The main inconvenience of computerization is
the removal of human work. This is because an electronic errand can
be executed quicker and with more prominent exactness than can be
accomplished by a human. For instance, Disney World has involved
self-driving vehicles for quite a long time to ship visitors around the
recreation area. Many individuals are anxious about the possibility that
this will bring about less employment for people.
ii. Requires significant investment of time and money: Computers have
been an integral business component for quite some time. Before switch-
ing to computerization, manufacturers must consider several unantic-
ipated consequences. One of these effects is the requirement for a
significant amount of financial investment to keep up with and manage
computerized systems. These systems are also more vulnerable to digital
attacks than manual systems, leaving businesses defenseless if their
foundation is not adequately protected. Manual systems would be more
effective at fending off physical attacks.
iii. Has the potential to become monotonous: Using computers is a valuable
solution to some problems. Nevertheless, this convenience may become
inconvenient in certain circumstances, such as when a familiar change
 8.7 Case Study - IoT for E-waste Recycling System 189

necessitates the adjustment of the mechanization. These adjustments

will only add to the organization’s responsibility, which may cause them
to lose valuable time and resources.
iv. Could present new security risks: Mechanization could present new
well-being dangers while working circumstances change suddenly. For
instance, it is conceivable that a driverless vehicle could be customized
to drive independently; however, it can, in any case, cause a mishap
when a passerby ventures out into the road under conditions that are
not so great (night/low permeability and so on).
v. Still requires human intercession: While the general advantages of
mechanization have been demonstrated, there are as yet specific under-
takings that require human mediation. We will think about the above
illustration of self-driving vehicles − these machines can distinguish
most obstructions out and about and can be customized to stop. In any
case, a few explicit circumstances can make these machines confound
data and produce undesired results − for example, passing through a
hindrance that is not obviously noticeable to the vehicle’s sensor.

8.7 Case Study - IoT for E-waste Recycling System

The waste generated from electrical or electronic devices, i.e., electronic
waste or e-waste, negatively impacts the environment due to the harmful
chemicals present in it. Figure 8.1 shows the categorization of e-waste based
on its components and composition. The circuit boards and batteries in e-
waste can cause damage to ecosystems. New natural resources must be
mined to manufacture electronic components that make a massive ecological
impact. In order to overcome this, the recovery of materials like copper, gold,
aluminum, plastic, and glass from e-waste is needed. To achieve this, e-waste
should be recycled.
Figure 8.1 shows the processes in the e-waste recycling system. First,
e-waste is collected through recycling bins, collection locations, or through
collection services. This mixed e-waste is transported to the e-waste catego-
rization center. In that center, e-waste is categorized into any of the e-waste
categories mentioned in Figure 8.2 based on its composition and components.
This step is critical because e-waste containing batteries mixed with other
waste can be very damaging. Next, the categorized e-waste is loaded into
appropriate pallets. The RFID tags are printed on each pallet with information
like the e-waste category and its storage location. The dock door of the
storage center is mounted with an RFID reader and an antenna. The pallets
190 Industrial Internet of Things (IIoT) for E-waste Recycling System

Figure 8.1 Categorization of e-waste.

are loaded into the forklift, and while the forklift passes through the entry
dock door, the RFID reader reads the RFID tag pasted on all the pallets.
The acquired data is passed to the cloud immediately and stored there for
further analysis. The pallets are stored in appropriate locations based on e-
waste categorization. When the pallets are dispatched for e-waste recycling,
the reader in the exit door dock reads the tag in the pallets when the forklift
passes through the exit door and sends the data to the cloud. In the cloud, the
received data is compared with the stored data at the time of storing pallets
in appropriate locations, and if the match is found, e-waste is dispatched for
recycling.
IoT-based e-waste recycling system is developed using NodeMCU and
Adafruit IO platform. All e-waste is loaded into proper pallets after e-waste
categorization. Based on the category of e-waste, appropriate RFID tags are
printed on each pallet. The RFID scanner in the dock doors of the storage
 8.7 Case Study - IoT for E-waste Recycling System 191

Figure 8.2 Process in e-waste recycling.

center scans the RFID tags in the pallets and stores the information in the
Adafruit IO cloud with the help of the ESP8266 Wi-Fi module. This informa-
tion is displayed in the Adafruit IO dashboard that can be used for analysis.
Figure 8.3 shows the hardware setup of the dock door in the storage center.
While forklift pallets pass through the entry dock door of the storage
center, the tags in the pallets are getting scanned, and information about e-
waste (like storage location and categorization of e-waste) is uploaded to

Figure 8.3 Hardware setup in dock door of the storage center.

192 Industrial Internet of Things (IIoT) for E-waste Recycling System

Figure 8.4 RFID tag information.

Figure 8.5 E-waste dispatched for recycling.

feeds in the Adafruit cloud as well as printed in serial monitor as shown

in Figure 8.4. Based on this information, e-waste is stored in appropriate
locations in the storage center.
Similarly, when the pallets are getting dispatched for e-waste recycling,
the reader in the exit door dock reads the tag in the pallets when the forklift
passes through the exit door and sends the data to the cloud, and e-waste is
dispatched, as shown in Figure 8.5.

8.8 Future Trends of IIoT

The eventual fate of IIoT is firmly combined with a pattern known as
Industry 4.0 [1]. Industry 4.0 is, basically, the fourth Modern Transforma-
tion [3]. Industry 4.0 is where we are today. Industry 4.0 depends on the
utilization of associated electronic gadgets − especially IoT gadgets. IoT
gadgets will assume a significant part in computerized changes, particularly
as associations endeavor to digitize their creation lines and supply chains.
Moreover, enormous information investigation will develop to consolidate
IIoT information. This will make it workable for associations to recog-
nize changing circumstances progressively and answer in like manner. Even
 8.8 Future Trends of IIoT 193

though IIoT gadgets have been around for quite a long time, actual reception
is still in its earliest stages. This makes certain to change as 5G turns out
to be progressively standard, and the sky is the limit from there, and more
associations start to acknowledge how IIoT can help them. Various assets are
accessible online for associations that need to find a workable pace on IoT and
IIoT. IoT innovations and the Modern Web of Things (WoT) are changing
assembling, transportation, and energy. The network permits organizations
to create new plans of action and further develop execution through better
information-driven bits of knowledge. Get a brief look at 10 patterns you can
expect for the eventual fate of IIoT.
More connected devices
The quantity of IIoT-associated gadgets is supposed to be twofold by 2022.
The following pages of IoT gadgets are intended to help current arising
advances, for example, computerized reasoning, the cloud, AI, portabil-
ity, augmented reality, and increased reality. These “more astute” gadgets
are additionally zeroing in on expanded interoperability to help telemetry
information trade among the Web of Things (WoT).
IIoT manufacturing as a service
Modern IoT assembling as a help is another way to deal with assembling in
light of a pay-more-only-as-costs-arise model, in which all parts are conveyed
or obtained from outsiders. In this course of action, makers consent to keep
up with specific norms and particulars yet have the opportunity to pick the
particular advancements utilized for execution. Many accept that this pattern
will, to some degree, shape the eventual fate of IIoT by 2022.
Cloud and edge computing
The utilization of cloud and edge registering will develop rapidly, with
the capacity to gather information from far-off gadgets on the edge of an
organization. The most recent age of IIoT gadgets can impart and convey
important data and guidelines through neighborhood passages that interface
them straightforwardly with the cloud for close quick access by an application
or administration. With this sort of execution, inertness is diminished, and
reaction time moves along.
Predictive maintenance
One more Modern Web of Things pattern expected in 2022 is proactive
upkeep. As IIoT advances keep on propelling, makers can foresee possi-
ble disappointments in the functional cycle while a gadget or framework
194 Industrial Internet of Things (IIoT) for E-waste Recycling System

is as of yet on the web and inside its guarantee period. This sort of
investigation permits organizations to limit unscheduled free time while
further developing general gear execution, lessening costs, and expanding
uptime.
Process data management
To remove meaningful experiences from IIoT information, producers need
to convey progressed examinations that can rapidly deal with enormous
measures of data. This will permit them to gather and dissect immense infor-
mation from their tasks and individual cycles while safeguarding it against
digital assaults. Makers will likewise require many associated gadgets to help
the shrewd association of data, examination, and interpretation of information
across frameworks.
Digital twins
Computerized twins are a fresher idea in the Modern Web of Things, where
they give a virtual portrayal of actual items and empower these items to con-
nect with outside clients. With this innovation, IIoT gadgets can be observed
on the web, with the ultimate objective of improving execution, anticipating
shortcomings, and diminishing free time. By 2022, advanced twins (virtual
imitations) will be utilized as a significant IIoT pattern for checking and
overseeing far-off hardware and resources. Producers can likewise utilize
computerized twins to unite information from different gadgets and sen-
sors to make a composite perspective on the framework for compelling
navigation.
Location tracking
The area following is an IIoT application that empowers organizations to
screen the area and state of resources, hardware, and workforce continuously.
The area following devices can further develop productivity in assembling
tasks by guaranteeing that materials are accessible when required. It likewise
empowers makers to comprehend where their items are in the store network,
making distinguishing possible issues and deferrals simpler.
Since this innovation utilizes continuous information streams, it will
be a significant concentration for IIoT execution in 2022. By gathering
data about area and development, this innovation offers knowledge into
the whereabouts of resources and merchandise. Coming soon for IIoT,
the area following will significantly affect ventures like transportation and
strategies.
 8.9 Conclusion 195

Blockchain IIoT for supply chain

Makers are expected to satisfy client needs for straightforwardness in their
stockpile chains, which can incorporate affirmations and reviews. Blockchain
innovation is a successful way that makers can arrive at this objective of
straightforwardness by 2022. Blockchain-controlled store network arrange-
ments will likewise assist with keeping up with item trustworthiness through-
out the production network process, even after it leaves the industrial facility
floor. Blockchain IIoT arrangements will be embraced by producers as a
method for giving more prominent perceivability into their stock chains,
making it simpler to fulfill client needs for straightforwardness and item
quality.
Smart manufacturing
In 2022, shrewd assembling will be a significant IIoT pattern that gives a
solitary perspective on all processing plant exercises and assists organizations
with working on functional effectiveness. Shrewd assembling can assist mak-
ers with distinguishing peculiarities in their work processes to forestall likely
issues before they become an issue. With this innovation, makers approach
continuous data about the situation with machines and creation processes.
This remembers information for engines, siphons, and other gear on the
production line floor.
Machine learning
Producers should arrive at savvy conclusions about their items and how they
deal with their resources to stay serious. AI can help producers by recogniz-
ing designs in cycles and work processes that can prompt more prominent
proficiency. Likewise, AI can foresee issues inside an interaction before they
end up decreasing free time and work on the plant’s general well-being.
In the future, IIoT reception increments to associate more gadgets
together, and AI will assume an undeniably significant part in assembling
to give ongoing examination. It can likewise assist makers with gaining from
their information to find significant experiences that change business results.

8.9 Conclusion
Modern mechanization is a developing pattern that indicates that things are
not pulling back. Proceeded with robotization, innovation will drive associa-
tions to a consistently more prominent degree of effectiveness and execution.
To remain cutthroat in your specific market, you must be one of the leaders.
196 Industrial Internet of Things (IIoT) for E-waste Recycling System

Organizations should embrace robotization through brilliant gadgets across

all endpoints to prevail from here on out. As well as clearing the way for
the up-and-coming age of modern computerization, a developing interest in
complex mechanical technology and robotization stages is expanding plant
proficiency and efficiency. Perpetually, a huge volume of information exists,
and the IoT gives the course to smoothing out it in a solid yet noticeable
manner. Perceiving new roads and forming innovation to integrate into your
organization’s tasks straightaway can go quite far toward getting the upper
hand.

References
[1] Tan, L.; Wang, N. Future Internet: The Internet of Things. In Proceed-
ings of the 3rd International Conference on Advanced Computer Theory
and Engineering, ICACTE 2010, Chengdu, China, 20–22 August 2010;
pp. V5-376–V5-380.
[2] Ashton, K. Internet of Things. Page 1—RFID Journal. Available online:
[Link] (accessed on 2 March
2020).
[3] Okano, M. T. IOT and Industry 4.0: The Industrial new revolution. In
Proceedings of the International Conference on Management and Infor-
mation Systems, ICMIS 2017, Bangkok, Thailand, 25–26 September
2017. Available online:[Link]
1057_IOT_and_Industry_40_The_Industrial_New_Revolution(access
edon17September2021).
[4] Rahim, M. A.; Rahman, M. A.; Rahman, M. M.; Asyhari, A. T.;
Bhuiyan, M. Z.; Ramasamy, D. Evolution of IOT-enabled connectivity
and applications in Automotive Industry: A Review. Veh. Commun.
2021, 27, 100285. [CrossRef]
[5] Rajmohan, R., Kumar, T. A., Pavithra, M., & Sandhya, S. G. (2020).
Blockchain: Next-generation technology for industry 4.0. In Blockchain
Technology (pp. 177-198). CRC Press. [CrossRef]
[6] Silva, C.; Silva, F.; Sarubbi, J.; Oliveira, T.; Meira, W.; Nogueira, J.
Designing mobile content delivery networks for the internet of vehicles.
Veh. Commun. 2017, 8, 45–55. [CrossRef]
[7] Lei, T.; Wang, S.; Li, J.; Yang, F. A cooperative route choice
approach via virtual vehicle in iov. Veh. Commun. 2017, 9, 281–282.
[CrossRef]
 References 197

[8] Bajaj, R.; Rao, M.; Agrawal, H. Internet of things (IoT) in the smart
automotive sector: A review. IOSR J. Comput. Eng. 2018, 9, 36–44.
[9] Fantian, Z.; Chunxiao, L.; Anran, Z.; Xuelong, H. Review of the key
technologies and applications in internet of vehicle. In Proceedings of
the 2017 13th IEEE International Conference on Electronic Measure-
ment & Instruments (ICEMI), Yangzhou, China, 20–22 October 2017;
pp. 228–232.
[10] Adil, M.; Khan, M. K. Emerging IOT applications in Sustainable Smart
Cities for COVID-19: Network security and data preservation chal-
lenges with future directions. Sustain. Cities Soc. 2021, 75, 103311.
[CrossRef] [PubMed]
[11] Mbunge, E. Integrating emerging technologies into COVID-19 contact
tracing: Opportunities, challenges and Pitfalls. Diabetes Metab. Syndr.
Clin. Res. Rev. 2020, 14, 1631–1636. [CrossRef] [PubMed]
[12] Otoom, M.; Otoum, N.; Alzubaidi, M. A.; Etoom, Y.; Banihani, R.
An IOT-based framework for early identification and monitoring of
COVID-19 cases. Biomed. Signal Process. Control 2020, 62, 102149.
[CrossRef] [PubMed]
[13] Vedaei, S. S.; Fotovvat, A.; Mohebbian, M. R.; Rahman, G. M.;
Wahid, K. A.; Babyn, P.; Marateb, H. R.; Mansourian, M.; Sami,
R. COVID-safe: An IOT-based system for Automated Health Moni-
toring and surveillance in post-pandemic life. IEEE Access 2020, 8,
188538–188551. [CrossRef]
[14] Gillis, S. A. What Is Internet of Things (IoT)? Available online:
[Link]
Things-IoT (accessed on 28 December 2021)
[15] Sundmaeker, H.; Guillemin, P.; Friess, P.; Woelfflé, S. Vision and
Challenges for Realising the Internet of Things; European Commis-
sion Information Society and Media: Brussels, Belgium, 2010. Avail-
able online: [Link] Clus-
terbook_March_2010.pdf (accessed on 7 November 2021).
[16] Perera, C.; Liu, C. H.; Jayawardena, S.; Chen, A. M. A Survey on
Internet of Things from Industrial Market Perspective. IEEE Access
2014, 2, 1660–1679. [CrossRef]
[17] Lampropoulos, G.; Siakas, K. V.; Anastasiadis, T. Internet of Things
(IoT) in Industry: Contemporary Application Domains, Innovative
Technologies and Intelligent Manufacturing. Int. J. Adv. Sci. Res. Eng.
2018, 4, 109–118. [CrossRef]
198 Industrial Internet of Things (IIoT) for E-waste Recycling System

[18] Atzori, L.; Iera, A.; Morabito, G. The Internet of Things: A survey.
Comput. Netw. 2010, 54, 2787–2805. [CrossRef]
[19] ITU Internet Reports 2005: The Internet of Things. Available online:
[Link] (accessed on
10 March 2021).
[20] Weber, R. H. Internet of Things–New security and privacy challenges.
Comput. Law Secur. Rev. 2020, 26, 23–30. [CrossRef]
 9
A Multi-hazard Industry Assessment
System Based on Unmanned Aerial
Vehicles (UAVs) for Bridges Crossing
Seasonal Rivers

Allan J. Wilson1 , A. Pon Bharathi1 , M. Leeban Moses2 ,

D. Vedha Vinodha3 , K. Kalaiselvi4 , and Kannan Pauliah Nadar5
1 Department of ECE, Amrita College of Engineering and Technology, India
2 Department of ECE, Bannari Amman Institute of Technology, India
3 Department of ECE, JCT College of Engineering and Technology, India
4 Department of ECE, Hindusthan College of Engineering and Technology,

India
5 Department of Electrical and Computer Engineering,

Institute of Technology, Jigjiga University, Ethiopia

Abstract
Digital Elevation Models automated multi-hazard performance assessment
system based on unmanned aerial vehicles (UAVs) was developed to meet
the demand for quick performance evaluation and performance prediction
of river crossing RC bridges. In the first stage of the process, the newly
developed system obtained the measurements from the UAVs to generate
three-dimensional DEMs of the river bed. These DEMs were used to analyze
the river bed. The hydraulic model was validated using the Q50 flood event
as the basis for the validation. The HEC-RAS software, in conjunction with
the flood simulation, provided an estimate of Q5 for the scour depths that a
probable flood would cause. After that, the scoured piles were added to the
bridge’s three-dimensional finite element model (FEM), which was generated
automatically using the written code. We were able to determine the flood
loads that would be placed on the virtual bridge with the assistance of the

199
200 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

HEC-RAS flood inundation map and the estimated water depths that would be
present close to the bridge piers. To conduct the seismic evaluation, nonlinear
time history analyses (THAs) were utilized. In addition, multiple-scaled
earthquake acceleration records were discovered to act in both principal axes
of the bridge simultaneously, which aligns with the surrounding area’s seis-
micity. The Boaçay-II Bridge in Antalya, Turkey, was the case to examine.
According to the findings of the analyses, the lateral displacements and the
pile internal forces increased as the scour depth increased. Still, the pier
column internal forces remained relatively constant throughout the process.
Consequently, the seismic displacement and load demands were tracked.
They are shifted from the pier columns to the piles in response to the growing
scour. As a direct consequence of this, the research report bridge was put to
use to verify that the proposed system is applicable.

Keywords: hazard, flood, piles, hydrodynamic, explore, earthquake

9.1 Introduction
Floods are the world’s deadliest natural calamity, claiming more lives and
causing more casualties than any other natural disaster. Every year, around
1700 people are killed by floods in India, and 39 lakh people are impacted
[1]. As a result of climate change and global warming, the frequency of
major floods has risen. Recurrent floods have resulted in many deaths and
a significant economic loss for the country. Floods are projected to have cost
the economy close to 1.5% of GDP [2]. Floods in rural areas are more severe
since most people live near rivers. In rural locations, disaster management has
been complicated by a lack of infrastructure and a remote location. Because
of its ability to collect data more precisely with fewer human interactions,
WSNs have gained a lot of interest as a disaster management tool. Recovery
efforts will be better coordinated, and data from the field will be collected
via the WSN. Unlike their wired counterparts, there are no downtimes for
wireless networks in the event of a natural catastrophe.
Additionally, they may be easily scaled and customized to meet spe-
cific needs. Using unmanned aerial vehicle (UAVs) to increase surveillance
has been the subject of previous studies on WSNs in disaster management
[3, 4]. Understanding the fundamental issues WSNs face is necessary before
realizing their full potential for disaster management.
 9.1 Introduction 201

Figure 9.1 Structure of warning system.

UAV-WSN systems for monitoring, exploring, and monitoring large areas

of interest are gaining favor. There will be a mix of data coming from the
ground, the air, and the Internet with this type of system. To improve the
speed of emergency response by making it easier to get to, moving about,
and reacting faster [2]. These collaborative methods have been praised as a
consequence.
Sensor networks may be utilized in indoor and outdoor areas and in
various disciplines to gather data [7]. There have been considerable tech-
nological advances due to the advancement of sensor networks and the
incorporation of data processing for onboard information. Many publications
in the last few years have examined the processing and data transmission
capacity, focusing on accessibility and production costs, dependability and
scalability, networking protocols, and specific physical limits like autonomy,
communication range, and energy consumption.
Environment monitoring, pre-disaster forecast and prevention, early
warning system (EWS), emergency preparedness, post-disaster response
(such as search and rescue), and structural health monitoring (SHM) have
all been successful uses of WSNs when dealing with disasters. We consider
SHM an example of disaster monitoring in this work because of its relevance
202 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

in preventing life loss due to natural catastrophes such as landslides, earth-

quakes, and tornadoes. Several approaches and techniques in conjunction
with the WSN have been presented to meet the demands of catastrophe moni-
toring, including network organizations and architectures, routing algorithms,
and sensor node structure. An energy-saving medium access control (ES-
MAC) protocol is used in the WSN power consumption process. There was a
downside to these methods: the amount of time it takes for many packets
to join the network increased. All network protocol stack levels must be
addressed to overcome these challenges. The optimum clustering technique
and routing protocol are usually formulated as optimization concerns. Recov-
ering sensor network management and attaining optimal QoS routing with
the least amount of energy consumption and the best path selection for data
transmission across the WSN may be achieved with bio-inspired intelligence
approaches.
Wireless sensor networks (WSNs) have found energy-efficient routing
algorithms to conserve resources while also increasing the network’s lifespan.
Protocols for detecting paths for transferring observed events help extend
the network’s lifespan despite sensor node limitations and the complex
environments in which they must operate. Health, medical, and agricultural
applications began to take hold of WSN as intelligent sensors spread across
a wide range of industries and sectors. Achieving energy efficiency and QoS
requirements for multiple application domains is a challenging task; however,
given the wide variety of traffic flows, changeable network settings, and the
ability to regulate network resources of the SNs QoS routing solutions at the
energy awareness network layer have attracted a lot of attention since they
are well-established. There has been a lot of focus on WSN energy usage due
to wireless communication.
This research is aimed to create an automated multi-hazard assessment
technique for river-crossing bridges using UAVs. The suggested UAV-based
approach may be easily and quickly deployed in areas where several bridges
span the same river, with modeling, assessment, and analysis taking up a lim-
ited time. Under both seismic excitations and flood loading, the present multi-
hazard performance under existing scour circumstances and the anticipated
performance under expected scour depths were calculated. The analyses were
carried out by the sub-objectives of: (a) examining the effect of scour on the
bridge’s seismic behavior, (b) using nonlinear explore time history analyses
(ETHAs) to observe the effect of scour on pile load and shear capacity, and
(c) evaluating the performance of a scoured bridge under flood loading with
expected to scour. Section 9.2 introduces similar research in the following
 9.2 Literature Survey 203

sections, including WSN, domain adaption, and flood sensor techniques. In

Section 9.3, we go over our framework in further depth. Sections 9.4 and
9.5 discuss the implementation process and undertake extensive experiments.
Finally, Section VI brings this work to a conclusion.

9.2 Literature Survey

Wireless sensor networks and remote sensing have played essential roles
in this investigation. However, early warning systems and remote sensing
have gained more attention in recent years. For early warning systems,
Marek et al. [4] utilized WSNs for real-time remote sensing and provided a
framework to aid in operating these systems. Pradhan et al. propose using
an ensemble to generate flood probability indices. These scientists used
geographic information systems (GIS) and remote sensing data (RSD). These
models rely on satellite data, which produces a large amount of data, so they
demand a sizable amount of resources to operate. Supercomputers can be
used to organize and process the massive amounts of data necessary for flood
forecasting. Countries with weak economies or low living standards do not
have the means to access these materials.
Naji et al. proposed an energy-efficient, decentralized, hierarchical,
cluster-based routing strategy for WSNs. To lessen the load on the power
supply from sending and receiving control messages, they employ techniques
like clustering and multichip routing [6]. Based on the results of the research
by Kumar et al., several improvements to the LEACH routing protocol for
WSN have been proposed [7]. Based on the LEACH protocol, Aravapalli
et al. [8] proposed a hierarchical protocol with enhanced power efficiency
and lifetime for the network. The team led by Dr. Aravapalli and colleagues
developed this method. A node’s energy level is irrelevant in LEACH; the
cluster head (CH) is randomly selected so that any node can become CH
regardless of its starting state. This affects the network’s resilience or ability
to function normally despite disruption and repair. Avi et al. proposed a track-
ing method for identifying and following people in thermal infrared images,
which relied on a particle filter in conjunction with background reduction
[9]. A group of researchers led by Daanoune et al. formulated an algorithm
for tracking people has been proposed to assess the efficacy of region-based
passive infrared motion sensors and their applicability in security systems
[10]. These algorithms’ numerous advantages necessitate a sizable investment
of computational resources for their deployment on individual sensor nodes.
To accomplish data hiding, WSNs employ EC-OU, which results in more
communication overhead. Girao et al. carried out this study [11]. There was
204 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

a hit to network performance due to the SLEACH algorithm methodology,

but Sec Leach Oliveira et al. [12] used symmetric essential cryptography
techniques to make the network more secure. Both fundamental methods have
been demonstrated to be safe and efficient for transmitting data. After gener-
ating a symmetric key, Hu et al. [13] hashed it using the ESODR algorithm.
But when image transmission is added, the size of the rudimentary micro-
scopes shrinks dramatically. The EDRLEACH approach is one of many that
rely on the SLEACH method. ORLEACH, proposed by Raj [14], increases
the time limit and the ratio of utilized energy while providing a practical
means of avoiding the need for energy consumption. A proposal for a chaotic
map and an ECC encryption scheme was presented by Shankar and Elhoseny
[15]. Using elliptic curve cryptography, Sumalatha and Nandalal [16] created
the attribute-based encryption method. We used the user’s private keys and
ciphertext annotated with various metadata tags. The proposed method has
been shown to outperform the state-of-the-art alternatives. Chen et al. [17]
ciphered the image using lightweight ciphers (LWCs). The proposed security
model offers a method to select the best possible key. In addition, the pro-
posed method requires the least time to produce the key for image decryption.
The existing algorithms regarding security accuracy regarding input images
are Horse, Baboon, Barbara, and Lena. Farouk et al. [18] proposed a fuzzy-
based cross-layer security approach to identify malicious nodes that disrupt
packet delivery. Installing false monitoring is aided by the enhanced convolu-
tional neural network (ECNN). Jiang et al. [19] favored the elliptic curve
Hill cipher method to ensure security in WSNs. The permutation of keys
aids in improving their size for the selection of image matrix size. Because
of a novel framework that uses clusters and the ECC method, as well as a
cluster-based encrypting routing algorithm, the authors achieved increased
security, decreased setback, and improved packet release percentage. Yavari
et al. present a remote user validation strategy for WSNs based on active
ID and temporal credentials [20]. The authors preferred Burrows–Abadi–
Needham (BAN) logic for scheme validation. The proposed techniques are
superior regarding low-energy usage and computational cost for authentica-
tion. Rao et al. [21] showed that a hierarchical K-means cluster formation
with knowledge QoS may be used to choose firefly heads for efficient energy
use. QoS parameters are reduced when data is sent between the node and
its destination. Using the protocol, KF-MAC was used to assess QoS factors
like bandwidth, latency, and jitter. Even with minimal power consumption,
the KF-MAC protocol could not transport data without encountering
collisions.
 9.3 Methodology 205

Quality of service (QoS) and energy efficiency routing based on data

categorization was presented by Zhang et al. [22] for industrial WSNs.
Quality of service and energy-efficient methods were used to transmit data
from industrial sensors (QoS). The research shows that data from events
can be sent in real-time with high reliability. The QoS parameter, however,
did not improve results regarding either network life or latency. A qualified
and heterogeneous cluster routing protocol (QHCR) has been developed,
as stated by Amjad et al. [23]. In addition to reducing network energy
consumption, this protocol offers the fastest path for time-critical programs.
Variable-power WSNs reduced latency in time-critical applications and
increased system stability. Actuators and WSNs can benefit from latency-
and power-aware QoS-aware routing protocol as stated by Yahiaoui et al.
[24]. Each cluster within the network is reported to a chief who oversaw
the entire system (CHs). Because of this, the network’s dependability is
improved, and communication delays are cut down significantly. Both com-
munication latency and energy consumption were found to have decreased
noticeably.
Attributable to the work of Gao et al. [25], Mobile Ad Hoc Networks
(MANETs) now have distributed trust measures that can adjust to their sur-
roundings. Counting the number of data packets exchanged between nodes,
forecasting trust based on the path that such value took, and scaling total
trust by comparing its historical significance to its projected value are all
parts of the proposed method for increasing trust in communication. Direct
trust can be calculated by comparing the requested and extended trust. Trust
between nodes was calculated using a combination of direct trust and propa-
gation distance. The results of the studies reveal that the suggested technique
effectively prevented malicious node attacks. For mobile education, Gao et al.
[25], have presented an energy-balanced uneven clustering routing technique,
considering network splits. It has been recommended that data from each
network node be sent to the base station via nodes closer to it, such that nodes
near the base station form a circle around it and split the network region based
on how far away they are from one another. They also created unequal clusters
by generating a variety of viable radii, which helps to balance the network’s
energy consumption. The protocol’s testing results show that it can minimize
node mortality rates, extend network life, and offset each node’s power loss.
The technique of intelligent environment monitoring based on WSNs is
examined in this study, as well as the need for such a sensor monitoring
system and the hurdles we must overcome to achieve it.
206 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

9.3 Methodology
As shown in Figure 9.2, the system depicts the approach for using the sug-
gested method for analyzing hazard bridges. As shown, four main procedures
were defined: (a) UAV-derived DEM generation for seasonal flights via 3D
reconstruction using the Structure from Motion algorithm (SM), (b) hydraulic
analyses using flood-induced scour analyses and flood load/scour depth esti-
mation by HEC-RAS 2D hydraulic modeling, (c) automated 3D FEM gener-
ation for the inspected bridge by SAP2000, and (d) multi-hazard assessment
methods using flood loading as well as earthquake excitation. The diagram
outlined the sub-procedures for getting the DEM, 3D FEM, conducting
hydraulic studies, and assessing the bridge’s multi-hazard performance.

9.3.1 UAV-derived DEM generation by 3D style

The methodology utilized to build a high-resolution DEM incorporated the
point cloud production process using UAV-derived aerial pictures and the
SM technique rather than standard survey methods. One hundred and fifty-
two ground control points (GCPs) were taken along the river region, stream
channel area of approximately 18 km and along the inspected bridge during
the UAV flights with real-time kinematic (RTK) GPS measurements in the
field via a 28 mm focal length lens with a 78.8◦ field of view at f/2.2 via

Figure 9.2 Proposed system.

 9.3 Methodology 207

152 GCPs were taken along the river region. The average flight altitude
was set at 70 m, the overlapping ratios for both the frontal and lateral
directions optimized at 85% and a middle ground sampling distance (GSD)
of 12 cm per pixel was ensured in order to obtain the 3D surfaces in the
riverbed and bridge regions with high accuracy. After camera alignment and
reference, the resulting accuracy in the created point cloud is evaluated during
model creation. Planimetric and altimetric accuracies of 0.050 and 0.080 m,
respectively, were achieved.
Furthermore, the scour depth under bridge piers was calculated using
manual flight height designations with camera angles ranging from 45 to
55 degrees nadir-off angles. The DEM of the riverbed was created after
the dense point cloud was created and before the mesh model was created.
Change detection analyses were performed using the ESRI ArcGIS AddIn-
Geomorphic Change Detection Tool GCD v7.2 and point cloud comparison
to trace changes in riverbed morphology.
To put the SM method into action, the photogrammetric software Pix4D
was utilized. Clusters of motion-integrated two-dimensional images were
used to construct three-dimensional (3D) structural models of terrain, build-
ings, and landforms, and their corresponding geometry, camera position, and
orientation data were simultaneously solved. This allowed the models to
be constructed quickly and accurately. Following the edges and corners of
remotely sensed objects from one image to the next helped establish a connec-
tion between the pictures, and feature trajectories provided evidence for the
object’s location in three dimensions. Using the high-resolution orthophotos
and dense point clouds collected from the riverbed, we could estimate the
scour depths that were present close to the piers and piles of the examined
bridge. The explore depths measured at the ports were used as input data
in developing a 3D FEM of the bridge’s substructure. The aforementioned
UAV-based measuring methodologies are only applicable and helpful for
shallow river systems with dry riverbeds for most of the year. This is the
only type of river system in which these methodologies can be used. As a
result, measurements of bridges located on deep-water rivers are outside the
scope of this research.

9.3.2 Hydrodynamic analyses

The river basin DEM was immediately imported into the HEC-RAS program,
designed for 2D hydrodynamic assessments of rivers with multi-way and non-
prismatic cross sections to estimate flood water depth, flood velocity, flood
208 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

load, and appropriate scour depth. The maximum discharge for the region was
calculated using the Hydrologic Engineering Center’s Hydrodynamic Model-
ing System (HEC-HMS) rainfall–runoff model. Tropical Rainfall Monitoring
Mission (TRMM) satellite, which was developed to provide real-time and
high-resolution rainfall data in the tropics and sub-tropics, can be used to
track rainfall–runoff changes in rivers where the stream gauge network is
insufficient and the intervals of temporal records are inconsistent.

9.3.3 3D FEM generation

The SAP2000 programmer was used to create the 3D FEM for the RC bridge,
which included substructure components such as UAV scour depth observa-
tions and HEC-RAS scour depth estimates for the bridge piles. Nonlinear 3D
beam elements were created to describe RC components like piers and piles
and linear springs for elastomeric bearings. Regarding the project designs,
the composite deck was treated as a linear element. As a result, instead of
manually specifying the bridge members and soil spring characteristics in
SAP2000 software (Figure 9.2), an input file was prepared using MATLAB.

9.3.4 Tectonic evaluation

Nonlinear ETHA was carried out in the seismic evaluation utilizing the 3D
FEM of the bridge, which directly integrated the nonlinear and inelastic
load-deformation characteristics of the bridge elements. The bridge was
subjected to 11 lateral earthquake excitations, represented by ground motion
acceleration histories chosen from the Pacific Earthquake Resistant Research
Center ground motion database as compatible with the bridge’s location
concerning nearby faults, fault mechanisms, and site-specific geotechnical
characteristics. As a result, the recordings were amplitude scaled so that the
average of the suite’s maximum direction spectra did not fall below 90% of
the target elastic design spectrum for the 0.2L1–2L1 time range. According
to the Turkish Earthquake Code, the target spectrum was established as the
5% damped adaptable design spectrum for the location. L1 denotes the first
basic period of the bridge. Simultaneously in the longitudinal and transverse
directions of the bridge, scaled acceleration data were enforced.

9.3.5 Soil modeling

The number, thickness, kind, distinctive features, and number of segments
of each soil layer were utilized as the critical parameters in the created code
 9.3 Methodology 209

for the input file. The soil deposits around the substructure components (i.e.,
pier columns, abutments, and piles) were described by lateral and vertical
bidirectional nonlinear spring elements using py and tz curves, respectively,
to integrate soil structure interaction (SSI). The preliminary factors used to
define the curve features for sand and limestone layers at a given depth are
the water table height, soil type, and relative height of the soil layer to the
ground surface. Each soil layer was separated into numerous soil segments
for multilayer soil deposition, with the nonlinear soil spring characteristics
presumed constant along the way. Because the soil spring properties were
described in terms of the difference between the mid-height of the soil
segment and the ground surface, the new ground level after scouring was
used to determine the soil spring properties. The substructure’s piles and other
underground portions were designed to have the same number of segments as
the soil layers to ensure compatibility and direct assignment of soil spring
characteristics to the bridge components.
The ordinates of the py curves (i.e., p values) at the mid-height of each
soil segment were computed by multiplying the height of the pile segment
under consideration by the ordinates of the py curves (i.e., p values). After
multiplying the prescribed vertical unit resistances by the surface area of the
pile segment, curve characteristics were assigned to the mid-height of the pile
segments. The lateral and vertical resistance supplied by each soil segment
was given to the appropriate pile segment using bidirectional and nonlinear
spring elements. A similar procedure was used on bridge piers and abutments
that were wholly or partially immersed in the soil. The impact of passive or
active earth pressures was neglected since the py curves for the abutments
were derived in both longitudinal and transverse orientations. To organize the
soil spring assignments along the piles, the input file employed the outputs
of seasonally observed scour depths or predicted flood-induced scour depths.
The soil spring assignments (py and tz curves) were not considered for the
scoured sections of the piles.
As a result of the altered ground level, soil type, and relative height to
ground level after scouring, the soil spring characteristics were computed
and allocated to the pile segments below the scour depth. Furthermore, the
negative impact of regularly aligned piles on pile groups’ lateral resistance
was recorded using p-multipliers, which represented the worsening of lateral
soil resistance when shadowing was considered.
Figure 9.3 illustrates the modeling stages used to convert meteorological
data into flood inundation data that is important to the effects on various
sectors of society. Figure 9.4 is an example of output from the modeling chain
210 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

Figure 9.3 Modeling procedures to access flood consequences from meterological effects.

Figure 9.4 Model data visualization: (a) precipitation (input), (b) runoff, (c) mizuRoute river
flow, and (d) flood depth.

for a flood event in a small watershed. The following are the modeling phases
in this framework.
• preparation of meteorological input;
• rainfall–runoff modeling: we use the modular modeling framework
FUSE (Framework for Understanding Structural Errors).
 9.3 Methodology 211

To develop an ensemble of conceptual hydrological models:

• river routing: we utilize a stand-alone river routing programmer to
predict flow along river channels;
• flood inundation modeling: we utilize a 2D flood inundation model to
evaluate flood danger.

9.3.6 Bridge modeling

The created code described all bridge members, including the deck (slab and
supporting girders), elastomeric bearings, pier caps, piers, abutments, pile
caps, and piles. 3D beam components were used to simulate the piles, pier
columns, pier caps, and supported girders that sat on elastomeric bearings.
Elastomeric bearings were also conceptualized as link members that provided
axial and shear stiffness. The existence of the RC shear keys between the
rafters, however, precluded lateral motions. The RC bridge deck, which
consisted of an RC slab and evenly spaced pre-stressed concrete (PC) girders,
was represented as a linear 3D beam element with the same cross-section
area, the moment of inertia, weight, and mass as the composite superstructure.
Furthermore, at abutments and piers, longitudinally oriented gap compo-
nents designated at the end of each girder reflected the spacing between two
successive rafters. Due to girder hammering, very high rigidity was expected
after closing the stipulated gap length. Firm beam components were used to
simulate the pile caps, guaranteeing a rigid connection between the piers and
the piles.
The pile cap connection was viewed as one of the plastic hinge points
for heaps. Sequential pushover investigations in longitudinal and transverse
directions and the continuous recording of the maximum bending moment
region throughout the pile’s length were used to determine the likely locations
of all additional hinges. It was determined that the effective stiffness of most
of the structural components was half of their gross stiffness during pushover
tests. Plastic hinges’ lateral and curvature capacities were determined using
standard section studies, considering the bridge’s position along it and its
axial forces during pushover analyses. Additionally, each axial stress level
and orientation angle of the member were bi-linearized, allowing it to be
included in plastic hinge requirements. These plastic hinge lengths were
assumed to match the depth of cross-sectional area for ease of calculation.
Federal highway administration (FHWA) computed the flood loads
operating on the submerged bridge deck using the primary parameters of
free stream velocity (v), inundation ratio (h), water density (*), and deck
212 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

Figure 9.5 (a) Parameters that have been set; (b) the pressures and moments acting on the
flooded bridge deck.

dimensions (U, T, and s for deck width, length, and height, respectively) as
shown in Figure 9.5 (a).
The inundation ratio eqn (9.1) was calculated by dividing the height
difference between the free surface during a flood (hu ) and the lowest chord
of the bridge deck (hb ) by the deck height (s), including the battlements.
hu − hb
h∗ = . (9.1)
s
Higher inundation ratios indicate that the bridge will be submerged more.
Following the hydrodynamic analyses, the stream velocity and estimated
water level following a flood event were determined.
 
CT = FT 0.5ρr2 T U (9.2)
 
Ngc = FN 0.5ρr2 T U 2 . (9.3)
After the hydraulic analyses, the computed parameters of floodwater
depth and floodwater velocity were utilized to determine the flood loads in the
created input file. The flood loads were computed by multiplying the surface
area of the bridge deck and the scoured sections with the expected inverted
triangle pressure distribution along the depth while considering flooding.
Nonlinear ETHA was carried out in the seismic evaluation utilizing the
3D FEM of the bridge, which directly integrated the nonlinear and inelastic
load-deformation characteristics of the bridge elements. The bridge was
subjected to 11 lateral earthquake excitations, which were represented by
ground motion acceleration histories chosen from the Pacific Earthquake
Engineering Research Center ground motion database as being compatible
with the bridge’s location in relation to nearby faults, fault mechanisms, and
site-specific geotechnical characteristics.
As a result, the recordings were amplitude scaled so that the average of the
suite’s full-direction spectra did not fall below 90% of the target elastic design
 9.4 Result and Discussion 213

spectrum for the 0.2T1–2T1 time range. According to the Turkish Earthquake
Code, the target spectrum was established as the location’s 5% damped elastic
design spectrum. The first basic period of the bridge is denoted by T1.
Simultaneously, in the longitudinal and transverse directions of the bridge,
scaled acceleration data were enforced.

9.4 Result and Discussion

The DEM of the river basin was obtained through the SM-derived point
clouds and top-view elevation model for the case study bridge by first using
high-resolution UAV imageries and combining GCPs acquired along the river
and the bridge. Between November 2016 and January 2021, 11 UAV flights
were used to detect seasonal variations in scour depths around bridge piers.
The examined bridge was photographed using seasonal UAV high-resolution
images in Figure 9.6.
The analyzed bridge was exposed to ETHA following the nonlinear dead
load scenario with the selected and scaled suite of ground motion records
with the desired spectrum to assess the present performance of the bridge
(Table 9.1 and Figure 9.6).

9.4.1 Scour depth and flood load calculations by hydraulic

modeling
The Boaçay River basin border was recovered using TanDEM-X (TerraSAR-
X add-on for Digital Elevation Measurement) digital elevation data with a

Figure 9.6 UAV images taken between the seasons (a)–(k) November 2016 and January
2021.
214 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

Table 9.1 Ground motion collection.

RJB VS30 PGA PGV PGD
Earthquake Station Mw Comp. Scale factor
km m/s g cm/s cm
012 0.270 24.8 9.3
Imp. Val. 06 Chihuahua 6.5 7.3 242 2.25
282 0.254 29.9 7.7
140 0.365 36.0 25.1
Imp. Val. 06 E1 [Link].#11 6.5 12.6 196 1.89
230 0.380 44.6 21.3
270 0.352 75.5 57.1
Imp. Val. 06 El [Link]. 6.5 5.1 202 1.37
360 0.481 40.9 16.4
000 0.357 48.0 19.3
[Link] 02 El [Link]. 6.5 18.2 192 1.90
090 0.260 41.8 21.9
270 0.475 41.2 7.7
[Link] 02 Poe Road 6.5 11.2 317 1.96
360 0.286 29.0 11.4
090 0.173 23.5 15.0
[Link] 02 [Link] Sta. 6.5 13.0 194 2.33
180 0.211 32.3 22.3
000 0.283 27.6 18.2
Landers Coolwater 7.3 19.7 353 1.73
090 0.417 43.4 15.2
000 0.276 33.6 26.6
Kobe Amagasaki 6.9 11.3 256 1.38
090 0.327 44.8 23.8
000 0.296 24.5 7.6
Kobe Tadoka 6.9 31.7 312 2.96
090 0.194 14.7 10.3
180 0.312 58.8 44.0
Kocaeli Duzce 7.5 13.6 282 1.32
270 0.364 56.6 25.0
000 0.739 55.9 25.6
Duzce Bolu 7.1 120 294 0.94
090 0.806 65.9 13.1

spatial resolution of 12 m before hydraulic modeling (Figure 9.7 (a)). The

flow network, sub-basin boundaries, and adjoint basins by which the basin
boundaries were obtained were disclosed by identifying flow directions and
flow concentration zones (Figure 9.7 (b)). The day TRMM streamflow was
defined with a spatial resolution of 0.25◦ (about 25 km) in latitude and
longitude for each grid, and it was related to its corresponding region in the
basin borders, which comprised the whole UAV measured part in the Boaçay
River basin. Between 1998 and 2021, daily average rainfall time data were
acquired from the TRMM satellite inside the TRMM grid, representing the
study region for flood assessments.

Figure 9.7 (a) Digital elevation data from TanDEM-X; (b) drainage lines with sub-basins.
 9.4 Result and Discussion 215

Figure 9.8 2D hydraulic model of the research area.

The UAV-derived surface models were coupled with a resampled DEM

digital surface model to be utilized in HEC-RAS 2D hydraulic modeling to
offer additional spatial information in the floodplain and to correctly represent
all of the basin areas in the model Figure 9.8.
After considering the nonlinear loading history owing to later flood and
dead load events, the bridge was subjected to ETHA for multi-hazard evalua-
tion. The bridge FEM was created in this example using the maximum scour
depth results from the HEC-RAS scour analysis.
Figure 9.9 shows the flood loading; the bridge members showed a linear
elastic response. When comparing the existing condition of the bridge to the
multi-hazard condition, the displacement demand (UX and UY ) rose in both
directions. It nearly doubled in the transverse direction (y).
The flood-induced scour was demonstrated to have no impact on the inter-
nal forces of the pier columns (Figure 9.10 (a) and (b)). However, shear details
in the transverse direction (Vy ) were found to have increased somewhat from
1600 to 2200 kN, as illustrated in Figure 9.10. For the abutment shear walls,
the moments along the longitudinal axis (i.e., strong axis) (Mx) were seen to
almost quadruple from 7000 to 14,000 kNm.
216 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

Figure 9.9 The scaled suite’s RotD100 spectrum, the selected record, and the desired
spectrum.

Figure 9.10 For current and multi-hazard circumstances, the maximum pier column dis-
placement at the higher ends.

For the multi-hazard situation, a significant fluctuation in internal forces

was seen concerning the rise in scour depth (Figure 9.11 (a) and (b)). Due to
 9.4 Result and Discussion 217

Figure 9.11 For current and multi-hazard circumstances, the maximum pier column internal
forces.

the stiffness degradation in piles caused by the loss of the surrounding soil
layers after scouring, the internal stresses increased by two to six times. This
was especially true for shear forces in the bridge’s longitudinal direction (Vx )
(Figure 9.12). The lateral response of the bridge, considering plastic hinge
locations and the hysteretic behavior of the selected bridge members, was
monitored during seismic loading by ETHA using the El Centro Imp—Co.
Cent. ground motion record, which was chosen at random from the ground
motion suite.
The rotation performance levels of plastic hinges were tracked using
ASCE/SEI 41-17, which considered rotation limits from A to E, with imme-
diate occupancy (IO), life safety (LS), and collapse prevention (CP) limit
states as acceptance criteria. All pier columns were monitored for rotation
levels between IO and LS beyond yielding in the current state, although the
abutments and piles were proven to respond elastically. In addition to the pier
columns, several of the banks in the multi-hazard scenario showed inelastic
reaction via plastic hinging with a lower rotation level across B and IO than
the piers panels.
Between A and B, the abutment’s lateral reaction remained elastic. The
pier column and pile in Pier 2 were chosen for lateral response comparison
in terms of hysteretic moment – rotation behavior under bi-directional lateral
218 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

Figure 9.12 For current and multi-hazard scenarios, the maximum pile of internal forces.

cycles due to the selected record to track rotation levels. The chosen mem-
bers’ lateral displacement, pier column internal force, and pile internal force
variations were shown (represented as the red star) with the selected ground
motion record. Regarding the moments along the x-axis, which defined the
bridge’s transverse behavior, the pier column and pile were found to be in
the elastic range for the present situation. However, minimal plastic hinging
was detected for the bank under the multi-hazard situation. The pier column
displayed inelastic behavior with a significant plastic rotation level in the lon-
gitudinal response (i.e., moments around the y-axis). Still, the pile exhibited
elastic behavior in the current circumstances. The pier column, on the other
hand, displayed an inelastic reaction with a lower plastic rotation level under
the multi-hazard situation, as the piling was found to reach the inelastic range
with modest plastic hinging.
As a result, the seismic load and displacement demands for the bridge
with a more considerable scour depth were demonstrated to migrate from
the pier columns to the piles after exploring in a multi-hazard scenario. This
is because in the absence of soil after researching, the internal forces that
were shared by the pile and the surrounding soil before grinding were alone
resisted by the piles. As a result, it may be deduced that following scouring,
the internal pressures previously resisted by the soil were transferred to the
 References 219

piles. This behavior resulted in a rising pile of internal forces (Figure 9.11 (a)
and (b)) and, as a result, increased rotations and lateral displacements (Fig-
ure 9.9), whereas the pier column internal forces remained relatively constant
(Figure 9.10). Furthermore, increased pile flexibility following the removal
of adjacent soil layers resulted in increased lateral displacements. The lateral
stresses operating in the longitudinal direction caused plastic hinging in the
pier columns instead of the pile group in the examined bridge due to the
arrangement of the columns such that their weak axis corresponded with the
bridge’s transverse axis.
Furthermore, the flood lateral forces were resisted by the pier columns’
strong axis, and the scoured pile group opposed the flood loads after explor-
ing, resulting in a linear elastic response because the flood loads were found
insignificant in terms of the pier column and pile group’s lateral load capac-
ities. As a result, it was determined that scour significantly impacted the
behavior of laterally laden piles and that the interaction of bridge components
and soil includes multi-hazard performance evaluations of bridges.

9.5 Conclusion
This work proposed an automated UAV-based bridge multi-hazard assess-
ment system for RC bridges crossing seasonal rivers under flood loading and
seismic excitations, including the impacts of scouring. Because the UAV-
based measuring approach proved unable to DEM buildings in submerged
regions, the proposed method can only be used on bridges located on rivers
with a dry riverbed for most of the year. The floodwater depths were deter-
mined after the HSC-RAS hydrodynamic assessments, and the consequent
flood-induced scour depths were derived to forecast the multi-hazard per-
formance. After that, the bridge FEM was created automatically. Finally,
after flood loading, nonlinear ETHA was used to assess the bridge’s multi-
hazard performance with flood-induced scour depths, considering lateral
displacements, internal forces, and plastic rotations. Furthermore, offering
strong security to the WSN environment motivates the creation of security
algorithms.

References
[1] Bhatt, Harshil, G. Pranesh, Samarth Shankar, and ShriyashHaralikar.
“Wireless Sensor Networks for Optimisation of Search and Res-
cue Management in Floods.” In 2021 IEEE International Conference
220 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

on Electronics, Computing and Communication Technologies (CON-

NECT), pp. 1-6. IEEE, 2021.
[2] Al-Mashhadani, Mohammad A., Mustafa MaadHamdi, and Ahmed
Shamil Mustafa. “Role and challenges of using a UAV-aided WSN
monitoring system in large-scale sectors.” In 2021 3rd International
Congress on Human-Computer Interaction, Optimization and Robotic
Applications (HORA), pp. 1-5. IEEE, 2021.
[3] Prasad, Devendra, Afshan Hassan, Deepak Kumar Verma, Pradeep-
taSarangi, and Sunny Singh. “Disaster Management System using Wire-
less Sensor Network: A Review.” In 2021 International Conference on
Computational Intelligence and Computing Applications (ICCICA), pp.
1-6. IEEE, 2021.
[4] Kumar, P. Praveen, T. Ananth Kumar, Pavithra Muthu, Rajmohan
Rajendirane, and R. Dinesh Jackson Samuel. “Security Analysis of UAV
Communication Protocols: Solutions, Prospects, and Encounters.” In
The Internet Of Drones, pp. 167-190. Apple Academic Press, 2022.
[5] H. Mojaddadi, H. Pradhan Biswajeet Nampak, N. Ahmad, and A. H.
b. Ghazali, “Ensemble machine learning- based geospatial approach for
flood risk assessment using multi-sensor remote-sensing data and gis,”
Geomatics, Natural Hazards, and Risk, vol. 8, pp. 1080–1102, 2 2017.
DOI 10.1080/19475705.2017.1294113.
[6] M. Sabet and H. R. Naji, “A decentralized energy efficient hierarchical
cluster-based routing algorithm for wireless sensor networks,” AEU -
International Journal of Electronics and Communications, vol. 69, pp.
790–799, 2015. DOI: 10.1016/[Link].2015.01.002.
[7] Kumar, T. Ananth, S. Arunmozhi Selvi, R. S. Rajesh, and G. Glorindal.
“Safety wing for industry (SWI 2020)–an advanced unmanned aerial
vehicle design for safety and security facility management in industries.”
In Industry 4.0 Interoperability, Analytics, Security, and Case Studies,
pp. 181-198. CRC Press, 2021.
[8] R. K. Kodali and N. K. Aravapalli, “Multi-level leach protocol using
ns-3,” 2014 International Conference on Circuits, Systems, Commu-
nication and Information Technology Applications, CSCITA 2014, pp.
132–137, 2014. DOI: 10.1109/CSCITA.2014.6839248.
[9] Avi, Alessandro, Matteo Zuccatti, Matteo Nardello, Nicola Conci,
and DavideBrunelli. “Infrared dataset generation for people detection
through the superimposition of different camera sensors.” In Proceed-
ings of the IEEE/CVF International Conference on Computer Vision,
pp. 307-316. 2021.
 References 221

[10] Daanoune, Ikram, Abdennaceur Baghdad, and AbdelhakimBallouk.

“Improved LEACH protocol for increasing the lifetime of WSNs.” Inter-
national Journal of Electrical & Computer Engineering (2088-
8708) 11, no. 4 (2021).
[11] E. Mykletun, J. Girao, and D. Westhoff, “Public key-based crypto
schemes for data concealment in wireless sensor networks,” IEEE
Int. Conf. Commun., vol. 5, no. c, pp. 2288–2295, 2016, doi:
10.1109/ICC.2006.255111.
[12] Oliveria, L., A. Ferreira, M. A. Vilaca, H. Wong, M. Bern, R. Dahab,
and A. A. F. Loureiro. “SecLEACH-On the security of clustered sensor
network.” Journal of Signal Processing 87, no. 12 (2007): 2882-2895.
[13] D. Wu, G. Hu, G. Ni, W. Li, and Z. Zhang, “Research on secure routing
protocols in wireless sensor networks,” Chinese J. Sensors Actuators,
vol. 21, no. 7, pp. 1195–1201, 2018.
[14] Raj, E. D. (2012). An Efficient Cluster Head Selection Algorithm
for Wireless Sensor Networks–Edrleach. IOSR Journal of Computer
Engineering (IOSRJCE), 2(2), 39-44.
[15] Shankar, K., & Elhoseny, M. (2019). Secure Image Transmission in
Wireless Sensor Network (WSN) Applications. Springer International
Publishing.
[16] Sumalatha, M. S., Nandalal, V. An intelligent cross-layer security based
fuzzy trust calculation mechanism (CLS-FTCM) for securing wire-
less sensor network (WSN). J Ambient Intell Human Comput (2020).
[Link]
[17] C. T. Chen, C. C. Lee, and I. C. Lin, “Efficient and secure three-
party mutual authentication key agreement protocol for WSNs in
IoT environments,” PLoS One, vol. 15, no. 4, pp. 1–28, 2020, doi:
10.1371/[Link].0232277.
[18] Elhoseny, M., Farouk, A., Batle, J., Shehab, A., & Hassanien, A. E.
(2017). Secure image processing and transmission schema in a cluster-
based wireless sensor network. In Handbook of research on machine
learning innovations and trends (pp. 1022-1040). IGI Global.
[19] Q. Jiang, J. Ma, F. Wei, Y. Tian, J. Shen, and Y. Yang, “An untraceable
temporal-credential-based two-factor authentication scheme using ECC
for wireless sensor networks,” J. Netw. Comput. Appl., vol. 76, pp. 37–
48, 2016, doi: 10.1016/[Link].2016.10.001.
[20] A. Yavari, D. Georgakopoulos, P. R. Stoddart and M. Shafiei, “Inter-
net of Things-based Hydrocarbon Sensing for Real-time Environmen-
tal Monitoring,” 2019 IEEE 5th World Forum on Internet of Things
(WF-IoT), 2019, pp. 729-732, doi: 10.1109/WFIoT. 2019.8767320.
222 A Multi-hazard Industry Assessment System Based on Unmanned Aerial

[21] Rao, Meena, and Neeta Singh. “Energy efficient QoS aware hier-
archical KF-MAC routing protocol in MANET.” Wireless Personal
Communications 101, no. 2 (2018): 635-648.
[22] Zhang, W., Liu, Y., Han, G., Feng, Y., & Zhao, Y. (2018). An energy-
efficient and QoS-aware routing algorithm based on data classification
for industrial wireless sensor networks. IEEE Access, 6, 46495-46504.
[23] Amjad, M., Afzal, M. K., Umer, T., & Kim, B. S. (2017). QoS-aware
and heterogeneously clustered routing protocol for wireless sensor
networks. IEEE Access, 5, 10250-10262.
[24] Yahiaoui, S., Omar, M., Bouabdallah, A., Natalizio, E., & Challal, Y.
(2018). An energy-efficient and QoS-aware routing protocol for wireless
sensor and actuator networks. AEU-International Journal of Electronics
and Communications, 83, 193-203.
[25] Zhang, D. G., Gao, J. X., Liu, X. H., Zhang, T., & Zhao, D. X. (2019).
A novel approach of distributed & adaptive trust metrics for MANET.
Wireless Networks, 25(6), 3587-3603.
 10
Air Quality Prediction using Machine
Learning Techniques for Intelligent
Monitoring Systems

Marimuthu Rajendran Ezhilkumar1 , Singaram Karthikeyan2 ,

Dimplekumar N. Chalishajar3 , and Rajappa Ramesh4
1 Department of Civil Engineering,
Sri Krishna College of Engineering and Technology, India
2 Centre for Environmental Studies, Anna University, India
3 Department of Applied Mathematics,

Virginia Military Institute (VMI), USA

4 Department of Science & Humanities,

Sri Krishna College of Engineering and Technology, India

E-mail: [Link]@[Link]; ksingaram @[Link];
dipu17370@[Link]; rameshrajappa1982@[Link]

Abstract
Real-time large data collection in air quality studies was found to be compli-
cated. To overcome such scenarios, machine learning and IoT systems help
researchers and regulatory systems to improve air quality and make decisions.
This study has used the ANN approach to predict the air quality from the
measured data for a street in Chennai, an urban metropolitan of Tamil Nadu.
This study studied particulate pollutants of size 2.5 and 10 microns in street
canyons at three different heights above ground level. Also, three different
street canyons were chosen in this experimental investigation. Results showed
that the concentration of the pollutants was not similar at all heights in
all three streets. Seasonal changes were also analyzed as the monitoring
period was one completed year. No similar profiles were identified among
the seasons in all three street geometries. However, a minimum concentration

223
224 Air Quality Prediction using Machine Learning Techniques for Intelligent

load was observed during summer than other monitored seasons. Further, a
back propagation neural network model was implemented in this study to
predict the vertical concentration trend of PM2.5 and PM10 pollutants. The
results of the models showed good prediction and confirmed the suitability of
the model to predict the vertical trend of PMs in any street canyon study.
It also justifies the importance of street geometry, emission source, and
meteorological factors in predicting the air quality of a typical street canyon.
This mature field is being spun into commercial applications for developing
intelligent air quality monitoring devices.

Keywords: particulate matter, street canyon, meteorological factor, neural

network, ANN

10.1 Introduction
Machine learning and intelligent systems for air quality prediction play a
major role in the current trend as they are found to be cost-effective and
advanced techniques. Air quality in the present era is of serious concern, as
they lead to significant health issues. Moreover, these issues are often reported
in epidemic studies conducted in an urban environment due to urbanization
and industrialization. Urbanization has brought a scarcity of construction
space in the urban boundary and a new change in the building style to accom-
modate dense populations in small areas to satisfy the shelter needs of the
migrants moving from rural to urban regions. The new change is the establish-
ment of high-rise buildings, having more dwelling units constructed vertically
one over the other in the defined plot area. Though it is a well-appreciated
construction, there is a negative impact too, i.e., weakening the ventilation
capacity inside the street, leading to pollution hotspots. Hence, researchers
from various fields showed interest in understanding and improving the air
quality in such environments.
“Street Canyon” (SC) refers to the continuous presence of buildings along
either side of the road [1]. SCs are known to be hot-spots of the urban due
to the confinement of traffic emissions and low ventilation inducing poor
atmospheric dispersion [2]. Urban air quality studies have been conducted
by many researchers and have reported that PM concentration seems to
be higher in the urban environment compared to other environments. The
sources of pollutant emission in this type of environment are road traffic.
This again increases the health risk of occupants and pedestrians in the
 10.1 Introduction 225

street environment. Hence, researchers from various fields show interest in

understanding and improving the air quality in such environments.
Monitoring and modeling spatial air quality are standard and conventional
research. Whereas changes in the building orientation, architectural aspects,
and city infrastructure demand transformation in air quality studies. In this
way, the concept of vertical air quality monitoring is performed by many
researchers worldwide on a laboratory scale through field investigation and
simulation models. The parameters that were taken into consideration in
the street canyon air quality studies include the geometry of the street [3],
emission sources [4, 5], meteorological parameters [6], and solar radiation
[7]. More of these sectoral studies were carried out using simulation and
prediction models. Artificial neural network (ANN) is a widely used pre-
diction model for air quality prediction studies [8]. Compared to earlier
statistical methods, ANN techniques can increase forecast accuracy. Back-
propagation neural network (BPNN), multilayer perceptron (MLP), radial
basis function (RBF), and adopted neuro-fuzzy inference system (ANFIS)
are a few examples of ANN types.
Djebbri and Rouainia [8] performed a study using the ANN model to
predict the importance of meteorological factors in deciding the ambient air
pollutant concentration of the pollutants, i.e., NOx and CO, measured in the
four major parts of Algeria. The model was trained for three different scenar-
ios: in scenario 1, only meteorological variables were trained; in scenario 2,
only pollutant concentrations were trained; and in scenario 3, both pollutant
concentration and meteorological parameters. Model 3 performed well in
the simulation, and predictions were significant compared to the other two
scenarios. Kurt et al. [9] used a feedforward neural network (FNN) to predict
the concentrations of SO2 , PM10 , and CO in the Greater Istanbul Area. The
prediction was performed in two stages of data training. First, the data from
day 1 to day 3 were trained and predicted. Secondly, the day 2 and day 3
concentration profiles were predicted using the first stage predicted dataset.
Later the size of the dataset was optimized. Utilizing 3–15 historical days in
the training dataset results in the most significant modeling performance with
the lowest error rate. Authors see improved predictions with greater accuracy
when the day of the week is included as an input parameter. Siwek et al. [10]
conducted a forecasting study on PM10 pollutants using a neural network
approach. In this study, various ANN models of a similar kind were used to
identify the best prediction models. MLP produces better results, which were
significant compared to the observed results.
226 Air Quality Prediction using Machine Learning Techniques for Intelligent

Cortina-Januchs et al. [11] applied the ANN model to predict the con-
centration of SO2 pollutants for an industrial zone in Salamanca, Mexico.
The authors trained the model for two results: (1) to identify the dataset size
for best prediction; (2) to establish the significant meteorological parameter
influencing the pollutant concentration. The prediction accuracy of the trained
models was validated using MAE and RMSE. The authors used FNN in the
training model. The results showed that the most miniature dataset predicted
better results and defined the importance of meteorological parameters in
the model for the forecasting results. Arhami et al. [12] investigated the
applicability of ANN and Monte Carlo simulations (MCSs) combination
to predict the hourly concentration of air pollutants, i.e., nitrogen oxides
(NOx), nitrogen dioxide (NO2 ), nitrogen monoxide (NO), ozone (O3 ), carbon
monoxide (CO), and particulate matter of diameter of less than 10 μm (PM10 )
for Tehran city. In addition, the major influencing meteorological parameters
were also studied in the model. The trained cum optimized ANN model
showed the best prediction results for the pollutants, i.e., CO, NOx, NO2 ,
NO, and PM10 , with R2 of 0.82 – 0.92 between the observed and predicted
values. While predictions show a poor correlation between O3 pollutants with
the observed values. The study concludes that ANN can be used as a good
prediction model provided proper input, the architecture of the model, and the
order of input variables play a significant role in making the model reliable.
ANN is a well-established model with high nonlinear relationships and
can be trained precisely when new datasets are used for predictions. In this
way, this paper discusses the use of ANN in predicting air quality concen-
tration in three geometrically varying street geometries, which has a scope in
intelligent monitoring systems used for air monitoring in industries.

10.2 Materials and Methods

The field-based investigation was conducted in Chennai, Tamil Nadu, India.
The streets chosen in this region represent southern India’s typical street
models. Here, for the study, three street models were chosen. They are: (1)
non-street canyon (NSC), (2) asymmetrical street canyon (ASC), and (3)
symmetrical street canyon with viaduct (SSCV). The sampling locations are
shown in Figure 10.1. The details of the street can be referred to in [12]. The
parameters considered in this work are listed in Table 10.1.
Particulate matter (PM) was measured using ambient fine dust samplers
(AFDSs). The vertical measurement of PM was done by placing three AFDSs
 10.2 Materials and Methods 227

Figure 10.1 Map showing the sampling locations.

on the open terrace of the buildings at three different heights. The meteo-
rological parameters were monitored using a portable weather monitoring
station placed at the top of the building. The classified vehicle count study
was performed by video surveying method.

Table 10.1 List of parameters measured and monitored in this study.

[Link]. Description Parameters
1 Pollutants monitored PM2.5 , PM10
2 Meteorological factors Wind speed, temperature, humidity, pressure
3 Other influencing factors Vehicle traffic, street orientation
4 Frequency of monitoring Season:
of the above data ([Link]. 1 winter, summer, southwest monsoon, and north-
to 3) except street orienta- east monsoon
tion No of sampling days per season:
6 days per sampling location per season
5 Sampling height above NSC: 3 m, 16 m, 37 m
street level ASC: 3 m, 16 m, 38 m
SSCV: 3 m, 17 m, 37 m
228 Air Quality Prediction using Machine Learning Techniques for Intelligent

In the ANN model, a typical neural network type, i.e., feedforward

neural network (FNN), was used. The model layer is shown in Figure 10.2.
The input layers were categorized into three sectors: I 1 – pollutant type,
concentration, and sampling height; I 2 – meteorological parameters; and I 3
– vehicle traffic and street geometry. The model was run in different phases to

Figure 10.2 Neutral network architecture was developed for this study.

Figure 10.3 Mass concentration profile of NSC.

 10.2 Materials and Methods 229

Figure 10.4 Mass concentration profile of ASC.

Figure 10.5 Mass concentration profile of SSCV.

230 Air Quality Prediction using Machine Learning Techniques for Intelligent

Figure 10.6 PM concentration prediction by ANN model for NSC.

 10.2 Materials and Methods 231

Figure 10.7 PM concentration prediction by ANN model for ASC.

232 Air Quality Prediction using Machine Learning Techniques for Intelligent

Figure 10.8 PM concentration prediction by ANN model for SSCV.

 10.4 Conclusion 233

predict: (O 1) – pollutant concentration concerning each monitoring height;

(O 2) – major factors influencing the pollutant fate inside the respective street
canyons. The statistical methods that were used for the analysis were the root
mean square error (RMSE) and the mean absolute error (MAE) to define the
linear regression between observed and predicted values.

10.3 Results and Discussion

This section presents the report of the PM concentrations, ANN prediction
of PM concentrations, and the major influencing factors in all three street
geometries.
The mass concentration profiles (Figures 10.3–10.5) show that maximum
concentrations were observed in winter for both pollutants in all three street
geometries. Among the three sampling locations, PM2.5 and PM10 were
maximum at ASC. This could be due to various reasons, such as the street’s
geometry, local emission sources, and no better dispersion conditions.
The ANN model showed a good concentration prediction for the mea-
sured pollutants in all three street geometries (Figures 10.6–10.8). The
prediction accuracy was about 94 percent from the overall observation, with
an RMSE of 0.0114 for PM2.5 and 0.0132 for PM10 . At the same time, MAE
for PM2.5 and PM10 were 0.0623 and 0.0498, respectively. The adequacy of
the model with FNN showed 76 percent in the concentration prediction.
Similarly, the ANN prediction results of influencing factor showed sig-
nificance for three parameters, i.e., street geometry, wind speed, and vehicle
traffic.

10.4 Conclusion
This work investigated the applicability and efficiency of the ANN tool for
predicting PM vertical concentration of size 2.5 and 10 in three varying street
geometries. In addition, the factors influencing the pollutant profile inside the
street geometries.
For the observed four seasons, the average correlation coefficient (R)
between measured and predicted PM2.5 and PM10 were 0.86 and 0.88, respec-
tively. The accuracy of the ANN-developed model showed 94 percent for the
prediction of PM concentrations in all three street geometries. Further, three
significant factors were predicted to influence determining the PM pollutant
concentration inside the street, i.e., street geometry, wind speed, and vehicle
traffic.
234 Air Quality Prediction using Machine Learning Techniques for Intelligent

Hence it is put forth to the research community that ANN models are a
way forward tool in predicting air quality scenarios for street canyons. This
model can be used in intelligent air monitoring devices to predict the quality
of air using measured data. Further, the model can be trained with other
environmental conditions and local geographical factors to make it more valid
and reliable, which opens the scope for future work in this area.

References
[1] SJ. Jeong and MJ. Andrews, Application of the k–e turbulence model
to the high Reynolds number skimming flow field of an urban street
canyon, Atmospheric Environment, 2002.
[2] AT. Buckland, and DR. Middleton, Nomograms for calculating pollution
within street canyons, Atmospheric Environment, 1999.
[3] MF. Yassin, and M. Ohba, Experimental simulation of air quality in
street canyon under changes of building orientation and aspect ratio,
Journal of Exposure Science and Environmental Epidemiology, 2012.
[4] Y. Zhou, and JI. Levy, The impact of urban street canyons on population
exposure to traffic-related primary pollutants, Atmospheric Environ-
ment, 2008.
[5] T. Zheng, B. Li, X. Li, Z. Wang, S. Li, and Z. Peng, Vertical and
horizontal distributions of traffic-related pollutants beside an urban arte-
rial road based on unmanned aerial vehicle observations, Building and
Environment, 2021.
[6] H. Zhang, T. Xu, Y. Zong, H. Tang, X. Liu, and Y. Wang, Influence
of Meteorological Conditions on Pollutant Dispersion in Street Canyon,
Procedia Engineering, 2015.
[7] J. Dong, Z. Tan, Y. Xiao, and J. Tu, Seasonal Changing Effect on Air-
flow and Pollutant Dispersion Characteristics in Urban Street Canyons,
Atmosphere, 2017.
[8] N. Djebbri, and M. Rouainia, Artificial neural networks based air
pollution monitoring in industrial sites, International Conference on
Engineering and Technology (ICET), Turkey, 2017.
[9] A. Kurt, B. Gulbagci, F. Karaca, and O. Alaghab, An Online Air
Pollution Forecasting System Using Neural Networks, Environment
International, 2008.
[10] K. Siwek, S. Osowski and M. Sowinski, “Neural predictor ensemble for
accurate forecasting of PM10 pollution,” The 2010 International Joint
Conference on Neural Networks (IJCNN), 2010.
 References 235

[11] M. G. Cortina-Januchs, J. M. Barron-Adame, A. Vega-Corona and D.

Andina, “Prevision of industrial SO2 pollutant concentration apply-
ing ANNs,” 2009 7th IEEE International Conference on Industrial
Informatics, 2009.
[12] Arhami, M., Kamali, N. & Rajabi, M. M. Predicting hourly air pollutant
levels using artificial neural networks coupled with uncertainty analysis
by Monte Carlo simulations. Environ Sci Pollut Res, 2013.
[13] M. R. Ezhilkumar, and S. Karthikeyan, Vertical Measurement of PM2.5
and PM10 in Street Canyons and Cohort Health Risk Estimation at
Chennai, South India, Environmental Engineering Science, 2020.
 11
Facial Emotion Classification for Industry
Automation using Convolutional Neural
Networks

R. Bharath Raam1 , Balaji Srinivasan1 , Prithiviraj Rajalingam2 ,

and Dinesh Jackson Samuel3
1 Loyola-ICAM College of Engineering and Technology, India
2 Assitant
Professor, Department of Electronics & Communication
Engineering, SRM Institute of Science & Technology, Kattankulathur,
603203
3 Biomedical Engineering, University of California, USA

Abstract
In today’s world, where businesses are constantly expanding and a need
to transition to a virtual mode of operation exists, all sellers must have a
direct line of sight into what their customers think about their idea, product,
and the people associated with it. This is because moving to a purely dig-
ital infrastructure is necessary for modern business. The proposed solution
analyzes the customer’s macro expressions during a video call. We aim to
utilize a convolutional neural network to classify the emotions expressed
by the customer while on a video call. The proposed solution achieved an
accuracy of 69.90 percent on the FER-2013 dataset and 73.24 percent on the
JAFFE dataset. The proposed solution can achieve better results than the mini
Xception architecture.

Keywords: convolutional neural network, emotion recognition, computer

vision, business intelligence

11.1 Introduction
Emotion recognition is a subsidiary of artificial intelligence technology
related to facial recognition, which aims to interpret the subject’s emotional

237
238 Facial Emotion Classification for Industry Automation using Convolutional

state purely based on facial expressions, physical cues, and body vital signals,
such as heart rate and brain activity. But, in this project, we restrict ourselves
to the least invasive and easily observable indicator of human emotion,
the human face. This is a profitable avenue of research at the forefront of
the development of service robots capable of being cognitive enough to
understand human emotions and act accordingly.
The priority of affective computing is to remove the “robotic” feel for
humans when interacting with computers. Today’s systems are being bottle-
necked by their inability to interpret and reciprocate emotions, which limits
their intelligence and prevents them from interacting with the user, taking into
account the user’s emotional state. The system must be sensitive to human
emotions to be more user-friendly and practical. In order to get an accurate
interpretation of any given information, we need to consider both linguistic
and nonverbal information. It is much needed to plan and build a system that
can recognize a person’s emotions using gestures, facial expressions, acoustic
properties of speech, verbal information, and more, as it can help provide
valuable business intelligence about how customers perceive the product.
In reality, it is viable to detect predetermined emotions by measuring body
signals and facial expressions recorded in real time by sensors and cameras.
According to conventional wisdom, nonverbal communication accounts for
70–90 percent of human communication. Albert Mehrabian’s study in 1967
established the “3V rule,” which stated that verbal, vocal, and visual commu-
nications constitute 7 percent, 38 percent, and 55 percent, respectively, in any
interaction. This explains why nonverbal communication is so exciting and
fortifies our approach to detecting emotions from facial expressions.
One possibility is to use a convolutional neural network (CNN) to investi-
gate the facial expressions of human subjects while they are engaged in video
calls to deduce how those individuals are feeling. In order to obtain more
consistent results when locating the network’s region of interest, the convo-
lutional neural network uses the multi-task cascaded convolutional networks
(MTCNN) framework. This framework was developed for both face detection
and face alignment. We were able to create both a system for recognizing
faces and a system for skeletonizing faces by using this framework.

11.2 Related Works

Understanding and detecting human emotions has been an enigma over the
centuries and plays a significant role in human–human interaction. However,
 11.2 Related Works 239

with the plethora of data and in a world of ever-evolving and increasing

computing power, we are much closer than before to detecting and under-
standing the emotional response of humans. As stated concisely by Pantic
et al. [1], the emotional intelligence of humans is integral for successful
interaction rather than their IQ.
In a world progressing toward intelligent computers and AI, aiming
toward emulating human intelligence, the ability to understand emotions
is a non-negotiable trait it must possess. It goes beyond the fact that it
would make computers human-like as there is significant evidence that the
rational thinking of the homo sapiens is dependent on emotions [2]. Hence,
affective computing is critical to advancing AI [3]. Affective computing is the
study and development of systems and devices that can recognize, interpret,
process, and simulate human effects, such as human emotions. Affective tries
to extract ideas, sentiments, and emotions from people’s behaviors, which can
be collected automatically through their writings, facial expressions, speech,
physiological signs, and movements.
Emotion, sentiment, and opinions are examples of affective data from
the Web, which is the primary data source. Social signal processing (SSP)
aims to analyze and model social interactions to provide computers with an
emotional quotient [4]. The development of human–computer interfaces with
an emotional quotient is the ultimate goal of SSP. Affective computing is
a branch of SSP that combines artificial intelligence and cognitive technol-
ogy to label the emotions of a human subject using data mined from their
facial feature and body gesture [5]. Highly evolved humans can display their
emotions through various means, from facial expressions to body movements,
gestures, and micro-expressions. The tone, choice of words, and inflections in
how people talk can also portray the underlying emotion of the conversation.
Scientific studies show that it is possible to deduce the emotion experienced
by a person by observing the electrical activity in their brain and heart via
electroencephalography (EEG) and electrocardiography (ECG), respectively.
We can also perceive the emotions detected through other physiological
signals such as respiration rate analysis (RR), electromyogram (EMG), and
galvanic skin response (GSR), to name a few [6]. In essence, visual, text,
audio, and physiological signals can be processed and observed to understand
human emotions.
The human face and eyes are the portals to understanding human emo-
tions, as they are very responsive to humans’ underlying emotions. A Simple
video camera equipment is required to extract the emotion expressed by
a person in real-time, and not much-specialized equipment is required [7].
240 Facial Emotion Classification for Industry Automation using Convolutional

The reactions to emotions are very much expressive and observable through
human facial features and are the key to decoding the enigma of understand-
ing emotions. Facial expressions provide a credible, non-invasive method for
communicating with another person and comprehending their emotional state
and the intentions they are trying to convey. The truthfulness of this cannot be
called into question. The area of psychology known as emotion modeling sees
a lot of research activity, and the methods that psychologists use to categorize
and model emotions can vary significantly from one researcher to the next.
Distinctive and dimensional emotion theories are the two main categories into
which the study of emotions can be broken down. According to the discrete
emotion theory, anger, disgust, fear, happiness, sadness, and surprise are the
core categories of emotions. These discrete emotion models detect and cat-
egorize emotions by monitoring facial expressions like lip pursing, eyebrow
raising, and eye blinking. The Ekman model [8] has shown the most promise
among similar frameworks. Emotion recognition from facial expressions can
be accomplished in several ways, from simple histogram of oriented gradient
(HOG)-based methods to highly sophisticated convolutional neural networks
(CNNs). Support vector machine (SVM) of HOG features and multilayer
perceptron (MLP) artificial neural network (ANN) of HOG features are two
examples of HOG-based methods that show promise but fall short when
applied to real-world data [10]. Since our scenario takes place indoors, any of
these solutions would work. As a result, researchers are looking into deep
learning (DL) techniques to boost the generalization of different types of
input data during deployment and to offer more accurate representations.
Facial expression recognition is just one of many types of image recognition
where DL methods like CNN have significantly improved results. The use of
deep learning techniques yields more trustworthy and accurate results [11],
according to a study that compared numerous approaches.
SVMs are among the most reliable classifiers available when the best
possible kernel can be found. For facial emotion recognition, Ma Xiaoxi et al.
[12] investigated the efficacy of an SVM-based fusion method. The authors
have combined SVM with deep Boltzmann machine (DBM), a shallow model
that can make accurate predictions and classifications based on data generated
in a lab. Using the straightforward fusion approach proposed in [13], we
combine the two models mentioned above. The models were trained using
data from FERA 2015 (Second Facial Expression Recognition and Analysis
Challenge) and tested using data from the SEMAINE database containing
examples of emotional facial expressions. Although the SVM-DBM fusion
model could achieve an accuracy of 91percent, this would not apply to our
 11.3 Dataset Description 241

use case because it relied on static feature engineering. To solve computer

vision problems, CNNs have surpassed physics-based models and other
preprocessing techniques because they permit “end-to-end” learning directly
from input images [14]. As a result, CNNs are now the go-to method for
resolving computer vision issues. By analyzing the results of Breuer and
Kimmel’s [15] comprehensive study of CNN’s performance across different
FER datasets, we were able to zero in on the connection between the features
generated by an unsupervised learning process and the Ekman model of
labeling data. These facial expressions are among the most telling of inner
thoughts.
Facial expression recognition can benefit significantly from access to
the temporal context. However, developing helpful features manually is a
time-consuming process. For the specified problem statement, two-stage
convolutional neural network (CNN) models [16] were also considered as
a possible means of relief. Images are used to extract temporal appearance
features in the first stage. The second step involves extracting temporal
geometry features from facial landmark points in the temporal domain. These
two steps are necessary to address the issue at hand. They are combined using
joint fine-tuning to improve the performance of the facial expression recog-
nition system. The multiple-tier CNN model we proposed for incorporating
temporal data into the learning process is computationally intensive because
it necessitates fine-tuning the parameters of two separate CNNs. A hybrid
approach [17] was developed to better account for time differences. To do
so, this technique combines a convolutional neural network (CNN) for the
spatial features of each frame with a long short-term memory (LSTM) for
the temporal features of the following frames. This was done to expedite the
process of making the necessary adjustments.
RNN-CNN [18] models, which combine data via simple temporal aver-
aging, outperformed the baseline CNN model, which did not use RNNs. The
2015 dataset from the Emotion Recognition in the Wild Challenge was used
for this analysis. The validation accuracy of the RNN-CNN hybrid model
was 39 percent, which was higher than the accuracy of the CNN model
alone (30%). The 3D Inception-ResNet architecture, proposed by Hasani and
Mahoor [19], is followed by an LSTM unit. When coupled with the LSTM
unit, this architecture can analyze the temporal and spatial relationships
between faces in multiple video frames. It is worth noting that this network,
too, takes facial features as inputs. As a result, the parts of the face that are
more crucial to conveying emotion are highlighted, while the less crucial parts
are downplayed.
242 Facial Emotion Classification for Industry Automation using Convolutional

11.3 Dataset Description

According to emotion theory, emotions can be treated categorically, which
proposes only six basic, distinct, and universal emotions exist, or dimen-
sionally, where we consider emotions to be a mixture of multiple factors.
The categorical emotion model supports the proposed solution of using a
convolutional neural network, as they provide hard and fast rules to classify
emotions. The dimensional model of emotion is not widely used due to its
innate complexity and the number of variables involved in classifying a single
emotion.
Several datasets available can aid computer vision-based emotion recog-
nition from facial expressions, but very few can be used to develop a
generalized and robust model. The research analyzes and builds solutions
using the FER 2013 and JAFFE datasets are shown in Figs. 11.1 and 11.2.
Pierre-Luc Carrier and Aaron Courville develop the FER-2013 as part
of their ongoing project, and it is publicly available. This dataset used the
Ekman model to label the images into seven distinct emotions following the
discrete emotion theory. “Happy” has the highest frequency in the dataset.
The baseline for random guessing is 24.4 percent across the dataset.
JAFFE – The Japanese Female Facial Expression dataset – is an image
dataset that depicts facial expressions posed by Japanese women, accompa-
nied by semantic ratings on nouns describing the expressions. The JAFFE
dataset was developed by recording the facial expression of 10 different
Japanese female subjects. The dataset has seven classes of emotions, six basic
expressions, and neutral. Around 60 annotators labeled the images, and the
semantic rating is the average rating of all the annotators. The dataset utilizes
the Ekman model as a basis to label emotions.

11.4 Model Architecture

A succinct summary of the findings and an elucidation of this work’s advance-
ment or contribution to the field are included. MLP and SVM are machine
learning techniques popularly used for emotion recognition but did not yield
the desired results. Even with further research to improve feature extraction,
these methods were not viable in real-time scenarios.
Deep learning (DL) approaches learn different levels of abstraction for
input data representations, with higher representational levels providing more
relevant features for differentiation and classification. This was the main
 11.4 Model Architecture 243

driving force to look into using CNN for emotion recognition, as they gave
better accuracy and generalizability.

Figure 11.1 Samples from FER-2013 dataset [12].

Figure 11.2 Samples from JAFFE dataset [13].

244 Facial Emotion Classification for Industry Automation using Convolutional

Figure 11.3 Model visualization.

 11.5 Model Training 245

Neural networks are not preprogrammed to function for the required task;
they learn and adapt to do a specific task when they are being trained for
the required task. An input layer will get the data that has to be fed to
the CNN, whose dimension is determined by the dimensions of the input
data. A deep neural network is a neural network composed of several hidden
layers, where a few hidden layers will help the CNN to learn complex data
interactions. An output layer that will give the final result, for instance,
classification probabilities. The number of classes that the data is partitioned
into determines the size of this layer in the hierarchy. The typical structure of
a CNN consists of several layers, all of which are linked to one another. This
indicates that all of the neurons in one layer are connected to all the neurons
in the layer beneath it. We propose using a fully connected feedforward
convolutional neural network for this problem. This has been proposed as
a potential answer. This network has a 48×48×1 input layer and a 7×1
output layer. The letterpress denotes the model’s activation function, and the
max pooling technique is used to sample the input more densely along its
spatial dimensions. We prefer the “ReLu” activation function because it is
sparse, which helps when dealing with dense representations, and because its
gradient is less likely to vanish.
There are 1,328,167 parameters in the model, of which 1,325,991 can be
trained and 2176 cannot. ADAM is the optimizer employed, with categorical
cross-entropy as the loss function of choice. Dropout dilution, or DropCon-
nect, is used at a rate of 0.2 in this context. Through preventing intricate
co-adaptations on training data, this regularisation method is used to lessen
the likelihood of overfitting in artificial neural networks. The dilution that
occurs in a dropout layer is another name for it. This technique allows neural
networks to perform model averaging efficiently.

11.5 Model Training

The model was trained using the train set of the FER-2013 dataset. The input
data for training are grayscale images of 48×48 size, which are loaded as
a array. Table 11.1 shows the pseudo code of the model training. . The input
data’s label is one-hot encoded to represent categorical data as binary vectors,
as convolutional neural networks cannot work with data directly. The images
and one-hot–encoded labels are then split into train and validation set in the
ratio 80:20.
The model was supposed to be trained for 200 epochs but could be
stopped early if the validation loss did not go down. With a learning rate
246 Facial Emotion Classification for Industry Automation using Convolutional

of 0.001, if the validation loss stays the same for more than 12 iterations, the
learning rate goes up by 0.1. It was found that the model was 69.60 percent
right. The same setup was used to train on the JAFFE dataset, and the results
showed that it was 73.24 Which is shown in Fig. 11.4.

Figure 11.4 Model training and validation: (a) accuracy and (b) loss.

Table 11.1 Pseudocode for model training.

Pseudo code for training the model
1 Import libraries
2 Load the dataset
3 Convert images to grayscale
4 Resize the image to size (48×48)
5 Split into test and train data
6 Build the model
Initialize [Link]() object
• Add Conv2D, activation, MaxPooling2D layer, Batch normalization, flattened
and dense layers as shown in Figure 11.3
7 Compile the model
• Set the metrics as accuracy
• Initialize ADAM optimizer with a learning rate of 0.001
• Monitor losses
6 Set hyperparameter: Batchsize, learning rate, early stopping criteria
7 Fit the model
• If val_loss has no improvement over 12 epochs, reduce the learning rate by a
factor of 0.1.
• If val_accuracy has no improvement over 50 epochs, stop training.
8 Save the model weights : [Link](’weights.h5’)
 11.6 Model Metrics 247

11.6 Model Metrics

Analyzing a model’s performance from a qualitative and quantitative per-
spective is required, and model metrics help us understand the strength and
pitfalls of the model. Accuracy, precision, recall, and F1-score, to name a
few metrics, were computed for the proposed solution across both datasets.
Since the JAFFE dataset is of a smaller volume, there are high possibilities
to overfit or underfit the model, contributing to the high-high and low-low
values while metrics are computed. The proposed solution for the FER-2013
dataset achieved an F1-score in the range of .65–.8 over all the classes. The
proposed model achieved an accuracy of 69.90 percent and 73.24 percent for
the FER-2013 and the JAFFE datasets, respectively.
A table known as a confusion matrix can be used to provide a concise
summary of the performance of your prediction model. A confusion matrix
is a table where each cell reveals the percentage of times the model correctly
identified the classes and the number of times it misidentified them. This
matrix provides a graphical representation of true positive false negative
network (TPFNN) representing true positives, false positives, and false nega-
tives. Table 11.2 shows the different model metrics for different emotions.
• The rate at which a classifier successfully and accurately predicts a
positive class is called the true positive (TP) rate.
• A classifier is said to have a “true negative” (TN) value if it can correctly
predict a negative class by a certain percentage.
• “False positives,” also known as FPs, are instances in which the classifier
incorrectly predicts that the negative class belongs to the positive class.

Table 11.2 Model metrics.

Emotions Recall Precision F1-score
JAFFE FER- JAFFE FER- JAFFE FER-
2013 2013 2013
Anger 67% 60% 80% 64% 0.73 0.62
Disgust 60% 71% 67% 78% 0.63 0.74
Fear 100% 44% 56% 60% 0.71 0.51
Happy 86% 90% 92% 88% 0.89 0.89
Sad 64% 54% 64% 56% 0.64 0.55
Surprise 64% 79% 100% 78% 0.78 0.79
Neutral 88% 78% 58% 61% 0.7 0.68

Macro avg 74% 69% 75% 68% 0.73 0.68

Weighted avg 77% 69% 73% 70% .0.74 0.69
248 Facial Emotion Classification for Industry Automation using Convolutional

Figure 11.5 Confusion matrix.

• The frequency with which a classifier incorrectly identifies a positive

class as a negative class is measured by a particular type of prediction
error known as false negative (FN).
From the confusion matrix shown in Figure 11.5, we can observe the extent
of misclassification. It gives a one-shot view of the classification and misclas-
sification. This also helps us analyze the misclassification trends. We can see
that the class happy is adequately classified about 83 percent of the time, and
the classifier confuses the images labeled fear as sad for about 26 percent.
This indicates that the features that distinguish fear and sadness are much
correlated with each other.

11.7 Activation Maps

Activation maps are a qualitative method to observe the discriminative
regions in an image utilized by CNN to distinguish among classes which is
shown in Fig. 11.6.
 11.8 Implementation Workflow 249

Figure 11.6 Activation maps.

This helps us observe the features the CNN considers to classify the input
data. Activation maps illustrate that deep learning networks already have
some built-in attention mechanisms. We can observe from the activation maps
that the model also extracts facial features such as eyes and pursing of lips to
classify emotions. This is similar to how the Ekman model uses facial features
to classify emotions.

11.8 Implementation Workflow

The proposed solution can currently work with Google Meet, and the plan is
to extend to other platforms. The entire application is cloud-hosted, alleviat-
ing any burden on the user hardware for processing the customer’s emotions.
The image is extracted from the video call and queued in the MongoDB
database for further processing. The implementation workflow is repre-
sented in Figures 11.7 and 11.8. Multi-task cascaded convolutional network
(MTCNN) is used for localization and extraction of region of interest (ROI),
which is passed onto for preprocessing, whose output is to be fed into the
model to classify emotions. MTCNN has better non-maximum suppression
and consistency in localizing faces in an image than Haar cascade classifiers.
Simply, it helps us give good-quality input data to receive good-quality output
from the model.
The frames from the video feed of the live conferencing application such
as Google Meet are extracted via a screenshot API provided by Google
250 Facial Emotion Classification for Industry Automation using Convolutional

Figure 11.7 Implementation workflow.

Figure 11.8 MTCNN architecture [22]: P-NET, R-NET, and O-NET.

Chrome. This video frame is then sent to a MongoDB database, which stores
the image data for processing in the FER module. The data is stored in
Base64 format in the database and converted back to a NumPy array, where
each element represents the pixel intensity, to feed as an input to MTCNN.
MTCNN extracts the ROI, which is essentially the region in the picture where
the faces are present. The ROI is then further converted to the grayscale color
space and resized to 48×48 before being fed to the CNN model as the input.
The prediction of each emotion is displayed as probability scores, and the
class of emotion which achieves the highest probability score is displayed as
the module’s output. The entire processing from the start to end is done in the
cloud, using an AWS EC2 instance, alleviating the hardware dependency on
the user end when deployed in real time.
The novelty of the proposed solution rests in its simplicity. The proposed
solution can provide an instantaneous real-time response to the given input
with high accuracy. It does not utilize complex architecture like the mini-
Xception model but still significantly outperforms it. Using MTCNN in
tandem with the CNN to extract ROI helps generate the ROI with high
confidence, which improves the quality of input to the CNN model. The
 References 251

model performs well on a single dataset and retains the same performance
across the JAFFE dataset. When talking about the real-world implementation
of the solution, it is capable of handling real-world data and producing outputs
instantaneously and is optimized for both cloud and edge-based deployment
as it requires very minimal processing power to deploy a simple CNN as the
one proposed in the paper.

11.9 Conclusion
The proposed solution aids in gaining insight into the customer’s emotional
response when interacting over a video call. It utilizes a convolutional neu-
ral network to classify emotions and show the emotions exhibited by the
customer. The model achieved an accuracy of 69.90 percent while using
the FER-2013 dataset and provides good scalability when in the future;
the model is fitted to accommodate new data. This model outperforms the
mini-Xception model, which achieved only 66 percent accuracy [11] on the
FER-2013 dataset. With the growing online presence of shops and significant
business meetings happening over video calls, the proposed system has the
potential to become a ubiquitous part of companies’ sales and CRM tools.
Currently, the model can classify macro emotions only, in extension in the
future; this can be extended to classify micro emotions too. If possible, we
can generate a transcript of the meeting conversation and tag the emotions
observed over each time stamp.

References
[1] Maja Pantic, Nicu Sebe, Jeffrey F. Cohn, and Thomas Huang. 2005.
Affective multimodal human-computer interaction. In Proceedings of
the 13th annual ACM international conference on Multimedia (MUL-
TIMEDIA ’05). Association for Computing Machinery, New York, NY,
USA, 669–676.
[2] Picard, R. 1997. Affective computing. Boston: The MIT Press.
[3] Arumugam, Devi, Kavya Govindaraju, and Ananth Kumar Tamilarasan.
“AIIoT-Based Smart Framework for Screening Specific Learning Dis-
abilities.” In Machine Learning for Critical Internet of Medical Things:
Applications and Use Cases, pp. 103-124. Cham: Springer International
Publishing, 2022.
252 Facial Emotion Classification for Industry Automation using Convolutional

[4] Vinciarelli, Alessandro, and Maja Pantic. “Hervé Bourlard, 2009.‘.”

Social Signal Processing: Survey of an Emerging Domain’in Image and
Vision Computing 27: 1743-1759.
[5] Shoumy, Nusrat J., Li-Minn Ang, Kah Phooi Seng, DM Motiur
Rahaman, and Tanveer Zia. “Multimodal big data affective analytics:
A comprehensive survey using text, audio, visual and physiological
signals.” Journal of Network and Computer Applications 149 (2020):
102447.
[6] Joseph, Abin John, Nidhin Sani, K. Suresh Kumar, T. Ananth Kumar,
and R. Nishanth. “Towards a Novel and Efficient Public Key Manage-
ment for Peer-Peer Security in Wireless Ad-Hoc/Sensor Networks.” In
2022 International Conference on Smart Technologies and Systems for
Next Generation Computing (ICSTSN), pp. 1-4. IEEE, 2022.
[7] Marechal, C. et al. (2019). Survey on AI-Based Multimodal Methods for
Emotion Detection. In: Kołodziej, J., González-Vélez, H. (eds) High-
Performance Modelling and Simulation for Big Data Applications.
Lecture Notes in Computer Science(), vol 11400. 307-324
[8] Tracy, J. L., & Randles, D. 2011. Four Models of Basic Emotions: A
Review of Ekman and Cordaro, Izard, Levenson, and Panksepp and
Watt. Emotion Review, 3, 397 - 405.
[9] Colombetti, Giovanna. “From affect programs to discrete dynamical
emotions.” Philosophical Psychology 22, no. 4 (2009): 407-425.
[10] C. Mayer, M. Eggers, and B. Radig, 2014. Cross-database evaluation for
facial expression recognition, Pattern recognition, and image analysis,
vol. 24, no. 1, 124–132.
[11] Kumar, Tamilarasan Ananth, Rajendrane Rajmohan, Muthu Pavithra,
Sunday Adeola Ajagbe, Rania Hodhod, and Tarek Gaber. “Automatic
face mask detection system in public transportation in smart cities using
IoT and deep learning.” Electronics 11, no. 6 (2022): 904.
[12] Xiaoxi, M., Weisi, L., Dongyan, H., Minghui, D., & Li, H. (2017). Facial
emotion recognition. IEEE 2nd International Conference on Signal and
Image Processing (ICSIP). doi:10.1109/siprocess.2017.8124509
[13] Pugazhendiran, P., Suresh Kumar, K., Ananth Kumar, T., & Sundaresan,
S. (2022). An Advanced Revealing and Classification System for Plant
Illnesses Using Unsupervised Bayesian-based SVM Classifier and Mod-
ified HOG-ROI Algorithm. In Contemporary Issues in Communication,
Cloud and Big Data Analytics (pp. 259-269). Springer, Singapore.
[14] Walecki, R., Rudovic, O., Pavlovic, V. Schuller, B. Pantic, M (2017).
Deep structured learning for facial action unit intensity estimation. In
 References 253

Proceedings of the IEEE Conference on Computer Vision and Pattern

Recognition, Honolulu, HI, USA; pp. 3405–3414.
[15] Breuer, R.; Kimmel, R. A deep learning perspective on the origin of
facial expressions. (2017)arXiv.
[16] Jung, H.; Lee, S.; Yim, J.; Park, S.; Kim, J. (2015) Joint fine-tuning in
deep neural networks for facial expression recognition. In Proceedings
of the IEEE International Conference on Computer Vision, Santiago,
Chile; pp. 2983–2991
[17] Ko, Byoung Chul. 2018. “A Brief Review of Facial Emotion Recog-
nition Based on Visual Information” Sensors 18, no. 2: 401. https:
//[Link]/10.3390/s18020401
[18] Kahou, S. E., Michalski, V. , Konda, K.,(2015) Recurrent neural net-
works for emotion recognition in video. In Proceedings of the ACM on
International Conference on Multimodal Interaction, Seattle, WA, USA,
pp. 467–474.
[19] Hasani, B.; Mahoor, M. H. Facial expression recognition using enhanced
deep 3D convolutional neural networks. In Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition Workshops,
Hawaii, HI, USA, 21–26 July 2017; pp. 1–11
[20] Goodfellow, I. J., Erhan, D., Carrier, P. L., Courville, A., Mirza, M.,
Hamner, B., Cukierski, W., Tang, Y., Thaler, D., Lee, D. H. and Zhou,
Y., 2013, November. Challenges in representation learning: A report on
three machine learning contests. In International conference on neural
information processing, Springer, Berlin, Heidelberg 117-124.
[21] Lyons, Michael, Kamachi, Miyuki, & Gyoba, Jiro. (1998). The Japanese
Female Facial Expression (JAFFE) Dataset
[22] Kaipeng Zhang and Zhanpeng Zhang and Zhifeng Li and Yu Qiao,
(2016), Joint Face Detection and Alignment Using Multi-task Cas-
caded Convolutional Networks, IEEE Signal Processing Letters, 23,
(1499-1503)
 12
Automatic Hand Sanitizer Dispenser
by Industrial Automation using Arduino
and Photodiode

R. Kiruthika1 , T. Prabhu2 , and Naveenbalaji Gowthaman3

1 Electronics and Communication Engineering,
SNS College of Technology, India
2 Electronics and Communication Engineering, Presidency University, India
3 Electronic Engineering, University of KwaZulu-Natal, South Africa

E-mail: [Link].2020@[Link]; prabhucbe1206@[Link];

[Link]@[Link]

Abstract
The 21st century is evolving with the emergence of new diseases. In this
pandemic Covid, we must protect ourselves from the harmful germs that
mainly enter the human body through our skin, especially our hands. To
be able to maintain hand hygiene, hand sanitizers provide an efficient way.
The main problem with those hand sanitizer bottled containers is that they
provide the chance to contaminate germs on their contact point. An effi-
cient and elegant solution to this is given by the industrial automation of
automatic hand sanitizer dispenser machines that are processed based on
sensors, and it is wholly touchless such that infections and the spread of
contagious diseases can be avoided. The problem with these automatic hand
sanitizer dispensers is that the sensors fail in performance due to changes
in atmospheric temperature, sunlight, humidity, and setup conditions. Most
of the available brands of automatic hand sanitizer dispensers make use
of ultrasonic and infrared sensors that are replaced by photodiode in our
proposed design. The proposed design is based upon an Arduino circuit in

255
256 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

which a signal from the microcontroller controls the reception of the signal
from the user, also giving output by activating the motor. The photodiode
proves to be worthy of selection by quickly detecting the user in a period
that is very much less compared to that of the other sensors. The designed
circuit was automated more than 25 times with varying parameters, and the
results were discussed. The proposed design proved its worthiness in several
experimental conductions. Designing an efficient system that meets all our
needs and vision is essential, and we need to consider all the parameters that
make necessary alterations in the existing system to help it function at its best.
The main motive or aim of the designed performance is to empathize with the
customers’ needs. Hence this proposed system was found to be user-friendly
and safe to use.

Keywords: pandemic, hand hygiene, automatic, automation, photodiode,

Arduino, faster

12.1 Introduction
In December 2019, the pandemic outbreak was first identified in Wuhan,
China. All the developing as well as the developed countries, were worst
affected by this pandemic. India, which has the second largest population in
the world, is suffering severely from COVID-19, and a substantial negative
impact was observed on the economic growth of the country [1, 2]. The
spread of COVID from an infected person to an average person was recorded
mainly through the air molecules inhaled by them, and a considerable amount
of viruses was found to be transmitted through the hands. Hence, hand
hygiene behaviors were expected to be strictly followed in hospitals and
public areas to prevent the pandemic from affecting the maximum number of
people [3, 4]. Hand sanitizers were used to wash the hands frequently to kill
the viruses on the hands, and they were also used to sanitize the surfaces sus-
pected to be contaminated by germs [5]. Alcohol-based hand sanitizers were
effective in inactivating more than 99 percent of the contaminated viruses
when asked to rub both palms with the ABHS (alcohol-based hand sanitizer)
[6]. There has been much work done by researchers from many fields to figure
out how to make it easier to manufacture automatic hand sanitizer dispensers
using state-of-the-art technologies. Automatic hand sanitizer dispensers have
been developed and are currently available for purchase. Some notable char-
acteristics of these dispensers include their ability to be wall-mounted and
their incorporation of automatic sprayers. However, certain advancements
 12.2 Background 257

have been made in the automation industry to ensure that users have access
to high-quality products and are satisfied with those products. Automatic
hand sanitizer dispensers [7–10] are triggered by an infrared obstacle detector
and employ a simple monostable multivibrator. Some of the automatic hand
sanitizer dispensers contained not only an automated dispensing mechanism
but also temperature measurement, which initially used rechargeable batteries
that were charged by solar panels and used two Arduino UNO boards with
voltage regulators [11]. Several automatic hand sanitizer dispensers with
sensor-imposed systems were developed and implemented for daily use in
the market. The primary sensors used for this development were ultrasonic
[13–17], but they did not match the expected results since their action was
affected by the interference of temperature and environmental factors.
A deep study about the disadvantages faced by the existing system in
industrial automation was made, and before producing an automatic hand
sanitizer dispenser that meets all the consumer needs [21, 22], the sensor that
is used to detect human hands [24] was changed in our study in a way that
yields the best outcome. A photodiode was used to detect human hands faster,
and the designed prototype also contains some of the added advantages, such
as dispensing only once for the first detection and waiting for a few seconds
for the subsequent detections to happen so that crowds can also be avoided.
This system also proves to be cost effective and time efficient. The result
of the comparative analysis of the existing systems with the proposed one
proved successful.

12.2 Background
Arnes Sembiring et al. developed an automatic hand sanitizer dispenser
using ultrasonic sensors and a mini DC water pump in which ultrasonic
sensors prevented the people from directly touching the sanitizer container
[23]. Although dispensers with an ultrasonic sensor proved efficient, they
sometimes failed to provide a suitable industrially automated device due
to damage, and their predetermined distance (3–5 cm) wasted the sanitizer
liquid because of poor sensing and improper recognition of the user [25–
28]. Automatic hand wash dispensers with battery-imposed systems and IR
sensors for sensing were developed by MM Srihari. This contains an RC
timer delay setup and a pipe to control the liquid flow, yet the consumer was
inconvenient because of the additional expenses [29]. Since the impact, scien-
tists and engineers have developed automatic hand sanitizers with advanced
258 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

technological aspects. Innovative research has been done in this field, and
effective outcomes have been recognized [30–34].

12.2.1 The goal

The prime aim of the proposed structure of automatic hand sanitizer dis-
pensers is mainly to provide enhanced performance at an affordable cost.
With more minor modifications that can bring a more significant advantage,
the proposed work system is designed to meet the customer’s needs and
improve product quality. With the advent and the development of semicon-
ductor technology, the efficiency and the time of operation of hand sanitizer
dispensers have been widely improved

12.2.2 The extent of the sanitizer dispenser with photodiode

Every conventional system in the market uses either an ultrasonic sensor or an
infrared sensor to detect the presence of human hands to dispense the sanitizer
liquid. Nevertheless, no one can deny that the photodiode sensors that are the
best results of developing semiconductor technologies are one of the perfect
sensors for humans. They are capable of detecting the infrared radiations
emitted by humans. All black bodies emit invisible IR radiation to human
eyes, and the photodiodes can detect that radiation. The novel change of using
photodiodes instead of ultrasonic sensors or infrared sensors can make the
proposed work system stand as the best among other systems available in
the market. The future-based novel design of the automatic hand sanitizer
dispenser using Arduino and photodiode can be used as a hygienic way of
cleaning the hands using the sanitizer liquid in schools, public areas where
the crowd of people is above the desired rate, hospitals, public and private
offices, for the usage of clinical persons and physicians, etc. This can also be
used in place of the hand-pressed sanitizer machines as a means of improved
health and hand hygiene.

12.3 Design of Extensive Hand Sanitizer

The automatic hand sanitizer dispenser circuit was designed to reduce the
spread of COVID-19 and safeguard people from the pandemic. Various
components and materials were selected with specific goals on the bucket list.
The design was made such that it should be time-efficient, faster in action, and
cost-effective. These prime aims were achieved and experimentally proved by
simulating the designed circuit several times for precise action.
 12.3 Design of Extensive Hand Sanitizer 259

12.3.1 Requirements of the design

The Arduino UNO board is chosen as the main microcontroller, which
controls the working of the whole circuit. The heart and soul of the embedded
systems and the automated circuits are the microcontrollers that can function
without needing additional external peripherals. The Arduino UNO consists
of both analog and digital pins. This can take inputs from the analog pins for
sensors and convert the analog voltage value with the help of an analog to
digital converter. The Arduino board takes in the input from the photodiode
sensor and gives its output through the LED and DC motor. The LED is used
in the circuit to indicate the turning ON of the DC motor. A 12 V power
supply is given as input to the DC motor. An NPN transistor is connected
to the DC motor through a 1 KΩ resistor that amplifies the DC motor’s
power.
All the connections for the circuit of the automatic hand sanitizer dis-
penser have been made to the available components in the simulation software
called Tinkercad, ensuring good internet connectivity. The tinkercad circuits
can be stored and tinkered with or simulated for n number of times, and proper
calculations can be made. The potentiometer, which is the variable resistance,
contains three pins, out of which one is connected to the ground, the second
is connected to the analog input pin of Arduino, and the other is the Vcc pin.

12.3.2 Development of the hand sanitizer dispenser framework

The photodiode inputs are given into the Arduino board’s analog input pins
A0 and A1, as shown in Figure 12.1, which converts the analog values into
integer values. A red LED connected to the power pin through a resistor
glows as an indication of when the system is ready to take up the sensor
input. Another white LED glows when the photodiode gives the input to the
microcontroller, and the DC motor turns on for a specific period specified by
the resistance value in the potentiometer that can be tuned according to our
needs.

12.3.3 Default pins and conditions associated with the sensor

Simply adjusting the potentiometer’s knob allows the user to change the
device’s resistance, which is also known as the rotary potentiometer or pot. It
typically has three pins: a GND pin connected to the GND potential, a Vcc
pin connected to Vcc (5 V or 3.3 V), and an output pin connected to the
analog input pin on the Arduino. The voltage is discharged through this pin.
260 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

The method that is used to calculate the voltage output that is provided by the
potentiometer for the various turns of the knob is as follows:
angle
OutputVoltage = ()cc . (12.1)
anglemax
Arduino’s analog inputs are pins A0 through A5, as indicated by their
names. The analog input pin rounds the voltage up or down to an integer
value (between 0 V and Vcc). The ADC value, also known as the analog
value, is a positive integer between 0 and 1023. The Arduino’s analog input
can take on values between 0 and 1023, and the on/off delay can be adjusted
from zero milliseconds to tens of seconds.
When the potentiometer is set at its maximum angle, it is expected to
turn on the device for the maximum ADC value in the code and then turn
it off automatically. This implies that if the potentiometer is at its maximum
resistance by tuning it to the maximum angle, the output voltage will be equal
to Vcc (i-e) 5 V, and the expected time is 10 s. Practically, results very close
to the theoretical values are obtained by simulating the designed circuit with
various potentiometer values.

12.3.4 Adjusting procedure of the potentiometer

One more additional work in the designed system is that by changing the
value of resistance in the potentiometer, the amount of sanitizer liquid needed
to be dispensed out can be customized according to our needs and places
where we make use of the automatic hand sanitizer dispensers for maintaining
good hand hygiene.
The potentiometer chosen here is of 60k resistance at its maximum range.
By tuning it to different angles, different output voltages corresponding to the
tuned angle are obtained and calculated as follows:
When the knob is set at 0◦ angle, which is the minimum angle and
contains 0k, the expected output voltage is also zero.
angle 0◦
()cc . (12.2)
anglemax 270◦
As expected, when the knob was at 0 resistance, the pot gave no output
voltage, and no sanitizer was dispensed.
This value can be seen clearly from the serial monitor when the knob of
the potentiometer is kept at 0 resistance.
 12.4 Theoretical and Mathematical Proofs 261

12.4 Theoretical and Mathematical Proofs

For a more detailed study and more substantial proof, the theoretical and the
simulated values for the half-tuning of the potentiometer were studied. When
the potentiometer is tuned to half of its maximum angle, it gives out half of
its Vcc, giving out 2.5 V. Hence the device is expected to dispense for half of
its maximum period i.e., for 5 seconds. This is proved as follows:
angle halfoftheanglemax 135◦ Vcc
()cc ∗ 5V= ◦
∗ 5V=2.5V . (12.3)
anglemax anglemax 270 2
Hence, the simulated result, which was observed in the serial monitor,
shows a very close observation of the theoretical values. The maximum
time for the tuning on of the device was set to be 10 seconds. When
the angle is half in the potentiometer, it gives half of its maximum resis-
tance, 60 KΩ, making the device turn on for 5 seconds and then turn off
automatically.
Similar proofs and calculations were observed for tuning of the poten-
tiometer to its maximum angle (anglemax ). When the knob of the potentiome-
ter is tuned to 270◦ , it is at its maximum resistance i.e., 60 KΩ. Hence it gives
out the output voltage, which is equal to Vcc. It makes the on-off time reach
its maximum value, and the device is on for 10 seconds.
angle halfoftheanglemax 270◦
()cc ∗ 5V= ∗ 5V=5V Vcc . (12.4)
anglemax anglemax 270◦
The potentiometer’s ratio to control the sanitizer’s volume to the on and
off time should be 1000:10 at its maximum range. The simulated result
was very close and precise to the expected values, i.e., the serial monitor
shows 1000:9, which is much closer to the expected values. The closer value
of the experimental result to the theoretical value suggests that the system
works with incremental steps or procedures adapted to make the poten-
tiometer function with different values of resistances at different working
conditions.
The wastage of hand sanitizer or unwanted dispensing can be avoided
with this action introduced in our proposed system. The amount of sanitizer
that has to be dispensed out can also be decided by the customer according to
their needs. A prescribed amount of hand sanitizer liquid is to be dispensed
out, which varies with the age and conditions. Children need only a tiny
amount of hand sanitizer, while doctors and clinical persons may require
a large amount. By varying the resistance value in the potentiometer, the
262 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

amount of hand sanitizer to be dispensed is varied, which accounts for an

enhanced adaptation of the system. So, the proposed system of automatic
hand sanitizer dispensers proved to be worthwhile in all situations. Adjusting
the probe of the potentiometer can make desirable changes in the circuit, and
even rural people can quickly learn how it works and customize the machines
according to their needs and requirements.

12.5 Schematic Layout of the Proposed Work

Figure 12.1 Schematic layout of the automatic hand sanitizer dispenser.

12.6 Examination of the Proposed Work

Every design of an engineer done in the lab with the help of tools and software
has to be checked for performance and efficiency. The automatic sanitizer
dispenser designed with the simulation software called Tinkercad must be
 12.6 Examination of the Proposed Work 263

checked and analyzed for its performance. Analyzing and experimenting

with the proposed design system with different conditions and frequently
changing specific parameters can be used to produce the required result of the
experimental study. The main motive of the design was sensing the process
that happens with photodiodes. Since the findings of the proposed system’s
operation are less distorted or impacted by the circumstances in the virtual
environment, it is preferable to conduct precise experimental investigations in
the virtual environment to produce more accurate results. There are also addi-
tional perks and advantages with the experiments that are done in the virtual
environment using simulation software, like the presence of serial monitors
that can be used to show the exact experimental readings. These readings can
be used to compare the experimental values with the theoretically available
values so that we can conclude that our system is working at its best. Usually,
the conventional methods of checking a proposed work can be experimenting
with all the designed systems after fabrication in real-time environments,
calculating the recorded values, or taking the readings manually. However,
the disadvantage of this system is that some of the fabricated machines can
become waste when automated due to the faults caused by the conditions that
affect their performance, so the designer may conclude that his design is a
failure. However, there is no such chance with the experiments done through
simulations, and they give the best results. Hence, the designed or proposed
automatic hand sanitizer dispenser system can be tested, and the results are
tabulated.
A comparative analysis is needed to prove that our system is the best com-
pared to the existing ones in the industry. By comparing the results obtained
from the experiment using software with the theoretically available readings,
we stand strong and firm in proving that our idea is the best in comparison to
the other machines available in the market. The working of the automatic
hand sanitizer dispenser was analyzed by simulating the designed circuit
using simulation software with different experimental conditions. On the
whole, to showcase the working of the proposed system in varying conditions,
the simulation was made at certain conditions of the potentiometer, and also,
the simulation was done by varying the distance of the user from the sensor.
The rpm of the DC motor was increased only when the distance between the
sensor and the user was found to be higher than 78 mm. Hence the sanitizer
liquid will not be dispensed if the distance of the human hand is less than
the prescribed distance from the sanitizer machine. Additionally this could
enhance the functionality of the automatic hand sanitizer dispenser and stop
the sanitizer liquid from being wasted due to false sensing. Table 12.1 depicts
264 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

the results of the comparative analysis made with the available machines
in the market and the proposed system. Comparing various parameters of
top brands of hand sanitizer dispensers available in the market and their
performance with that of the proposed system shows that the proposed system
is working at its best to satisfy all desirable conditions. With simple changes
in the needed arena of the system, the best results can be obtained to meet the
customer’s needs and improve satisfaction.
Once the result values of the proposed system of automatic hand sanitizers
are closer to the expected value, they can be fabricated and released into
the market for use by the public. The designed circuit of the automatic hand
sanitizer dispenser is so easy and compact, occupies significantly less space,
and is even cost-effective to buy.

12.7 A Comparative Study with the Market Products

Table 12.1 A comparative study of the parameters of the designed prototype with the
available dispensers in the market.
[Link] Name of Price Time Distance Sensor Input Battery
the sanitizer in | consumption of used power
machines (sensor sensing
dispensing
time)
1. NETBOON 4491 1 second 12 Ultra 90– 1.5V
automatic inches Sonic 270 V AA
hand AC/50
sanitizer Hz
dispenser
2. KENT 2490 3.25 10 cm IR 4.5 V 4C
touchless seconds sensor (3 no’s) type
automatic (1.5 W) bat-
sanitizer teries,
dispenser 1.5V.
Bt
3. SVAVO 8316 0.38 2–12 IR Power 4 AA
automatic seconds cm sensor adaptor batteries
hand
sanitizer
dispenser
 12.8 Drawbacks of the Conventional System of Dispensers 265

Table 12.1 Continued.

[Link] Name of Price Time Distance Sensor Input Battery
the sanitizer in | consumption of used power
machines (sensor sensing
dispensing
time)
4. Hi-genic 1899 0.93 3–5 cm IR 110 V 6V/1A
automatic seconds sensor power
bulge adaptor
sanitizer
dispenser
5. BELTER 3199 Varies with 2.4–4.5 Motion 110– 4C
automatic conditions inches sensors 240 V type
hand DC 6V batter-
sanitizer adaptor ies
dispenser
6. Proposed 900 0.32 78 mm Photod 12 V Not
design of milliseconds iode power needed
automatic adap-
hand tor/ 12
sanitizer V from
dispenser power
using gener-
Arduino ator

12.8 Drawbacks of the Conventional System of Dispensers

The result of the literature review proved the worth of the proposed system.
Most of the available dispensers in industrial automation were known to func-
tion with an ultrasonic or infrared sensor. The sensing time, or the time the
sensors take, varies with the changing environmental setup conditions. The
sensor action varies, and the efficiency of the sanitizer and machines decrease
following the changes. Most of these devices in the market cost more than
our proposed device, so our proposed system proves to be cost-efficient. The
time taken by the photodiode to sense the user is just around 0.32 milliseconds
which is much faster compared to that of other machines whose ultrasonic and
infrared sensors are equipped with automatic hand sanitizers like NETBOON,
HI-GENIC, and all others as compared in Table 12.1. Most of such identified
devices work with AA batteries which have a faster discharging rate. Due to
this faster discharging rate, the batteries need to be changed frequently, and
they account for additional costs, including the maintenance charge. Another
disadvantage of battery replacement is that rural people may need the help of
educated ones as they might be unaware of the steps to follow during battery
266 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

replacement. Since, this kind of hand sanitizer dispensers are available in the
market, replacing conventional machines is the need of the hour.

12.9 Performance Comparison of Automatic Hand

Sanitizer Dispensers
Batteries are required for the operation of the vast majority of self-service
machines. Consequently, they need to have their batteries charged consis-
tently and on time as part of their maintenance. If the batteries run out of
juice, the consumer must make additional financial and time investments to
replace them. Despite this, many Indian manufacturers of hand sanitizer dis-
pensers have begun incorporating automated features in their products since
the beginning of the pandemic. The customer’s overall shopping experience
should become more streamlined as a primary focus. Because contemporary
hand sanitizer dispensers are powered by electricity, you will never need to
worry about them running out of liquid. Because of this, the batteries in
these sanitizer machines need to be replaced consistently, which can be an
inconvenience and may increase the risk associated with device management.

12.10 The Desirable Properties of the Proposed Idea

The proposed system uses no battery and works wholly based on a 12 V
power supply with a current of just 5 A. Hence, we can also use a 12 V supply
from the power generator instead of preferring batteries that need additional
expenses once discharged. The main microcontroller of the proposed system
is the Arduino UNO. The maximum output capacity of the Arduino is 5
V. In addition to the output produced by the Arduino, we can make the
system work with sufficient power by adding a simple 9 V battery. Even
the amplification part in the proposed design that contains an NPN transistor
with the resistor can contribute to power usage by amplifying it. Hence, no
additional requirements are necessary here for power supply, and it makes the
system run more efficiently than other machines in the market.

12.11 Additional Perks of the Industrial Automated System

Others disagree with the viewpoint held by some individuals who maintain
that automatic dispensers always require human supervision. Because some
areas become congested due to the automatic hand sanitizer dispenser, it is
essential to maintain cleanliness in those areas. This is another reason why
 12.12 Conclusion 267

this place is so filthy and unclean. On the other hand, electric dispensers that
spray a fine mist rarely require maintenance or repair. They are also very
proud of how clean they are because this is the case. Even though you have to
manually refill the hand sanitizer dispenser, doing so does not require much
time or effort. There is also a time delay of 1000 milliseconds for the sanitizer
machine to start for the next sensing as soon as it has dispensed the required
amount of sanitizer for the previous sensing. This also ensures additional
safety and hygiene in terms of avoiding crowding of people. The setting made
in the software part of the microcontroller of the proposed system to make
the sanitizer dispenser dispense the sanitizer liquid only for the determined
distance is yet another added advantage of the system. It is determined and
coded in the Arduino coding such that the sanitizer will be dispensed only
when the distance of the human hand is more than 78 cm. This will add credit
to your work because it prevents dispensing the hand sanitizer liquid for any
fake sensing of the human hand when they are not sensed in the actual time.

12.12 Conclusion
The proposed system of automatic hand sanitizer dispensers will prove their
efficiency in performance because of the novel change made in the existing
system of dispensers. Replacing the batteries in the available machines in
the market was difficult to replace and maintain the entire system. Hence the
proposed system is designed in such a way as to overcome the disadvantage
of many in which battery replacement is an essential one. The disadvantage
can be overcome by using the output from the microcontroller and adding a
battery of 9 V as an extra power supply. In the future, to expand this work,
the addition of an amplifier stage can be used to amplify the power from
the Arduino output. The amplifier circuit consists of the NPN transistor with
a tank circuit that amplifies the voltage divider bias input. These are more
straightforward and cost-effective machines that can be used for hand hygiene
in various places, as discussed already, including homes, hospitals, and public
places where the crowd of people is even slightly above the normal range. As
the COVID precautions are increasing daily with the advent of new diseases
like monkey fever, advanced COVID attacks, etc., in addition to the safety
precautions declared by the health ministry, maintaining good hand hygiene
is also very important. The proposed system of automatic hand sanitizer
dispensers using Arduino and photodiode was expected to be an excellent
way of preventing the spread of viruses through industrial automation. The
research for adding an ESP8266 WiFi module and a temperature-sensing
268 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

system is on hand. With the addition of the WiFi module, the amount of
hand sanitizer dispensed out for each detector in the human hand and the
temperature sensed by the temperature sensors can be recorded. In case of
want of any temperature data regarding the person affected with COVID-19
or that of a person with the symptoms of the disease, the proposed system,
with its extended additions of temperature sensors and WiFi modules, can
help the health ministry a lot. These recorded data can also be used to
further analyze persons with high temperatures and will also help check out
frequently if the system is working with total efficiency. The recorded data by
the microcontroller can be shared and processed from anywhere in the world
through the WiFi module using the Internet of Things concept. The Internet of
Things has an efficient way of adding performance efficiency to the systems
by collecting data, storing data which is followed by the processing of data to
make necessary modiïňAcations
˛ in the system or to process the stored data.

References
[1] Ghosh, A., Nundy, S., & Mallick, T. K. (2020). How India is dealing
with the COVID-19 pandemic. Sensors
[2] Moore, L. D., Robbins, G., Quinn, J., & Arbogast, J. W. (2021). The
impact of the COVID-19 pandemic on hand hygiene performance in
hospitals. American Journal of infection control, 49(1), 30-33.
[3] Huang, F., Armando, M., Dufau, S., Florea, O., Brouqui, P., & Boud-
jema, S. (2021). COVID-19 outbreak and healthcare worker behavioral
change toward hand hygiene practices. Journal of Hospital Infection,
111, 27-34.
[4] Ciotti, M., Ciccozzi, M., Terrinoni, A., Jiang, W. C., Wang, C. B., &
Bernardini, S. (2020). The COVID-19 pandemic. Critical reviews in
clinical laboratory sciences, 57(6), 365-388.
[5] Tamimi, A. H., Carlino, S., Edmonds, S., & Gerba, C. P. (2014). Impact
of an alcohol-based hand sanitizer intervention on the spread of viruses
in homes. Food and environmental virology, 6(2), 140-144.
[6] Santhosh, R., & Mahalakshmi, R. (2021, September). Low-cost auto-
matic hand sanitizer dispenser for the Covid-19 pandemic period. In
2021 Third International Conference on Inventive Research in Comput-
ing Applications (ICIRCA) (pp. 534-538). IEEE.
[7] Das, A., Barua, A., Mohimin, M., Abedin, J., Khandaker, M. U., &
Al-Mugren, K. S. (2021, April). Development of a Novel Design and
 References 269

Subsequent Fabrication of an Automated Touchless Hand Sanitizer Dis-

penser to Reduce the Spread of Contagious Diseases. In Healthcare (Vol.
9, No. 4, p. 445). Multidisciplinary Digital Publishing Institute.
[8] Suryawanshi, V. R., Surani, H. C., & Yadav, H. R. (2021). Sensor-based
automatic hand sanitizer dispenser. Medical Journal of Dr. DY Patil
Vidyapeeth, 14(5), 543.
[9] Eddy, Y., Mohammed, M. N., Daoodd, I. I., Bahrain, S. H. K., Al-
Zubaidi, S., Al-Sanjary, O. I., & Sairah, A. K. (2020). 2019 Novel Coro-
navirus Disease (Covid-19): Smart contactless hand sanitizer-dispensing
system using IoT based robotics technology. Revista Argentina de
Clínica Psicológica, 29(5), 215.
[10] Zainudin, Z. I. B., San, L. Y., & Abdulla, R. (2022). Smart hand sani-
tizer dispenser. Journal of Applied Technology and Innovation (e-ISSN:
2600-7304), 6(1), 10.
[11] Ciotti, M., Ciccozzi, M., Terrinoni, A., Jiang, W. C., Wang, C. B., &
Bernardini, S. (2020). The COVID-19 pandemic. Critical reviews in
clinical laboratory sciences, 57(6), 365-388.
[12] Boudjema, S., Dufour, J. C., Aladro, A. S., Desquerres, I., & Brouqui, P.
(2014). MediHandTrace: a tool for measuring and understanding hand
hygiene adherence. Clinical Microbiology and Infection, 20(1), 22-28.
[13] Janga, P., Khan, M. S., Suneel, L., & Reddy, M. A. (2021, July).
Design and implementation of automatic hand sanitization technique
using Arduino and ultrasonic sensor. In AIP Conference Proceedings
(Vol. 2358, No. 1, p. 050029). AIP Publishing LLC.
[14] Kumar, K. S., Mani, A. R., Sundaresan, S., Kumar, T. A., & Robinson,
Y. H. (2021). Blockchain-based energy-efficient smart green city in IoT
environments. In Blockchain for Smart Cities (pp. 81-103). Elsevier.
[15] Cure, L., & Van Enk, R. (2015). Effect of hand sanitizer location on
hand hygiene compliance. American Journal of infection control, 43(9),
917-921.
[16] Sitorus, A., syahri Cebro, I., & Bulan, R. (2021). Experimental push
and pull force data utilizing self-developed automatic liquid dispensers.
Data in Brief, 38, 107308.
[17] Bördlein, C. (2020). Promoting hand sanitizer use in a University
Cafeteria. Behavior and Social Issues, 1-9.
[18] Anderson, J. L., Warren, C. A., Perez, E., Louis, R. I., Phillips, S.,
Wheeler, J., . . . & Misra, R. (2008). Gender and ethnic differences in
hand hygiene practices among college students. American Journal of
infection control, 36(5), 361-368.
270 Automatic Hand Sanitizer Dispenser by Industrial Automation using Arduino

[19] Iqbal, M. Z., & Campbell, A. G. (2021). From luxury to necessity:

Progress of touchless interaction technology. Technology in Society, 67,
101796.
[20] Ward, M. A., Schweizer, M. L., Polgreen, P. M., Gupta, K., Reisinger, H.
S., & Perencevich, E. N. (2014). Automated and electronically assisted
hand hygiene monitoring systems: a systematic review. American Jour-
nal of infection control, 42(5), 472-478.
[21] Meng, M., Sorber, M., Herzog, A., Igel, C., & Kugler, C. (2019).
Technological innovations in infection control: A rapid review of the
acceptance of behavior monitoring systems and their contribution to
improving hand hygiene. American Journal of infection control, 47(4),
439-447.
[22] Zeng, Z., Chen, P. J., & Lew, A. A. (2020). From high-touch to high-
tech: COVID-19 drives robotics adoption. Tourism Geographies, 22(3),
724-734.
[23] Syahputra, M. I., Khair, U., & Sembiring, A. (2021). Automatic Hand
Sanitizer Dispenser. ALGORITMA: JURNAL ILMU KOMPUTER
DAN INFORMATIKA, 5(2)
[24] Gupta, a., Kumar, r., & Gupta, r. (2021). Novel design and performance
analysis of automatic sanitizer dispenser machine based on the ultra-
sonic sensor. In recent trends in communication and electronics (pp.
320-325). CRC press.
[25] Deshpande, S., Aggarwal, A., Ponraj, A. S., & Jackson, J. C. IoT-
Based Smart Hand Sanitizer Dispenser (COVID-19). In Smart Buildings
Digitalization (pp. 281-302). CRC Press.
[26] Nof, S. Y. (2009). Automation: What it means to us around the world.
In Springer handbook of automation (pp. 13-52). Springer, Berlin,
Heidelberg.
[27] Kumar, B. K., Thanusha, N. J., Hiremath, S., Soujanya, K., Vanishree,
B. C., & Navadeep, G. U. (2021, August). A Solar Powered Kiosk for
Contactless Body Temperature Sensor and Hand Sanitizer Dispenser to
Monitor and Control Covid-19 Disease. In 2021 International Confer-
ence on Recent Trends on Electronics, Information, Communication &
Technology (RTEICT) (pp. 884-888). IEEE.
[28] Lastovicka-Medin, G., & Vanja, B. (2021, June). From Contactless
Disinfection Intelligent Hand Sanitizer Dispenser for Public & Home
towards IoT Based Assistive Technologies for Visually Impaired Users
Institutional Responses to the COVID-19 Pandemic. In 2021 10th
 References 271

Mediterranean Conference on Embedded Computing (MECO) (pp. 1-4).

IEEE.
[29] Srihari, M. M. (2020, July). Self-activating sanitizer with the battery-
imposed system for cleansing hands. In 2020 Second International Con-
ference on Inventive Research in Computing Applications (ICIRCA)
(pp. 1102-1105). IEEE.
[30] Ajayan, A. (2020, December). Automizer-An Automatic Sanitizer Dis-
penser. In 2020 IEEE International Women in Engineering (WIE)
Conference on Electrical and Computer Engineering (WIECON-ECE)
(pp. 153-156). IEEE.
[31] Bal, M., & Abrishambaf, R. (2017, March). A system for monitor-
ing hand hygiene compliance based on Internet-of-Things. In 2017
IEEE International Conference on Industrial Technology (ICIT) (pp.
1348-1353). IEEE.
[32] Ganeshkumar, D. (2021, October). Design and Implementation of
Hands-Free Electronic Sanitizer Dispenser. In 2021 Smart Technolo-
gies, Communication and Robotics (STCR) (pp. 1-4). IEEE.
[33] Nayana, D. K., Mallikarjun Patil, K. S., Manoj, A. M., Mallikarjun, G.
N., & NS, S. (2021). AUTOMATIC HAND SANITIZER DISPENSER
USING ARDUINO. NEW ARCH-INTERNATIONAL JOURNAL OF
CONTEMPORARY ARCHITECTURE, 8(1s), 283-286.
[34] Syahputra, M. I., Khair, U., & Sembiring, A. (2021). Automatic Hand
Sanitizer Dispenser. ALGORITMA: JURNAL ILMU KOMPUTER
DAN INFORMATIKA, 5(2).
 13
Energy Efficiency Investigation in Massive
MIMO Communication for Industry
Automation using Precoding Schemes

R. Surender1 , S. Sudharsan2 , S. Sundaresan3 ,

and Sunday Adeola Ajagbe4
1 Department of ECE, Sathyabama Institute of Science and Technology,
Chennai
2 Department of ECE, Rajalakshmi Engineering College, India
3 Department of Electronics & Communication, NIT Karaikal, India
4 Department of Computer Engineering,

Ladoke Akintola University of Technology, LAUTECH, Nigeria

E-mail: [Link]@[Link]; sudharsan.s@[Link];
sundaresanece91@[Link]; saajagbe@[Link]

Abstract
Fifth-generation wireless access technology aims at consistent quality of
everything (QoE), and this demand can be accomplished by enhancing the
network throughput. Massive multiple input multiple output (mMIMO) sys-
tem is a promising candidate that can be employed at the base station (BS)
and afford high reliability and network sum rate, improving the network
energy efficiency (EE). The network employing mMIMO system is ana-
lyzed in the presence of interference to attain better throughput and network
energy efficiency. Different precoding schemes like maximal ratio combining
(MRC), zero-forcing (ZF), and pilot zero-forcing (P-ZF) are employed in the
network for mMIMO communication system analysis. Further, the work is
analyzed by incorporating regular and superimposed pilot in the maximal
ratio combining and zero-forcing precoding schemes. Performance attributes

273
274 Energy Efficiency Investigation in Massive MIMO Communication

like average rate, BS energy efficiency, user equipment energy efficiency,

and network power consumption are investigated for upgrading the mMIMO
network performance in industrial automation.

Keywords: QoE, mMIMO, precoding schemes, superimposed pilot, industry

automation.

13.1 Introduction
The tremendous growth in development of electronic gadgets and vari-
ous emerging applications like artificial intelligence, augmented reality, big
data analytics, virtual reality, ultra–high-definition transmission video, three-
dimensional (3D) media, etc., have encapsulated the wireless communication
field and created a substantial evolution in the wireless networks data volume.
In the interim, mobile networks have become essential services for computing
personal devices. As new technologies have emerged, the need for fast wire-
less services like switched traffic, multimedia, and IP data packets has grown.
Without these services, there would be no way for mobile communication
systems to work. On the other hand, the system’s design should meet the
customers’ needs without making them pay extra. Balancing complexity,
adaptability, data rate, quality-of-service (QoS), and cost is essential in
commercial applications. Modern improvements in microelectronics, signal
processing, mobile computing, and other related fields have enabled high
spectral efficiency and high degrees of adaptability [1].
The conventional mobile system has difficulty fulfilling the traffic demand
for data and spectral scarcity. The wireless system’s spectral efficiency
depends on attributes like signal-to-noise ratio and other attributes like spatial
correlation, accuracy in channel estimation, hardware impairment, and signal
processing resources. So, it is essential to enhance the spectral efficiency
as it increases the demand for wireless services. Recently, advancements in
the 5G systems have been exposed by deploying additional antennae at the
transceiver. The capacity of the wireless system can be increased substantially
without an increase in the bandwidth and transmission power. The design
incorporating multiple antenna elements at uplink and downlink is a MIMO
system [2].
In mobile communication [3], the signals transmitted are attenuated
by fading and obstructed in between the transceiver by shadowing. This
sequence yields a primary challenge for seamless communication, and the
MIMO system can achieve it. The system uses spatial multiplexing to achieve
 13.2 Evolution of Wireless Communication System 275

multiplexing gain and improves the communication system’s capacity. The

effort to achieve the gain by spatial multiplexing has been lifted from MIMO
to mMIMO. The base station serves several users simultaneously; even each
user has a single antenna, which overcomes the limitation of propagation in
the channel.

13.2 Evolution of Wireless Communication System

Wireless communication has become a vibrant part of the present commu-
nication system, transforming the present communication and living system.
As the wireless system progress from one generation to other generation ,
there is upgradation in the system’s data rate, mobility, coverage, and the
operator’s spectral capabilities. In the early 1980s, the 1G system was united
with analog services with 2.4 Kbps which is more prone to noise and easily
vulnerable [4]. The 2G system user digital services for voice communication
uses 64 kbps data rate, whereas 2.5G systems like GPRS and EDGE have
evolved with 144 kbps data rate.
In late 2000, the 3G wireless system imparted data rates up to 2 Mbps
by merging mobile access with high-speed to IP-based services. This feature
enhances the improvement in QoS and global roaming at more power con-
sumption in the network. At the intermediate with an enhanced data rate of
5–30 Mbps, Universal Mobile Telecommunication System (UMTS) evolved
[5, 6]. Later in 4G, to supplement the network capacity and to facilitate
a more significant number of users with services like video on demand,
file sharing, and Web services evolved to accomplish it. The 4G system is
generally highlighted to be the inheritance of 2G & 3G standards, with a
3G partnership project (3GPP) standardizing the networks. The 4G wireless
system improves the conventional approach by fully imparting an IP-based
reliable solution. This makes ubiquitous communication a possible one when
compared to traditional techniques of communication [7].
LTE enhances wireless systems’ network capacity and mobility by using
the core network and radio interface. LTE does not fulfill the service capacity
of a 4G wireless. LTE-Advanced is an upgraded standard of the LTE, which
delivers a high data rate, unlike current LTE networks, and is also known as
True4G. LTE-A incorporates many hardware and software technologies to
accomplish high network standards. They are:
i. Increased uplink and downlink peak data rate
ii. Comparatively high spectral efficiency
276 Energy Efficiency Investigation in Massive MIMO Communication

iii. More number of active subscribers

iv. Improved performance at cell edges
v. Carrier aggregation
vi. Enhanced usage of multi-antenna techniques
vii. Relay nodes support
The ubiquitous system exhibits an exponential increase in data traffic as
it incorporates a more significant number of users. This data traffic makes
the 4G system be replaced by a 5G system with beam division multiple
access (BDMA) and filter bank multi-carrier (FBMC) . BDMA increases
the system by allocating an orthogonal beam to each user; it will share the
same shaft according to the location of MS. The 5G network with BDMA
addresses the following challenge, which is not discussed by 4G [8]. They
are high system capacity and data, latency reduction in end-to-end services,
connectivity with massive device, and providing better quality of experience.
To fulfill the user demands and to attain the challenges of the 5G system,
a new network was designed for 5G wireless architecture. Here, the base
station uses antennas on a large scale for network communication with users
using BDMA technology, hence the name MIMO. Since conventional MIMO
system has the constraint of using antenna terminals, it limits the attributes of
improving the performance in the network. As MIMO uses an antenna on a
large scale, the beneficial features obtained by the system are high in terms
of capacity and gain in the network [9].

13.3 Challenges in the 5G Communication System

Fifth-generation cellular network demands high area throughput for the expo-
nential evolution of wireless data traffic. To accompany the traffic growth in
5G technology, the system’s throughput must be improved a thousand times.
It can be enhanced by [10]
i. providing a high user data rate;
ii. allocating more bandwidth to serve the 5G network;
iii. densifying the network with independent access points and small cell;
iv. improving the network efficiency.
Various applications in wireless network may lead to many technical chal-
lenges for the users. To provide a reliable seamless service for these wireless
applications and the requirements of 5G [11], the following components are
involved:
 13.4 Massive Multiple Input Multiple Output 277

With the advent of transmission waveforms, novel resource management

methods, and multiple access control are developed.
To develop multi-hop technologies and advance inter-node coordination
schemes, antennas in large-scale configurations are designed with multimode.
The design consideration of the network considers the data traffic,
mobility handling capability, and novel and efficient way of interference
management, which involves complexity in the integration of the network.
The usage of the spectrum includes a band of operation in the prolonged
region of the spectrum. The process in this region forms a new set of concepts
completely for the prolonged spectrum, and it must be prudently used in the
system.
mMIMO is a multi-antenna technique to multiplex the user terminal over
the entire bandwidth spatially and reduces the data traffic growth in the 5G
cellular system. By reducing 5G network traffic, the system throughput can be
significantly increased. It also provides other benefits, such as lower latency,
cheaper parts, and robustness against intentional jamming.

13.4 Massive Multiple Input Multiple Output

mMIMO is the advanced and improved version of MIMO in which the anten-
nas are grouped at the transceiver to enhance throughput and efficiency. It is
a MIMO technology with a multiuser where a BS is deployed with antenna
elements on a large scale. It employs these antenna elements to connect with
single antenna UEs over a similar time and frequency band since it uses
space division multiple access (SDMA). The SDMA makes the mMIMO
into a scalable form that makes coherent signal processing over a large scale
of antennas. In the downlink of the mMIMO system, transmit precoding is
incorporated, and it is made to focus each signal to send at the intended
terminal. In the uplink, receive combining is incorporated to eliminate the
transmitted signals across various terminals [12]. The illustration of mMIMO
transmission is shown in Figure 13.1.
Time duplexing mode is used in which the transmission of the signal
occurs at identical frequencies but at different times. This condition highlights
the responses of the channel are equal in both uplink and downlink directions.
The system has many advantages to operate in time division duplexing (TDD)
mode. First, the channel conditions must be known only to the base station
for coherent processing using antenna elements. Second, overhead in the
estimation of the uplink is proportional to the UE. Third, the approximation
structure of channel responses will not degrade by using additional antennas
278 Energy Efficiency Investigation in Massive MIMO Communication

Figure 13.1 Illustration of transmission in an mMIMO.

at BS in TDD mode. It uses spatial multiplexing that depends on BS to

estimate channels with proper information on the channel (CSI) in uplink
and downlink. In downlink estimation, the pilots are placed orthogonal to
one another, which varies the time and frequency of resources. The quality
of the estimated channels might be reciprocated in TDD mode in the uplink,
where it depends on the BS antennas [13]. The spatial multiplexing technique
transmits the signal to the typical user and reduces the impact of interference
on the other user. The intended signal quality is optimized by weighing and
summing the antenna signals in the various antenna arrays.
The importance of multiple antenna systems is improvement in array
gain, reduction in interference, and diversity gain. It provides a high mul-
tiplexing boost by exploiting both the transmit and receive antenna using a
traditional MIMO system [14]. The improvement in array gain at the receiver
is retrieved in the transceiver by coherent signal processing. The system gain
is attained properly when the transmitter gains knowledge of the CSI and
proper correlation with the received signals.
The concept of frequency reuse in wireless channels introduces network
interference and degrades the performance of the wireless network. By using
different spatial signatures and proper estimation of CSI, the interference in
the network can be reduced. Diversity is another method to combat interfer-
ence by reducing fading in the channels. The signal in deep fade for one user
will not be the same for the other. The more independent the fading channel
 13.4 Massive Multiple Input Multiple Output 279

the higher the probability that the signal is not in the deep fade. The reason to
use space diversity is that it shares the same frequency and time resource. To
utilize this resource, the antennas must be appropriately spaced; otherwise,
signals will correlate with interference and reduce the diversity gain.
To enhance the system performance, which is limited by interference, the
transmitter has to be adopted a precoding scheme, as shown in Figure 13.2.
A precoding scheme is a design process of the transmitting signal before
transmission, making communication effective to the intended user in a
multiuser MIMO system. Each antenna individually transmits a stream of
data in a precoding scheme which is a linear combination of all data streams
employed with beamforming weights which makes sharing a standard data
DL at the transceiver and exploiting the DL reciprocity concept.
Wireless communication toward 5G makes energy efficiency an essential
criterion for design in sustaining the evolution. The information and com-
munication technology (ICT) sector focuses on rapid progress in evolving
5G, which expects to integrate at most everything across the world into the

Figure 13.2 Illustration of precoding scheme at the transmitter.

280 Energy Efficiency Investigation in Massive MIMO Communication

Internet. The network has adequate access to power sources to serve several
devices in an unprecedented manner, which is known as densification. The
ultimate aim of densification is to provide capacity 1000 times than the
conventional network. The 5G network operators spend more on network
operation, emitting 2 percent of total carbon dioxide [16]. To overcome this
constraint, the energy efficiency perspective is adopted, referred to as green
communication.
mMIMO is a 5G technology that extracts energy and spectral efficiency
over an LTE-A network with an extensive array of antennas. Energy-efficient
communication has the advantage of exploiting the interference in co-
channel. It also reduces the environmental effects like the dissipation of heat
and electronic pollution. Compared to the conventional system, there are two
ways to achieve energy efficiency on a large scale. First, to obtain the fixed
throughput in a given system, the power transmission in the UE must be
reduced by upgrading the transmitter side antennas. Second, more gain in
the network can be achieved by incorporating more UEs in the system.
In the mMIMO system, an increase in BS and UEs does not guarantee
energy efficiency improvement because circuit power consumption increases
with the number of BS and UEs. Subsequently, an increase in BS can improve
energy efficiency by reducing the UE’s transmission power. Similarly, energy
efficiency can be enhanced by incorporating more UEs at the dominance of
power expenditure. As energy efficiency is inversely proportional to circuit
power consumption, it can be attained by considerably diminishing overall
network power consumption [17].
MIMO is a challenging multi-antenna technology to upgrade the data rate
in a 5G system for seamless communication. As the mMIMO is characterized
by the deployment of BS in large numbers, the network gets affected due to
interference which reduces the network performance. The interference must
be considerably reduced to enhance the performance. The interference in the
system appears in the following ways: interference caused by the same UE
in the same cells and UE formed in different cells is recognized as intercell
interference and intracell interference, respectively. This interference can be
moderated by using proper precoding schemes, which diminishes the interfer-
ence and enhances the performance of MIMO. Another factor that degrades
the mMIMO performance is pilot contamination which gets introduced due
to the frequent use of pilot symbols and limitation in coherence block. By
increasing the duration of the coherence block and proper channel estimation
techniques, the system’s performance can be upgraded by combating the
interference that occurs due to pilot contamination.
 13.4 Massive Multiple Input Multiple Output 281

Various studies have focused on maximizing the performance of the

mMIMO communication system over the conventional MIMO system. How-
ever, most problems addressed in the published research papers are associated
with traditional MIMO systems regarding the interference problem. Inter-
estingly, such attempts have made a better pathway for other categories of
research works toward different future advanced versions of MIMO networks.
This section discusses the research contribution of interference reduction in
the mMIMO network using a precoding scheme.
The first, LTE, was proposed by NTT DOCOMO of Japan, commercially
launched by Sweden and Japan, and followed by the United States. More
LTE networks were installed as GSM, UMTS, and CDMA 2000. Long-term
evolution advanced is the enhancement of LTE, and it is used in 4G to support
high-end applications [18, 19]. The essential characteristic is to improve the
channel capacity and data rate in wireless application, and it can be attained
by the MIMO system, which the idea was introduced by authors in [20].
As the conventional MIMO system is limited in base station antennas, the
network’s data rate is also limited. By deploying antennas on a large scale,
the LTE-A network achieves the ability of a 5G network. The usage of the
antenna on a large scale is said to be mMIMO which introduces interference
from unintended users.
HQ Ngo et al. analyzed the uplink of the systems in perfect and imperfect
CSI with precoding schemes like MRC, ZF, and minimum mean square error.
The study of the proposed work demonstrated that moderate usage of the
antenna on a large scale could considerably increase spectral efficiency [21].
Emil et al. [22] considered single-cell and multicell MU-MIMO networks to
analyze the network’s performance in perfect CSI and imperfect CSI using the
precoding schemes ZF, MRC, and MMSE processing. The simulated result
concluded that energy efficiency performance is preferable in the region of
high SNR and can be attained by suppressing interference using the precoding
scheme.
Wang et al. [23] demonstrated that ZF processing could seamlessly cancel
the self-interference (SI) in a Rician fading, where the transceiver antennas
are large. The study of the proposed work in the paper indicated that the
SI could be canceled for short-range communications by using limited base
station antennas. De Mi et al. [24], in the existence of reciprocity errors,
deduced signal-to-interference noise ratio expressions (SINR) for ZF and
MRT, and further investigation highlighted that effect of reciprocity errors
could reduce the SNR. Lee et al. [25] investigated the mMIMO performance,
and the numerical results proved that SE is reduced due to a high clipping
282 Energy Efficiency Investigation in Massive MIMO Communication

ratio causing degradation in energy efficiency. The proposed iterative clipping

and filtering technique gave better gain with the ZF precoding scheme.
Hoydis et al. [26] have presented a study on multicell mMIMO networks
to analyze the performance of achievable rates in the uplink and downlink
using different precoding schemes like regularized zero-forcing (RZF) and
matched filter (MF). The authors have considered the work based on the
imperfect CSI, path loss, and pilot contamination; and the outcome of the
result proved that RZF preceding scheme can perform better than the MF
preceding scheme with a limited number of BS antennas. Bjornson et al.
[27] considered the scenario in the downlink of the single-cell system to
analyze the energy efficiency in the cellular network with a modified form
of precoding scheme known as multi-flow RZF with small cell access (SCA).
The precoding scheme is introduced with a low-complexity algorithm that
consumes static and hardware power in the network, improving the quality of
everything (QoE).
Li et al. [28] investigated the performance in a cellular network with
BS antennas on a large scale and analyzed the impact of both BS antennas
and user equipment density in the network. The simulated outcome of the
study highlighted that an increase in the thickness of BS would enhance
the throughput linearly and then, in turn, increases the energy efficiency of
the mMIMO network. Gao et al. [29] presented a study on performance
measurement of massive antennas using ZF and MMSE processing in the
downlink channel and compared it with dirty paper coding (DPC). The
paper’s outcome proved that using linear precoding schemes, 98 percent of
the sum rate can be achieved compared to DPC with a limited number of
antennas in BS.
Hu et al. [30] presented a study on the performance of SE in network
MIMO compared to mMIMO architecture employing ZF precoding schemes.
The web MIMO is operated by grouping the user equipment in the corre-
sponding locations with time-frequency slots. The study’s outcome proves
that the proposed work achieves better users’ spectral efficiency in different
areas. The work in [31] analyzed the specific system sum rate with ZF and
MRC precoding schemes. The outcome of the work proved the given data
rate is achieved by minimizing the total power consumed at BS with optimal
transmission power, BS antennas, and users in the network.
Hao et al. [32] proposed a framework to analyze the downlink of heteroge-
neous networks employing a low-complexity algorithm. The proposed work
enhanced the tradeoff with optimal antennas in BS significantly. The authors
in [33] presented the work on the downlink of mMIMO using MRC and ZF
 13.4 Massive Multiple Input Multiple Output 283

precoding schemes to study the problem of max-min fairness (MMF). The

proposed work in multiantenna system incorporating multicasting, identified
the optimal power for uplink pilots and downlink precoders which provided
better spectral efficiency for modes of precoding schemes which can alter
between them based on the system’s parameters.
Hu et al. [34] incorporated the same set of pilot sequences in the mMIMO
network and assumed it is reused in all cells. It mitigates the interference
by exploiting covariance matrices. The proposed architecture in this paper
is based on cooperative base stations and achieved a comparable spectral
efficiency with the ZF precoding scheme. The authors in [35] used another
method to combat interference, having longer pilot sequences than the users
per cell using a similar set of pilots. This method reduced the impact of
pilot contamination at the overhead estimation penalty, decreasing the data
symbols incorporated in the coherence block.
Muller et al. [36, 37] proposed a method of subspace prediction to
enhance channel estimation with power-controlled hand-off in multi-antenna
systems. The proposed blind method does not need pilot data to identify
the proper subspace, and it can mitigate interference without coordination
between the cells. Yin et al. [38] presented a method of pilot allocation for
desired and interfering UE in correlated channels. This novelty undertook
the problem by allowing low-rate cell coordination in the channel estima-
tion duration. It provided additional user channel information and proved
a way to discriminate against interfering users even when pilot sequences
correlate strongly. A pilot decontamination array processing method has
been presented for the use of channels with discrete paths which is finite
in number. This method improved the performance of the system compared
to the existing approaches.
Takeuchi et al. [39] analyzed the large-scale antenna system using the
replica method under the replica symmetry assumption. The study of the tech-
nique proved that pilot information is the finest in the condition of achievable
data rates of successive decoding receivers for the antenna on a large scale.
The study in the proposed results suggests that it can significantly gain the
average SNR regimes. In [40], the authors proposed segregating method to
improve the throughput. The simulation results depict that performance can
be enhanced with reuse-1 architectures. Li et al. [41] considered an uplink
system that uses pilot reuse to obtain the optimum pilots reuse factor between
the cells using the MRC precoding scheme. The numerical results proved that
pilot reuse could achieve high throughput in cellular networks, in which the
traditional interference term controls the pilot contamination term. Noh et al.
284 Energy Efficiency Investigation in Massive MIMO Communication

[42] proposed an algorithm for pilots to exploit the interference that degrades
the network’s performance. The resulting low-complexity design generates
optimal pilot sequence beam patterns for the given parameters and proves the
importance of the proposed algorithm.
Upadya et al. [43] investigated the DL performance of a network that
uses SP to estimate the channel to reduce pilot contamination. The authors
used a normalized mean-squared error channel estimate for matched filter and
compared it with Bayesian Cramer-Rao lower bound. The former channel
estimation proved that the system’s throughput decreases with an increase
in antenna number. It further improved the DL throughput using SP than
the regular pilots. Jing et al. [44,45] stated that contamination of the pilot
and interference in the data are two essential features that reduce the SP-
based channel estimation accuracy. The authors proposed an approach to
improve the interference and proved that the proposed method mitigates the
interference and improves the channel estimation and SE.
Zhang et al. [46] presented an SP method for the estimation of channel
schemes and studied the outcome of SP on a very large-scale antenna. This
method established that the SP scheme’s outcome results could improve pilot
contamination. It is proficient in increasing the frame size and enhancing
the spectral efficiency. Li et al. [47] used SP to estimate channels in the
mMIMO system using simple linear detection like MRC and ZF. The pro-
posed scheme yielded a superior performance over the conventional method.
Upadhya et al. [48] proposed a scheme for uplink channel estimation based
on SP employing matched filter precoding scheme for channel estimate. This
precoding scheme utilized reduced the overall interference. The study of
the simulation proved the effectiveness of the proposed schemes. Li et al.
[49] proposed a superimposed pilot in the mMIMO system using MRC and
ZF precoding schemes. The study of the proposed work demonstrated that
channel estimation with SP yielded a better result by maximizing the network
performance. Verenzuela et al. [50] evaluated the network performance with
MRC estimated with different methods. They proved that the proposed SP
work outperforms the other estimated methods by reducing interference and
enhancing energy efficiency.
Tominaga et al. [51] investigated the performance of mMIMO and
revealed that NOMA-based mMIMO provides significant gain compared
with different precoding schemes. David et al. [52] enabled the usage of
critical technologies with mMIMO, like the carrier design, advanced sleep
mode, and machine learning, and highlighted the need for QoS demands
for end-users. The authors also stated that MRC, characterized by lower
 13.5 Precoding Schemes 285

complexity, achieves higher energy efficiency than ZF. But ZF having the
ability to combat inter-cell interference outperforms MRC in terms of average
sum-rate EE.
This work aims to investigate the 5G network performance by imple-
menting mMIMO communication. This will allow seamless communication
between industrial automation systems. Further, the work contributes to
achieving seamless communication by incorporating different precoding
schemes with regular and superimposed pilots considering the interferences
in the network. The work in this chapter is advanced to implement a reduced
interference network for a compatible real-time application for 5G IoT-based
industry automation.

13.5 Precoding Schemes

To simplify the receiver’s structure and reduce the receiver’s consumable
power, the precoding process is used at transmit side instead of receive side.
Precoding is a linear processing technique that exploits convey diversity, i.e.,
the source transmits the coded information of the transmitted signal to the
receiver to realize the CSI. Therefore, the precoding technique is employed
at the transmit signal before transmission, which reduces the performance
degradation caused by the interference and channel fading [53].
In the mMIMO LTE-A network, each BS is employed with an M array
of antennas to communicate simultaneously to K single antenna UEs. To
recover the signal from a noisy channel, coherent processing is accomplished
by employing transmit precoding and receive combining in the downlink and
uplink, respectively.

13.5.1 Maximal ratio combining

In MRC, signal amplification process is carried out by adding coherent com-
ponents to the receiver. This process is called array gain, which is proportional
to number of M antennas in the BS. Typically, the transmit signal vector is
√
given by x = ρW s,
where s is the transmitted signal before precoding
ρ is average BS transmit power
W is the precoding matrix.
The downlink’s conjugate transpose matrix is used√to represent the
matched filter (MF) precoder and it is represented as W = αH ∗ . Therefore,
286 Energy Efficiency Investigation in Massive MIMO Communication

the received signal vector (yMR ) of an MRC is given by

√
yMR = ρα s H T H ∗ + n0 , (13.1)

where α is the scaling factor to normalize signal power

H T is of downlink channel matrix transpose
H ∗ is of downlink channel matrix conjugate
n0 is the noise matrix.
The conjugate transpose of the channel matrix is the downlink precoder.
When the BS is large in number compared to UEs in the network, the optimal
solution is attained by the diagonal matrix of the term H T H ∗ [54].

13.5.2 Zero forcing

It is a precoding scheme that considers inter-user interference, irrespective of
noise. By sending the signal in the direction of the desired user ZF cancels
out the interference of other users. An array gain that is proportionate to the
(M-K) antennas in BS is produced by the signal amplification in ZF and is
given by
√  −1
yZF = ρα s H T H ∗ H T H ∗ + n0 (13.2)
The H T H ∗ matrix’s diagonal elements represent the channels’ relative
power imbalance, while the off-diagonal elements reflect the channels’ inter-
dependence. When additive noise is absent, ZF is an optimal precoding
scheme; otherwise, it is not optimal [55, 56]. Eqn (13.1) and (13.2) are
the general form of the received signal vector of MRC precoding and ZF
precoding, respectively.

13.5.3 Pilot zero-forcing (P-ZF) precoding scheme

Intercell interference is mitigated across K channels in a conventional ZF
precoding scheme. P-ZF combining proposed in [57] mitigates intercell and
intracell interference. Channel directions that are estimated are orthogo-
nalized to combat interference that is known at the BS by synchronizing
combinations across B cells. Hence,
P-ZF precoding scheme will combat both interferences and enhances the
network performance.
The P-ZF channel matrix is given by
H
yP−ZF = Hγj (Hγj Hγj )−1 , (13.3)
 13.6 Superimposed Pilots 287

where all channel users are represented H = [H1 , H2 · · · HK ]. Here ξ are the
pilot signals that originated from the code book and it is provided by

 
H B, b1 = b2
ξ = {ν1 , · · · νB } , where νb1 νb2 = , (13.4)
0, b1 = b2

with B signals of the pilot to form orthogonality in BS on the constraint of

1 ≤ B ≤ CB .
The network coherence block has a limited size where the channel is
assumed as static and frequency flat. The channels are estimated using
orthogonal pilot sequences in finite numbers. As the pilot sequences used
are the same used in multicell mMIMO system, coherence interference is
generated known as pilot contamination between UEs. This pilot contami-
nation reduces the channel estimation and mMIMO spectral efficiency. To
minimize interference in the conventional or regular pilot (RP), the pilot and
data transmission are accompanied individually within the coherence block.

13.6 Superimposed Pilots

A superimposed pilot is one in which the data symbols and transmission
pilot both take place concurrently within a coherence block. The idea behind
superimposed pilots is to aggregate the data samples and the pilot samples
into a single total, rather than separating them based on time and/or frequency.
The UE broadcast signal can be employed to estimate data or pilots or to
superimpose the two, depending on the methodology.

Figure 13.3 Regular and superimposed pilot protocol.

288 Energy Efficiency Investigation in Massive MIMO Communication

Figure 13.3 [58] illustrates the regular and superimposed pilots’ respec-
tive coherence frame formats, which can be used to gain an understanding
of the regular and superimposed pilot concepts. The pilot contamination has
been decreased as a result of using the complete coherence block of SP pilots,
which has improved the channel estimate. Additionally, there is no penalty
incurred in the pre-log factor.

13.7 System Model

The multicell mMIMO network for the uplink model is considered here in
which each base station with M antennas serves K user equipment. The
number of channels obtained by dividing the network bandwidth BW by
the coherence bandwidth BC is BBW C
. The time duration of every coherence
block TC contains τC = TC BC complex samples. Because they provide an
acceptable data rate, the uncorrelated Rayleigh fading channels are used in
multicell mMIMO systems.
The channel between the U El  i and M antennas BSl is modeled as
hll  i a small-scale fading condition cll  i ∼ CN (0, ζll  i IM )∀l ,l  ∈ φλ with
ζll  i ≥ 0 large-scale fading condition. For instance, the distance between K
single antennas and BS antennas is considered to be significant in number for
studying large-scale fading. The condition also represents the received signal
y0 from the base station BS0 [58].
K
 
y0 = h0l  i Xl  i + n0 , (13.5)
l  ∈φλ i=1

whereh0l  i is the channel vector, xl  i is the transmitted signal, and n0 is the

noise vector

13.8 Regular Pilots

In each regular
 pilot coherence block, τp samples are used for pilot, and a
τc −τp
fraction τc of samples are used for data. Each base station allocates
τp (τp ≥ K) pilot sequences to UEs in its cell given as Φl  i . It is given as to
recognize the users sharing the same pilot U E0k in different cells P0k RP =
 ∗ T
{l , i } : ϕ0k ϕl  ,i = 0 . The U E0k sends pilot Φ0k with other users for τp
 13.10 Energy Efficiency for Different Precoding Schemes 289
RP received at BS is [58]
instances of eqn (13.5), and the signal Z0k 0

K
 
RP √
Z0k = ql  i h0l i ϕTl i + N̄0 , (13.6)
l ∈φλ i=1

where ql  i is the transmission power of U El  i , the pilot symbol and noise

Φ∗0k
RP (eqn (13.6)) grew √
matrix, N̄0 Now the received signal sequence Z0k τp
to correlate pilot signals with equivalentU E0k pilot sequences. The despread
received signal is given by [58]
 √
RP
z0k = ql  i τp h0l i + n̄0 , (13.7)
R
{l ,i} ∈P0k

N̄ Φ∗  
where n̄0 = √0 τp0k is a noise represented as n̄0 ∼ CN 0, σ 2 IM with no
information vanished in the despreading process.

13.9 Superimposed Pilot

Coherence block samples are used in this transmission to send the pilot and
data symbols. Each BS assigns τc (τc ≥ K) pilot sequence to its UE in its cell
and represents ψl  i to identify the different cell users sharing similar pilot
U E0k s. The U E0k sends pilot sequence ψ0k T for other users and it is given

by [58]
K
 K

SP √ √
Z0k = ql  i h0l i ψ T l i + pl i h0l i sTl i + n0 , (13.8)
l ∈φ λ i=1 l ∈φ λ i=1

where pl  i is data symbol power, sl  i is a vector that encapsulates the

transmission data symbols. n0 is the noise vector  represented by n0 =
[n01 , . . . , noτc ], whereas n0j ∼ CN 0, σ 2 IM . The SP signal Z0SP
ψ∗
(eqn (13.8)) grew √0kτc to correlate the pilot signals with respect pilot
sequenceU E0k , and it produces

SP
 √
K
 p l i T ∗
τc
 [ψ0k ]∗j
z0k = q τc h
l i 0l i + h0l i sl i ψ0k +
 n0j √ .
τc τc
{l ,i}∈PS l ∈φλ i=1 j=1
0k
(13.9)
290 Energy Efficiency Investigation in Massive MIMO Communication

13.10 Energy Efficiency for Different Precoding Schemes

The average sum rate per unit area to the ratio of the average power con-
sumption per unit area can be used to calculate the EE in multicell mMIMO
systems [67]. It is represented as

Energy Efficiency (EE)

Average sum rate per UE (R0k )
=
Average power consumption per BS(PBS )

13.11 Average Rate per Unit Area

The channel for both conventional precoding scheme and precoding with SP
has been LMMSE estimated using eqn (13.7) and (13.9) is computed between
the BS and UE, respectively [58].
The achievable rate for U E0k with P-ZF precoding scheme and conven-
tional or RP is given by [58]

RP τp  RP

R0k = 1− Bw log2 1 + γ0k , (13.10)
τc
RP is the SINR of the regular pilot and is provided by
where γ0k

RP Gf p0k ζ00k
γ0k =  K 2
ζ0l
 K .
Gf p l  i ql  i i 1 2
τp 
l ∈φλ i=1 q0k ζ00k + ϕRP 
l ∈φλ p  ζ
i=1 l i 0l  i + σ
0k
(13.11)
ϕRP
0k is the channel estimate quality of RP and is provided by

q0k τp ζ00k
ϕRP
0k =  K . (13.12)
q0k τp ζ00k + 2 + σ2
l  ∈φλ i=1 pl  i ζ0l i

Gf is a factor scale caused by array gain from a different precoding

strategy as
Gf = M for MRC
Gf = M − K for ZF
Gf = M − B for P-ZF.
The achievable rate for U E0k incorporating SP precoding schemes is
given by  
SP SP
R0k = Bw log2 1 + γ0k , (13.13)
 13.13 Results and Discussion 291
SP is the SINR of superimposed pilot and is given by
where γ0k
SP
γ0k
Gf p0k ζ00k
= Gf  K p  q  ζ 
2 Gf  K p2 q  ζ 2   K p2
l i l i 0l i + l i l i 0l i + τ12 l i
l ∈φ l ∈φλ l ∈φλ
τc λ i=1
2
q0k ζ00k
 τc
K
i=1 q0k ζ00k
 c
i=1 q0k
ζ0l 1
ζ00k
i
+ ϕSP l  ∈φλ i=1 pl  i ζ0l  i + σ 2 ,
0k
(13.14)

where ϕSP
0k is the channel estimate quality of SP and is given by

q0k τc ζ00k
ϕSP
0k =  K  K
q0k τc ζ00k + l  ∈φλ i=1 ql  i ζ0l  i + l ∈φλ i=1 pl i ζ0l i + σ 2
(13.15)

13.12 Average Power Consumption

The power consumption per unit area (PBS )is computed by considering the
network transmit and circuit power requirements. The power combination in
linear processing PLP and channel estimation PCE studies the system power
consumption. The linear power consumption consumes two times of power
for SP, and it is given by [58]
 Bw
PLP + PCE = L MK f or RP − ZF
,
2Bw
L MK f or SP − ZF

where L is the processor’s computational efficiency.

The EE of the network using different precoding schemes with SP is
given by
R0k K
EE = ,
PT X + C0 + C1 K + D0 (M − K) + PLP + PCE + A (R0k ) K
(13.16)
Γ(α/2)+1
where PT X = ρΥη α/2 is the average transmit power for UE andη is
(πλ)
the overall power amplifier efficiency. Here C0 is the static power consumed
at the BS, D0 which is the power consumed by the base station transceiver
chains, which scales with M base station antennas, and C1 K is the power
consumed by the user equipment at the base station during signal processing.
A is the proportionality constant for coding and decoding.
292 Energy Efficiency Investigation in Massive MIMO Communication

Figure 13.4 Cell under investigation surrounded by a cluster.

13.13 Results and Discussion

A multicell scenario of each 500 m ×500 m cell size is employed for
simulation using the latest MATLAB software version, which can be used
for developing a real-time 5G network, as shown in Figure 13.4 [22]. The
users are dispersed evenly throughout the network, and the Monte-Carlo
simulation was used to simulate the scenario, which is likely to experience
user interference in all directions. An energy efficiency investigation in a 5G
network using different precoding schemes and the superimposed pilot has
been performed. The simulation is carried out using a propagation model
based on the 3GPP standard [59] listed in Table 13.1.

13.13.1 Performance of various precoding schemes

Figure 13.5 highlights the energy efficiency of different precoding schemes
under different ranges of BS antennae for mMIMO communication. The P-
ZF energy efficiency increases to 16.32 Mbit/J until M = 210 and becomes
steady with increased BS antennas.
This is because P-ZF mitigates intercell and intracell interference and
enhances the performance associated with that of another precoding scheme.
However, SP-ZF energy efficiency increases to 12.89 Mbit/J till M = 180,
and ZF energy efficiency increases to 9.72 Mbit/J till M = 140, decreasing
gradually as the ZF precoding scheme mitigates only intercell interference.
The SP-MRC energy efficiency rises to 7.44 Mbit/J at M = 155, and MRC
 13.14 Power Amplifier Power of Different Precoding Schemes 293

Table 13.1 Simulation parameters.

Parameter Value
Number of BS antennas, (M) 300
Number of UEs, (K) 120
Transmission bandwidth, (Bw ) 20 MHz
Coherence block length 400
Propagation loss (Υ) 120 dB
The efficiency of power amplifier, (η) 0.40
Symbol time, (τ ) 0.5 × 107 (s/symbol)
Static power consumption, (C0 ) 10 W × τ (J/symbol)
UEs circuit power (C1 ) 0.1W × τ (J/symbol)
BS antenna circuit power (D0 ) 0.2W × τ (J/symbol)
Signal processing coefficient, (D1 ) 1.56 × 10−10 (J/symbol)
Coding/decoding/backhaul power, (A) 1.15 × 10−9 (J/bit)
 2
Noise variance, σ 10−20 (J/symbol)

Figure 13.5 Energy efficiency of various precoding schemes under different BS antennas.

energy efficiency rises to 6.36 Mbit/J at M = 140 and then reduces gradually
as the MRC precoding scheme mitigates only intracell interference.
294 Energy Efficiency Investigation in Massive MIMO Communication

13.14 Power Amplifier Power of Different Precoding

Schemes
Figure 13.6 impacts the network’s power consumption effects for various
precoding schemes under different BS antennas. The power utilized by
the precoding scheme incorporating superimposed pilots is comparatively
high than that of the precoding scheme incorporating regular pilots. Since
a superimposed pilot employs twice the power of a traditional pilot for
channel estimation and signal processing. The RF power consumed by the
P-ZF precoding scheme is considerable because it estimates the channel
with B pilot signals and utilizes the ability to mitigate intercell and intracell
interference.

Figure 13.6 Total power amplifier power of various precoding schemes under different BS
antennas.
 13.14 Power Amplifier Power of Different Precoding Schemes 295

13.14.1 Average sum rate of various precoding schemes under

different BS antennas
The average rate per UE under various precoding schemes is shown in
Figure 13.7. The P-ZF precoding scheme estimates the channels with pilot
signals across different B cells with the BS antenna and transmits data streams
in the network. Hence, it achieves a comparatively higher sum rate than
other precoding schemes. A precoding scheme employing a superimposed
pilot accommodates pilot sequences in large numbers for channel estimation
and allows more data streams for transmission across different BS stations.
The ZF precoding scheme uses τp coherence block samples, reducing the
channel estimate quality and achieving a relatively low average sum rate
per UE.

Figure 13.7 Average sum rate for various precoding schemes under different BS antennas.
296 Energy Efficiency Investigation in Massive MIMO Communication

13.14.2 Performance of various precoding schemes for different

UEs
Figure 13.8 portrays the performance of various ranges of UEs in the network.
When K = 47, the UEs EE incorporating the P-ZF scheme rises up to 19.6
Mbit/J and decreases. As the P-ZF precoding scheme estimates more cells,
it includes more UEs and improves the performance than other precoding
schemes. When K = 48, the energy efficiency of UEs employing SP-ZF
increases to 15.53 Mbit/J, and when K= 42, ZF, energy efficiency rises to
12.13 Mbit/J and then decreases gradually. The SP-MRC EE rises to 7.44
Mbit/J at K = 22, and MRC energy efficiency upsurges to 6.36 Mbit/J at K
= 20 and then drops. Superimposed pilot incorporates more pilot samples for
channel estimation, employs more UE, and comparatively enhances energy
efficiency than regular pilots in the network.

Figure 13.8 Energy efficiency of UEs in the network.

 13.14 Power Amplifier Power of Different Precoding Schemes 297

13.14.3 Average rate per UEs using different precoding

schemes for various coherence block length
Figure 13.9 investigates UEs average rate for different precoding at various
coherence block lengths. The result infers that the average sum rate per
UE increases gradually with coherence block length increase. Precoding
scheme incorporating SP uses entire τc coherence block samples for both
data and pilot transmission, which remains static for the whole duration and
comparatively achieves a high sum rate per UE. The P-ZF and ZF precoding
schemes use τp and τc samples for transmission, which limit the range in
the coherence block. This makes the pilots to be reused in the conventional
precoding scheme, which is defined in the precoding scheme using SP.
Thus, the SP precoding scheme outperforms the traditional precoding scheme
regarding the average rate per UE. But P-ZF mitigates both the interferences
and achieves an equivalent sum rate than that of the SP precoding scheme.

Figure 13.9 Energy efficiency of the network for different coherence block length.
298 Energy Efficiency Investigation in Massive MIMO Communication

13.14.4 Energy efficiency for various levels of SNR

Figure 13.10 implicits the sum rate per UE of different precoding for various
SNRs in mMIMO network. Average sum rate per UE is obtained using the
term SINR. When the network interference vanishes, the performance of the
5G network improves gradually. The P-ZF precoding scheme can mitigate
both the interference and achieves high SNR compared to other precoding
schemes. The average sum rate per UE for precoding scheme using SP
increases with increased SNR. The precoding scheme using SP utilizes the
τc samples of coherence block length for data transmission, achieving high
SNR. The conventional precoding scheme operates the uses τp out of τc
symbols for data transmission, and hence it achieves a relatively low SNR.

Figure 13.10 Average sum rate per UE for different SNRs in the network.
 References 299

13.15 Conclusion and Future Scope

This chapter examines how precoding schemes can improve 5G network
energy efficiency by reducing network interference. The 5G Internet of
Things industry automation-based network uses several precoding strategies
to test its performance. P-ZF, along with other precoding schemes, saves
more energy when a large number of BS antennas are used, according to the
research. It reduces interference and improves mMIMO network performance
compared to other precoding strategies. A precoding scheme that superim-
posed multiple pilots investigated 5G network interference caused by pilot
contamination. SP in precoding schemes with longer coherence blocks can
reduce network interference. It improves communication channel estimation
methods and reduces pilot interference, reducing pilot contamination. P-
ZF reduces both types of network interference and achieves high energy
efficiency for 5G IoT industry automation. SP-ZF only reduces intercell
interference. The following potential areas that might be interesting for
researchers to pursue and explore in the future are efforts that can be made
to use MMSE precoding in 5G system-based IoT industry automation and
interference reduction with a power control technique. Further, industry
automation performance can be improved by employing energy-efficient
algorithms and better decoding algorithms. Further, 5G network-based energy
efficiency investigation would be necessary considering hardware impairment
techniques that are not utilized as system parameter optimization.

References
[1] C. J. Anumba, X. Wang, Mobile and Pervasive Computing in Construc-
tion, John Wiley & Sons, Technology & Engineering, 2012
[2] D. Tse and P. Viswanath, “Fundamentals of Wireless Communication,”
Cambridge University Press, Cambridge, UK, 2005.
[3] [Link], M. Kountouris, R. W. Heath Jr., C. B. Chae, and T.
Sälzer, “Shifting the MIMO paradigm,” IEEE Signal Processing
Magazine,vol.24,no.5,pp.36-46, Sep.2007
[4] T. Rappaport, Wireless Communications: Principles and Practice,
Prentice-Hall, Englewood Cliffs, NJ, 1996.
[5] [Link], J. Romero, and [Link], GSM, GPRS, and EDGE Perfor-
mance: Evolution Towards 3G/UMTS. New York, Wiley, 2003.
[6] K. R. Santhi, V. K. Srivastava, G. Senthil Kumaran, and A. Butare,
“Goals of true broad band’s wireless next wave (4G-5G),” in
300 Energy Efficiency Investigation in Massive MIMO Communication

Proceedings of IEEE 58th Vehicular Technology Conference, Orlando,

FL, USA, vol. 4., pp. 2317–2321, Oct. 2003.
[7] B. Furht and S. A. Ahson, Long Term Evolution: 3GPP LTE Radio and
Cellular Technology. Boca Raton, CRC Press, 2009.
[8] S. Sesia, I. Toufik, and M. Baker, Eds., LTE: The UMTS Long Term
Evolution. John Wiley and Sons, 2009.
[9] C.-X. Wang, F. Haider, X. Gao, X.-H. You, Y. Yang, D. Yuan, H.
Aggoune, H. Haas, S. Fletcher, and E. Hepsaydir, “Cellular architecture
and key technologies for 5G wireless communication networks,” IEEE
Communication Magazine., vol. 52, no. 2, pp. 122–130, Feb. 2014.
[10] T. V. Chien, and E. Bjornson, “MMIMO Communications,” in 5G
Mobile Communications, pp. 77-116, Springer, 2017.
[11] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia,
O. Queseth, M. Schellmann, H. Schotten, H. Taoka, H. Tullberg, M.
A. Uusitalo, B. Timus, and M. Fallgren, “Scenarios for 5G mobile
and wireless communications: the vision of the METIS project,” IEEE
Communications Magazine, Vol. 52 , Issue 5, pp. 26 – 35, May 2014.
[12] T. L. Marzetta, “Non-cooperative cellular wireless with unlimited
numbers of base station antennas,” IEEE Transactions on Wireless
Communications, Vol. 9, Issue 11, pp. 3590–3600, November 2010.
[13] T. L. Marzetta, E. G. Larsson, H. Yang, and H. Q. Ngo, Fundamentals
of MMIMO. Cambridge Press, 2016.
[14] Tamilarasan, Ananth Kumar, Suresh Kumar Krishnadhas, Sundaresan
Sabapathy, and Arun Samuel Thankamony Sarasam. “A novel design
of Rogers RT/droid 5880 material based two turn antenna for intracra-
nial pressure monitoring.” Microsystem Technologies 27, no. 9 (2021):
3579-3588.
[15] Rajesh, R. S., G. Tamilselvia, and P. Sivananainthaperumala. “A New
PAPR reduction Technique based on Precoding and Gamma Compand-
ing Technique for OFDM system.” Scientific Journal of PPI-UKM 3, no.
4 (2016): 182-186.
[16] A. Yazdan, J. Park, S. Park, T. A. Khan, R. W. Heath, “Energy-Efficient
MMIMO: Wireless-Powered Communication, Multiuser MIMO with
Hybrid Precoding, and Cloud Radio Access Network with Variable-
Resolution ADCs,” IEEE Microwave Magazine, Vol. 18, Issue 5, pp.
18-30, August 2017.
[17] K. N. R. S. V. Prasad, E. Hossain, and V. K. Bhargava, “Energy Effi-
ciency in MMIMO-Based 5G Networks: Opportunities and Challenges,”
IEEE Wireless Communications, Vol. 24, Issue 3, pp. 86-94, June 2017.
 References 301

[18] Harri Holma, Antti Toskala, LTE for UMTS - OFDMA and SC-FDMA
Based Radio Access, John Wiley & Sons 2009.
[19] Moray Rumney, LTE and the Evolution to 4G Wireless: Design and
Measurement Challenges, Agilent Technologies Publication 2009.
[20] A. Paulraj, R. Nabar & D. Gore, Introduction to Space-time Communi-
cations. Cambridge University Press, 2003.
[21] H. Ngo, E. Larsson, and T. Marzetta, “Energy and spectral efficiency of
very large multiuser MIMO systems,” IEEE Transactions on Wireless
Communications, vol. 61, no. 4, pp. 1436–1449, February 2013.
[22] E. Bjornson, L. Sanguinetti, J. Hoydis, M. Debbah, “Optimal Design of
Energy-Efficient Multiuser MIMO Systems: Is MMIMO the Answer?”
IEEE Transactions on Wireless Communications, vol. 14, no. 6, pp.
3059-3075, June 2015.
[23] S. Wang, Y. Liu, W. Zhang, and H. Zhang, “Achievable Rates of
Full-Duplex MMIMO Relay Systems Over Rician Fading Channels,”
in IEEE Transactions on Vehicular Technology, vol. 66, no. 11, pp.
9825-9837, Nov. 2017.
[24] D. Mi, M. Dianati, L. Zhang, S. Muhaidat, and R. Tafazolli. “MMIMO
performance with imperfect channel reciprocity and channel estimation
error,” IEEE Transactions on Communications, Vol. 65, no. 9, pp. 3734-
3749, March 2018.
[25] B. M. Lee, “Improved Energy Efficiency of MMIMO-OFDM in Battery-
Limited IoT Networks,” in IEEE Access, vol. 6, pp. 38147-38160, 2018.
[26] J. Hoydis, S. ten Brink, and M. Debbah, “MMIMO in the UL/DL of
cellular networks: How many antennas do we need?” IEEE J. [Link]
Commun., vol. 31, no. 2, pp. 160–171, February 2013.
[27] E. Björnson, M. Kountouris, M. Debbah, “MMIMO and Small Cells:
Improving Energy Efficiency by Optimal Soft-Cell Coordination,” Pro-
ceedings of International Conference on Telecommunications (ICT),
Casablanca, Morocco, May 2013.
[28] C. Li, J. Zhang, and K. B. Letaief, “Throughput and Energy Efficiency
Analysis of small Cell Networks with Multi-antenna Base Stations,”
IEEE Transactions on Wireless Communications, Vol. 13 , Issue 5, pp.
2505-2517, May 2014.
[29] X. Gao, O. Edfors, F. Rusek, and F. Tufvesson, “MMIMO performance
evaluation based on measured propagation data,” IEEE Trans. Wireless
Commun., vol. 14, no. 7, pp. 3899–3911, July 2015.
[30] H. Huh, G. Caire, H. C. Papadopoulos, and S. A. Ramprashad,
“Achieving mMIMO” spectral efficiency with a not-so-large number of
302 Energy Efficiency Investigation in Massive MIMO Communication

antennas,” IEEE Trans. Wireless Commun., vol. 11, no. 9, pp. 3226–
3239, September 2012.
[31] . V. Cheng, D. Persson, E. Björnson and E. G. Larsson, “MMIMO at
night: On the operation of mMIMO in low traffic scenarios,” IEEE
International Conference on Communications, pp. 1697-1702, London,
UK, June 2015.
[32] Y. Hao, Q. Ni, H. Li, S. Hou, “On the Energy and Spectral Efficiency
Tradeoff in MMIMO-Enabled HetNets with Capacity-Constrained
Backhaul Links,” IEEE Transactions on Communications, Vol. 65, Issue
11, pp. 4720 – 4733, November 2017
[33] H. Huh, G. Caire, H. C. Papadopoulos, and S. A. Ramprashad, “Achiev-
ing mMIMO” spectral efficiency with a not-so-large number of anten-
nas,” IEEE Trans. Wireless Commun., vol. 11, no. 9, pp. 3226– 3239,
September 2012.
[34] H. Yin, D. Gesbert, M. Filippou, and Y. Liu, “A coordinated approach
to channel estimation in large-scale multiple-antenna systems,” IEEE
[Link]. Areas Commun., vol. 31, no. 2, pp. 264–273, February 2013.
[35] J. Vinogradova, E. Bjornson, and E. G. Larsson, “On the separa-
bility of signal and interference-plus-noise subspaces in blind pilot
decontamination,” in Proc. IEEE ICASSP, pp. 3421–3425, March 2016.
[36] H. Q. Ngo and E. Larsson, “EVD-based channel estimation in multicell
multiuser MIMO systems with very large antenna arrays,” in Proc.
IEEE Int. Conf. on Acoustics, Speech and Signal Processing (ICASSP),
Kyoto, pp. 3249–3252, March 2012.
[37] R. R. Muller, L. Cottatellucci, and M. Vehkaperĺa, “Blind pilot decon-
tamination,” IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp.
773–786, Oct 2014.
[38] H. Yin, D. Gesbert, M. Filippou, and Y. Liu, “A coordinated approach
to channel estimation in large-scale multiple-antenna systems,” IEEE J.
Sel. Areas Commun., vol. 31, no. 2, pp. 264–273, Feb. 2013.
[39] K. Takeuchi, R. R. Muller, [Link]́a, and T. Tanaka, “On an achiev-
able rate of large rayleigh block-fading MIMO channels with no CSI,”
IEEE Trans. Inf. Theory, vol. 59, no. 10, pp. 6517–6541, Oct. 2013.
[40] H. Papadopoulos, G. Caire, and S. Ramprashad, “Achieving large
spectral efficiencies from MU-MIMO with tens of antennas: Location-
adaptive TDD MU-MIMO design and user scheduling,” in Proc. IEEE
ASILOMAR, Nov. 2010, pp. 636–643.
[41] Y. Li, Y. H. Nam, B. L. Ng, and J. Z. Zhang, “A non-asymptotic
throughput for mMIMO cellular uplink with pilot reuse,” in Proc. IEEE
 References 303

Globalcom, Anaheim, CA, USA, pp. 4500–4504, Dec. 2012.

[42] S. Noh, M. D. Zoltowski, Y. Sung, and D. J. Love, “Pilot beam pattern
design for channel estimation in mMIMO systems,” J. Sel. Topics Signal
Process., vol. 8, no. 5, pp. 787–801, Oct. 2014.
[43] K. Upadhya and S. A. Vorobyov, “An array processing approach to
pilot decontamination for mMIMO,” in Proc. IEEE 6th Int. Work-
shop on Computational Advances in Multi-Sensor Adaptive Processing
(CAMSAP), Cancun, pp. 453–456, Dec. 2015.
[44] X. Jing, M. Li, H. Liu, S. Li, and G. Pan, “Superimposed Pilot Optimiza-
tion Design and Channel Estimation for Multiuser MMIMO Systems,”
in IEEE Transactions on Vehicular Technology, vol. 67, no. 12, pp.
11818-11832, Dec. 2018.
[45] J. Ma, C. Liang, C. Xu and L. Ping, “On Orthogonal and Superim-
posed Pilot Schemes in MMIMO NOMA Systems,” in IEEE Journal
on Selected Areas in Communications, vol. 35, no. 12, pp. 2696-2707,
Dec. 2017.
[46] J. Zhang, Y. Wei, E. Björnson, Y. Han and S. Jin, “Performance Anal-
ysis and Power Control of Cell-Free MMIMO Systems with Hardware
Impairments,” in IEEE Access, vol. 6, pp. 55302-55314, 2018.
[47] J. Li, H. Zhang, D. Li and H. Chen, “On the Performance of Wireless-
Energy-Transfer-Enabled MMIMO Systems with Superimposed Pilot-
Aided Channel Estimation” IEEE Journals and Magazines, vol. 3, pp,
2014 - 2027, October 2015
[48] K. Upadhya, S. A. Vorobyov, and M. Vehkapera, “Superimposed pilots
are superior for mitigating pilot contamination in mMIMO,” IEEE
Trans. Signal Process., vol. 65, no. 11, pp. 2917–2932, June 2017.
[49] J. Li, H. Zhang, D. Li and H. Chen, “On the Performance of Wireless-
Energy-Transfer-Enabled MMIMO Systems with Superimposed Pilot-
Aided Channel Estimation” IEEE Journals and Magazines, vol. 3, pp,
2014 - 2027, October 2015.
[50] D. Verenzuela; E Björnson; and L. Sanguinetti, “Spectral and Energy
Efficiency of Superimposed Pilots in Uplink MMIMO” IEEE Transac-
tions on Wireless Communications, Volume: 17, no. 11, Pages: 7099 -
7115, August 2018 (23)
[51] Eduardo n. Tominaga, Onel l. A. López, Hirley Alves, Richard Demo
Souza, and João Luiz Rebelatto, “ Performance Analysis of MIMO-
NOMA Iterative Receivers for Massive Connectivity,” IEEE Access,
Volume: 10, pp: 46808 – 46822, April 2022
304 Energy Efficiency Investigation in Massive MIMO Communication

[52] David López-Pérez, Antonio De Domenico, Nicola Piovesan, Geng

Xinli, Harvey Bao, Song Qitao, and Mérouane Debbah,“A Survey on
5G Radio Access Network Energy Efficiency: MMIMO, Lean Carrier
Design, Sleep Modes, and Machine Learning” IEEE Communications
Surveys & Tutorials, Vol: 24, Issue: 1, pp:653 - 697, January 2022
[53] N. Fatema, G. Hua, Y. Xiang, D. Peng, and I. Natgunanathan, “MMIMO
Linear Precoding: A Survey,” IEEE Systems Journal, Vol. 12, Iss. 4, pp:
3920–3931, Dec. 2018
[54] J. Zhu, R. Schober, and V. K. Bhargava, “Secure downlink transmission
in mMIMO system with zero-forcing precoding,” in Proc. 20th Eur.
Wireless Conf., Barcelona, Spain, pp. 1–6, January 2014.
[55] X. Gao, O. Edfors, F. Rusek, and F. Tufvesson, “Linear precoding
performance in measured very-large MIMO channels,” in Proc, of IEEE
Veh. Technol. Conf., San Francisco, CA, USA, pp. 1–5, April 2011.
[56] T. Parfait, Y. Kuang, and K. Jerry, “Performance analysis and compari-
son of ZF and MRT based downlink mMIMO systems,” in Proc. 6th Int.
Conf. Ubiquitous Future Network, Shanghai, China, pp. 383–388, June
2014.
[57] E. Bjornson, E. G. Larsson and M. Debbah, “MMIMO for Maximal
Spectral Efficiency: How Many Users and Pilots Should Be Allocated?”
IEEE Transactions on Wireless Communications, vol. 15, no. 2, pp.
1293-1308, February 2016.
[58] D. Verenzuela; E Björnson; and L. Sanguinetti, “Spectral and Energy
Efficiency of Superimposed Pilots in Uplink MMIMO” IEEE Transac-
tions on Wireless Communications, Volume: 17, no. 11, Pages: 7099 -
7115, August 2018
[59] A. Mezghani and J. A. Nossek, “Power efficiency in communication
systems from a circuit perspective,” in Proc. IEEE Int. Symp. Circuits
and Systems (ISCAS), pp. 1896–1899, March 2011
 14
Internet of Things and Cybersecurity
Mechanism for Industrial
Automation Systems

Shitharth Selvarajan1 , Gouse Baig Mohammed2 , Mohamed Sirajudeen

Yoosuf3 , D. Shivaprasad3 , Mustapha Hedabou4 , and C. K. Yogesh5
1 Cyber Security & Digital Forensics, School of Built, Environment,
Engineering and Computing, Leeds Beckett University, LS1 3HE
Leeds, U.K.
2 Department of Computer Science & Engineering, Vardhaman College of

Engineering, India
3 School of Computer Science and Engineering, VIT-AP University, Andhra

Pradesh, India
4 School of Computer Science, University Mohammed VI Polytechnic,

Morocco
5 School of Computer Science and Engineering, VIT Chennai Campus

Email: [Link]@[Link]; gousebaig@[Link];

[Link]@[Link]; siva.22phd7130@[Link];
[Link]@[Link]; yogesh.c@[Link]

Abstract
Cyberattack on our country’s infrastructure could be easily prevented if we
had access to our personal information or the information of our own business
network communities. IT professionals and networked businesses face an
infinite number of cyberattacks. Machine-to-machine (M2M) communica-
tion, used to link and transport data from one computer to the other, is a
common use of the Internet of Things (IoT). However, cybersecurity threats
are also being communicated so an attacker can access and follow the data.
Manufacturing processes and efficiency can be improved by implementing

305
306 Internet of Things and Cybersecurity Mechanism for Industrial Automation

the Industrial Internet of Things (IIoT) idea. Existing hierarchical models

must be converted to a fully connected vertical model to accomplish this. IIoT
is a novel method, and as such, the ecosystem is vulnerable to cyberthreat
vectors and challenges with standardization and interoperability. New com-
munication models and technologies are required to accomplish the needed
levels of data security in the IIoT M2M. These include 5G, TSN ethernet, self-
driving networks, etc. Malicious actors may take advantage of system flaws
caused by the faulty implementation of security standards if no measures
are in place to assess the risks and vulnerabilities. A cybersecurity project
for Industry 4.0 is presently underway, and the findings in this report are
based on that work. Converged/hybrid cybersecurity standards are explained
in this research, and best practices are reviewed. A roadmap for identifying,
aligning, and implementing the correct cybersecurity standards and tactics for
protecting M2M communications in the IIoT is also provided.

Keywords: cybersecurity, IoT/M2M, Industry 4.0, Industrial IoT, threats

14.1 Introduction
Cybersecurity has been a significant subject for as long as the Internet has
existed, and its relevance has only grown in recent years due to a greater
reliance on IT infrastructure for telecommuting, online entertainment, and
online business, to mention a few significant areas. In recent years, the
amount of data transmitted over networks has increased dramatically due to
faster internet connections, increased users, and increased number of devices
[1]. Businesses and society might lose much money if there are interrup-
tions in data flow. Organizations are under constant threat from increasingly
sophisticated and complex security threats and assaults, many of which
make use of automation. For security specialists, the time-consuming task of
patching and analyzing systems and identifying dangers can be automated
in order to make systems more secure against a wide range of threats.
There is much space for progress in the automation of complicated security
jobs. Automating security operations ranges from simple scripts to more
complicated methods, such as incorporating machine learning and artificial
intelligence into the software [2].
The privacy of IoT users has also been a hot topic in related industries.
Recent studies have focused on web technologies, big data analytics systems,
NoSQL data storage, data stream management, and RFID technologies (e.g.,
[3]). For example, [4]. These approaches, however, fall short of solving the
new dilemma that IoT ecosystems face. The complexity of data flows between
 14.1 Introduction 307

IoT devices and back-end systems makes it easy for consumers to lose control
of how their data are transported and handled. The lack of control over how
multiple IoT devices’ data are combined to generate new information about
individuals further worsens this situation. On the other hand, we believe that
giving users more control over IoT platform data management is critical.
In response to this problem, we have developed an essential framework to
give users more say over how IoT systems use their personal data. We offer
a unique approach to facilitate the specification of privacy choices governing
data analysis in IoT environments and associated enforcement methods [5].
It is hoped that the suggested framework will help prevent the inference
of sensitive and confidential user information from the aggregation of data
provided by numerous IoT devices belonging to the user. It is vital, in terms of
data management, to let customers specify their privacy choices for how their
personal information is used, which they may do by utilizing well-established
privacy management principles like “purpose,” “duration,” and “disclosure to
third parties.” Because of this, various IoT systems have created a common
IoT platform framework that can be used. By way of illustration, it restricts,
among other things: (1) access to and the aggregation of personal user data;
(2) what analytics algorithms cannot deduce from user data. Because of this,
it is easy for users to create new privacy settings that regulate the processing
of new data acquired via analytics activities. Because of a lack of technical
skills and competence, the ordinary user of an IoT device may not be able to
set their privacy choices. To make the framework even more user-friendly, we
want to add tools that aid users in customizing privacy settings in the future.
Although the Internet of Things has provided numerous benefits, it has also
created problems for those using it [6].
Data security and privacy concerns are front of mind for the researchers
and security professionals interviewed for this article. Two issues have many
companies and government entities perplexed. Cyberattacks on the Inter-
net of Things (IoT) have been demonstrated to be susceptible. There is
a vulnerability in the Internet of Things that needs the implementation of
additional security measures [7]. Security and privacy are the most significant
impediments to the widespread use of the Internet of Things. Unfortunately,
most customers are unaware of the security concerns until there has been a
breach, resulting in huge losses like the theft of sensitive data. Services that
do not sufficiently secure client information have seen a reduction in customer
interest. In a recent study on privacy and security, the Internet of Things made
for consumers did not do well. The safety features of cars today are not as
reliable as they could be. Industry 4.0/Industrial Internet of Things (IIoT) is
a concept for a smart and innovative factory that uses disruptive technologies
308 Internet of Things and Cybersecurity Mechanism for Industrial Automation

Figure 14.1 The study’s scope, structure, and objectives.

(like IoT and cloud computing) and innovative solutions (like IIoT and
automation) to improve the production environment in terms of reduced
costs, increased agility, increased efficiency, remote operations, etc. Data and
network security are critical when operating in a completely autonomous
system. Security threats to IIoT applications and services have widened in
scope, necessitating the development of new controls and safeguards [2–4].
During the pandemic, Internet threats and breaches must not be over-
looked. A breach would be critical in the Industrial Internet of Things
(IoT) arena because of exposures connected to M2M communication and
surroundings [8]. Connected factories rely significantly on 5G and time-
sensitive networking for their communication networks. The linked factory’s
infrastructure is built around self-auto-mating, self-guiding, and self-learning
networks. Future M2M gadgets will incorporate artificial intelligence and
machine learning algorithms to operate autonomously and make decisions.
Because of the potential for disruptive risks, operational security is necessary
for certain sorts of equipment. The IIoT environment is more exposed than
ever to existing security risks because of the significant gap between current
 14.1 Introduction 309

IT and operational technology (IT/OT) domains. Hackers could exploit many

points of entry created by industrial IoT/M2M devices, resulting in the loss
of valuable assets and confidential data. Without sufficient standards and
security measures, it will be difficult to detect the cyberthreat impact and
the information that has been modified. The entire environment [5] is at
risk if the breach is not noticed in time. Data analysis and cybersecurity are
required for advanced use cases such as smart industrial robots and digital
twins to safeguard vertical industries. It is possible to reduce the risk of
Industry 4.0 (I4.0) and industrial control system (ICS) incidents by bridging
the security metrics gap. This will help ensure end-to-end (E2E) security
in ICS. The study’s distinctive features include a thorough examination of
cybersecurity standards, recommendations for designing/converging IT/OT
security architectures, an emphasis on the need to implement interoperable
and hybrid standards connecting various complex interfaces, the bridging of
the IT/OT divide, and a strategic alignment to mitigate IT/OT cyberrisks in
the IIoT. The National Institute of Standards and Technology conducted the
research (NIST). All these elements work together to help readers fully grasp
the topic.

14.1.1 IoT security

This is understandable in light of the growing popularity of the Internet of
Things (IoT). Hackers will soon gain access to your entire network by break-
ing into your coffee machine. Businesses worldwide are more vulnerable to
cyberattacks because of the Internet of Things. The Internet of Things also
raises concerns about data sharing and privacy. When billions of gadgets are
networked, think about the issues that will arise in the future. Businesses
will have to deal with the enormous volumes of data that these gadgets will
generate. Keeping the massive volumes of data collected while allowing easy
access, tracking, and analysis will be a significant challenge. Security and risk
management should not be taken for granted when developing new Internet
of Things applications.

14.1.2 IoT and cloud computing

The Internet of Things encompasses all the devices and services we use daily
linked to the Internet (IoT). Cloud computing, which acts as a front end for
IoT, will also help it succeed. One way cloud computing can help with IoT is
by allowing users to perform normal computer activities using services made
available through the Internet. Because of memory or space limits on their
310 Internet of Things and Cybersecurity Mechanism for Industrial Automation

Figure 14.2 Hierarchy of ransomware families.

computer, an employee may be forced to abandon a large project that needs

to be submitted to their management. If a program is hosted on the Internet,
memory and storage limits can be reduced. Workers can use cloud computing
services to complete their tasks because the data is stored elsewhere. You may
also need to reinstall or reformat your mobile device’s operating system if it
has a problem. Google Photos allows you to share your images online. After
the reformatting or reinstalling process, you may move your photographs
back to your device or access them online at any time [9]. Figure 14.2 present
the ransomware classifications.
Because of this, personal information like the user’s location, contact
information, accounts, and photographs are in danger. Antivirus and malware
detection apps are also missing from most Android smartphones [4, 6].
Sandboxes, access controls, and a signature and authorization mechanism
have been offered. API calls are used to deliver M0Droid’s signatures to
client devices for hazard identification [9]. The author in [10] has developed a
new malware detection method based on application data consumption. Also,
[16] built a system for monitoring the frequency of Dalvik activity codes
on Android, which might be used to detect malicious programs. The author
in [11], introduced Crowdroid framework out of the box for client and server
components [12]. The client uses the Linux trace function for Android system
calls. According to [13], Copper Droid is a framework for detecting and
executing Java code. Previous Android malware detection research has relied
heavily on a small number of features to identify malware [5, 6, 14, 18, 20].
The combination of blockchain and machine learning for intelligent
model training and safe information sharing is gaining traction in the
fight against this dilemma. Because of its high credibility and efficiency,
 14.2 Literature Survey 311

blockchain technology is more famous for exchanging IoT data because it

is more secure and reliable.
The remaining paper is structured as follows: Literature Survey in Section
14.2, security requirements are the focus of Section 14.3, detailed method-
ology and proposed research model discussed in Section 14.4, while the
breadth of cyberthreats attached for IoT are discussed in Section 14.5. For
the sake of this research, Section 14.6 includes data security attacks. Section
14.6 includes cybersecurity challenges in IIoT, followed by Section 14.7,
results and discussion, and Section 14.8, which gives detailed information on
challenges in IIoT, a roadmap to a unified standard framework for minimizing
cyberthreats and standardization challenges that develop in the I4.0/IIoT
environment as a result of IT/OT convergence gaps in Section 14.9 followed
by conclusion and future scope in Section 14.10.

14.2 Literature Survey

Technologies like cloud computing and the Internet of Things should make
it possible to analyze and store data so it can be shared worldwide. Cloud
solutions like Microsoft Azure, Amazon Web Services (AWS), and Google
Docs are expected to provide standard gateways for integrating physical
objects with computing and communication capabilities across various appli-
cations, services, and technologies. There are a number of interesting new
technologies that could change the way ubicomp infrastructure is built shortly
(IoT). Most people think that cloud computing will have something to do
with this [4]. Using big data analytics and cloud computing, the data from
IoT devices will be looked at and figured out.
Also, more and more cloud-based services are adding web-based inter-
faces that make it easier for users to talk to each other and work together.
According to [14], traditional computing and Internet connectivity platforms
must be expanded beyond the typical mobile communication that connects
people. This is necessary for the Internet of Things to work (IoT). This
framework can be used to make connections that are smart and aware of their
surroundings. People think that the Internet of Things will create many data
that will need to be stored, processed, and shown in a way that is easy to
understand. This is because billions of devices are expected to be connected
to the ecosystem. Cloud computing is a must for Internet of Things services
to work with a virtual infrastructure [15]. Data can be tracked and stored,
computations (and analytics) can be done, data can be visualized, and services
can be offered to customers [2]. Gubbi and his coworkers say that the IoT’s
312 Internet of Things and Cybersecurity Mechanism for Industrial Automation

vision is clear when seen from the point of view of both a “Thing” and the
“Internet.” Because it is built around the Internet, the Internet of Things has
changed its focus from Internet-as-a-services to device data. IoT services and
apps that use a thing-centric architecture [16] focus on the connected things
rather than the network that links them.
Many experts agree that networked “smart environment” devices will be
the next big technological step [2]. As we move from www (online static
pages) to www2 (web of social networking) to web3 (web of things), there
will be much demand for data-on-demand searches that are both smart and
easy to use (ubiquitous computing, also known as the web of things). The
post-PC era has begun, and smartphones and other forms of technology have
significantly changed our surroundings and how we talk to each other. The
situation is improving because the format has changed to be more about
teaching and getting people involved. Mark Weiser says that a “new ecosys-
tem” refers to a “smart environment” that has “sensors, actuators, displays,
and computational elements seamlessly integrated into everyday things” [8].
Pervasive computing is growing because of cloud computing and the
Internet of Things. In his article [18], Kevin Ashton says that putting RFID
and other sensors into everyday objects would lead to a new era of machine
awareness and help set up the Internet of Things. Since then, there has been
a lot going on around this idea. Burhan et al. [10] describes the Internet of
Things (IoT) as a group of parts that can be found, addressed, and read by
any other part. This means that services can be sent and received instantly
from anywhere and on any device. Because of this, people, pets, computers,
books, cars, appliances, and even food will all be connected to the Internet of
Everything (Figure 14.3).
The term “Internet of Things” was initially introduced in the annual
ITU Network Reports of 2005 [19]. In the study, the Internet of Things
is described as a scenario where various devices interact intelligently. For
instance, a car alarm can notify drivers of errors, a briefcase can remind its
carrier of forgotten items, and clothing can transmit preferences for colors and
temperatures to a washing machine, among other capabilities. (Figure 14.4)
The European Research Cluster on the Internet of Things (IERC) defines
IoT as an integral aspect of the future Internet with the following features
(Figure 14.5).
The Internet of Things (IoT) has been the subject of numerous studies
since then. Using distributed systems as an example, [20] examined Internet
of Things features and security problems. Issues ranging from identification
and authentication to access control and protocol were covered.
 14.3 Security Requirements 313

Figure 14.3 Futuristic Internet of Everything (IoE) ecosystem [10].

Figure 14.4 Characteristics of a future IoE [21].

314 Internet of Things and Cybersecurity Mechanism for Industrial Automation

Figure 14.5 IERC definition of IoT [22].

14.3 Security Requirements

In this section, I will review the previously discussed layers of vulnerability
for the major Internet of Things technologies. In addition, I examine the most
typical IoT scenarios and the issues they raise regarding ensuring user data
privacy in these settings. In order to address some of these concerns, I then
look at some of the most significant security measures that other academics
have recommended. Finally, I discuss how blockchain technology might be
used to safeguard IoT data. All of these ideas are broken down into four
primary categories: securing IoT using blockchain, IoT vulnerabilities, IoT
security measures, and more [6, 21, 23, 24]. There has been a tremendous
increase in assaults in recent years as the number of connected devices has
increased. We need to protect our privacy when we use electronic devices,
such as smart locks and security systems controlled by the Internet of Things
(IoT). If two or more devices are linked via wireless networks, it makes no
difference how secure the network is. The proliferation of protocols and inter-
faces within the same ecosystem, which makes it impossible to manage them
all, is another difficulty with today’s sensor and connected device ecosystem.
We are likelier to use separate applications to operate many devices if they
are connected. Web, cloud, and mobile interfaces are all part of the problem.
Physical access is the most straightforward method of attacking a gadget. It
is critical to be aware that multiple persons may be using the same device and
 14.3 Security Requirements 315

to use all available safeguards. Many types of cyberattacks can target Internet
of Things (IoT) devices. There are a few things to want to have a safe Internet
of Things:

a. Authentication: It is the procedure used to verify that a person truly

is who they say they are. It is one of the most pressing issues in the
Internet of Things because of the sheer volume of connected devices.
Figuring out if someone or something is who or what they claim to be.
When it comes to IoT, many entities must be authenticated for these
interactions (such as service providers, devices, people, and processing
units). In order to keep corporate networks safe, all devices must be
authenticated. Authorization is the act of enabling someone else to do or
have something that they want. Based on the user’s unique identification,
it grants them access to specific resources within the system. In most
security systems, the first step is to authenticate the user, and the second
is to authorize them.
b. Integrity: Techniques for limiting access to data modification to those
granted permission. Because the Internet of Things relies on many con-
nected devices sharing data, the data must be kept secure. Protection of
integrity also includes safeguarding against sabotage. The IoT system’s
fault tolerance and resilience are essential determinants of data integrity.
Password-protected systems keep your data safe.
c. Availability: Internet of Things (IoT) users should have access to all
data at all times and from any location. IoT systems must maintain their
desired levels of availability and meet their applications’ performance
requirements.
d. Privacy: Maintaining the information in secure locations and employing
effective security measures is the goal of this requirement. Many privacy
concerns have surfaced in the IoT’s rapid expansion over the last year,
as well as the fact that many products now fail to provide all of the
warranties users expect.
e. Confidentiality: It is critical that the data is only accessible to those
granted permission to use it. These users may include equipment, ser-
vices, people, and other items within the system. Encryption techniques
provide privacy protection.
f. Non-repudiation: An entity’s involvement can be proven through a
variety of methods. It is frequently used for digital signatures, contracts,
and e-mails. Digital signatures, confirmation services, and timestamps
can all be used to ensure non-repudiation.
316 Internet of Things and Cybersecurity Mechanism for Industrial Automation

g. Lightweight solutions are a new type of security measure offered by

IoT devices because of their limited power and processing capabilities.
Communication protocols on a local or vast area network that are
simpler, faster, and easier to maintain is called simplified protocols.
h. Heterogeneity: A wide range of devices and systems can be connected
to the Internet of Things (IoT). Because of this, protocols must be
developed to work in various settings and on a wide range of devices.
The Internet of Things (IoT) intends to connect different networks and
devices. In comparison to other devices, each one works and communi-
cates differently. Other factors like privacy, integration, and identity may
be hampered due to this problem.
The security of devices in the perception layer is critical since these devices
typically lack the memory required for full security technology. At the
nodes where data is transferred to the transport layer, these assaults occur
from external sources. One of the main goals of assaults on this layer is
to disrupt the ability of sensor nodes to identify objects accurately. Intru-
sion detection and wireless encryption are two countermeasures that can
be found in this layer. The primary technologies utilized at the perception
layer, such as RFID(Radio-Frequency IDentification), WSN (Wireless Sensor
Network), and NFC(Near Field Communication) are examined for security
vulnerabilities.

14.4 Methodology
The Internet of Things (IoT) is becoming a powerful amplifier for cyber-
attacks as it gains popularity [7]. Their vulnerability is the weakest link
in today’s computer networks’ security chain. According to an HP poll,
IoT gadgets are vulnerable to hackers and identity thieves because they
have simple or no passwords and use unencrypted network services. Mali-
cious software (malware) and private information can easily be installed on
unprotected IoT devices in a vulnerable environment [12]. Malicious botnets
significantly threaten the IoT (Internet of Things). Botnets are compromised
computers exploited by cybercriminals for malicious purposes on the Internet
[13]. Malware known as Mirai [8] was responsible for Distributed Denial
of Service(DDoS) attacks against Krebs on Security and Dyn provider,
which generated more than 1 Tbps of network traffic. AutoBotCatcher uses
dynamic analysis based on network traffic flow to detect botnets in IoT
device communities. Agents and block generators are crucial players in a P2P
 14.4 Methodology 317

Figure 14.6 AutoBot Catcher system flow.

botnet detection approach for the blockchain-based Internet of Things (IoT).

(Figure 14.6)

14.4.1 Privacy and technology knowledge in IoT

As part of a person’s right to privacy, they can choose who can see their
personal information [9]. Identity privacy, data privacy, attribute privacy, and
task privacy are some of the many things that privacy can cover [10, 11].
When dealing with data that can be changed, deleted, or added to, there
are worries about the security of information that can be used to identify a
person. “Privacy” includes more than just user-initiated actions and data. It
also includes technologies like crowd-sensing, which uses devices to collect
location data without the user’s permission [12], and the unexpected use of

Figure 14.7 Proposed research model.

318 Internet of Things and Cybersecurity Mechanism for Industrial Automation

personal data [13]. As people learn more about privacy and security issues,
they are putting more pressure on device makers to fix those [14]. Privacy-
protecting and pseudonymous authorization frameworks that allow people to
own and govern their data are decentralized. A lack of security can jeopardize
an IoT application’s integrity and existence. By their very nature, Internet
of Things (IoT) devices are delicate. Many products and gadgets in an IoT
ecosystem are connected and share sensitive data. Sensitive data and device
communication must be safeguarded in such a setting. More than one million
devices are evaluated for compliance with necessary digital certificates in
the IoT. Devices can be identified digitally, making it impossible to manip-
ulate the blocking. It is also possible to increase scalability by dynamically
updating device information. Incorporating blockchain technology can make
distributed data exchanges more secure [4]. Regarding privacy and security,
research shows that blockchain technology can connect IoT devices.
In order to ensure the security of the new technology, companies must first
check and comprehend it, thoroughly examine the security risks involved,
and devise strategies for resolving them. The author addresses cloud com-
puting security challenges and hazards in this paper and outlines methods an
organization may take to mitigate these risks and protect its resources. Cloud
computing has its strengths and limitations, as well as its applications in risk
management [58].

14.5 Cyberthreats on IoT

In recent years, there has been an increase in malicious software such
as malware, spyware, phishing attacks, ransomware, and zero-day attacks.
Because of this, the influence of these threats and risks on new communi-
cation models and standards needs to be evaluated. M2M communication in
IIoT/I4.0 depends on particular specifications such as URLLC(Ultra Reliable
Low Latency Communications), mMTC (Massive Machine-Type Commu-
nications), and eMBB (Enhanced Mobile Broadband) [25], which provide
real-time data with low latency and good reachability in a low-latency fash-
ion [1]. Guidelines for cybersecurity should be thoroughly comprehended
and implemented to lessen the impact of these dangers [27, 28]. However,
only a few industries have been able to implement them successfully, and
even fewer are aware of their company’s level of security maturity concern-
ing the risks currently facing their industry. All three aspects of improved
ICS(Integrated Communications System) cybersecurity—detection, predic-
tion, and prevention—depend on these actions. ENISA(European Union
 14.5 Cyberthreats on IoT 319

Agency for Network and Information Security) threat landscapes [2] evaluate
the risks and vulnerabilities associated with IIoT across 5G networks. This
assessment also considers the network resources and the sensitivity of the
IIoT. When you use this model, you are provided with a more in-depth
technical view of the 5G architecture, sensitive assets, cyberthreats that
influence those assets, and threat agents. The information used to create the
threat environment came from sources open to the public. These sources
included 5G standardization groups and bodies (such as ETSI(European
Telecommunications Standards Institute (ETSI)) and 3GPP) [26] and 5G
investors such as mobile carriers, technology providers, and national and
international organizations.

14.5.1 Device attack

Hacking the Internet of Things devices is possible with this method of attack.
Its primary purpose is to make the system’s architecture more vulnerable
(depending on the devices involved). If an attacker successfully compromises
the base device, they can take down the entire RFID-powered inventory
control system (e.g., the server is the target). As a result of a device attack,
disruptions to the neighborhood area network (NAN) of an electrical grid
could lead to distributed denial-of-service attacks on the entire grid. Device
attacks can be caused by various malfunctions at the middleware layer of the
operating system, including IP configuration errors, memory corruption, and
incorrectly executed code, to name a few.

14.5.2 Network attack

In this assault, messages are either delayed or lost, compromising inter-
device communication. Network assaults have the potential to devastate IoT
configuration systems’ computational activities. Attacks like these are aimed
at disrupting the functionality of home network equipment, such as cameras
and routers (Home Area Network). Similarly, a similar assault might prohibit
linked devices from accessing important information from other devices in
nearby networks (NANs). SQL injection, denial-of-service assaults, and code
execution are all examples of this type of attack. Table 14.1 shows the vul-
nerabilities in Wireless Sensor Network; Table 14.2 shows the vulnerabilities
in RFID and Table 14.3 shows the vulnerabilities in WiFi.
“Cyber threats” come in many forms, but the most common ones
involve stealing, changing, or destroying data in cyberspace (exploiting data
confidentiality, integrity, and availability). Criminals commit these crimes
320 Internet of Things and Cybersecurity Mechanism for Industrial Automation

Table 14.1 WSN vulnerabilities (main threats and countermeasures).

WSN vulnerabilities
Main threats Countermeasures
Wormhole Synchronized clocks are used. A combination of directional antennas
and multidimensional scaling is required.
Sybil Verification of location and economic incentives are all part of a trusted
certification. Random essential dispersal and testing of resources.
Spoofing A WSN authentication protocol and data encryption should be imple-
mented.
Ping flood Perimeter adjustments to a firewall detection of IDS are required (Intru-
sion detection system).

Table 14.2 RFID vulnerabilities (main threats and countermeasures).

RFID vulnerabilities
Main threats Countermeasures
DoS attacks Incorporate a tighter mechanical link between tags and merchandise.
Assign an alarm function to tags that are in use.
Eavesdropping Encryption of the tag-reader communication.
Skimming Incorporation of blockchain-based monetary instruments.
Replay attack Schemes that are countered schemes based on the passage of time.
Side channel Reduce the amount of data that is made public. Remove the connection
attack between the information that’s been released and the information that’s
been kept concealed.

Table 14.3 Wi-Fi vulnerabilities (main threats and countermeasures).

Wi-Fi vulnerabilities
Main threats Countermeasures
Wi-Fi password cracking, malware installation, WPA2, and web content screening should
be introduced. Data Thieves and Bad Neighbours.

using many different methods, such as eavesdropping, denial of service (DoS)

attacks, viruses and other types of malware, and denial of service attacks
(i.e., hacktivists, state-sponsored, knowledgeable insider, organized crime,
hackers, amateur, etc.). Direct or indirect cyberattacks can threaten industrial
control systems (ICS). However, passive attacks are harder to spot, putting
systems at risk by letting hackers listen in on conversations. This could put
both security and reliability at risk.

14.6 Data Security Attacks

Keeping up with current threats and attacks is a significant challenge for
data security. Public cloud services may require different privacy, security,
 14.6 Data Security Attacks 321

and confidentiality standards for processing information. In order to develop

better security methods to secure data in a cloud computing environment, it
is required to identify potential security threats and assaults. We are simply
thinking about web application data security attacks when it comes to cloud
data security threats and attacks. Using Web 2.0 (the leading technology for
using SaaS in cloud computing) vulnerabilities, attacks are launched against
Web 2.0 sites. Many attacks on cloud data security have been reported.
However, only malware injection attacks on cloud web applications have been
included in this study. In order to cause harm or obtain access to data, malware
injection attacks target cloud systems by injecting a malicious service or
virtual machine into the system. Malicious code is injected into the web appli-
cation as a new instance of service implementation and begins redirecting
legitimate services to the attacker’s instance, allowing the attacker to steal
private session information. In our study of Malware injection attacks on
cloud computing web applications, we primarily focused on the most frequent
attacks, such as SQL injection and XSS(Cross Site Scripting) assaults.

14.6.1 SQL injection attack

In order to perform a SQL injection, security flaws in the software must be
exploited. Attackers take advantage of the vulnerabilities in web servers and
inject malicious code to avoid authentication and obtain access to the database
without authorization. Botnets (private networked computers infected with
malicious software) are commonly used to launch this attack, also known
as a zombie army. If the attacker is successful, he/she can remotely execute
system instructions, retrieve confidential data, change database contents, and
seize control of the web server. This attack is carried out by thousands of
bots equipped with a SQL injection kit, which can be seen in the botnets.
Botnets have infected millions of URLs at various websites throughout the
world using SQL injections. Merchants use SaaS (Software as a Service)
applications to host and sell their products online in the cloud computing
environment. SQL injection can be divided into three different types.

14.6.2 Attack by cross-site scripting

Phishing attacks (fraudulent attempts) are used to obtain data this way. To
carry out this attack, a malicious hyperlink is typically used to acquire data.
An output page appears to be a legitimate piece of content on the website, but
it contains the malicious data initially given to the web application [7]. It is
the most prevalent method of leaking information contained in user cookies,
322 Internet of Things and Cybersecurity Mechanism for Industrial Automation

which can lead to a breach in security. The attacker receives access to the
session information, such as the user’s ID, password, credit card information,
etc. With this attack, an attacker can alter the user’s preferences and take
control of their account, steal cookies, or allow intrusive advertising. It is
possible to run arbitrary code on a victim’s computer if cross-site scripting
is used with other flaws [8]. Malicious pages or parameter values can be the
source of a cross-site scripting attack. It is possible to accurately detect cross-
site scripting assaults by knowing the attack’s source, time, and signatures
[9]. A cross-site scripting attack uses a weakness in cross-site scripting to
bypass the system’s access controls. Many intrusion prevention systems miss
cross-site scripting, even though it is a simple assault to detect. Many security
experts believe it poses a minor risk [8].

14.7 Results and Discussion

Cyberattack has long relied solely on the work of Yandex, Google, and
other counterparts to blacklist fraudulent sites. Machine learning and artificial
intelligence techniques and methods are employed in modern anti-phishing
defences and various other techniques and methods. A wide variety of algo-
rithms are constantly being created and refined in order to identify phishing
sites. In most anti-phishing algorithms, sites are searched and compared to the
original, such as by comparing how the site’s name is written and displayed in
the address bar and the content on the site itself. Thanks to such algorithms,
it is possible to develop browser extensions that alert users to the unreliability
of a site and the likelihood of phishing, as well as spam filters in e-mail
accounts. Users’ credentials are protected by storing their hashes rather than
their passwords to prevent phishing. However, despite tremendous investment
in research and development, no technologies can guarantee 100 percent
protection from phishing attempts, regardless of how much money is spent on
this research and development. Phishing assaults typically involve the victim
in some way, which is why most anti-phishing defences include technical and
social engineering capabilities.

14.7.1 Cyberattack in Industrial IoT devices

Using the experimental data, the significance of the extracted components
of the proposed framework’s suggested framework is proven. Clustering
findings for a noisy and high-dimensional dataset are shown in Figure 14.8.
The red color indicates dangerous samples, and the blue color represents
benign samples over two axes in clustered data after feature reduction.
 14.7 Results and Discussion 323

Table 14.4 Methods of prevention.

Scheme Type Solution
Social engi- Fake ICO Prevent dangerous finan- using bookmarks
neering cial investments by thor- instead of links;
phishing oughly inspecting project installing anti-phishing
paperwork and site traffic. extensions on browsers;
not clicking on links or
downloading suspicious
attachments; verifying
the SSL certificate
before using services;
educating users about
phishing; launching
offline copies of crypto
wallets; using two-
factor authentication;
using complex
passwords (at least
14 symbols); avoiding
public Wi-Fi; and using.
Bloating
Pyramids,
ponzi
Clones Secure mail servers and
databases for employees,
customers, and investors;
monitoring activities on cor-
porate and community web
pages.
Aimed
phishing
Social
networking
Fake
cryptowallets
Technical DNS based Create an alternative to
phishing DNS, such as ENS (the
Ethereum name service).
Hijacking Ensure that the sender and
receiver addresses are cor-
rect.
Malware Attachments should not be
opened or installed.
324 Internet of Things and Cybersecurity Mechanism for Industrial Automation

Table 14.4 (Continued.)

Scheme Type Solution
Key loggers Keep an eye on processes
in the device’s task man-
agement, verify signatures,
use on-screen keyboards,
and keep a password wallet
handy.

Figure 14.8 The benign and malware detection clustering.

The depiction of two distinct clusters is clearly dominated by smoothing

over the subspace region of the feature. On the other hand, it is difficult to
tell the two groups apart because of the point distribution representation.
Hyperplane placement can be seen more clearly when two clusters of data
points are projected onto each other. It is difficult to draw such data pro-
jections since it necessitates an in-depth investigation of each data point
throughout the separation process. Examples of the outcomes of several
sample data projections are shown in Figures 14.9 and 14.10, illustrating that
the technique we recommend effectively distinguishes between benign and
malicious samples.
Malware data was extracted from the blockchain using machine learning
techniques such as clustering and classification. The history of all mal-
ware information can be communicated through the blockchain, allowing
for excellent detection of new viruses. The clustering technique determines
 14.7 Results and Discussion 325

Figure 14.9 Accuracy of classifiers based on machine learning.

Figure 14.10 Comparing false-positive and true-positive rates.

326 Internet of Things and Cybersecurity Mechanism for Industrial Automation

the weights of each feature set and repeatedly removes the unnecessary
features that can be highly efficient in discriminating between malware and
non-malicious programs. In order to attain high classification accuracy and
robustness, the Bayes classifier, which uses a decision tree to deal with multi-
feature problems, is also used. Our system’s permissioned blockchain-based
malware database allows us to detect malware more accurately and precisely
in real time.

14.8 Cybersecurity Challenges in IIoT

This section discusses several security challenges (such as denial of ser-
vice, data theft, manipulation, eavesdropping, etc.) [8–11]. As a result of
unresolved vulnerabilities in the IIoT/I4.0 sector, industrial environments are
vulnerable to security breaches [12, 13].

14.8.1 IT/OT (Data security issues)

Over the past few years, there has been a dramatic rise in the number
of attacks on operational technology (OT) [14, 15]. This is most likely
attributable to the increased flexibility and remote access afforded to new
and old OT systems (such as SCADA, PLCs, etc.) by new interfaces (such
as IT systems, the cloud, etc.). Researchers and policymakers have become
interested in this field of study due to the alarming increase in cyberattacks
against ICS. One example of this increase is the most significant fuel-pipeline
ransomware attack in the United States [16]. Because the IIoT relies on
innovative forms of communication, there is cause for concern regarding the
confidentiality of trade secrets in the manufacturing industry [3]. Whereas
operational technology (OT) is more concerned with the security of industrial
control systems (ICS), information technology (IT) is more concerned with
the security of applications and supporting technologies. Because opera-
tional technology (OT) and information technology (IT) systems do not
communicate with one another, targeted cyberattacks are possible [17].

14.8.2 Cybersecurity threats to the Industrial Internet of Things

Machine-to-machine (M2M) communications rely on cutting-edge networks
like 5G, Wi-Fi 6, and time-sensitive networking (TSN) to realize their
high throughput and low latency goals. Due to the unique communication
types and devices used in IIoT environments, these spaces are open to
 14.8 Cybersecurity Challenges in IIoT 327

Figure 14.11 Cybersecurity taxonomy for IIoT.

various cyberthreats. IoT device endpoints present several security chal-

lenges, including assessment, monitoring, and security (e.g., lack of standard
regulations for all IoT devices from different vendors [9, 10], and trusting
third-party cloud vendors). It is essential to consider a variety of poten-
tial dangers when securing the M2M communication ecosystem. Lack of
328 Internet of Things and Cybersecurity Mechanism for Industrial Automation

M2M communication standardization, malicious IoT nodes, and improperly

configured devices are some dangers (IoT, cloud, edge, etc.). The severity
of the damage caused by compromised M2M devices can be gauged using
a combination of impact type (a threat materializes, compromising confiden-
tiality, availability, and/or network integrity) and impact factor/scale (the total
number of users, the duration of downtime, the number of cells affected, and
the information altered or accessed).

14.9 The Roadmap for Unifying IIoT Standards

A unified approach is required to protect and standardize the fully connected
and intelligent industrial environment. In the Industrial Internet of Things
(IIoT), vulnerabilities can originate from sources other than IT and OT, such
as a lack of essential skill sets, inefficient processes, etc. A unified strategy
developed by the author helps bridge the gaps between various communica-
tion standards and security objectives and other areas that impact the whole
cybersecurity domain. Figure 14.12 presents the interoperability roadmap that
has been developed for I4.0’s unified standards. In order to have an environ-
ment that is both autonomous and interoperable, there needs to be compliance
with a number of different standards (hybrid standards). The same method is
applied to analyzing common metrics used in hybrid standards. This causes
gaps and vulnerabilities to transition from unknown (reactive) to known
(proactive). Heterogeneous network architecture is essential to the Internet
of Things (IoT) and Industry 4.0 (I4.0). A sector’s expansion prospects and
performance indicators could be hampered if forced to rely exclusively on
a single communication network. BP1-CPNI [109], IEC 62351 [12], and
NIST 800-82 [10, 11] are three methods that can be utilized to segment a
network successfully. You can anticipate receiving direction from them when
implementing security standards such as NIST 800-53 or ISO 27001. Since
standards can only be practical if they are selected and evaluated following
a company’s particular goals and objectives, the authors of I4.0 strongly
advise production environment designers to do so for the time being. This
recommendation is based on the fact that standards can only be effective if
they are done so. The level of security that ICS/SCADA and IoT devices
provide should be considered when determining the order of priority for these
types of devices. The traffic flow should be organized according to which
lanes serve the most important functions, and then it should be redirected
to other areas. Cyberattacks that are successful at this layer can result in
denial of service (DoS) and financial and asset losses, but they can also alter
 14.10 Conclusion and Future Scope 329

Figure 14.12 Unified IIoT standards roadmap [23].

the quality of the product or produce defective items that threaten public
safety. Implementing SIEM and SOAR tools at this layer will help you
better understand the reactive strategy for detecting, preventing, and quickly
recovering from attacks or failures involving ICS. [8] Extending or modifying
the table’s dimensions is possible depending on the operational and functional
capacities of the production environment.
The importance of the data and other factors can be considered when
establishing security goals with the help of the unified roadmap for IIoT
standards. Is there a cap on how far the information can travel? To what extent
are your options limited? etc. If the user is familiar with ICS/SCADA and all
it can do, a comprehensive gap analysis encompassing security standards,
communication protocols, and threat intelligence frameworks can be carried
out. Users can accomplish this by using this procedure from beginning
to end.
330 Internet of Things and Cybersecurity Mechanism for Industrial Automation

14.10 Conclusion and Future Scope

Even though many cybersecurity standards and protocols have been put in
place, the IIoT/I4.0 is still vulnerable to a wide range of cyberthreats and
dangers, as mentioned in the studies above. For an IIoT/I4.0 fully connected,
autonomous factory to be built, there needs to be a set of universal standards
that all vendors can use. At the same time, it is essential to fix the fact
that the different control systems and standards do not work well together
regarding security. One way to reach this goal is by making rules that apply
everywhere. To meet each smart factory’s unique service and security needs,
it is essential to create a comprehensive security strategy that considers
the many different cyber standards already in place while also making the
threat landscape less dangerous. A unified roadmap has many benefits, such
as providing a framework for implementing different levels of protection,
(ii) providing alignment between IT and OT, and (iii) providing guidelines
for risk assessment and mitigation of new cybersecurity-based threats in a
heterogeneous production environment. This means that it shows how to
make a lot of different standards work together and how to use them in the
IIoT/I4.0 ecosystem. Protocols like these are used in M2M, 5G, the cloud, the
edge, and many other places.

References
[1] Gouse baig Mohammad and Dr. Prabhakar Kandukuri, “Detection of
Position Falsification Attack in VANETs using ACO,” in the Journal
of International Journal of Control and Automation. 2019; Vol 12, No
6:715-724,
[2] Sicari, S., Rizzardi, A., Grieco, L. A., & Coen-Porisini, A. Security,
privacy and trust in Internet of Things: The road ahead. Computer
Network. 2015; 76: 146–164.
[3] S. Karnouskos, A. W. Colombo, T. Bangemann, “Trends and Challenges
for Cloud-Based Industrial Cyber-Physical Systems,” in Industrial
Cloud-Based Cyber-Physical Systems. 2014; 231-240.
[4] S. L. P. Yasakethu, J. Jiang, “Intrusion Detection via Machine Learning
for SCADA System Protection,” in Proceedings of the 1st International
Symposium on ICS & SCADA Cyber Security Research. 2013;101-105.
[5] L. A. Maglaras, J. Jiang, “Intrusion Detection in SCADA Systems Using
Machine Learning Techniques,” Science and Information Conference
(SAI), London. 2014; 626-631.
 References 331

[6] Gouse Baig Mohammad, [Link], Puranam Revanth Kumar’ Inte-

grated Machine Learning Model for an URL Phishing Detection’,
International Journal of Grid and Distributed Computing, vol. 14, Issue
1, pp. 513-529, 2021.
[7] Joseph, A. J., Sani, N., Kumar, K. S., Kumar, T. A., & Nishanth, R.
(2022, March). Towards a Novel and Efficient Public Key Management
for Peer-Peer Security in Wireless Ad-Hoc/sensor Networks. In 2022
International Conference on Smart Technologies and Systems for Next
Generation Computing (ICSTSN) (pp. 1-4). IEEE.
[8] R. C. B. Hink, J. M. Beaver, M. A. Buckner, T. Morris, U. Adhikari,
S. Pan, “Machine Learning for Power System Disturbance and Cyber-
Attack Discrimination,” in Proceedings of the 7th International Sympo-
sium on Resilient Control Systems (ISRCS). 2014; 1-8.
[9] [Link], Sangeetha, Praveen Kumar, ’Integrated Probability rele-
vancy classification (IPRC) for IDS in SCADA,’ Design Framework for
a wireless network, Lecture notes in network and systems, Springer ,
vol. 82, Issue 1, 2019, pp. 41-64
[10] Burhan, M., Rehman, R. A., Khan, B., & Byung-Seo, K. IoT elements,
layered architectures and security issues: A comprehensive survey.
Sensors. 2018; 18(9),
[11] Chatterjee, S., Kar, A. K., & Dwivedi, Y. K. Intention to use IoT by aged
Indian consumers. Journal of Computer Information Systems. 2018; 1–
12.
[12] Carrion, G., Nitzl, C., & Rold’an, J. Mediation analyses in partial
least squares structural equation modeling: Guidelines and empirical
examples. Partial least squares path modeling. 2017; 173–195.
[13] B. Thirumaleshwari Devi, [Link], ’An Appraisal over Intru-
sion Detection systems in cloud computing security attacks’, 2nd
International Conference on Innovative Mechanisms for Indus-
try Applications, IEEE Explore, ICIMIA -2020, pp. 122 , DOI:
10.1109/ICIMIA48430.2020.9074924
[14] Krishna Keerthi Chennam, Rajinikanth Aluvalu and [Link],’ An
Authentication Model with High Security for Cloud Database’, Archi-
tectural Wireless Networks Solutions and Security Issues, Lecture notes
in network and systems, Springer , vol. 196, Issue 1, 2021, pp. 13-26.
[15] Duan, R., & Guo, L. Application of blockchain for internet of things:
A bibliometric analysis. Mathematical Problems in Engineering. 2021;
1–16.
332 Internet of Things and Cybersecurity Mechanism for Industrial Automation

[16] J. M. Beaver, R. C. Borges-Hink, M. A. Buckner, “An Evaluation of

Machine Learning Methods to Detect Malicious SCADA Communi-
cations,” in 12th International Conference on Machine Learning and
Applications (ICMLA), 2013; vol. 2: 54-59.
[17] N. Erez, A. Wool, “Control Variable Classification, Modeling and
Anomaly Detection in Modbus/TCP SCADA Systems,” International
Journal of Critical Infrastructure Protection. 2015; vol. 10: 59-70.
[18] [Link], [Link], [Link], Abhay Chaturvedi, Pravin R.
Kshirsagar, [Link], ’Grow of Artificial Intelligence to Challenge Secu-
rity in IoT Application,’ wireless personal communication, Springer,
[Link]
[19] M. Hedabou. A Frobenius Map Approach for an Efficient and Secure
Multiplication on Koblitz curves. International Journal of Network
Security, Vol. 3, N. 2, PP.233-237. 2006. (+4)
[20] K. Stefanidis, A. G. Voyiatzis, “An HMM-Based Anomaly Detection
Approach for SCADA Systems,” in IFIP International Conference on
Information Security Theory and Practice. 2016; 85-99.
[21] E. Byres, J. Lowe, “The Myths and Facts Behind Cyber Security Risks
for Industrial Control Systems,” in Proceedings of the VDE Kongress,
vol. 116, 2004.
[22] S. Shitharth, K. M. Prasad, K. Sangeetha, P. R. Kshirsagar, T. S. Babu
and H. H. Alhelou, “An Enriched RPCO-BCNN Mechanisms for Attack
Detection and Classification in SCADA Systems,” in IEEE Access, vol.
9, pp.156297-156312, 2021, doi: 10.1109/ACCESS.2021.3129053.
[23] Sanfilippo, M. R., Shvartzshnaider, Y., Reyes, I., Nissenbaum, H., &
Egelman, S. Disaster privacy/privacy disaster. Journal of the Association
for Information Science & Technology. 2020; 71(9): 1002–1014.
[24] Schneier, B. BOTNETS of things. Most Technology Review. 2017;
120(2): 88–91.
[25] Sharma, M., Joshi, S., Kannan, D., Govindan, K., Singh, R., & Purohit,
H. C. (2020).
[26] Internet of Things (IoT) adoption barriers of smart cities’ waste man-
agement: An Indian context. Journal of Cleaner Production, 270, Article
122047.
[27] Gouse baig Mohammad and Dr. Uppu Ravibabu, “RCAC: A Secure
and Privacy-Preserving RFID based Cloud-Assisted Access Control
to IoT Integrated Smart Home, ”in the International Journal of
 References 333

Innovative Technology and Exploring Engineering (IJITEE). 2019;

8(4): 1-8.
[28] M. Hedabou. Cryptography for addressing Cloud Computing Security,
Privacy and Trust Issues. Book on Computer and Cyber Security : Prin-
ciples, Algorithm, Applications and Perspective. CRC Press, Francis,
and Taylor Publisher. USA, 2018
 15
Illustrating the Mesmerizing Edge AI
Journey for Industry Automation

Pethuru Raj1 and Chellammal Surianarayanan2

1 EdgeAI Division, Reliance Jio Platforms Ltd., India

2 Centrefor Distance and Online Education, Bharathidasan University, India
E-mail: peterindia@[Link]; chellammals@[Link]

Abstract
Edge AI is emerging and evolving fast to be a promising and potential
paradigm for envisaging and implementing scores of real-time and intelligent
applications and services for businesses and people. Edge AI is the cutest and
cognitive combination of two hugely popular domains of edge computing
and artificial intelligence. Edge AI represents the next and new version of
edge data analytics. All kinds of the Internet of Things (IoT) edge devices
interact with one another. This purposeful interaction results in a massive
amount of multi-structured data. It is therefore essential to collect, cleanse,
and crunch edge data in order to extract actionable insights in time. Thus, the
aspect of edge analytics has become an indispensable tool for bringing forth
context-aware, real-time, and people-centric services and applications.
Going forward, the traditional data analytics methods are being advanced
through the surging popularity of AI algorithms (machine and deep learning
(ML/DL)) and models. Highly optimized AI models are being produced
through the training of ML/DL algorithms on datasets (training, validation,
and testing) for automating and accelerating a variety of professional and
personal tasks. So, it is not an exaggeration to state that the AI domain is
being seen as a paradigm shift in the business world. With the accumulation
of IoT edge devices in our mission-critical environments (homes, hotels,
hospitals, etc.), computing, which generally happens in nearby and faraway
cloud server clusters, moves over to IoT edge devices, which are called

335
336 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

networked embedded systems. This fusion of AI and edge computing is being

proclaimed as the most inspiring venture in the ensuing knowledge era.

Keywords: Edge AI, Edge computing, Artificial intelligence, IoT edge

devices, Data analytics, Industry automation

15.1 Delineating the Paradigm of Cloud Computing

There are predominantly two types of IoT edge devices and sensors
(resource-intensive and resource-constrained). Typically, resource-intensive
IoT devices are termed IoT servers/fog devices. There are plentiful IoT mid-
dleware solutions (IoT gateways/hubs/buses/brokers), which are resource-
intensive and hence contribute as IoT edge servers. That is, these IoT edge
devices are primarily for real-time data aggregation, enrichment, encryption,
processing, and storage. On the other hand, there are resource-constrained
IoT sensors, which primarily are leveraged for data capturing from their envi-
ronments and from the assets on which those sensors are embedded internally
and externally. Further on, these IoT sensors transmit the captured data to a
nearby IoT middleware solution to be stocked and analyzed. Thus, the idea
of on-device data processing for generating real-time edge intelligence is
gaining the attention of many these days. The much-discussed concept of
real-time and intelligent computing is gathering momentum with the arrival
and convergence of several cutting-edge technologies and tools. There are
integrated platforms for performing edge data analytics through the leverage
of traditional streaming analytics methods.
In the recent past, the phenomenon of running well-defined machine and
deep learning (ML/DL) models directly on IoT edge servers is attracting
the attention of technical experts. Businesses are also excited about this
transition. It is expected that there will be scores of fresh possibilities and
opportunities. This chapter is fully dedicated to explaining the expanding
scope of edge AI. Before understanding edge AI, it is ideal to have some
details about the currently popular cloud AI.

15.2 Enterprises Embrace the Cloud-first Strategy

There are many state-of-the-art digitization and digitalization technologies
dramatically and distinctively empowering the digital era. The Internet of
Things (IoT) and communication technologies such as 5G fuse well for pro-
ducing connected digitized entities out of all kinds of physical, mechanical,
 15.3 Tending Toward Cloud-native Computing 337

electrical, and electronic systems. When these digitized elements interact

with one another purposefully, a humongous amount of digital data gets
generated. With a multitude of pioneering integration technologies such as
middleware solutions (message brokers, hubs, buses, gateways, adaptors,
queues, etc.) emerging and evolving fast, all kinds of business workloads and
information technology (IT) services are being linked up to give integrated
information and applications.
Clouds represent the highly organized and optimized IT infrastructure.
Further on, clouds are consolidated, centralized/federated, and shared IT
environments with a host of automated tools for meticulously monitoring,
measuring, managing, and maintaining cloud operations. In short, the cloud
idea guarantees much-insisted IT affordability, agility, and adaptivity. The
cloudification aspect supports and sustains the IT phenomenon. The hard-to-
solve challenges and concerns such as IT elasticity, application scalability,
continuous availability, high reliability, heightened security, easy maneuver-
ability, etc., are being simplified and speeded up through the fast-growing
cloud paradigm.

15.3 Tending Toward Cloud-native Computing

Due to the inherent and sustainable strength of the cloud phenomenon,
there are noteworthy evolutions and revolutions being unearthed and rolled
out in the cloud space. Firstly, the wider adoption of microservices archi-
tecture (MSA) and event-driven architecture (EDA) has resulted in event-
driven microservices, which turn out to be the optimal software building
block. Microservices are front-ended with application programming inter-
faces (APIs) thereby microservices are self-contained to be autonomous,
business-centric, and technology-agnostic. Microservices are interoperable,
portable, independently deployable, horizontally scalable, publicly discover-
able, network-accessible, shareable, and composable. Enterprise-class soft-
ware applications are being built and upgraded through a collection of
different microservices. In a nutshell, the microservices architecture has laid
down a stimulating foundation for visualizing and realizing state-of-the-art
software products, solutions, and services. The ready availability of no-code
and low-code application development platforms fast tracks software produc-
tion. The service composition platforms further accelerate the realization of
composite services, which are business-critical, process-aware, and adaptive.
Service mesh is an important module in linking up microservices toward
accomplishing bigger and better tasks in a deterministic fashion.
338 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

Secondly, the tenet of containerization has permeated into the cloud

world. With the surging popularity of containers as the next-generation run-
time environment for software services, applications, platforms, middleware,
and databases, the containerization era has dawned. Especially, containers are
being projected as the most optimal solution for hosting, running, and man-
aging microservices. The combination of microservices and containerization
is being seen as a win-win situation for business enterprises and IT service
providers. Containers guarantee the most important aspect of portability for
software applications. Due to their low memory footprint, microservices-
hosted containers ensure real-time scalability as they can be provisioned in
a matter of seconds. Both stateless and stateful microservices can be run
in containers comfortably. With a deeper understanding, container images
are being formed and deposited in public as well as private image reposi-
tories. Such centralized stores facilitate quick discovery and smart leverage.
There are container-enablement platform solutions to make containerization
penetrative, pervasive, and persuasive.
Thirdly, there is widespread adoption of DevOps practices and tools to
simplify and speed up the challenging task of taking your software services
and applications to production environments. Gradually it has become the
norm for Web-scale companies to deploy their applications several times
per day. Thus, with the growing frequency, the role and responsibility of
DevOps solutions go up sharply. DevOps techniques and tools contribute
immensely to fulfilling the varying requirements for continuous integration,
delivery, deployment, feedback, and improvement. There are optimized pro-
cesses and products toward agile programming. We need agile techniques
to easily and swiftly take developed software to its runtime environment.
Highly competent DevOps toolkits are emerging and evolving to automate
and accelerate the process of deployment and management of containerized
microservices. The critical need of incorporating the security feature into the
DevOps process gets fulfilled through the advanced paradigm of DevSecOps.
Thus the end-to-end software lifecycle management gets accelerated through
agile software production, deployment, and management methods.
Fourthly, the arrival of Kubernetes platform solutions is being seen as a
boon for business houses and IT service providers. Kubernetes is the leading
container orchestration platform. Self-scaling, healing, and other complex
cloud operations are being automated through the power of Kubernetes. The
container lifecycle management tasks are being simplified through Kuber-
netes. In short, the faster maturity and stability of Kubernetes platform
solutions are being seen as a boon in the cloud era. The strategically sound
 15.4 The Edge AI Enablers 339

combination of microservices, containers, container orchestration platforms,

and DevOps toolsets is to shape up the software engineering field.
In the realm of microservices, leveraging a service mesh enhances com-
panies by amalgamating security and operations into a unified infrastructure
layer situated between the containerized application and the network, effec-
tively mitigating risks such as service impersonation, unauthorized access,
and data exfiltration attacks. A service mesh can help manage encryption,
authentication, authorization, policy control, and configuration. And with
role-based access control (RBAC), service mesh supports the zero-trust
philosophy of “trust no one, authenticate everyone.”
The other prominent advancements include site reliability engineering
(SRE) and value stream management (VSM). The operational aspects of
business workloads and IT infrastructure and platform services are very
vital for the ensuing knowledge era. Sometime back, someone predicted that
software eats the world. The continuous availability, high reliability, unbreak-
able security, portability, easy findability, accessibility, manageability and
maneuverability of software solutions are being termed as the most signifi-
cant factors for ensuring real business transformation. The above-mentioned
technologies, tools, and processes come in handy in envisaging and realizing
state-of-the-art software applications, which turn out to be very critical and
crucial for solid and sustainable business empowerment. The non-functional
and tough-to-implement characteristics (portability, interoperability, compos-
ability, scalability, shareability, reliability, and availability) are being hugely
simplified through the adoption of the mesmerizing cloud idea. That is, cloud
applications are being innately provided with such qualities without much
sweat. With this powerful motivation, enterprising businesses, independent
software vendors, and IT service organizations ensure that enterprise-grade
applications are natively built with cloud distinctions. Having understood the
impending needs of business organizations, massive and monolithic appli-
cations are being methodically dismantled and modernized as cloud-native
applications, which are then migrated to cloud environments to reap all the
originally expressed cloud benefits.

15.4 The Edge AI Enablers

The device ecosystem is growing rapidly. There are purpose-specific and
agnostic devices. Devices are becoming miniaturized and multi-faceted.
There are wearables, handhelds, hearables, portables, mobiles, and nomadic,
implantable, fixed, and wireless devices being manufactured in plenty to meet
340 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

up the growing needs of people and businesses. Devices interact with one
another in the vicinity. Also, they get hooked on the Internet and are also
integrated with cloud-based software services, platforms, middleware, and
databases. For every complex and mission-critical device on the ground, a
corresponding digital twin is running in cloud environments. Thus, connected
devices (networked embedded systems) are pervasive and persuasive too.
Through a host of innovations in the information and communication tech-
nologies (ICT) space, connected devices (also termed IoT devices) can be
emboldened to offer pioneering capabilities. This is the first and foremost
step toward the era of edge AI.
Secondly, the deployment of cellular and low-range communication net-
works across the globe has set in motion for everyday devices to provide
low-latency applications to their users. Especially, 5G is regarded as the
one for enabling edge devices to be powerful and path-breaking in their
operations, offerings, and outputs.
Thirdly, the chip industry has gone through a lot of mesmerizing trans-
formations. We have been fiddling with CPUs for a long time. Now we have
high-performance chips to tackle specific requirements. We have graphics
processing units (GPUs), vision processing units (VPUs), tensor processing
units (TPUs), neural processing units (NPUs), etc. These are application-
specific and AI-centric processors. Such hardware accelerators go a long way
in adequately facilitating edge processing and analytics.
Finally, the techniques and tools for producing highly optimized AI
models are fast maturing and stabilizing. Pruning, quantization, federated
learning, knowledge distillation, online learning, etc., are some of the promis-
ing methods for creating and running highly advanced yet lightweight AI
models for tackling various problems such as machine vision and intelligence,
natural language processing, streaming analytics, etc.
A flurry of distinct developments happens in the edge space so that we
will be surrounded by autonomous edge devices. Our places (working, walk-
ing, and wandering) will be decisively stuffed and saturated with AI-enabled
devices, which are self-, surroundings-, and situation-aware. In short, with
the fusion of many cutting-edge technologies, the much-expected paradigm
of edge AI is going to be ubiquitous and a utility.
The recent phenomenon of cloud-native computing is being replicated
in edge devices. Edge devices are typically distributed and heterogeneous.
They follow different data formats and data transmission protocols. Also,
different operating systems are being installed on different devices. To ful-
fill the long-pending demand for service and application portability and
 15.5 Machine and Deep Learning (ML/DL) Algorithms on Edge Devices 341

interoperability, the aspect of containerization is being replicated in edge

devices. By incorporating a container engine, a layer of abstraction is being
incorporated to fulfill the portability need. With more resources being stuffed
in edge devices, multiple containers are being created and run in an edge
device.
Now containers are found to be a highly optimal runtime for microser-
vices. Therefore, microservices are containerized and container images are
being stocked in image repositories so that software developers can use
those containerized microservices. Further on, multiple microservices can
be composed to create process-aware and business-critical applications. For
ensuring high availability and reliability, there can be multiple containers
hosting and running the same microservice. For creating and managing multi-
container composite services, the arrival of container orchestration platforms
for taking care of the end-to-end container lifecycle management activities is
being seen as a positive omen.
DevOps toolkits are emerging and evolving fast to simplify and stream-
line service flows. Data pipelines are also being built to smoothen data
flow from a source to its corresponding sink. Thus, containerization, orches-
tration, microservices and event-driven architectural styles, DevOps, value
stream management (VSM), and other technological paradigms immensely
contribute to elevating edge computing to be an enterprise transformation
method. Edge-native applications are being constructed and released to
sharply enable the penetration of the edge phenomenon.
Going forward, many edge devices can be clustered to form device clus-
ters or clouds. Kubernetes, a market-leading container orchestration platform,
is enabling the setting up and sustenance of edge device clouds, which are
gaining popularity for performing bigger and better tasks. Thus, with the
arrival and accumulation of a myriad of competent and cognitive technolo-
gies, the edge computing model is gathering momentum across industry
verticals.

15.5 Machine and Deep Learning (ML/DL) Algorithms on

Edge Devices
Several years back, we could deploy Apache Edgent, which is an open-source
programming model and runtime for analytics, in edge devices.
eKuiper ([Link] is a lightweight IoT
data analytics/streaming software to be run on edge devices. This streaming
analytics solution can be run on all kinds of resource-constrained edge
342 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

devices. It can migrate cloud-based real-time streaming analytics frameworks

such as Apache Spark, Apache Storm, and Apache Flink to run on edge
devices to facilitate edge analytics. Thus, the shrunken version of data analyt-
ics platforms is being installed in edge devices to extract actionable insights
in time. Now with the surging popularity of AI algorithms (ML and DL),
the data analytics scene is changing forever. That is, ML and DL models are
generated, downsized, and deployed in edge devices to empower edge devices
with all kinds of right and timely insights.
Machine learning algorithms are comparatively smaller in size and hence
running their models in edge devices is quite straightforward. However, that
is not the case with deep learning (DL) models, which are naturally bulky.
Deep learning teaches computers to learn by example.
Deep learning models can achieve state-of-the-art accuracy and even sur-
pass human-level performance. DL models are typically trained on big data
and the neural network (NN) architectures may contain several hidden layers.
Computer vision (CV) and natural language processing (NLP) applications
are being enabled through deep learning algorithms. In certain scenarios,
there is an insistence on real-time detection, recognition, notification, and
action. Data security and privacy besides the need for affordability have been
mandated for running lightweight DL models on edge devices. The current
operating model is to train DL models in cloud servers and take them to edge
devices for inferencing. There are research works initiated and implemented
to perform model training, retraining, and refinement in edge devices. In the
next section, we talk about cloud AI and how edge AI can solve some of
the toughest problems of cloud AI. The widely used ML use cases include
malware detection, business process optimization and automation, predictive
maintenance, spam filtering, etc.

15.6 The Emergence of TinyML

Considering the resource constrictions at the edge, tiny ML models are being
insisted upon. TinyML is all about arriving at lightweight or downsized ML
models. We all know about the budding concept of frugal ML. ML model
optimization, hardware accelerators, architectural styles, energy efficiency,
and other methods are being used to arrive at efficient ML models that can
run comfortably on IoT edge devices. TinyML can perform on-device sensor
data analytics.
With the massive increase in data generation and storage, data pro-
cessing at the edge is gathering speed. Edge AI is being touted as the
 15.8 Cloud AI vs. Edge AI 343

most promising digital transformation technology for enterprising busi-

nesses across the world. There will be fresh edge AI use cases and there
will be more investments in this space by technology giants and pio-
neers. With edge device clusters and clouds, bulkier AI workloads can be
handled.

15.7 The Significance of Edge Security

For the concept of edge AI to thrive, edge security is extremely important.
With connected devices flourishing in important junctions, the quality and
quantity of cyberattacks vary greatly. The connectivity enlarges the attack
surface. For securely accomplishing data analytics through AI models, edge
security turns out to be an important factor and facet.
Edge security is for ensuring unbreakable and impenetrable security for
edge data. Similarly, the security and safety of edge devices are also equally
important. Additional measures are being taken to boost edge security. These
key security measures include the creation of a secure perimeter for tight-
ening access to edge devices and their resources through encrypted tunnels,
firewalls, and access control. Secondly, edge-native applications have to be
comprehensively secured beyond the network layer. Security analytics has
to be consciously performed to predict any impending threats and to identify
breaches proactively. With security insights in hand, it is possible to eliminate
any security implications proactively and pre-emptively. Edge devices have to
be empowered through automated patching to surmount any security attacks
and minimize the attack surface. Vulnerability management is very important
toward edge security. When edge device data gets compromised, then the
decision and conclusion being arrived may go wrong.

15.8 Cloud AI vs. Edge AI

This is an interesting topic to be deliberated deeply and decisively. We have
been running AI models on cloud servers to crunch all kinds of log, opera-
tional, performance/throughput, health condition, security, transactional, and
analytical data. Further on, business, machine, device, sensor, people, IT
application and service, and IT infrastructure data are also being meticulously
gathered, cleaned, stocked, processed, and mined to emit useful information
and insights through the power of AI models. Thus, cloud environments
play a stellar and sparkling role in transitioning data to information and
344 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

to knowledge. Cloud environments are stuffed with a lot of powerful yet

affordable server machines, storage appliances, and network components to
AI-powered data analytics. Clouds store and process big data. Databases,
data warehouses, and lakes are being run on cloud environments to store
and facilitate AI-backed data analytics. Thus, clouds are being positioned
as the next-generation one-stop solution for all kinds of analytics on big
and streaming data. Clouds have both the hardware and software capacities
and capabilities to ensure high-quality data analytics. But there is a problem.
Cloud environments typically fall short and fail/fumble on fulfilling the goal
of real-time data analytics because the network latency is being seen as
the main culprit. In other words, both public and private clouds, which are
situated far away from the place, wherein all eventful things happen, cannot
guarantee producing and supplying real-time insights out of event data.
Here comes the mesmerizing idea of edge AI. That is, engineering,
evaluating, optimizing, and deploying AI models directly on IoT edge devices
is being presented as the workable idea for achieving real-time insights.
That is, the much-celebrated computing moves over to the places wherein
all things happen. In other words, the processing logic reaches to where the
data resides. Previously data moves whereas logic stays. But in the big data
era, data stays whereas logic traverses. AI models are optimized and com-
pressed through a host of powerful methods such as pruning, quantization,
knowledge distillation, federated learning, transfer learning, etc., and then
the models are migrated to edge devices to empower the aspect of local data
analytics. That is, on-device processing for proximate analytics is termed as
the future of knowledge discovery and dissemination. Several breakthroughs
are being unearthed and rolled out in order to empower edge devices to
participate in data storage and processing. AI-centric processing units are
plentifully produced and marketed. AI model optimization techniques and
tools are flourishing with consistent nourishment from AI researchers and IT
professionals spread across the globe. Edge devices are being stuffed with
more memory, storage, and processing power to contribute for the much-
insisted edge analytics. With the dramatic power of 5G communication, edge
computing and analytics have become the new normal. Slowly yet steadily,
the hybrid version of cloud and edge computing paradigms takes over. There
are containerization and serverless technologies and platforms for simplifying
and speeding up the setting up and sustenance of edge device clusters/clouds.
Thus, the faster maturity and stability of multiple path-breaking digitization
and digitalization technologies has come in handy in fulfilling the edge AI
ideals.
 15.9 The Key Motivations of Edge AI 345

15.9 The Key Motivations of Edge AI

The flourishing field of edge computing brings data storage and processing
to the place wherein data gets generated and gathered. This technology-
backed transition paves the way for real-time data capture and crunching.
Low-latency services can be realized through edge processing. The expensive
network bandwidth is being saved. Newer services can be envisaged and
elegantly fulfilled. Computing happens in a distributed and decentralized
manner. There are many digitization and digitalization technologies and tools
such as containerization for ensuring application portability. Microservices
architecture (MSA) orchestrates enterprise-scale applications through the
assembly of decentralized, self-contained, and configurable elements. Further
on, the lightweight versions of the Kubernetes platform are emerging and
evolving fast to speed up the process of setting edge device clouds, which are
famous for ingeniously tackling big data and tasks.
The other noteworthy instigation is the steady generation of massive
amounts of multi-structured digital data. With edge analytics through tradi-
tional analytics platforms and AI algorithms, the aspect of real-time edge
intelligence has been gaining the attention of many in the recent past. As
the quantity of operational, transactional, and analytical data is going up
exponentially, business houses must make use of them for squeezing out
actionable insights in time to produce real-time competencies. Every bit of
data has to be meticulously collected, cleansed, and analyzed to emit viable
intelligence. The data value is great and appreciable if it is subjected to a
variety of deeper and more decisive investigations immediately. The value
of data is bound to go down if it is not processed immediately. Thus, for
the widespread acceptance and adoption of the edge AI paradigm, the ready
availability of data plays a very vital role.
With the mammoth surge in devices (purpose-specific and agnostic), the
quantity of edge data is bound to grow exponentially in the days ahead.
There are multi-faceted edge devices, machinery, drones, robots, medi-
cal instruments, equipment, appliances, wares, utensils, etc., being stuffed
in many of our everyday assets and environments such as manufacturing
floors, warehouses, ports, railway stations, eating joints, entertainment plazas,
hypermarkets, vehicles, crowded locations, tunnels, expressways, defense
areas, aircraft, auditoriums, and other sensitive and secure places. Further
on, purpose-specific and agnostic IoT sensors are being embedded in all
kinds of tangible artifacts in our personal and professional places. Thus,
there are wearables, handhelds, hearables, mobiles, portables, and nomadic,
346 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

wireless, and implantable devices to assist us in our typical assignments and

engagements. When these entities find, bind, and interact with one another
purposefully, a tremendous amount of useful data gets generated and stocked.
Thus, for the paradigm of edge AI to arrive and thrive, the role of edge data
is undoubtedly paramount.
Thus, the vital ingredients empowering the edge AI concept are the
growing device ecosystem, the exponential growth of edge data, the maturity
and stability of data analytics platforms, data stores, knowledge visualization
dashboards, etc.
In a nutshell, the edge AI conundrum carries several business and techni-
cal advantages. The local storage and proximate processing enabled through
edge computing and analytics can take pressure off centralized cloud servers.
Edge data goes through filtering locally so that the repetitive and irrelevant
data gets eliminated at the source. Such a shift has led to saving network
bandwidth. With edge computing in place, data security and privacy along
with the huge savings of network bandwidths are being achieved.
By utilizing AI chips, Edge AI is capable of guaranteeing superior per-
formance. As we all know, AI is destined to empower ordinary systems to
be extraordinary in their actions and reactions. In short, intelligent machines,
devices, and things will flourish everywhere with the clever usage of AI.

15.10 The Edge AI Implications

Edge AI can unlock several benefits for businesses and common people
in the long run. A growing number of hitherto unheard innovations and
disruptions are being unearthed and rolled out for the benefit of end-users,
employees, and executives. Industries are looking out to visualize and realize
fresh possibilities and opportunities with the evolutionary and revolutionary
characteristics of edge AI. The much-anticipated real business transforma-
tions can be availed and fulfilled through the smart leverage of edge AI. The
distinct advancements in machine and deep learning (ML/DL) are directly
and indirectly contributing to the unprecedented success of the edge AI
paradigm.
Experts have pointed out that tactic and strategic benefits are being
accrued out of the fledgling domain of edge AI. Multiple industry verticals
are keenly exploring and experimenting with the power of edge AI. Manufac-
turing, healthcare, telecom, retail, inland security, supply chain, agriculture,
and other verticals are seriously working toward offering edge AI-inspired
 15.11 The edge AI advantages 347

premium and breakthrough services to their consumers and customers.

Digitally transformed homes, hotels, hospitals, warehouses, airports, cities,
etc., are being realized through edge AI.

15.11 The edge AI advantages

The aspect of edge AI is emerging as a new frontier through the seamless and
spontaneous combination of edge computing and artificial intelligence. All
the local and real-time computing capabilities are being fluidly extended to
AI. The prominent advantages of edge AI are as follows

15.11.1 Enabling real-time data capture and analytics

Edge devices can gather their environments’ data and make sense of edge
data quickly. Thus, the long-standing goal of real-time analytics is to emit
real-time insights. With edge AI, there is no need for transmitting edge data
to faraway cloud platforms for storage and processing. Edge AI astutely
contributes to a myriad of tasks such as prediction, classification, asso-
ciation, clustering, regression, recognition, detection, tracking and tracing,
translation, summarization, captioning, etc.

15.11.2 Edge AI supplements with cloud AI

Edge AI takes some of the loads off the cloud platform. Thereby the precious
and expensive network resources get saved. The network-induced latency
comes to nil. The cybersecurity attacks on confidential and personal data
are getting nullified. Edge devices are stuffed with additional modules to
be computational, communicative, and vision-enabled to be sensitive and
perceptive, AI-powered to be decision-making, articulative, and active. In-
memory database systems are being embedded to store a lot of sensitive and
temporal data. With the fast proliferation of heterogeneous, connected edge
devices in our daily environments, it is possible to add additional devices on
demand to fulfill fresh resource requirements. That is, the much-discussed
horizontal scalability is being achieved in the device world also.
Thus, with more resources ingrained into new-generation devices, edge
computing is to become a utility. Edge computing and analytics will become
ubiquitous. Devices join with one another in the vicinity and with remote
devices through networking. It is anticipated that edge devices emerge as
the new IT powerhouse for the ensuing digital era. The scalability and
348 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

versatility of edge devices are being seen as a good omen for the future.
With the voluminous production and deployment of multi-faceted edge
devices, there will be a multitude of edge-native applications to delight
users in the days to unfurl. There are integrated development, deployment,
and management platforms being made available by product, platform, and
tool vendors to take the edge concept to the next level. The promising
cloud-native principles are being replicated in the edge environment to pro-
duce edge-native services, which, in turn, can be composed to realize and
deliver process-aware, business-critical, people-centric, and context-sensitive
applications in real time. It is a fact that machines generate more data
than men.

15.11.3 Transitioning to intelligent machinery and devices

In a nutshell, with the accumulation of high-end and resources-intensive
devices embedded with state-of-the-art machine and deep learning (ML/DL)
models in our working, walking, and wandering places, individuals and inno-
vators can embark on envisaging ground-breaking use cases. Newer services
and applications can be visualized and realized toward business and people
empowerment. Precisely speaking, machines, robots, drones, equipment, and
instruments are all set to be cognitive in their decisions, deals, and deeds. The
edge AI conundrum has the wherewithal to grow and glow across industry
verticals. Establishing and sustaining intelligent and real-time enterprises will
see the reality sooner than later. Further on, smarter homes, offices, stores,
warehouses, clinics, and cities will be quickly set up and sustained. The
ingenuity factor of edge AI is to transform the entire world substantially.
Researchers have proclaimed extensively about the plausible use cases of
edge AI through a bevy of peer-reviewed articles. There are case studies,
well-written blogs, white papers, e-books, webinars, and podcasts explaining
the growing power and promise of the edge AI paradigm.
The edge AI power is being attached in multiple systems, solutions, and
services these days with the sole aim of empowering them adequately to be
deft and decisive in their operations, offerings, and outputs. The renowned
examples include autonomous vehicles, traffic control systems, voice assis-
tant apps, home security products, secure banking services, edutainment
games, metaverse environments, smart manufacturing facilities, humanoid
robots, cognitive drones, etc. Airport security is ensured through edge AI,
which is getting widespread recognition and reward for automating many
manual activities with ingenuity and alacrity.
 15.12 Edge AI – Individual and Industrial Use Cases 349

15.12 Edge AI – Individual and Industrial Use Cases

With a greater and deeper understanding by business executives and IT
pundits, there comes a slew of fresh use cases for individuals, institutions, and
innovators. This new and next-generation paradigm has all the ingredients
and wherewithal to create a lasting impact on many industry verticals. In
this section, we are to focus on a few domains and use cases to convey the
point clearly. Every worthwhile domain is keenly strategizing and planning
to embrace this pioneering paradigm to deliver premium services to their
consumers and customers.

15.12.1 Speech recognition

This is one of the finest use cases of edge AI. Edge devices can recog-
nize human voices and act upon them with clarity and alacrity. The latest
innovations and disruptions in the field of deep learning (DL) have raised a
glimpse of empowering edge devices to recognize human communication.
The recognition accuracy is going up steadily. There are automated speech
recognition tools and libraries gathering attention lately. Today we have
highly advanced speech recognition systems guaranteeing the feature of
speech-to-text transcription going beyond the basic feature of speech-to-text
transcription. Experts have pointed out that this functionality (7 AI features
changing speech recognition | Enable Architect ([Link])) results in a few
interesting use cases.
1. Sentiment analysis – This is to extract and elegantly articulate the
emotions and sentiments in a speaker’s speech.
2. Entity detection – This feature is to identify and categorize noteworthy
entities in a language text.
3. Speaker identification – This capability identifies different speakers in
an audio or video file.
4. Content safety detection – This is for checking content if there is any
hate and harmful information in the content.
5. Personal information removal – This is to identify and remove person-
ally identifiable information (PII).
6. Summarization – This generates a summary of a speech.
There arise several real-world use cases. Voice commands are being rec-
ognized by edge devices and such an emerging capability is being widely
used for making online purchases. Emails are being dictated through voice
instructions. Meeting discussions and court arguments are being noted down.
350 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

Resource-intensive smartphones and digital assistants are empowered with

such unique capabilities. Voice-activated payment systems are getting a lot
of adoption. Voice has become important biometry for security enablement.
AI models are being prepared and deployed in edge devices to receive
and recognize voice commands. Several allied tasks such as background
noise cancellation and speech enrichment are also being addressed through
edge AI.
Speech processing needs high-performing, energy-efficient, and highly
affordable chips. Fortunately, there are AI-specific chips such as GPUs,
VPUs, NPUs, TPUs, etc. Precisely speaking, with the immense contributions
of the state-of-the-art chips and compressed AI models, speech recognition
and synthesis at the edge get accomplished elegantly.

15.12.2 Smart healthcare

Elegant edge devices such as sensors and actuators are being attached to
patients to minutely monitor and measure various body parameters. There are
miniaturized and multi-faceted wearables, hearables, and implantable devices
being built and manufactured in large quantities nowadays. These aid in
precisely tracking the intended patients’ progress. Various decision-enabling
details are continuously captured and cognitively crunched to understand
the factual condition of the patients. Such empowerment helps doctors, spe-
cialists, surgeons, nurses, and other caregivers to correctly diagnose and to
recommend expert medication.

15.12.3 Ambient assisted living (AAL)

This is another formidable requirement for people who are diseased, dis-
abled, and debilitated. Edge devices empowered with AI capability are being
proclaimed as the prominent contributors to ambient assisted living. Patient
monitoring, management, and maintenance are being facilitated through the
edge AI phenomenon. There are several other possibilities being enabled
through the distinct power of edge AI.

15.12.4 Intelligent robots and drones

Increasingly all kinds of mission-critical environments are being stuffed with
robots and drones to automate a variety of repetitive tasks. Robots and
drones are being advanced through software enablement. By running AI
models on these devices, it is possible to have intelligent robots and drones
 15.12 Edge AI – Individual and Industrial Use Cases 351

to accomplish bigger and better things. There is no need for any manual
intervention, instruction, or interpretation to activate AI-enabled drones and
robots. Warehouses, retail stores, healthcare facilities, defense zones, man-
ufacturing floors, oil rigs and fields, tunnel and expressway-building, and
other vital junctions are to benefit immensely through AI-powered drones
and robots. All the noteworthy advancements being realized in the domains
of computer vision (CV) and natural language processing (NLP) are being
embedded in robots and drones to excel in their operations, offerings, and
outputs. In a nutshell, everyday robots and drones are becoming connected,
competitive, and cognitive by expertly leveraging the growing power of edge
AI. So there is no doubt that edge AI is penetrating every sunshine domain.

15.12.5 Predictive maintenance

This is a well-known use case of edge AI. Machinery, equipment, appliances,
utensils, flight engines, IT hardware, and other business-critical instruments
ought to be monitored to ensure their continuous functioning. Any slowdown
or breakdown is to impact business operations and the brand value of service
providers. Thus, the tenet of predictive maintenance has acquired special
attention. Edge AI has the wherewithal to provide such a facility to elongate
the lifeline of all kinds of critical infrastructures.

15.12.6 Edge AI in retail industry

There are online as well as offline stores for retailing and merchandizing.
Edge AI is to bring in a series of automation for retail service providers. Cus-
tomer management, inventory management, and replenishment management
activities are dramatically enhanced through edge AI. Customer delight will
be fulfilled through the smart leverage of the distinct improvisations being
achieved in the edge AI space.

15.12.7 Edge AI for video surveillance system

Real-time detection and counteractions are important for ensuring the tight-
est security and safety for properties and people in crowded places. The
delectable innovations accomplished in computer vision have brought in a
series of advancements. Edge devices are being vision-enabled to see and
understand what is happening around them. Cameras are being stuffed with
CNN and RNN models to gain vision capability and to interact with nearby
machines and men naturally. Thus, the concept of on-device data processing
352 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

to emit edge intelligence has laid down a sophisticated foundation to envisage

and build ground-breaking use cases for business behemoths and start-ups.
Machine vision and intelligence provided by edge AI are to empower airports,
railway stations, nuclear installations, bus bays, warehouses, tunnels, oil
wells, refineries, pipelines, etc. Smart traffic cameras automatically adjust
light timings to nullify traffic snarls and congestion. Autonomous vehicles
facilitate the automated platooning of truck convoys. This is to allow drivers
to be removed from all trucks except the one in front.

15.12.8 Edge AI in smart homes

As we all know, smart homes provide several exciting features and facilities
to homeowners and occupants. Energy management, home security, and asset
management are well-known use cases. Security cameras, video doorbells,
and alarms emit a lot of security-related data, which gets locally collected
and crunched to empower security-enablement systems to act quickly so
that any kind of security breaches can be nipped in the bud. There are
competent AI models for extracting security insights in time. Edge devices
hosting such lightweight AI models guarantee the utmost security for people
and properties. Similarly, entertainment, education, water, garbage, drainage,
and gardening management activities are made smart through edge AI. A
security camera recognizes your face as you walk into your house and it then
automatically adjusts the room temperature, lights, and music as per your
previously set preferences.
With CNN models embedded inside security cameras, familiar faces are
recognized. Otherwise, cameras send an alert message to the concerned if an
unknown person comes in.

15.12.9 Edge AI for banking and financial services

Customer management becomes highly personalized by promoting the right
financial products and services. For example, SoftBank’s Robotic Pepper is
a robot designed for people. It connects with, assists, and shares knowledge
with customers. This robot is friendly and engaging and capable of creating
unique experiences to form real relationships. Pepper interprets and responds
to client information requests using natural language processing, and it can
even detect and respond to basic human emotions. This can process massive
amounts of data in real-time. Thus, edge devices stuffed with competent AI
solutions can be a game-changing affair for service providers and consumers.
Edge AI enhances customer satisfaction.
 15.12 Edge AI – Individual and Industrial Use Cases 353

15.12.10 Edge AI in automobile: Enhancing real-time capabilities

Edge AI is increasingly prominent in autonomous vehicles, smart cities,
process-centric manufacturing, healthcare, and improving monitoring and
management of product assembly lines, etc. Its scope continuously enlarges
across multiple industry verticals. Image, audio, and video analytics happen
at edge devices directly thereby real-time actions can be guaranteed. AI work-
loads are getting moved to edge devices from cloud servers. AI-assisted edge
devices can showcase greater power and efficiency. Low-latency applications
are being attempted through edge AI.

15.12.11 Edge AI is transforming Enterprises

Business operations are being meticulously automated and accelerated
through the leverage of information and communication technologies. With
the wider acceptance and adoption of AI and edge computing technologies,
enterprising businesses are bound to achieve extreme automation. As we
discussed above, every worthwhile industry vertical is methodically explor-
ing and experimenting to know how edge AI is tactically and strategically
beneficial.
As businesses keenly embrace digitization and digitalization aspects to
journey in the right direction toward the designated destination, many praise-
worthy digital innovations and disruptions are being experienced. The idea
of digital intelligence is also acquiring special significance with the solid
improvements in the data analytics space. Therefore, the mission of digi-
tal transformation gets activated and accomplished across several business
domains with all the alacrity and clarity. Edge AI, a breakthrough digital tech-
nology, is getting increased attention as it has the wherewithal to succulently
transform worldwide enterprises in a risk-free and rewarding manner.

15.12.12 Edge AI impacts the IoT

Every commonly found and cheap thing in our personal, social, and pro-
fessional environments is being digitized through a host of digitization and
edge technologies (sensors, actuators, microcontrollers, RFID tags, barcodes,
stickers, beacons, LED lights, smart dust, specks, etc.). With a host of con-
nectivity technologies, local and global networks are being formed out of all
kinds of digitized entities. Such a network of digitized entities turns out to be
game-changing for business houses to dream and deliver big things for their
clients, employees, and end-users. Besides digitizing ordinary objects, IoT
technologies (a combination of digitization and connectivity technologies)
354 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

enable digitized elements to interact with one another in the vicinity and with
remotely held software applications and data stores through networks such as
the Internet. Such purposeful connectivity, collaboration, and correlation can
generate a lot of digital data, which has to be cleanly gleaned and subjected
to a variety of deeper investigations to squeeze out viable and venerable
information. Conventionally, digital data thus produced is taken to cloud
storage and processing systems to produce digital intelligence.
Now with the ready availability of the fabulous edge AI capability, edge
data is being locally collected and subjected to proximate processing to
instantaneously discover hidden patterns in data mountains. Knowledge, thus
discovered, goes a long way in empowering business and IT systems in many
ways. That is, traditional systems are systematically emboldened to exhibit
adaptive behavior in their service deliveries. The bewildering edge AI is being
pronounced as the silver bullet for industrial automation.

15.12.13 Edge AI for smarter homes

Our living spaces are adorned with different kinds of sensors and actuators.
Energy management, security enablement, device management, provisioning
of choice, convenience, care, and comfort for the owners and occupants, etc.,
are being accomplished through edge AI. Homes need such breakthrough
technologies.

15.12.14 Edge AI for smart cities

Cities across the globe are bubbling with population. Drainage, water,
garbage, and utility management ought to be technologically enabled to
smoothen city life for the people. Traffic congestion avoidance is a big need
for city people. Apartments, hospitals, IT company premises, hotels, and
other prominent locations in any city have to be carefully monitored and
managed. Herein, the role and responsibility of edge AI are definitely vital.

15.12.15 Edge AI for smart manufacturing

A variety of manufacturing activities are precisely accelerated and automated
through AI-enabled machines, assembly lines, and industrial assets. Intelli-
gent asset management becomes critical for industry automation. Real-time
edge intelligence elevates factories to a greater height. Business partners,
employees, and customers all will be succulently provided with a bevy of
real-time services.
 15.13 The Distinctions of Edge AI 355

15.12.16 Surveillance
Continuous monitoring and real-time decision-making are essential for deriv-
ing and delivering a number of fresh capabilities. Edge AI opens up new
possibilities and opportunities across industry verticals. This is one of the
well-known use cases for the runaway success of edge computing. Surveil-
lance, proximate data processing, real-time knowledge discovery, and dis-
semination go a long way in safeguarding expensive properties and ensuring
people’s security. AI-enabled CCTV cameras are being liberally used these
days in important junctions, eating joints, entertainment plazas, stadiums,
auditoriums, retail malls, educational campuses, tunnels and expressways,
nuclear installations, etc., to accurately pinpoint problems and activate the
concerned ones to plunge into appropriate countermeasures in time. Any kind
of deviation and deficiency can be detected proactively and proper solutions
can be articulated by predictive and prescriptive AI models.

15.12.17 Mining, oil, and gas and industrial automation

The business value of edge-based ML models (increasingly enabled through
TinyML frameworks) becomes obvious in the oil, gas, and mining industries.
Employees are supposed to work in remote and rough environments. This
is full of risks. Herein, AI-powered robots and machinery come in handy.
These empowered systems capture a variety of decision-enabling data and
make accurate decisions by analyzing the captured and cleaned data instan-
taneously. That is, they work autonomously. For complicated decisions, the
control goes to people put up at remote control centers. Such an arrangement
goes a long way in ensuring the safety and security of people and properties.
Critical assets across industry verticals are being technologically monitored
so that any kind of deviation and disturbance can be pre-emptively pinpointed
and suitable countermeasures can be considered and activated to nullify any
incoming disasters.

15.13 The Distinctions of Edge AI

With the flourishing of IoT edge devices in our everyday environments
such as homes, hotels, hospitals, etc., the data processing steadily moves
over to edge devices. That means edge devices join mainstream computing.
Increasingly edge devices are being embedded with more memory, storage,
and processing power. Analytics platforms are installed in resource-intensive
356 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

edge devices to do proximate data analytics. Especially streaming analytics

is surging in the edge world.
Edge streaming analytics platforms can collect and analyze edge device
data to find useful findings with minimal latency. Herein, there is no need
to transmit edge data to nearby or faraway cloud environments to perform
data analytics. Instead, edge devices with sufficient computing and storage
resources can host a streaming analytics platform to receive and crunch edge
data quickly to emit real-time insights. Thus, the long-standing goal of data-
driven insights and insights-driven decisions is getting fulfilled.
The much-celebrated centralized and shared cloud environments across
the globe are enabling this expectation. Now with edge devices individually
and collectively emerging as the new IT powerhouse, the aspect of data ana-
lytics moves over to high-end edge devices. This shift is capable of bringing
in advanced use cases for people as well as enterprising businesses. Several
start-ups have emerged in the recent past to make use of this strategically
sound phenomenon. In short, any bit of data getting generated is meticulously
gathered and analyzed immediately to create full and long-lasting value out of
data. Due to edge analytics, the data processing happens in real-time thereby
real-time insights are produced and shared to empower the process of build-
ing real-time services and applications. Precisely speaking, edge analytics
through traditional analytical methods and the latest AI algorithms have laid
down a stimulating and sparkling foundation for real-time enterprises. Thus,
running highly optimized and purpose-specific AI models getting installed in
edge devices is to fulfill a dazzling array of real-time and intelligent products,
solutions, and services.
There are several advantages to on-device data processing. Real-time
analytics results in timely insights, which can be fed to actuation systems
to embark on counteractions immediately with all clarity and confidence.
Thus, edge AI stands in contrast to the proven paradigm of cloud AI, in
which AI models are being developed and run entirely on cloud platforms.
Cloud AI enjoys all the originally expressed benefits of the indomitable cloud
paradigm.

15.14 Conclusion
Edge AI is going to be a sophisticated combination of edge computing and
artificial intelligence. The real digital transformation initiatives by business
houses and government organizations are being hugely simplified and spear-
headed through the methodical usage of edge AI. The enabling technologies
 15.15 The Glossary of Terms 357

and tools of edge AI are simply mesmerizing. Product vendors, research labs,
IT companies, and academic institutions are striving hard and stretching fur-
ther to make edge AI pervasive and persuasive. Edge devices are being stuffed
with more memory and storage capacities and processing capabilities. Edge
devices emerge as data-processing entities. With edge devices increasingly
getting ordained with AI capabilities, it is possible to visualize a dazzling
array of next-generation services through edge devices. Edge devices can
form clusters to tackle bigger problems. That is, device clusters/clouds will
become the new normal in the days to come. There are competent technolo-
gies such as lightweight versions of Kubernetes for forming and managing
edge clouds to do edge data analytics. Especially, the prospect of performing
edge analytics by leveraging AI models brightens. Thus, the journey humbly
started with edge computing for on-device data processing. Then came the
transition toward forming and firming up edge clouds for doing complicated
tasks. Edge analytics is one of the crucial requirements for extracting action-
able insights out of edge data in real-time. Now we are heading toward the
idea of edge AI, which is for accomplishing AI-powered analytics on edge
device data.
Thus technological evolutions and revolutions have vividly led to the
realization of applying highly optimized AI models on edge data to unravel
the useful patterns hidden in data heaps.

15.15 The Glossary of Terms

1. Artificial intelligence (AI) refers to a collection of technologies and
tools to analyze (through batch or stream processing) a huge amount of
data to emit actionable insights, which comes in handy in automating
and accelerating business processes. That is, data becomes information
and knowledge. By receiving any discovered knowledge, our everyday
devices including computers, communicators, robots, drones, and other
digital assistants can exhibit adaptivity and ingenuity in their operations,
outputs, and offerings.
2. AI Chips – These refer to high-performing processing units such as
GPUs, VPUs, TPUs, NPUs, etc., to handle AI workloads efficiently. The
usage of general-purpose chips like CPUs can go down in the AI era. AI
chips can complete large-scale computing through parallelization.
3. Bandwidth refers to the amount of information that can be sent
over a connection in a measurable amount of time. This is generally
358 Illustrating the Mesmerizing Edge AI Journey for Industry Automation

measured in kilobits, megabits, gigabits, etc. In the forthcoming 6G

communication, it may reach terabits per second.
4. Cloud environments represent highly optimized and organized IT envi-
ronments comprising hundreds of thousands of IT infrastructure mod-
ules such as computing machines, storage appliances, network solutions,
and security devices. Development, deployment, delivery, orchestra-
tion, integration, security, government, and management platforms are
being installed on cloud infrastructures. Analytical, operational, and
transactional applications are being run on these platforms.
5. Cloud AI is to leverage cloud platforms, databases, middleware, etc., to
train and run AI workloads.
6. A convolutional neural network (CNN) is a type of artificial neural
network (ANN) for computer vision (CV) tasks, which include the
extraction of useful and usable information from images, videos, and
other visual data.
7. Edge computing – Data gets collected and processed at the source of the
data. Instead of sharing the data through risky Internet infrastructure,
edge or local devices embark on data storage and processing. Such a
transition facilitates real-time data capture, analytics, and knowledge
discovery. This empowerment results in the formation of real-time
services and applications
8. Edge AI combines AI and edge computing. With AI at the edge, doing
real-time analytics of edge data gets strengthened to provide real-time
insights.
9. Machine learning – There are powerful algorithms for empowering
machines to learn automatically from input data. These algorithms
pinpoint useful patterns in the dataset.
10. On-device learning – IoT devices are empowered with memory, pro-
cessing, and storage power. Further on, there are additional modules
such as sensing, actuation, communication, etc., being attached to IoT
devices. By deploying AI models, devices are empowered to learn from
data. Such on-device learning yearns for a lot of technical and business
benefits. Devices can continuously learn from data streams to exhibit
intelligent behavior.
11. Few-shot learning is the ability to empower AI models to learn
and adapt to a few labeled data samples. Consider keyword spotting,
where the task is to identify when a keyword is spoken, such as “Hey
Snapdragon,” using always-on machine learning.
 References 359

12. Continuous learning is to empower AI models to continue learning on

new data even after their deployment.

References
[1] Edge computing: 4 use cases for the industrial sector | The Enterprisers
Project
[2] Applied Edge AI: Concepts, Platforms, and Industry Use Cases - 1st Edi
([Link])
[3] What Is Edge AI and How Does It Work? | NVIDIA Blog
[4] Exploring the edge AI space: Industry use cases - ScienceDirect
[5] Edge AI | Industrial Use Cases 2023 ([Link])
 16
Edge Computing and Analytics: A New
Computing Paradigm for Better
User Experience

Santosh Das1,2 and Pethuru Raj3

1 SWAN Lab, Indian Institute of Technology, Kharagpur
2 Department of Computer Science & Engineering,
OmDayal Group of Institutions, India
3 Edge AI Division, Reliance Jio Platforms Ltd., India

E-mail: santoshdascse@[Link]; peterindia@[Link]

Abstract
The term “edge computing” refers to the compute power and infrastructure
offered by devices located closer to the network’s periphery. Considering that
the number of sensors and devices deployed at the network’s periphery is
expected to reach the billions in the next years, the volume of data being
produced by these devices is staggering. Edge data processing is becoming
more popular as it is realized that all edge data contains some actionable
insight. The timely discovery of useful information in massive datasets
depends on it. Since the value of data decreases over time, processing data
as soon as possible is essential for extracting time-sensitive business value.
The distributed architecture at the edge of the network is tried and true.
Edge computing is a decentralized approach, while cloud computing is very
centralized. When it comes to gathering and processing data at the network’s
edge, several heterogeneous devices freely work together. Containerization
addresses the issue of device heterogeneity. Additionally, container lifecycle
management technologies like Kubernetes make it possible for a wide variety
of devices to contribute to business computing.

Keywords: EDGEAI, IIoT, Digital Twin, Edge Computing, Blockchain,

Edge computing

361
362 Edge Computing and Analytics: A New Computing Paradigm

16.1 Introduction
With edge computing, data from various clients is processed at the network’s
edge, as near to its point of origin as possible. Business intelligence and
the ability to monitor and manage mission-critical functions in real time are
only two examples of the many uses for data in today’s businesses. Huge
volumes of data may be routinely acquired from sensors and IoT devices
running in real time from remote places and harsh working environments
practically anywhere in the world, leaving modern organizations adrift in a
sea of data. But this digital deluge is also altering how organizations approach
computing. Moving ever-increasing rivers of real-world data presents signif-
icant challenges for the old computer paradigm based on a centralized data
center and the everyday Internet. Such efforts can be hampered by a conflu-
ence of factors, including low bandwidth, high latency, and random network
outages. Companies are using edge computing architecture to address these
data issues. In its simplest form, edge computing involves relocating some
amount of data storage and processing capacity away from a centralized data
center and closer to the location where the data was originally generated.
Data processing and analysis are carried out at the point of origin, be it
a storefront, a manufacturing facility, a utility’s vast service territory, or
an entire smart city. Only the results of this computing effort at the edge,
such as real-time business insights, predictions for equipment repair, or other
actionable solutions, are sent back to the central data center for review and
other human interactions. As a result, edge computing is influencing profound
changes in the IT and business computing sectors. Explore the definition,
operation, cloud’s impact, edge use cases, trade-offs, and implementation of
edge computing.

16.1.1 Edge devices

We already use devices that do edge computing every day – like smart
speakers, watches, and phones – devices which are locally collecting and
processing data while touching the physical world. Internet of Things (IoT)
devices, point of sales (POS) systems, robots, vehicles and sensors can all be
edge devices – if they compute locally and talk to the cloud.

16.1.2 Network edge

It is not necessary to set up a special “edge network” in order to do edge
computing (it could be located on individual edge devices or a router, for
example). Since this is just another node on the road between users and the
 16.3 The Recent Technological Fusion 363

cloud, 5G can be useful in this scenario when a different network is in play.

With 5G’s low latency and high cellular bandwidth, edge computing may now
support intriguing new uses, such as autonomous drones, remote telesurgery,
smart city projects, and more. When high responsiveness is needed yet it
would be too expensive and inconvenient to place computation on premises,
the network’s edge can be a great resource (meaning the cloud is too
distant).

16.1.3 On-premises infrastructure

These are for managing local systems and connecting to the network and
could be servers, routers, containers, hubs, or bridges.

16.2 Importance of Edge Computing

The majority of computing is now performed at the edge, in areas like hospi-
tals, factories, and stores, where the most sensitive data and powering of key
equipment must occur reliably and safely. Solutions for these areas should
have minimal latency and function without a network. Edge’s promise lies in
its ability to revolutionize enterprise processes everywhere they are applied,
from front-facing customer interactions to back-end manufacturing. Edge
computing enhances proactive and adaptable business activities, frequently
in real time, resulting in novel and improved user experiences.
With edge, companies can bridge the gap between the digital and the
real world. The incorporation of digital resources, such as databases and
recommendation engines, into traditional storefronts making it possible for
machines to teach themselves and for humans to learn from machines con-
structing protective and comfortable habitats. What all of them have in
common is the use of edge computing, which is allowing businesses to
locally execute mission-critical applications with stringent time-to-market,
data-volume, and reliability requirements. In the end, this paves the way for
increased speed of innovation, the launch of brand-new items and services,
and the development of fresh income channels. Edge’s potential to revolution-
ize companies across the board is what makes it so intriguing. This includes
front-and back-office activities, as well as front- and back-facing interactions
with customers.

16.3 The Recent Technological Fusion

At the edge, centralized and decentralized systems coexist. The edge and the
cloud both contribute to the creation of novel user experiences. Data is created
364 Edge Computing and Analytics: A New Computing Paradigm

or collected in a variety of places, but once in the cloud, where computation

is centralized, it can be processed efficiently and cheaply at scale. Edge
computing employs locally generated data to offer real-time responsiveness,
allowing for the creation of novel experiences in a secure and cost-effective
manner. By completing tasks closer to their points of origin rather than in the
faraway cloud and waiting for a response, edge computing reduces latency.
• 5G guarantees the transfer of crucial control messages that allow devices
to make autonomous decisions, making edge deployments seamless. By
bridging the gap between the edge and the Internet’s backhaul, this
solution guarantees that edge devices are properly configured with the
appropriate software-defined network settings.
The Internet of Things (IoT) and other linked devices are valuable
resources that must be protected and properly identified in the cloud.
The “Edge” is located close to or even on these data stores.
• Containers give programmers a uniform setting in which to construct
and distribute their applications. Containers are designed to be portable,
so they can be run on any computer system, regardless of the hardware’s
specifications or software.
Data and services that are stored in different containers and databases
at the edge can be deployed and queried via a service and data mesh.
These meshes provide a unified front end that automatically manages
and routes data and services. This crucial enabler allows for queries to
be performed in bulk for vast populations within the edge, rather than on
each individual device.
Users are able to set up their own overlay networks with the help of
software-defined networking. It also simplifies the process of configur-
ing routing and bandwidth for inter-edge device and cloud connectivity.
• The digital twin facilitates the integration of the physical and digital, as
well as the cloud and the edge. The twin eliminates the need to manually
configure data and applications based on database tables and message
streams, and instead allows them to be set up in terms of assets and
production lines. Domain experts (as opposed to software engineers) can
use digital twins to set up applications for edge sensing, reasoning, and
action.
Edge is strengthened by other technologies, such as artificial intelligence and
blockchain. When AI processes data at the edge, for instance, less processing
power is required at the data center. Furthermore, edge enhances blockchain
by reducing the potential for human error and increasing confidence in the
 16.4 Edge Computing Benefits and Application 365

system as a whole. Through the rising usage of sensors and cameras on the
edge, more and richer data will become available for analysis and action
in real time, and machines can gather and send this data directly. When it
comes to automation, edge is at the forefront of a revolution that is shifting
the focus from systematic procedures in regulated settings like factories to
intricate shows in more natural settings like farms.

16.4 Edge Computing Benefits and Application

When combined with the cloud, edge will provide companies the ability to
completely rethink customer experiences. Edge computing now has potential
uses outside of traditional manufacturing and the Internet of Things. Edge
can be used to increase relevance at each touch point, allowing for faster
decision-making and better user experiences. Now, with the support of the
cloud as a backbone, edge is crucial in the generation of novel insights and
user experiences.
Some benefits of edge computing include:
Rapid response: Transmitting data requires time. There isn’t enough time
for data to travel to the cloud and back in some applications, such as self-
driving cars or remote surgery. When response speed is of the essence, edge
computing makes perfect sense.
High data volume: Although large amounts of data can be stored on the
cloud, there is still the matter of transmission costs and the inherent limits of
network capacity to consider. It may make more sense to process the data at
the edge in some circumstances.
Privacy: Sometimes, users need to keep private information off of the cloud
for legal or ethical reasons.
Remote areas: Both physically distant uses (like an offshore oil drilling plat-
form) and functionally distant uses (like some home automation systems) fall
under the category of “remote” (involving mobile or transportation-related
scenarios using edge).
Cost sensitivity: Various locations on the cloud continuum have different
costs associated with processing data, which can be adjusted to reduce overall
system costs.
Autonomous operations: Where connectivity to the cloud is not possible – or
likely to be intermittent or unreliable – users may need end-to-end processing
within the local environment to keep operations up and running.
366 Edge Computing and Analytics: A New Computing Paradigm

The prime advantage of edge computing is clear: User experience

improves because relevance increases with edge. Additionally, edge unlocks
valuable data to shape new opportunities and innovation for the future. More
sensors generate more data, and there is more processing at the location where
the data is created – which is faster, more reliable, and safer. Integrated with
knowledge from the cloud, the system yields better predictions and more
relevant information, repeating in a cycle of continuous improvement. Other
characteristics of edge use cases include:
Intelligent machines and real-time productivity: Edge let users process
data with velocity, enabling robots and sensors to make split-second decisions
and complete tasks in smarter, faster, and safer ways. This is revolutionizing
everything from smart signage to assembly-line quality assurance.
Optimized close to consumption: Using the Internet of Things, real-time
data, and AI makes the creation and use of digital products better and faster.
Experience with extended reality: These use cases can incorporate dig-
ital twins and optimize rich experiences in healthcare, the workforce, and
entertainment, from smart health to mixed-reality gaming.
Privacy and security by default: By processing sensitive data on the edge,
these use cases improve reliability and protect privacy. Examples include
wearable health devices and the processing of regulated data.
Always-on and untethered: Edge allows for decision-making and process-
ing independent of connectivity for mission-critical and remote applications,
like POS or autonomous operations.

16.5 Edge Computing Examples

Edge, in conjunction with cloud computing, will allow businesses to reimag-
ine the customer service they give. This opens up new possibilities for edge
computing beyond IoT and conventional manufacturing. Through the usage
of edge, we can improve the quality of our interactions with users at every
stage of the user journey, leading to quicker decision-making and more
satisfying interactions overall. Now that the cloud is acting as a foundation,
edge is essential for creating new insights and user experiences.
Below are just a few of the many advantages of edge computing:
Instinctive reaction: Data transmission is a slow process. Some applica-
tions, like self-driving cars or remote surgery, don’t have time to wait for
 16.5 Edge Computing Examples 367

data to go to and from the cloud. Edge computing is ideally suited for use in
situations where a quick response time is crucial.
Heavy data loads: While it’s true that massive volumes of data can be
kept in the cloud, it’s also important to keep in mind the costs of data transfer
and the constraints imposed by any given network. In some cases, it could be
preferable to perform data processing at the periphery.
Users may want to avoid storing sensitive data in the cloud for legal or
ethical reasons requiring anonymity.
Remote regions include both those that are physically removed (such
as an offshore oil drilling platform) and those that are merely functionally
removed (such as some home automation systems) (involving mobile or
transportation-related scenarios using edge).
Sensitivity to cost: The total system cost can be lowered by shifting data
processing to a cheaper point on the cloud continuum.
Autonomous operations: Users may require end-to-end processing within
the local environment to keep operations up and running where access to the
cloud is not possible or is likely to be intermittent or unreliable.
The primary benefit of edge computing is obvious: better user experience
due to increased relevance. Furthermore, edge provides access to vital data
that can be used to mould future possibilities and innovations. Faster, more
reliable, and safer data processing can be achieved by increasing the number
of sensors and bringing that processing closer to the point of data creation.
The system improves with each iteration, thanks to the incorporation of
cloud-based knowledge, which allows for more accurate forecasting and more
pertinent analysis. Some further features of edge applications are:
Edge allows users to digest data quickly, allowing robots and sensors to
make split-second judgments and carry out operations in a more efficient,
timely, and secure manner. This has far-reaching implications, improving
processes from digital signage to quality control on the production line.
Edge computing, used in contexts such as content delivery or on off-
shore oil wells, is an example of digital production and consumption being
optimized for the best experience and lowest cost close to the point of
consumption.
These applications of extended reality, from smart health to mixed-reality
games, can optimize immersive experiences in fields as diverse as healthcare,
the workplace, and entertainment.
These use cases boost dependability and protect privacy by processing
sensitive data at the edge. Wearable medical gadgets and the handling of
restricted data are two such examples.
368 Edge Computing and Analytics: A New Computing Paradigm

Edge computing enables always-on, decentralized decision-making and

processing for crucial, off-network applications like point-of-sale systems
and autonomous vehicles.

16.6 Case Studies in Edge Computing

Let’s examine a couple of current edge use cases that are only going to get
better when 5G expands and new advancements are made.
Store of Tomorrow is a new integrated vision for the near future of
commerce that centers on a flexible, customer-centered experience. A crucial
enabler for the human-centered experiences at the model’s core is edge tech-
nology, which will soon become a central retail competency. The checkout
process at stores can be made easier with the help of edge. Stores lose an
estimated $37.7 billion in revenue every year in the United States due to
customers abandoning their purchases due to long waits.
Customers can avoid standing in line by leaving the store past a self-
service kiosk that appropriately charges their accounts using data acquired
by in-store cameras and artificial intelligence (AI) that has been trained to
recognize inventory items. It helps stores deliver better customer service,
reduce instances of shoplifting, and optimize their supply chains.
In the medical field, robots have made surgery more manageable for
doctors and shorter, less traumatic, and more efficient for patients. As a result
of using edge computing, the incisions can be smaller, the surgeon can sit
down and have a better view of the location, and the controls can be more
natural and intuitive.

16.7 Challenges and Opportunities in Edge Computing

When businesses try to reap the rewards of edge computing, they sometimes
run across resistance. Finding the optimal plan for your company’s competi-
tive advantage might be challenging, but trial and error is essential for making
progress. Typically, we encounter these sorts of problems:
The necessary infrastructure (cloud provider(s), network, and devices) is
needed to get started with edge; however, there is a lack of standard and
integrated architectures. It is common for businesses to employ a variety of
incompatible technology stacks, all of which must be synchronized for edge
to function properly.
In a rapidly evolving ecosystem with a wide variety of technological
options, it is essential to make informed choices quickly. The situation
 16.8 Next-generation Analytics on the Cutting Edge 369

is made more complicated by the ongoing development of new network

capabilities, such as MEC and 5G.
Value for business not yet realized at the edge: It might be challenging
for businesses to fully appreciate the value that can be unleashed by edge
solutions. For edge computing to provide a sustainable return on investment,
businesses must look beyond simple use cases that generate short-term profits
and instead invest in experiences that customers will find appealing, practical,
and viable.
Unfortunately, many businesses aren’t built to quickly adapt and scale
beyond the proof of concept stage, leading to innovation fatigue and pilot
purgatory.
Edge isn’t about retooling, especially for firms that are already embracing
the cloud, and there is a lack of cloud professionals who understand what
belongs at the edge, why, and when. The goal is to bring those capabilities
to the periphery. Deploying at the edge requires only a basic hardware
connection, so you can take advantage of your current cloud expertise.
The edge presents its own unique security concerns, as security must be
maintained in all potential edge instances, from the cloud on out, but IoT and
edge security is significantly different from IT security. Time-critical, secure,
and self-driving tasks abound in the periphery. Security models account for
the long product lifecycle and existing base of edge devices. Production and
security can be jeopardized with each reboot, and they become comparatively
antiquated quickly. Devices may also be deployed in untrusted or far-flung
locations, necessitating a multi-layered approach to security that incorpo-
rates both cyber and physical measures. The distribution of security patches
is made more difficult by heterogeneous hardware, software, and network
configurations.

16.8 Next-generation Analytics on the Cutting Edge

By bringing cloud resources (such as computing, storage, and network)
closer to the edge devices (smart devices where the data are produced and
consumed), edge technology can improve performance and efficiency. Two
new concepts in edge technology – edge computing and edge analytics –
have emerged as a result of the incorporation of processing and application
into edge devices. Data from edge devices can be analyzed with the help of
various methods and algorithms used in edge analytics. Edge devices are now
fully functional thanks to the development of edge analytics. The analytical
370 Edge Computing and Analytics: A New Computing Paradigm

Figure 16.1 Venn diagram showing the overlap between edge devices, edge computing, and
edge analytics.

methodologies are not fully supported by edge analytics at the moment. Due
to limitations in power, memory, resources, etc., edge devices are unable to
run complex analytics.

16.9 The Urgent Requirement for Edge Analytics

IoT’s meteoric rise around the world has prompted researchers and businesses
alike to focus on edge analytics. Significant work has gone into the develop-
ment of efficient and effective frameworks for carrying out edge analytics
for the various scenarios that can arise in practice. In order to facilitate the
linking of devices and the processing of data streams from devices, numerous
IoT platforms have been developed and put into operation. Popular services
include Arkessa, ThingSquare, and Thing Worx. Previous work by us also
offered a cloud-based solution for processing of actionable analytic events in
real time. Typically, data from IoT devices is collected and processed using
the Platform as a Service (PaaS) cloud service architecture and lightweight
messaging protocols like Messaging Queue Telemetry Transport (MQTT)
and Constrained Application Protocol (CAP) (CoAP).
Stream processing must be relocated closer to or onto the edge, and this
has given rise to the concepts of edge computing and fog computing, as the
computational power of the edge has increased significantly in recent years
and cloud IoT analytic services incur high latency and unstable networking
to edges. Still, they haven’t figured out how to define and deploy processing
closer to or onto the edge in a flexible manner. By reducing the time and
 16.9 The Urgent Requirement for Edge Analytics 371

effort required to build and deploy changes to the processing logic on the
fog node, real-world IoT applications would benefit from a more flexible fog
node. To limit the amount of data devices send to the cloud from the edge,
but keep the essential events of users’ interest, researchers are focusing on
the heterogeneity of device data in order to execute flexible data processing
on the edge. Edge-based IoT data reduction can make use of a variety
of strategies, the most prominent of which being time series compression
and aggregation and event or rule-based data filtering. Edge-based IoT data
reduction can make use of a variety of strategies: time series compression
and aggregation, and event or rule-based data filtering. These strategies are
predicated on locating “critical points”, whereas piecewise approximation,
another technique for data reduction, is used for fitting curves to the data.
However, these solutions tend to be application-oriented, leaving users with
less leeway when it comes to using them in their own IoT context. Such
solutions, however, tend to be application-oriented, leaving users with less
leeway when it comes to using the aforementioned methodologies in their
own IoT context.
In contrast, rule-based data filtering streamlines IoT information in accor-
dance with industry-specific policies rather than time series features. As an
illustration, Cisco’s open-source project Krickitt is working on defining a data
format and mechanism for “telling the network-edge devices” which data to
forward and in what manner. These methods still rely on edge developers
to implement data reduction logic, but go no farther than building rule
engines and policy languages for expressing data reduction. To attract the
growing market, cloud service providers are looking for ways to simplify the
management of edge analytics by making it possible for users of Internet of
Things (IoT) applications to create and deploy their analytic logic on the edge
in a manner that allows for maximum flexibility. So, the three largest cloud
service providers in the world – Microsoft Azure, Amazon Web Services
(AWS), and IBM Watson – are all pushing their edge analytic services –
Azure IoT, AWS Greengrass, and IBM Watson IoT Platform, respectively.
All have a generally similar architecture, with a microservice on the
edge managing the connectivity of edge analytics and device data, a cloud-
based messaging service, and a cloud-based service managing the operation
of devices and analytics. They also offer a standard method for defining
analytical logic for the edge in the cloud, making that logic available for
use in edge microservices. While Azure IoT and IBM Watson IoT Platforms
give programming-free “task” and “rules-based analytic model” approaches
to edge analytic logic and RESTful APIs for service composition, AWS
Greengrass offers local lambda programming capability in Python to increase
372 Edge Computing and Analytics: A New Computing Paradigm

analytic flexibility. Our proposed analytic model for the IBM Watson Plat-
form is able to access any local services via an edge resident broker, whereas
AWS Greengrass core, which is responsible for local lambda function execu-
tion, has restrictions on accessing external services (such as local file systems
or local 3rd party service). In contrast to the AWS Greengrass core and the
Azure IoT edge SDK, which use multi-threaded runtime to handle many
analytic processes in parallel, our single-threaded edge engine can process
rule-based analytic with minimal overhead.

16.10 Definitions of Edge Analytics

Data gathered by edge devices might be either visual or textual, and this
distinction informs how edge analytics are applied. There are two main types
of edge analytics – image analytics and video analytics – depending on the
type of media being processed. Again, there are two distinct kinds of edge
analytics based on textual data: descriptive analytics and regression/predictive
analytics. Analyses performed at the edge of the network can be further
categorized using sound samples. There have been no proposals for edge
analytics; hence we have not included them in the taxonomy. The field of edge
analytics is broken down into two distinct subfields, image edge analytics
and video edge analytics, depending on whether the underlying data source is
images or videos. Video is used as the foundation for picture edge analytics. A
video is an animated series of still images called frames. Accordingly, a frame
is the starting point for picture edge analytics. A video is used as the input
in video edge analytics. In order to process videos with limited computing
power and resources, videos are broken down into groups of video frames
before being processed.
GigaSight is a VM-based cloud with a distributed hybrid architecture that
allows for video edge analytics. Cloudlets are edge nodes in the architecture
that execute video analytics, modify video streams to safeguard privacy, and
retain films for finite amounts of time in accordance with pricing regulations.
Each video is altered in a separate virtual machine (VM) within the cloudlet,
with the level of denaturation being automatically decreased based on the
fidelity of the video stream the user selects. To avoid overwhelming the
computer with too much work, the denaturing procedure is done to a small
subset of the video at a time. When a video is denaturated, two files are
generated: a thumbnail and an encrypted version of the video that is saved to a
different cloud service; the thumbnail video serves as an outline of the video’s
content for indexing and query purposes. Video frames are passed unaltered
through a binary filtering process based on metadata (such as location and
 16.11 ML and Near-real-time Analytics on the Periphery 373

time). The computational complexity of this filtering is low, and it filters

based on the content. The indexing operation’s throughput is proportional
to the number of objects requiring detection. For efficiency’s sake, indexing
is handled in a separate VM while the classifier is applied to the database’s
most vital objects.
To find videos that are relevant to a given user, we first conduct a standard
SQL search using metadata and tags obtained from the index, and then
we conduct a second, more computationally intensive search on the actual
content to provide users with more relevant videos. Also, GigaSight can
conduct time-sensitive video searches. Video processing output and metadata
are stored in the cloud. Because of its reliance on virtual machines, GigaSight
is limited in the number of queries it can handle at once. Since edge devices
have fewer resources, a smaller number of efficient virtual machines is pos-
sible. The number of concurrent requests is further reduced because indexing
occurs in a separate virtual machine.

16.11 ML and Near-real-time Analytics on the Periphery

Edge technology is an opportunity for researchers to develop machine learn-
ing algorithms with reduced computational requirements. Edge nodes can
generate actions for events without the need for programming by incorpo-
rating AI and machine learning algorithms into edge devices. For new data,
inference requires less computing power to run and produce results. There-
fore, inference helps provide a low-latency response to the user’s request
when executed at edge nodes. Microsoft’s Azure IoT edge allowed for the
integration of machine learning and edge computing. Using models stored in
the cloud, AWS Greengrass performs machine learning inference locally on
edge devices. The edge nodes have a lot of issues because of machine learning
techniques. Some of the issues are as follows.
• An edge node only receives a fraction of the whole data. Consequently,
the performance of the machine learning algorithm when deployed to
edge devices is inadequate.
• The information in real time is always changing. Therefore, the training
that the machine learning algorithm does becomes obsolete after some
time has passed.
• Every single machine learning method has the same requirement: a lot
of computer processing time. Thus, such algorithms cannot be supported
by edge nodes due to their limited resources, power, processing, etc.
374 Edge Computing and Analytics: A New Computing Paradigm

In what follows, we’ll take a look at some of the methods that have been
proposed to bring together machine learning and edge analytics. The edge-
deployed convolution neural network (ECNN) is a distributed data analytics
architecture that enhances the efficacy of data gathering and analytics in smart
grids. Many of the first layers are convolutional or filtering or pooling or
fully connected (FC) or use the Softmax function to classify data. On both
the cloud and edge devices, you’ll find a plethora of convolution and pooling
layers. Further, classification is performed by FC layers at the edge nodes.
These nodes’ local data generates local inference, while more data-intensive
inference is generated in the cloud. Since local inference is conducted on
a small scale, transmission costs are reduced. ECNN is used to aggregate
data, which helps in cross-regional inference and decision-making. An exam-
ination of federated learning could be undertaken to develop data-intensive
inferences without resorting to the cloud. Sarabia-Jacome et al. developed a
proposal for an edge gateway with a predictive analytics module that makes
use of deep learning models.
The system is private, secure, and compatible with other systems.
Container-based virtualization is used for managing scarce resources. Pre-
dictive analytics uses cloud-based services based on a deep learning model to
discover and predict trends in data collected from edge devices. An enormous
amount of time and energy is needed to create an inference system for a
deep neural network. That’s why we draw conclusions about the world from
data stored online. The results are then transmitted to the edge nodes. Edge
devices with three-axis accelerometers can collect user feedback in the form
of textual information. In their proposal for a data analysis framework, Moon
et al. proposed delegating specific analysis tasks to various edge nodes. With
this architecture, you can leverage the power of the cloud and the speed of the
edge nodes. Edge nodes collect data while also monitoring and operating the
actuators in real time. Edge analytics uses machine learning algorithms in the
cloud to create models based on information acquired by edge devices.
Not all information gets uploaded to the cloud, however, to protect
users’ privacy. Since then, the cloud has been limited to using only pub-
licly available information. The cloud houses the modules responsible for
managing IoT data and machine learning models, while edge nodes host the
IoT interaction module. Module 1 is where we collect and process training
data. In the second part, the data is trained. The IoT interaction module
makes forecasts according to the user profile. The models employed in edge
analytics eventually become obsolete, leading to subpar output. The system
is more reliant on the cloud because the edge analytics necessitate constantly
 16.12 Edge and Big Data Analytics 375

refreshed forecasts. Restricting some data owing to privacy concerns reduces

the accuracy of the forecast. Dey and Mukherjee developed a deep learning-
based architecture for analytics at the network’s periphery. Based on a deep
convolutional neural network model, the framework provides procedures for
dividing up resources between edge devices (DCNN). However, tuning the
DCNN’s hyperparameters consumes a lot of computing power. There are
three phases of edge analytics’ resource management: intermediate resource
standardization, intermediate resource supply, and offloading execution to
serve small devices. Standardized resources have many advantages, including
low latency and data management.
In addition to increasing customer confidence and facilitating magnetiza-
tion, these features also facilitate the commercialization of services created
by third parties and analytics designers. Intermediate resource provisioning
is utilized for efficient load balancing. Different configurations of control
unit, graphics processing unit (GPU), and random access memory (RAM)
are used to test out the performance of a deep learning model. They are
profiled and used as data in the form of training exercises. In order to offload
the processing of the analytics tasks associated with deep learning, they are
split up and sent to the many external devices that are connected to the local
network. The optimal offload implementation makes use of a simple capacity-
based segmentation method. It is not possible to separate the processing
of each layer for parallel processing due to the coarse granularity of the
partitioning used in the architecture. Since deep learning is a computationally
intensive process, it is possible to complete the operation without the cloud
by distributing it among a number of edge devices. Also, if you divide the
work up according to the available resources, you should be able to finish it
much more rapidly.

16.12 Edge and Big Data Analytics

Data analysis is performed in two ways by the many edge devices that are
linked to the edge gateway: via edge analytics and via big data analytics.
Analytics for massive amounts of data can be done on the cloud. When large
amounts of data are collected that lack clarity, exhibit unusual patterns, or
yield unpredictable outcomes, this is when big data analytics comes into play.
Big data analytics necessitates a robust cloud environment. One of the main
challenges of big data analytics at the edge is ensuring data quality and con-
sistency - especially when dealing with heterogeneous and distributed data
sources . We can see a connection between big data analytics and analytics
at the edge. scale, or the number of computers, networks, storage devices,
376 Edge Computing and Analytics: A New Computing Paradigm

etc., that are spread out across a large area. Supporting these complicated
and cutting-edge technologies that need to calculate the data also necessitates
powerful analytics and large-scale computing systems. The cloud is home to
big data analytics because it is capable of accommodating the humongous
amounts of storage, computing power, and network bandwidth required by
these applications.
Integrating data analytics into edge devices is a primary focus of edge
computing. Because it’s possible that the analysis will be performed by the
data source itself in that case. Meanwhile, the development of edge analytics
has just begun. Edge computing allows for the compact storage and process-
ing of massive datasets (i.e., router). It will decrease both transmission costs
and bandwidth utilization, among other benefits. However, the edge analytics
applications are not yet highly developed or sophisticated. Getting the kind
of power source that can support such high-tech applications has become
an issue. Therefore, in the event of a request needing robust compute, edge
analytics must rely on the cloud.

16.13 Applications
Several sectors make use of edge analytics, such as the energy sector, the
logistics sector, the manufacturing sector, and the retail sector. Edge analytics
can be used for a wide variety of purposes, such as analyzing client behavior
in retail settings, keeping tabs on industrial and logistics equipment, keeping
an eye on energy usage remotely, and detecting fraud in financial institutions
(ATMs). This section discusses how edge analytics can be applied to address
problems in a variety of disciplines. Edge devices such as Wi-Fi networks,
sensors, and CCTV are widely used by the retail industry to collect raw data.
The number of customers at different times of day, the frequency with which
discounts were used, and the quantity of images and videos taken are all
examples of the information that was collected. By analyzing the raw data,
businesses may better anticipate consumer behavior and tailor promotions to
entice repeat business. Transferring data to the cloud for processing, however,
will cause a delay in getting the results back. Thus, timely analysis is required
to entice shoppers before they leave the business. Edge analytics are useful in
these kinds of circumstances.
Sony et al. provided a service-oriented framework for personalized and
real-time suggestion based on an interactive screen-based edge analytics
system. Edge information and social media profile data are matched using
matching algorithms based on common criteria like name and address. This
ensures that analytics are consistently applied across all client channels inside
 16.13 Applications 377

the edge analytics framework. Social media engagement is a leading indicator

of consumer intent to buy. As an example, edge analytics just needs to con-
sider a customer’s recent purchases and a small portion of their social media
activity to provide them with real-time, customized suggestion services. It
can be difficult to properly connect two entities, a social media entity and
a business entity, because a single client may have many accounts on social
media with different or similar names, as well as multiple bank accounts.
Privacy should be prioritized in the edge analytics since some customers may
want to use the service while keeping some purchases or items private.
The use of sensors to track which animals enter fields of crops, automated
irrigation systems, real-time weather updates, and other forms of smart tech-
nology are all helping to elevate the agricultural sector. However, a human
presence is still required round-the-clock, in case the owner needs to be
notified to move the animals away from the crop field, for example. The use
of edge analytics can help cut down on manual processes. We can keep an
eye on the animals owing to video edge analytics. The alarm will ring if
the animals go too close to the crop fields, and scare them away if they do.
In the event that any livestock grazes outside of the designated area, or if
another animal or unauthorized person enters the area, the owner is notified
immediately. A weather report is essential for preserving the quality of the
harvest. Say it’s late in the day, and there’s a probability of rain. After that
point, morning irrigation of crops is no longer performed. In a similar vein,
if the temperature is high, more crops will need to be watered. Bhargava
et al. have proposed using edge analytics as a useful strategy for achieving
precision in the dairy industry. The data is gathered by collars, in-field sensor
nodes, and edge gateways. Monitoring the environment and the weather, the
infield nodes are essential to any modern farm. Dairy cows are equipped
with collar technology that monitors their vital signs and movement. These
gadgets feed information into edge gateways, which subsequently forward it
to the cloud. Due to the typically poor Internet connectivity present in a farm
setting, the solution employs a delay-tolerant application. Because of this,
it’s possible to seamlessly move information between edge gateways and the
cloud.
The GPS data is analyzed to get an idea of where the cow is. Because
of storage constraints, the GPS data is now only sampled once every 15
minutes. By setting it up in this way, the device will last, on average,
longer while consuming less power. The technology encrypts and com-
presses the raw data collected by the collar devices. Linear-SIP data com-
pression is used to expand storage while decreasing processing time. The
378 Edge Computing and Analytics: A New Computing Paradigm

edge analysis helps keep an eye on the cows without the need for human
intervention.

16.14 Transportation
A smart city’s intelligent mobility system (ITS) allows for convenient and
safe transportation via a network of sensors, wireless connections, and
autonomous vehicles. Smart sensors installed in automobiles pave the way
for ITS. These edge devices collect a plethora of disparate information
about the passengers, the immediate environment, the car itself, and other
nearby vehicles, all of which must be processed in real time. For efficient
transportation, the vehicle, rather than the driver, must make a number of
crucial decisions. Having these data transported to the cloud, analyzed, and
then delivered back so that action may be taken; however, has a major impact
on quality of service. As an added downside, postponing decisions can cause
disastrous results like traffic bottlenecks and accidents. Because of this, ITS
requires efficient edge analytics to facilitate timely decision-making based on
real-time data.
There are a variety of sensors built into the automobile that keep tabs
on things like speed, gas mileage, and driver fatigue. As storage on edge
devices is sometimes meager and data overflows can lead to the loss of critical
information, it is essential that unneeded data be purged. The amount of data
that needs to be processed is still immense, even after filtering; therefore
edge analytics must be efficient to quickly calculate such a vast quantity
of real-time data. When an edge device can swiftly calculate to identify the
optimal path, ITS’s quality of service improves. A self-driving automobile
must make its own decisions about how quickly to accelerate, where to go,
and how to turn. These factors are, however, affected by the presence of
nearby vehicles, people, and traffic signals. Congestion has a major impact
on the route taken. Traffic congestion and delays can occasionally be caused
by unforeseen circumstances, such as accidents. Rapid computation of real-
time data is required to suggest a quick course correction to the driver in order
to avert the driver from traveling such a road. The cars can be driven either
manually by the driver or automatically. Mode switching in these automobiles
is lightning fast. When the driver is in charge, the car is constantly giving them
suggestions, such as when they are too near to another car or the best route
to go.
However, it is difficult to provide such guidance since human drivers
frequently make unanticipated decisions. The stakes are high if an attacker
 16.14 Transportation 379

can trick an edge analytics machine into drawing the erroneous conclusions,
therefore the technology must be solid. When a car is in self-driving mode,
the human driver may not be able to take over the wheel immediately, making
the situation extremely dangerous. Edge analytics can occasionally employ
erroneous data, leading to poor conclusions, thus it’s important to think
about the safety of edge devices. To improve ITS, Jiang et al. presented a
blockchain-based video edge analytics system. A vehicle and a node known
as a road side unit (RSU) make up its two edge devices. Edge analytics are
located in the road side unit, whereas the camera is located in the automobile
node. The vehicle extracts individual frames from a video and sends them via
vehicular networks to edge analytics, where they are processed. Blockchain
technology allows for the secure storage of video in the RSU node and
the exchange of data between several vehicles. The process of video edge
analytics is broken down into sub-steps due to a lack of resources. The video
decoding module is responsible for real-time decoding and decompression
of the encoded video. The video is broken down into individual frames by
the protocols used for real-time streaming. The preprocessing module makes
adjustments to the brightness, contrast, and color temperature to fix problems
caused by uneven lighting.
Semantic objects of a certain class can be located using the object detec-
tion feature. It is essential for the progress of driverless cars. The use of
deep learning techniques to improve object recognition has led to significant
advances in computer vision applications. The object tracking component
determines where the object or objects have been at various periods. It’s also
used to make educated guesses about how busy the roads will be along the
path the car takes. The path’s environment can be better understood with the
help of the semantic segmentation module, which can then be used to decide
whether to make a course correction. A label is assigned to each pixel in a
photograph, and that label is then assigned to one of several categories. The
stereo module is responsible for transforming the camera’s 2D images into
3D data. The optical flow component provides data for use in other programs
as well as doing its own motion estimation and tracking. Through the use
of the suggested edge analytics, the video offloading and resource allocation
problem is converted into a discrete Markov decision process in which the
reward is optimized. We apply deep reinforcement learning methods to solve
the discrete Markov decision problem. Due to the unpredictability of traffic
conditions, object detection processes need to be sped up. The computational
demands of deep learning make it inappropriate for use in object detection.
380 Edge Computing and Analytics: A New Computing Paradigm

The program should also perform well with dim or fuzzy videos (e.g., the
video takes at a rainy season or during rain).
At busy intersections and elsewhere, the INcreasing TRAffic SAFety with
EDge and 5G (InTraSafEd5G) system will detect bikes and pedestrians in
the blind spots of cars and issue alerts. Computing and communication are
the system’s two primary components. The calculation is a joint effort of the
hardware and the software. Installing edge devices, calculating space require-
ments, and placing cameras in optimal locations are all part of the hardware
setup for ensuring constant surveillance of key intersections and walkways.
The software part is known as edge analytics. Edge analytics receives footage
from edge devices and then processes it with deep learning object detection
techniques. The communication module relays user messages sent by the
edge analytics in real time. An integral part of every communication sys-
tem is the network and the applications that use it. The network piece is
responsible for transmitting the data to the many drivers who are approaching
the crossroads in a timely manner. The application’s job is to keep tabs on
the car’s whereabouts and notify the driver of any problems via the app’s
user interface. TraSafEd5G facilitates communication via 5G networks. 5G
is essential for instantly notifying motorists. To do this, TraSafEd5G uses
cameras installed at busy junctions and pedestrian crossings. The videos are
captured and sent to the camera’s in-house edge analytics. Edge analytics
are constantly processing the videos to identify people on foot and cyclists.
When a vehicle is approaching the area, the edge analytics sends a notification
with details on the number of items and a picture of each item detected. The
driver’s mobile device runs an app that can display and play audible alerts.
Information and videos are not stored for privacy concerns. If object
detection is always active, it could take more time to process the video frames
containing the item. The camera also requires constant power in order to func-
tion. There’s a chance that the gadget can also gauge how fast the approaching
vehicles are moving. Then, a special color on the traffic lights can alert
pedestrians and cyclists. The (SCADA) infrastructure requires a connection to
pre-existing remote telemetry units (RTUs). The inference feedback from the
edge gateway to the RTU is handled independently. When the edge gateway
uploads the latest data to the cloud, MLM(Masked Language Modelling) is
refreshed to reflect the changes. Rod pumps and artificial lift systems are
used to boost hydrocarbon production in producing wells. The effectiveness
of rod pumps is measured by the use of dynagraph cards (i.e., position vs.
load graphs). Generating a dynagraph card is a quick and simple process that
requires no more than a few seconds. As a result, there is an issue with the
 16.14 Transportation 381

storage of these cards. Data file management and archiving on a local level
are employed for this purpose. The information is gathered by the RTU and
sent to the SCADA host. It takes time for a specialist to spot any mechanical
or manufacturing flaws. There is a window of time between gathering data
and the risk of losing it. For this reason, these tasks are typically executed
by means of machine learning algorithms. A supervised learning approach is
utilized for cloud training.
To train a generic model with machine learning methods, historical
dynagraph cards are digitized and then pixelized. The ability to foresee
manufacturing issues is honed using a historical database of 6000 dynagraph
cards. Subject matter experts label and categorize the cards prior to being
used by operators or rod pumps. The assembling method combines MLM’s
prediction, giving more weight to the model that performed best for a given
class of problems. The model is trained multiple times to get the desired level
of precision. Once the MLM has reached the required accuracy, it is sent to the
edge gateways. These inferences can make predictions in a matter of seconds
while requiring fewer resources in practice. The training dynagraph cards
require expert labeling. However, the constant generation of dynagraph after
small intervals is the cause of the frequent modification of interference. As a
result, edge analytics rely heavily on involvement from humans. Boguslawski
et al. proposed an edge analytics strategy for rod pump monitoring. It makes
use of ML approaches.
At the cutting edge, analytics may take the form of either assembly or
generalization. The assembled technique develops a meta-model from the
learned models. Data is trained on the edge devices. The failure of a machine
or process can be more easily pinpointed with the help of the meta-model.
Those inferences allow for the usage of less expensive hardware. The edge
devices save the data to disc on a regular basis. The edge analytics continue
to work even if there is no Internet connection. Disconnection in the cloud
may lead to less precise inferences due to the small training dataset. Once the
connection is established, the training model and the conclusions it generates
are sent to the cloud. In cases where the user values secrecy, they can employ
this technique. Machine learning methods that necessitate a lot of compute
can’t be run on edge devices because of the amount of effort required to
assemble the techniques. Therefore, edge devices cannot support the imple-
mentation of a wide variety of machine learning methods. Edge analytics can
deliver inferences without the cloud, albeit with reduced precision. A diagram
showing how federated learning works. There are multiple devices, such as
smartphones and laptops, that have local data and models. They communicate
382 Edge Computing and Analytics: A New Computing Paradigm

with a central server that aggregates the local models and updates the global
model. Federated learning allows the devices to learn from each other without
sharing their data, thus preserving privacy and reducing communication costs.
Edge analytics for end-to-end water leak localization and burst detection
scheme proposed by Kartakis et al. takes advantage of the difference in arrival
time of changes measured at the sensor location, as opposed to the traditional
scenario of periodic reporting between backend servers and sensor devices.
The event’s timestamp is relayed to the backend, allowing for precise local-
ization of the water explosion while simultaneously reducing communication
costs. The decision control procedure decides how the pipe network will be
modified using remotely controllable valves. The timestamp of the data is
an input for edge analytics, which uses it to identify occurrences. The data
from the sensor node pair is analyzed iteratively using edge analytics to
spot outliers. Edge analytics can take compressed data as input, although the
compression method is inefficient and has a limited storage capacity (10,000
rows maximum). Instability of the data can be influenced by the efficiency of
the compression process. The system could use more memory-efficient data
compression methods and more edge devices to lessen data volatility. The
degree of data volatility changes according to the topology’s edge nodes.

16.15 Hopes and Plans for the Future

In the near future, edge analytics will likely replace other technologies as the
industry standard, creating new opportunities. The most promising avenue is
the development of miniaturized hardware with enhanced processing power
and storage capacity. Algorithms/techniques must be highly efficient and
accurate while utilizing as few system resources as possible, according to
software engineers. The backbone of edge analytics is information derived
from either textual sources or images/videos. Music, like other forms of
media, conveys information. The good news is that we can strengthen national
security without punishing law-abiding citizens. Voice calls and the telephone
can facilitate both domestic violence (including stalking) and international
terrorist plots. A warrant and supporting proof are typically required before
authorities can access a person’s phone records in many countries. As a means
of thwarting terrorist plans and other criminal enterprises, the government is
lobbying for the ability to track the whereabouts and phone histories of its
citizens. A common counterargument is that such widespread surveillance
violates individual rights.
 16.15 Hopes and Plans for the Future 383

Edge analytics is the solution to this problem. At the edge, analytics can
examine each call for context. Phone numbers and other identifying infor-
mation are passed along to the proper authorities whenever a call involves
potentially dangerous language, such as threats or harassment. To ensure the
privacy of both parties, the system will delete any and all data associated with
a regular or encrypted call. To this end, scientists are focusing on cutting-
edge technologies with the help of federated learning (FL). For the purposes
of federated learning, which is a type of machine learning, an algorithm is
trained in a decentralized manner, meaning that it is learned by a network of
devices. Edge analytics greatly benefits from federated learning due to the
latter’s ability to make the most of the limited resources available. A machine
learning algorithm’s training request from the edge analytics is often referred
to as a “FL server.” The FL server selects several edge devices from which to
collect data for model training. From time to time, edge devices will update
the FL server with new model data, which may include revised weights and
parameters. In order to get the appropriate level of accuracy, this process is
done as many times as necessary. There are numerous issues that need to be
resolved, despite the fact that federated learning is a fantastic technique to
employ machine learning algorithms in data analytics.
It’s possible that not all edge devices utilized for federated learning have
access to the same dataset. This increases the cost of communications as the
dataset must be distributed over multiple edge devices. The trained model’s
performance may suffer if there is an unbalanced distribution of classes. In
such a case, an additional rebalancing step is carried out. The selected edge
devices have their data for the subclasses enhanced during pretraining. The
cost of both sending and receiving texts is a major issue as well. Throughout
each training session, there is a great deal of back-and-forth communication
between edge devices and FL servers. In addition, a high-dimensional dataset
necessitates additional rounds of transmission. Choosing which edge devices
will take part in the network without using a load balancing mechanism is
another difficulty. In any case, a staggered node will cause a delay in the
training process. In light of the importance of achieving true cloud indepen-
dence for edge analytics, FL is a viable option worth examining. In the sixth
generation of communication networks, edge devices will be essential (6G).
One of the most important specifications for 6G is a latency of less than 1
millisecond, and edge devices play a crucial role in achieving this.
The 6G network will rely heavily on edge devices to guarantee global
coverage, high network node density, privacy, secrets, and security. Edge
computing is one of the technologies that enables sixth-generation mobile
384 Edge Computing and Analytics: A New Computing Paradigm

networks. Edge devices will enable 6G and provide high-speed Internet with
fewer interruptions. There will be real-time analysis of data. Edge devices
(like drones) will act as key network nodes in times of crisis. At this time,
the network’s nodes are unable to communicate with one another. Wireless
networking will be utilized to connect the edge devices that will be put
throughout the network at crucial spots. Having a way to stay in touch with
the rest of the world is made much easier by having access to a network like
this. The timely distribution of food, medicine, and medical care is hampered
by inadequate communication.
Edge devices will manage cutting-edge technology like holographic in
addition to the home, traffic, grid, and environment. Augmented and virtual
reality, telesurgery, tactile web browsing, and other similar technologies are
all in the works.

16.16 Extreme 5G Networking

The market has recently been flooded with 5G advertisements from various
telecommunications companies and service providers. Indeed, why not? Edge
computing’s advent and widespread use has spurred innovation in AI applica-
tions, while 5G has altered the landscape of edge computing significantly. It is
fueling and hastening the adoption of cutting-edge and AI technology. In the
contemporary epidemiological setting, there are numerous examples of 5G in
usage. These are promoted as having high bandwidth and low latency, and
they can be used for a broad variety of purposes, from distance education and
gaming to watching live sports. There is potential to improve performance,
strengthen data security, and elevate digital experiences by combining edge
computing with 5G technologies. This essay will provide an introduction to
5G from an edge computing perspective.

16.16.1 A vocabulary for the 5G frontier

As we have seen in prior blog posts, edge computing provides new opportu-
nities for businesses across many different industries. In order to fully grasp
what or how 5G enhances various use cases, it is helpful to first become
familiar with basic 5G-related terminology.
The 5G base station, or gNodeB, is in charge of transmitting and receiving
data from and to UEs on the mobile network. Novel radio, a novel radio
access method, is used (NR).
mmWave is a common kind of transmission for 5G networks. Due to the
shorter range and the possibility of interference from building walls, the cells
 16.17 Edge Data Classification and Storage Analysis 385

can only be as large as microwaves. Millimeter wave networks can employ

much smaller antennas than older cellular networks did.
MEC: In a multi-access edge computing network architecture, supple-
mentary processing is stored on the network’s periphery, where it may be
accessed quickly by users running latency-sensitive applications.
Virtualization, also known as network function virtualization (NFV), is an
approach to network design that makes use of IT virtualization technologies
to implement full virtualization of entire classes of network node operations.
Users’ devices are linked to the rest of the mobile network by a radio
connection at the RAN. 5G is able to accommodate both virtual radio access
networks and the more traditional, physical aspects of a network (vRAN).
Software-defined networking (SDN) is an approach to managing network
devices like routers and switches via a network using standard protocols for
configuration and management.
In the 5G era, the packet gateway is known as user plane function
(UPF). The parts in this set facilitate packet forwarding and routing, network
connectivity, and policy enforcement. UPF is sometimes referred to as the
data plane.
The 5G design has integrated network function virtualization (NFV)
with native cloud standards/technologies to simplify network and service
deployment, operations, and management.

16.17 Edge Data Classification and Storage Analysis

When using edge computing to assess IoT data at the edge, we are met with
a deluge of information. The vast majority of the data that is generated every
second by all devices—whether it be audio, video, sensory, or telemetry—is
never stored locally, analyzed, or transmitted north to enterprise data centers
or public clouds. Business organizations do not have the resources to store
and handle all of this information. So now what are we supposed to do with
all of this information? It is possible for businesses to properly classify edge
data by adhering to a few criteria that will help them determine which data is
essential and which can be safely discarded.
It’s common to classify information according to the context in which
it was gathered, be it a certain discipline or method. Statistics makes use
of both quantitative and qualitative information. Data science takes it a step
further by classifying data as nominal, ordinal, discrete, or continuous. In
the world of database management systems, there are three types of infor-
mation: transient, persistent, and useless. In order to better manage security
386 Edge Computing and Analytics: A New Computing Paradigm

and compliance, firms can classify data as either public, internal-only, con-
fidential, or restricted. Additionally, there are methods for automatically
classifying data based on its context and content. However, the data indus-
try relies on three distinct classification methods: content analysis, context
analysis, and user analysis.

16.17.1 Classifying information for a cutting-edge approach

There are two types of data that can be used in an edge solution:
Information about the end-user generated by the solution’s devices, as
well as information about the solution itself that is required for configuration
and operation.

16.18 Edge-clustering Phenomenon

In many cases, the restricted resources of IoT devices and the high computa-
tional power required for processing and storing data render it impossible to
run applications that rely heavily on IoT devices. Companies can use edge
clusters to give their operations teams the adaptable, dependable environ-
ments they require to handle their data’s storage, compute, low latency, high
performance, and high bandwidth demands. In addition, the scalability that
Kubernetes clusters are meant to provide can be required by some highly
accessible on-premises shared services, such as edge cloud deployments.
An organization’s use of an edge cluster as a logical resource barrier is
common. The cutting-edge technologies are expensive for businesses to pur-
chase. A large number of older, specialized devices with fixed functionality
may have previously been installed and paid for. As a means to improve
and secure their applications for the future, businesses can leverage edge
cluster technology. An IoT platform or device management solution connects
these devices via a small-footprint edge cluster. A previous blog entry titled
“Policies at the Edge” demonstrated how to manage and run an edge cluster
like an edge device.

16.18.1 The primary benefits of edge clusters are as follows

Make it easy and possible for users to increase the cluster’s capacity as
needed, a feature known as “scalability.” The system may be set up to scale
automatically, which reduces the workload on the operations staff.
 16.20 Alternatives in Edge Architecture 387

Because the edge cluster processes and analyzes more data, you gain
insights and can take action more quickly without spending as much time
or money transmitting data to the cloud.
In addition to meeting data residency and isolation needs, processing
data at the edge cluster also aids in meeting regulatory compliance needs.
Highly available: The failback options on the cluster and the flexibility to
add new nodes as needed ensure that the application will continue to run
smoothly even if some nodes fail. Safe: No unsecured data transfers between
the cluster’s various app servers.

16.19 Safety on the Fringe

The large number and dispersed location of these devices makes their man-
agement a nightmare from a security perspective. Now that I think about it,
the centralized nature of the solution and cloud-based control systems have
caused problems with the IoT method in that very case. Prospective buyers
are naturally interested in knowing how the benefit will be safeguarded.
What are the best practices for shielding data and workloads at the edge
against intrusion attempts? Can a modified device be detected and rendered
ineffective? Can problem edge nodes be dealt with? Customers want to know
whether there are any special precautions or new processes they should learn
before venturing to the edge.
In an effort to answer these questions, this blog has evolved into a two-
part story, with a high-level overview in the first part and more specifics in
the second. Like this, IT safety is achieved.
Due to the nature of an edge solution, complete stack solutions are
uncommon. The typical infrastructure includes hardware from a server
provider, software from an IT technology supplier, sensors and devices from
a wide variety of manufacturers, and (perhaps) the network from a telecom
provider. It makes sense to rethink how we protect our IT networks now that
they are largely hosted in data centers and the cloud.
All the equipment in a factory, for instance, is usually hermetically sealed
so that no contaminating air may get in. In response, those who make indus-
trial machinery have come up with highly specialized software for monitoring
and controlling devices. We had to make those environments more open to
give people more options, and it left them more vulnerable to attacks from
hackers. Both Industry 4.0 and the Internet of Things played a role in this.
388 Edge Computing and Analytics: A New Computing Paradigm

16.20 Alternatives in Edge Architecture

16.20.1 A boundary architectural choice
There is a wide variety of novel solution architectures at the periphery.
Differences emerge as a natural consequence of the heterogeneity of edge
topologies’ many constituent parts, which might range from edge devices and
compute servers to networks and clouds.
The key design concerns and the reasoning behind the chosen solutions
are documented in well-known IT architectural decisions. In most cases, they
are deliberate choices made throughout the design process that affect either
the entire software system or one or more of its fundamental components and
connectors. The same is true for developing cutting-edge answers. Solution
architects are frequently put in the position of deciding which protocol, 5G
vs. Private 5G, or edge server is best suited to receive data from a certain
device. These, along with a great many other architectural details, require
careful thought and decision-making.
Every AD, along with its justification, ramifications, and any rebuttals,
needs to be recorded for future reference. In this article, we’ll discuss some
of the most frequently overlooked considerations while planning an edge
architecture. To restate, edge includes everything from devices at the far edge
to cloud-based processing and storage.

16.21 A Peripheral Approach to Automation

There are two ways to interpret the article’s title. For those in the telecom
industry, it’s analogous to edge automation of networks. In the enterprise
IoT/edge computing space, it evokes images of autonomous robots and other
gadgets. Both statements are correct and relevant since network automa-
tion at the network edge is required to enable the autonomous devices at
the network’s very tip (i.e., the device edge). Network automation in the
telecom business is enabled by 5G and software-defined networks (SDNs),
while automation at the enterprise edge is enabled by smart devices, app
auto-deployment, and AI/ML inferencing.
As mentioned in a previous article titled “5G at the Edge,” edge com-
puting and 5G connection offer near real-time automation, unlocking oppor-
tunities to enhance digital experiences and performance at the edge. The
topology diagram in Figure 16.1 should help you recall this. The challenge
is in determining which part to automate first and how to do it in a way that
causes minimal disturbance.
 16.23 Problems and Obstacles 389

16.22 The Edge and Data Sovereignty

Data sovereignty and sovereign clouds may seem at odds with edge com-
puting and distributed clouds; however, this is not the case. The location
of the command and control plane under administration appears to be the
top priority. The cloud control aircraft is often asked to operate under strict
limits set by customers. So long as the edge sites don’t leave the country’s
borders, they can take part in the distributed cloud model while still meeting
the sovereignty condition. Some examples will help:
• An international, remotely managed control plane flies beyond national
airspace.
• Both remote-controlled and management-controlled aircraft are permit-
ted within the territorial airspace of a nation.

16.23 Problems and Obstacles

The following are some of the most pressing concerns and difficulties
associated with edge analytics:
Workload optimization involves taking into account energy consump-
tion, latency, bandwidth, and construction and maintenance costs in order
to determine how best to optimize the workload. Since edge devices are
compact, their resources, battery, etc., are also compact. The complexity of
the optimization problem increases as a result of the additional constraints
placed on the four parameters.
Edge devices have limited memory because of their small form factor,
which prevents them from processing or generating large files. This is prob-
lematic for both the input and the output, as data training for predictive edge
analytics requires a large number of datasets, but edge analytics can only use
a subset of these data for training.
Edge devices’ limited storage space means that any data created by their
sensors must be filtered and sent to the cloud for safekeeping; otherwise, the
data will be destroyed from the edge devices to avoid memory overflow and
to protect users’ anonymity.
Powered by its own energy supply, computation and data analytics at
the edge can function independently of a central data center. The small size
of edge devices precludes the installation of large-sized batteries; hence,
batteries used in edge analytics must be compact, secure, long-lasting, and
rechargeable. One option is to use energy harvesting. The energy harvest-
ing system includes the energy harvester, a power regulator, and a storage
390 Edge Computing and Analytics: A New Computing Paradigm

capacitor. Nonetheless, the output power is typically low and erratic, which
is a major drawback. Because of this, there are constant blackouts. Due to a
loss of power, the current execution has been aborted. Non-volatile memory
is the answer to this problem. Process state and the currently used stack can
be saved in non-volatile memory thanks to non-volatile processors (NVM).
Checkpointing is the term for this process. Power management presents
another challenge, as does balancing hardware and software needs. The
majority of the time and energy spent while writing to NVM is caused by
software. In hardware, the capacitor is used to hold the charge while waiting
for the power supply to be restored. Another source of energy waste is the
capacitor. More energy is consumed during charging if the capacitor’s desired
charging voltage is high.
Edge devices receive raw data, most of which is redundant. As a result,
edge analytics requires the use of data-filtering techniques. In addition, the
raw data needs to be transformed in some way before it can be used by the
analytics program.
Powerful computational applications, such as deep learning, are being
implemented by edge analytics. These highly computational apps cannot
be deployed on edge devices because they lack the necessary infrastruc-
ture. As an added bonus, edge analytics allows for the use of numerous
computationally-intensive algorithms in a single edge device. Therefore, it
is challenging to schedule and manage resources for these algorithms.
In the case of computational offloading, the most crucial task is the
construction of an optimal decision engine. The decision engine evaluates
each task and decides whether or not to parallel split the computation and
offload it to a remote server or the cloud. Due to the complexity of the
decision-making process, offloading adds cost (e.g., offload time, latency,
computation time).

16.24 Conclusion
When considering the speed at which the globe and its data are expanding, it
becomes clear just how sluggish the cloud may be. To address this issue,
a new type of technology called “edge computing” is being developed to
place users’ devices in physical proximity to cloud servers. In this work,
we demonstrate how edge devices can perform analytics with the help of
edge analytics. Several edge applications that can process text, images, and
 16.24 Conclusion 391

videos are also discussed. Using data filtering performed at the edge, cloud
storage and bandwidth are conserved. In addition, it explains how machine
learning and AI algorithms may make edge analytics smarter, allowing them
to function autonomously (as in industry) while still aiding healthcare and
smart devices. But edge analytics is just getting started; it faces several
obstacles in its infancy. The phrase “need to be addressed” implies that the
challenges are urgent or important and cannot be ignored or postponed.
 Index

A Emotion recognition 237, 241, 242

ANN 225, 233, 240 Explore 165, 202, 299, 362
Arduino 257, 266 Edge computing 20, 193, 347, 355,
Artificial intelligence 27, 61, 123, 363
239
Automatic 138, 256, 263 F
Automation 12, 45, 86, 257, 353 Faster 60, 111, 153, 201, 267
Flood 199, 211, 384
B
Business intelligence 77, 238, 362 H
Blockchain 13, 195, 314, 379 Hand hygiene 258, 260, 267
Hazard 163, 183, 310
C Hydrodynamic 208, 212, 219
CRM 93, 251
Circular economy 37, 164, 172 I
Cloud computing 6, 61, 309, 321, IIoT 2, 169, 182, 308
366 Industrial IoT 6, 51, 78, 322
Computer vision 86, 135, 142, 379 Industrial automation 17, 72, 185,
Conventional system 258, 280 355
Convolutional neural network 153, Industry 4.0 6, 30, 129, 307
204, 251 Industry 5.0 58, 77, 80
Cyber-physical system 60, 77, 164 Industry automation 80, 285, 354
Cybersecurity 113, 306, 318 Internet of Everything 168, 312
IoT 5, 18, 78, 168, 299
D IoT/M2M 309
Digital Twin 29, 34, 340, 364
Data analytics 10, 89, 115, 336, 375 M
Machine learning 27, 70, 107, 284,
E 381
ERP 181 Machine to machine (M2M) commu-
Earthquake 200, 208 nication 169, 326

393
394 Index

Machine-to-machine communication S
164, 184 Smart production 34, 62
Meteorological factor 225 Street canyon 224, 234
mMIMO 273, 284, 298 Superimposed pilot 273, 287, 296

N T
Neural network 88, 120, 143, 245, Threats 29, 114, 308, 383
358
U
P Use cases 28, 40, 97, 309
Pandemic 150, 256, 308
Particulate matter 226 W
Performance metrics 155 Waste-to-energy approach 178
Photodiode 257, 264, 267
Piles 207, 211, 219
Precoding schemes 280, 292, 299

Q
QoE 273, 282
Quantum 102, 109, 112, 121, 130

R
Real-time 6, 42, 203, 285, 358
Robotic process automation 44, 76,
77
 About the Editors

Pethuru Raj PhD works as a chief architect at Reliance Jio Platforms

Ltd. (JPL) Bangalore. Previously, he worked as cloud architect in the IBM
global Cloud Center of Excellence (CoE), an enterprise architect in Wipro
consulting services (WCS), and software architect in Robert Bosch Cor-
porate Research (CR). In total, he has gained more than 22 years of IT
industry experience and 8 years of research experience. He finished his
CSIR-sponsored Ph.D. degree at Anna University, Chennai and continued
with UGC-sponsored postdoctoral research in the Department of Computer
Science and Automation, Indian Institute of Science (IISc), Bangalore. There-
after, he was granted a couple of international research fellowships (JSPS and
JST) to work as a research scientist for 3.5 years in two leading Japanese
universities.

Dr. Abhishek Kumar is currently working as an Assistant Direc-

tor/Associate Professor in the Computer science & Engineering Department
in Chandigarh University, Punjab, India. He has a doctorate in computer
science from University of Madras and is post-doctoral fellow in Ingenium
Research Group Ingenium Research Group Lab, Universidad De Castilla-La
Mancha, Ciudad Real, and Ciudad Real Spain. He has total of more than
11 years of academic teaching experience along with 2 years of teaching
assistantship. He has more than 100 publications in reputed, peer reviewed
national and international journals, books & conferences. He has guided more
than 30 [Link] and M.S. Projects at national and international level and
guided 6 Ph.D. Scholars at national and international Level. He is acting
as Series Editor for three books series, Quantum Computing with Degruyter
Germany, Intelligent Energy System with Elsevier, and Sustainable Energy
with Nova, USA.

Dr. T. Ananth Kumar works as Associate Professor and R & D head in

IFET college of Engineering (Autonomous) affiliated to Anna University,
Chennai. He received his Ph.D. degree in VLSI Design from Manonmaniam

395
396 About the Editors

Sundaranar University, Tirunelveli. He received his master’s degree in VLSI

design from Anna University, Chennai and bachelor’s degree in electronics
and communication engineering from Anna University, Chennai. He has pre-
sented papers in various national and international conferences and journals.
His fields of interest are networks on chips, computer architecture and ASIC
design. He has received awards such as Young Innovator Award, Young
Researcher Award, Class A Award – IIT Bombay and Best Paper Award at
INCODS 2017. He is the life member of ISTE, IEEE and few membership
bodies. He has many patents in various domains. He has edited four books
and has written many book chapters for Springer, IET Press, and Taylor &
Francis press.

Dr. Neha Singhal has about 13 plus years of experience in teaching. She is
presently working as an Assistant Professor, Dept. of Computer Science at
Christ University, Central Campus, Bangalore, India. She obtained her Ph.D.
Degree from VTU, Belgaum and M. Tech in Computer Science degree from
Banasthali University, Rajasthan. Her teaching and research interests are in
the field of web services, AI,ML and IoT. She is a professional member
of ISTE and IEEE society. Dr. Singhal has published research papers and
book chapters in various Scopus-indexed and ESCI indexed journals. She
has delivered technical talks and has been invited as a resource person to
several colleges in Bangalore and India. She received four government funded
projects from various funding agencies during 2018 to 2020. She has received
an “Appreciation Award” for her exemplary services at RRCE.