Manufacturing Letters 18 (2018) 20–23

Contents lists available at ScienceDirect

Manufacturing Letters
journal homepage: www.elsevier.com/locate/mfglet

Industrial Artificial Intelligence for industry 4.0-based manufacturing

a r t i c l e i n f o a b s t r a c t

Article history: The recent White House report on Artificial Intelligence (AI) (Lee, 2016) highlights the significance of AI
Received 20 July 2018 and the necessity of a clear roadmap and strategic investment in this area. As AI emerges from science
Received in revised form 2 September 2018 fiction to become the frontier of world-changing technologies, there is an urgent need for systematic
Accepted 9 September 2018
development and implementation of AI to see its real impact in the next generation of industrial systems,
Available online 10 September 2018
namely Industry 4.0. Within the 5C architecture previously proposed in Lee et al. (2015), this paper pro-
vides an insight into the current state of AI technologies and the eco-system required to harness the
Keywords:
power of AI in industrial applications.
Industrial AI
Industry 4.0
Ó 2018 Society of Manufacturing Engineers (SME). Published by Elsevier Ltd. All rights reserved.
Big data
Smart manufacturing
Cyber physical systems

1. Introduction to industrial Artificial Intelligence operation of industries in a flexible, efficient, and green way. Since
Industrial AI is in infancy stage, it is essential to clearly define its
Artificial Intelligence (AI) is a cognitive science with rich structure, methodologies and challenges as a framework for its
research activities in the areas of image processing, natural lan- implementation in industry. To this end, we designed an Industrial
guage processing, robotics, machine learning etc. Historically, AI ecosystem, which covers the essential elements in this space
Machine Learning and AI have been perceived as black-art tech- and provides a guideline for better understanding and implement-
niques and there is often a lack of compelling evidence to convince ing it. Furthermore, the enabling technologies that an Industrial AI
industry that these techniques will work repeatedly and consis- system can be built upon are described. Fig. 1-a provides a sche-
tently with a return on investment. At the same time, the perfor- matic comparison of the desired system performance of Industrial
mance of machine learning algorithms is highly dependent on a AI with other learning systems over time.
developer’s experience and preferences. Hence, the success of AI
in industrial applications has been limited. On the contrary, Indus- 2. Key elements in Industrial AI: ABCDE
trial AI is a systematic discipline, which focuses on developing, val-
idating and deploying various machine learning algorithms for The key elements in Industrial AI can be characterized by
industrial applications with sustainable performance. It acts as a ‘ABCDE’. These key elements include Analytics technology (A),
systematic methodology and discipline to provide solutions for Big data technology (B), Cloud or Cyber technology (C), Domain
industrial applications and function as a bridge connecting aca- knowhow (D) and Evidence (E). Analytics is the core of AI, which
demic research outcomes in AI to industry practitioners. can only bring value if other elements are present. Big data tech-
AI-driven automation has yet to have a quantitatively major nology and Cloud are both essential elements, which provide the
impact on productivity growth [1]. Besides present day industries source of the information (data) and a platform for Industrial AI.
are facing new challenges in terms of market demand and compe- While these elements are essential, domain knowledge and Evi-
tition. They are in need of a radical change known as Industry 4.0. dence are also important factors that are mostly overlooked in this
Integration of AI with recent emerging technologies such as Indus- context. Domain knowhow is the key element from the following
trial Internet of Things (IIoT) [3], big data analytics [4–6], cloud aspects: 1) understanding the problem and focus the power of
computing [7–9] and cyber physical systems [2,10–11] will enable Industrial AI into solving it; 2) understanding the system so that
right data with the right quality can be collected; 3) understanding
⇑ Corresponding author. the physical meanings of the parameters and how they are associ-
E-mail address: [email protected] (J. Singh). ated with the physical characteristics of a system or process; and

https://doi.org/10.1016/j.mfglet.2018.09.002
2213-8463/Ó 2018 Society of Manufacturing Engineers (SME). Published by Elsevier Ltd. All rights reserved.
 J. Lee et al. / Manufacturing Letters 18 (2018) 20–23 21

Industrial AI

Performance
(Systematic Learning Approach)
AI and Machine Learning Utilize New Value Creation
Knowledge/ using
Expert Systems (Rule-based) Technologies
Avoid Smarter Information
Expert’s Experiences For Value-added For Unknown Knowledge
Improvement

Solve
Problem Solving
Upgrade Utilize New Methods/
Experiences Through Continuous
Techniques to Solve
Discontinued Improvement and
The Unknown Problems
Standard Work
Time Visible Invisible

Fig. 1. a) Comparison of Industrial AI with other learning systems; b) The impact of Industrial AI: from solving visible problems to avoiding invisible.

4) understanding how these parameters vary from machine to Self-compare, Self-predict, Self-optimize and Resilience. This chart
machine. Evidence is also an essential element in validating Indus- also includes four main enabling technologies including Data Tech-
trial AI models and incorporate them with cumulative learning nology (DT), Analytic Technology (AT), Platform Technology (PT)
ability. By gathering data patterns and the evidence (or label) asso- and Operations Technology (OT). These four technologies can bet-
ciated with those patterns can only we improve the AI model to ter be understood when put in the context of the Cyber-Physical
become more accurate, comprehensive and robust as it ages. Systems (CPS), proposed in [2]. As depicted in Fig. 3, these four
Fig. 1-b shows how AI can drive us from visible space to invisible, technologies (DT, AT, PT and OT) are the enablers for achieving suc-
and from solving the problems to avoiding them before they cess in Connection, Conversion, Cyber, Cognition and Configura-
surface. tion, or 5C. This section of the paper provides a brief description
of each of the mentioned technologies.
3. Industrial AI eco-system
3.1. Data technologies (DT)
Fig. 2 shows the proposed Industrial AI ecosystem, which
defines a sequential thinking strategy for needs, challenges, tech- Data Technologies are those technologies, which enable suc-
nologies and methodologies for developing transformative AI sys- cessful acquisition of useful data with significant performance
tems for industry. Practitioners can follow this diagram as a metrics across dimensions. Therefore, it becomes a co-enabler of
systematic guideline for developing a strategy for Industrial AI the ‘Smart Connection’ step in the 5C architecture by identifying
development and deployment. Within the targeted industry, this the appropriate equipment and mechanism for acquiring useful
ecosystem defines the common unmet needs such as Self-aware, data. The other aspect of data technologies is data communication.

Fig. 2. Industrial AI Eco-system.

Fig. 3. Enabling technologies for realization of CPS in manufacturing.

Communication in Smart Manufacturing extends beyond the rela- the provided insight. This machine-to-machine collaboration can
tively straight-forward transfer of acquired data from its source to be between two machines in a shop floor, or machines in two dif-
the point of analysis. It involves: 1) Interaction between manufac- ferent factories far apart. They can share their experience on how
turing resources in the physical-space. 2) Transfer and storage of adjusting specific parameters can optimize performance, and
data from machines and the factory floor to the Cloud. 3) Commu- adjust their production based on the availability of other machines.
nication from physical space to cyber space. 4) Communication In an industry 4.0 factory, Operations technology is the last step
from the cyber-space to the physical-space. In addition, DT needs leading to the following four capabilities: 1) Self-aware 2) Self-
to address 3B issues of data systems, namely, broken, bad, and predict, 3) Self-Configure and 4) Self-Compare.
background of data [6].
4. Case Study: Intelligent spindle system
3.2. Analytics technologies (AT)
This section describes the application and implementation of
Analytics Technology converts the sensory data from critical the Industrial AI architecture framework described in Section 3
components into useful information. Data driven modeling uncov- to machine tool spindle of a Computer numerical control (CNC)
ers hidden patterns, unknown correlations and other useful infor- machine. In manufacturing industry, the health condition of the
mation from manufacturing systems. This information can be used machine tool spindle is of major importance and this case study
for asset health prediction, such as generating a health value or a aims to show how Industrial AI-powered by the four enabling tech-
remaining useful life value, which can be used for machine prog- nologies can provide a full solution for real-time monitoring and
nostics and health management. Analytic Technologies integrate performance prediction of a machine tool spindle. This system is
this information with other technologies for improved productivity designed to minimize maintenance costs and optimize product
and innovation. quality, simultaneously. Based upon Fig. 3, the first step in the
guideline is to consider the common unmet needs in this applica-
3.3. Platform technologies (PT) tion area. For addressing the unmet needs (a self-aware and self-
optimizing machine) the challenges of 1) data quality, 2) multi-
Platform technologies include the hardware architecture for regime complexity, 3) machine-to-machine variation, 4) incorpo-
manufacturing data storage, analysis and feedback. A compatible rating expert system and 5) complexity of multi-source data need
platform architecture for analyzing data is a major deciding factor to be considered. Fig. 4 provides an overview of how DT, AT, PT and
for realizing smart manufacturing characteristics such as agility, OT are utilized to address these challenges and develop an intelli-
complex-event processing, and so on. Three major types of platform gent spindle system.
configurations are generally found – stand-alone, embedded and
cloud. Cloud computing is a significant advancement in Information 5. Challenges of Industrial Artificial Intelligence
and Communication Technologies with regard to computational,
storage and servitization capabilities. The cloud platform can pro- The expectations from Industrial AI are versatile and enormous
vide rapid service deployment, high level of customization, knowl- and even a partial fulfilment of these expectations would represent
edge integration, and effective visualization with high scalability. unique and real challenges of applying AI to industries. Among the
existing challenges and complexities, the following ones are of
3.4. Operations technology (OT) higher importance and priority:

Operation technology here refers to a series of decisions made 5.1. Machine-to-machine interactions
and actions taken based on the information extracted from data.
While delivering machine and process health information to the While AI algorithms can accurately map a set of inputs to a set
operators is valuable, an Industry 4.0 factory goes beyond and of outputs, they are also susceptible to small variations in the
enables machines to communicate and make decisions based on inputs caused by variations from machine to machine. It needs to
 J. Lee et al. / Manufacturing Letters 18 (2018) 20–23 23

Fig. 4. Platform technology for intelligent spindle.

ensure that individual AI solutions do not interfere/conflict with ing, this paper aims to provide a guideline for strategizing the
the working of other systems, further down the line. efforts toward realization of Industrial AI systems.

5.2. Data quality References

[1] Lee K. Artificial intelligence, automation, and the economy. The White. House
AI algorithms require massive and clean data sets with mini- Blog; 2016.
mum biases. By learning from inaccurate or inadequate data sets, [2] Lee J, Bagheri B, Kao HA. A cyber-physical systems architecture for industry
the downstream results can be flawed. 4.0-based manufacturing systems. Manuf Lett 2015;3:18–23.
[3] Da Xu L, He W, Li S. Internet of things in industries: A survey. IEEE Trans Ind Inf
2014;10(4):2233–43.
5.3. Cybersecurity [4] Lee J, Lapira E, Bagheri B, Kao HA. Recent advances and trends in predictive
manufacturing systems in big data environment. Manuf Lett 2013;1(1):38–41.
[5] Shi J, Wan J, Yan H, Suo H. November. A survey of cyber-physical systems. In:
The increasing use of connected technologies makes the smart Wireless Communications and Signal Processing (WCSP), 2011 International
manufacturing system vulnerable to cyber risks. Currently, the Conference on. IEEE. 2011. p. 1–6.
[6] Lee J, Ardakani HD, Yang S, Bagheri B. Industrial big data analytics and cyber-
scale of this vulnerability is under-appreciated and the industry
physical systems for future maintenance & service innovation. Procedia CIRP
is not prepared for the security threats that exist [12]. 2015;38:3–7.
[7] Zhang L, Luo Y, Tao F, Li BH, Ren L, Zhang X, et al. Cloud manufacturing: a new
manufacturing paradigm. Enterprise Inf Syst 2014;8(2):167–87.
6. Conclusion [8] Wu D, Greer MJ, Rosen DW, Schaefer D. Cloud manufacturing: Strategic vision
and state-of-the-art. J Manuf Syst 2013;32(4):564–79.
[9] Yang S, Bagheri B, Kao HA, Lee J. A unified framework and platform for
As AI emerges from science fiction to become the frontier of
designing of cloud-based machine health monitoring and manufacturing
world-changing technologies, there is an urgent need for system- systems. J Manuf Sci Eng 2015;137(4):040914.
atic development and implementation of AI to see its real impact [10] Baheti R, Gill H. Cyber-physical systems. Impact Control Technol 2011;12
(1):161–6.
in the next generation of industrial systems, namely Industry 4.0.
[11] Leitao P, Karnouskos S, Ribeiro L, Lee J, Strasser T, Colombo AW. Smart agents
This study aims to define the term Industrial AI and put it into in industrial cyber–physical systems. Proc IEEE 2016;104(5):1086–101.
the perspective of Industry 4.0 paradigm. In addition, by providing [12] Tuptuk N, Hailes S. Security of smart manufacturing systems. J Manuf Syst
an overview of the Industrial AI eco-system in today’s manufactur- 2018;47:93–106.