DEPARTMENT OF MECHANICAL ENGINEERING

RV COLLEGE OF ENGINEERING®
(Autonomous Institution affiliated to VTU, Belagavi)

PROJECT REPORT

On

Development of a Predictive Maintenance System using Edge AI for

Monitoring Motor Vibration and Temperature
Submitted in partial fulfilment of First year IDEA Lab

Submitted by

Aditi P Rao 1RV24CD001

Ramya Bhat 1RV24CD040
Varshini S 1RV24CD057
Yarragunta Chandrakala 1RV24CD062

Under the guidance of

Prof. Jinka Ranganayakulu Dr. Prapul Chandra A C

Assistant Professor Assistant Professor
Dept. of Mechanical Engineering. Dept. of Mechanical Engineering.

2024 – 2025
 RV COLLEGE OF ENGINEERING®, BENGALURU–560059
(Autonomous Institution Affiliated to VTU, Belagavi)
DEPARTMENT OF MECHANICAL ENGINEERING

CERTIFICATE

This is to certify that the project work titled “Development of a Predictive Maintenance System
for Motors using Vibration and Temperature Monitoring with Edge AI” carried out by Aditi P
Rao (1RV24CD001), Ramya Bhat (1RV24CD040), Varshini S (1RV24CD057), Yarragunta
Chandrakala (1RV24CD062), in partial fulfilment of the IDEA Lab (ME121DL) course
introduced in the first year for all branches, during the academic year 2024-2025. It is certified
that all necessary corrections/suggestions indicated during internal assessment have been duly
incorporated in the final report. The project work has been approved as it satisfies the academic
requirements in respect of the IDEA Lab Experiential Learning Project as prescribed by the
institution.

Prof. Jinka Ranganayakulu Dr. Prapul Chandra A C Prof. & Head

Assistant Professor Assistant Professor
Department of Mechanical Department of Mechanical
Department of Mechanical Engineering Engineering
Engineering

External Examiners

Name of Examiners Signature with Date

1. __________________ _______________________

2. __________________ _______________________
 ABSTRACT
This project presents the development of a low-cost, real-time predictive maintenance system
designed for industrial machinery using an Arduino-based IIoT framework. The system is
engineered to detect and classify early-stage faults through continuous monitoring of temperature
and vibration data, captured using the DS18B20 temperature sensor and the MPU6050 inertial
measurement unit, respectively. Traditional maintenance strategies often rely on time-based
scheduling or reactive repairs after failures, leading to unexpected downtime and increased
operational costs. This system addresses those challenges by enabling proactive fault detection
using sensor fusion and Edge AI.

Data collected under normal and simulated fault conditions were used to train a lightweight
machine learning model using the Edge Impulse platform. The model was optimized and
deployed directly on the Arduino microcontroller to perform on-device inference. The system
classifies input data into three conditions: normal operation, temperature anomaly, and vibration
anomaly. Upon detecting a fault, it can trigger a relay to shut down equipment and optionally
send data to a cloud dashboard using Wi-Fi (ESP8266/ESP32), enabling IIoT-based alerts and
remote monitoring.

The system demonstrated over 90% classification accuracy during testing, with reliable fault
detection in real time. Its impact lies in minimizing machine downtime, improving safety, and
enabling affordable predictive maintenance for small to medium-scale industries. The project
showcases a scalable framework that combines embedded systems, machine learning, and IoT to
move from reactive to predictive maintenance strategies.

.
 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

TABLE OF CONTENTS
Title Page No.

COVER PAGE i

CERTIFICATE ii

ABSTRACT iii

TABLE OF CONTENTS iv

CHAPTER 1: INTRODUCTION 1

1.1 Background 1

1.2 Problem Definition 1

1.3 Motivation 1

1.4 Objectives 1

1.5 Scope 2

CHAPTER 2: LITERATURE REVIEW 2

2.1 Overview of Existing Technologies 2

2.2 Prior Art and Research Studies 2

2.3 Gaps Identified 3

2.4 Research Opportunity 4

CHAPTER 3: PROJECT METHODOLOGY 4

3.1 Overview 4

3.2 Component Selection 5

3.3 Block Diagram 5

3.4 Working Principle 6

3.5 Design Considerations 6

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 4

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

CHAPTER 4: HARDWRE AND SOFTWARE IMPLEMENTATION 6

4.1 Hardware Implementation 6

4.2 Circuit Diagram Description 7

4.3 Software Implementation 7

4.4 Integration Steps 8

CHAPTER 5: EXPERIMENTAL SETUP 9

5.1 Testing Objectives 9

5.2 Testing Procedure 9

5.3 Results Captured 10

5.4 Performance Graphs 10

5.5 Output Images 12

CHAPTER 6: RESULTS AND DISCUSSION 15

6.1 Interpretation of Results 15

6.2 Key Findings 15

6.3 Limitations 16

CHAPTER 7: CONCLUSION AND FUTURE WORK 17

7.1 Conclusion 17

7.2 Future Work 17

7.3 Final thoughts 18

REFERENCES 18

Appendices – Arduino code snippets 19

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 5

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 1
INTRODUCTION

1.1 Background

In industrial environments, machinery and equipment are critical assets that require
regular maintenance to ensure safe and efficient operation. Traditional maintenance
strategies are either time-based (preventive) or failure-based (corrective), both of
which have significant drawbacks. Preventive maintenance often leads to
unnecessary service and resource use, while failure-based approaches result in costly
downtime and safety risks. Predictive maintenance, enabled by real-time monitoring
and data-driven decision-making, offers a smarter alternative. Recent advancements
in embedded systems, affordable sensors, and edge AI make it feasible to implement
intelligent fault detection at low cost and low complexity.

1.2 Problem Definition

Most small and medium-scale industries lack access to high-end monitoring systems
due to cost and complexity. They rely heavily on manual inspection or fixed
maintenance cycles, which cannot detect subtle changes in machine condition. This
leads to unplanned failures, increased repair costs, and reduced operational
efficiency. A compact, low-power system that can monitor physical parameters such
as temperature and vibration and intelligently identify early-stage faults is essential
for bridging this gap.

1.3 Motivation

This project is inspired by the principles of Industry 4.0, which promotes smart
automation and intelligent data processing in manufacturing. The goal is to bring
predictive maintenance to resource-constrained environments using simple and cost-
effective tools. Arduino boards, when combined with sensor fusion and on-device
AI, offer an ideal platform for this purpose. The ability to locally process data and
respond to anomalies without relying on cloud servers makes the system fast,
reliable, and scalable.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 6

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

1.4 Objectives

➢ Design a compact hardware system for real-time machine condition

monitoring.
➢ Interface vibration and temperature sensors with Arduino to collect
operational data.
➢ Develop a machine learning model capable of classifying operational states.
➢ Deploy the model on the microcontroller using Edge Impulse/TinyML
frameworks.
➢ Trigger control actions (e.g., relay shutdown) and enable IIoT features for
remote alerts.

1.5 Scope

This project is limited to detecting temperature and vibration anomalies in single-

phase machinery setups. It uses the MPU6050 for multi-axis vibration sensing and
DS18B20 for temperature monitoring. A lightweight classification model is trained
on labelled sensor data and deployed on the Arduino platform. The system
demonstrates edge-based inference and IIoT communication (via ESP8266/ESP32)
to a cloud dashboard. Though currently limited to two types of sensors and three
conditions (normal, temperature anomaly, vibration anomaly), the architecture
allows future integration of more parameters (e.g., current, pressure) and extension to
industrial-grade systems.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 7

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 2
LITERATURE REVIEW

2.1 Overview of Existing Technologies

Predictive maintenance (PdM) is a modern maintenance strategy that uses sensor

data, analytics, and machine learning to forecast equipment failures before they
happen. Over the past two decades, industries have gradually moved from traditional
preventive maintenance to more intelligent, data-driven predictive approaches. Many
high-end systems employ advanced technologies like industrial-grade
accelerometers, thermographic cameras, and programmable logic controllers (PLCs)
in conjunction with cloud-based machine learning platforms.
Some of the key technologies used in current PdM systems include:
➢ Vibration Analysis Tools: High-precision piezoelectric accelerometers are
commonly used in large-scale industrial setups to detect imbalance,
misalignment, and bearing faults. These are often paired with spectrum
analysers or FFT-based software tools.
➢ Infrared Thermography: Non-contact infrared cameras monitor
temperature anomalies in rotating or electrical equipment and are widely used
for fault detection in transformers and motors.
➢ Cloud-Based Predictive Platforms: Many industries rely on platforms like
IBM Maximo, GE Predix, or Siemens MindSphere, which use sensor data
streamed to cloud servers where ML algorithms run on large datasets.
➢ Edge AI and Embedded ML: Recently, microcontroller-based systems (e.g.,
those using Raspberry Pi or Arduino) have begun supporting Edge ML
models for localized anomaly detection, offering a cost-effective and
responsive alternative to cloud dependency.

2.2 Prior Art and Research Studies

Several academic studies and industry reports have shown the effectiveness of
combining vibration and temperature data to predict faults in rotating machinery:

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 8

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

➢ A 2020 study published in IEEE Transactions on Industrial Informatics

demonstrated that integrating accelerometer and temperature data using
neural networks significantly improved fault classification accuracy.
➢ Research from MIT's Industrial Performance Center explored real-time data
logging on embedded devices and emphasized the importance of Edge AI for
latency-sensitive environments.
➢ Commercial implementations such as Bosch's Vibration Monitoring Unit
have set benchmarks for early failure detection in motors using high-end
MEMS sensors and AI.

2.3 Gaps Identified

Despite significant progress in the field, several key gaps remain—especially for
small to medium enterprises:
➢ Cost and Accessibility: Many predictive maintenance systems are
prohibitively expensive and require high-end infrastructure, making them
inaccessible to smaller businesses.
➢ Lack of Edge-Level Inference: Most systems rely on cloud platforms for
model training and inference, leading to latency and reliance on continuous
internet connectivity.
➢ Sensor and Hardware Complexity: High-precision industrial sensors often
require skilled installation and calibration, limiting widespread deployment in
smaller setups.
➢ Limited Customizability: Existing commercial systems are often rigid and
lack the flexibility to be adapted for varied use-cases or scaled modularly.
➢ Underutilization of Arduino-Class Devices: While Arduino and similar
boards are widely available, their potential for real-time anomaly detection
using AI is underexplored in literature and practice.

2.4 Research Opportunity

This project bridges the above gaps by implementing a low-cost, edge-enabled,

Arduino-based predictive maintenance system. It uses common sensors (MPU6050

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 9

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

and DS18B20), trained via accessible tools like Edge Impulse, and includes built-in
IIoT capabilities for real-time remote alerts. It brings intelligent maintenance
monitoring within reach for educational institutions, start-ups, and MSMEs.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 10

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 3

SYSTEM DESIGN AND ARCHITECTURE

3.1 Overview

The system is designed to monitor real-time temperature and vibration data from
industrial machinery and detect early-stage anomalies using an Edge AI model. It
consists of five functional layers:
➢ Sensing Layer
➢ Data Processing Layer
➢ Decision-Making Layer (ML Model)
➢ Actuation Layer (Relay, Alert)
➢ Communication Layer (IIoT/Cloud)
Each component is selected to ensure low cost, energy efficiency, and compatibility
with edge inference using microcontrollers.

3.2 Component Selection

This table outlines the key components chosen for the system based on their
functionality, efficiency, and compatibility. Each component—ranging from
microcontrollers to sensors—was selected to ensure low power usage, real-time
monitoring, and seamless connectivity.

Component Function Reason for Selection

Arduino Low power, open source, large

Main microcontroller
Uno/Nano/ESP32 community support

Vibration sensing 3-axis accelerometer and

MPU6050
(Accelerometer + Gyroscope) gyroscope in one chip

Accurate, digital one-wire

DS18B20 Temperature sensor
communication

Relay Module Actuation Simple digital control, isolatable

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 11

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

3.3 Block Diagram

This block diagram illustrates the overall workflow of an Arduino-based IIoT

predictive maintenance system. It begins with data collection under normal and
abnormal conditions, followed by preprocessing and model training. The trained
model is integrated with Arduino for edge AI, enabling cloud connectivity,
predictive strategies, and final output actions like alerts and logging.

3.4 Working Principle

➢ Sensor Data Acquisition:

• The DS18B20 records temperature at regular intervals.
• MPU6050 captures accelerometer data in 3 axes (X, Y, Z).
• Both sensors are read via Arduino’s I2C and OneWire interfaces.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 12

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

➢ Edge Inference:
Sensor readings are fed into a pre-trained ML model deployed using Edge
Impulse or TensorFlow Lite for Microcontrollers.
The model classifies the current condition as:
• Normal
• Temperature Anomaly
• Vibration Anomaly
➢ Decision Making and Actuation:
• Based on model output:
If “Normal”: no action taken.
If anomaly: a relay is activated to cut off a load or send an alert.
• A local display can show live status and class label.
➢ IIoT Communication:
If an ESP8266/ESP32 module is integrated, the system:
Sends real-time data to platforms like ThingSpeak, Firebase, or via MQTT to
Node-RED.
Enables cloud dashboards, remote alerts, or logging.

3.5 Design Considerations

➢ Real-time Response: Ensured by using on-device ML inference.

➢ Scalability: Can add more sensors (e.g., current sensor, pressure) in future.
➢ Low Power: Ideal for battery-operated or embedded environments.
➢ Cost-Effective: All components are affordable and open-source compatible.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 13

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 4

HARDWARE AND SOFTWARE IMPLEMENTATION

4.1 Hardware Implementation

4.1.1 Circuit Details

The hardware setup integrates key components for sensor monitoring, load
simulation, machine learning inference, and cloud communication. A DC motor is
included to simulate industrial machinery, and is controlled via a relay module for
automated shutdown during abnormal conditions.

Major Components and Interfacing:

This table lists the major hardware components interfaced with the Arduino, along
with their corresponding pins and functions. It includes essential modules for sensing
vibration, temperature, and controlling a DC motor through a relay. Additionally, it
supports Wi-Fi communication and real-time display for effective predictive
maintenance.

Component Arduino Pin Purpose

MPU6050 A4 (SDA), A5 (SCL) Measures 3-axis vibration and gyroscope

DS18B20 D2 (Data) Measures surface temperature

Relay Module D8 Controls power to the DC motor

DC Motor Relay-controlled external circuit Simulates rotating machinery

ESP8266/ESP32 TX/RX (or native Wi-Fi) Sends data to the cloud

OLED Display A4/A5 Real-time condition display

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 14

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Power Configuration:
The DC motor is powered via a separate 9V/12V supply connected through the relay.
Arduino is powered via USB or external 5V adapter.
Sensors and relay share a common ground for stable operation.

4.2 Circuit Diagram Description

Key Highlights:

➢ MPU6050 and OLED share the I2C bus.

➢ DS18B20 uses a pull-up resistor between its data and VCC lines.
➢ The relay switches the DC motor’s positive terminal; the common (COM) is
connected to +V, and Normally Open (NO) goes to the motor’s power pin.
➢ Ground of the motor power supply is connected to Arduino GND for proper
relay control.

4.3 Software Implementation

4.3.1 Development Tools

➢ Arduino IDE: Program and upload firmware
➢ Edge Impulse Studio: Train ML model and export as .h file
➢ ThingSpeak / Firebase / MQTT (Optional): IIoT cloud integration
➢ Python + Excel: For CSV dataset creation and preprocessing
4.3.2 Functional Modules
➢ Sensor Reading
MPU6050: Reads real-time vibration values
DS18B20: Captures temperature in °C
➢ Feature Calculation
Compute vibration magnitude (RMS or raw XYZ)
Prepare feature vector for inference
➢ Model Inference
• Uses Edge Impulse/TinyML .h model
• Model classifies the machine state as:
0 = Normal

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 15

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

1 = Temperature Anomaly
2= Vibration Anomaly

➢ Motor Control
• If anomaly detected:
Relay activates → DC motor shuts off
• If normal:
Relay stays inactive → Motor keeps running
➢ IIoT Integration (Optional)
• ESP8266/ESP32 sends:
Sensor readings
Predicted status
• Platform: ThingSpeak, Firebase Realtime DB, or MQTT Broker +
Node-RED

4.4 Integration Steps

➢ Build the circuit on breadboard or PCB.

➢ Collect sensor data under:
• Normal condition (motor running freely)
• Heat anomaly (apply heat to DS18B20)
• Vibration anomaly (disturb motor rotation)
➢ Label and export data as CSV using Excel.
➢ Upload data to Edge Impulse, train model, and download .h file.
➢ Embed model in Arduino sketch using #include "model.h".
➢ Upload code to Arduino. Observe:
• Real-time prediction
• Automatic motor shutdown via relay
• Optional data upload to cloud

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 16

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 5

HARDWARE AND SOFTWARE IMPLEMENTATION

5.1 Objective of Testing

The aim of the experimental setup was to collect real-time sensor data under various
machine conditions, label the data accordingly, and train a machine learning model
capable of predicting potential anomalies. The system's goal was to identify:
➢ Normal operating conditions
➢ Temperature anomalies
➢ Vibration anomalies
This setup was meant to replicate industrial machine behavior in a scaled-down lab
environment using a DC motor, DS18B20 temperature sensor, and MPU6050 IMU
sensor.

5.2 Testing Procedure

Step 1: Hardware Setup

➢ DS18B20 and MPU6050 sensors were mounted near a running DC motor.
➢ Sensor data was read using an Arduino and printed via Serial Monitor.
➢ A relay was connected for response simulation (cutoff).

Step 2: Data Collection

➢ Sensor readings were collected under three conditions:
➢ Normal: Motor running with no load or external heat.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 17

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

➢ Temperature Anomaly: External heat applied to DS18B20 to simulate

overheating.
➢ Vibration Anomaly: Vibration induced by unbalancing the motor shaft
(weights or obstruction).
Collected parameters:
• Temperature (°C)
• Accelerometer X, Y, Z
• Gyroscope X, Y, Z
~97 labelled data points were recorded in Excel and labelled manually.
Step 3: Feature Engineering
➢ Acceleration Magnitude calculated from √(AccX² + AccY² + AccZ²)
➢ Features used for training: Temp (°C) and Accel_Magnitude
Step 4: Preprocessing
➢ Checked for missing values (none found)
➢ Scaled features using StandardScaler for normalization
Step 5: Train-Test Split
➢ Data split: 77 training samples, 20 test samples
➢ Stratified to maintain class proportions

5.3 Data Collected Summary

This table summarizes the collected dataset comprising 97 samples categorized into
normal, temperature anomaly, and vibration anomaly conditions. It provides a
balanced foundation for training and testing predictive maintenance models.

Condition Number of Samples

Normal 44

Temp_Anomaly 31

Vib_Anomaly 22

Total 97

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 18

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

5.4 Performance graphs

Here are the key performance insights based on our training output:
Label Distribution:
➢ Normal: 44 samples
➢ Temp_Anomaly: 31 samples
➢ Vib_Anomaly: 22 samples

Confusion matrix:
This confusion matrix represents the performance of the classification model across
three classes: Normal, Temperature Anomaly, and Vibration Anomaly. The model
accurately predicted 44 normal, 31 temperature anomaly, and 21 vibration anomaly
samples. There was only one misclassification where a vibration anomaly was
labeled as normal. Overall, the matrix indicates high accuracy and strong class
separation in predictions.

Classification Report:
This classification report summarizes the model’s performance using precision,
recall, and F1-score for each class. It shows perfect detection for temperature

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 19

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

anomalies and strong performance for normal and vibration anomaly classes. The
overall model accuracy is 95%, indicating reliable fault classification.

Class Precision Recall F1-Score

Normal 0.90 1.00 0.95

Temp_Anomaly 1.00 1.00 1.00

Vib_Anomaly 1.00 0.80 0.89

Accuracy 0.95

Feature Plots:
1. Scatter Plot – Temperature vs Accel_Magnitude

This graph compares actual and predicted labels based on temperature and
acceleration magnitude features. The close alignment between both plots indicates
the model's strong prediction capability across normal, temperature anomaly, and
vibration anomaly conditions. Minimal misclassification suggests high model
accuracy and generalization. The scatter plots compare actual and predicted labels
using temperature and acceleration magnitude features. The close match between
both plots reflects the model’s strong classification accuracy and minimal
misclassification across all operational states.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 20

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

2. Histogram – Accel_Magnitude Distribution

This graph shows the distribution of acceleration magnitude categorized by anomaly

type. Most values cluster around a central magnitude, with outliers indicating
potential vibration anomalies. It helps visualize how different conditions impact
vibration intensity, aiding in anomaly detection.

3. Line Graph – Temp over Time

This line graph shows temperature variation over time, with data points colored by
anomaly type. It highlights a clear temperature rise during anomaly periods, aiding in
identifying abnormal thermal behaviour.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 21

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

5.5 Output Images

This graph shows the accelerometer readings (X, Y, Z axes) over time, with values
represented as data point indices. A clear spike and oscillation in all three axes after
the 70th index indicate strong vibrations, signaling potential anomalies. It effectively
illustrates dynamic motor behavior under different conditions.

The graph depicts the acceleration magnitude over time, represented as data point
indices. A noticeable spike in readings occurs after the 70th data point, indicating
increased vibrations and potential system anomalies. This pattern highlights dynamic
motor behavior under varying operational conditions. Such trends are crucial for
identifying faults and ensuring timely maintenance in electromechanical systems.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 22

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

The graph presents gyroscope readings (GyroX, GyroY, GyroZ) plotted against data
point indices. A noticeable rise in fluctuations occurs after the 70th data point,
indicating abnormal rotational motion in the system. This data helps in identifying
potential mechanical instabilities and anomalies in real-time.

The graph illustrates the gyroscope magnitude over data point indices, representing
overall rotational activity. A significant increase in magnitude is observed after the
60th data point, indicating abnormal rotational behavior. This trend is valuable for
detecting anomalies and potential faults in the motor system’s operation. Continuous
monitoring of such patterns aids in predictive maintenance and improves equipment
reliability.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 23

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

The correlation matrix displays the relationships between various sensor readings,
including temperature, accelerometer, and gyroscope values. Strong positive and
negative correlations are visible, with values close to +1 or -1 indicating strong direct
or inverse relationships.

The graph illustrates the decision boundaries of a Random Forest Classifier applied
to scaled temperature and vibration magnitude data. Different colored regions
represent classification zones for normal, temperature anomalies, and vibration
anomalies. The plot effectively visualizes how the model separates various
operational conditions based on sensor inputs.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 24

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

5.6 Summary

The experimental setup validated the system’s ability to:

➢ Collect clean, labelled sensor data
➢ Differentiate between normal, thermal, and mechanical anomalies
➢ Train an ML model with 95% accuracy
➢ Perform real-time inference on Arduino for smart machine monitoring
The process confirmed the system's readiness for deployment in low-cost, embedded
IIoT applications.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 25

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 6

EXPERIMENTAL SETUP
6.1 Interpretation of Results

The predictive maintenance model was trained using labeled sensor data collected
from the prototype system consisting of a DC motor, DS18B20 temperature sensor,
and MPU6050 accelerometer. After preprocessing and feature engineering, two key
features were selected:
➢ Temperature (°C)
➢ Vibration Magnitude (calculated as √(AccX² + AccY² + AccZ²))
These features were scaled using StandardScaler, and the dataset (97 samples) was
split into 77 for training and 20 for testing using stratified sampling to preserve label
distribution.
The classification report shows:
➢ Normal: Precision = 0.90, Recall = 1.00
➢ Temp_Anomaly: Precision = 1.00, Recall = 1.00
➢ Vib_Anomaly: Precision = 1.00, Recall = 0.80
This indicates that:
The model is highly reliable in detecting normal and temperature-based anomalies.
Graph:

Here’s the graph visualizing the Precision and Recall values for the three classes
from your classification report. It clearly shows strong performance across all classes.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 26

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

6.2 Key Findings

The table summarizes the key performance metrics and deployment details of the
developed predictive maintenance system. It highlights a test set accuracy of 95%,
with temperature anomalies achieving perfect precision. The model is optimized for
microcontroller deployment via micromlgen, ensuring efficient real-time inference
on embedded systems.

Parameter Value

Accuracy (Test Set) 95%

Input Features Temperature, Accel_Magnitude

Algorithm Random Forest Classifier (Scikit-learn)

Best Precision Class Temp_Anomaly (100%)

Most Confused Class Vib_Anomaly (Recall = 80%)

On-Device Inference Supported via micromlgen

Final Model Size Compact enough for microcontroller (optimize if needed)

6.3 Limitations

Despite strong performance, the system has a few constraints:

➢ Memory Usage on Arduino: Random Forest models may consume more
memory than available on basic boards like Arduino Uno. ESP32 or
simplified models like DecisionTreeClassifier may be needed in
production.
➢ Dataset Imbalance: There were fewer vibration anomaly samples, slightly
reducing model generalization for this class.
➢ Real-World Noise: The training data was collected in a controlled lab
setting; performance in noisy industrial environments may vary.
➢ Fixed Thresholds: Feature engineering (e.g., magnitude calculation) assumes
consistent orientation and sensor mounting.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 27

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

6.4 Summary

The trained machine learning model demonstrated excellent accuracy and on-device
compatibility, making it highly suitable for real-time predictive maintenance in
embedded systems. With minor improvements in data diversity and model size, this
solution can scale effectively into industrial applications.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 28

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Chapter 7

CONCLUSION AND FUTURE WORK

7.1 Conclusion

This project successfully demonstrates the design and implementation of a low-cost,

intelligent predictive maintenance system using Arduino, vibration and temperature
sensors, and Edge AI. By monitoring real-time data from an industrial DC motor
setup, the system was able to detect early signs of abnormality such as overheating
and excessive vibration. It classified machine conditions using a machine learning
model deployed directly on the Arduino, thus eliminating the need for continuous
cloud connectivity.
Key achievements include:
➢ Real-time fault detection through local inference.
➢ Relay-based motor control based on sensor data.
➢ Edge deployment of a classification model trained using Edge Impulse.
➢ Optional IIoT capability for remote monitoring via platforms like
ThingSpeak.
The system bridges the gap between traditional low-cost embedded systems and
modern Industry 4.0 requirements. It provides an affordable solution for MSMEs
(Micro, Small & Medium Enterprises) and educational labs to implement predictive
maintenance without relying on expensive infrastructure.

7.2 Future Work

While the current system is functional and reliable within its design scope, there are
several areas for future enhancement:
A. Hardware Improvements:
➢ Add more sensors such as:
Current sensors (e.g., ACS712) to monitor electrical load.
Microphones for acoustic monitoring.
Pressure/strain sensors for structural monitoring.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 29

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

➢ Shift to a more powerful board like ESP32 to support larger models and
higher sampling rates.
B. Software Enhancements:
➢ Use more advanced ML algorithms (e.g., 1D CNN, LSTM) for higher
accuracy.
➢ Implement adaptive learning that allows the system to re-train with new data.
➢ Include fail-safe modes to reduce false positives.
C. IoT and Cloud Integration:
➢ Develop a custom mobile/web dashboard for real-time visualization and
alerts.
➢ Integrate with platforms like Firebase, AWS IoT Core, or Blynk for better
scalability.
➢ Enable SMS/email notifications using cloud services or GSM modules.
D. Scalability & Deployment:
➢ Expand to monitor multiple machines simultaneously using a central
controller.
➢ Implement mesh networking (LoRa) for remote or large factory installations.
➢ Package the system into an industrial-grade enclosure with user interface for
plug-and-play deployment.

7.3 Final Thoughts

This project lays a strong foundation for future innovations in embedded predictive
maintenance systems. With further development, such systems can play a vital role
in minimizing downtime, extending machine lifespan, and increasing industrial
efficiency—all while remaining accessible to resource-constrained users.

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 30

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

REFERENCES
[1] S. Yin, X. Li, H. Gao, and O. Kaynak, “Data-based techniques focused on
modern industry: An overview,” IEEE Transactions on Industrial Electronics, vol.
62, no. 1, pp. 657–667, Jan. 2015.
https://doi.org/10.1109/TIE.2014.2308133
[2] K. Worden, G. Manson, and N. R. J. Fieller, “Damage detection using outlier
analysis,” Journal of Sound and Vibration, vol. 229, no. 3, pp. 647–667, Jan. 2000.
https://doi.org/10.1006/jsvi.1999.2441
[3] A. Susto, S. Schirru, S. Pampuri, S. McLoone, and A. Beghi, “Machine learning
for predictive maintenance: A multiple classifier approach,” IEEE Transactions on
Industrial Informatics, vol. 11, no. 3, pp. 812–820, June 2015.
https://doi.org/10.1109/TII.2014.2349359
[4] Edge Impulse. “Edge Impulse: Embedded Machine Learning Platform.” [Online].
Available: https://www.edgeimpulse.com/
[5] Arduino. “Arduino Uno Rev3.” [Online]. Available:
https://store.arduino.cc/products/arduino-uno-rev3
[6] Adafruit. “MPU6050 6-DoF Accelerometer and Gyroscope Sensor Datasheet.”
[Online]. Available: https://learn.adafruit.com/mpu6050-6-dof-accelerometer-and-
gyroscope

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 31

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Appendices – Arduino code snippets

Code Snippet for DS18B20 Temperature Sensor Testing:

Main Code:

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 32

 Development of a Predictive Maintenance System using Edge AI for
Monitoring Motor Vibration and Temperature

Expenditure Split:
Total
Component Description Qty
(INR)
Arduino Uno
Microcontroller board 1 ₹450
(Clone)
Vibration sensor
MPU6050 1 ₹200
(accelerometer + gyroscope)
DS18B20 Temperature sensor (digital) 1 ₹75
Breadboard +
For wiring the setup 1 ₹125
Jumper Wires
DC Motor (12V) To simulate rotating machine 1 ₹175
12V Adapter or Power supply for motor +
1 ₹150
Battery Pack Arduino
Mounting
Cardboard 1 ₹125
Platform
USB Cable (A to Arduino Uno programming
1 ₹75
B) cable
Buzzer / LED
For local alerts 1 ₹30
(Optional)

Data Pre-processing and Cleaning:

Department of Mechanical Engineering, Idea Lab, RVCE, Bengaluru Page 33