CSE328 Term Project - Motor Vibration Monitoring

Project Summary

ESP32-based IoT system for real-time motor vibration monitoring, fault detection, and remote control. The project reads vibration data from an ADXL345 accelerometer and surface temperature from an MLX90614 IR thermometer. An Edge Impulse TinyML model runs on the ESP32 to classify motor behavior as STOP, NORMAL, or WARNING. Anomaly detection is used to catch unseen mechanical imbalance patterns and latch the system into FAULT. The OLED shows live state, confidence, anomaly, temperature, Wi-Fi, RSSI, cloud status, and IP information. Arduino IoT Cloud receives live telemetry and provides dashboard controls for motor ON/OFF and fault reset. A 5V relay cuts motor power when a fault is detected.

GitHub repository: https://github.com/aftekeli/CSE328-Term-Project

Repository Contents

firmware/
  MotorVibrationMonitoring.ino      Final ESP32 sketch
  thingProperties.h                 Arduino IoT Cloud variables and connection handler
model/
  ei-motor-vibration-monitoring-arduino-1_0_3.zip
data/
  STOP/                             70 CSV recordings
  NORMAL/                           70 CSV recordings
  WARNING/                          70 CSV recordings
docs/
  proposal.pdf
  system_architecture.drawio
  system_architecture.png
images/
  wiring_diagram.png
  hardware_setup.jpg
  dashboard.png
  oled_page_a.jpg
  oled_page_b.jpg
  oled_fault.jpg
  ei_*.png
videos/
  project_demo.mp4                  Live demonstration video

Components

Component	Type	Purpose
ESP32 Dev Board	Microcontroller	Runs sensor acquisition, TinyML inference, state machine, cloud sync, and actuator control
ADXL345	3-axis accelerometer	Primary vibration sensor for TinyML input
MLX90614 / GY-906	IR thermometer	Secondary sensor for motor surface temperature and overheat fault detection
SSD1306 128x64 OLED	Display	Shows live state, confidence, anomaly, temperature, motor status, Wi-Fi/RSSI, cloud status, and IP
5V relay module	Actuator	Switches motor power and cuts power in FAULT state
DC motor + propeller	Mechanical target	Simulates normal and abnormal vibration behavior
RGB LED	Status output	Green / blue / magenta / red / white state indication

System Architecture

The system has two phases. During training, vibration CSV files are collected from the ESP32 and uploaded to Edge Impulse to train the model. During runtime, the ESP32 samples the ADXL345 at 50 Hz, builds a 2-second feature window, runs the exported Edge Impulse model locally, publishes telemetry to Arduino IoT Cloud, and controls the motor through the relay.

Wiring

Module	Module Pin	ESP32 / Connection	Notes
ADXL345	VCC	3V3	Accelerometer supply
ADXL345	GND	GND	Common logic ground
ADXL345	SDA	GPIO21	Shared I2C data line
ADXL345	SCL	GPIO22	Shared I2C clock line
MLX90614	VCC	3V3	IR thermometer supply
MLX90614	GND	GND	Common logic ground
MLX90614	SDA	GPIO21	Shared I2C data line
MLX90614	SCL	GPIO22	Shared I2C clock line
OLED SSD1306	VCC	3V3	128x64 I2C OLED
OLED SSD1306	GND	GND	Common logic ground
OLED SSD1306	SDA	GPIO21	Shared I2C data line
OLED SSD1306	SCL	GPIO22	Shared I2C clock line
Relay module	IN	GPIO32	Active-low relay control in firmware
Relay module	VCC	Ext. PSU 5V	Relay module power from external supply
Relay module	GND	Ext. PSU GND	Relay module ground from external supply
Relay CH1	COM1	Ext. PSU 3.3V	Motor power source through relay
Relay CH1	NO1	Motor (+)	Motor positive terminal
DC Motor	(−)	Ext. PSU GND	Motor negative terminal
RGB LED	R	GPIO25	FAULT red channel
RGB LED	G	GPIO26	STOP green channel
RGB LED	B	GPIO27	NORMAL blue channel
RGB LED	Common (K)	GND	Common-cathode, connected to ESP32 GND

Firmware pin definitions are in firmware/MotorVibrationMonitoring.ino:

#define SDA_PIN 21
#define SCL_PIN 22
#define RELAY_PIN 32
#define LED_R 25
#define LED_G 26
#define LED_B 27

The ESP32 GND and the external power supply GND are connected together on the breadboard to form a common ground.

Cloud Setup

Platform: Arduino IoT Cloud.

The Cloud Thing exposes telemetry variables for dashboard monitoring and command variables for dashboard control.

Variable	Direction	Update	Purpose
systemState	READ	On change	Text state: STOP, NORMAL, WARNING, FAULT, SENSOR_ERROR
systemStateCode	READ	On change	Numeric state for charts: 0..4
faultReasonCloud	READ	On change	Fault source: NONE, VIB, TEMP, SENS, EI
motorRunning	READ	On change	Actual relay/motor state
faultLatchedCloud	READ	On change	Whether a fault is latched
objectTempCloud	READ	1 second	MLX90614 object temperature
anomalyCloud	READ	1 second	Edge Impulse anomaly score
confidenceCloud	READ	1 second	Model confidence percentage
motorCommand	READWRITE	On change	Dashboard motor ON/OFF command
resetFaultCommand	READWRITE	On change	Dashboard fault reset command

Dashboard widgets used:

Widget	Variable
Switch	motorCommand
Push button / switch	resetFaultCommand
Status / text	systemState, faultReasonCloud
Value cards	objectTempCloud, anomalyCloud, confidenceCloud
Chart	systemStateCode, anomalyCloud, objectTempCloud
Indicator	motorRunning, faultLatchedCloud

Dashboard evidence:

Edge Impulse Model

The model is exported as an Arduino library and stored in:

model/ei-motor-vibration-monitoring-arduino-1_0_3.zip

Training data is stored as balanced time-series CSV recordings:

Class	File Count	Meaning
STOP	70	Motor off / low vibration baseline
NORMAL	70	Motor running in normal condition
WARNING	70	Higher vibration condition

Impulse configuration:

Parameter	Value
Sampling frequency	50 Hz
Window size	2000 ms
Window increase	200 ms
Axes	accX, accY, accZ
Processing block	Spectral Analysis
Learning block	Classification
Anomaly block	K-Means anomaly detection

Model screenshots:

Firmware Behavior

The ESP32 firmware runs a non-blocking loop:

1. ArduinoCloud.update() keeps dashboard communication active.
2. Serial commands are handled locally: 1 starts the motor, 0 stops it, r resets a latched fault.
3. MLX90614 temperature is read every second and published to Arduino IoT Cloud.
4. ADXL345 acceleration is sampled every 20 ms, which gives 50 Hz sampling.
5. A 2-second feature buffer is passed to the Edge Impulse classifier.
6. The state machine stabilizes transitions with two consecutive confirmations.
7. The relay cuts motor power when vibration, temperature, sensor, or inference fault is detected.

State behavior:

State	Trigger	OLED	RGB LED	Relay
STOP	Motor off or STOP classification	STOP, confidence, anomaly, temperature, network page	Green	OFF
NORMAL	Normal model output	NORMAL live status	Blue	ON
WARNING	Warning model output below fault threshold	WARNING live status	Magenta	ON
FAULT	Anomaly above threshold twice, overtemperature, sensor fault, or inference error	Full-screen motor stopped warning	Red	OFF
SENSOR_ERROR	Repeated I2C failures	Sensor error warning	White	OFF

Fault safety rules:

Rule	Implementation
Motor cannot start while a fault is latched	motorStart() forces relay OFF and clears motorCommand
Motor OFF ignores model output	State remains STOP unless fault is latched
Startup transient is ignored	3-second grace period after motor start
Vibration fault requires confirmation	Anomaly must exceed threshold for 2 consecutive inferences
Relay is fail-safe on fault	latchFault() immediately calls setRelayRaw(false)

Final firmware thresholds:

Constant	Value
OBJECT_TEMP_FAULT_C	32.0 C
ANOMALY_THRESHOLD	2.5
MOTOR_FAULT_GRACE_MS	3000 ms
STATE_CONFIRM_COUNT	2
VIB_FAULT_CONFIRM_COUNT	2

OLED Display

The OLED is not a title screen; it shows live operating data. In normal operation, it alternates between Page A and Page B every 3 seconds.

Page A:

OLED Item	Meaning
State	Large state label: STOP, NORMAL, or WARNING
Temperature + anomaly	MLX90614 object temperature and Edge Impulse anomaly score
Confidence + motor	Classification confidence and relay/motor state
Wi-Fi + cloud + RSSI	Wi-Fi status, Arduino Cloud status, and signal strength

Page B:

OLED Item	Meaning
State	Large state label: STOP, NORMAL, or WARNING
Temperature + anomaly	MLX90614 object temperature and Edge Impulse anomaly score
Confidence + motor	Classification confidence and relay/motor state
IP address	Full local IP address for network verification

Evidence:

How To Run

1. Install Arduino IDE.
2. Install ESP32 board support through Arduino Boards Manager.
3. Install required Arduino libraries:

ArduinoIoTCloud
Arduino_ConnectionHandler
Adafruit SSD1306
Adafruit GFX Library
Adafruit MLX90614 Library
Wire
WiFi

4. Install the Edge Impulse Arduino library from:

model/ei-motor-vibration-monitoring-arduino-1_0_3.zip

5. Open:

firmware/MotorVibrationMonitoring.ino

6. Configure Wi-Fi and device credentials in Arduino IoT Cloud.
7. Upload the sketch from Arduino IoT Cloud. The Cloud upload flow provides SECRET_SSID, SECRET_OPTIONAL_PASS, and SECRET_DEVICE_KEY automatically; no local credential file is committed.
8. Open Arduino IoT Cloud dashboard.
9. Use motorCommand to start/stop the motor and resetFaultCommand to clear a latched fault.

Local serial commands are also available at 9600 baud:

Command	Action
1	Start motor
0	Stop motor
r / R	Reset latched fault

Evidence

The system was tested with the physical hardware setup. Dashboard screenshots, OLED states, hardware photos, and a live demonstration video are included in this repository.

Demo video: Project demonstration video

External streaming link: Watch the demo video on SharePoint

Results

The final system satisfies the project requirements:

Requirement	Status
ESP32 microcontroller	Implemented
At least 2 sensors	ADXL345 + MLX90614
OLED with meaningful live information	State, confidence, anomaly, temperature, motor status, RSSI, Wi-Fi, cloud, IP
Additional actuator/output not OLED or LED	5V relay controls motor power
Wi-Fi networking	Arduino IoT Cloud connection handler
Cloud publishing	Temperature, anomaly, confidence, state, fault reason, motor state
Cloud dashboard monitoring	Dashboard screenshot included
Cloud dashboard control	motorCommand and resetFaultCommand
Wiring table + diagram	Included
Evidence	Dashboard, hardware, OLED, Edge Impulse screenshots, demo video

Limitations

• All collected samples were used for training without a separate test split. The reported accuracy (99.9%) reflects training set performance. Despite this, the system performed correctly during live testing on the physical prototype.
• The three motor states (STOP, NORMAL, WARNING) produce clearly different vibration patterns, which makes classification easier. A real industrial motor with more subtle fault types would need more training data.
• The temperature fault threshold (32.0 C) is set low for demo purposes and should be adjusted for real applications.