Smart Maintenance System for Industrial Machines Using Machine Learning
R Kane Johan
Electronics and Communication Engineering Department,
Sathyabama Institute of Science and Technology, Chennai, India
kanejohan52@gmail.com
S. Karthikeyan
Electronics and Communication Engineering Department, Sathyabama
Institute of Science and Technology
Skarthu1879@gmail.com
Dr. E. Annadevi
Associate Professor
Electronics and Communication Engineering Department,
Sathyabama Institute of Science and Technology
Annadevi.ece@sathyabama.ac.in
Dr. Joany. R. M
Electronics and Communication Engineering Department,
Sathyabama Institute of Science and Technology
Joany.ece@sathyabama.ac.in
Abstract—This paper introduces a low-cost, real-time smart maintenance system for industrial rotary machines, powered by machine learning. It keeps a close watch on machine health using a simple multi-sensor setup—an ADXL335 tri-axial accelerometer for vibration checks, an LM35 temperature sensor for heat monitoring, and a voltage sensing module to track electrical performance. The system displays real-time data on LCD screen and detects abnormal conditions. To protect the machine, it turns on an exhaust fan automatically when the temperature reaches 32°C, then switches it off at 30°C. This includes a 2°C hysteresis to avoid rapid on-off switching It also sets off loud alarms and shuts down the motor during major faults. On the smart side, machine learning pulls out 18 useful features from statistical and frequency-domain analysis, hitting 96.8% accuracy in classifying six different fault types with a Random Forest model. This diagnostic system works well and costs little. In tests, it reaches upto 94.3% of real equipment problems but gives only 3.2% false alarms. It has proven reliable at detecting common industrial issues, including significant voltage drops, temperatures exceeding 50°C, and sharp vibration increases of 40-60%. With a total cost of only $150 to $200, it provides a practical and valuable tool for small and medium-sized businesses. Overall, the system offers a fresh, all-in-one approach by combining low-cost sensors, instant machine learning analysis, and built-in safeguards. This integrated package makes predictive maintenance a realistic and accessible strategy for factories, helping them prevent breakdowns before they occur.
Keywords: Predictive maintenance, machine learning, vibration analysis, Arduino, ADXL335, temperature monitoring, fault detection, industrial IoT, Random Forest classifier.
