Fault Detection and Diagnosis in Induction Motors Using Machine Learning-Based Vibration Analysis

Abstract

Three-phase induction motors are essential to contemporary industrial automation, powering vital equipment in sectors like manufacturing, power generation, and processing facilities. Unexpected downtime due to motor malfunctions can cause significant financial setbacks and interrupt operations. Conventional methods of condition monitoring, including regular visual checks and simple threshold-based vibration monitoring, often fall short in identifying early-stage multi-fault scenarios or adjusting to changing load conditions.

This study introduces an all-encompassing Machine Learning (ML) framework designed for the real-time Fault Detection and Diagnosis (FDD) of induction motors through the analysis of multi-axis vibrations. High-frequency vibration signals are recorded by tri-axial MEMS accelerometers across different operational loads, specifically at 0%, 25%, 50%, 75%, and 100% of the rated load. The preprocessing of these signals involves the use of Fast Fourier Transform (FFT) and Continuous Wavelet Transform (CWT) to derive features from multiple domains, including statistical time-domain parameters, spectral harmonics, and time-frequency energy distributions.

A diagnostic pipeline that integrates Random Forest (RF), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN) has been created and evaluated using experimental baseline datasets for comparison.The proposed system achieves an overall classification accuracy of 98.7% across healthy states and four primary fault modes:

1. Inner and outer race bearing defects
2. Broken rotor bars
3. Stator winding inter-turn short circuits
4. Shaft misalignment

Research findings indicate that integrating time-frequency CWT feature maps with a 2D-CNN architecture achieves exceptional results in isolating faults at early stages, sustaining an accuracy rate of 97.4% despite the presence of noise and varying load conditions in industrial settings. This framework offers a scalable and low-latency option for predictive maintenance systems in edge computing.