Stator Current Based Multi-Class Fault Diagnosis of Three Phase Induction Motor using Machine Learning Framework

Stator current based fault diagnosis of three phase induction motor is a simple and cost effective solution for industrial automation. However, with low volume of training samples, raw stator current based multi-class motor fault diagnosis is really challenging due to the predominant power frequency component. Keeping this in mind, the present investigation is a comprehensive exploration to demonstrate the impact of various data related hyper-parameters in multi-class fault diagnosis using machine learning framework. Powerful classifiers like Decision Trees (DT), Random Forest (RF) and 1D Convolutional Neural Network (1D CNN) have been considered for comparison purpose. Raw stator current sampled at 1280 Hz has been directly fed to the classifiers for fault classification. Experimental results clearly indicate that the optimum data length and volume of training samples have an important role in robust model building. Ensemble based RF has been found much superior to DT in motor fault classification. However, due to self-synthesized feature extraction through convolution process, deep learning based 1D CNN outperforms. Finally, stator current envelope based approach has been discovered with promising future for fault diagnosis of induction motor