Diagnosis of Induction Motor Faults Using Frequency Occurrence Image Plots—A Deep Learning Approach

Abstract

Accurate diagnosis of induction motor faults is important for reliable and safe operation of industrial processes. Majority of the faults which occur in induction motors are mainly diagnosed using motor current signature analysis. However, the accuracy of fault detection depends on selection of suitable features from motor current, the failure of which may result in incorrect interpretation. Considering the aforesaid fact, this paper presents an image processing aided deep learning framework for reliable diagnosis of induction motor faults, which eliminates the need of separate feature extraction stage. To this end, the motor current signals under different types of fault conditions were procured and were subsequently processed into frequency occurrence plots. The frequency occurrence image plots for different fault scenarios were finally used as inputs to a deep convolution neural network for the purpose of classification. Transfer learning technique was adopted to reduce the computation time of Convolution Neural Network and classification of motor faults was done at five different loading conditions. Four types of classification tasks have been addressed here and comprehensive analysis was done using a variety of CNN architectures. It has been observed that the proposed method returns a highest mean classification accuracy of 96.67% in segregating different types of faults which can be implemented in real-life for condition monitoring of induction motors.