Diagnosis of Stator Winding Fault of Single-Phase Induction Motor Employing Wavelet Induced Residual-Convolutional Neural Network

Single-phase induction motors (SPIM) are quite effectively used in household machines and equally efficiently for industrial purposes. Mostly, current signature-based techniques are often used for fault diagnosis, which is not always a cost-effective solution. This paper presents a stator winding fault detection scheme of a single-phase induction motor employing a wavelet induced residual-convolutional neural network (W-RCNN) based on a deep learning framework. The structure of W-RCNN possesses a residual block with a skip connection. The importance of using the skip connection is not only to transfer the information of the input data towards the output layer but also to produce an alternative path for gradient transformation as well as feature fusion. In this work, experimental data collected from a healthy 1kW SPIM and another similar rated machine with three-types of winding fault are used as inputs to the W-RCNN architecture for the diagnosis of automated machine faults. The proposed detection scheme is protracted to a statistical study which shows marked improved performance metrics compared to the state-of-the-art methods.