Amalgamation of Transfer Learning and Deep Convolutional Neural Network for Multiple Fault Detection in SCIM

Abstract

The modern industries are driven by the Squirrel cage induction motors (SCIMs), and zero downtime is the need of the hour. Condition-based maintenance is pivotal for achieving zero downtime. The ability of automatic feature extraction of Deep learning has effectively used in fault diagnosis in SCIMs. This paper proposes a novel transfer learning (TL) based deep convolutional neural network (CNN) fault detection model for bearing fault and broken rotor bar detection in SCIM, both individually and jointly. The transfer learning enables the faster learning and accelerates the training of deep CNN based fault detection model. Compared with the deep CNN model trained from scratch, the developed method is meticulous and computationally efficient. This paper has used a current analysis for fault detection in SCIMs. The proposed method owing to its deep structures and inherent ability, automatically learns the features from current signals for fault detection. The proposed fault detection model has achieved a mean accuracy of 99.40%. Also, the proposed method overcomes the disadvantages of deep CNN by applying for the knowledge transfer through transfer learning.