Sensitive Inter-Tum Fault Identifcation in Induction Motors Using Deep Learning Based Methods

Abstract

Inter Turn Short Circuit Fault (ITSCF) is one of the frequented electrical machine failures, which accounts for more than 30% of faults. Thus, the detection of ITSCF at an early stage is an important subject to investigate, since induction machines form the workhorses for the industries. This work has leveraged the advantages of deep learning-based approaches for diagnosis of ITSCF at early stage in induction motor using ID Convolutional Neural Network (CNN) and one of Recurrent Neural Network models which is Long Short-Term Memory (LSTM). Initially, the mathematical analysis of induction motor with ITSCF in arbitrary reference frame has been proposed. Then, the novel deep learning methods, namely ID CNN and LSTM have been investigated to diagnose ITSCF at very low level and isolate the fault, from ambiguous conditions like voltage imbalances and load variations using motor currents. The performance of the proposed methods has been compared, and the results show that CNN is more superior in accuracy to LSTM which provides good classification performance for fault detection at early stages of fault development, so that CNN is more effective for fault diagnosis at an incipient stage, than LSTM.