A LSTM-based Neural Strategy for Diagnosis of Stator Inter-turn Faults with Low Severity Level for Induction Motors

This paper outlines a neural-based strategy to diagnose stator-inter-turn faults (STIFs) at low severities. The proposed strategy involves the study of a state-space model for healthy and faulty SITF configuration of squirrel cage induction machine (IM). Following the acquisition of healthy and faulty 3-phase currents from the state-space models, exploratory analysis of data are conducted using principal component analysis (PCA) and independent component analysis (ICA). Thereafter, various neural and non-neural-based classifiers are trained respecting appropriate data divisions. After considering factors like the least number of trainable parameters, confidence level of the outputs and highest classification metrics, the best classification model belonged to the family of LSTM neural networks.