Robust Active Learning Multiple Fault Diagnosis of PMSM Drives with Sensorless Control under Dynamic Operations and Imbalanced Datasets

Abstract

This paper proposes an active learning scheme to detect multiple faults in permanent magnet synchronous motors in dynamic operations without using historical labelled faulty training data. The proposed method combines the self-supervised anomaly detector based on a local outlier factor (LOF) and a deep Q-network (DQN) supervised reinforcement learner to classify interturn short-circuit, local demagnetisation and mixed faults. The first fault, which is detected by LOF and verified by an expert during maintenance, is used as training data for the DQN classifier. From that point onward, the LOF anomaly detector and DQN fault classifiers are working in tandem in the identification of new faults, which require expert intervention when either of them identifies a fault. The robustness of the scheme against dynamic operations, mixed fault and imbalanced training datasets is validated via a comparative study using stray flux data from an inhouse test setup.