Mixed-Fault Diagnosis for Permanent Magnet Motor With Few-Shot Learning Based on the Prototypical Network

This paper proposes an AI-driven few-shot learning approach for fault diagnosis in permanent magnet synchronous motors (PMSMs), utilising a prototypical network to accurately differentiate among healthy conditions, single-fault states (ITSCF or LDMF) and mixed-fault scenarios (i.e., when the two types of faults occur simultaneously) with limited training data. Addressing these concurrent faults is particularly significant due to the potential interactions between their underlying mechanisms (e.g., high current spikes from an ITSCF causing possible magnet demagnetisation) and the increased complexity of their combined diagnostic signatures, posing significant challenges to accurate diagnosis. The study first simulates and analyses motor stator current characteristics, identifying them as key diagnostic signals for both fault types. Experimental validation measures stator current from both healthy and faulty motors. Through training with a minimal amount of data, the proposed model using a prototypical network achieves over 98% accuracy in diagnosing mixed faults (i.e., ITSCF, LDMF or a combination of both), significantly outperforming convolutional neural network (CNN)-based methods (80%). Furthermore, demonstrating a key advancement for few-shot learning in this domain, when trained on only a few labelled fault patterns, the model correctly classifies unseen faults with 81% accuracy, compared to CNN's 70%, demonstrating strong generalisation and scalability for real-world applications.