Broken magnets fault detection in pmsm using a convolutional neural network and svm

Abstract

The Permanent Magnet Synchronous Motor (PMSM) stands as a pivotal component in
various applications, yet it remains susceptible to an array of faults within both its rotor and
stator, there arises an imperative to swiftly and intelligently address these issues. In this
study, a novel approach was undertaken wherein a PMSM design was conceptualized within
the Ansys Maxwell program, followed by the deliberate introduction of a fault at the rotor's
magnetic level. Specifically, three distinct fault scenarios were delineated based on the
number of broken magnets (BM), namely 2, 3, and 4, localized within specific rotor areas.
Notably, the magnetic flux density was selected as the focal parameter for this investigation.
To effectively detect and diagnose faults stemming from BM, an innovative Convolutional
Neural Network (CNN) architecture was devised. Leveraging images of the PMSM design
captured during operational phases at various time intervals, the CNN exhibited remarkable
efficacy in discerning and categorizing fault instances. Upon analysis of the derived
outcomes, it becomes evident that the CNN exhibited unparalleled accuracy in fault
detection, achieving a remarkable 100% success rate when juxtaposed with alternative
methodologies such as Support Vector Machines (SVM) and K-Nearest Neighbors (KNN),
which yielded accuracy rates of 97%.