Training Scheme for Convolutional Neural Network Based Multiple Fault Classifier of Permanent Magnet Synchronous Motors Under Variable Speed and Load Conditions

This paper introduces a novel methodology for training Convolutional Neural Network models to detect multiple faults in permanent magnet synchronous motors (PMSMs). The proposed method aims to mitigate memory overloading and overfitting for a neural network. Within the framework, measured current and flux signals from a PMSM under variable speed condition in time domain are converted into spectrogram images using short-time Fourier transformation. The suggested algorithm can classify inter-turn short-circuit faults, local de-magnetization faults, and mixed-fault in PMSMs. The proposed scheme is validated using data from an in-house experimental setup.