Beta-Variational Autoencoder for Intern-Turn Short Circuit Fault Diagnosis in Dual Three-Phase Synchronous Motors

Dual Three-Phase Permanent-Magnet Synchronous Machines (DTPPMSMs) are an appealing alternative for a variety of applications that require high power levels and reliability. Regardless of the machine architecture, the winding insulation of an electrical machine is frequently subjected to thermal and mechanical stress as a result of high-performance requirements. This may cause Inter-Turn Short Circuit (ITSC) failures, compromising the machine's health and reducing the overall reliability of the drive system. Diagnosing the presence of an ITSC may be challenging, especially when only a few turns are shorted and the machine operates at low speed. In this circumstance, the effects of a short circuit are externally undetectable until the speed reaches a critical value. This research introduces a method utilizing a Beta-Variational Autoencoder ($\beta$-VAE) neural network to assess the presence of an ITSC by using appropriate features for training the Machine Learning (ML) model. Moreover, a statistical approach is introduced to diagnose the fault for different confidence intervals. The proposed method is assessed under various operating conditions using a prototype of DTPPMSMs.