Implementation of ANN for PMSM Interturn Short-Circuit Detection in the Embedded System

The problem of condition monitoring and fault detection in powertrain systems becomes more critical with the increasing use of fail-operational systems. These systems are essential in the automotive industry, robotics, and other industrial applications. One of the critical features of such a system is recognizing the fault and suppressing its influence. The paper describes a feed-forward artificial neural network-based diagnostic of interturn short-circuit faults in a dual three-phase permanent magnet synchronous motor. The paper focuses on using multi-layer perceptron network (MLPN) and convolutional neural network (CNN) for interturn short-circuit detection and, more importantly, their real implementation into the automotive AURIX TC397 microcontroller. The paper presents the achieved neural network inference times as well as data preprocessing computation time. The behavior of the artificial neural networks (ANNs) is tested on an experimental configurable multi-phase permanent magnet synchronous motor (PMSM) with the possibility to emulate interturn short-circuit fault using prepared winding taps. The paper includes the essential aspects that should be respected during ANN design and implementation into the microcontroller.