Early detection of PMSM faults at static and dynamic operating points by using artificial neural network for automated electric vehicle

The realization of autonomous driving requires maximum safety and reliability of powertrain system and components (e.g., PMSM). While today's vehicles have a fallback level provided by the driver, this is completely or at least temporarily eliminated in autonomous driving systems. It is thus of high importance that the safety–critical faults of the powertrain is detected early, safely and reliably. An electrical fault in the PMSM in the stator while driving can lead to a failure of the electric vehicle. This can create a hazard for passengers and other road users. Therefore, highly reliable, safe and early detection of the PMSM faults is an important requirement to meet these safety requirements. In the literature, PMSM faults have been considered only for steady-state operating points and not for dynamic operating points. In this study, a new early diagnosis method is developed using artificial neural networks. This method is applied to three PMSM faults at both fixed and dynamic operating points (speed and torque). The fault diagnosis is successfully achieved (accuracy of about 99.7%) at both fixed operating points and dynamic operating points of an electric vehicle. In this way, the automated driving vehicle is brought to a safe state without using a redundant powertrain. In this way, weight, installation space and costs of automated driving vehicles are reduced, and safety requirements are met.