Sensor Placement and Fault Detection in Electric Motor using Stacked Classifier and Search Algorithm

Efficient health monitoring for high power energy systems has become an imperative research area in the field of reliability engineering. Novel systems, such as permanent magnet synchronous motors (PMSM), have become prominent in many impactful applications. These include but are not limited to propulsion aircraft, electric vehicles, ultra-high-speed elevators, and industrial manufacturing. Therefore, determining an optimal fault detection framework is a significant task. However, due to the newness of this system, there is little to no experimental data to analyze, so finite element simulation data is a necessity for determining the monitoring system. In this study, a design optimization approach is implemented for sensor placement and fault detection on a PMSM with hall effect sensors. This system is prone to short-winding faults, which can lead to catastrophic failures. The proposed method simultaneously determines the optimal placement of sensors while training an optimal classifier. The sensor placement is identified with a genetic algorithm, which uses the classifier's accuracy as the fitness function. In this case, the classifier structure is 'stacked,' which means it combines multiple classification models and makes a final output with a meta-learner. This advanced classifier enables not only fault detection, but the severity of said fault, which is a significant improvement over present methodologies. Overall, this proposed structure converges to a design that has high accuracy for detection of faults, as well as the severity level.