Real-Time Fault Diagnosis of Permanent Magnet Synchronous Motor and Drive System

Abstract

Permanent Magnet Synchronous Motors (PMSMs) have gained massive popularity in industrial applications such as electric vehicles, robotic systems, and offshore industries due to their merits of efficiency, power density, and controllability. PMSMs working in such applications are constantly exposed to electrical, thermal, and mechanical stresses, resulting in different faults such as electrical, mechanical, and magnetic faults. These faults may lead to efficiency reduction, excessive heat, and even catastrophic system breakdown if not diagnosed in time. Therefore, developing methods for real-time condition monitoring and detection of faults at early stages can substantially lower maintenance costs, downtime of the system, and productivity loss. In this dissertation, condition monitoring and detection of the three most common faults in PMSMs and drive systems, namely inter-turn short circuit, demagnetization, and sensor faults are studied. First, modeling and detection of inter-turn short circuit fault is investigated by proposing one FEM-based model, and one analytical model. In these two models, efforts are made to extract either fault indicators or adjustments for being used in combination with more complex detection methods. Subsequently, a systematic fault diagnosis of PMSM and drive system containing multiple faults based on structural analysis is presented. After implementing structural analysis and obtaining the redundant part of the PMSM and drive system, several sequential residuals are designed and implemented based on the fault terms that appear in each of the redundant sets to detect and isolate the studied faults which are applied at different time intervals. Finally, real-time detection of faults in PMSMs and drive systems by using a powerful statistical signal-processing detector such as generalized likelihood ratio test is investigated. By using generalized likelihood ratio test, a threshold was obtained based on choosing the probability of a false alarm and the probability of detection for each detector based on which decision was made to indicate the presence of the studied faults. To improve the detection and recovery delay time, a recursive cumulative GLRT with an adaptive threshold algorithm is implemented. As a result, a more processed fault indicator is achieved by this recursive algorithm that is compared to an arbitrary threshold, and a decision is made in real-time performance. The experimental results show that the statistical detector is able to efficiently detect all the unexpected faults in the presence of unknown noise and without experiencing any false alarm, proving the effectiveness of this diagnostic approach.