Fault Diagnosis of Broken Rotor Magnets in PMSMS Using FFT Features and Machine Learning: MLP and SVM Models

Permanent magnet synchronous motors (PMSMs) have become commonly employed in various critical applications such as industrial automation, electric vehicles, aerospace, robotics, and HVAC/R systems. In this study, the detection of rotor magnet breakage faults in PMSMs was investigated using two artificial intelligence (AI) techniques: Multilayer Perceptron (MLP) and Support Vector Machine (SVM). Fault conditions were experimentally induced by introducing controlled breaks in the rotor magnets of PMSM samples. Stator current signals were collected using a current probe and oscilloscope, then preprocessed to remove noise components. Fast Fourier Transform (FFT) was applied to convert the time-domain signals into the frequency domain, allowing extraction of characteristic fault-related features. These frequency spectrum features served as inputs to train and test the MLP and SVM classifiers. Both AI models achieved high classification accuracy in distinguishing healthy and faulty motor states, with overall accuracies exceeding 95%. Comparative analysis showed that while both models performed effectively, the SVM demonstrated slightly superior precision in fault detection. The proposed approach confirms that frequency-domain analysis combined with AI classification provides a reliable, non-invasive method for timely detection of rotor magnet faults in PMSMs, which is crucial for improving system reliability and minimizing unexpected downtime.