Model-Based Data Normalization for Data-Driven PMSM Fault Diagnosis

The diagnosis of numerous defects in permanent magnet synchronous motors is crucial to the motor system's safe operation. However, fluctuating working conditions can distort current waveforms, making defect diagnosis more difficult. In addition, the closed-loop controller influences the faulty components of current signals. This article proposes a model-based data normalization method to convert raw current data into a standard format in order to address the issue. In this article, the mathematical model is presented under both healthy and different faulty conditions. Luenberger observer is constructed to calculate the current estimation error between the actual motor and the constructed model. The error under interturn short-circuit fault, uniform demagnetization fault, and partial demagnetization fault is theoretically demonstrated as a proof of the effectiveness of the approach. Then, the error is further normalized to eliminate the influence of speed on both signal amplitude and fundamental frequency. The method proposed eliminates the influence of working conditions and controller bandwidth on current signals. After the whole process, the data can be processed to be almost only related to the fault type and fault severity. The effectiveness of the proposed method is verified by theoretical derivation, simulation, and experimentation.