Inter-Turn Short Circuit Diagnosis of Permanent Magnet Synchronous Motor Based on Siamese Convolutional Neural Network Under Small Fault Samples

Inter-turn short circuit (ITSC) fault is the most common and harmful fault of permanent magnet synchronous motor (PMSM). Preventing from subsequent severe damage to the motor and related systems, the detection of ITSC becomes an important issue. In order to detect the ITSC problem efficiently and accurately, particularly when fault data are scarce, this article proposes an effective method, which can eliminate redundant information and extract valuable features of the ITSC current signal of PMSM using a data processing method. Subsequently, Pearson coefficient optimized Wasserstein generative adversarial network (PWGAN) is designed to generate fault data and filter out low-quality data to construct a hierarchical training strategy, which allows Siamese convolutional neural network (SCNN) have better detection of ITSC problem in the case of a small size of fault samples and reduces substantial computational burden. Due to large number of healthy data being put into model in actual use, a soft threshold method is adopted to preclassify most of healthy data to improve the stability of diagnosis system. Experimental test is performed to substantiate that the hierarchically trained model, when furnished with a substantial number of fault samples, is capable of achieving an accuracy rate of 98.75%. In comparison with other machine learning methodologies, SCNN demonstrates superior accuracy, particularly in scenarios where ITSC fault data are limited. The application of the soft threshold method allows for the preliminary diagnosis of up to 94.7% of healthy data inputs, thereby enabling the overall diagnostic system to attain an accuracy rate of 98.97%.