Custom Phase Space Reconstruction Image-Driven Fault Diagnosis for PMSM Under Few-Labeled Samples

The existing intelligent fault diagnosis methods for motors mainly focus on the supervised learning of 1-D signals, which ignores the global features of the signal and requires large training samples of labeled data. For this reason, this article proposes a custom double-sided phase space reconstruction (CDPSR) image-driven fault diagnosis method for permanent magnet synchronous motor (PMSM) under few-labeled samples. To reveal 2-D features in the signal, a CDPSR method is proposed to convert the fault signal into data images. The two-parallel diagnostic framework is designed, which includes 1) two semisupervised stacked autoencoders for image feature extraction and primary classification and 2) the weighted decision fusion layer to output the final diagnosis result. The extensive diagnostic results on the experimental platform verify the effectiveness of the proposed method for PMSM fault diagnosis.