A New Collaborative Learning Method for Fault Diagnosis of Traction Motors Considering Information Missing

As the core component of complex equipment such as tanks, the health of traction motors directly affects the safety and reliability of the equipment. In practical applications, motors often face the problem of missing information caused by multi-rate sampling of sensors. However, existing methods have the following limitations: 1) They directly ignore the missing data and labels. 2) Supervisory signals that rely on tagged data in the absence of data and cannot effectively utilize untagged data. To overcome these limitations, this paper presents a collaborative learning strategy (CESSL), implemented in two stages. The first stage uses an unsupervised data reconstruction strategy, merging a parallel self-learning feature extraction network with a Generative Adversarial Network that includes identity mapping residual blocks, to efficiently extract features from incomplete data. The second stage employs a semi-supervised learning model based on this feature extraction network for label completion and classification, enhanced by a multi-layer Local Maximum Mean Discrepancy (LMMD) module to evaluate domain differences under various operating conditions and integrates predictive and consistency regularization losses into the loss function for improved model performance. The effectiveness of the proposed method is verified through fault simulation experiments on a three-phase motor. The CESSL method outperforms other state-of-the-art diagnostic methods, maintaining high accuracy even when both data and labels are missing.