Cross-Domain Compound Fault Diagnosis of Machine-Level Motors via Time–Frequency Self-Contrastive Learning
Yiming He, Chao Zhao, Weiming Shen
Abstract
Intelligent fault diagnosis of industrial motors under different operating conditions is a valuable and challenging topic. Compound faults are inevitable to occur under different operating conditions. Due to the high cost of dismantling and testing, it is impractical to collect and label all compound fault types under different operating conditions. This article proposes a cross-domain compound fault diagnosis framework of machine-level motors without target domain information. A novel time-frequency self-contrastive learning (TFSCL) strategy is proposed to enhance domain-irrelevant feature extraction. TFSCL generates homogeneous and heterogeneous information of the input itself from time domain and frequency domain to construct self-contrastive pairs. The multiscale spatial convolution structure and the cross time–frequency information interaction strategy are designed to further provide fusion and interaction. Ablation experiments are performed on real industrial motor signals. TFSCL achieved F1-scores of over 90% on six cross-domain tasks, which is superior to compared advanced lightweight models.