Fault semantic knowledge transfer learning: Cross-domain compound fault diagnosis method under limited single fault samples

The coupling of faults leads to an exponential growth of compound fault types, making it impractical to collect complete labeled compound fault data in real-world scenarios. While cross-domain compound fault diagnosis (the target-domain does not have labeled compound fault data) is crucial for system reliability, existing methods often rely on abundant single-fault samples and rarely validate the reliability when single-fault data is limited. To overcome this limitation, we propose a novel fault semantic knowledge transfer learning framework. Specifically, FSKTL incorporates inter-class semantic distance loss in the source-domain, enabling fault classification through low-dimensional fault semantics and identifying the optimal fault semantic correlation function. Subsequently, FSKTL introduces inter-domain semantic alignment loss in the target-domain. This approach not only preserves the semantic space optimized by the source-domain for fault classification, but also achieves domain adaptation, enhancing the cross-domain generalization of the optimal fault semantic correlation function. Finally, extensive experiments are conducted on two publicly available datasets to validate the effectiveness of the proposed method. The results demonstrate that compared to other methods, this approach achieves the highest accuracy in cross-domain compound and single fault diagnosis.