Cross Varying-Speed Fault Diagnosis Method Based on Multi-Sensor Information Fusion and Domain Adaptation

Current researches in deep transfer learning for fault diagnosis predominantly assume the constant-speed operating condition. However, in real-world industrial scenarios, dynamical production environments often result in time-varying speed conditions. Therefore, this study proposes a novel cross-domain diagnostic approach integrating energy weighted signal fusion (EWSF) with a domain adaptation feature enhancement network (DAFEN). Firstly, the EWSF technique amalgamates data from diverse sensors, amplifying the key information within signals. Subsequently, a global-local feature enhancement module (GLFEM) is developed to capture the invariant features across various domains. Finally, an integrated DAFEN with GLFEM is designed to achieve cross-domain diagnosis under time-varying speed conditions. The proposed method was implemented and verified on planetary gearbox dataset. Experimental results show that the proposed EWSF-DAFEN method outperformed comparative methods, achieving an impressive average accuracy of 96.24% across seven transfer diagnosis tasks, thereby proving its robust domain adaptation and cross-domain fault diagnostic capability.