A novel multi-sensor information fusion method for fault diagnosis of rotating machinery with missing signals

Multi-sensor systems are a cornerstone of rotating machinery diagnostics, with information fusion significantly enhancing diagnostic accuracy and robustness. However, many existing methods fail to balance the consistency and specificity of sensor signals, resulting in incomplete fault information extraction. Moreover, these methods often assume the availability of all sensor signals, rendering them ineffective when signals are missing. To address these limitations, this paper proposes a novel multi-sensor information fusion fault diagnosis (MSIFFD) method designed specifically for scenarios with missing signals. The method employs a feature encoder that integrates private and shared features to extract comprehensive fault information. A learnable prompt-based module reconstructs missing features by integrating existing signals with prior knowledge. Additionally, intra-source and inter-source fusion modules are utilized to further enhance feature integration. Experimental results demonstrate the superior performance of the proposed method. Even in tasks where three sensor channels are missing, the method achieves an impressive accuracy of 97.65%, providing a robust solution for fault diagnosis in complex industrial environments.