A novel dual networks-guided self-assessment framework for bearings fault mode diagnosis considering early fault feature diversity

In the early stage of fault occurs, due to the fault-related information has not yet fully manifested, different fault modes may exhibit similar degradation features. However, traditional fault diagnosis methods fails in effectively predicting future fault mode according to the early stage of fault characteristics. For performing predictive maintenance, it is urgent to have the ability of fault mode diagnosis based on the early fault-stage data. In this work, we propose a novel dual networks-guided self-assessment framework for bearings fault mode diagnosis considering early fault feature diversity. A regression network-based failure time point (FTP) identification method is developed to obtain the starting time of early fault by calculating the distribution discrepancy between healthy data and degradation data. Afterwards, multimodal information is embedded into the dual networks-guided fault diagnosis frameworks to realize the fault feature in-depth extraction from the early fault-stage data. Based on these, a novel fault mode self-assessment strategy is designed to adaptively evaluate the diagnosis performance of each network under various fault characteristics, achieving the most reliable fault mode diagnosis result by in-time updating network credibility indicator and dynamically adjust weights between dual diagnose networks. The competitive of the proposed method was validated by compared with other state-of-the-art methods through XJTU-SY dataset. Comparison results demonstrate that the proposed method can accurately deduce and predict the fault mode through only early fault data.