Shrinkage mamba relation network with out-of-distribution data augmentation for rotating machinery fault detection and localization under zero-faulty data

Abstract

Data-driven fault detection (FD) or diagnosis methods are key technologies to ensure safe operation of rotating machinery. These methods rely on a requisite volume of fault data. However, acquiring fault data from rotating machinery is typically problematic and can be entirely unattainable. The critical challenge is to accurately detect and localize the fault states of rotating machinery under the absence of any fault data. Therefore, a newly shrinkage Mamba relation network (SMRN) with out-of-distribution data (OODD) augmentation is proposed for FD and localization in rotating machinery with zero-faulty data. Firstly, the corresponding sensors are arranged for the key detection locations on the rotating machinery. The data generator is referenced to generate OODD for the health data at each detection locations, assisting in mining of intrinsic state information from health data. Then, feature pairs are built in health data and OODD to reveal inter-state attribute relationships. Finally, the location of faults in rotating machinery is determined by evaluating the similarity between feature pairs at each detection location. The SMRN method effectiveness is verified by using self-built propulsion shaft system experiments and rolling bearing cases. The experimental results show the SMRN method can effectively detect and localize fault state of rotating machinery in multiple fault modes, compound fault scenarios, and variable operating conditions.