Simulation Model-Based Adaptive Graph Information Aggregation Method for Bearing Fault Diagnosis

To address the issue of insufficient model generalization caused by the scarcity of real data in the field of fault diagnosis, this paper proposes an adaptive graph-based information aggregation method for bearing fault diagnosis based on a simulation model. First, a non-stationary simulated bearing fault model is constructed according to the fault mechanism, which can generate various types of simulated fault data under specified bearing structure parameters and operating conditions. Then, a graph structure is designed using a local-global similarity threshold to effectively integrate knowledge from both simulated and real fault data. An adaptive graph information aggregation module is developed to dynamically adjust edge weights, enhancing noise filtering and feature extraction. Finally, graph-based visualization demonstrates high feature similarity between simulated and real data, and the proposed method maintains competitive classification performance even under extremely limited real sample conditions, validating the effectiveness of the approach and the feasibility of using simulated data to compensate for the lack of real data.