As a critical load-bearing component of railway vehicles, the performance of rolling bearings directly affects the operational safety and reliability of trains. To meet the requirements of health monitoring and fault diagnosis for rolling bearings, a digital twin-driven dynamic modeling and signal generation framework is proposed. This framework introduces a 4 degrees-of-freedom dynamic model of rolling bearings, combined with a joint parameter-noise optimization mechanism based on digital twin technology. In this mechanism, both model parameters and the signal-to-noise ratio of added noise are treated as optimization variables. A coherence function-based objective function guides the particle swarm optimization to update the parameters and SNR simultaneously. The purpose of the framework is to correct the dynamic model, reduce distribution differences, and generate twin signals under normal and faulty conditions to achieve realistic mapping of the twin signals. A convolutional neural network is then employed for multi-class fault identification. Experimental results demonstrate that the proposed framework effectively enhances signal fidelity and improves the accuracy and robustness of fault diagnosis.  This study provides a reliable solution for building a virtual-real fusion diagnostic model within bearing digital twin systems and offers important support for fault prediction and maintenance decision-making of railway vehicles.
 

digital twin,dynamic modeling,parameter update,coherence function,fault diagnosis