Research on digital twin-assisted bearing fault diagnosis method based on virtual-real mapping

Abstract

The deployment of intelligent fault diagnosis for bearings within the industrial field is significantly challenged by the issue of limited sample sizes. Digital Twin (DT) technology facilitates the replication of rotating machinery operations within a virtual environment, thereby enabling the acquisition of equivalent or superior information regarding physical entities at a reduced cost, introducing a novel method for fault diagnosis in scenarios characterized by limited sample sizes. Nevertheless, the disparity in data distribution across virtual and physical realms poses challenges to deploying digital twin-based fault diagnosis methods. In response to this challenge, this paper proposes a digital twin-assisted bearing fault diagnosis method based on virtual-real mapping. Firstly, a bearing dynamics model is constructed in the virtual space using finite element methods to reflect the bearing's vibration response in physical space. Secondly, an efficient multi-scale attention cycle-consistent generative adversarial network (PACycleGAN) with a perceptual loss function is proposed as a bridge between virtual and physical spaces, reducing the data distribution differences through data mapping. Finally, a multi-index evaluation framework was established to validate the effectiveness of the simulation data after mapping, and through two case studies, the proposed method's ability to effectively address the small sample issue was confirmed.