A Deep Generative Digital Twin for Vehicle Steering System Health Monitoring with Limited Healthy Monitoring Samples*

The intelligent health management for vehicle steering system often faces problems like history data dependency and limited faulty samples. Digital twins (DT) have the potential to solve these problems. However, the advantages of DT might be lost due to the prior physics knowledge, data heterogeneity, and the limited healthy physical samples. In this paper, the Weighted Feature Space Measure Barycenter Averaging-Wasserstein Generative Digital Twin architecture (WFSMBA-WGDT) is proposed. The WFSMBA algorithm is designed through a designed low-dimension feature space measure to synthesize samples whose distributions are similar to healthy physical samples efficiently. The WGDT is constructed combining the Wasserstein generative adversarial network with gradient penalty (WGAN-GP) trained with the synthetic samples and the proposed consistency test. This test can guarantee the high fidelity of the virtual samples without large time consumption and prior knowledge. Afterwards, the WFSMBA-WGDT can be used to carry out multiple health management tasks. The effectiveness of the WFSMBA-WGDT is verified through case studies.