176 / 2025-06-11 22:01:49

A deep generative digital twin for the health monitoring of rotating machines

Data augmentation,deep digital twin,health monitoring,rotating machines

The deep digital twin of rotating machines for health monitoring has the advantages of not relying on historical monitoring data and prior knowledge of fault, but it still requires a large number of health monitoring samples. To address this problem, this paper first designs a feature space measure based on the statistical features of monitoring data closely related to the potential fault characteristics in the time and frequency domains, and proposes a weighted feature space measure barycenter averaging method (WFSMBA) to efficiently synthesize any number of samples based on a small number of health monitoring samples. The generative adversarial network (GAN) in the deep digital twin model is trained based on these synthesized samples. This paper further introduces the Wasserstein loss function and gradient penalty mechanism to alleviate the training instability of the GAN. A comparative study of the proposed method is carried out and the results show that the proposed method can effectively detect the early faults of rotating machines with only a small number of healthy samples.