Digital twin-assisted interpretable transfer learning : a novel wavelet-based framework for intelligent fault diagnostics from simulated domain to real industrial domain

Abstract

Rolling bearings are crucial components in a wide range of rotating machinery, playing a vital role in maintaining safe and reliable industrial production. Transfer learning techniques have shown significant promise for the real-time monitoring of bearings, boosting the safety of machinery and equipment operations. Nonetheless, the scarcity of adequately labeled fault data presents challenges for the training of transfer diagnosis models, leading to notable constraints in actual industrial applications. To address these challenges, this article proposes a digital twin-assisted diagnostic method, which incorporates a bearing dynamic model into an improved transfer learning network to improve diagnostic performance. The primary contributions of this research include 1) the development of a five-degree-of-freedom (5-DOF) dynamic model for rolling bearings to produce simulated signals that accurately represent the status of the bearings; and 2) the exploration of an effective integration of wavelet-based feature learning with discriminative learning mechanisms, culminating in a novel stacked discrete wavelet-based transfer learning network (SDWTN). SDWTN can sensitively locate and reinforce critical fault information and effectively construct discriminative status representations, thereby boosting diagnostic accuracy. Extensive experiments demonstrate that SDWTN surpasses other leading methods in diagnostic performance.