Digital twin-assisted imbalanced fault diagnosis framework using subdomain adaptive mechanism and margin-aware regularization

Yan, Shen and Zhong, Xiang and Shao, Haidong and Ming, Yuhang and Liu, Chao and Liu, Bin (2023) Digital twin-assisted imbalanced fault diagnosis framework using subdomain adaptive mechanism and margin-aware regularization. Reliability Engineering and System Safety, 239. 109522. ISSN 0951-8320 (https://doi.org/10.1016/j.ress.2023.109522)

[thumbnail of Yan-etal-RESS-2023-Digital-twin-assisted-imbalanced-fault-diagnosis-framework-using-subdomain-adaptive-mechanism-and-margin-aware-regularization]
Preview
 Text. Filename: Yan-etal-RESS-2023-Digital-twin-assisted-imbalanced-fault-diagnosis-framework-using-subdomain-adaptive-mechanism-and-margin-aware-regularization.pdf
Accepted Author Manuscript
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo
Download (1MB)| Preview
[thumbnail of AAM] Text. Filename: JRESS_D_23_00961_R1.pdf
Accepted Author Manuscript
Restricted to Repository staff only until 25 July 2025.
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo
Download (1MB) | Request a copy

Abstract

The current data-level and algorithm-level based imbalanced fault diagnosis methods have respective limitations such as uneven data generation quality and excessive reliance on minority class information. In response to these limitations, this study proposes a novel digital twin-assisted framework for imbalanced fault diagnosis. The framework begins by analyzing the nonlinear kinetic characteristics of the gearbox and establishing a dynamic simulation model assisted by digital twin technology to generate high-fidelity simulated fault data. Subsequently, a subdomain adaptive mechanism is employed to align the conditional distribution of the subdomains by minimizing the dissimilarity of fine-grained features between the simulated and real-world fault data. To improve the fault tolerance of the model's diagnosis, margin-aware regularization is designed by applying significant regularization penalties to the fault data margins. Experimental results from two gearboxes demonstrate that, compared to the recent data-level and algorithm-level based imbalanced fault diagnosis methods, the proposed framework holds distinct advantages under the influence of highly imbalanced data, offering a fresh perspective for addressing this challenging scenario. In addition, the effectiveness of subdomain adaptive mechanism and margin-aware regularization is verified through the ablation experiment.

ORCID iDs

Yan, Shen, Zhong, Xiang, Shao, Haidong, Ming, Yuhang, Liu, Chao and Liu, Bin ORCID logoORCID: https://orcid.org/0000-0002-3946-8124;
CORE (COnnecting REpositories)