文献类型：期刊文献

英文题名：A Domain Generalization Method for Fault Diagnosis: Integrating Causal Learning and Distributionally Robust Optimization

作者：Qi, Zhikuan[1];Luo, Zhi[1,2];Zhao, Ming[3];Zhou, Shaoping[1,2]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Key Lab Pressure Syst & Safety, Minist Educ, Shanghai 200237, Peoples R China;[3]Xi An Jiao Tong Univ, Sch Mech Engn, Xian 710049, Peoples R China

年份：2025

卷号：74

外文期刊名：IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

收录：;EI(收录号：20251218090670);WOS:【SCI-EXPANDED(收录号:WOS:001457791200049)】；

基金：This work was supported by the Natural Science Foundation of China under Grant 52275127.

语种：英文

外文关键词：Training; Fault diagnosis; Data mining; Optimization; Feature extraction; Prognostics and health management; Uncertainty; Representation learning; Power engineering; Noise; Causal learning; distributionally robust optimization (DRO); domain generalization; intelligent fault diagnosis (IFD)

摘要：Domain adaptation has been widely used in variable condition fault diagnosis of mechanical equipment, due to its ability to effectively address the degradation of model generalization performance caused by differences in data distribution. However, the success of domain adaptation methods typically depends on sufficient access to target domain data, which significantly limits their practical application scenarios. To tackle this problem, this article proposes a novel domain generalization method called integrating causal learning and distributionally robust optimization (ICLDRO). In this method, a causal learning-based encoding-decoding system is designed to generate augmented data that maintains consistent semantic information and constructs uncertainty sets by the augmented data. Distributionally robust optimization (DRO) is then executed on the uncertainty set to enhance the robust domain generalization performance of the model on unknown target domains. The effectiveness of ICLDRO is validated through experiments on one public dataset and two private datasets. The results demonstrate that ICLDRO outperforms several state-of-the-art methods across most generalization tasks.