Journal:Expert Systems with Applications
Key Words:Fault diagnosis; Imbalanced domain adaptation; Hypergraph neural network; Domain adversarial learning; Rotating machinery
Abstract:Reliable fault diagnosis for rotating machinery becomes challenging when operating conditions change across domains and fault data exhibit long-tailed class distributions. Under such settings, conventional deep learning models often suffer from degraded transferability and biased recognition of rare but critical faults. This paper proposes IDAHGCN, an Imbalanced DomainAdversarial Hypergraph Convolutional Network that jointly addresses domain shift and class imbalance. IDAHGCN constructs a hypergraph to encode group-wise relations among samples, which enables structural representation learning beyond pairwise connections. On this basis, a domain-adversarial hypergraph convolutional network is employed to learn features that remain discriminative while reducing cross-domain discrepancy. To further improve minority-fault recognition, IDAHGCN integrates class-balanced reweighting, class-wise alignment, and marginbased regularization. Experiments on two widely used rotating machinery datasets under multiple imbalanced domain adaptation protocols demonstrate consistent improvements over representative unsupervised and imbalanced domain adaptation baselines, measured by accuracy as well as imbalance-sensitive metrics such as Macro-F1 and G-mean. Additional analyses on noise robustness and computational efficiency further support the practicality of the proposed framework for safetycritical industrial diagnosis.
Co-author:Yuansheng Luo,Miao Chen,Yaguang Guo
First Author:Zhangjun Wu
Indexed by:Journal paper
Correspondence Author:Haidong Shao
Discipline:Engineering
Document Type:J
Translation or Not:no
Date of Publication:2026-02-26
Included Journals:SCI