A structure information-assisted generalization network for fault diagnosis of out-of-round wheels of metro trains

被引：0

作者：

Jiang, Jinnan ^{[1
]}

Tao, Gongquan ^{[2
]}

Liang, Hongqin ^{[1
]}

Zhang, Kai ^{[1
]}

Xie, Qinglin ^{[2
]}

Lu, Chun ^{[1
]}

Wen, Zefeng ^{[2
]}

Xiao, Qian ^{[3
]}

机构：

[1] Southwest Jiaotong Univ, Sch Mech Engn, Chengdu 610031, Peoples R China

[2] Southwest Jiaotong Univ, State Key Lab Rail Transit Vehicle Syst, Chengdu 610031, Peoples R China

[3] East China Jiaotong Univ, Key Lab Conveyance & Equipment, Minist Educ, Nanchang 330013, Peoples R China

来源：

MEASUREMENT | 2025年 / 244卷

基金：

中国国家自然科学基金;

关键词：

Domain generalization; Triplet generalization loss; Structural information auxiliary branch; Adaptive weighting strategy;

D O I：

10.1016/j.measurement.2024.116519

中图分类号：

T [工业技术];

学科分类号：

08 ;

摘要：

Wheel out-of-roundness (OOR) is a common problem with metro wheels, negatively affecting the comfort and safety of metro trains. In recent years, most data-driven models for diagnosing OOR faults have overlooked the problem of degraded diagnostic performance due to the data distribution discrepancies caused by different working conditions. Therefore, a structural information-assisted generalization network (SIAGN) is proposed. Firstly, a triplet generalization loss is introduced to encourage the model to generate a discriminative decision boundary by forming triplet pairs across different domains and learn the domain-irrelevant features. Additionally, a structural information auxiliary branch is proposed to help the model learn the discriminative basis of faults, while an adaptive weighting strategy is proposed to assist the auxiliary branch in optimizing the classification performance of the model. Finally, the feasibility of SIAGN is validated on simulated and measured datasets, and its superiority compared to some domain generalization methods is demonstrated.

引用

页数：17