基于改进DANN和注意力机制的轴箱故障诊断方法  

Axlebox Fault Diagnosis Method Based on Improved DANN and Attention Mechanism

作　　者：王健 邬娜 杨建伟 吕百乐 WANG Jian;WU Na;YANG Jianwei;LV Baile(School of Mechanical-Electronic and Vehicle Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Beijing Key Laboratory of Service Performance Guarantee of Urban Rail Transit Vehicles,Beijing 100044,China)

机构地区：[1]北京建筑大学机电与车辆工程学院,北京100044 [2]城市轨道交通车辆服役性能保障北京市重点实验室,北京100044

出　　处：《机床与液压》2024年第16期194-199,共6页Machine Tool & Hydraulics

基　　金：国家自然科学基金项目(51975038,52272385);北京建筑大学青年教师科研能力提升计划(X21055,X22011)。

摘　　要：针对变工况条件下现有深度学习网络模型对滚动轴承故障分类效果不佳的问题,以滚动轴承实验台数据为研究对象,提出迁移学习和注意力机制相结合的滚动轴承故障诊断方法。结合域对抗神经网络(DANN)与宽卷积核卷积神经网络(WDCNN)得到新的网络诊断模型(WDAANN),并通过对目标域的带标签数据进行训练以优化网络参数;结合注意力机制方法使所提网络获得更好的分类能力,从而实现变工况下的滚动轴承故障诊断。最终将该方法与传统CNN、DANN、WDAANN等模型进行对比验证。结果表明:所提方法的准确率提高,且模型的跨域诊断能力提高;所提网络的性能相比WDCNN、CNN及WDAANN网络明显提升,验证了所设计模型的优越性。To address the problem of poor performance of the existing deep learning network model on rolling bearing fault classification under variable operating conditions,taking the experimental data obtained from a rolling bearing bench as the research object,a rolling bearing fault diagnosis method was proposed based on the combination of migration learning and attention mechanism.A new network diagnosis model of WDAANN was obtained by combining domain adversarial neural network(DANN)and deep convolutional neural networks with wide first-layer kernel(WDCNN),and the network parameters were optimized by training the labeled data in the target domain;the proposed network was combined with the attentional mechanism to obtain a better classification ability,so as to realize the fault diagnosis of rolling bearings under variable operating conditions.Finally,the method was validated by comparing with traditional CNN,DANN and WDAANN models.The results show that the accuracy of the proposed method is improved,and the cross-domain diagnostic ability of the model is improved;compared with WDCNN,CNN and WDAANN network,the performance of the proposed network is significantly improved,which verifies the superiority of the designed model.

关 键 词：深度学习 迁移学习 变工况 注意力机制 分类准确率 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP391[自动化与计算机技术—计算机应用技术]