域适应网络与平衡动态分布自适应的轴承变工况故障迁移诊断研究

STUDY OF DOMAIN ADAPTIVE NETWORK AND BALANCED DYNAMIC DISTRIBUTION ADAPTIVE FAULT MIGRATION DIAGNOSIS OF BEARING UNDER VARIABLE CONDITIONS

机械设备在工业现场下的工况复杂多变,导致故障样本分布不均,给传统机器学习带来巨大的困扰。针对上述问题,提出了一种基于域适应神经网络与平衡动态分布自适应的轴承故障迁移诊断方法。首先,利用小波变换改进卷积神经网络的卷积层,并自适应提取轴承样本特征。其次,利用最大均值差异度量和权重正则化在损失函数处理所生成的特征,改善样本分布差异,获取域适应神经网络模型。最后,利用A-distance距离改进平衡分布自适应,使其具备动态特性,进一步改善样本分布差异,通过KNN分类器实现轴承迁移诊断。经过实验验证,所提方法在同试验台和跨试验台案例验证中,能够较为精确地迁移出轴承故障状态,证明该方法可有效解决无标签样本在变工况条件下样本分布不均的问题,具备有效性与鲁棒性。

The complex and changeable working conditions of mechanical equipment in industrial field lead to uneven distribution of fault samples,which brings great trouble to traditional machine learning.In order to solve this problem,proposes a bearing fault transfer diagnosis method based on domain adaptive neural network and balanced dynamic distribution adaptive.Firstly,According to the characteristics of bearing vibration fault samples,the convolution layer of convolutional neural network is improved by wavelet transform,and the characteristics of bearing samples are extracted adaptively.Then,Maximum Mean Discrepancy measure and weight regularization are used to process the generated features in the loss function to reduce the difference in sample distribution and obtain the domain adaptive neural network model.Finally,A⁃distance is used to improve the equilibrium distribution adaptive to make it have dynamic characteristics,further improve the difference of sample distribution,and realize bearing transfer diagnosis by KNN classifier.Through experimental verification,the proposed method can accurately migrate the bearing fault state in the same bench rig cases and cross bench rig cases,proving that the method can effectively solve the problem of uneven distribution of unlabeled samples under variable working conditions,and has the effectiveness and robustness.