采用多层次一致性和半监督深度网络的轴承域适应故障诊断方法  

Bearing domain adaptive fault diagnosis method using multi-level consistency and semi-supervised deep networks

作　　者：沈建军[1] 于树源 贾峰 蒋昭宇 SHEN Jianjun;YU Shuyuan;JIA Feng;JIANG Zhaoyu(Key Laboratory of Road Construction Technology and Equipment of Ministry of Education,Chang an University,Xi'an 710064,China)

机构地区：[1]长安大学道路施工技术与装备教育部重点实验室

出　　处：《机电工程》2025年第2期267-276,共10页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(52105085);陕西省重点研发计划项目(2023-YBGY-327)。

摘　　要：在滚动轴承的故障诊断中,深度学习智能故障诊断的成功很大程度上依赖于充足的标记数据;然而,实际的情况是,收集大量标记数据常常面临困难和高昂成本,而大量未标记的数据则未被有效利用。针对这一难题,提出了一种基于多层次一致性半监督深度网络(MLC-SDN)的滚动轴承智能故障诊断方法。首先,将轴承原始信号经过数据预处理转为二维时频图,建立了特征提取器模块,利用深度卷积网络将轴承样本映射到高维特征空间;然后,在域间层面,采用基于样本的最优传输方法,利用目标样本不同视图的优缺点,稳健准确地对齐源域和目标域;在样本层面上,将弱增强视图的预测设置为强增强视图的伪标签,以保证一致性,同时,将非目标类的预测分布纳入优化目标,避免其与目标类的竞争,从而提高了伪标签生成的预测置信度;最后,为了验证MLC-SDN的有效性,利用三种轴承数据集进行了对比实验。研究结果表明:该方法在不同数据集上均取得了预测精度超过95%的结果。MLC-SDN方法不仅可以充分利用有限标记数据,同时在处理未标记数据和实现高精度故障诊断方面具有广泛的适用性。In recent years,fault diagnosis of rolling bearing has attracted increasing attention and has emerged as an important area of concern.The success of deep learning-based intelligent fault diagnosis largely relies on an abundance of labeled data.However,in practical scenarios,collecting large quantities of labeled data is often challenging and costly,and a significant amount of unlabeled data is not effectively utilized.To address this issue,a semi-supervised deep network with multi-level consistency(MLC-SDN)was proposed for intelligent fault diagnosis of bearings.Firstly,after data preprocessing,the original bearing signal was transformed into a two-dimensional time-frequency graph.The feature extractor module was established,and bearing samples were mapped to the high-dimensional feature space through a deep convolutional network.Then,at the inter-domain level,the optimal transmission method based on samples was employed,and the strengths and weaknesses of different views of the target sample were exploited to align the source domain and the target domain robustly and accurately.At the sample level,the prediction of the weakly enhanced view was set as the pseudo-label of the strongly enhanced view to ensure consistency.Simultaneously,the prediction distribution of non-target classes was incorporated into the optimization objective to avoid competition with the target class,thereby enhancing the prediction confidence generated by the pseudo-label.Finally,in ordor to verify the effectiveness of MLC-SDN,comparative experiments were conducted using three bearing datasets.The research results demonstrate that the prediction accuracy of this method exceeds 95% on different datasets.The MLC-SDN method demonstrates its wide applicability in processing unlabeled data and achieving high-precision fault diagnosis while making full use of limited labeled data.

关 键 词：轴承智能故障诊断 多层次一致性 半监督深度网络 领域自适应 伪标签 一致性正则化 

分 类 号：TH133.3[机械工程—机械制造及自动化] TH17