自监督学习结合对抗迁移的跨工况轴承故障诊断

Self-supervised Learning Combined with Adversarial Transfer for Cross-conditions Bearing Fault Diagnosis

轴承智能故障诊断应用中,由于实际工况复杂多变,极难获得足够的真实故障数据,且目标域和源域信号存在较大差异,导致深度模型的跨工况迁移识别也出现特征提取及分类困难、模型泛化性弱。考虑到目标域存在大量无标签数据,引入无监督思想,提出基于自监督学习结合对抗迁移的改进方法。首先根据信号本身特点创建辅助任务,对大量无标签数据学习,建立源域与目标域故障类别之间的内在联系;再通过对抗域适应和联合最大平均差异将源域知识迁移到目标域中,结合辅助任务优化两域差异,最终实现目标域准确的故障分类。用2个公开的轴承数据集上验证了所提方法的性能,实验结果表明,所提方法的故障诊断识别准确率在多数情况下均高于98%。

In the application of intelligent bearing fault diagnosis,it is extremely difficult to obtain sufficient real fault data due to the complexity and variability of the actual working conditions,and there exist large differences between the signals in the target and source domains,leading to the problems of difficult feature extraction and classification and weak generalization of the model in the cross-working condition transfer recognition of the deep model.Considering the existence of a large amount of unlabelled data in the target domain,an unsupervised ideas and proposed an improved method based on self-supervised learning combined with adversarial transfer is introduced.Firstly,the pretext tasks are created based on the characteristics of the signal itself to learn from a large amount of unlabeled data and establish the intrinsic connections of fault categories between the source domain and the target domain;then the knowledge learned from the source domain is transferred to the target domain through adversarial domain adaptation and joint maximum mean difference,and finally,it is combined with the pretext task to optimize the difference between the two domains and achieved accurate fault classification in the target domain.Experimental results have verified the feasibility and good performance,accuracy is all higher than 98%in most cases.