基于层级对比学习的小样本轴承故障诊断方法

打开文本图片集

关键词：轴承；对比学习；注意力机制；小样本学习；故障诊断中图分类号： U270.331+ .2 文献标志码：B doi：10.20213/j.cnki. tdcl.2024.12.24.01

Small-sample Bearing Fault Diagnosis Method Based on Hierarchical Contrastive Learning

ZHAO Zhihong1，² ， TAO Xu² （1. State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures， Shijiazhuang Tiedao University，Shijiazhuang O5oo43，China；2. School of Information Science and Technology，Shijiazhuang Tiedao University，Shijiazhuang O5oo43，China）

Abstract：In recent years，the deep learningtechnology is widely applied in the field of bearing fault diagnosis. The efects of these methods significantly rely on the support of massive flag data，however，it’s difficult to obtain masive fault samples in the actual production environment. How to use limited fault samples to train a stable bearing fault diagnosis model is a critical problem to be overcome urgently. The article proposes a smal-sample bearing fault diagnosis method （CMBDC） based on hierarchical contrastive learning. By using the characteristics of contrastive learning in shortening the distance between similar samples and increasing the distance between different samples，a hierarchical contrastive learning model is designed to perform the contrastive learning in both shallw and deep parts of feature extraction，helping the model learn richer feature representations at diferent feature abstraction levels. Then，the feature vectors of similar samples are concatenated and input into the parallel attention fusion module.The feature vectors obtained by multiple attention mechanisms are weighted and averaged to obtain the final category representation. Final， the predicted fault categories are obtained using the brownian distance covariance measure（DeepBDC） method. In orderto verify the effectivenessof the method described inthe article，tests have been conducted on the bearing datasets of Case Western Reserve University，Paderborn University and wheelsets of railway freight car.Under the setup of cross load test，the test accuracies have reached 99.50% ， 99.88% and 99.48% respectively.The testresults show that the method described in thearticle allws to recognize well the fault categories and has a good generalization performance and a certain engineering application value.

Key words： bearing； contrastive learning；attention mechanism; small-sample learning；fault diagnosis

故障诊断是智能制造的关键环节，维护机械设备在使用过程中的安全和健康[1]。（剩余13720字）