基于WPD_CNN_Bilstm自主更新的轴承故障诊断模型  

The Independently Updated Bearing Fault DiagnosisModel Based on WPD_CNN_Bilstm

作　　者：赵奇 张骞[1,2] ZHAO Qi;ZHANG Qian(Mechanical and Electrical Engineering College,Qingdao University,Qingdao 266071,China;National Engineering Laboratory for High-speed Train System Integration,CRRC Qingdao Sifang Co.,LTD.,Qingdao 266111,China)

机构地区：[1]青岛大学机电工程学院,山东青岛266071 [2]中车青岛四方机车车辆股份有限公司高速列车系统集成国家工程实验室,山东青岛266111

出　　处：《青岛大学学报(工程技术版)》2024年第4期20-25,共6页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：山东省自然科学基金资助项目(ZR2019PEE011)。

摘　　要：针对列车滚动轴承故障诊断中存在的小样本条件与错误样本影响模型准确度的问题,提出一种基于小波包分解(wavelet packet decomposition, WPD)、二维卷积神经网络(2D-convolutional neural network, 2D-CNN)与双向长短期记忆网络(bidirectional long short-term memory, Bilstm)的列车轴承故障状态诊断方法。通过提取选定频段内特征的方法构建样本,结合2D-CNN与Bilstm搭建神经网络模型,增加自主更新功能并添加数据过滤的结构。通过轴承故障数据,验证所提出方法。在小样本量和样本数据不平衡的训练条件下,故障诊断模型对测试集的分类准确度为98.15%,无数据过滤功能模型的分类准确度降低0.36%,而有数据过滤功能模型的分类准确度提高0.28%,对未知样本的分类准确度提高0.33%。实验结果表明,该方法使模型在小样本条件下具有较高的分类准确度,减少错误样本对模型自主更新的影响。In the field of fault diagnosis for rolling bearings in trains,the challenge of dealing with small sample sizes and the impact of erroneous samples on model accuracy is a significant issue.A diagnostic method for train bearing fault conditions is proposed,which is based on Wavelet Packet Decomposition(WPD),Two-Dimensional Convolutional Neural Network(2D-CNN),and Bidirectional Long Short-Term Memory network(Bilstm).Samples are constructed by extracting features within selected frequency bands.A neural network model is construted by combining 2D-CNN with BiLSTM,enhancing it with an autonomous updating feature and incorporating a data filtering structure.It validates the proposed method through bearing fault data.Under the training conditions of small sample size and imbalanced sample data,the fault diagnosis model achieved a classification accuracy of 98.15%on the test set.The classification accuracy of the model without the data filtering function decreased by 0.36%,while the model with the data filtering function increased accuracy by 0.28%,and the classification accuracy for unknown samples improved by 0.33%.The experimental results indicate that the proposed method enables the model to maintain a high classification accuracy under small sample conditions,reducing the impact of erroneous samples on the autonomous updating of the model.

关 键 词：小波包分解 CNN-Bilstm算法 自主更新 数据过滤 

分 类 号：TN822[电子电信—信息与通信工程]