基于LWKConv-DRSN-FPN的旋转机械故障诊断  

Fault Diagnosis of Rotating Machinery Based on LWKConv-DRSN-FPN

作　　者：伍兴 李志伟[1] 宁文乐 郑照 WU Xing;LI Zhiwei;NING Wenle;ZHENG Zhao(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201600,China;Hangzhou Industry Internet Technology Co.,Ltd.,Hangzhou 310000,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201600 [2]杭州工互科技有限公司,杭州310000

出　　处：《噪声与振动控制》2024年第5期133-139,共7页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(61705127)。

摘　　要：针对传统旋转机械故障诊断方法难以应对强噪声干扰以及诊断准确率较低的问题,提出一种Laplace小波核卷积层(Laplace Wavelet Kernel Convolutional Layer,LWKConv)、深度残差收缩网络(Deep Residual Shrinkage Networks,DRSN)和特征金字塔网络(Feature Pyramid Networks,FPN)相结合的故障诊断方法。具体地,在DRSN模型结构基础上,构造LWKConv,通过更新尺度因子和平移因子,多尺度提取故障引起的突变冲击特征;引入FPN融合深层和浅层特征,提高模型对浅层细节信息的利用程度,实现对旋转机械的故障诊断。研究表明:所提的LWKConv-DRSN-FPN方法基于轴承和齿轮数据集的诊断准确率最高能达到100%,尤其在-4 dB强噪声干扰条件下的诊断准确率达到97.75%,能有效提取突变冲击特征,具有较好的通用性和抗强噪声干扰能力。In order to solve the problem that the traditional fault diagnosis methods of rotating machinery are difficult to deal with the strong noise interference and their low diagnostic accuracy,a new fault diagnosis method combining Laplace wavelet kernel convolutional layer(LWKConv),deep residual shrinkage networks(DRSN)and feature pyramid networks(FPN)is proposed.Specifically,based on the DRSN model structure,the LWKConv is constructed to extract the mutation impact features caused by faults from multi-scale by updating the scale parameter and translation parameter.FPN is introduced to fuse the deep and shallow features to strengthen the use of the shallow details of the model,and realize the fault diagnosis of rotating machinery.The research shows that the diagnostic accuracy of the proposed LWKConv-DRSN-FPN method on bearing and gear dataset can reach 100%at most,especially under the condition of strong noise interference of-4 dB,the diagnostic accuracy rate can reach 97.75%.The proposed method can effectively extract mutation impact features,and has good versatility and the ability to resist strong noise interference.

关 键 词：故障诊断 旋转机械 Laplace小波核卷积层 深度残差收缩网络 特征金字塔网络 

分 类 号：TH133[机械工程—机械制造及自动化] TP206[自动化与计算机技术—检测技术与自动化装置]