基于MSDCNN-BiGRU-SVM的滚动轴承故障诊断  

Rolling bearing fault diagnosis based on MSDCNN-BiGRU-SVM

作　　者：洪乐 文传博(指导)[1] HONG Le;WEN Chuanbo(School of Electrical Engineering,Shanghai Dianji University,Shanghai 201306,China)

机构地区：[1]上海电机学院电气学院,上海201306

出　　处：《上海电机学院学报》2025年第1期1-6,共6页Journal of Shanghai Dianji University

基　　金：国家自然科学基金资助项目(61973209);上海市科技行动计划资助项目(22010501100)。

摘　　要：针对传统故障诊断方法特征提取不充分,复杂场景下诊断准确率低的问题,提出了一种结合神经网络特征提取能力与支持向量机(SVM)分类性能的故障诊断方法。首先,通过宽卷积核提取特征中的低频信息,并利用多尺度空洞卷积神经网络(MSDCNN)进行自适应特征提取;其次,通过坐标注意力机制(CA)自适应确定不同通道的特征权值,并利用双向门控循环单元(Bi GRU)进一步提取振动信号中的时序特征;最后,将所提取的特征信息归一化后输入SVM分类器,并输出故障诊断结果。实验结果表明:该方法与其他智能诊断方法相比,在噪声干扰和变负载条件下有更好的故障诊断性能。To address the issues of insufficient feature extraction in traditional fault diagnosis methods and low diagnostic accuracy in complex scenarios,a fault diagnosis method that combines the feature extraction capability of neural networks with the classification performance of support vector machines(SVM)is proposed.First,low-frequency information is extracted from the features using wide convolutional kernels,followed by adaptive feature extraction using multi-scale dilated convolutional neural networks(MSDCNN).Next,the feature weights of different channels are adaptively determined using the coordinate attention(CA)mechanism.Then,a bidirectional gated recurrent unit(BiGRU)is employed to further extract temporal features from the vibration signals.Finally,the extracted feature information is normalized and fed into the SVM classifier to produce the fault diagnosis results.Experimental results demonstrate that the proposed method achieves superior fault diagnosis performance under noise interference and variable load conditions compared to other intelligent diagnostic approaches.

关 键 词：轴承故障诊断 支持向量机 多尺度空洞卷积神经网络 坐标注意力机制 双向门控循环单元 

分 类 号：U226.8[交通运输工程—道路与铁道工程]