强噪声干扰下基于SVMD-FFCNN的深沟球轴承故障分类模型  

Fault classification model of deep groove ball bearing under strongnoise interference based on SVMD-FFCNN

作　　者：李友家 张忠伟 焦宗豪 李新宇 秦贺 LI Youjia;ZHANG Zhongwei;JIAO Zonghao;LI Xinyu;QIN He(College of Transportation and Vehicle Engineering,Shandong University of Technology,Zibo 255000,China)

机构地区：[1]山东理工大学交通与车辆工程学院,山东淄博255000

出　　处：《机电工程》2025年第4期686-696,共11页Journal of Mechanical & Electrical Engineering

基　　金：山东省自然科学基金青年基金资助项目(ZR2021QE108)。

摘　　要：针对滚动轴承振动信号易受到外界噪声的干扰,导致故障特征信号微弱甚至被淹没,难以提取有效的故障特征的问题,提出了一种基于逐次变分模态分解与特征融合卷积神经网络(SVMD-FFCNN)的故障诊断方法。首先,利用SVMD对原始振动信号进行了模态分解,得到了固有模态函数(IMF)分量,并计算了皮尔森相关系数,筛选出相关程度大的分量,对信号进行了重构,完成了信号的降噪工作,并以降噪后的信号作为输入数据;然后,搭建了特征融合卷积神经网络模型(FFCNN),对卷积神经网络(CNN)提取到的浅层特征以及利用不同映射方法获取的深层特征成分进行了融合,提取了更具代表性的故障特征;最后,以SoftMax作为分类器,进行了深沟球轴承故障的分类任务,采用SKF6203深沟球轴承,并利用搭建的轴承故障模拟实验台采集了深沟球轴承振动数据,对SVMD-FFCNN方法进行了实验验证,并将其与其他方法进行了对比分析。研究结果表明:SVMD方法能够有效降低噪声的干扰,相较于未经过SVMD降噪处理的信号,实测实验信号信噪比提升了116.22%,均方根误差减低了56.10%;SVMD-FFCNN方法在噪声环境下的平均准确精度达到了99.37%,且三个转速工况下的诊断精度均达到了99%以上。上述结果表明,该方法在噪声环境下具有更优越的故障诊断性能。Aiming at the problem that rolling bearing vibration signals are easily disturbed by external noise,it leads to weak or even submerged fault feature signals,and it is difficult to extract effective fault features.A fault diagnosis method based on successive variational modal decomposition and feature fusion convolutional neural network(SVMD-FFCNN)was proposed.Firstly,SVMD was used to decompose the original vibration signal into intrinsic mode function(IMF)components,and Pearson correlation coefficient was calculated to select the components with high correlation degree for signal reconstruction,achieving signal denoising.The denoised signal was used as input.Then,a feature fusion convolutional neural network model(FFCNN)was built to fuse the shallow features extracted by the convolutional neural network(CNN)and the deep feature components obtained through different mapping methods,in order to extract more representative fault features.Finally,using SoftMax as the classifier,the classification task of deep groove ball bearing faults was carried out.SKF6203 deep groove ball bearings were used,and vibration data of deep groove ball bearings were collected using the constructed bearing fault simulation experimental platform to experimentally verify the SVMD-FFCNN method.It was compared with other methods.The research results show that the SVMD method can effectively reduce the interference of noise.Comparing with the signal without SVMD denoising treatment,the measured experimental signal signal-to-noise ratio increases by 116.22%,and the root mean square error decreases by 56.10%.The SVMD-FFCNN method has an average accuracy of 99.37%in noisy environments,and the diagnostic accuracy under three speed conditions has reached over 99%.This indicates that the method has superior fault diagnosis performance in noisy environments.

关 键 词：滚动轴承 强噪声干扰 智能故障诊断 逐次变分模态分解 特征融合卷积神经网络 SoftMax分类器 

分 类 号：TH133.33[机械工程—机械制造及自动化]