VMD可视化及深度学习的滚动轴承故障诊断  

Fault Diagnosis of Rolling Bearing with VMD Visualization and Deep Learning

作　　者：魏航信[1] 程欢 吴伟[1] 王晓荣 WEI Hang-xin;CHENG Huan;WU Wei;WANG Xiao-rong(College of Mechanical Engineering,Xi’an Shiyou University,Shaanxi Xi’an 710065,China;Oil and Gas Technology Research Institute,Changqing Oilfield,Shaanxi Xi’an 710018,China)

机构地区：[1]西安石油大学机械工程学院,陕西西安710065 [2]长庆油田油气工艺研究院,陕西西安710018

出　　处：《机械设计与制造》2024年第7期210-214,220,共6页Machinery Design & Manufacture

基　　金：国家自然基金青年科学基金项目(51405385);陕西省科技厅科技攻关项目(2014K07-20)。

摘　　要：滚动轴承故障检测信号具有非线性、不平稳的特点,且特征量难以提取,因此提出变分模态分解(VMD)信号的可视化与深度学习神经网络相结合的方法以诊断轴承故障。首先对轴承原始振动信号进行VMD分解,滤除信号噪声;其次采用希尔伯特黄变换消除VMD存在的“欠包络”问题;接着对一维时间序列信号进行可视化变换,提取信号的格莱姆角视场(GAF)二维特征图,以充分反映不同故障的特征。最后采用卷积神经网络(CNN)对可视化图形进行诊断,CNN网络包括两个卷积层和两个池化层,卷积层的内核均为(5×5),池化层内核均为(2×2),卷积层深度分别为20和32。对采集的10类轴承振动信号进行诊断,训练集样本数量为3791,训练精度为96.5%,测试集样本数量为209,测试精度为95.2%,证明了本方法的有效性。The fault detection signal of rolling bearing has the characteristics of non-linearity and unevenness,and the characteristic quantity is difficult to extract.Therefore,a method combining variation modal decomposition(VMD)signal visualization and deep learning neural network is proposed to diagnose bearing faults.Firstly,VMD method on the original vibration signal of the bearing is performed to filter out the signal noise.Secondly,Hilbert-Huang algorithm is used to eliminate the“under-envelope”problem of VMD.Thirdly,the one-dimensional time series signal is visualized,and the two-dimensional feature map based on the Gramian Angular Field(GAF)is extracted.Finally,the convolutional neural network(CNN)is used to diagnose visualized images.The CNN network includes 2 convolutional layers and 2 pooling layers.The kernel of the convolutional layer is(5×5),and the pooling layer kernel is(2×2).The depth of two convolutional layers are 20 and 32 respectively.10 kinds of vibration signals collected are diagnosed.The number of samples in the training set is 3791,and the training accuracy is 96.5%.The number of the samples in test set is 209,and the test accuracy is 95.2%.Therefore,the effectiveness of this method is proved.

关 键 词：深度学习 VMD 格莱姆角视场 故障诊断 滚动轴承 

分 类 号：TH16[机械工程—机械制造及自动化] TH133.33TH165.3