基于双通道CNN与SSA-SVM的滚动轴承故障诊断

Rolling bearing fault diagnosis based on dual channel CNN and SSA-SVM

为有效提取滚动轴承故障信号特征,解决分类器对提取特征存在较强依赖的问题,本文提出了一种双通道特征融合卷积神经网络(CNN)与麻雀搜索算法优化的支持向量机(SSA-SVM)相结合的滚轴承故障诊断方法.通过建立一维卷积神经网络和二维卷积神经网络并行的双通道结构对轴承数据中的特征进行提取,并将双通道CNN提取到的故障特征在融合层融合,将全连接层结果作为SSA-SVM分类层的输入.通过SSA对SVM的参数进行优化来提高模型的分类精度.最后,在凯斯西储大学轴承数据集上将双通道CNN与SSA-SVM模型跟传统一维卷积神经网络和二维卷积神经网络进行对比以验证其有效性.实验结果表明,该模型有着更高的故障识别准确率.本文中所有代码与实验结果均已开源,可在https://github.com/suisuisuiaa/tbysuisui获取.

In order to effectively extract the features of rolling bearing fault signals and solve the problem that classifiers have strong dependence on the extracted features,this paper proposes a rolling bearing fault diagnosis method combing dual-channel feature fusion convolutional neural network(CNN)and support vector machine optimized by sparrow search algorithm(SSA-SVM).The features of bearing data are extracted by establishing a parallel dual channel structure of onedimensional CNN and two-dimensional CNN.Then the fault features extracted by dual channel structure are fused in the fusion layer,and the results of full connection layer are used as the input of SSA-SVM.In addition,SSA is exploited to optimize the parameters of SVM to improve the classification accuracy.Finally,the proposed method is compared with the traditional one-dimensional CNN and two-dimensional CNN using CaseWestern Reserve University bearing data set to verify its effectiveness.Experimental results demonstrate that the proposed method has higher fault identification accuracy.All code and experimental results of the method have been open source and available at https://github.com/suisuisuiaa/tbysuisui.