多特征融合CNN网络的旋转机械故障诊断研究

Research on Fault Diagnosis of Rotating Machinery Based on Multi-feature Fusion CNN Network

为提高旋转机械故障诊断中故障分类的准确率,以及针对故障数据特征不充足而带来的泛化能力较差问题,提出一种多特征融合卷积神经网络(CNN)的旋转机械故障诊断方法。首先利用连续小波变换将一维原始信号转换成二维小波时频图,构建多特征融合CNN网络模型。其中,原始振动信号为1DCNN模型输入,小波时频图为2DCNN模型输入;然后根据上面两个维度的输入进行网络模型训练;最后将测试集中的数据输入到已经训练好的网络模型,对不同旋转机械故障进行分类。在凯斯西储大学的轴承数据集、机械故障预防技术(MFPT)的轴承数据集上进行实验验证,结果表明,该方法与其他同类方法相比具有更高的故障诊断准确率,达到了99.78%。

To improve the accuracy of fault classification in fault diagnosis of rotating machinery,and solve the problem of insufficient fault data features resulting in poor generalization ability,a multi-feature fusion convolutional neural networks(CNN)network is proposed for fault diagnosis of rotating machinery.Firstly,the method converts the vibration signal of one-dimensional rotating machinery into the wavelet time-frequency diagrams by using continuous wavelet transform.After,a multi-feature fusion CNN network model is constructed,where the original vibration signals are the input of the 1DCNN model,and the wavelet time-frequency diagrams are the input of the 2DCNN model.Then,the network model is trained with the input of the above two dimensions.Finally,the test set is input into the trained network model to realize the specific classification of different rotating machinery faults.This method is verified on the bearing dataset of Case Western Reserve University and MFPT(Machinery Failure Prevent Technology).Compared with other similar methods,the results show that this method has a higher fault diagnosis accuracy,and its accuracy rate of fault diagnosis reached 99.78%.