基于多通道一维卷积神经网络特征学习的齿轮箱故障诊断方法

Gearbox fault diagnosis based on feature learning of multi-channel one-dimensional convolutional neural network

为了解决单通道图像信号输入不能全面表达故障特征的问题,提出基于多通道一维卷积神经网络(Multi-Channel One-dimensional Convolutional Neural Network,MC-1DCNN)的故障特征学习方法。利用经验模态分解(Empirical Mode Decomposition,EMD)方法对信号进行处理,得到多通道一维信号;构建MC-1DCNN模型,对多通道一维信号进行特征提取。在MC-1DCNN的全连接层后接堆叠降噪自编码器(Stacked Denoised Autoencoder,SDAE)层,进一步进行维度缩减和特征提取并实现特征分类。通过某型号齿轮箱故障诊断实验对所提方法进行验证,实验结果表明,所提方法的特征提取能力和故障诊断效果显著优于典型的深度学习方法和机器学习分类器。

A new DNN model,called multi-channel one-dimensional convolutional neural network(MC-1DCNN)was proposed in order to solve the problem of using single-channel signal images as input,which can not express the fault characteristics hidden in the vibration signals effectively.Firstly,empirical mode decomposition(EMD)was used to obtain the multi-channel one-dimensional signals.Secondly,MC-1DCNN was constructed to perform the feature extraction of the multi-channel one-dimensional signals.Finally,stacked denoised autoencoder(SDAE)was embedded after the fully connected layer for further feature extraction and classification.The effectiveness of the proposed method was verified on the gearbox test rig.The experimental results show that the proposed method has better performance on feature extraction and fault diagnosis than typical DNNs and other regular classifiers.