CNNLSTM深度神经网络在滚动轴承故障诊断中的应用

An Application of Convolution Neural Network and Long Short-Term Memory in Rolling Bearing Fault Diagnosis

针对大数据下的滚动轴承振动信号自适应故障特征提取与智能诊断问题,提出了一种结合卷积神经网络(CNN)与长短时记忆网络(LSTM)的故障诊断模型。首先通过网格搜索算法寻找到当前模型的最优初始参数;然后以原始一维振动信号作为模型的输入,利用网络CNN层自适应提取短时特征信并降维后作为LSTM层输入;接着利用LSTM层学习特征信息并训练神经网络模型;最后,网络输出层利用Softmax函数实现多故障模式识别,完成故障诊断。使用Spectra Quest机械故障综合模拟试验台实测数据集对模型进行验证,试验结果表明,与多层感知器、LSTM网络以及经典的LeNet5、AlexNet、VGG相比,所提出的CNN-LSTM模型的分类平均准确率可达99%以上,且模型结构比其他模型更简单,训练时间更短;同时,通过K折叠交叉验证算法对模型进行评价,结果表明CNN-LSTM模型计算误差较小且网络训练充分,未出现过拟合或欠拟合情况。

Aiming at the problem of adaptive fault feature extraction and intelligent diagnosis of rolling bearing vibration signals based on big data,a fault diagnosis model combining convolutional neural network(CNN)and long and short time memory network(LSTM)is proposed.Firstly,the optimal initial parameters of the current model are found by a grid search algorithm.Then,the original one-dimensional vibration signal is used as the input of the model,and the short-time feature signal is extracted by the network CNN layer adaptively and then the dimension of the extracted signal is reduced and is used as the input of the LSTM layer.Then the LSTM layer is used to learn the feature information and to train the neural network model.Finally,the network output layer uses Softmax function to realize multi-fault pattern recognition and complete fault diagnosis.Experimental results and comparisons with multilayer perceptron,LSTM network,LeNet5,AlexNet,and VGG show that the average classification accuracy of the proposed CNN-LSTM model reaches more than 99%,and the model structure is simpler than other models and the training time is shorter.Moreover,the K-fold cross-validation algorithm is used to evaluate the model,and the results show that the CNN-LSTM model has a small calculation error and sufficient network training,and there is no over-fitting or under-fitting.