Research On Fault Diagnosis And Remaining Life Prediction Of Rolling Bearing Based On Deep Learning

Rolling bearing is one of the most widely used core support components in rotating machinery,which is often used as the rotating component of the supporting shaft in the gearbox of high-speed railway EMU.When the rolling bearing fails,it is generally to collect equipment signals to reflect the running state of the rolling bearing and express the fault characteristic information.However,the traditional vibration signal analysis method is too dependent on data,and it is difficult to select and extract the characteristic information.With the advent of artificial intelligence era,fault diagnosis has a tendency to develop towards intelligent diagnosis.In view of the above difficulties,in this paper,vibration signals and deep learning methods are used to realize the fault diagnosis and remaining life prediction of rolling bearings.The specific content is as follows:(1)Fault diagnosis experiment of rolling bearing and vibration signal analysis method.Based on the acquisition process of vibration signals.This paper designs and builds a fault diagnosis experiment platform,and produces a data set of vibration signals for different damage positions of rolling bearings and different speeds.and the fault types are identified by EEMD decomposition and envelope analysis,at the same time,the Maximum Correlated Kurtosis Deconvolution(MCKD)is used for better denoising,and the denoised signal is filtered adaptively by empirical wavelet transform(EWT),and the components are obtained to realize fault diagnosis.(2)Using the vibration signals at different damage positions and different speeds,the deep learning method is used to realize the fault diagnosis and identification of the rolling bearing.An improved convolutional neural network(CNN)model is proposed,which improves the accuracy of fault diagnosis.And different data preprocessing methods and the fault diagnosis results under the CNN model are studied.The results show that the preprocessing method of full-wave rectification after normalizing the data according to the speed category is better.Combining the one-dimensional vibration signal into a two-dimensional vibration image as input,compared to using the vibration signal as the input directly,it can give full play to the advantages of the two-dimensional convolution network in processing image features.The recognition accuracy can reach 96.8%;the improved convolutional neural network proposed in this research,which combines the one-dimensional convolution model and the two-dimensional convolution model to extract features together,the classification is realized after feature fusion,the recognition accuracy was the highest,reaching 99.3%,indicating that the improved convolutional neural network can simultaneously express the timing characteristics of vibration signals and image characteristics,and suitable for vibration signals fault diagnosis.(3)Using full-period vibration signals,deep learning method is used to predict the remaining life of the rolling bearing and the failure stage.Firstly,the degradation trend of rolling bearings is predicted,and the characteristic indexes which can represent the degradation trend of rolling bearings are extracted;Then,the remaining life of rolling bearing is predicted by using temporal convolutional network(TCN),which using the single and combined features as the model input,the results show that compared with the LSTM network,the loss function of the TCN decreases faster,the training time is shorter,and the prediction trend is closer to the real situation;Finally,the TCN model is applied to predict the failure stage of the vibration signal at different time by using the vibration signal at different time as the input.The results show that the neural network can identify the operation stage of the vibration signal and judge the operation state of the bearing,so as to prevent the failure of the rolling bearing in advance,and the recognition accuracy can reach up to 93%.