Research On Rolling Bearing Fault Diagnosis And RUL Prediction Based On Deep Learning

Rolling bearing is an important part in mechanical equipment,its health or not will have a significant impact on the operation of mechanical equipment.In the real industrial environment,the rolling bearing may break its inner and outer rings and wear and tear the rolling body due to excessive load,oil pollution erosion and insufficient lubrication,so that the failure of mechanical equipment affects the normal operation,and the serious situation will cause difficult to recover.Based on this,it is of great significance to accurately and efficiently realize bearing fault diagnosis and Remaining Useful Life(RUL)prediction.It can provide the basis for the maintenance and maintenance of mechanical equipment,but also can reduce the maintenance cost of mechanical equipment and improve its safety.This paper takes rolling bearing as the research object,mainly from rolling bearing fault diagnosis and RUL prediction.The main research work is as follows:(1)In order to solve the problem of insufficient mining of fault and degradation characteristic information of rolling bearing vibration signals,this paper uses S transform to preprocess bearing vibration signals,and obtains the corresponding time-frequency graph data set,which is used as the experimental data of the model.(2)In order to solve the problems that the existing rolling bearing fault diagnosis methods can not adapt to select features under different load and cross working conditions,the fault recognition rate is not high and the fault discrimination time is too long.In this paper,the attention mechanism Sim AM is introduced into the fault diagnosis model to redistribute the weight among different feature dimensions in the time-frequency graph to enhance the model’s attention to important feature information.Meanwhile,the adaptive activation function Mate-ACON is introduced to further improve the performance of the model.Experimental results show that the proposed fault diagnosis model has high accuracy.(3)In order to solve the problem of a small number of samples in the rolling bearing RUL prediction data set in real environment,in this paper,time series generation Adjunct network(Time GAN)was used to augment the experimental data set,and t-SNE was used to visualize the feature distribution of the original experimental data and the generated experimental data.Then the CNN-LSTMRUL prediction model is used to predict the original experimental data set and the augmented experimental data set respectively.The experimental results prove the validity of the data generated by Time GAN and achieve the purpose of augmenting the experimental data set of rolling bearings.(4)In order to solve the problem of insufficient extraction of bearing degradation features in time-frequency graphs due to the single convolution kernel size of traditional CNN network model,a SAM-MSCDN feature extraction model was constructed in this paper to fully extract bearing degradation features at different scales.Then,the deep features extracted by SAM-MSCDN are used as the input of LSTM network to predict the RUL of rolling bearings.The experimental results show that the prediction results of the RUL prediction model constructed in this paper are closer to the real remaining service life of the rolling bearing,which proves the validity of the RUL prediction model.