Health Evaluation Of Rolling Bearing Based On Deep Learning

Bearing is the core component of locomotive and vehicle transmission system,and its safety and reliability are closely related to the smooth operation of the train.The health monitoring of rolling bearing is of great significance to improve the stability and reliability of train operation and reduce the loss caused by bearing failure.With the continuous development of artificial intelligence and deep learning,fault diagnosis and health management(PHM)technology came into being.The data-driven residual service life(RUL)prediction technology of rolling bearings has been widely studied by many scholars at home and abroad.This thesis mainly studies the health evaluation of rolling bearing based on RUL prediction.Starting from the two key steps of RUL prediction: the construction of health indicators and prediction models,this thesis studies two different construction methods of health indicators to reduce the dependence on the prior knowledge of expert experience in the traditional construction methods,and verifies the effectiveness of the health indicators built by the prediction model,so as to achieve the rolling bearing health status assessment.The main research contents are as follows:(1)Performance degradation evaluation of rolling bearings based on SE＿LSTM.In order to reduce the dependence of expert knowledge and prior experience in the construction process of rolling bearing performance degradation indicators and improve the generalization ability of the degradation performance indicators,a multi-feature fusion based on channel domain attention mechanism(SE)and LSTM based rolling bearing performance degradation prediction model(SE＿LSTM)is proposed.First,extract the time and frequency domain characteristics of rolling bearing,and input them to SE after increasing the dimension＿In the LSTM model,the channel attention network(SENet)in the model assigns a suitable weight value to the characteristics of each channel to represent the importance of each feature,and then adds and fuses the weighted features of each channel to generate a one-dimensional performance degradation index,and then inputs the performance degradation index into the LSTM network to achieve performance degradation prediction.Through the experimental verification using PHM2012 bearing data set,the bearing data under three different working conditions is used as the training set,and the bearing data under one working condition is used as the test set.Compared with the best single feature model built for the performance degradation index,this method reduces 19.12% in RMSE and 29.51% in MAE.It is proved that the constructed performance degradation index is effective and has certain generalization ability.The model can achieve good results in the prediction of bearing performance degradation degree.(2)Rolling bearing health monitoring based on multi-layer noise reduction and self-attention mechanism.In order to reduce the impact of environmental noise on vibration signals used for health monitoring,a multi-layer noise reduction module based on singular value decomposition(SVD),wavelet threshold noise reduction and local mean decomposition(LMD)was built.Then,taking advantage of the advantages of LSTM and self-coder in processing timing signals and extracting characterization features,a self-coder structural model based on LSTM network is constructed to generate characterization features of input signals.The feature sequence is comprehensively evaluated using correlation,predictability and robustness,and the health index with the highest evaluation index is obtained.Finally,the health status of rolling bearing is predicted through the prediction model.(3)Online monitoring system of rolling bearing health status.In order to make the rolling bearing health status assessment technology practical and verify its effectiveness in solving practical problems,a set of online monitoring system for rolling bearing health status based on Python graphical framework pyqt5 has been developed,which combines Python’s in-depth learning model and data processing capabilities with pyqt5’s graphical interface,so that it is ready to effectively realize real-time monitoring and intelligent prediction of rolling bearing health status,In addition,it can set equipment parameters and send alarm information in time when the service life of the equipment reaches the limit threshold.It improves the reliability and safety of rolling bearing in operation.At the same time,the system can use the latest vibration data for secondary training of the original model to continuously improve the accuracy and reliability of the model prediction.