Rolling Bearing Remaining Life Prediction Research Based On Convolutional Neural Network

In order to ensure the safe operation of mechanical equipment,it is necessary to carry out condition monitoring and life prediction for its main components as much as possible.Knowing the operation and fault information of rolling bearings is an important part of realizing the health and reliability of equipment.Carrying out analysis and use relevant methods to predict the remaining useful life of rolling bearings,which can bring reference value to preventive maintenance decisions,so as to formulate maintenance plans and avoid safety accidents.Since the vibration signal contained in the running process of the rolling bearing can highlight the state information of the bearing,it is very suitable to carry out RUL research by collecting the relevant characteristic information through the vibration signal.In this way,early and effective maintenance actions on bearing components can be implemented to prevent bearing defection.In this thesis,by exploring the relevant methods of bearing prediction,the research on the rolling bearing practiced in this thesis is carried out.The research contents of this thesis are as follows:1.This thesis starts from the operation state of rolling bearing and the application of the prediction of remaining useful life.The structural features and categories of rolling bearings are expounded,and the feasibility and necessity of using vibration sensors to collect relevant signals of bearings and predict their remaining useful life are further discussed.The feature of rolling bearing vibration signal are also analyzed,and then the feature extraction method of rolling bearing vibration signal based on three aspects: time domain feature,frequency domain feature and time-frequency domain feature is given.2.According to the vibration signal collected from the rolling bearing,the thesis can carry out feature extraction from three aspects: time domain feature,frequency domain feature and time-frequency domain feature,and obtain each principal element characterizing the bearing degradation feature by principal component analysis method.Then,according to the different degrees of sensitivity of each pivot element to the bearing degradation trend,feature screening is adopted for the pivot element obtained by principal component analysis,and the first two pivot element features that can better capture the bearing degradation state signal are selected as the final degradation feature set.This is the basis for the next modeling step.Then,a method based on CNN is used to build a rolling bearing degradation model,so as to establish a CNN model,and the life degradation curve of the bearing can be obtained through the CNN model.3.In order to make full use of the CNN’s characteristics of spatial pooling,local connection,weight sharing and the ability to extract relevant information of local deep features,the frequency domain amplitude features of the bearing are deeply excavated,so that the features can be reduced in dimension.And the thesis can take advantage of the LSTM to process time series,and use the bearing degradation state modeling method of LSTM network to build an LSTM model to predict the life of the rolling bearing and then obtain the life degradation curve.4.Finally,the double exponential method is adopted for the degradation value of the bearing life degradation curve to predict the remaining useful life of the bearing.Through the above two models,the verification process was carried out on the public data set.The outcomes indicate that the prediction results of the two models can more accurately approximate the real life value,And the prediction accuracy of the model constructed in Chapter 4 is better than that of the model constructed in Chapter 3.In addition,compared with the results of the life prediction method based on the recurrent neural network,the prediction accuracy of the two models in this topic is better,which proves that the two models constructed in this topic have better prediction accuracy and validity of the model for predicting the remaining useful life of bearings.