Deep Learning Based Health State Assessment And Remaining Life Prediction Of Rolling Bearings

As an important component of rotating machinery,the health status of rolling bearing directly affects the reliability of the whole equipment.However,due to the complex working environment of rolling bearing,its health status will gradually decline,and the possibility of failure will gradually increase.Once the fault occurs,it will be stopped for maintenance,which will affect the normal work plan and cause economic losses;if the fault is serious,it will cause safety problems and casualties.Therefore,it is of great practical importance to evaluate the health state and predict the remaining useful life(RUL)of rolling bearings.In this paper,rolling bearings are taken as the research object.Based on the vibration signals of rolling bearings,the deep learning algorithm is used to extract the degradation features,construct the health indicators(HI)and predict the RUL of rolling bearings.The main research contents are as follows:Firstly,the common time-domain features,frequency-domain features and time-frequency domain features are extracted from the vibration signals of rolling bearings,and then the sensitive degradation features which can reflect the bearing degradation process are selected by using the monotonicity index,correlation index,robustness index and identifiability index.To solve the problem of information redundancy in sensitive degradation feature set,the sparse autoencoder(SAE)is used to fuse sensitive degradation feature set to get low dimensional fusion sensitive degradation feature set.Secondly,the whole life cycle of bearing is divided into health stage and degradation stage by determining the starting point of bearing degradation.Then,the mapping relationship between the fusion sensitive degradation features and the rolling bearing HI is established by bidirectional gated recurrent unit(BiGRU).Finally,the predicted RUL of the rolling bearing is calculated according to the corresponding relationship between the constructed HI and RUL.Thirdly,in order to give full play to the advantages of deep learning algorithm,in view of the problem that the extraction of degradation features needs a lot of professional knowledge of signal processing,the powerful feature extraction ability of convolutional neural network(CNN)is utilized to automatically extract the deep degradation features contained in the vibration signals of rolling bearings.Then,the deep degradation features are input into the BiGRU model to construct HI of rolling bearing,and finally the predicted RUL of bearing is obtained.Finally,the RUL prediction method proposed in this paper is verified on PHM2012 data set and XJTU-SY data set,and the prediction results are compared with those obtained by other commonly used prediction models.The results show that the BiGRU model has smaller prediction error for the degradation starting point and RUL of rolling bearing.At the same time,in XJTU-SY data set with less sample data,the RUL prediction error obtained by CNN feature extraction is smaller than that obtained by using fusion sensitive degradation features.However,the opposite result is obtained in PHM2012 data set with more sample data,which shows that CNN model for feature extraction needs a lot of training data to have a better performance,so the RUL prediction method based on CNN feature extraction and BiGRU is more suitable for a large number of sample data.