Degradation State Assessment And Remaining Life Prediction Of Rolling Bearing Based On CNN And GRU

Rolling bearings are the most commonly used and most vulnerable mechanical parts in the industry,any unexpected failure of rolling bearing may bring huge loss to the factory.In addition,the vibration signal of rolling bearing contains abundant state information,which lays a foundation for the prediction of remaining life by data-driven method.In order to prevent bearing failure,the remaining useful life(RUL)of rolling bearing can be predicted by using its vibration signal,therefore maintenance can be carried out in advance.Based on the vibration signal of rolling bearing,this thesis studies the degradation state modeling and remaining life prediction of rolling bearing under data-driven method.Firstly,this thesis studies degenerate feature extraction of rolling bearing,and constructs two kinds of degradation features.By analyzing the time domain,frequency domain and time-frequency domain feature extraction methods of vibration signal of rolling bearing,the similarity features are constructed based on similarity,and sensitive features are selected as the first degradation features of rolling bearing.Then,in order to reduce the complexity of constructing the degradation features,the cumulative amplitude feature(SF)in frequency domain is proposed,which is more smooth and stable than the amplitude feature in frequency domain.Secondly,two kinds of health indicators of rolling bearing are established by using Convolutional Neural Network(CNN)and Gated Recurrent Neural Network(GRU).First of all,according to degradation features of rolling bearing based on similarity,the CNN-HI model is constructed by using the advantages of local connection,weight sharing and space pooling of CNN.According to the model,the degradation value of bearing is predicted and the degradation curve is obtained.Then,according to the SF feature,using the advantage of CNN to extract deep features,the dimension of SF feature is reduced to 64 dimensions.Considering the advantage of using GRU to process time series,GRU-HI model is established to predict the degradation value of rolling bearing and output the life degradation curve.Finally,the particle filter algorithm is used for prognosis.For the degradation curve of rolling bearing,the double exponential model is used to fit the curve to obtain the initial parameters,and then the model parameters are optimized by using the particle filter algorithm,which has the excellent prediction ability and uncertainty management ability.according to updating the state value continuously to predict the degradation value.Finally,degradation value is mapped to remaining life of the rolling bearing.The above models are verified on the PHM2012 public data set.The results show that,the degradation state modeling method which based on SF feature and GRU-HI has higher prediction accuracy than the similarity feature and CNN-HI method.In order to further verify the effect of the model,they are compared with the remaining life prediction method based on RMS and recurrent neural network.The results show that the prediction accuracy of two methods proposed in this thesis are better than the comparison model.