Remaining Life Prediction Of Bearing Based On SPSO Optimized TWSVM And Bayesian Update Exponential Model

Rolling bearing is an important part of mechanical equipment.Its working condition and remaining using life are related to the healthy operation of machinery and equipment.The remaining using life of the rolling bearing is discrete.On the one hand,when the actual service life of the rolling bearing is shorter than the design service life,people may neglect the early weakness fault of the rolling bearing,which will threaten the performance and safety of the mechanical equipment in operation;on the other hand,the actual service life of the rolling bearing is greater than that of its design life,at the same time,the scrapping of rolling bearing is not conducive to maximizing the utilization of resources.Therefore,it is great significance to study effective methods to predict the remaining using life of rolling bearings accurately.With the rise of large data,cloud computing and artificial intelligence in recent years,rolling bearing life prediction method based on the data driven has been greatly applied.Feature extraction,feature evaluation,feature fusion and degenerate state evaluation are the big data base of rolling bearing life prediction.There are many nonlinear factors in the rolling bearing during high speed operation.The traditional linear and stationary feature extraction technology can easily lose important nonlinear state information and can not extract the effective state characteristics that reflect the essence of nonlinear vibration from complex nonlinear signals.Therefore,the fractal dimension,entropy value,recursive analysis and trigonometric feature are proposed in this paper.The nonlinear characteristic measurement methods are very sensitive to the recognition of rolling bearing degradation trend,and have good statistical characteristics.Considering the degradation law of the rolling bearing,this paper applies the trend index to the nonlinear feature evaluation method to select the better characteristics,and at the same time,a weighted fusion method based on trend index is proposed to achieve the fusion of the final features.In the evaluation stage of degradation,a method based on Chebyshev inequality and 25? criterion are proposed to identify the decline period and failure period of rolling bearings.The above methods have achieved good results in the experimental stage.In order to meet the demand of rolling bearing life prediction under different working conditions,this paper puts forward two methods of predicting the remaining life of rolling bearings.One method is to use a typical bearing degradation data as the training sample set to establish the prediction model,and then to predict the remaining life of the test bearing.This method is suitable for operating conditions of accumulating bearing data with sufficient and complete.The other method is to predict the remaining life of the bearing at the present time by using the historical data of the bearing to predict the remaining life of the bearing at present.The method does not need to accumulate the prior data of other bearings and directly uses the historical degradation data of the bearing to build a prediction model.The two methods are:(1)residual life prediction of rolling bearings based on SPSO-TWSVM(SPSOTWSVM,Simple Particle Swarm Optimization-Twins Support Vector Machine): this method can form a multivariable input matrix by phase space reconstruction after the fusion of features,and use TWSVM to train and learn to find the mapping association between the multivariable characteristic matrix and the remaining life,and the TWSVM parameter is optimized by SPSO method.This method is applied to the prediction of the remaining life of a pair of rolling bearings,and the prediction method is simultaneously with the SPSO-SVM(SPSO-SVM,Simple Particle Swarm Optimization-Support Vector Machine).The comparison proves the superiority of SPSO-TWSVM proposed in this paper.(2)The remaining life prediction of rolling bearing based on SPSO optimization Bayesian renewal exponential model: this method uses the Bayesian process to evaluate the parameters of the exponential model by using the rolling bearing state data as a priori information,and selects the remaining life as the final remaining life when the remaining life probability density function is the maximum.In addition,the SPSO method is used to evaluate the non random parameters of the exponential model,thus avoiding the sensitivity of the expectation maximization(EM,Expectation Maximization)parameter evaluation method to the initial iteration parameters,as well as improving the accuracy of the remaining life pretest.In this paper,the method is applied to the prediction of remaining life of a single rolling bearing,and a high prediction accuracy is obtained.At the same time,the evaluation of the non random parameters of the exponential model by the EM method is compared,the sensitivity of the initial iterative parameters of the method is shown,and the stability of the SPSO optimization of the non random unknown parameters proposed in this paper is compared with the stability of EM method.The strong contrast of accuracy proves the superiority of SPSO optimization method.