The Research Of EMU Gearbox Reliability Based On Degradation Analysis

The work is mainly about the reliability of EMU gearbox,which combines reliability degradation method,involves a piece of EMU gearbox oil detection data.Also,training machine learning models to forecast Fe element content.Firstly,plotting oil mileage-Fe element content scatter and train mileage-Fe element content scatter,then calculating oil mileage-Fe element content Spearman Rank Correlation Coefficient and train mileage-Fe element content Spearman Rank Correlation Coefficient.It is obviously found out that oil mileage-Fe element content and train mileage-Fe element content is statistically positively related and oil mileage-Fe element content correlation is stronger.Secondly,first hitting time of the Wiener process obeys the inverse Gaussian distribution.The model with drift and diffusion parameters are constants and the model with drift parameter is a random number are established.The inference formulas of the parameters are given by MLE method.It is found out that,the reliability function and the CDF of the 2 model are similar,when the failure threshold is 120ug/g,the reliability rapidly declines;the threshold is 350ug/g or 5 00ug/g,the CDF and reliability are firstly slow change,then dramatically change.Thirdly,according to the properties of the first hitting time of the Gamma process,the CDF and the reliability function are given.The inference formulas of the parameters are given by means of the moment estimation method.Through the analysis of the reliability function,When the failure threshold is 120ug/g,the reliability is quickly reduced to a lower value.The failure threshold is 120ug/g or 500ug/g,the CDF and reliability are slowly changed and then drastically changed.Considering the reliability function,the oil will be changed at 400,000 km.Finally,using SVM,artificial neural network and random forest algorithm to fit Fe content.The k-fold CV method is used to optimize both the parameters of the three models,so that the errors of the training set and the test set are minimized at the same time.So the random forest model is best for the data.Through the analysis about EMU gearbox oil Fe element model and the reliability function,the conclusion is made that,the oil should be changed at 400,000 km.The model and method in this paper is completely suitable for gearbox reliability modeling.