Research On Vehicle Reliability Data Modeling Based On Support Vector Regression

With the continuous development of the manufacturing industry and the continuous improvement of product quality requirements,the reliability of products has become an important factor for consumers to make purchase choices.At the same time,the reliability of products has also become an important factor in making the product warranty strategy for manufacturers.If reliability can be accurately predicted by product reliability data,it is easier for enterprises to make production,quality assurance,budget and strategy.The reliability data needed is derived from the warranty and claim behavior during the warranty period.On the other hand,it can be derived from the production life test or production investigation.At present,owing to the value expansion after production sales,the analysis of product reliability data has attracted widespread attention.In the existing traditional research,some prediction methods have been developed,such as the traditional Weibull distribution model,the log-linear Poisson prediction model,the Kalman filter model and the time series model.However there are two main problems about current forecasting methods:(1)they are based on certain distribution models,but the effect on reliability of factors is often uncertain,which greatly restricts the accuracy of prediction;(2)they are developed based on maintenance rate,which deduct the utility of data,failed to fully discover the value of reliability data.With the evolution of computing and storage capacity of hardware,machine learning technology is developing rapidly,many machine learning algorithms remaining in the theoretical research for long time are implemented,which realizes practical value for production activities in the era of big data.To overcome the limitations of the traditional methods,this thesis introduces the support vector machine regression(SVR)method to construct the reliability modeling.Support vector machines(SVM)has the advantages of strong robustness,strong generalization ability and high accuracy,and it breaks through the traditional distributed control.In addition,modeling discussion is implemented,such as parameter optimization,the amount of data,the choice of kernel function and so on,meanwhile,the relationship between mileage and operating time under the same reliability is studied,at last other machine learning methods are introduced to expand the depth of research.