Warranty Claims Forecasting Based on Weibull Distribution for the First Failure
Based on the warranty claim of the first failure,the failure number n in month-in-service(MIS)is taken as the random variable,and used the two-parameter Weibull distribution to fit the warranty claim data,and used the maximum likelihood mothed to estimate the parameters of Weibull distribution,and established the prediction model of the cumulative failure quantity.The results show that,using warranty data in the past 12 months,the mean relative error(MRE)of predicting for the future from 1 to 3 months is 0.9％,and MRE of predicting from 4 to 6 months is 3.8％,but MRE in predicting the number of failures beyond 6 months is very bigger.In order to validate the influence of the sample size of truncated data on the prediction accuracy,the prediction models with truncated data T=6 and T=9 are established,and MRE in predicting the number of faults in the next 3 months is 16.1％and 2.4％,respectively,which shows that increasing the sample data size can effectively improve the prediction accuracy.The warranty data of T=12 is used to predict the number of faults in the next 3 consecutive months,and the prediction accuracy is very good.If only the number of claim faults in the next one month is predicted,the warranty data of T=9 can be used,because the prediction accuracy can also reach the analysis requirements.However,it is not recommended to use warranty data less than 6 months for forecasting.
reliabilitywarranty data analysiswarranty claim predictionWeibull distribu-tionmaximum likelihood estimation
NETL
NSTL
万方数据
