Reliability Assessment For New Products By Using Small Sample And Masked Data Information

The reliability assessment of new products plays a vital role in its further research and development,but the problem of the assessment is there is not enough data for effective statistical inference.Therefore this paper products three models,using small-sample field data,masked historical life data and accelerated life test(ALT)data,to make reliability assessment for new products.The main content is as follows.Firstly,in the cases that the component life distribution is exponential and Weibull,based on the idea of the least squares method,this paper provide a robust model which is easy to calculate,incorporating the concept of relative entropy.The model changes the objective function of the least squares method to fusing small-sample field data and ALT data.Considering the promotion of the model,the ALT data fused in this model comes from accelerated life tests with double stresses.The Monte Carlo simulation proves the effectiveness of the model,and when the size of sample is small,the model is also robust.Secondly,in the cases that the component life distribution is exponential and Weibull,this paper provides a model to use the historical system life data of each component in the new product.The model improves the derivation of the original masked data likelihood function by using statistical learning method.Then this paper derives the Bayesian estimation of component life distribution parameters,as well as the reliability function of new products.When the component life distribution is Weibull,the parameter estimation algorithm of MCMC method is given.Monte Carlo simulation shows that the results of this model are between those of other methods,and in the case of a small sample,the model results have no major deviation.As the data simulation sample size increases,the estimation tends to stabilize,which shows that the model is effective and robust.Finally,in the cases that the component life distribution is exponential and Weibull,through the independent fusion method,field information and historical system life data information are used to jointly determine the prior distribution of the component life distribution parameters,and the ALT data is used as the sample joint density function to finally obtain the Bayesian estimation of the component life distribution parameters in the new product.When the component life distribution is Weibull,the parameter estimation algorithm of MCMC method is given.The Monte Carlo simulation shows that this model is effective and robust,which is similar to the previous model.The estimation of this model has a slightly larger deviation than the previous model,which is might due to the influence of field data.