Statistical Inference On The Reliability Of Censord Life Data

The proposal of the "Made in China 2025" plan has aroused a new round of scientific and technological revolution and industrial reform in China,bringing wonderful opportunities and creativity improvement for industrial engineering.However,the impact of COVID-19 has intensified the competition among enterprises,and the industry has further increased the requirements for product quality,making the analysis and evaluation of product reliability more important.Statistical inference on the reliability of censord life data is the most widely used reliability evaluation technology in engineering practice nowadays,which has great theoretical value and practical significance.This paper discusses the statistical inference and application of industrial reliability in different occasions through different censord life data and accelerated life tests.The overall research content is as follows:(1)Based on constant-stress accelerated life test with multiple-stress,reliability inference of life products with stress interaction effect is discussed under Type-II censoring.When the lifetime of testing products follows a generalized inverse exponential distribution,a life-stress relationship is constructed by using generalized linear Eyring model with stress interaction effect.Maximum likelihood,least square and Bootstrap methods are used respectively to obtain point and interval estimates of the model parameters and reliability indices.Finally,extensive numerical simulations,sensitivity analysis and a real-life example are provided to illustrate the effectiveness of our methods.(2)Based on a k stage constant-stress accelerated life test,a life-stress model is established between the accelerated stress and the p quantile life of the product under the Weibull distribution.The unknown parameters of the model and the reliability index of the product are inferred by the two-step estimation method.In the first step,the point and interval estimation of Weibull life parameters and p quantile life under acceleration condition are established.In the second step,the acceleration coefficients of the life-stress model are solved by the iterative reweighted least square method.Then,the Weibull life parameters and p quantile life of products under normal stress are extrapolated by inverse estimation method.Simulation and case analysis show that the generalized estimation based on pivot quantity has higher estimation accuracy than the classical maximum likelihood estimation.(3)The Matusita measure estimation problem is discussed under the progressive firstfailure censored data.For two independent populations subject to generalized inverse exponential distribution,the Matusita measure estimation is established for the cases of equal scale parameters and unequal parameters,including maximum likelihood estimation and its existence and uniqueness,generalized pivot estimation,and Bootstrap confidence interval estimation.The results of simulation and case analysis show that both the likelihood method and the generalized method can obtain satisfactory estimation results.But the generalized estimation of CC pivot quantity shows the best performance among all the estimation methods.