Análise e performance de métodos de estimação de parâmetros em tempos de falha, utilizando testes de vida acelerados

Abstract

Accelerated life tests consist of exposing components / systems to stress loads in order to reduce the failure to acceptable levels. The information obtained under accelerated conditions is considered in the forecasting of the behavior of the system under normal operating conditions. To relate the times obtained in stress condition with measures of reliability under normal conditions are considered physical-statistical models. In this work we used the Inverse Power Law model that is based on the Weibull distribution and is indicated when the stress factor is the voltage. The Reverse Power Law stipulates That model the parameter β of the Weibull distribution takes the same value at all levels of the stress factor Allowing the various tests to be accelerated Compared to each other. When this assumption is not verified it is usual to consider an estimate for the average of the estimates of this parameter in each level of the factor. This work proposes another solution for this problem, which uses the median β, but for this it is necessary to check the best estimation method for this parameter of interest. Throughout the simulation study, it was possible to notice the superiority of the Bayesian estimation method, in comparison with the maximum likelihood method and the least squares method. Regarding the implementation of the methods in real data, in 44 samples of polyester film, submitted to the levels of the stress factor are: 5, 7, 10 and 15 kV. In the simulation test duration of 910,425 hours, the different methods results are close estimates the reliability of the indicators of interest and the values of the Obtained estimates for the mean and median times, are higher by MMV.