The effect of discounting failures and weighting data on the accuracy of some reliability growth models

Abstract

The effect of two parametric failure discounting methods on the accuracy of three discrete and two continuous reliability growth models is analyzed. Similar comparisons are made for two data-weighting methods. Graphs are used to make comparisons on the accuracy of these models without discounting or weighting, with discounting only, and with weighting only. The accuracy comparisons are made using Monte Carlo methods. The results show that cumulative growth models such as the AMSAA and maximum likelihood models have greater bias than the noncumulative regression models for the cases simulated. The results also show that the cumulative models appear to be more sensitive to failure discounting and thus more susceptible to yielding optimistic estimates of reliability than the regression-type models when failure discounting is employed. Failure discounting applied too frequently (e.g. after each successful test) can adversely affect the accuracy of any of the models analyzed.<>