PREDICT: a case study, using fuzzy logic

Los Alamos National Laboratory, Design For Reliability, Inc., and others, have worked together to develop PREDICT, a new methodology to characterize the reliability of a new product during its development program. Rather than conducting testing after hardware has been built, and developing statistical confidence bands around the results, this updating approach starts with an early reliability estimate characterized by large uncertainty, and then proceeds to reduce the uncertainty by folding in fresh information in a Bayesian framework. A considerable amount of knowledge is available at the beginning of a program in the form of expert judgment that helps to provide the initial estimate. This estimate is then continually updated as substantial and varied information becomes available during the course of the development program. This paper presents a case study of the application of PREDICT, including an example of the use of fuzzy logic, with the objective of further describing the methodology.