Leveraging prior knowledge into life test design

Abstract

Life tests are generally very time-consuming and expensive. For a highly reliable product, even an accelerated life test takes too much time to meet business and product development needs. This is particularly true in automotive component testing. Traditional life test design with classical statistics does not consider prior knowledge or previous test results. In real applications, one typically has some prior knowledge before designing a new testing. This knowledge can come from previous test, expert opinion, engineering analysis, past similar product performance, or combination of them. Utilizing this prior knowledge can reduce the test sample size or increase the confidence, and more importantly, help to make decisions quicker. Some studies in the area of utilizing prior knowledge are developed through Bayes approach. [1-4] are good examples in automotive industry application. However, the acceptance of Bayesian statistics proves to be much more challenging than the acceptance of classical statistics by management as well as by the engineering community and even by many reliability engineers. In this paper, we present a method that utilizes prior knowledge for the life test design but within a classical statistics framework. A Weibull time-to-failure distribution with a known shape parameter is used throughout the discussion. The prior knowledge of reliability and its confidence is translated into an imaginary or virtual life test to obtain an equivalent total time of testing and an equivalent number of failures. They are then quantitatively taken as the prior knowledge for a classical confidence bound based new life test design. This method will be compared with Bayes method. The prior knowledge determination and applications of the method developed here are discussed with examples.