Monday, February 25, 2002
11:30 am, Science Building, Small Auditorium

Michelle H. Capozzoli

Bristol-Myers Squibb Company

New Bayesian Approach to Modeling Accelerated Life Test Data

Abstract: An accelerated life test (ALT) is often used to obtain timely information for very reliable items on test. In studying the reliability of high-tech components and products, even those that are nearly infallible, failures can often be traced to manufacturing or engineering faults. Under accelerated environmental stress conditions, even benign manufacturing faults or "flaws" may eventually cause the unit to fail. In this paper, we propose a new model which is motivated by the actual failure process of units on test and allows for modeling units that remain free of manufacturing faults and may not fail in their "technological lifetime." The approach described herein yields a new class of models and associated analyses under a Bayesian framework, which are unique to the classical ALT procedures. For illustration, the data set from Meeker and LuValle (1995) on the lifetimes of printed circuit boards subjected to different levels of relative humidity will be investigated.