On the analysis of reliability data

Abstract

Fitting models to observed failure data is an important topic in reliability. The resulting models can be useful both for the manufacturer, as well as for the end-user of the observed system. In this study, we first provide a summary of some useful results from reliability theory. We then present two procedures, which have been developed by extracting methods from the existing literature, that can be used as a starting point when analysing and fitting models to failure data. In particular we focus on obtaining analytical estimates of the lifetime distribution (for data from nonrepairable systems) and of the intensity of a NHPP (for data from a repairable system). Simulated data and claims data from the warranty database of a major car manufacturer are used to illustrate selected methods and tests from these procedures.