Analysis of Degradation Data in Reliability

Reliability studies often measure the degrading performance of units before failure occurs. We identify three types of variability: (1) intrinsic variability or variation of the observation from its expected value because of the probabilistic nature of the degradation process, (2) variability due to measurement errors, and (3) extrinsic variability or variability due to unit-to-unit variations. This paper identifies Markov processes as an important class of degradation models and provides estimators of these three types of variability, as well as goodness of fit tests, for continuous time Markov chains. Estimators of the intrinsic variability and of the variability due to the measurement errors are derived from the degradation data measured on a single unit.