Contributions to Reliability and Lifetime Data Analysis.

Abstract

This dissertation deals with problems in reliability and lifetime data analysis. The first part focuses on the study of graphical estimators from probability plots with right censored data. The second part deals with reliability inference for repairable systems.

Probability plots are popular graphical tools for assessing parametric distributional assumptions among reliability engineers and other practitioners. They are particularly well suited for location-scale families or those that can be transformed to such families. When the plot indicates a reasonable conformity to the assumed family, it is common to estimate the underlying location and scale parameters by fitting a line through the plot. This quick-and-easy method is especially useful with censored data. Indeed, the current version of a popular statistical software package uses this as the default estimation method. Part I of the dissertation investigates the properties of graphical estimators with multiply right-censored data and compares their performance to maximum likelihood estimators. Large-sample results on consistency, asymptotic normality, and asymptotic variance expressions are obtained. Small-sample properties are studied through simulation for selected distributions and censoring patterns. The results presented in this study extend the work of Nair (1984) to right-censored data.

Analysis of failure data arising from repairable systems has received considerable attention in the statistical, engineering, computer software, and medical literature. Data pertaining to a repairable system is viewed as some type of `recurrent event'. Part II of the dissertation investigates some models and methodologies for analyzing failures from repairable systems with multiple failure modes. We consider the case where the cause-specific failures (from each failure mode) follow some counting processes with an emphasis on nonhomogeneous Poisson processes (NHPPs). Some properties of the data are characterized and estimation methods are studied, both from a single system and multiple systems assuming independence of the failure modes. Some results are also developed when there is partial masking of the failure modes. The NHPP case with a power law intensity function is studied in detail.