Bayesian Computations for Reliability Growth Modeling

Alaattin Erkanli, Thomas A. Mazzuchi, Refik Soyer
Duke University, George Washington University

Nov 30 1993

In this paper reliability growth model for both the attribute and variable data are considered. Bayesian analysis of these models requires inference over ordered regions. Even though closed form results for posterior quantities can be obtained in some cases [see for example, Mazzuchi and Soyer (1993)], in general when the number of test stages gets large computations become burdensome and more importantly the results become inaccurate due to the problems in evaluation of gamma functions involve the computations. In such cases the posterior and predictive analyses can be done more efficiently using Markov Chain Monte Carlo (MCMC) methods. We discuss use of the MCMC methods for inference in both attribute and the variable data reliability growth models and discuss extension of our approach to different reliability growth scenarios. We illustrate the implementation of the approach by using reliability growth data.


Bayesian inference, Markov chain Monte Carlo methods, ordered dirichlet distribution, gibbs sampling


PDF icon 1994-39.pdf