Methodology for Assessing Reliability Growth Using Multiple Information Sources

Abstract

The research presented here examines the assessment of the reliability of a system or product utilizing multiple data sources available throughout the different stages of its development. The assessment of the reliability as it changes throughout the development of a system is traditionally referred to as reliability growth, which refers to the discovery and mitigation of failure modes within the system, thereby improving the underlying reliability. Traditional models for assessing reliability growth work with test data from individual test events to assess the system reliability at the current stage of development. These models track or project the reliability of the system as it matures subject to the specific assumptions of the models.

The contributions of this research are as follows. A new Bayesian reliability growth assessment technique is introduced for continuous-use systems under general corrective action strategies. The technique differs from those currently in the literature due to the allowance for arbitrary times for corrective actions. It also provides a probabilistic treatment of the various parameters within the model, accounting for the uncertainty present in the assessment. The Bayesian reliability growth assessment model is then extended to include results from operational testing. The approach considers the posterior distribution from the reliability growth assessment of the prior for the operational reliability assessment. The developmental and operational testing environments are not a priori assumed to be equivalent, and the change in environments is accounted for in a probabilistic manner within the model. A Bayesian reliability growth planning model is also presented that takes advantage of the reduced uncertainty in the combined operational assessment. The approach allows for reductions in the amount of demonstration testing necessary for a given level of uncertainty in the assessment, and it can also be used to reduce high design goals that often result from traditional operating characteristic curve applications. The final part of this research involves combining various sources of reliability information to obtain prior distributions on the system reliability. The approach presents a general framework for utilizing information such as component/subsystem testing, historical component reliability data, and physics-based modeling of specific component failure mechanisms.