Augmenting system tests with component tests for reliability assurance

Abstract

To successfully determine if a system meets some reliability standard, typically assurance tests are conducted on the full system. This proven approach is extremely effective. However, it can be cost prohibitive when full system tests are expensive. Previous work has incorporated component and subsystem data to supplement the component prior distributions before a full system assurance test. Although this can be helpful, we propose an additional advance of allowing component and subsystem tests as part of the assurance test. We show that by augmenting full system tests with component and subsystem tests, in some cases, we can reduce the cost of the assurance test. For a system that can be decomposed into independent series and parallel parts, the theory works perfectly. In the case where we cannot accurately classify parts using the series or parallel structure, we add a generalization that works for systems that are “almost-series” or “almost-parallel.”

Keywords:

• Bayesian testing
• component test
• consumer’s risk
• cost reduction
• demonstration test
• producer’s risk

Acknowledgements

Data sharing is not applicable to this article as no new data were created or analyzed in this study. We thank two anonymous referees whose insightful comments on an earlier version helped significantly improve this article.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Notes on contributors

Richard L. Warr

Richard L. Warr is an Associate Professor of Statistics at Brigham Young University. Dr. Warr graduated Magna Cum Laude with a BS in Mathematics from Southern Utah University in 1996. He earned an MA in Mathematics from the University of Nebraska at Omaha in 2005 and his MS and PhD degrees in Statistics at the University of New Mexico in 2009 and 2010. Dr. Warr’s research is focused on statistical methodology, specifically in the areas of reliability, random partition models, computational methods, and stochastic processes. In these areas he often uses a Bayesian nonparametric approach.

Jace Ritchie

Jace E. Ritchie is currently a PhD student at Utah State University. Jace graduated Summa Cum Laude with a BS and MS in Statistics from Brigham Young University in 2023 through an integrated program. Jace’s research has included optimal design of experiments, evolutionary algorithms, reliability, Bayesian statistics, and automobile crash safety. He also has experience utilizing machine learning to predict E-commerce metrics.

Michael Hamada

Michael Hamada is a retired Scientist from Los Alamos National Laboratory (LANL). He earned a Ph.D. in Statistics from the University of Wisconsin-Madison. He is a Fellow of LANL, ASQ, and ASA. His research interests include design of experiments, reliability, quality control, and measurement system assessment.