A Bayesian approach for predicting risk of autonomous underwater vehicle loss during their missions. (February 2016)

• Record Type: Journal Article
• Authors: Brito, Mario; Griffiths, Gwyn
• Is Part Of: Reliability engineering & system safety. Volume 146(2016:Feb.)
• Page Start: 55
• Page End: 67
• Publication Date: 2016-02
• Subjects: Bayesian networks -- Survival statistics -- Expert judgment elicitation -- Autonomous vehicles; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2015.10.004
• Languages: English
• ISSNs: 0951-8320

Abstract

Abstract: Autonomous Underwater Vehicles (AUVs) are effective platforms for science research and monitoring, and for military and commercial data-gathering purposes. However, there is an inevitable risk of loss during any mission. Quantifying the risk of loss is complex, due to the combination of vehicle reliability and environmental factors, and cannot be determined through analytical means alone. An alternative approach – formal expert judgment – is a time-consuming process; consequently a method is needed to broaden the applicability of judgments beyond the narrow confines of an elicitation for a defined environment. We propose and explore a solution founded on a Bayesian Belief Network (BBN), where the results of the expert judgment elicitation are taken as the initial prior probability of loss due to failure. The network topology captures the causal effects of the environment separately on the vehicle and on the support platform, and combines these to produce an updated probability of loss due to failure. An extended version of the Kaplan–Meier estimator is then used to update the mission risk profile with travelled distance. Sensitivity analysis of the BBN is presented and a case study of Autosub3 AUV deployment in the Amundsen Sea is discussed in detail. Highlights: Novel method to estimate risk of autonomous vehicle loss in uncertain environments. A framework to integrate frequentist and subjective probability modelling. A Bayesian belief updating method for capturing variation in operating environment. Graphical approach for sensitivity analysis, applicable to any BBN model validation. Pragmatic case studies showing the application of the proposed framework.