Bayesian network modelling of an offshore electrical generation system for applications within an asset integrity case for normally unattended offshore installations. (November 2018)

• Record Type: Journal Article
• Authors: Loughney, Sean; Wang, Jin
• Is Part Of: Proceedings of the Institution of Mechanical Engineers. Volume 232:Number 4(2018:Nov.)
• Page Start: 402
• Page End: 420
• Publication Date: 2018-11
• Subjects: Offshore safety -- Integrity Case -- Bayesian networks -- offshore installations -- electrical generation systems; Marine engineering -- Periodicals; 623.805
• DOI: 10.1177/1475090217704787
• Languages: English
• ISSNs: 1475-0902

Abstract

This article proposes the initial stages of the application of Bayesian networks in conducting quantitative risk assessment of the integrity of an offshore system. The main focus is the construction of a Bayesian network model that demonstrates the interactions of multiple offshore safety critical elements to analyse asset integrity. The majority of the data required to complete the Bayesian network was gathered from various databases and past risk assessment experiments and projects. However, where data were incomplete or non-existent, expert judgement was applied through pairwise comparison, analytical hierarchy process and a symmetric method to fill these data gaps and to complete larger conditional probability tables. A normally unattended installation–Integrity Case will enable the user to determine the impact of deficiencies in asset integrity and demonstrate that integrity is being managed to ensure safe operations in situations whereby physical human-to-machine interaction is not occurring. The Integrity Case can be said to be dynamic as it shall be continually updated for an installation as the quantitative risk analysis data are recorded. This allows for the integrity of the various systems and components of an offshore installation to be continually monitored. The Bayesian network allows cause and effect relationships to be modelled through clear graphical representation. The model accommodates for continual updating of failure data.