Risk Modelling of a Hydrogen Refuelling Station Using a Bayesian Network

Fault trees and event trees have for decades been the most commonly applied modelling tools in both risk analysis in general and the risk analysis of hydrogen applications including infrastructure in particular. It is sometimes found challenging to make traditional Quantitative Risk Analyses sufficiently transparent and it is frequently challenging for outsiders to verify the probabilistic modelling. Bayesian Networks (BN) are a graphical representation of uncertain quantities and decisions that explicitly reveal the probabilistic dependence between the variables and the related information flow. It has been suggested that BN represent a modelling tool that is superior to both fault trees and event trees with respect to the structuring and modelling of large complex systems. This paper gives an introduction to BN and utilises a case study as a basis for discussing and demonstrating the suitability of BN for modelling the risks associated with the introduction of hydrogen as an energy carrier. In this study we explore the benefits of modelling a hydrogen refuelling station using BN. The study takes its point of departure in input from a traditional detailed Quantitative Risk Analysis conducted by DNV during the HyApproval project. We compare and discuss the two analyses with respect to their advantages and disadvantages. We especially focus on a comparison of transparency and the results that may be extracted from the two alternative procedures.