DYNAMIC PROBABILITY FAILURE USING BAYESIAN NETWORK FOR HYDROGEN INFRASTRUCTURE MODELING

To produce large scale hydrogen production, it requires adequate and efficient risk
control. For decades, fault tree analysis was the most widely used tool for risk
assessment for industrial sector generally and hydrogen infrastructure particularly in
terms of risk and consequences associated to it. The limitation to this tool is it tends
to be static and do not develop over time which can give unreliable estimation of
risk.
The purpose of this project is to study the suitability and efficiency of dynamic
Bayesian Networks in terms of projecting the risk probability failure that develop
over time for hydrogen infrastructure as the alternative of the fault tree analysis. In
this study, only the risk probability failure is covered without further exploration on
the consequences of the risk. The process involved by the conversion of fault tree to
Bayesian Networks model by using appropriate framework. Then, the conditional
probability table is assigned to each node where the numbers of CPT depend on the
numbers of relationship between nodes. Finally the temporal reasoning is done to
show the time-invariant between each node and the beliefs is updated to get the
results.
The ways of inference use for this study are filtering and smoothing. The results
show that generally, the OR gates contribute to higher risk probability compare to
AND gates. Besides that, the probability for hydrogen activities increase from year to
year with the assumption the accident did not happen the previous year. In addition,
the instantaneous release incident is relatively low and unlikely to happen compare to
the continuous release.