Treatment of uncertainty in Probabilistic Risk Assessment using Monte Carlo analysis

Probabilistic Risk assessment (PRA) keeps a key-place to support the decision-making in current process safety management approaches. Unfortunately, the scarcity of data (failures probabilities or frequencies) for PRA is sometimes used for discrediting the validity of such analyses. In order to increase the confidence in the PRA results and therefore to improve efficiently the safety level, uncertainties induced by the lack of knowledge should be accurately taken into account. In this context, the present paper deals with parametric uncertainty using Monte Carlo simulation on the basis of the bow-tie diagram which is the most used technique for performing PRA studies. Precisely, uncertainty study aims (1) to show how the uncertainty of input parameters spreads onto the output of the model at hand, i.e. accident frequency (uncertainty propagation) and (2) to inspect how the variation in the output of that model can be apportioned to the variation in the input parameters (sensitivity analysis). An illustrative example related to a steam boiler system is analyzed.