Uncertainty Quantification Implementation for Quantitative Risk Assessments of Hydrogen Systems

Quantitative risk assessments (QRAs) can provide deterministic estimates of risk criteria but involve significant assumptions and uncertainties. Uncertainty quantification (UQ) methodologies enable propagation of model input uncertainties through deterministic models. This propagation of uncertainty can be leveraged to assess the relative impact of each input uncertainty on the output risk uncertainty in a sensitivity analysis (SA). Latin hypercube sampling is used to produce risk predictions with quantified uncertainty. Surrogate-estimated sensitivity indices and plots are used to estimate the sensitivity of the risk predictions to each input uncertainty. Results from an indoor hydrogen forklift fueling facility risk assessment calculation show significant uncertainty in the predicted risk, given significant uncertainties in various input parameters, though the median risk prediction is close to the deterministic prediction. SA results for this example indicate that occupant location is the primary driver of risk uncertainty in this specific case; when conditioned on a fixed location, the next leading uncertainty source is the leak frequency of full-bore instrumentation leaks. Leak frequencies for full-bore leaks from other components contributed less than 10% of the uncertainty in risk. Applying UQ methods to QRA calculations provides both a quantitative understanding of the uncertainty associated with its risk predictions as well as an understanding of which uncertainty sources are driving that uncertainty.