A novel approach combining bootstrapped non-intrusive reduced order models and unscented transform for the robust and efficient CFD analysis of accidental gas releases in congested plants

The risk assessment for safety-critical, complex systems is a very challenging computational problem when it is performed with high-fidelity models, e.g. CFD, like in the case of accidental gas releases in congested systems. Within this framework, a novel CFD approach, named Source Box Accident Model, has been recently proposed to efficiently model such phenomena by splitting the simulation of the gas release and its subsequent dispersion in the system in two steps. In this view, the present paper proposes a non-intrusive, Proper Orthogonal Decomposition-Radial Basis Functions reduced order model that exploits the two-step nature of the SBAM approach, to mimic the behaviour of the original, long-running CFD model code at a significantly lower computational cost. Moreover, the paper presents a methodology combining the bootstrap and unscented transform approaches to efficiently assess the ROM uncertainty in the safety-critical simulation output quantities of interest, e.g. the flammable volume. The results obtained in a test case involving a high pressure, accidental gas release in an off-shore Oil & Gas plant are in very satisfactory agreement with those produced by CFD, with a relative error smaller than 10% and a reduction in the computational time of about three orders of magnitude.