Uncertainty consideration in CFD-models via response surface modeling: Application on realistic dense and light gas dispersion simulations

Abstract

Major accidents in the process industry often lead to the release of light or dense gases, which can mean a thread to employees, local residents or to the environment. Possible scenarios are therefore analyzed and evaluated in advance for approval issues. There is a trend, where simple empirical models are being replaced with more complex numerical models. Gaussian dispersion models or models based on dimensional analysis approaches are for example, increasingly replaced by CFD simulations. The main reason for this is the potentially higher accuracy. However, usually scenarios using sharp parameter values are calculated, since comprehensive consideration of parameter distributions via Monte Carlo or Latin Hypercube Sampling fails due to the numerical effort. This includes the risk that the influence of uncertainties on the simulation results is not taken into account. Response surface methods offer an alternative, with which the CFD problem can be mapped onto an algebraic surrogate model. If this is sufficiently precise, parameter sampling can also be carried out with the surrogate as well, as shown in some publications. Previous investigations only demonstrated the basic principle using trivial dispersion models. In this paper two realistic CFD simulations from the plant safety area are considered: VOC emissions from a storage tank and near-ground dense gas emissions. The entire procedure of response surface determination and parameter studies was automated and parallelized for high-performance-computing, and is carried out on the underlying CFD grids. For the CFD simulations as well as for all visualizations, the commercial software ANSYS CFX and the open source software OpenFOAM were used. The aim of this paper is to demonstrate the method using industry-relevant applications as well as to show how this can be used in practical engineering applications. The quality of surrogate modeling, the numerical effort and advantages that can result from the procedure are discussed as well as advantages which may result from taking parameter uncertainties into account in safety studies.