CONSEQUENCE MODELLING OF INDUSTRIAL FIRES USING COMPUTATIONAL FLUID DYNAMICS AND SURROGATE MODELLING

Abstract

Consequence modelling performed with computational fluid dynamics (CFD) promotes better understanding of complex hazards, and it can be used to create a more accurate quantitative risk assessment (QRA). However, this is impractical given the extensive computational resources required to simulate thousands of hazard scenarios. In this thesis, we develop a new methodology for QRA by integrating sampling algorithms, surrogate models and error measures. The proposed methodology maximizes knowledge gained from CFD-based consequence models and new applications for process safety can be created. This research is the first to train surrogate models for the prediction of net radiation flux from pool fires and jet fires. In addition, a new methodology for CFD-based QRA is proposed, which is applied in 3 case studies with a range of problem dimensionality to evaluate its performance.