Application of Bayesian Regularization Artificial Neural Network in explosion risk analysis of fixed offshore platform

Abstract

Computational Fluid Dynamics (CFD) is routinely used in Explosion Risk Analysis (ERA), as CFD-based ERA offers a good understanding of underlying physics accidental loads. Generally, simplifications were incorporated into CFD-based ERA to limit the number of simulations. Frozen Cloud Approach (FCA) is a frequently used simplification in the dispersion part of the CFD-based ERA procedure. However, its accuracy is questionable in the complex and congested environment such as offshore facility. Furthermore, in explosion part, some specific techniques, e.g. linear/double bin-interpolated techniques have been proposed while the corresponding accuracy is still unknown since the developers did not yet check their accuracy by considering the explosion computational data as the benchmark.This study presents a more accurate algorithm, namely Bayesian Regularization Artificial Neural Network (BRANN) and accordingly proposes the frameworks regarding BRANN-based models for the CFD-based ERA procedure. Firstly, the framework is proposed to develop the Transient-BRANN (TBRANN) model for transient dispersion study. In addition, the framework to determine the BRANN model for explosion study is developed. The proposed frameworks are explained by a case study of the fixed offshore platform. Consequently, this study confirms the more accuracy of the TBRANN model over FCA and the accuracy of BRANN model for CFD-based ERA.