Development and validation of a novel method to predict flame behavior in tank fires based on CFD modeling and machine learning

Ensuring storage tank farm safety involves systematic engineering. Tank fire with a large ullage height is a common type of accident and poses a serious threat to tank farms due to the air restrictions by ullage height. This study investigates the impact of ullage height on flame morphology, air entrainment, and burning behaviors through experiments and computational fluid dynamics (CFD) simulations. Results showed that ullage height of the tank significantly affect burning rate, flame morphology and air entrainment. Three burning regimes were identified as ullage height changes. Experimental and simulation data were then used in a machine learning (ML) model, which combines particle swarm optimization (PSO) and back-propagation neural networks (BPNN) to predict the mass burning rate and internal flow field. The input datasets included the tank diameter, ullage height, experimental mass burning rate, and the internal flow field predicted by the CFD model. The predicted results by the ML model agree well with the experimental and numerical data. It was shown that the larger number of the training datasets, the more accurate predictions. The new model provides a fast and efficient way to predict the burning behaviors and supports risk assessment for tank fire accidents with limited experimental and numerical inputs.