Temperature field model for large spatial structures: Experiment, simulation and ANN prediction

Abstract

Large spatial stainless steel structures can be susceptible to accidental events such as fires, given their high occupancy, wide range of combustible materials and diverse usage. Consequently, a comprehensive investigation into the temperature field in large spatial stainless steel structures in the event of a fire has been conducted, encompassing experiments, simulations, assessments and predictions. Eight scaled temperature field tests were performed using a pool fire as the fire source, a common scenario in fire incidents. Building on the completed tests, a calibrated CFD model was developed using the Fire Dynamics Simulator (FDS) software and employed to further analyse the temperature field in large spatial structures under various fire powers and radii. A total of 8064 sets of three-dimensional large spatial temperature field data were acquired. Existing temperature field models, both from codes and other research studies, were evaluated against a substantial dataset. The results indicated that current models tend to be conservative, especially in areas near the fire source. In response to these findings, a novel approach utilizing Artificial Neural Networks to predict the 3D spatial temperature field in large spatial stainless steel structures is introduced. In addition, compared with the complex calculation formulae of traditional models, the model proposed herein based on Artificial Neural Networks is more convenient to use in practice and exhibits better accuracy.