Performance-based egress safety assessment of underground tunnels: simulation and artificial neural network approaches

Abstract In this study, an artificial neural network (ANN) model was proposed to evaluate the egress safety of underground tunnels during fire. Fire simulations were carried out using the Fire Dynamics Simulator (FDS) for underground tunnels with a general cross-section, considering fire size as a key variable. In addition, egress simulations were performed using the Pathfinder program, with the spacing of cross-passage and the width of fire doors set as variables. Through this process, the available safe egress time (ASET), required safe egress time (RSET) and the number of casualties were derived for each variable, and the egress safety characteristics of underground tunnels under various parameter combinations were analysed in detail. Based on the derived data, an ANN model was developed to derive the ASET, RSET and survival rate in underground tunnels during fire incidents. The proposed ANN model is expected to efficiently evaluate the egress safety of underground tunnels with general dimensions without the need to perform additional fire and egress simulations. The ANN-based prediction model achieved coefficients of determination (R²) of 0.99 for ASET, 0.99 for RSET and 0.98 for survival rate, with average error rates of 1.72%, 1.36% and 2.31%, respectively, demonstrating very high accuracy.