Evacuation modelling in road tunnel fires

Abstract: Evacuation model capabilities are rapidly improving, allowing the simulation of ever more complex scenarios in different types of environments. The definition of the best evacuation modelling approach for safety assessment is a key point for optimizing engineering work and ensuring the desired safety conditions. In the present thesis, a wide comparison between modelling approaches has been provided for the study of road tunnel evacuations. The models employed are FDS+Evac, buildingEXODUS, Gridflow, STEPS, Pathfinder and Simulex while the calculations provided in the Society of Fire Protection Engineering handbook have been used to compare the results of the computational models to the hydraulic method. Models have been used individually and a new framework has also been presented, namely the multi-model approach. The predictive capabilities of the modelling approaches employed have been tested for both hypothetical evacuation scenarios and a set of new tunnel experiments performed at the Department of Fire Safety Engineering and Systems Safety of Lund University, Sweden. The aim was to identify the appropriate approach in relation to the complexity of the scenario under consideration. Two key aspects have been analysed through modelling tools: 1) the influence of smoke on movement speeds and 2) the impact of way-finding installations on exit choice. Models are tested through an a priori vs a posteriori result comparison based on the collected experimental data. Results show that: 1) analytical calculations are not a sufficient method to simulate evacuation scenarios involving exit choice, 2) the use of model default settings produces significant differences in the results, 3) the calibration of model input requires different degrees of effort in relation to the embedded sophistication of the model, 4) an individual use of the model is sufficient if the evacuation modeller has the necessary information to calibrate the input, 5) the presented multi-model approach is required in the case of very complex scenarios; it has been used to test the sensitivity of the results to the model employed and provide an estimate of the uncertainty related to the input, the models and the data-sets embedded in the models.