Numerical and experimental validation of a fast-response Lagrangian modeling approach for hazardous substance releases in industrial and urban environments

The release of hazardous substances into the atmosphere is a major concern for both public authorities and industry to ensure the safety of people and infrastructure. During accidental events, to optimize interventions and decision making, rapid response atmospheric dispersion models can be implemented. For these models to meet the challenges of crisis management, they must meet the criteria of speed, accuracy, reliability and limited available computing resources. The scientist's challenge is to develop a model that meets all these criteria. It is with this objective that the development of the BUILD software was initiated, within a project resulting from a collaboration between the Fluid Mechanics and Acoustics Laboratory (LMFA) and the Military Applications Division of the French Alternative Energies and Atomic Energy Commission (CEA-DAM). In parallel to the development, the knowledge of the potential of a model goes through a validation process in which this thesis work is part, with the support of an additional partner, TotalEnergies. This step consists of evaluating the model through different configurations using reference data. Several configurations were studied to rigorously qualify the results of the BUILD model. To support our results, CFD simulations (RANS and LES) have been implemented to have an additional support to the experiments for the comparisons. These simulations allow quantifying some parameters that the experiment prevents from measuring. They have thus enabled to refine the analysis of the flow behavior and the species transport and to provide the experimental results. Within the framework of this thesis, the actions undertaken consisted in setting up CFD simulations on the different configurations, studying the results obtained and comparing them to the simulations from BUILD. However, the lack of information for some experimental conditions or the presence of physically inconsistent information can make the parameterization of CFD simulations complex. This work was the opportunity to propose solutions to overcome these problems. Finally, the intercomparison between the experiment, the CFD models and BUILD make to evaluate the performance of BUILD possible for a limited computing time and resources. In conclusion, these studies show the current capabilities of the BUILD model and highlight the priority perspectives considered to improve it. (author)