L’objectif de la thèse est d’effectuer l’analyse de sensibilité paramétrique du logiciel Phast de modélisation de la dispersion atmosphérique de gaz toxiques et/ou inflammables. La technique a consisté à coupler Phast et l’outil d’analyse de sensibilité SimLab, ce qui permet d’exécuter automatiquement un grand nombre de simulations en faisant varier l’ensemble des paramètres du modèle de façon simultanée. La méthode d’analyse de sensibilité globale choisie, E-FAST, est basée sur l’analyse de la variance des sorties du modèle pour le calcul des indices de sensibilité. Nous avons étudié des scénarios de rejet continus pour six produits différents (monoxyde d’azote, ammoniac, chlore, azote, n-hexane et fluorure d’hydrogène), sélectionnés pour couvrir une large gamme de caractéristiques physiques et de conditions de stockage. L’analyse du modèle de dispersion de Phast, Unified Dispersion Model, a été séparée en deux étapes : étape de « screening » avec pour but de comparer l’influence de l’ensemble des paramètres puis étude de l’influence globale des paramètres de modélisation, autres que les paramètres météo et du terme source, sur une plage large de valeurs. Pour chaque produit, nous avons décomposé les scénarios de base en sous-scénarios correspondant à des conditions de rejet différentes. Ce travail a notamment permis de classifier les paramètres du modèle selon leur degré d’influence sur la variabilité de différentes sorties et d’effectuer une analyse comparative par produit indiquant, pour des conditions de rejet données, quels paramètres sont les plus influents sur les sorties. Une étude complémentaire a consisté à effectuer une analyse de sensibilité locale de ces paramètres autour de leur valeur par défaut. / We have undertaken a parametric sensitivity analysis of the Phast software tool’s models for atmospheric dispersion of toxic and/or inflammable gases. We have coupled Phast with the sensitivity analysis tool SimLab, and have automated the execution of a large number of simulations while varying simultaneously selected model parameters. The global sensitivity analysis method used, E-FAST, is based on analysis of the variance of model outputs, and allows us to estimate sensitivity indices. We have studied continuous release scenarios for six different products (nitric oxide, ammonia, chlorine, nitrogen, n-hexane and hydrogen fluoride), which were chosen to cover a wide range of physical characteristics and storage conditions. Our analysis of Phast’s Unified Dispersion Model comprises two phases: a screening phase which allows the sensitivity of a wide range of parameters to be compared, followed by a phase focusing on the sensitivity of internal model parameters (excluding weather and source term variables), over a wide input range. For each product, we have broken down base-case scenarios into a number of sub-scenarios corresponding to different release conditions. This work has allowed us to rank model parameters according to their influence on the variability of a number of model outputs. It also includes a per-product comparative analysis indicating, for each release condition studied, which parameters have the most influence on the outputs. In the final part of the work, we have analyzed the local sensitivity of these parameters in a narrow range around their default values.

In this work, a new approach to optimize facility layout for toxic release, fire and explosion scenarios is presented. By integrating a risk analysis in the optimization formulation, safer assignments for facility layout and siting have been obtained. Accompanying with the economical concepts used in a plant layout, the new model considers the cost of willing to avoid a fatality, i.e. the potential injury cost due to accidents associated with toxic release near residential areas. For fire and explosion scenarios, the building or equipment damage cost replaces the potential injury cost. Two different approaches have been proposed to optimize the total cost related with layout. In the first phase using continuous-plane approach, the overall problem was initially modeled as a disjunctive program where the coordinates of each facility and cost-related variables are the main unknowns. Then, the convex hull approach was used to reformulate the problem as a Mixed Integer Non-Linear Program (MINLP) that identifies potential layouts by minimizing overall costs. This approach gives the coordinates of each facility in a continuous plane, and estimates for the total length of pipes, the land area, and the selection of safety devices. Finally, the 3D-computational fluid dynamics (CFD) was used to compare the difference between the initial layout and the final layout in order to see how obstacles and separation distances affect the dispersion or overpressures of affected facilities. One of the CFD programs, ANSYS CFX was employed for the dispersion study and Flame Acceleration Simulator (FLACS) for the fires and explosions. In the second phase for fire and explosion scenarios, the study is focused on finding an optimal placement for hazardous facilities and other process plant buildings using the optimization theory and mapping risks on the given land in order to calculate risk in financial terms. The given land is divided in a square grid of which the sides have a certain size and in which each square acquires a risk-score. These risk-scores such as the probability of structural damage are to be multiplied by prices of potential facilities which would be built on the grid. Finally this will give us the financial risk. Accompanying the suggested safety concepts, the new model takes into account construction and operational costs. The overall cost of locations is a function of piping cost, management cost, protection device cost, and financial risk. This approach gives the coordinates of the best location of each facility in a 2-D plane, and estimates the total piping length. Once the final layout is obtained, the CFD code, FLACS is used to simulate and consider obstacle effects in 3-D space. The outcome of this study will be useful in assisting the selection of location for process plant buildings and risk management.

The Biscayne Aquifer is a primary source of water supply in Southeast Florida. As a coastal aquifer, it is threatened by saltwater intrusion (SWI) when the natural groundwater flow is altered by over-pumping of groundwater. SWI is detrimental to the quality of fresh groundwater sources, making the water unfit for drinking due to mixing and reactions with aquifer minerals. Increasing water demand and complex environmental issues thus force water utilities in South Florida to sustainably manage saltwater intrusion and develop alternative water supplies (e.g., aquifer storage and recovery, ASR). The objectives of this study were to develop and use calibrated geochemical models to estimate water quality changes during saline intrusion and during ASR in south Florida. A batch-reaction model of saltwater intrusion was developed and important geochemical reactions were inferred. Additionally, a reactive transport model was developed to assess fate and transport of major ions and trace metals (Fe, As) at the Kissimmee River ASR. Finally, a cost-effective management of saltwater intrusion that involves using abstraction and recharge wells was implemented and optimized for the case of the Biscayne Aquifer. Major processes in the SWI areas were found to be mixing and dissolution-precipitation reactions with calcite and dolomite. Most of the major ions (Cl, Na, K, Mg, SO4) behaved conservatively during ASR while Ca and alkalinity were affected by carbonate reactions and cation exchange. A complex set of reactions involving thermodynamic equilibrium, kinetics and surface complexation reactions was required in the ASR model to simulate observed concentrations of Fe and As. The saltwater management model aimed at finding optimal locations and flow rates for abstraction and recharge wells. Optimal solutions (i.e., minimum total salt and total cost Pareto front) were produced for the Biscayne Aquifer for scenarios of surface recharge induced by climate change-affected precipitation. In general, abstraction at the maximum rate near the coast and artificial recharge at locations much further inland were found to be optimal. Knowledge developed herein directly supports the understanding of SWI caused by anthropogenic stressors, such as over-pumping and sea level rise, on coastal aquifers.