Lors que nous examinons numériquement des phénomènes multiphasiques suite à un accidentgrave dans le réacteur nucléaire, la dimension caractéristique des zones multi-fluides(non-réactifs et réactifs) s’avère beaucoup plus petite que celle du bâtiment réacteur, cequi fait la Simulation Numérique Directe de la configuration à peine réalisable. Autrement,nous proposons de considérer la zone de mélange multiphasique comme une interface infinimentfine. Puis, le solveur de Riemann réactif est inséré dans la Méthode des ÉquationsDiscrètes Réactives (RDEM) pour calculer le front de combustion à grande vitesse représentépar une interface discontinue. Une approche anti-diffusive est ensuite couplée avec laRDEM afin de précisément simuler des interfaces réactives. La robustesse et l’efficacité decette approche en calculant tant des interfaces multiphasiques que des écoulements réactifssont à la fois améliorées grâce à la méthode ici proposée : upwind downwind-controlled splitting(UDCS). UDCS est capable de résoudre précisément des interfaces avec les maillagesnon-structurés multidimensionnels, y compris des fronts réactifs de détonation et de déflagration. / When numerically investigating multiphase phenomena during severe accidents in a reactorsystem, characteristic lengths of the multi-fluid zone (non-reactive and reactive) are foundto be much smaller than the volume of the reactor containment, which makes the directmodeling of the configuration hardly achievable. Alternatively, we propose to consider thephysical multiphase mixture zone as an infinitely thin interface. Then, the reactive Riemannsolver is inserted into the Reactive Discrete Equations Method (RDEM) to compute highspeed combustion waves represented by discontinuous interfaces. An anti-diffusive approachis also coupled with RDEM to accurately simulate reactive interfaces. Increased robustnessand efficiency when computing both multiphase interfaces and reacting flows are achievedthanks to an original upwind downwind-controlled splitting method (UDCS). UDCS is capableof accurately solving interfaces on multi-dimensional unstructured meshes, includingreacting fronts for both deflagration and detonation configurations.

This dissertation focused on development and utilization of numerical and experimental approaches to improve the CFD modeling of fluidization flow of cohesive micron size particles. The specific objectives of this research were: (1) Developing a cluster prediction mechanism applicable to Two-Fluid Modeling (TFM) of gas-solid systems (2) Developing more accurate drag models for Two-Fluid Modeling (TFM) of gas-solid fluidization flow with the presence of cohesive interparticle forces (3) using the developed model to explore the improvement of accuracy of TFM in simulation of fluidization flow of cohesive powders (4) Understanding the causes and influential factor which led to improvements and quantification of improvements (5) Gathering data from a fast fluidization flow and use these data for benchmark validations. Simulation results with two developed cluster-aware drag models showed that cluster prediction could effectively influence the results in both the first and second cluster-aware models. It was proven that improvement of accuracy of TFM modeling using three versions of the first hybrid model was significant and the best improvements were obtained by using the smallest values of the switch parameter which led to capturing the smallest chances of cluster prediction. In the case of the second hybrid model, dependence of critical model parameter on only Reynolds number led to the fact that improvement of accuracy was significant only in dense section of the fluidized bed. This finding may suggest that a more sophisticated particle resolved DNS model, which can span wide range of solid volume fraction, can be used in the formulation of the cluster-aware drag model. The results of experiment suing high speed imaging indicated the presence of particle clusters in the fluidization flow of FCC inside the riser of FIU-CFB facility. In addition, pressure data was successfully captured along the fluidization column of the facility and used as benchmark validation data for the second hybrid model developed in the present dissertation. It was shown the second hybrid model could predict the pressure data in the dense section of the fluidization column with better accuracy.