A Three-Dimensional Numerical Study of Gas-Particle Flow and Chemical Reactions in Circulating Fluidised Bed Reactors

Three-dimensional Computational Fluid Dynamics (CFD) simulations of Circulating Fluidized Beds (CFB's) have been performed. The computations are performed using a 3D multiphase computational fluid dynamics code with an Eulerian description of both gas and particle phases. The turbulent motion of the particulate phase is modeled using the kinetic theory for granular flow, and the gas phase turbulence is modeled using a Sub-Grid-Scale model. A computational study of a cold flowing CFB riser has been performed. The results have been compared to experimental findings of particle volume fraction, particle axial velocity, and pressure drop provided as a blind test in connection with the 10th International Workshop on Two-Phase Flow Prediction held in Merseburg, Germany, 2002. The simulated profiles are in good qualitative agreement with the experiments, but the extend of the radial solid segregation is under predicted in the lower part of the riser. In the upper dilute part of the riser there is good agreement between simulated and measured values both for the solid volume fraction and the pressure profile. A cold flowing riser fluidized with FCC catalysts has been studied. The results were submitted to a blind-test in connection to the 10th international workshop on two-phase flow prediction held in Merseburg, Germany, 2002. The results are validated against experimental findings of particle mass flux across the riser and pressure profile along the riser. The calculations show good agreement with experimental findings of both mass flux and pressure profile, but further improvements are proposed and investigated. A parameter study shows that mesh refinement, choice of particle diameter and choice of drag model are crucial when simulating FCC riser flow. The isothermal decomposition of ozone has been implemented in the CFD code FLOTRACS-MP-3D. The decomposition reaction is studied in a 3D representation of a 0.254 m i.d. riser, which has been studied experimentally by Ouyang et al. (1993). Comparison between measured and simulated time-averaged ozone concentration at different elevations in the riser shows good agreement. The 3D representation of the reactor geometry gives better predictions of the radial variation in concentration than in a similar 2D simulation, Samuelsberg and Hjertager (1995). A parameter study is performed to investigate improvements in the predicted pressure drop profile. When using two-fluid modeling to predict riser flows there have been difficulties in predicting the solids hold up in risers represented by the correct pressure drop profile. Mesh refinement has shown to improve the axial segregation of particles in the riser, but when simulating a riser with a large L/D like an FCC riser in 3D, one is still limited to using rather coarse grids. Another way to get better results would be to use more than one solids phase to represent the actual particle size distribution. The introduction of more particle phases and use of a fine mesh resolution leads to an increase in CPU time. To reduce the wall-clock time for the simulations, the in-house code, FLOTRACS-MP-3D, has been parallelized using domain decomposition. In order to study the influence of mesh resolution and the addition of Eulerian phases, two test cases are investigated. Flow in a lab-scale CFB of the boiler type with a low L/D of 8 and flow in a FCC riser with a larger L/D of 48. For both risers comparison between measured and simulated pressure profile along the riser is performed. For the CFB riser comparisons are made for solids velocities and for the FCC riser measured and simulated solids fluxes are compared. In both cases the simulations compare well to the experiments. For the FCC riser it is concluded that the introduction of more particle phases only give minor improvements in the predictions. For the CFB riser the use of a very fine mesh gives indications of a dense bottom zone that has been almost absent in previous simulations of the case. The simulations have been performed using a parallelized CFD code. It is demonstrated that the number of partitions has no effect in the time averaged results obtained from simulations.