Parallelization and performance optimization of a dynamic PDE fixed bed reactor model for practical applications

Abstract

An important inherent limitation of dynamic multiphase reactor flow simulations is the computational time requirements, making long time statistics intractable. A parallel CFD model has therefore been developed intended for the simulation of multi-phase reactors. The present version of the model simulates 2D reactive flows in a fixed bed reactor. The simulations are performed on two grids of different resolutions. The predicted profiles are in accordance with results reported in the literature. Parallelization and performance optimization of the model have been performed to reduce the computational time. Further reductions have been achieved by applying compiler optimization. The most expensive part of the numerical solution algorithm is the implicit solution of the Poisson equation for the pressure. To solve the Poisson equation a TDMA-algorithm with and without a global block correction procedure, several variations of the conjugated gradients-algorithm and a bi-orthogonal conjugate gradient-algorithm were tested. The optimization work performed has shown that, compared to the serial non-optimized version of the code, the computational time spend solving the model has been reduced by more than an order of magnitude by using an optimized algorithm combined with optimal compiler options. Further reductions in computational time has been achieved by parallelizing the program. With this type of model performance optimization, the multiphase reactive flow systems in chemical reactors are expected to be simulated within feasible time limits.