Estudo de parâmetros do método Multigrid para sistemas de equações 2D em CFD
Resumo
Resumo: A influência de alguns parâmetros do método multigrid geométrico sobre o tempo de CPU para diferentes modelos matemáticos é investigada. Os parâmetros investigados são: número de iterações internas do solver (); número de níveis de malha (L); tamanho do problema (N); esquemas CS e FAS para dois modelos; e o efeito causado pelo número de equações diferenciais em dois modelos matemáticos. Os parâmetros são estudados para a equação de Laplace, equações de Navier (Termoelasticidade linear), equações de Burgers e de Navier-Stokes para escoamento incompressível; nas equações de Navier-Stokes discute-se também o efeito do número de Reynolds. Para o método multigrid, são feitas simulações com iterações internas = 1, 2, 3, · · · , 10 e = 15, e, no caso das quações de Navier-Stokes, o necessário para confirmar a tendência. O número de níveis de malhas varia de L = 2 a L = Lmáximo com N = 5×5, 9×9, 17×17, · · · , 1025×1025. O desempenho do método multigrid nas equações de Navier-Stokes é comparado nas formulações função de corrente e velocidade ( - v) e função de corrente e vorticidade ( - !). As equações são usadas na forma bidimensional e em regime estacionário. Os algoritmos multigrid CS (Correction Scheme) e FAS (Full Approximation Scheme) são implementados para a equação de Laplace e equações de Navier. Para as equações de Burgers e de Navier-Stokes implementa-se o algoritmo FAS. As equações diferenciais parciais são discretizadas com o Método de Diferenças Finitas em malhas uniformes nas duas direções. Os sistemas de equações algébricas são resolvidos com o solver MSI (Modified Strongly Implicit), e no caso das equações de Navier-Stokes com o SOR (Successive Over-Relaxation), ambos associados ao método multigrid geométrico com ciclo V e razão de engrossamento dois. As informações foram transferidas entre as malhas com injeção na restrição e interpolação bilinear na prolongação. Apenas na formulação função de corrente e velocidade utilizouse a ponderação completa na restrição. Verificou-se principalmente que: o esquema FAS apresentou melhor desempenho que o CS nos problemas lineares; a redução do fator de aceleração do método multigrid não é causado pelo acoplamento das equações; a formulação - v apresentou maiores fatores de aceleração que a formulação - !, mas o tempo de execução do singlegrid com -! é menor que -v; as soluções da formulação - v são mais acuradas que as soluções da formulação - !, inclusive em malhas grossas. Os resultados foram comparados com método singlegrid e resultados disponíveis na literatura. Abstract: The influence of some parameters of the geometric multigrid on the CPU time for different mathematical models is investigated. The parameters investigated are: the number of inner iterations in the solver (); the number of grid levels (L); the size of the problem (N); CS and FAS schemes for two models; and the number of differential equations of the mathematical models. The mathematical models adopted are: Laplace, Navier (thermoelasticity), Burgers and Navier-Stokes equations for incompressible flow; in Navier-Stokes equations is studied also the Reynolds number Re. For the multigrid method, simulations are carried out with inner iterations = 1, 2, 3, · · · , 10 and = 15, and, in the case of the Navier-Stokes equations, the necessary to confirm the tendency. The number of grid levels ranging from of L = 2 to L = Lmaximum with N = 5 × 5, 9 × 9, 17 × 17, · · · , 1025 × 1025. The Navier-Stokes equations are solved according to formulations streamfunction-velocity ( - v) and streamfunction-vorticity ( - !) formulations, whose performances were compared in the lid-driven square cavity flow classic problem. All the problems are twodimensional steady-state. The multigrid algorithms CS (Correction Scheme) and FAS (Full Approximation Scheme) are implemented for Laplace and Navier equations. For Burgers and Navier-Stokes equations the FAS algorithm is implemented. The partial differential equations are discretized using the Finite Difference Method in niform grids in both directions. The algebraic equations are solved with MSI (Modified Strongly Implicit) and in the case of the Navier-Stokes equations with SOR (Successive Over-Relaxation) solver, both associated to geometric multigrid method with V-cycle, by using a coarsening ratio two. The information was transferred among the grids with the injection in restriction and bi-linear interpolation within prolongation. In the streamfunction-velocity formulation was used the full weighting in restriction only. It was verified mainly that: the performance of the FAS scheme is better than CS in linear problems; reducing the efficiency of the multigrid method is not caused by the coupling of the equations; the - v formulation presented higher values of speed-up’s that the - ! formulation, but the runtime of the singlegrid with -! formulation is smaller; the solutions of the -v formulation are more accurate than the solutions of the - ! formulation also in coarse grids. The results were compared with the singlegrid method and other ones available in the literature.
Collections
- Teses & Dissertações [10546]