Proceedings of ENCIT 2010 13 th Brazilian Congress of Thermal Sciences and Engineering Copyright © 2010 by ABCM December 05-10, 2010, Uberlandia, MG, Brazil PARALLEL COMPUTING SIMULATION OF MULTIPHASE FLOWS EMPLOYING THE VOF METHOD Karime Louise Zenedin Glitz, karime@sinmec.ufsc.br António Fábio Carvalho da Silva, afabio@sinmec.ufsc.br Carlos Newmar Donatti, newmar@sinmec.ufsc.br Clovis Raimundo Maliska, maliska@sinmec.ufsc.br SINMEC – Computational Fluid Dynamics Lab. Mechanical Engineering Department, UFSC. CEP: 88040-970, Florianópolis-SC, Brazil. Phone: +55 48 3721 9562 Abstract. In the last years numerical simulation became a very important tool in solving engineering problems, due to its versatility in dealing with problems of different engineering’s areas. However, as problems grow in complexity, such a technique is limited by the current capability of the single-processor computers. This reason and the development of cluster of computers lead to a wide use of parallel computing for solving flow problems in very large domains or in detailed scales. The parallelization of a computer code requires, however, modifications in the serial algorithm in order to deal with the exchange of data among the processors, and such changes depend on the method used to solve the flow. In this paper a parallel algorithm employed to solve transient multiphase flows using the Volume-of-Fluid method (VOF) with Piecewise-Linear Interface Calculation (PLIC) reconstruction technique is presented. The solution of the Navier-Stokes equations is obtained by using the PRessure Implicit Momentum Explicit (PRIME) method to treat the pressure-velocity coupling in a staggered arrangement of variables, while the parallelization is carried out using the Multiblock method and the MPI library. Three two-dimensional tests are carried out in order to assess the performance of the new algorithm: the broken dam with and without an obstacle and the rising bubble in a resting fluid. The objective of the first case is to test the capability of the parallel algorithm to propagate the pressure gradient to all sub-domains when the water arm, which is formed after water flows against the obstacle, reaches the opposite wall of the domain. Whilst the second test aims in assessing the computational time spent – as well as the speed-up factor – as a function of the number of processors employed when this algorithm is used. This is also the objective of the third test, at which the effects due to the interfacial tension are considered. All the numerical simulations were carried out by the supercomputer SGI Altix ICE 8200 of the Computational Fluid Dynamics Lab (SINMEC), at the Federal University of Santa Catarina (UFSC). As shown by the results, the parallel algorithm proposed in this work is able to deal with complex flows such as the broken dam with an obstacle. It can also be verified that the employment of the proposed parallelization technique results in a significant economy of CPU time when more than one processor is used. As can be verified through the results, when the computational domain is divided into 16 sub-domains, the simulation is four times faster than the serial one for the rising bubble case. The results presented in this paper confirm that the proposed algorithm represents an alternative to expensive serial simulations of transient multiphase flows employing the VOF method. Keywords: parallel computing, multiphase flow, VOF, Multiblock 1. INTRODUCTION In the last years numerical simulation became a very important tool in solving engineering problems, due to its versatility in dealing with problems of different engineering’s areas. However, as problem grows in complexity, such a technique is limited by the current capability of the single-processor computers. This reason and the development of cluster of computers lead to a wide use of parallel computing for solving flow problems in very large domains or in detailed scales. The parallelization of a problem means to divide it into several sub-problems and to solve each of them in separated processors. Each process runs simultaneously to the others and some information, which is employed as boundary condition, is exchange among the processors. This is done in order to achieve the solution in less CPU time, that is, than that required in a serial procedure. There are several ways to parallelize a computer code. For instance, it can be done by parallelizing only the linear system solver or by subdividing the solution domain into sub-domains. Such strategy is called domain decomposition and it can be done in time or in space. Another possibility of parallelization is to employ the Multiblock method. This method is similar do the domain decomposition in space, however, in the Multiblock method, the geometry characteristics control the subdivision of the domain (Ferziger and Peric, 2002). In this paper a parallel algorithm employed to solve transient multiphase flows using the Volume-of-Fluid (VOF) method (Hirt and Nichols, 1981) with Piecewise-Linear Interface Calculation (PLIC) reconstruction technique is presented (Malik and Bussmann, 2004; Kothe et al., 1996). The solution of the Navier-Stokes equations is obtained by using the PRessure Implicit Momentum Explicit (PRIME) method to treat the pressure-velocity coupling in a staggered