Direct Simulation of the Motion of Particles in Flowing Liquids

Project Summary

The grand challenge proposed is to develop highly efficient methods for computing the three-dimensional motions of large numbers of particles in solid-liquid flows, under the action of the hydrodynamic forces and torques exerted by the suspending fluid, and to use these methods to elucidate the fundamental dynamics of particulate flows and solve problems of engineering interest. The goal is to develop high-performance, state-of-the-art software packages called particle movers, capable of simulating the motion of thousands of particles in 2-D and hundreds in 3-D domains, in both Newtonian fluids, governed by the Navier-Stokes equations, and in several popular models of viscoelastic fluids. Such simulations will be extremely computationally intensive. It is therefore imperative to develop the most efficient possible computational schemes, and to implement them on parallel machines, using state-of-the-art parallel algorithms.

To meet this challenge, we propose to develop two different computational schemes for simulating solid-liquid flows on parallel computers. The first is a generalization of the standard Galerkin finite element method in which both the fluid and particle equations of motion are incorporated into a single variational equation, containing both the fluid and particle velocities as primitive unknowns. The hydrodynamic forces and torques on the particles are eliminated in the formulation, so need not be computed as separate quantities. The computation is performed on an unstructured body-fitted grid, and an arbitrary Lagrangian-Eulerian moving mesh technique has been adopted to deal with the motion of the particles.

In the second approach, an embedding method, the fluid flow is computed as if the space occupied by the particles were filled with fluid. The no-slip boundary condition on the particle boundaries is enforced as a constraint using Lagrange multipliers. This allows a fixed grid to be used, eliminating the need for remeshing, a definite advantage in parallel implementations.

Both approaches have been initiated by us, for quite different kinds of applications. At present, one scheme does not fit all applications. Perhaps ultimately, a ``best'' universal scheme for moving particles may evolve, but it is not presently prudent to make a bet.

A crucial computational issue to be addressed is the efficient solution of the various algebraic systems which arise in the schemes. These systems can be extremely large for 3-D problems, and their solution can consume up to 95 of the CPU time of the entire simulation. It is therefore imperative to use efficient iterative solution methods, with matrix-free preconditioners, and to implement them on parallel architectures.

We plan to develop a library of parallel numerical algorithms to solve these systems. This parallel library will consist of algorithms for solving nonlinear algebraic equations using variants of Newton's method, preconditioned iterative solvers for sparse symmetric indefinite and nonsymmetric linear systems, and rapid elliptic and Stokes solvers on uniform grids. This library will be used for rapid prototyping of simulation codes for the application problems referred to above.

The library will be augmented with a collection of kernels to allow it to be efficiently portable across either the massive parallelism of the Cray T3-D or its successors, or cluster-based parallelism such as that of several interconnected SGI Power-Challenge workstations. Both architectures exhibit two-level parallelism that is ideally suited for schemes such as the embedding method on a fixed grid.

The new schemes will be used to study the microstructural (pair interaction) effects which produce clusters and anisotropic structures in particulate flows, to produce statistical analyses of particulate flows (mean values, fluctuation levels and spectral properties), to derive engineering correlations of the kind usually obtained from experiments, and to provide clues and closure data for the development of two-phase flow models and a standard against which to judge the performance of such models. They will also be used to solve practical problems of industrial interest such as sedimentation, fluidization and slurry transport of solid particles in Newtonian and viscoelastic fluids.

Finally, the results of all numerical simulations will be compared with experiments from the literature or experiments to be done in the Minnesota laboratory, or with field data from industry---especially from our industrial sponsors. The project will therefore advance the science of solid-liquid flow using all the available tools: theory, experiments, and numerical simulation.

Project Home

| AEM Home | Institute of Technology |

Solid-Liquid Flows, Summary

Last updated October 16, 2000