An Open Source CFD-DEM Perspective

Christoph GONIVA1, Christoph KLOSS1, Alice Hager1 and Stefan PIRKER1,2

1

Christian Doppler Laboratory on Particulate Flow Modelling

2

Institute of Fluid Mechanics and Heat Transfer

both Johannes Kepler University, Altenbergerstr. 69, 4040 Linz, Austria

ABSTRACT We report on the synthesis of the Discrete Element Method (DEM) modelling the behaviour of granular materials and a finite volume model for the continuous interstitial fluid using Computational Fluid Dynamics (CFD). As real industrial problems are often characterized by high computational effort, the need for a CFD-DEM solver able to run on large clusters is obvious. To take a first step into this direction, the authors developed a coupled solver joining the two software packages LIGGGHTS and OpenFOAM. While the former covers the DEM modelling, the latter describes the fluid dynamics of the continuous phase. The implementation of the code permits fully parallel simulations on distributed memory machines using MPI functionality. In a test case the correct prediction of pressure drop and minimum fluidization velocity is shown. NOMENCLATURE
Cd d Rsl Ff g p u up Rep xp drag coefficient diameter force density the particles exert on the fluid sum of all forces the fluid exerts on a single particle gravity constant pressure velocity relative particle velocity at contact point particle Reynolds number particle overlap at contact point f n p t fluid normal to contact point particle tangential to contact point

Greek letters: volume fraction dynamic viscosity Coulomb friction coefficient density stress tensor

Subscript indices:

INTRODUCTION Two main modelling strategies for particle flow can be applied. The continuum approach considers the multitude of particles as an artificial continuum and is based on the solution of the underlying conservation equations using CFD techniques. Of course, such an approach disregards the local behaviour of individual particles. The second modelling strategy does not rely upon continuum mechanics. It rather simulates the motion of each particle individually, with a special treatment for eventual collisions. The most important discrete model is the Discrete Element Method (DEM) and its derivates. Lately, the DEM comes more and more into the focus of engineers and researchers. Being discrete in nature, it is, in principal, capable of capturing all granular physical phenomena. In industrial application, single phase 'dry' granular flow rarely occurs. In the vast majority of natural or industrial processes concerning granular materials, a secondary fluid phase, such as air, is present and its effects like fluidization (aeration of particles by gas injection) play an important role. In some cases, such as pneumatic conveying, the fluid phase controls particle movement, but particle-particle interactions may still be an important issue that cannot be neglected, whereas in other application, such as hopper discharge, the particle flow induces fluid flow. A promising approach to model such coupled granular-fluid systems is a coupled CFDDEM approach. Excellent reviews on DEM and the CFD-DEM technique have been published by Yu et al. (2007 and 2008b). In Kloss et al. (2009), the authors presented a similar coupling approach for the commercial software packages EDEM and FLUENT along with experimental validation. They could show the feasibility of a modelling approach comprising the effect of (a) volume displacement by the particles, (b) drag of the fluid on the particles as well as (c) Magnus force due to particle rotation. With that approach, a wide range of coupled particle-fluid systems can be handled. Yet, due to shortcomings of the commercial software packages this approach is limited to shared memory machines. In this paper we will describe a CFD-DEM approach for the Open Source software packages OpenFOAM and FLUENT. We then present and discuss a test case before drawing final conclusions in the last section.

MODEL DESCRIPTION Discrete Element Method (DEM) The Discrete Element Method was introduced by Cundall and Strack (1979). In the frame of the DEM, all particles in the computational domain are tracked in a Lagrangian way, explicitly solving each particles trajectory. A very brief description of the method will be provided in this section. Further details on the contact physics and implementation issues are available in the literature (e.g. Campbell, 1990; Zhou et al., 1999; Mattutis et al., 2000; Bertrand et al., 2005). Each physical particle is mathematically represented by a sphere, another geometrically well-defined volume or a combination of them. The translational and angular accelerations of a sphere are based on the corresponding momentum balances. Generally, the particles are allowed to overlap slightly. The normal force tending to repulse the particles can then be deduced from this spatial overlap xp and the normal relative velocity at the contact point, un.

Figure 1: Simple spring-dashpot model The simplest example is a linear spring-dashpot model, shown in Fig. 1. The normal force is given by
F
n

= k n

x
p

+ cn

p ,n

(1)

The magnitude of the tangential contact force can be written as:

F
t t u = min k t t c ,0 p ,t

dt + c t

p ,t

, c

F , n

(2)

where Ft is the tangential force and up,t is the relative tangential velocity of the particles in contact. The integral term represents an incremental spring that stores energy from the relative tangential motion, representing the elastic tangential deformation of the particle surfaces that happened since the time when particles touched at t = tc,0. The second part, the dashpot, accounts for the energy dissipation of the tangential contact. The magnitude of the tangential force is limited by the Coulomb frictional limit, where the particles begin to slide over each other. Subsequently, the total force acting on a particle can then be expressed as:
F
t o t

F
n

F
t

F
f

F
b

(3)

Here, Ff is the force that the fluid phase exerts on the particles, which will be described in further detail later. Other body forces like gravity, electrostatic or magnetic forces are subsumed into Fb. Similar balances are necessary for the particles angular momentum which are not stated here for the sake of shortness. The strength of the DEM lies in its ability to resolve the granular medium at the particle scale, thus allowing realistic contact force chains and giving rise to phenomena induced by

particle geometry combined with relative particle motion, such as particle segregation by percolation. Thereby, it is able to capture many phenomena, describe dense and dilute particulate regimes, rapid flow as well as slow flow and equilibrium states or wave propagation within the granular material. Thanks to advancing computational power, the DEM has become more and more accessible lately. On actual desktop computers, simulations of up to a million particles can be performed. On very large clusters, the trajectories of hundreds of millions of particles can be computed. The DEM Solver LIGGGHTS LIGGGHTS is an Open Source software for modelling granular material by means of the Discrete Element Method. LIGGGHTS stands for LAMMPS Improved for General Granular and Granular Heat Transfer Simulations and is based on LAMMPS (Large Atomic and Molecular Massively Parallel Simulator), a successful open source Molecular Dynamics code by Sandia National Laboratories for massively parallel computing on distributed memory machines. LAMMPS is a classical molecular dynamics solver and provides potentials for soft materials (bio-molecules, polymers), solid-state materials (metals, semiconductors) and coarse-grained granular materials (see Plimpton, 1995). It can be used to model atoms or, more generically, as a parallel particle simulator at the atomic, meso, or continuum scale. If coarse-grained granular particles are simulated, this method is termed DEM as in the previous section. LAMMPS offers implementations for both linear (Hooke) and non-linear (Hertz) granular potentials. It also provides efficient algorithms for detecting and calculating the pair-wise interaction forces. This is crucial as the largest part of the runtime of a DEM algorithm is spent on these issues. LAMMPS is also starting GPU support for further simulation speed-up. Detailed descriptions can be found in the LAMMPS manual (Sandia, 2009) LIGGGHTS brings these features for granular simulations to a new level, implementing the following features on top of what is possible with LAMMPS: A re-write of the granular pair- and wall contact laws, including a macroscopic cohesion model Wall import of CAD, including stress analysis A moving mesh feature A 6 degree of freedom solver for bodies represented by a surface mesh A heat conduction model for spherical particles Further features (e.g. non-spherical particle handling) are under development. The code can be downloaded at www.cfdem.com. Both LIGGGHTS and LAMMPS run on single processors or in parallel using messagepassing techniques and a spatial-decomposition of the simulation domain. The code is designed to be easy to modify or extend with new functionality. Both LIGGGHTS and LAMMPS are distributed as open source codes under the terms of the GPL.

CFD Approach The motion of an incompressible fluid phase in the presence of a secondary particulate phase is governed by a modified set of Navier-Stokes-Equations, which can be written as:
u f + ( f t
f

( f

) = 0,
u
f f

(4)
u
f

) + (

) =

sl

(5)

Here, f is the volume fraction occupied by the fluid, f is its density, uf its velocity, and is the stress tensor for the fluid phase. Rsl represents the momentum exchange with the particulate phase. For each cell, Rsl is calculated following the Gidaspow model (described in the following section). CFD-DEM interaction: The solver used for coupling granular flow with fluid flow is an incompressible transient solver using a PISO loop for pressure velocity coupling. To implement the CFD-DEM coupling, the DEM solver (LIGGGHTS) and an OpenFOAM CFD solver are being run consecutively, each halting calculation after a predefined number of time-steps for the purpose of data exchange. Inside the time loop, the DEM solver is periodically called to calculate the particles positions and velocities. The coupling routine consists of several steps: (1) The DEM solver calculates the particles positions and velocities. (2) For each particle, the corresponding cell in the CFD mesh is determined. (3) For each cell, the volume fraction as well as a mean particle velocity is determined. (4) Based on this information, the momentum exchange terms between the gas phase and the particulate phase can be evaluated (see below). (5) Using the momentum exchange terms the fluid flow is calculated. (6) The fluid forces acting on the particles are calculated and sent to the DEM solver (7) The routine is repeated from (1) The most important contribution to particle-fluid momentum exchange is established by means of a drag force depending on the granular volume fraction. In contrast to the drag force acting on a single sphere, here the granular volume fraction has to be considered for the drag force calculation. For numerical reasons the momentum exchange term is split-up into an implicit and an explicit term:
R
sl

sl

u
f

u
p

).

(6)

K For the calculation of s l we make use of a drag model by Gidaspow (1992) which is a combination of the Wen and Yu (1966) model and the Ergun (1952) equation.

For f > 0.8 we use:

sl

3 f (1 f ) = Cd 4 dp

u
f

u
p

f 2.65
p

(7) (8) (9)

Cd =

f
=

24 R e

[1 + 0.15(
p

R e
f

) ]
0.687

R e
p

u
f

dp

For f 0.8 we use the Ergun equation: u u K (1 f )2 f + 1.75 (1 f ) f s l = 150 dp f d p2

p

(10)

Further literature on similar approaches can be found in Yu et al. (2008a), Yu et al. (2008b), Tsuji et al. (2008) and Kafui et al. (2002). Beside the drag force resulting from a relative velocity between the particle and the fluid, other forces, neglected in this work, may be relevant too. These may stem from the pressure gradient in the flow field (pressure force), from particle rotation (Magnus force), particle acceleration (virtual mass force) or a fluid velocity gradient leading to shear (Saffman force). As a first guess, the fluid forces acting on each particle Ff can be evaluated in a similar way as K described above for the calculation of s l (Enwald 1996). The time-step size for the CFD solver is chosen in a way that it is an integer multiple of the DEM time-step. This is in order to save computational time, i.e. the momentum exchange terms are only updated after a number of DEM steps (e.g. every 10th DEM time step). Chimera approach: One specific problem of a CFD-DEM coupling lies in the different size scales of the particles. Especially if the particles are not smaller than the mesh that is used for the CFD difficulties occur. To overcome this problem a chimera approach was chosen where the volume fraction as well as the momentum exchange field is calculated on a mesh different to that mesh for the CFD calculations.

Figure 2: CFD mesh (left), momentum exchange mesh (middle), DEM domain (right)

APPLICATION EXAMPLE CFD-DEM Coupling Validation The authors have already managed to implement a CFD-DEM coupling approach using the commercial software packages FLUENT and EDEM. This approach has been presented and discussed in Kloss et al. (2009) by means of three validation examples: A pneumatic conveying channel, a particle charging experiment, and the discharge of a hopper with enclosed standpipe.

In this paper, we want to focus on a fluidized bed application example calculated using our Open Source CFD-DEM coupling that is outlined in the previous section. The focus is on the correct prediction of pressure drop and onset of fluidization.
Fluidized Bed In order to show the feasibility of the CFD-DEM solver a simple test case is presented here. The granular and fluid flow of a small fluidized bed is calculated using the CFD-DEM approach described in this paper. The geometry is a cylinder of radius 0.0138 and height 0.553 m. On the bottom of the cylinder there is a uniform, fixed value velocity inlet. Inlet velocities range from 0 m/s to 0.02 m/s. The upper side of the cylinder is a fixed pressure outlet. At the side walls a slip condition is applied. For the granular flow all walls are solid walls. The particles are ideal spheres with a diameter of 0.001 m and a density of 200 kg/m3. Gravitational acceleration of 9.81 m/s2 is acting on the particles in negative z-direction. The kinematic viscosity of the interstitial fluid is 1.5*10-4 m2/s. The flow field is initialized with zero velocity and 10000 particles are packed on the lower side of the cylinder. The computational time for one second simulation time is approximately one hour on 2 2.4 GHz processors. Geometry and boundary conditions are shown in Fig. 3.

Figure 3: Geometry and velocity boundary condition (left), initial particle positions (right).

In Fig. 4, the particle positions at different inlet velocities are shown. At gas velocities smaller than the minimum fluidization velocity of ~0.01 m/s the bed is fixed and no particle motion occurs, see Fig. 4 (1). For gas velocities slightly higher than the minimum fluidization velocity the particles start moving, the voidfraction inside the bed increases and the pressure drop decreases, see Fig. 4 (2). If the gas flow is further increased, particles start moving Fig. 4 (3-4).

Figure 4: Simulation result for the fluidized bed test case. Particle positions, inlet velocity

The pressure drop p for a fixed bed of the height h can be calculated using the Ergun equation: u u (1 f )2 f f + 1.75 (1 f ) f f 2 . (11) p = 150 h f 3 pd p2 f 3 pd p u Where p denotes the sphericity of the particles, f the superficial velocity, d p the particle diameter and f the void fraction in the bed. If the upward force in terms of pressure drop is balanced with the gravitational force acting on the particles we can derive a relation for the state of incipient fluidization. For u f d p Re = < 20 we get the relation for the minimum fluidization velocity (Kunii 1992)
u
m f

d p ( p f )g f p
2 3

150 f

(1 ) .
f

(12)

In Fig. 5 and 6 the results of a simulation where inlet velocity is increased stepwise are shown. On the right hand side of Fig. 5 we can see that below the minimum fluidisation velocity of about 0.012 m/s the void fraction of the bed remains constant. If the inlet velocity is further increased, the particles start moving and the voidfraction increases.

Figure 5: inlet velocity (left), voidfraction o the bed (right).

In Fig. 6 the results of equation 11 compared to simulation results. For the pressure drop is in excellent accordance to the Ergun equation.

u
f

<

m f

= 0.0121 m / s

Figure 6: pressure drop at different inlet velocities, simulation vs. analytics. CONCLUSION AND OUTLOOK Coupled CFD-DEM simulations have to cover a huge variety of regimes. The physics must be described correctly irrespective of whether the fluid flow is induced by particle motion with hardly any feedback to the particles (e.g. hopper discharge), or the particle motion is controlled by the gas flow (e.g. in the case of pneumatic conveying). Even for the case of a fixed bed, the situation may occur that some particles are dispersed in the gas as it is shown in the test case above. In this work we could show that the synthesis of DEM and the CFD method leads to a very versatile tool capable to deal with fixed bed as well as incipient fluidization. The long term goal is to develop a CFD-DEM solver being robust and efficient enough to handle industrial granular flow applications. The ability to use the presented solver on distributed memory clusters will allow to handle flows with several millions of particles and therefore it will be capable for a wide range of industrial applications. At the present stage the solver is designed to show the feasibility of the coupling in terms of programming. Improvement of the physical modelling which is applied is planned for future work.

ACKNOWLEDGEMENTS This study was partly funded by the Christian Doppler Gesellschaft (www.cdg.at) of the Austrian government. REFERENCES
BERTRAND, F., LECLAIRE, L.-A. and LEVECQUE, G. (2005): DEM-based models for the mixing of granular materials, Chemical Engineering Science, 60, 2517 2531 CAMPBELL, C. S. (1990): Rapid Granular Flows, Annual Rev. Fluid Mech., 22, 57-92. CUNDALL, P.A. and STRACK, O.D. (1979): A discrete numerical model for granular assemblies. Geotechnique, 21, 47-65 ENWALD, H., PEIRANO, E., ALMSTEDT, E. (1996): "Eulerian Two-Phase Flow Theory Applied to Fluidization." Int. J. of Multiphase Flow, 22, 21-66 ERGUN, S. (1952): "Fluid Flow through Packed Columns." Chem. Eng. Prog., 48(2):89 GIDASPOW, D., BEZBURUAH, R., DING, J. (1992): "Hydrodynamics of Circulating Fluidized Beds, Kinetic Theory Approach", in Fluidization VII, Proceedings of the 7th Engineering Foundation Conference on Fluidization KAFUI, K.D., THORNTON, C., ADAMS, M. J. (2002): Discrete particle-continuum modelling of gassolid fluidised beds, Chemical Engineering Science, 57, 2395 2410 KLOSS, C., GONIVA, C., AICHINGER, G. and PIRKER, S. (2009): "Comprehensive DEM-DPM-CFD simulations: Model synthesis, experimental validation and scalability", Proceedings Seventh International Conference on CFD in the Minerals and Process Industries, CSIRO, Melbourne, Australia, December 9-11, (submitted for publication) KUNII, D., LEVENSPIEL, O. (1992): Fluidization Engineering, 2nd ed., Butterworth-Heinemann series in chemical engineering ISBN 0-409-90233-0 MATUTTIS, H.G., LUDING S. and HERRMANN, H.J. (2000): Discrete element simulations of dense packings and heaps made of spherical and non-spherical particles, Powder Technology, 109, 278292 PLIMPTON, S. J. (1995), "Fast Parallel Algorithms for Short-Range Molecular Dynamics", J. Comp. Phys., 117, 1-19 (LAMMPS homepage: http://lammps.sandia.gov). SANDIA (2009): LAMMPS User Manual, http://lammps.sandia.gov/doc/Manual.html, Sandia National Laboratories, USA WEN, C.-Y. and YU, Y. H. (1966), Mechanics of Fluidization, Chem. Eng. Prog. Symp.Series, 62:100 TSUJI, T., YABUMOTO, K. and TANAKA, T. (2008): Spontaneous structures in three-dimensional bubbling gas-fluidized bed by parallel DEMCFD coupling simulation, Powder Technology, 184, 132140 YU, A.B., ZHOU, Z.Y, ZHU, H.P., and YANG, R.Y. (2007): Discrete particle simulation of particulate systems: Theoretical Developments, Chemical Engineering Science, 62, 3378-3396 YU, A.B., ZHOU, Z.Y, ZHU, H.P., and YANG, R.Y. (2008b): Discrete particle simulation of particulate systems: A review of major applications and findings, Chemical Engineering Science, 63, 5728-5770 YU, A.B., WRIGHT, B., ZHOU, Z.Y, ZHU, H.P., and ZULLI, P. (2008a): Discrete particle simulation of gas-solids flow in a blast-furnace, Computers & Chemical Engineering, 32, 1760-1772 ZHOU, Y.C., WRIGHT, B.D., YANG, R.Y., XU, B.H., YU, A.B. (1999): Rolling friction in the dynamic simulation of sandpile formation, Physica A, 269, 536-553

10