THEORETICAL & APPLIED MECHANICS LETTERS 2, 012002 (2012)

Coupling of discrete-element method and smoothed particle

hydrodynamics for liquid-solid flows
Yrjö Jun Huanga) and Ole Jørgen Nydal
Department of Energy and Process Engineering, Norwegian University of Science and Technology (NTNU),
Trondheim, Norway
(Received 10 September 2011; accepted 27 September 2011; published online 10 January 2012)
Abstract Particle based methods can be used for both the simulations of solid and fluid phases in
multiphase medium, such as the discrete-element method for solid phase and the smoothed particle
hydrodynamics for fluid phase. This paper presents a computational method combining these two
methods for solid-liquid medium. The two phases are coupled by using an improved model from
a reported Lagrangian-Eulerian method. The technique is verified by simulating liquid-solid flows
in a two-dimensional lid-driven cavity. c 2012 The Chinese Society of Theoretical and Applied
Mechanics. [doi:10.1063/2.1201202]
Keywords discrete-element method, smoothed particle hydrodynamics, liquid-solid flows, lid-driven
cavity

Mesh generation is a time-consuming task in com- time steps to find the particle’s velocity and position.
putational simulation. Mesh free methods abandon this Comparing with the Eutherian methods for particulate
concept. Several meshless methods have been devel- flows, we can easily capture the key features and details
oped for fluid mechanics since 1970s. Among all these of the system to provide details which are helpful in the
methods, smoothed particle hydrodynamics (SPH) is development of accurate models.7,8 Hence, the DEM al-
one of the earliest methods.It was developed for solv- gorithm looks similar to SPH because both of them are
ing astro-physical problems.1,2 Later it was also used numerically intergraded subjected to forces applied by
for the simulations of different types of fluid flows, be- the neighboring particles. The major difference is that
cause the movement of particles in astrophysical flows the inter-particle forces in SPH are derived from the
is similar to those in liquid or gas flows. The idea of the Navier-Stokes equations and those forces in DEM are
SPH method is building a set of disordered particles in due to particle contact process.9
a continuum without grid or mesh. Each particle has A lot of researches using DEM to couple with
an associated mass, momentum and energy. The mo- mesh-based computational fluid dynamics (CFD) tech-
tion of the particle is calculated from the integration nique, namely Lagrangian (particle-based) and Eule-
of Newton’s second law. A property for each particle rian (mesh-based) methods (La/Eu), to simulate the
within the flow, such as density, is obtained from an liquid-solid flows have been carried out.10,11 The vol-
interpolation of the neighboring particles by using the ume fraction, φ, is an important parameter for coupling
following approximate averaging operator, which is also the phases in these simulations, because the drag force
called kernel function3,4 is described as a function of φ.10,12 These researches
 cannot get rid of the mesh generation process in the
f (xi ) = f (x )δ(x − x )dx ≈ simulations thoroughly. Ten years ago, Potapov et al.13
Ω
showed some simulations of coupling of the SPH and

N
DEM methods, which is different from the other pre-
f (xj )W (xi − xj , h)Vj , (1) vious approaches for liquid-solid flows. The size of the
j=1
solid beads in the system are much larger than the inter-
where δ is the Dirac delta function, W is the smoothing particle spacing for the fluid SPH particles in their sim-
kernel function and h is the smoothing length defined ulations. No-slip boundary conditions were applied to
the influence area of W , Vj is the space occupied by couple the fluid particles close to the solid surface and
the jth particle, Ω is the space of the integral which those solid particles composed the beads surface. Re-
contains x, also called support domain and N is the cently, Huang et al.14 showed their SPH simulations us-
number of discrete particles in Ω. ing multi-mass particles. This technique makes it easy
Also in 1970s, Cundall developed discrete-element to combine the SPH and DEM methods for the solid-
method (DEM) for the analysis of rock-mechanics fluid problems with tiny beads.
problems5 and then this method was applied to soils.6 This paper presents some results of an ongoing
Same as that in SPH, the motion of each particle in study of solid-liquid flow with tiny solid beads by com-
DEM is also described by the Newton’s law, that the bining SPH and DEM. It is a Lagrangian-Lagrangian
sum of all forces acting on it is intergraded over tiny (La/La) method and no mesh is required more.
Each particle carries individual mass and occupies
individual space in SPH method for single phase. The
a) Corresponding author. Email: jun.huang@ntnu.no. density of the ith particle, ρi , in a particulate sys-
012002-2 Y. J. Huang, and O. J. Nydal Theor. Appl. Mech. Lett. 2, 012002 (2012)

tem can be evaluated by using the summation density

approach4,15
N 
N  mj 
ρi = mj Wi,j Wi,j , (2)
j=1 j=1
ρj

where ρj is the density and mj is the mass of the jth

particle of the N fluid particles supporting domain. It is
well suited for the normalization condition and strictly
ensures the conservation of mass. We have adopted this
approach in this paper.
Based on the summation density approach, the
mass conservation equation and the momentum equa-
tion also can be written in particle form as16,17 Fig. 1. Location of the particles in the lid-driven square
cavity at t = 0 s. The big dots for solid beads and the
dρxi 1 N
mj vij ∂Wij
small dots are for fluid particles in the system. The initial
= ρi , (3) velocities of all particles are zero. The origin of the xy-
dt j=1
ρj ∂xxi 2 coordinate system is located in the lower-left corner of the
cavity.
and

dvix1  mj σ j
N x1 ,x2
∂Wij force, | Ft,i |, is obtained from Coulomb’s friction law
ρi = , (4)
dt ρj ∂xxi 2 that | Ft,i |= μ | Fn,i |. In the simulations shown in this
j=1
paper, e = 0.9 and μ = 0 were selected.
where v is velocity, t is time, σ is stress tensor and The drag force term in Eq. (6) presents the force on
x is coordinate in the kernel function. The stress ten- solid beads due to the velocity relative to the surround-
sor is made up of two parts, the isotropic pressure and ing fluids. For a single spherical bead, it can be written
the viscous stress. If external forces are considered, the as12
momentum equation Eq. (4) becomes FsD = 3πds ρf νf φβf (1 + 0.15Re0.678 )(vf − vs ), (8)
p
 
dvix1
N
mj σ xj 1 ,x2 ∂Wij Fext where Rep is the particle Reynolds number, Rep =
ρi = + . (5) φf ds |vf − vs |/νf , ds is the diameter of the solid bead,
dt j=1
ρj ∂xxi 2 Vi
ρf is the density, νf is the viscosity of the fluids, the ex-
The external forces include the gravitation and the ponent β varies with the particle Reynolds number, the
solid-fluid interaction. subscripts s and f are for solid and fluid particles respec-
The motion of each solid bead in a solid-liquid sys- tively. In our simulations, the density ρf is obtained by
tem can be described by the Langevin equation, as from Eq. (2) and the velocity of the surrounding fluids
for the solid particle i, vf,i is approached by averaging
dv the surrounding fluid particles
m = F i + F D + F b + F g. (6)
dt
N  N
where the four terms on the right hand side are due vi,f = vj mj mj (9)
to bead-bead contact, drag force and buoyancy force j=1 j=1
from the fluid phase, and gravity force. To simplify The buoyancy force term is given by
the problems, the shapes of the beads are assumed to
be spherical or spheroidal and each particle stands for Fsb = (ρf − ρs )Vs · k, (10)
a solid bead. A DEM algorithm given by Ramı́rez et where k is the unit vector paralleling with gravitation.
al.18 was chosen in our simulations, that the bead-bead The aforementioned drag force and the buoyancy
contact process could be modeled as a spring-dashpot force act on the solid beads, then the equal and op-
system. The normal interaction force is the sum of an posite forces should be acted upon the fluid particles.
elastic term and a viscous term Employing the kernel function, we approach these reac-
Fn = Fne + Fnd = Kn δnζ + ηn δnξ δ̇n , (7) tions on each fluid particle by
 N
where ζ = 3/2,19,20 , ξ = 0.5 by dimensionless D,b
Ff,k = Fs · (Vf,k Wk,j )
D,b
Vf,j Wi,j . (11)
analysis,21 δn is the overlap, δn = ri + rj − dij (δn > 0), j=1
Kn is the effective stiffness, ηn is the damping coeffi-
cient, ri,j is the radius of each bead and dij is the dis- Adding the fluid-solid interaction to the momentum
tance between the two centers of the beads. While for equation for fluid phase, we find that Eq. (5) becomes
a constant coefficient of restitution (COR) in the nor- x1
dvf,i N
mf,j σ xj 1 ,x2 ∂Wij Ff,i
b
+ Ff,i
D
mal direction, e, ζ the details of other coefficients are ρf,i = + . (12)
described in Refs. 7, 22. The magnitude of tangential dt j=1
ρf,j ∂xxi 2 Vf,i
012002-3 Coupling of discrete-element method Theor. Appl. Mech. Lett. 2, 012002 (2012)

Fig. 2. Location and velocity profiles of the lid-driven square cavity at different times. Here, the solid dots are for the solid
beads in the system.

Fig. 3. Dimensionless velocity along the vertical central line

of the cavity. The different types of dots are for the solid-
fluid flows at different time. The curve is for a single-phase Fig. 4. The flow field of fixed and concentrated dense solid
fluid flow with the same boundary conditions when it arrives beads, which can be used for simulation of porous materials.
steady state.

density of the fluid phase is 1 000 kg/m3 , and the vis-

The lid-driven cavity flow is perceived as a text- cosity νf = 1 000 m2 /s. Gravitation is not considered
book example and usually it is determined qualitatively in the simulations. The square simulational domain is
in the simplest case of the flow in a two dimensional sketched in Fig. 1 with side length L = 0.01 m and the
(2D) square cavity. The comparison with this classical velocity of the top boundary, vt , is 0.01 m/s. The mass
phenomena has been used to examine the ability and of each fluid particle is obtained by
efficiency of a simulation method.14,23–26 A simulation
of solid-liquid flow in a 2D lid-driven square cavity is 
N
mf,i = ρf,i dxdy · Vs,j Wi,j /Vh , (13)
presented. The simulation region consists of 40 × 40
j=1
fluid particles and 20 × 4 solid beads, with shear mov-
ing boundary conditions at the top side and bounce- where N is the number of solid particles in the ith par-
back boundary conditions at the left, right and bottom ticle’s supporting domain and Vh is space of the sup-
sides. The diameter of these solid beads is 1 mm, the porting domain. For 2D problems, Vh = 3πh2 /4, where
012002-4 Y. J. Huang, and O. J. Nydal Theor. Appl. Mech. Lett. 2, 012002 (2012)

h is the radius of the supporting domain. It is reported by an example of solid-fluid flow in a square lid-driven
that a prime eddy forms at 0.24L below the lid in the cavity.
mid-plane for water flow, when the Reynolds number
is 100, where the Reynolds number Re = Lvt /ν.27 Be- The authors thank Prof. Liu Moubin of the Key
cause gravitation is not considered in this simulation, Lab for Hydrodynamics and Ocean Engineering, Institute
the buoyancy force also approaches zero. of Mechanics, Chinese Academy of Sciences for using their
Figure 2 shows the transient behavior of the parti- Codes.4 . This work was partially supported by Department
cles in the cavity. The snapshots are taken at six dif- of Energy and Process Engineering, Norwegian University
ferent times: t = 0.05 s, 0.25 s, 0.5 s, 1.0 s, 5.0 s and of Science and Technology, Institute for Energy Technology
and SINTEF through the FACE (Multiphase Flow Assurance
10.0 s. As shown in Fig. 2(a), the motion of the solid
Innovation Center) Project.
beads affects the flow field in the beginning. Because
the motions of those solid beads are caused by the flow
of the fluid phase, there exists a delay between the mo- 1. L. B. Lucy, Astronomical J. 82, 1013 (1977).
tions of these two phases. The quasi-static solid beads 2. R. A. Gingold, and J. J. Monaghan, Monthly Notices of Royal
separate the cavity into two chambers and the fluid in Astro. Soc. 181, 375 (1977).
3. S. Li, and W. K. Liu, Appl. Mech. Rev. 55, 1 (2002).
each chamber forms an eddy. The flow field reaches a 4. G. R. Liu, and M. B. Liu, Smoothed particle hydrodynamics,
steady state gradually and the relative velocity between a meshfree particle method (World Scientific 2003).
the two phases approaches to zero. With the movement 5. P. A. Cundall, Proceeding of the Symposium of International
Society of Rock Mechanics (Nancy, Fance, 1971).
of the solid beads, the two chambers connect to each 6. P. A. Cundall, and O. D. L. Strack, Geotechnique 29, 47
other, and these two eddies combine to a prime one. (1979).
Figures 2(a)–2(c) present this process. Each particle 7. J. Huang, C. K. Chan, and P. Zamankhan, Phys. Rev. E 82,
in the system is propagating along a closed streamline. 031307 (2010).
8. Y. J. Huang, and O. J. Nydal, Communications in Computa-
To compare the velocity profiles, another SPH simula- tional Physics 10, 1027 (2011).
tion for single-phase liquid flow in the same cavity was 9. P. W. Cleary, and M. Prakash, Phil. Trans. R. Roc. Lond. A
carried out. All the dimensionless velocities along the 362, 2003 (2004).
10. Y. Tsuji, T. Kawaguchi, and T. Tanaka, Power Technology
vertical central line (x = L/2) at different times are 77, 78 (1993).
plotted in Fig. 3, as well as the single-phase liquid flow 11. P. Zamankhan, J. Huang, and S. M. Mousavi, J. Offshore &
at steady state. The dimensionless velocities and coor- Arctic Eng. 129, 176 (2007).
12. C. Y. Wen, and Y. H. Yu, Chem. Eng. Proc. Symp. Ser. 62,
dinate are defined as v ∗x = vx /vt and y ∗ = y/L, respec-
100 (1966).
tively. From Fig. 3, it also can be seen that the eddy 13. A. V. Potapov, M. L. Hunt, and C. S. Campbell, Powder Tech.
center of the solid-liquid flow approaches that of the 116, 204 (2001).
single-phase liquid flow after 0.5 s in our simulations. 14. Y. J. Huang, and O. J. Nydal, Particle refined and
coarsened method for SPH simulations in fluid dynamics,
On the other hand, if the mass of the solid particles http://www.youtube.com/watch?v:ZH2Rwrmrko.
were set as great enough, the two chambers shown in 15. P. W. Randles, and L. D. Libersky, Compu. Methods in Appl.
Fig. 2(a) become stable. The convection between the Mech. & Eng. 139, 375 (1996).
16. L. D. Libersky, A. G. Petscheck, and T. C. Carney, et al, J.
chambers becomes weak. The performance of the dense Compu. Phys. 109 67 (1993).
solid phase concentrated in the middle of the simulation 17. C. E. Zhou, G. R. Liu, and K. Y. Lou, Int. J. Compu. Methods
domain is similar to porous materials. Figure 4 shows 4 671 (2007).
the flow field of t = 0.1 s when the density of the solid 18. R. Ramirez, T. Poschel, and N. V. Brilliantov, et al, Phys.
Rev. E. 60, 4465 (1999).
phase is set as 108 kg/m3 . All the aforementioned sim- 19. H. Hertz, J. Reine Angew. Math. 92, 156 (1882).
ulation results and theoretical explanations show the 20. K. L. Johnson, Contact Mechanics (Cambridge Univ. Cam-
validation of the SPH-DEM coupling technique. bridge 1982).
21. G. Kuwabara, and K. Kono, Jpn. J. Appl. Phys. 26, 1230
This paper presents a La/La algorithm to simulate (1987).
solid-fluid phase flows. The motion of the solid phase is 22. R. D. Mindlin, J. Appl. Mech. 16, 259 (1949).
simulated by using DEM while the fluid aspect is simu- 23. S. Hou, Q. Zou, and S. Chen, et al, J. Compu. Phys. 118, 329
(1995).
lated by using SPH. The coupling of the two aspects are 24. L. Larchevêque, P. Sagaut, and I. Mary, et al, Phys. Fluids
followed from some solid-fluid interaction models for La- 15, 193 (2003).
Eu methods. Employing this method, we find that the 25. T. Watanabe, and K. Ebihara, Nuclear Eng. & Design 188,
sizes of the solid beads are very flexible, namely from 111 (1999).
26. P. N. Shankar, and M. D. Deshpande, Ann. Rev. Fluid Mech.
infinite to the same scale as the smooth length in the 32, 93 (2000).
SPH aspect, which is helpful to control and balance the 27. U. Ghia, K. N. Ghia, and C. T. Shin, J. Compu. Phys. 48,
CPU time in the two aspects. The method is verified 387 (1982).