You are on page 1of 59

DEM-CFD COUPLING

TECHNICAL MANUAL
dem-cfd coupling manual i
ROCKY

Copyright

copyright ©2019, esss. all rights reserved.

No part of this documentation may be reproduced in any form, by any means,


without the prior written permission of ESSS.

ESSS makes no representations or warranties with respect to the program material


and specifically disclaim any implied warranties, accuracy, merchantability or fitness
for any particular purpose. Furthermore, ESSS reserves the right to revise the
program material and to make changes therein without obligation to notify purchaser
of any revisions or changes except specific errors determined to be incorporated in the
program material. It shall be the responsibility of ESSS to correct any such errors in
an expeditious manner. In no event shall ESSS be liable for any incidental, indirect,
special, or consequential damages arising out of the purchaser’s use of program
material.
ANSYS, ANSYS Workbench, Fluent, and any and all ANSYS, Inc. brand,
product, service and feature names, logos and slogans are registered trademarks or
trademarks of ANSYS, Inc. or its subsidiaries in the United States or other countries.

© 2019, esss - all rights reserved


dem-cfd coupling manual ii
ROCKY

Contents

1 Introduction 1

1.1 Granular-fluid systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Why is it so complicated to model these systems? . . . . . . . . . . . . . 1

1.3 Why not just CFD? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2


1.3.1 The Eulerian approach . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.2 The Lagrangian approach . . . . . . . . . . . . . . . . . . . . . 3

1.4 Why couple DEM and CFD together? . . . . . . . . . . . . . . . . . . . . 3

1.5 How is this document organized? . . . . . . . . . . . . . . . . . . . . . . . 4

2 Governing equations 5

2.1 Particle phase modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5


2.1.1 Translational and rotational particle motion of a particle . . . 5
2.1.2 Energy balance in a particle . . . . . . . . . . . . . . . . . . . . 6

2.2 Fluid phase modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6


2.2.1 Mass and momentum conservation equations . . . . . . . . . 6
2.2.2 Energy conservation equation . . . . . . . . . . . . . . . . . . . 7

3 Particle-fluid interaction forces 8

3.1 Pressure gradient force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.2 Drag force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8


3.2.1 Single particle drag laws . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1.1 Schiller & Naumann (1933) . . . . . . . . . . . . . . 9
3.2.1.2 DallaValle (1948) . . . . . . . . . . . . . . . . . . . . 10
3.2.1.3 Haider & Levenspiel (1989) . . . . . . . . . . . . . . 10
3.2.1.4 Ganser (1993) . . . . . . . . . . . . . . . . . . . . . . 12
3.2.2 Dense flow drag laws . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2.2.1 Wen & Yu (1966) . . . . . . . . . . . . . . . . . . . . 13
3.2.2.2 Ergun (1958) . . . . . . . . . . . . . . . . . . . . . . 13
3.2.2.3 Gidaspow, Bezburuah & Ding (1992) . . . . . . . . 14

© 2019, esss - all rights reserved


dem-cfd coupling manual iii
ROCKY

3.2.2.4 Huilin & Gidaspow (2003) . . . . . . . . . . . . . . 14


3.2.2.5 Di Felice (1994) . . . . . . . . . . . . . . . . . . . . . 14

3.3 Virtual mass force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15


3.3.1 Virtual mass models . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3.1.1 Constant value . . . . . . . . . . . . . . . . . . . . . 16
3.3.1.2 Ishii & Mishima (1984) . . . . . . . . . . . . . . . . 17
3.3.1.3 Paladino (2005) . . . . . . . . . . . . . . . . . . . . . 17

3.4 Lift force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18


3.4.1 Lift laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4.1.1 Saffman (1968) . . . . . . . . . . . . . . . . . . . . . 19
3.4.1.2 Mei (1992) . . . . . . . . . . . . . . . . . . . . . . . . 20

3.5 Fluid generated torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20


3.5.1 Torque laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.5.1.1 Dennis, Singh & Ingham (1980) . . . . . . . . . . . 21

3.6 Turbulent dispersion force . . . . . . . . . . . . . . . . . . . . . . . . . . . 21


3.6.1 Turbulent dispersion model . . . . . . . . . . . . . . . . . . . . 22
3.6.1.1 Eddy fluctuating velocity . . . . . . . . . . . . . . . 22
3.6.1.2 Eddy lifetime . . . . . . . . . . . . . . . . . . . . . . 23
3.6.1.3 Particle-eddy interaction time . . . . . . . . . . . . 23

4 Heat transfer 25

4.1 Heat transfer modes in the DEM-CFD coupling . . . . . . . . . . . . . . 25

4.2 Heat transfer between fluid and particle . . . . . . . . . . . . . . . . . . . 26


4.2.1 Fluid-particle heat transfer correlations . . . . . . . . . . . . . 27
4.2.1.1 Ranz & Marshall (1952) . . . . . . . . . . . . . . . . 27
4.2.1.2 Whitaker (1972) . . . . . . . . . . . . . . . . . . . . . 27
4.2.1.3 Gunn (1978) . . . . . . . . . . . . . . . . . . . . . . . 27
4.2.2 Comparison of heat transfer correlations . . . . . . . . . . . . 28

5 Computational details 30

5.1 Fluid phase calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.2 Source terms linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . 30


5.2.1 Momentum terms . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2.2 Heat transfer terms . . . . . . . . . . . . . . . . . . . . . . . . . 31

5.3 Lagrangian to Eulerian mapping . . . . . . . . . . . . . . . . . . . . . . . 32


5.3.1 Uniform distribution L-E mapping . . . . . . . . . . . . . . . . 33
5.3.1.1 Known issues . . . . . . . . . . . . . . . . . . . . . . 35
5.3.2 Volumetric diffusion L-E mapping . . . . . . . . . . . . . . . . 35

© 2019, esss - all rights reserved


dem-cfd coupling manual iv
ROCKY

5.3.2.1 Formulation . . . . . . . . . . . . . . . . . . . . . . . 36
5.3.2.2 Operation . . . . . . . . . . . . . . . . . . . . . . . . 37
5.3.2.3 Mapping other physical quantities . . . . . . . . . . 38
5.3.2.4 Selection of fluid cell zones . . . . . . . . . . . . . . 39
5.3.2.5 Mapping across sliding interfaces . . . . . . . . . . 39

5.4 Two-way coupling algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 41

6 Best practices 43

6.1 One-way coupling simulation . . . . . . . . . . . . . . . . . . . . . . . . . 43


6.1.1 Mesh generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.1.2 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . 43

6.2 Two-way coupling simulation . . . . . . . . . . . . . . . . . . . . . . . . . 46


6.2.1 General information . . . . . . . . . . . . . . . . . . . . . . . . 46
6.2.2 Mesh generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2.3 CFD simulation setup . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2.3.1 Multiphase model . . . . . . . . . . . . . . . . . . . 47
6.2.3.2 Boundary conditions and initialization . . . . . . . 48
6.2.3.3 Numerics . . . . . . . . . . . . . . . . . . . . . . . . 48
6.2.3.4 Solution data export . . . . . . . . . . . . . . . . . . 48
6.2.4 Rocky simulation setup . . . . . . . . . . . . . . . . . . . . . . . 48
6.2.4.1 Particle size scale factor . . . . . . . . . . . . . . . . 49

7 Bibliography 50

© 2019, esss - all rights reserved


dem-cfd coupling manual 1
ROCKY

1 Introduction
1.1 Granular-fluid systems
Many processes in various industries involve the simultaneous flow
of fluids and particles. Just to cite a few examples:

• Slurry mills (mining industry)


• Cyclones, desanders, and drill cutting removal (oil and gas
industry)
• Pneumatic conveyors (multiple industries)
• Wastewater management (waste disposal industry)
• Grains drying and sorting (agriculture and food industries)
• Biomass reactors (energy industry)
• Fluidized beds and catalytic reactors (chemical and nuclear
industries)

In all these cases it is important to take the fluid flow into account
in order to get the correct behavior of the particles. Design and
scale-up, as well as optimization of such processes, require a deep
understanding of the thermo-hydrodynamics, which is determined
by the particle-level interactions between the fluids, particles, and
boundaries. So it is clear that we need a modeling strategy to deal
with granular-fluid systems.

1.2 Why is it so complicated to model these systems?


The complexity of the fluid-solid flow present in these systems
makes modeling them a challenging task. The primary source of
difficulties is due to differences in order of magnitudes amongst the
characteristic scales existing in the problem.
First of all, there is the device scale, which is naturally respected.
Secondly, the typical fluid-flow scales are captured in a CFD solution
by solving the flow at the mesh scale, which is generally much
bigger than the particle but yet quite small when compared to the

© 2019, esss - all rights reserved


dem-cfd coupling manual 2
ROCKY

device scale. Finally, there is the fluid-particle interactions scale,


which has the magnitude of the smaller particles, therefore making
it computationally prohibitive to solve the flow in a sub-particle
resolution for most industrial applications. These difficulties are what
makes the coupled DEM-CFD approach so promising: it provides an
intermediate level between using the sub-particle resolution for the
fluid and the mesh resolution for both fluids and particles.

1.3 Why not just CFD?


There are two main approaches to deal with solids in CFD: The
Eulerian approach and the Lagrangian approach.

1.3.1 The Eulerian approach

In the Eulerian approach, both the fluids and solid phases are treated
as interpenetrating continua in a computational cell that is much
larger than the individual particles, but still small compared to the
size of the process scale. Therefore, continuum equations are solved
for both phases with an appropriate interaction term to model them.
This, in turn, means that constitutive equations for inter- and intra-
phase interaction are needed. Since the volume of a phase cannot be
occupied by the other phases, the concept of phasic volume fraction is
introduced. Location-based mapping techniques are applied and local
mean variables are used in order to obtain conservation equations
for each phase. The advantage of this approach is its reasonable
computational cost for practical application problems, making it the
most used granular-fluid modeling technique in use today.1 1
Gidaspow, D. (1994). Multiphase Flow
and Fluidization. Academic Press, San
The problem relies on the fact that finding general equations for Diego
granular systems is difficult due to the changing nature of how
solids flow. However, the capacity of the continuous approach to
produce accurate results is directly dependent on the constitutive
relations adopted for modeling interactions between the phases and
the rheology of the particulate material, which are quite difficult to
obtain.2 2
Xu, B. and Yu, A. (1997). Numerical
simulation of the gas-solid flow in a
Moreover, due to the continuum interpenetrating approach, no fluidized bed by combining discrete par-
individual particle information is available, and this might be the data ticle method with computational fluid
dynamics. Chemical Engineering Science,
sought. Not to mention that prescribing a particle size distribution 52:2785–2809
can increase your computational cost, since in general several phases
are created to model several particle sizes.

© 2019, esss - all rights reserved


dem-cfd coupling manual 3
ROCKY

1.3.2 The Lagrangian approach

In the Lagrangian approach, the fluid is still treated as continuum


by solving the Navier-Stokes equations, while the dispersed phase is
solved by tracking a large number of particles through the flow field.
Each particle (or group of particles) is individually tracked along the
fluid phase by the result of forces acting on them by numerically
integrating Newton’s equations that govern the translation and
rotation of the particles.3 3
Cundall, P. A. and Strack, O. D. L.
(1979). A discrete numerical model for
This approach is made considerably simpler when particle-particle granular assemblies. Geotechnique, 29:47–
interactions can be ignored. This requires that the dispersed second 65

phase occupies a low volume fraction, which is not the reality in


the majority of the industrial applications. Due to the fact that
no particle interaction is resolved, the model is inappropriate for
4
Tsuji, Y., Kawaguchi, T., and Tanaka, T.
(1993). Discrete particle simulation of
modeling applications where the volume fraction of the second phase two-dimensional fluidized bed. Powder
Technology, 77:79–87
cannot be ignored, such as fluidized beds. For applications such
5
Hoomans, B., Kuipers, J., Briels, W.,
as these, particle-particle interactions need to be taken into account
and Van Swaaij, W. (1996). Discrete
when solving the dispersed phase. particle simulation of bubble and slug
formation in a two-dimensional gas
fluidised bed: a hard-sphere approach.
Chemical Engineering Science, 51:99–118
1.4 Why couple DEM and CFD together? 6
Hoomans, B., Kuipers, J., and
Van Swaaij, W. (2000). Granular
dynamics simulation of segregation
The coupled DEM-CFD approach is a promising alternative for phenomena in bubbling gas-fluidised
modeling granular-fluid systems since it can capture the discrete beds. Powder Technology, 109:41–48

nature of the particle phase while maintaining the computational 7


Xu, B., Yu, A., Chew, S., and Zulli,
P. (2000). Numerical simulation of the
tractability. This is accomplished by solving the fluid field at the cell gas-solid flow in a bed with lateral gas
level instead of at the detailed particle level. Due to the reduced fluid blasting. Powder Technology, 109:13–26

calculations required, this technique enlarges the range of equipment 8


Feng, Y. and Yu, A. (2004). Assessment
of model formulations in the discrete
and processes that can be studied with numerical simulations. particle simulation of gas-solid flow. In-
dustrial & Engineering Chemistry Research,
The coupling of DEM with a finite volume method for the solution
43:1713–1728
of fluid phase in a computational cell level was first reported by Tsuji 9
Ye, M., Van der Hoef, M., and Kuipers,
et al.4 and Hoomans et al.5 , using the soft-sphere model and hard- J. (2004). A numerical study of flu-
idization behavior of geldart a particles
sphere model for the interaction force between particles, respectively. using a discrete particle model. Powder
Since then, numerous authors have published their work using the Technology, 139:129–139

Euler-Lagrange type of model to study granular flow (Hoomans


10
Ye, M., Van der Hoef, M., and Kuipers,
J. (2005). From discrete particle model
et al.6 , Xu et al.7 , Feng and Yu8 , Ye et al.9 , 10 , Goldschmidt et al.11 ), to a continuous model of geldart a
particles. Chemical Engineering Research
proving this approach as a very promising method. and Design, 83:833–843
In the DEM-CFD method, the fluid flow is obtained by the 11
Goldschmidt, M., Beetstra, R., and
conventional continuum approach, providing information to calculate Kuipers, J. (2002). Hydrodynamic mod-
elling of dense gas-fluidised beds: com-
the fluid forces acting on individual particles while the motion of the parison of the kinetic theory of granular
flow with 3d hard-sphere discrete par-
particle is obtained by using a discrete particle method. ticle simulations. Chemical Engineering
Some specific benefits to using the DEM-CFD coupled method vs. Science, 57:2059–2075

© 2019, esss - all rights reserved


dem-cfd coupling manual 4
ROCKY

using CFD alone are listed below.

• Unlike continuum methods, the motion of every particle is


simulated, so particle-particle interactions are taken into account.
Therefore there is no need to provide equations of state-motion of
granular systems, which are quite difficult to derive.

• By the same token, there is no limitation of low particle size


concentration and particle size distribution is easily prescribed
without increasing CFD solver computational cost.

• It is possible to deal with non-spherical particles.

• It is possible to model adhesive-cohesive materials by modeling the


attractive force between a pair of particles and between particles
and walls.

• Additionally, particle-particle and particle-walls heat transfer as


well as the convective heat transfer with the fluid can be included.

1.5 How is this document organized?


This document aims to provide the Rocky-Fluent coupling implemen-
tation details.
In chapter 2, the governing equations for both fluid and particle
phases adopting the Euler-Lagrange approach are given.
In chapter 3, the calculation of interaction forces between particles
and fluid is detailed. Section 3.2 presents a description of the drag
laws implemented in the one-way and two-way Rocky DEM-CFD
coupling. Section 3.4 provides documentation on lift force calculations
and on the options of lift coefficient laws that are implemented in the
Rocky DEM-CFD coupling modules.
In chapter 4, the heat transfer calculation between particles and
fluids as well as during contacts is described and Nusselt laws
currently available in this DEM-CFD implementation are detailed.
In chapter 5, the computational details of the coupling between
Rocky and Fluent are detailed and the coupling algorithm is given.
Chapter 6 gives the best practices for both one-way and two-way
coupling between Rocky and Fluent.

© 2019, esss - all rights reserved


dem-cfd coupling manual 5
ROCKY

2 Governing equations
In the Rocky-Fluent coupling, the fluid flow is obtained by the
conventional continuum approach using ANSYS Fluent® , in which
the conservation equations for mass, momentum, and energy are
solved by the finite volume method. The solid phase flow is modeled
using the discrete approach within Rocky. The coupling between
solid and fluid is accomplished by interphase momentum and heat
transfer terms due to the interaction between phases.
In this section, governing equations solved for the fluid and solid
phases are provided and the coupling methodology is described in
detail.

2.1 Particle phase modeling


2.1.1 Translational and rotational particle motion of a particle

In the frame of the DEM, all particles within the computational


domain are tracked in a Lagrangian way, by solving explicitly Euler’s
first and second laws, that govern translational and rotational particle
motion, respectively:

dv p
mp = Fc + F f → p + m p g (2.1)
dt


ωp
Jp = Mc + M f → p (2.2)
dt

where m p is the particle mass, g is the gravitational acceleration vector,


Fc is the contact force that accounts for particle-particle and particle-
wall interactions, ω p is the angular velocity vector, J p is its moment of
inertia tensor and Mc is the net torque generated by tangential forces
that causes the rotation of the particle.
Due to the fluid interaction, two additional terms appear when
comparing with a pure DEM simulation: F f → p is the additional force
accounting for the interaction with the fluid phase and M f → p is the

© 2019, esss - all rights reserved


dem-cfd coupling manual 6
ROCKY

additional torque due to the fluid phase velocity gradient, whose


calculations are further described in section 3.

2.1.2 Energy balance in a particle

If the thermal model is activated1 , an additional equation for the 1


For more information on the thermal
model used in Rocky, refer to the Rocky
energy balance is solved along with the equations governing the DEM Technical Manual.
motion of the particle. In the current implementation, the particle
temperature is assumed to be uniform, i.e., no radial or circum-
ferential temperature variation is admitted. This approximation is
reasonable for small or highly conductive particles.
The temperature variation of a particle can be obtained over time
according to the differential equation:

dTp
mpcp = q̇ (2.3)
dt

where c p is the specific heat of the particle material and q̇ is the total
particle heat transfer rate.
This heat transfer rate accounts for the heat transfer that occurs
during the contact with other particles or walls, q̇c , and the convective
heat transfer between particle and fluid phase, q̇ f → p , according to the
expression:
q̇ = q̇c + q̇ f → p (2.4)

The convective heat transfer (q̇ f → p ) between particle and fluid


is calculated by assuming a lumped capacitance system (Bi < 0.1)
and using one of many correlations available in the literature for
dense particulate systems. The convective fluid-particle heat transfer
calculation is described in section 4.2.

2.2 Fluid phase modeling


2.2.1 Mass and momentum conservation equations

The fluid phases are described by the classical Navier-Stokes equations


averaged in volume.2 The averaged mass conservation equation is 2
Drew, D. (1993). Mathematical model-
ing of two-phase flow. Annual Review of
given by: Fluid Mechanics, 15:261–291
∂    
αf ρf + ∇ · αf ρf uf = 0 (2.5)
∂t

© 2019, esss - all rights reserved


dem-cfd coupling manual 7
ROCKY

whereas the averaged momentum conservation equation is written as:

∂    
αf ρf uf + ∇ · αf ρf uf uf =
∂t  
− α f ∇ p + ∇ · α f T f + α f ρ f g + F p→ f (2.6)

where α f stands for the fluid volume fraction, p is the shared pressure,
ρ f is the fluid density, u f is the fluid phase velocity vector and T f is
the stress tensor of the fluid phase, defined as:
 
  2
Tf = µf ∇u f + ∇uTf + λf − µf ∇ · uf I (2.7)
3

In equation (2.2.1), F p→ f represents the source term of momentum


from interaction with the particulate phase, calculated according to
the expression:
∑N
p =1 F f → p
F p→ f = − (2.8)
Vc

where Vc is the computational cell volume, N is the number of


particles inside the computational cell volume is and F f → p accounts
for the forces generated by the fluid on the particles, calculated
according to the equations in section 3.

2.2.2 Energy conservation equation

In order to describe energy conservation, a separate enthalpy equation


is written for each fluid phase, according to:

∂    
αf ρf hf + ∇ · αf ρf uf hf =
∂t
∂p f
αf + α f T f : ∇u f − ∇ · q f + Q p → f (2.9)
∂t

where h f is the specific enthalpy of the fluid phase, q f is the heat


flux and Q p→ f is the heat exchange between the fluid and particulate
phases.
The heat exchange with the particulate phase is calculated
according to the equation:

∑N
p=1 q̇ f → p
Q p→ f = − (2.10)
Vc

where q̇ f → p is the heat transfer rate between the fluid and particle,
the calculation for which is shown in section 4.2.

© 2019, esss - all rights reserved


dem-cfd coupling manual 8
ROCKY

3 Particle-fluid interaction forces


The fluid interaction force, F f → p , is commonly split into two terms:
the drag force, FD , and a second term composed by the remaining
(non-drag) forces, FN-D , in the following way:

F f → p = FD + FN-D (3.1)

Amongst the most common non-drag forces are the pressure


gradient force, F∇ p , the added (virtual) mass force, FVM and the
lift force, FL , so we could write also:

F f → p = FD + F∇ p + FL + FVM + Fothers (3.2)

Depending on the flow conditions, the majority of these forces can


be ignored and only the drag and pressure gradient forces need to
be considered, such as in cases where the specific mass difference
between fluid and particles is high ( ρ p  ρ f ).

F f → p = FD + F∇ p (3.3)

3.1 Pressure gradient force


The pressure gradient force, F∇ p , is calculated according to the
expression:
F∇ p = −Vp ∇ p (3.4)

where Vp is the volume of the particle and ∇ p is the local pressure


gradient.

3.2 Drag force


The drag force, FD , acting on the particles is calculated using the
definition of the drag coefficient CD :1 1
Pritchard, P. J. (2010). Fox and McDon-
ald’s Introduction to Fluid Mechanics, 8th
1

0
  Edition. John Wiley & Sons
FD = CD ρ f A u f − v p u f − v p (3.5)

2

© 2019, esss - all rights reserved


dem-cfd coupling manual 9
ROCKY

where u f − v p is the relative velocity between particle and fluid and


0
A is the projected particle area in the flow direction.
Various drag correlations based on particle shape (spherical and
non-spherical) and particle concentration (dilute or dense flows) are
available within the Rocky package for the calculation of the drag
coefficient and are presented in sections 3.2.1 and 3.2.2.
All correlations use the relative particle Reynolds number, Re p ,
defined using the particle diameter and the relative particle-fluid
velocity according to2 : 2
ANSYS (2013b). Fluent Theory Guide.
ANSYS, Inc., Canonsburg, PA
ρ f |v − u| d p
Re p = (3.6)
µf

3.2.1 Single particle drag laws

A collection of correlations for the drag coefficient (drag laws) can


be found in the extensive technical literature available on particle-
fluid interactions. Some of the most common and validated drag
correlations for single particle (dilute flow) are implemented in the
Rocky DEM-CFD coupling modules and apply to spherical and non-
spherical particles.

3.2.1.1 Schiller & Naumann (1933)

The Schiller & Naumann drag correlation for spherical particles


provides the drag coefficient CD for Re < 800 with a maximum
deviation of 5% in relation to experimental data.3 The standard 3
Pritchard, P. J. (2010). Fox and McDon-
ald’s Introduction to Fluid Mechanics, 8th
version of the correlation is given by:4 Edition. John Wiley & Sons

24   4
Crowe, C., Schwarzkopf, J., Sommer-
CD = 1 + 0.15Re p0.687 (3.7) feld, M., and Tsuji, Y. (2011). Multiphase
Re p Flows with Droplets and Particles, Second
Edition. Taylor & Francis
A modification commonly used for inclusion of drag inertial range
(Re > 1000)5 , 6 , 7 is given by: 5
Crowe, C., Schwarzkopf, J., Sommer-
feld, M., and Tsuji, Y. (2011). Multiphase
Flows with Droplets and Particles, Second
 
24  0.687

CD = max 1 + 0.15Re p , 0.44 (3.8) Edition. Taylor & Francis
Re p
6
ANSYS (2013a). CFX Solver Theory
Guide. ANSYS, Inc., Canonsburg, PA
The modified version of this drag law is implemented in Rocky
7
ANSYS (2013b). Fluent Theory Guide.
DEM-CFD coupling, being the recommended drag law to be used for ANSYS, Inc., Canonsburg, PA
simulations with spherical particles.

© 2019, esss - all rights reserved


dem-cfd coupling manual 10
ROCKY

3.2.1.2 DallaValle (1948)

The DallaValle drag law8 also provides the drag coefficient for 8
DallaValle, J. M. (1948). Micromeritics:
the technology of fine particles. Pitman Pub.
spherical particles. Its validity range is up to Re p < 3000 and its Corp
main difference to the modified Schiller & Naumann correlation is
that it is a continuous function. The DallaValle correlation is defined
by: !2
4.8
CD = 0.63 + p (3.9)
Re p

Figure 3.1 shows the comparison of equation (3.8) and equation


(3.9) to experimental data for spherical particles. It can be seen
that the modified version of Schiller & Naumann drag correlation
and the DallaValle correlation both fit very well into the spherical
experimental data.

Figure 3.1: Comparison of drag laws for


dilute flows to experimental data.

3.2.1.3 Haider & Levenspiel (1989)

Haider & Levenspiel9 compiled drag coefficient and terminal velocity 9


Haider, A. and Levenspiel, O. (1989).
Drag coefficient and terminal velocity
experimental data for spherical and non-spherical particles. Then, of spherical and nonspherical particles.
they developed explicit expressions for both types of particles. For Powder Technology, 58:63–70

spherical particles, the correlation coefficients have fixed values,


whereas for non-spherical values the are dependent on the sphericity,
which is defined below.

© 2019, esss - all rights reserved


dem-cfd coupling manual 11
ROCKY

The unified correlation for both types of particles is given by:

24   C
CD = 1 + A Re pB + (3.10)
Re p 1 + ReDp

For spherical particles, the values of the coefficients in this


correlation are:
A = 0.1806
B = 0.6459
(3.11)
C = 0.4251
D = 6880.95

On the other hand, for non-spherical particles, they are given by


the expressions:
 
A = exp 2.3288 − 6.4581φ + 2.4486φ2

B = 0.0964 + 0.5565φ
  (3.12)
C = exp 4.905 − 13.8944φ + 18.4222φ2 − 10.2599φ3
 
D = exp 1.4681 + 12.2584φ − 20.7322φ2 + 15.8855φ3

where φ is the particle’s sphericity, defined as:

Asph
φ= (3.13)
Ap

where Asph is the surface area of a sphere having the same volume as
the particle and A p is the actual surface area of the particle.
As an illustration of the accuracy of this correlation, Figure 3.2
presents the comparison of equation (3.10) with experimental data
for four different types of non-spherical particles, all with different
values of sphericity.
The Haider & Levenspiel drag law the recommended drag law for
isometric non-spherical particles and for non-isometric non-spherical
particles tolerating some loss of accuracy. The particle sphericity is
automatically calculated by Rocky based on the particle’s shape and
size.

© 2019, esss - all rights reserved


dem-cfd coupling manual 12
ROCKY

Figure 3.2: Drag coefficient for different


sphericities using Haider & Levenspiel
drag model.

3.2.1.4 Ganser (1993)

Ganser10 showed that both Stokes’ shape factor and Newton’s shape 10
Ganser, G. H. (1993). A rational
approach to drag prediction of spher-
factor are important for the prediction of drag. Stokes’ shape factor ical and nonspherical particles. Powder
(K1 ) is defined as the ratio between the drag coefficient of a spherical Technology, 77:143–152

particle and the drag coefficient for a particle with an arbitrary shape,
both in Stokes’ flow (Re p  1). Newton’s shape factor (K2 ) is defined
as the ratio between the drag coefficient of a particle with an arbitrary
shape and the drag coefficient of a spherical particle, both in Re p =
10000.
Thereby, Ganser developed a simplified drag equation that is a
function only of the generalized Reynolds number Re p K1 K2 . This
equation, applicable to all shapes and valid for Re p K1 K2 < 105 , is the
following:

CD 24 h 0.6567 i 0.4305
= 1 + 0.1118 Re p K1 K2 + (3.14)
K2 Re p K1 K2 1 + Re3305
p K K2
1

The suggested expressions of K1 and K2 for isometric and non-


isometric particles are, respectively, the following:
 −1
dp

1 dn 2 1
K1 = + φ− 2 − 2.25 (3.15)
3 dp 3 D
0.5743
K2 = 101.8148(− log10 φ) (3.16)

where dn is the diameter of a spherical particle with the same projected


area of the actual particle in the direction of the flow, and d p is the

© 2019, esss - all rights reserved


dem-cfd coupling manual 13
ROCKY

diameter of a spherical particle with the same volume of the actual


particle, and D is the diameter of the container.
This correlation calculates both the Stokes and Newton parameters,
considering the effects of shape and alignment of the particle with
the flow field to compute the drag coefficient. This is a more accurate
(but also more computationally expensive) option.

3.2.2 Dense flow drag laws

The drag laws presented in section 3.2.1 were developed (generally)


for a single particle in an infinite medium and can be applied
for collections of particles as long as the criteria for dilute flow is
satisfied.11 ANSYS (2013b). Fluent Theory Guide.
11

ANSYS, Inc., Canonsburg, PA


For a dense flow of particles, different drag laws must be used.
Some of these dense flow drag laws are corrections over the single
particle drag laws based on fluid volume fraction (α f ). Others are
completely independent equations.

3.2.2.1 Wen & Yu (1966)

For a relatively low particle concentration (αs < 0.2), Wen & Yu
developed a drag correlation based on a series of experiments
on fluidized beds conducted by Gidaspow12 . This correlation is 12
Gidaspow, D. (2012). Multiphase Flow
and Fluidization: Continuum and Kinetic
presented in terms of a correction (based on fluid volume fraction) Theory Descriptions. Elsevier Science
of the Schiller & Naumann correlation, using a superficial velocity
relative to the particle’s Reynolds number. The corresponding
mathematical expression is13 : 13
ANSYS (2013a). CFX Solver Theory
Guide. ANSYS, Inc., Canonsburg, PA
( 
24  0.687 
CD = α−
f
1.65
max 1 + 0.15 α f Re p , 0.44} (3.17)
α f Re p

3.2.2.2 Ergun (1958)

For higher particle concentrations, the Wen & Yu drag law deviates
from the experimental data. For these cases, which can have solid 14
Crowe, C., Schwarzkopf, J., Sommer-
feld, M., and Tsuji, Y. (2011). Multiphase
volume fraction αs < 0.2 up to the maximum packing limit (usually Flows with Droplets and Particles, Second
60 - 70%), Ergun developed the following correlation to the head loss Edition. Taylor & Francis

in fixed beds:14 , 15 15
Gidaspow, D. (2012). Multiphase Flow
and Fluidization: Continuum and Kinetic
αs 7 Theory Descriptions. Elsevier Science
CD = 200 + (3.18)
α f φ2 Re 3φ

© 2019, esss - all rights reserved


dem-cfd coupling manual 14
ROCKY

3.2.2.3 Gidaspow, Bezburuah & Ding (1992)

Gidaspow, Bezburuah & Ding16 have developed a simple connection 16


Gidaspow, D., Bezburuah, R., and
Ding, J. (1993). Hydrodynamics of
between the Wen & Yu and Ergun correlations to represent the circulating fluidized beds: Kinetic theory
complete range of solids volume fraction in a single drag law, by approach. In Engineering foundation
conference on fluidization, volume 7, pages
simply applying each law over its valid range. The Gidaspow, 75–82, Brisbane, Australia
Bezburuah & Ding correlation is then given by:

 0.687 
24
 

 1 + 0.15 α f Re p α−
f
1.65
, α f > 0.8, α f Re p < 1000



 α f Re p

CD = 0.44 α−f
1.65
, α f > 0.8, α f Re p ≥ 1000



 αs 7
200 + , α f ≤ 0.8


α f φ2 Re p 3φ
(3.19)

3.2.2.4 Huilin & Gidaspow (2003)

The Gidaspow, Bezburuah & Ding drag correlation covers the entire
range of solids (particle phase) volume fraction (from 0 up to the
maximum packing limit) but presents a discontinuity at the point
α f = 0.8. To make the transition between the Wen & Yu and Ergun
correlations in a smoother way, Huilin & Gidazpow17 have applied 17
Huilin, L. and Gidaspow, D. (2003).
Hydrodynamics of binary fluidization in
a blending function to promote the connection based on the fluid a riser: Cfd simulation using two gran-
volume fraction. The final drag correlation is given by: ular temperatures. Chemical Engineering
Science, 58:3777–3792

CDHuilin&Gidaspow = ψ CDErgun + (1 − ψ) CDWen&Yu (3.20)

The blending parameter ψ is defined as a function of the fluid


volume fraction, α f , given by:
h  i
arctan 150 · 1.75 0.8 − α f
ψ= + 0.5 (3.21)
π

3.2.2.5 Di Felice (1994)

Using experimental data, correlations from previous works and


analytical results, Di Felice18 derived a correction function for the 18
Di Felice, R. (1994). The voidage
function for fluid-particle interaction
single particle drag coefficient, in order to consider the case of dense systems. International Journal of Multi-
particle flow. The correlation is given by: phase Flow, 20:153–159

2− ς
CD = CD0 α f (3.22)

© 2019, esss - all rights reserved


dem-cfd coupling manual 15
ROCKY

where CD0 is the drag coefficient for a single particle, also calculated
using the particle’s Reynolds number based on the superficial relative
velocity. In his work, Di Felice used the DallaValle19 correlation for 19
DallaValle, J. M. (1948). Micromeritics:
the technology of fine particles. Pitman Pub.
single particle drag coefficient. Corp
The exponent ς in equation (3.22) is calculated by means of:
 h  i2 
 1.5 − log10 α f Re p
 

ς = 3.7 − 0.65 exp − (3.23)

 2 

A comparison among these drag laws for dense flows is presented


in Figure 3.3 for a constant fluid volume fraction of α f = 0.4. At this
value of volume fraction, all correlations seem to match, but one must
pay attention to the differences over a range of fluid volume fraction.

Figure 3.3: Comparison of the dense


flow drag laws for α f = 0.4.

Figure 3.4 shows the variation of these same laws for a fixed
Re p = 1000 and a varying fluid volume fraction. The variances are
significant and will generate strong differences on overall particle-
fluid interaction.

3.3 Virtual mass force


The virtual mass force is caused by the fact that when the particle
accelerates it drags some of the surrounding mass of the fluid. This
mass portion is supposed to attain the particle velocity, resulting in
a virtual increase in the particle mass. This force is significant when

© 2019, esss - all rights reserved


dem-cfd coupling manual 16
ROCKY

Figure 3.4: Comparison of the dense


flow drag laws for Re p = 1000.

the density of the fluid phase is similar or higher than the density of
the particles, such as in light particles flowing in water.
The virtual mass force is defined by the expression:

FVM = CVM ρ f Vp ar (3.24)

where CVM ρ f Vp represents the mass of fluid displaced by the particle,


whereas ar is the relative acceleration. The virtual mass coefficient,
CVM , represents the particle volume of displaced fluid that contributes
to the effective virtual mass of the particle, in terms of a fraction of
the particle volume.
The theoretical value of CVM for a sphere in an infinite medium is
0.5. However, for higher particle volume fractions, particles interaction
effects become important, increasing the fluid mass displaced. In
general, it depends on particle shape and particle concentration.

3.3.1 Virtual mass models

Three models are currently implemented in Rocky DEM-CFD


coupling and are described in the following sections.

3.3.1.1 Constant value

As already mentioned, the theoretical value for CVM for a sphere in an


infinite medium is 0.5. So, when the constant value option is chosen

© 2019, esss - all rights reserved


dem-cfd coupling manual 17
ROCKY

the virtual mass coefficient is set with that value:

CVM = 0.5 (3.25)

3.3.1.2 Ishii & Mishima (1984)

Ishii & Mishima20 calculated the virtual mass coefficient as a function 20


Ishii, M.; Mishima, K. (1984). Two-
fluid model and hydrodynamic consti-
of the volume fraction of the particulate phase, according to the tutive relations. Nuclear Engineering and
expression: Design, 82:107–126
2 − αf
CVM = (3.26)
2 αf

3.3.1.3 Paladino (2005)

Paladino21 suggested a virtual mass coefficient as a function of volume 21


Paladino, E. E. (2005). Estudo do
escoamento multifásico em medidores de
fraction of the particulate phase. This model showed results for the vazão do tipo pressão diferencial. PhD
differential pressure in excellent agreement with experimental data thesis, Universidade Federal de Santa
Catarina, SC
published by Lewis and Davidson22 . The Paladino virtual mass 22
Lewis, D. A. and Davidson, J. F. (1985).
coefficient is given by: Pressure drop for bubbly gas-liquid flow
through orifice plates and nozzles. Chem.
CVM = 1.5 − α f (3.27) Eng. Res. Des., 63:149–156

Figure 3.5 compares the behavior of each virtual mass model with
the increase of the solid phase volume fraction.

Figure 3.5: Virtual mass coefficients


calculated using two different models.

© 2019, esss - all rights reserved


dem-cfd coupling manual 18
ROCKY

3.4 Lift force


The lift force, F L , is a force transversal to the direction of the relative
velocity between particle and fluid velocities. This force is generated
by a series of factors, the most common are fluid velocity gradients or
a rotating particle that generates different pressure values at the sides
of the particle. In most cases, the lift force is insignificant compared to
the drag force, so there is no reason to include this extra term. If the
lift force is significant (for example, if the phases separate quickly), it
may be appropriate to include this term.
The lift force is usually calculated based on the lift coefficient (CL ),
and uses a cross product between the relative velocity and the curl of
the velocity, which naturally provides the force direction, as can be
seen in the following equation:

F L = CL ρ f Vp ur × (∇ × u) (3.28)

where F L is the actual lift force, Vp is the particle volume, ur is the


relative velocity between fluid and particle and ρ f is the density of
the fluid.
The value of the lift coefficient, CL , is generally calculated from
correlations derived from experimental data or analytical/numerical
solutions. The available laws for the calculation of the lift force
coefficient in Rocky are detailed in the following sections.

3.4.1 Lift laws

In all lift force laws supported by Rocky DEM-CFD coupling, the


common factors in the definition of the correlations are the usual
Reynolds number based on the relative velocity between the particle
and fluid (Re p ) as given by equation (3.6), and the vorticity Reynolds
number (ReΩ ) given by:

ρ f |∇ × u| d2p
ReΩ = (3.29)
µf

as well as the particle’s rotational Reynolds number (Reω ), defined as:

ρ f ω p d2p

Reω = (3.30)
µf
where ω p is the particle angular velocity vector.
It is common to consider the ratio of Re p and ReΩ , and it is usually

© 2019, esss - all rights reserved


dem-cfd coupling manual 19
ROCKY

expressed in terms of a variable χ, defined as:

1 ReΩ
χ= (3.31)
2 Re p

3.4.1.1 Saffman (1968)

The Saffman lift force23 , 24 is due to a pressure difference developed on 23


Saffman, P. G. (1965). The lift on a
small sphere in a slow shear flow. J.Fluid
a particle in a fluid velocity gradient. The higher velocity at one side Mech, 22:385–400
of the particle develops a lower pressure and the lower velocity on the 24
Saffman, P. G. (1968). Corrigendum
other side develops a higher pressure (according to the Bernoulli’s to: The lift on a small sphere in a slow
shear flow. J.Fluid Mech, 31:624
principle). Saffman analyzed the lift force for low (relative) Reynolds
numbers and reached the following expression for the force in a
one-dimensional configuration:
1
d2p du x du x − 2

1
FL = 6.46 ρ f ν f2
(u x − v x ) (3.32)
4 dy dy

This expression can be generalized for three-dimensional flows as:

  1 d2 1
p
ur × (∇ × u) |∇ × u|− 2
2
F L = 6.46 ρ f µ f (3.33)
4

Moreover, using the definition of the CL and ReΩ , the Saffman lift
coefficient can be written as:

3
CL = 6.46 √ (3.34)
2π ReΩ

It can be noted that if the relative velocity is positive, there is a lift


force toward the higher velocity of the continuous phase, whereas if
the relative velocity is negative, the lift force is oriented toward the
lower continuous phase velocity. The expression for the lift coefficient

in equation (3.34) is valid for 0 6 Re p 6 ReΩ 6 1.
Figure 3.6 compares the shear lift coefficient, CL , given by equation
(3.34) with the numerical results of Dandy and Dwyer25 used by 25
Dandy, D. S. and Dwyer, H. A. (1990).
A sphere in shear flow at finite Reynolds
Mei26 to evaluate the Saffman law for two different χ. Satisfactory number: effect of shear on particle lift,
agreement can be observed for χ = 0.1 and χ = 0.4. As Re p increases, drag, and heat transfer. J. Fluid Mech,
216:381–410
it was found that the lift coefficient decreases and it levels off around 26
Mei, R. (1992). An approximate expres-
Re p = 40. sion for the shear lift force on a spher-
ical particle at finite Reynolds number.
International Journal of Multiphase Flow,
18:145–147

© 2019, esss - all rights reserved


dem-cfd coupling manual 20
ROCKY

Figure 3.6: Comparison of Saffman and


Mei lift laws.

3.4.1.2 Mei (1992)

Mei27 expanded the validity of the Saffman law to higher relative 27


Mei, R. (1992). An approximate expres-
sion for the shear lift force on a spher-
Reynolds number values, by defining a correction term acting over ical particle at finite Reynolds number.
the Saffman lift coefficient in the following way: International Journal of Multiphase Flow,
18:145–147
3
CL = CL∗ 6.46 √ (3.35)
2π ReΩ

where the correction term is given by:

Re p
   
1 1
 1 − 0.3314χ 2 exp −
 + 0.3314χ 2 , Re p ≤ 40

CL = 10
 1
0.0524 χ Re p 2 , Re p > 40

(3.36)

With this correction, the correlation is still limited only to


spherical particles, but is valid for wider ranges: 0.1 6 Re p 6 100,
0.005 6 χ 6 0.4.

3.5 Fluid generated torque


The trajectory of a particle moving in a fluid can be significantly
influenced by its rotation, specially for large and/or heavy particles
with high moments of inertia. In order to account for particle rotation,
an ordinary differential equation for the angular momentum of the
particle, equation (2.2), is solved along with the equation for the

© 2019, esss - all rights reserved


dem-cfd coupling manual 21
ROCKY

particle translational motion, equation (2.1).


One additional form of interaction between particle and fluid is
the generation of an angular moment, or a torque, over the particle
when it moves in a fluid. In the Rocky DEM-CFD coupling, the fluid
generated torque is computed based on the torque coefficient (CT )
according to:
5
1 dp
M f → p = CT ρ f 5 |ω r | ω r (3.37)
2 2

where ω r is the relative particle-fluid angular velocity, given by:

1
ωr = ∇ × uf − ωp (3.38)
2

3.5.1 Torque laws

The torque coefficient is usually given as a function of the Reynolds


number based on the relative angular velocity Reωr , calculated as

ρ f d2p |ω r |
Reωr = . (3.39)
µf

3.5.1.1 Dennis, Singh & Ingham (1980)

Dennis et al.28 have investigated the torque necessary to keep a sphere 28


Dennis, S., Singh, S., and Ingham, D.
(1980). The steady flow due to a rotating
rotating at an angular speed ω p in an otherwise stagnant fluid. The sphere at low and moderate Reynolds
torque coefficient is given by: numbers. Journal of Fluid Mechanics,
101:257–279
128.64  p 
CT = 1 + 0.1005 Reωr (3.40)
Reωr

The range of validity for this expression is 20 < Reωr < 2000.

3.6 Turbulent dispersion force


When enabled by the user in Rocky under the CFD coupling settings,
the turbulent dispersion acts as a diffusion mechanism in dispersed
flows, resulting in a transference of particles from high volume
fraction regions to low volume fraction regions due to turbulent
fluctuations.
The model of turbulent dispersion detailed in this chapter assumes
that the instantaneous velocity of the fluid phase is a combination of

© 2019, esss - all rights reserved


dem-cfd coupling manual 22
ROCKY

a mean and a fluctuating value:

0
u f = ū f + u f (3.41)

0
where ū f is the mean velocity of the fluid phase and u f is the
fluctuation caused by turbulent effects. The turbulent force then
0
arises as a consequence of the presence of u f in equation (3.41), as
it will generate an additional drag force when u f is substituted in
equation (3.5).

3.6.1 Turbulent dispersion model

Rocky takes an approach similar to Gosman and Ioannides29 in order 29


Gosman, A. D. and Ioannides, E.
0 (1983). Aspects of computer simulation
to model u f , where the dispersion of particles due to turbulence of liquid-fueled combustors. Energy
in the fluid phase is predicted stochastically30 . In this model, the Journal, 7:482–490

turbulence is randomly sampled during each particle’s trajectory and


30
Unless mentioned otherwise, all equa-
tions of this section follow the model
allowed to influence its motion. The gross behavior of the turbulence proposed by Gosman and Ioannides.

in the simulated system emerges as a consequence of the averaging


that naturally occurs when the random sampling is performed for a
statistically significant number of particles.
Specifically, the influence of the turbulence over a particle is
simulated by means of the interaction with a succession of discrete
fluid phase turbulent eddies31 . Each eddy is characterized by a ANSYS (2013b). Fluent Theory Guide.
31

0 ANSYS, Inc., Canonsburg, PA


fluid velocity fluctuation u f and a time scale τe (the eddy lifetime).
0
When a particle is interacting with a turbulent eddy, u f influences the
particle by means of an additional drag force during the particle-eddy
interaction time τint .
0
The following sections detail how Rocky estimates u f , τe and τint .

3.6.1.1 Eddy fluctuating velocity


0
In order to estimate u f , it is assumed that the turbulence is isotropic.
The fluid fluctuating velocity can be decomposed as a magnitude
scalar and a direction unit vector:

0 0
u f = u f êu0 (3.42)

in which both the magnitude and the direction unit vector contain
random components as described ahead. At the start of the lifetime of
0
a turbulent eddy, u f and êu0 are sampled independently, and equation
0
(3.42) is used to estimate the eddy’s characteristic value u f .

© 2019, esss - all rights reserved


dem-cfd coupling manual 23
ROCKY

The magnitude of the fluctuating velocity is estimated as:

0
u f = |U| (3.43)

where U is a random variable distributed normally around zero:

U ∼ N (0, σu 2 ) (3.44)

The standard deviation σu of U is given by:


r

σu = (3.45)
3

where κ is the kinetic energy of the turbulence associated with the


flow32 . 32
κ is automatically provided by Fluent
when its k-epsilon viscous model is
The direction of the fluctuating velocity is assumed to be a random employed for two- and one-way coupled
variable given by: simulations. For constant one-way cou-
pled simulations, κ must be directly set
êu0 ∼ S 2 (3.46) by the user under the Constant One-Way
settings in Rocky.
where S 2 is a uniform distribution of points over the surface of a unit
sphere.

3.6.1.2 Eddy lifetime

The lifetime τe is another characteristic value of turbulent eddies.


An estimate of τe is made under the further assumption that the
characteristic size of the eddy is equal to the dissipation length scale
of the system, given by:

0.09κ 3
le = (3.47)
e

where e is the dissipation rate of the turbulent kinetic energy


associated with the flow33 . 33
e is automatically provided by Fluent
when its k-epsilon viscous model is
The eddy lifetime is then approximated as34 : employed for two- and one-way coupled
simulations. For constant one-way cou-
le pled simulations, e must be directly set
τe = (3.48) by the user under the Constant One-Way
σu
settings in Rocky.
34
In Gosman and Ioannides’s paper,
equation (3.48) has the denominator
3.6.1.3 Particle-eddy interaction time replaced by the magnitude of the fluid
fluctuating velocity. This was found to
A further assumption of this turbulent model is that each particle of cause unrealistically long eddy lifetimes
however.
the simulation has a one-to-one association with a turbulent eddy
during an interaction time interval τint . For estimating this particle-
eddy interaction time, two possible outcomes are considered:

© 2019, esss - all rights reserved


dem-cfd coupling manual 24
ROCKY

1. The particle moves sufficiently slowly relative to the fluid in order


to remain within the influence of the eddy during its whole lifetime
τe ;

2. The relative velocity between the particle and the fluid is high
enough to allow the particle to transverse the eddy in a transit time
τr shorter than τe .

The particle-eddy interaction time is therefore defined as the


minimum of the above, i.e.:

τint = min (τe , τr ) (3.49)

The transit time τr is estimated from the following solution of a


simplified form of the motion equation of a small particle in a fluid
medium:  
le
τr = −τp ln 1 −  (3.50)
τp u f − u p

where τp is the particle relaxation time defined as:

4ρ p d p
τp = (3.51)
3ρ f CD u f − u p


In cases where le > τp u f − u p , equation (3.50) has no solution.

This can be interpreted as the particle being "captured" by the


turbulent eddy, in which case τr → ∞ in equation (3.49) and
consequently τint = τe .

© 2019, esss - all rights reserved


dem-cfd coupling manual 25
ROCKY

4 Heat transfer
4.1 Heat transfer modes in the DEM-CFD coupling
There are different modes of heat transfer in a dense particulate
system. According to Vargas and McCarthy1 , the most common 1
Vargas, W. L. and McCarthy, J. J. (2002).
Stress effects on the conductivity of
modes of heat transfer pertaining to the particulate phase occur by particulate beds. Chemical Engineering
the following mechanisms: Science, 57:3119–3131

• Thermal conduction through the solid


• Thermal conduction through the contact area between two particles
• Radiant heat transfer between the fluids within neighboring voids
• Radiant heat transfer between the surfaces of neighboring particles
• Thermal conduction through the fluid between the neighboring
particles
• Heat transfer by inter-particle convection, if the fluid is flowing
• Frictional heating between the particles and particle-surface

According to Zabrodsky2 , radiation heat transfer can be neglected 2


Zabrodsky, S. S. (1966). Hydrodynamics
and Heat Transfer in Fluidized Beds. MIT
at low temperatures, typically T < 700 K. The CFD coupling options Press, Cambridge, MA
in this version of Rocky do not take into account radiant heat transfer.
Disregarding the radiant heat transfer, Batchelor and O’Brien3 3
Batchelor, G. K. and O’Brien, R. W.
(1977). Thermal or electrical conduction
have shown that the conduction through the solid phase (inside and through a granular material. Proceedings
between particles) dominates the conduction process when: of the Royal Society of London, 355:313–333

k s dc
1 (4.1)
k f d¯p

where k s is the conductivity of the solid phase, k f is the conductivity


of the interstitial media, dc is the diameter of the contact spot and d¯p
is the average diameter of the particles. This expression is satisfied,
for instance, in cases with high conductivity ratios (k s /k f ) or cases
with solid particles in a vacuum (k f → 0). Molerus4 pointed out that 4
Molerus, O. (1997). Heat transfer in
moving beds with a stagnant interstitial
under these conditions, the contact conductance between adjacent gas. International Journal of Heat and Mass
particles in the presence of a stagnant interstitial gas is the controlling Transfer, 40:4151–4159

process in both slowly moving and static beds of particles.

© 2019, esss - all rights reserved


dem-cfd coupling manual 26
ROCKY

Besides the thermal conduction due to particle-particle and particle-


surface contact that is already taken into account if the thermal model
is activated5 , the available CFD coupling modes in this version of 5
For more information on the thermal
model used in Rocky, refer to the Rocky
Rocky also consider the convective heat transfer between particles DEM Technical Manual.
and the fluid6 . 6
Thermal conduction through the fluid
between neighboring particles is ig-
nored, as well as the frictional heating
between particles and particle-surfaces.
4.2 Heat transfer between fluid and particle
The heat transfer rate between a particle and fluid, q̇ f → p , can be
calculated using Newton’s law of cooling, which states that the rate
of heat loss of a body is directly proportional to the difference in the
temperature between the surface of the body and its surroundings.
Assuming that the heat transfer coefficient is relatively independent
of the temperature difference between the body and the surroundings,
the heat transfer hate is given by:
 
q̇ f → p = hA p T f − Tp (4.2)

where Tp is the temperature of the particle’s surface, T f is the local


fluid temperature and A p is the particle surface area.
The heat transfer coefficient h depends upon fluid physical
properties, as well as of the operating conditions. For instance,
for the same fluid and particles, turbulent flows give higher heat
transfer coefficients when compared to laminar flows. Therefore, a
heat transfer coefficient needs to be provided for the analyzed system.
In Rocky, the average convective heat transfer coefficient, h, is
calculated based on the Nusselt number, Nu, according to the
expression:
k f Nu
h= (4.3)
dp
where k f is the fluid thermal conductivity and d p is the equivalent
diameter of the particle.
Different correlations can be found in the literature for calculating
the Nusselt number, usually in function of the Reynolds number, Re p ,
and Prandtl number, Pr, defined as:
cf µf
Pr = (4.4)
kf

where µ f is the fluid’s dynamic viscosity, and c f is its specific heat.

© 2019, esss - all rights reserved


dem-cfd coupling manual 27
ROCKY

4.2.1 Fluid-particle heat transfer correlations

4.2.1.1 Ranz & Marshall (1952)

The Ranz & Marshall (1952) correlation7 , 8 is valid for estimating the 7
Bergman, T., Incropera, F., DeWitt, D.,
and Lavine, A. (2011). Fundamentals of
heat transfer between a spherical particle and its surroundings. The Heat and Mass Transfer. Wiley
correlation is given: 8
E, M. (2006). Particles, Bubbles And
Drops: Their Motion, Heat And Mass
Transfer. World Scientific Publishing
Nu = 2 + 0.6 Re1/2
p Pr
1/3
(4.5) Company

where Re p is the relative Reynolds number based on the diameter


of the particle and the relative velocity and Pr is the Prandtl
number, computed according to equation (4.4). The Ranz & Marshall
correlation is valid for Re p < 5 · 104 .

4.2.1.2 Whitaker (1972)

The Whitaker (1972) correlation9 , 10 is defined as: 9


Bergman, T., Incropera, F., DeWitt, D.,
and Lavine, A. (2011). Fundamentals of
!1/4 Heat and Mass Transfer. Wiley
  µf∞
Nu = 2 + 0.4 Re1/2
p + 0.06 Re2/3
p Pr 0.4
(4.6) 10
Bejan, A. (2013). Convection Heat
µfw Transfer. Wiley

where µ f ∞ is the viscosity of the fluid at the free stream temperature,


whereas µ f w is the viscosity of the fluid at a temperature equal to the
temperature of the sphere’s surface.
The Whitaker correlation is suitable for the estimation of the
Nusselt number for a single spherical particle with Prandtl number
within 0.71 < Pr < 380, Reynolds number in the range of 3.5 < Re p <
7.6 · 104 , and viscosity ratio within 1 < µ f ∞ /µ f w < 3.2.
For a no-flow condition, this correlation returns Nu = 2, which
is the estimation for steady radial pure conduction between a
spherical surface and the motionless, infinite, conducting medium
that surrounds the particle.

4.2.1.3 Gunn (1978)

Gunn11 derived a correlation for the Nusselt number based on 11


Gunn, D. (1978). Transfer of heat or
mass to particles in fixed and fluidised
analytical (simplified) solutions and four asymptotic relations that beds. Int. J. Heat Mass Transfer, 21:467–
delineate the bounds of the Nusselt number for heat transfer to 476

particles at low and high Reynolds number. These known asymptotic


relations are: the Nusselt number for a single particle at low Reynolds
number, the experimentally-established dependence of the Nusselt

© 2019, esss - all rights reserved


dem-cfd coupling manual 28
ROCKY

Figure 4.1: Nusselt correlations for fixed


bed of spherical particles

number for single particles and fixed beds at high Reynolds numbers
and a limiting value derived in the paper for low Reynolds numbers
in a packed bed. Using these relations, the derived expression is:
  
Nu = 7 − 10α f + 5α2f 1 + 0.7Re0.2
p Pr
1/3
 
+ 1.33 − 2.4α f + 1.2α2f Re0.7
p Pr
1/3
(4.7)

where α f is the fluid volume fraction. The empirical correlation of


Gunn is useful for calculating the heat transfer rate in a fixed or 12
Littman, H. and Silva, D. E. (1970).
fluidized bed of particles within the fluid volume fraction range of Gas-particle heat transfer coefficient in
packed beds at low Reynolds number.
0.35 6 α f 6 1, for gases and liquids, and for Reynolds numbers up to
In Proceeding of 4th International Heat
Re p 6 105 . Transfer Conference, Versilles, France
13
Gunn, D. and De Souza, J. F. C. (1974).
Heat transfer and axial dispersion in
packed beds. Chemical Engineering Sci-
4.2.2 Comparison of heat transfer correlations ence, 29:1363–1371
Turner, G. A. and Otten, L. (1973). Val-
14

Figure 4.1 shows a comparison of the Nusselt number prediction ues of thermal (and other) parameters in
packed beds. Ind. Eng. Chem. Proc. Des.
using the correlation of Gunn and experimental results obtained by Dev., 12:417–424
different authors, for different Reynolds number values 12 , 13 , 14 , 15 . 15
Miyauchi, T., Kikuchi, T., and Hsu, K.-
Figure 4.2 presents a comparison of predictions for the Nusselt H. (1976). Limiting sherwood number of
sphere packed beds by electrical method.
number using the correlation of Ranz & Marshall (1952), Whitaker Chemical Engineering Science, 31:493 – 498
(1972) and Gunn (1978) with experimental data for a single spherical
particle in water under different Reynolds numbers.

© 2019, esss - all rights reserved


dem-cfd coupling manual 29
ROCKY

Figure 4.2: Nusselt number prediction


for different correlations for a single
spherical particle

© 2019, esss - all rights reserved


dem-cfd coupling manual 30
ROCKY

5 Computational details
5.1 Fluid phase calculation
In the ANSYS Fluent® Eulerian multiphase model1 , the fluid phase 1
ANSYS (2013b). Fluent Theory Guide.
ANSYS, Inc., Canonsburg, PA
conservation equations are identical to the fluid equations of the
DEM-CFD coupling presented in section 2. This enabled the usage
of Fluent® solver in this coupling implementation with Rocky for
one-way and two-way coupling modes2 . 2
Constant one-way does not require the
Fluent solver.
The choice of using ANSYS Fluent® to solve the fluid phase was
mainly due to its user-defined functions support (UDFs) for the
customization of some parts of the code. This feature has allowed
the definition of the source terms in the fluid phase conservation
equations that carried out the integration of DEM model for modeling
the solid phase.
ANSYS Fluent® solver is based on the standard finite volume
method3 which is cell centered and adopts an implicit scheme for 3
Patankar, S. (1980). Numerical Heat
Transfer and Fluid Flow. Series in com-
time stepping. A block algebraic multigrid solver is used for the putational methods in mechanics and
solution of the linearized equations.4 thermal sciences. Taylor & Francis

The momentum and energy equations of the phase that represents


4
Hutchinson, B. and Raithby, G. (1986).
A multigrid method based on the ad-
the discrete phase are not solved by Fluent® , the dispersed phase ditive correction strategy. Numer. Heat
Transfer, 9:511–537
velocity and temperature fields are solved by Rocky. A source term
is included on the disperse phase continuity equation to impose the
disperse volume fraction calculated at the DEM side. The momentum
exchange term is not calculated in the CFD solver, but on the DEM
side, as explained in section 2, and included through a source term
in the continuous phase momentum equation. The PC-SIMPLE
algorithm, which is the SIMPLE algorithm extended to multiphase
flows, is used for the pressure-velocity coupling. More detailed
information about this coupling segregated pressure-based method
can be found in the paper by Vasquez and Ivanov5 . 5
Vasquez, S. A. and Ivanov, V. (2000).
A phase coupled method for solving
multiphase problems on unstructured
meshes. In Fluids Engineering Division
5.2 Source terms linearization Summer Meeting, Boston, Massachusetts.
ASME

In order to improve convergence, a semi-implicit treatment is adopted


for the momentum and heat source terms on the CFD side of the

© 2019, esss - all rights reserved


dem-cfd coupling manual 31
ROCKY

coupling. In this procedure, rather than simply using the value of the
drag force and convective heat transfer computed in the DEM solver,
these terms are divided into an explicit and an implicit part.

5.2.1 Momentum terms

To apply this procedure, equation (2.8) is rewritten using equations


3.2 and 3.5:
∑N
p=1 FD + FN-D
F p→ f = − (5.1)
Vc
 
∑N
p =1 β p f v p − u f ∑N
p=1 FN-D
F p→ f = − − (5.2)
Vc Vc

In this last equation, β p f is the momentum exchange coefficient


between solid and fluid phase, defined as:

1
0

βp f = CD ρ f A u f − v p (5.3)

2

where CD is the drag coefficient, calculated using the correlations


0
presented in section 3.2 and A is the area of the particle projected in
the force direction.
Since in the classic finite volume method the fluid velocity is
considered constant within a cell, the first term on the right hand side
of the equation (5.2) can be subdivided into two terms and written as:

∑N
p =1 β p f v p ∑N
p =1 β p f ∑N
p=1 FN-D
F p→ f = − − uf − (5.4)
Vc Vc Vc

This expression can be further rewritten, splitting the interaction


forces into an explicit term, A, and an implicit term, B, that is function
of the fluid velocity on the current time step:

∑N
p=1 β p f v p + FN-D ∑N
p =1 β p f
F p→ f = − − u f = A + Bu f (5.5)
Vc Vc

5.2.2 Heat transfer terms

The heat exchange between particles and fluid is calculated in Rocky


and sent to Fluent via a source term. This source term is linearized to
enhance stability on the CFD side of the simulation.
The expression for the heat transfer rate between particle and fluid

© 2019, esss - all rights reserved


dem-cfd coupling manual 32
ROCKY

is:  
q = h̄A p Tp − T f (5.6)

Equation 5.6 can be split into two terms following the general
linearized source term on ANSYS6 at section Linearized Source Terms: 6
ANSYS (2013b). Fluent Theory Guide.
ANSYS, Inc., Canonsburg, PA

Sφ = A + Bφ (5.7)

Then, the following equivalences arise from the source term


linearization:
φ = Tf (5.8)
A = h̄A p Tp (5.9)
B = −h̄A p (5.10)

5.3 Lagrangian to Eulerian mapping


The DEM-CFD coupling method was developed for the ideal case in
which the particles are smaller than the CFD cells (Figure 5.1), so that
the particulate phase volume fraction of a cell can be calculated as:

∑N
p=1 Vp
αp = (5.11)
Vc

where Vc is the computational cell volume, Vp is the particle volume


and N is the number of particles whose centroids lie inside the cell.

Figure 5.1: Schematic representation of


many particles with volumes smaller
than CFD cells, with volume fraction
values calculated by equation (5.11) rep-
resented as cell colors. In this situation
the volume fraction distribution is real-
istic.

Equation (5.11) is accurate as long as Vp  Vc . As the relative sizes


of the particles increase compared to the sizes of the cells, however,
the discrete representation of particles in the DEM domain cause
equation (5.11) to lose accuracy when mapping solid volume to the
CFD domain, as represented by Figure 5.2.
In order to refrain from sending unrealistic values of solid volume
fraction to the CFD domain, in which the flow is described following

© 2019, esss - all rights reserved


dem-cfd coupling manual 33
ROCKY

Figure 5.2: Schematic representation of 3


particles with volumes greater than CFD
cells, with volume fraction values cal-
culated by equation (5.11) represented
as cell colors. In this situation the vol-
ume fraction distribution is unrealistic –
there are cells with more than 100% of
solid volume while neighboring cells are
empty of solid volume.

an Eulerian approach, a subsequent step that redistributes the solid


phase is performed7 . Rocky provides two options to perform 7
The mapped information covers the
solid fraction, interaction forces and
this Lagrangian-to-Eulerian (L-E) mapping: uniform distribution and exchanged heat amounts.
volumetric diffusion.

5.3.1 Uniform distribution L-E mapping

The uniform distribution L-E mapping is used when the Mapping Method
is set to Uniform Distribution in the Fluent Two-Way Coupling
settings in Rocky. This method works by considering an averaging
radius value r∆ that defines a new cell, herein called "super-cell", as
shown in Figure 5.3. This super-cell is formed by the original cell and
all the neighboring cells having their centroids located inside a sphere
with radius equal to r∆ .

Figure 5.3: Schematic depiction of a


super-cell and its averaging radius.

In this mapping, the averaging is performed considering a uniform


distribution over the cells contained inside the super-cell. The idea
of the mapping is to distribute the volume fraction and interaction
forces of the central cell equally to all cells composing the super-cell.
In the uniform distribution L-E mapping, the averaging radius,

© 2019, esss - all rights reserved


dem-cfd coupling manual 34
ROCKY

r∆ , is a user input and may be given either as an absolute value


or as a fraction of the maximum particle diameter according to
the parameter Averaging Radius Type in the Fluent Two-Way Coupling
settings in Rocky.
In order to obtain the contribution to the volumetric fraction for all
cells forming a super-cell, the volume of all particles located at the
center of the super-cell are summed and divided by the total volume
of the super-cell:
Np
∑ p=1 Vp
α p,i = (5.12)
VSC
where α p,i is the volume fraction contribution for all cells inside the
super-cell, Vp is the volume of each particle, Np is the number of
particles that have their centroids located inside the center cell of the
super-cell and, finally, VSC is the volume of the super-cell, computed
according to:
Nc
VSC = ∑ Vc (5.13)
c =1

where Nc and Vc are, respectively, the number and the volume of the
cells composing a super-cell.
This procedure is repeated for all NT cells in the fluid domain, i.e.,
NT super-cells are generated and the contribution of each one of them
is added to obtain the final volume fraction of particles at each one of
the NT cells in the fluid domain. Figure 5.4 shows an example of the
addition of contributions of two overlapped super-cells.

Figure 5.4: Addition of contributions


of two overlapped super-cells when
uniform averaging scheme is used: (a)
In order to obtain the quantities that have to be sent to the CFD contribution for the super-cell delimited
by the red circle, (b) contribution for the
solver, this same averaging process is also performed for the total super-cell delimited by the orange circle,
interaction force as well as the heat trasfer rate with the fluid. (c) added contributions from the two
previous super-cells.

© 2019, esss - all rights reserved


dem-cfd coupling manual 35
ROCKY

5.3.1.1 Known issues

The uniform distribution L-E mapping tends to perform sub-optimally


under the following circumstances:

• Heterogeneous meshes. By default the super-cell is 3.5 times larger


than the largest cell of the mesh8 . If the mesh contains regions that 8
Using smaller values for the super-
cell perimeter may impact the mapping
are heavily refined compared to its largest cells, the computational quality at the largest cells of the mesh.
cost of searching cells inside the perimeter of the super-cell
increases at these regions due to the increased amount of cells
that may fit inside the perimeter.

• Internal boundaries. To determine which cells are part of the super-


cell, the search algorithm considers only the spatial location of
the cells and disregards its topological information, as the latter
is too computationally complex for the algorithm to handle. In
corner cases of cells separated by a thin internal boundary, this
may lead to volume fraction being "leaked" across the boundary by
the mapping process if the cells are small enough (and positioned
close enough) to eventually fit inside the super-cell at the same
time.

To overcome these issues, the volumetric diffusion L-E mapping


(section 5.3.2) has been implemented.

5.3.2 Volumetric diffusion L-E mapping

The volumetric diffusion L-E mapping is enabled when the Mapping Method
is set to Volumetric Diffusion in the Fluent Two-Way Coupling settings
in Rocky. This method works by solving iteratively a discretized
diffusion equation on the CFD mesh. With each iteration, calculated
amounts of exceeding solid volume (and other physical quantities
related to the particle phase as well) are exchanged between each cell
and its immediate neighbors. These amounts are proportional to the
difference of values between the cell and its immediate neighbors,
and are tuned by automatic detection of individual optimal diffusion
coefficients for each cell. Iterations are repeated until a target
maximum volume fraction is achieved over the mesh, or a maximum
number of iterations is reached.
The next sections explain the volumetric diffusion L-E mapping
method in detail.

© 2019, esss - all rights reserved


dem-cfd coupling manual 36
ROCKY

5.3.2.1 Formulation

The volumetric diffusion L-E mapping handles the problem of


smoothing the distribution of solid volume as a process of molecular
diffusion in a fluid medium. Fick’s second law of diffusion predicts
how diffusion causes the concentration α of a given substance to
change as time passes in an isotropic homogeneous medium:

∂α
= D ∇2 α (5.14)
∂t

where D is the diffusion coefficient for α in the medium.


In a discrete homogeneous mesh, the continuous Laplacian can be
expressed as a graph Laplacian. Then for any given cell c of the mesh,
equation (5.14) can be expressed as:

∂αc
= D ∑ (αn − αc ) (5.15)
∂t n∈N c

where αi is the substance concentration at cell i, and Nc is the set of


all cells that are immediate neighbors of c.
Equation (5.15) can be solved numerically by an explicit iterative
method. For this numerical solution a unit time step can be assumed,
since an accurate transient solution is not of interest – but only the
final state in which the concentration α has been diffused after enough
iterations. The explicit iterative form of equation (5.15) is:

αc,new = αc + D ∑ (αn − αc ) (5.16)


n∈Nc

Equation (5.16) offers an iterative approach to smooth an initial


distribution of volumetric fraction values in homogeneous meshes.
For heterogeneous meshes, in order to conserve the total amount
of solid in the solution domain as the iterative process evolves,
the diffusion coefficient D can be expressed as being inversely
proportional to the volume of the cell. Equation (5.16) can then
be adapted to:

K
αc,new = αc +
Vc ∑ (αn − αc ) (5.17)
n∈Nc

where K is a constant.
Equation (5.17) shows that the smaller the K value, the longer it
takes to smooth the solid volume field iteratively. Therefore K is
desired to be as large as possible; however a maximum limit must be
defined for K to guarantee the stability of the iterative process.

© 2019, esss - all rights reserved


dem-cfd coupling manual 37
ROCKY

It was determined heuristically that the following value of K


provides a satisfactory compromise between the speed and the
stability of the volumetric diffusion method in homogeneous meshes:

Vmin
Kmax = (5.18)
2Nmax

where Vmin is the minimum cell volume found among all cells of the
mesh and Nmax is the maximum number of immediate neighbors
found among all cells of the mesh9 . 9
This criteria was deduced by consider-
ing a regular unbounded mesh – the
For heterogeneous meshes, Rocky takes advantage of the fact that shape of the cells is irrelevant – con-
K does not need to be the same for all cells of the mesh10 . Instead, taining a regular distribution of volume,
where half of its cells are filled with
the only requirement is that each pair of cells that undergo a transfer the same amount of solid volume and
half are empty. After one iteration, this
of solid volume must agree on the same K. Rocky uses this fact to
value of K causes all cells of the mesh
automatically determine optimum diffusion constants for each cell as: to contain the same amount of solid
volume.
Vc
10
In fact, forcing a unique value of K
Kc = (5.19) tuned to the smaller cell of a heteroge-
2Nc neous mesh would impose an unnec-
essary burden to larger cells whose L-E
where Nc is the number of immediate neighbors of cell c. mapping could be performed faster with
their own larger values of K.
Then in equation (5.17), the same value of K must be used whenever
the same pair of cells undergo a solid transfer. Since each cell has its
own Kc – which happens to be the maximum allowable value for the
cell to undergo stable solid exchanges – the smaller value of the pair
must be chosen. Considering this requirement, equation (5.17) can be
rewritten to:

1
αc,new = αc +
Vc ∑ [min(Kn , Kc )(αn − αc )] (5.20)
n∈Nc

where Kn is the value of K for an immediate neighbor of the cell c.


Equation (5.20) is the basis of the volumetric diffusion L-E mapping
iterative algorithm implemented by Rocky. With each iteration, Rocky
applies this algorithm to all cells of the CFD domain to calculate new
volume fraction values11 . Iterations are repeated until the stop criteria 11
Section 5.3.2.3 explains how this pro-
cess works for other physical quantities.
described in section 5.3.2.2 are met.

5.3.2.2 Operation

Prior to starting a simulation in which the L-E mapping method has


been chosen, Rocky calculates and stores the following information:

• Sets of immediate neighbors of all cells of the mesh.

• Diffusion constants (K) of all cells of the mesh according to equation


(5.19).

© 2019, esss - all rights reserved


dem-cfd coupling manual 38
ROCKY

Every time the L-E mapping must be performed during the


simulation, equation (5.20) is applied to all cells of the mesh to
achieve a smoothed state of the solid volume distribution. This
action comprises one iteration of the method. Rocky then decides if
more iterations are required based on the following parameters that
are set by the user under the Fluent Two-Way Coupling settings:

• Maximum Volume Fraction Target – Desired maximum volume fraction


over the mesh (default 65%). After each iteration the maximum
volume fraction is queried over the mesh and, if no greater than
the Maximum Volume Fraction Target , the mapping is declared finished.

• Maximum Iterations – Maximum number of iterations that must be


performed (default 500). If the number of performed iterations
exceeds this value, the iterative process terminates in order for the
solid volume distribution to be sent to Fluent.

Figure 5.5 illustrates the application of the volumetric diffusion


L-E mapping with a default Maximum Volume Fraction Target of 65% to a
regular two-dimensional Cartesian mesh containing in its center cell
a particle whose volume is two times greater than the cell’s volume.
After two iterations the mapping is complete with a maximum solid
volume fraction of 56% over the whole mesh.

Figure 5.5: Representation of 2 iterations


of the volumetric diffusion L-E mapping
over a regular two-dimensional Carte-
sian mesh containing in its center cell
a particle whose volume is two times
greater than the cell’s volume.

5.3.2.3 Mapping other physical quantities

Section 5.3.2.1 describes how the volumetric diffusion L-E mapping


smooths a field of solid volume fraction values. A similar approach is
considered for mapping force- and thermal-related quantities, with
the following considerations over the exchange of values between a
cell and its neighbors.
Forces are weighted by the specific mass of the fluid contained
in the cells at the moment the mapping is performed. The iterative

© 2019, esss - all rights reserved


dem-cfd coupling manual 39
ROCKY

equation for mapping forces12 is: 12


A similar equation applies to mapping
time derivatives of forces.
  
1 Fn Fc
Fc,new = Fc +
Vc ∑ min(Kn ρn , Kc ρc )
ρn

ρc
(5.21)
n∈Nc

where Fi is a force component13 at cell i. 13


The L-E mapping is performed for
each force component independently.
Heat amounts are weighted by the volumetric heat capacity of the
fluid contained in the cells at the moment the mapping is performed.
The iterative equation for mapping heat amounts14 is: 14
A similar equation applies to mapping
time derivatives of heat amounts.
  
1 Qn Qc
Qc,new = Qc +
Vc ∑ min(Kn cn ρn , Kc cc ρc )
cn ρn

cc ρc
(5.22)
n∈Nc

Lastly, the volumetric diffusion iterations of force- and thermal-


related quantities are coupled with the iterations of the solid volume
mapping. Consequently there are no maximum target values for these
quantities.

5.3.2.4 Selection of fluid cell zones

The volumetric diffusion L-E mapping can be further tuned in two-


way coupled setups that contain more than one fluid zone by selecting
what fluid zones are part of the mapping15 . 15
The selection of the fluid zones is
performed in Rocky via the Fluent Two-
The L-E mapping does not happen in fluid zones that are not Way settings, Zones and Interfaces tab,
selected, increasing the computational efficiency of simulations where Coupling Fluid Cell Zones section.

particles are known to never reach such zones.

5.3.2.5 Mapping across sliding interfaces

Setups that contain moving meshes connected through sliding


interfaces are supported by the volumetric diffusion L-E mapping
algorithm16 . 16
When setting up the simulation, the
user must select which interfaces the
In order to correctly apply the L-E mapping across sliding volumetric diffusion L-E mapping is
interfaces, Rocky keeps a further list of neighbors that are in contact allowed to map across. The selection
of the interfaces is performed in Rocky
through the interfaces, as well as updated diffusion constants for via the Fluent Two-Way settings, Zones
and Interfaces tab, Mapping Cell Zone
nearby cells17 . The volumetric diffusion L-E mapping then honors the
Interfaces section.
user selection of interfaces by adding this temporary set of neighbors This information is constantly up-
17

to the static set of neighbors that was queried initially for all meshes dated during the simulation as meshes
move relative to each other.
individually.
Consequently the volumetric diffusion L-E mapping distributes
solid volume (and other physical quantities as well as per section
5.3.2.3) across the sliding interfaces that were selected by the user,
while interfaces that were not selected behave as barriers to the

© 2019, esss - all rights reserved


dem-cfd coupling manual 40
ROCKY

mapping. For illustrating this concept, consider Figure 5.6, which


represents a particle near to the sliding interface between two meshes
that rotate relative to each other.

Figure 5.6: A particle near the sliding


interface between two rotating meshes.

If the sliding interface is selected under Mapping Cell Zone


Interfaces in the UI, the volumetric diffusion L-E mapping expands
solid volume towards the mesh at the other side of the interface as well
(Figure 5.7). This is the indicated setting when particles are intended
to move freely through the sliding interface in the simulation, as
the fluids at both sides of the interface are supposed to be equally
disturbed by the solid volume.

Figure 5.7: Resulting volume fraction


distribution when the situation repre-
sented by Figure 5.6 is mapped with
the sliding interface selected under Map-
ping Cell Zone Interfaces.

If the sliding interface is not selected under Mapping Cell Zone


Interfaces in the UI, the mapping does not expand solid volume
across the interface (Figure 5.8). This is the indicated setting when
the interface is intended to act as a free gateway for fluids in the CFD
domain while blocking particles in the DEM domain18 . 18
Typical examples are DEM-CFD cou-
pled simulations containing geometries
that act as sieves.

© 2019, esss - all rights reserved


dem-cfd coupling manual 41
ROCKY

Figure 5.8: Resulting volume fraction


distribution when the situation repre-
sented by Figure 5.6 is mapped with
the sliding interface not selected under
Mapping Cell Zone Interfaces.

5.4 Two-way coupling algorithm


The way in which coupling is implemented allows both solvers,
the DEM solver and the CFD solver, to work in parallel, using N
processors for the fluid phase solution and M processors for the
particulate phase solution (or using GPU processing to solve the DEM
part).
Figure 5.9 shows schematically the two-way coupling algorithm.
The following operation sequence is performed during the particulate
system solution:

1. DEM solver time step calculation.

2. Initial fluid flow field calculation (velocity, pressure and physical


properties) on the CFD solver and transfer to the DEM solver.

3. Particulate phase volume fraction and initial interaction terms


calculation (interaction forces and heat transfer rates) on the DEM
solver and transfer to the CFD solver.

4. CFD solver time step correction in order to have an integer multiple


of the DEM solver time step.

5. Initial solid phase field update on CFD solver.

6. Parallel execution of one CFD solver time step and n DEM solver
time steps.

7. Semi-implicit transfer of the interaction forces and explicit transfer


of the heat transfer rate from the DEM solver to the CFD solver.

8. Velocity, pressure and physical properties in each cell transfer from


CFD solver to the DEM solver.

9. Repetition of the process until reaching the total simulation time.

© 2019, esss - all rights reserved


dem-cfd coupling manual 42
ROCKY

Figure 5.9: Coupling algorithm.

© 2019, esss - all rights reserved


dem-cfd coupling manual 43
ROCKY

6 Best practices
6.1 One-way coupling simulation
This section applies to one-way steady-state DEM-CFD coupling with
Fluent.

6.1.1 Mesh generation

Standard CFD recommendations for mesh generation are also applied


to DEM-CFD coupling cases using the one-way coupling approach.
For example, to avoid poor-quality cells and large variations in
adjacent cells volumes, apply denser meshes in zones with strong
quantity gradients, and so on.
In the one-way coupling approach, it is assumed that the particles
do not affect the flow field, as there is no information transfer from
Rocky into the CFD solution. Because of this, there are no additional
recommendations nor limitations imposed in the CFD simulation by
the coupling scheme itself. Therefore, it is important that the user
focus on providing the best CFD simulation possible.
It is important to highlight that the mesh size has an impact on
the simulation cost, since the finer the mesh, the bigger the number
of cells Rocky will search in order to find particles on which to
perform calculations. This search process is done in order to obtain
local pressure, fluid velocity, and temperature for forces and heat
exchanges with the particle.

6.1.2 Simulation setup

The current Rocky formulation calculates buoyancy and additional


pressure gradient forces based on pressure derivatives extracted from
the CFD solutions, which are exported from ANSYS Fluent. It is
recommended, even in single-phase cases with constant properties,
that gravity is turned on, and the reference density and pressure
location are defined manually (assuming there are no boundary

© 2019, esss - all rights reserved


dem-cfd coupling manual 44
ROCKY

conditions defining pressure references). In the momentum equation


solved by ANSYS Fluent1 : 1
ANSYS (2013b). Fluent Theory Guide.
ANSYS, Inc., Canonsburg, PA

(ρv) + ∇ · (ρvv) = −∇ p0 + ∇ · τ̄¯ + ρg (6.1)
∂t

the solution of the modified pressure field represents the static


pressure variation due to the flow and the hydrostatic pressure at the
reference density, as given by:

p 0 = p − ρ0 g · x (6.2)

By setting the reference density to 0 kg/m3 , all the hydrostatic


contributions will be explicitly accounted within the pressure solution
and will be transferred to Rocky when you export the data. If the
simulation is stable, you may also set the reference pressure to 0
Pa. Although it is not necessary, it allows you to have the absolute
pressure including hydrostatic pressure to post-process within ANSYS
Fluent and then later, within Rocky after the Fluent results are
imported.
Gravity, reference (operating) pressure, pressure reference location
(if needed), and the specified operating density (reference density) can
all be set up in the Operating Conditions dialog from the Cell Zone
Conditions panel inside ANSYS Fluent, as shown in Figure 6.1. Please
check the ANSYS Fluent documentation2 to find further information 2
ANSYS (2013b). Fluent Theory Guide.
ANSYS, Inc., Canonsburg, PA
on the modified pressure and reference (operating) values. Some
additional care should be taken when defining pressure boundary
conditions (such as pressure outlets); in this way, specification of the
correct pressure distribution (instead of a constant value) may be
needed.
If the thermal model is activated, an additional equation for the
energy balance is solved along with the equations governing the
motion of the particle. The temperature variation of a particle is a
function of the heat transfer that occurs during the contact with other
particles or walls, and the heat transfer between particles and the
fluid phase.
The heat transfer rate between a particle and fluid is calculated
using the average convective heat transfer coefficient, the particle
area, and the temperature difference between the particle and the
fluid. The average convective heat transfer coefficient, by its turn, is
computed based upon the Nusselt number that is calculated using the
selected correlation within Rocky3 . These correlations are a function 3
Refer to chapter 4 for details on how
heat transfer is accounted by Rocky
of the particle Reynolds number and the Prandlt number. In order DEM-CFD coupling.

© 2019, esss - all rights reserved


dem-cfd coupling manual 45
ROCKY

Figure 6.1: Operating Conditions panel


in ANSYS Fluent version 18.2.

to compute these numbers, Rocky needs fluid properties, including


fluid-specific heat and thermal conductivity.
These thermal properties are not exported to Fluent when a
constant value is prescribed within Fluent setup. Therefore, when
running cases with the Rocky Thermal Model enabled and constant
values set for fluid thermal properties, specific heat and thermal
conductivity both need to be changed from constant to polynomial
using the constant value as the first polynomial value, as shown in
Figure 6.2.

Figure 6.2: Fluid material panel in


ANSYS Fluent version 18.2.

© 2019, esss - all rights reserved


dem-cfd coupling manual 46
ROCKY

6.2 Two-way coupling simulation


6.2.1 General information

In a DEM-CFD two-way coupling simulation, the particles are part


of the fluid flow and will affect it in a two-way interaction, i.e., the
particle movement is affected by the interaction with other particles
and the fluid around it while the flow is also affected by the particle
presence.
On the Rocky side, the fluid flow will exert forces upon the particle,
including pressure gradient (including buoyancy) force, drag force,
and virtual mass force (optional). On the CFD side, the reaction of
the forces upon the particles will be applied over the fluid phases.
If geometries are shared between the two programs and movement
of these geometries are also desired, Moving Meshes can be enabled
for the geometry on the Fluent side and those settings will be
automatically converted into a Rocky Motion Frame upon importing
the CAS file (Figure 6.3). In this way, a consistent motion for the
shared eometry can be achieved between the two programs.

Figure 6.3: Motion settings on the Fluent


side (left) are automatically converted
into a new Fluent Motion Frame on the
Rocky side (right)

The scheme used in the Rocky DEM-CFD two-way coupling


module allows both solvers to run in parallel, which can considerably
decrease simulation time if the solvers do not compete for resources.
During setup within the Rocky coupling module interface, you can
control the number of processors required by each solver to avoid
resource competition. Also, since Rocky can make use of GPU
and Multi-GPU processing, an alternative to increasing the coupling
performance is to set the CFD solver to run on CPU cores and set
Rocky to run on the GPU cards.

© 2019, esss - all rights reserved


dem-cfd coupling manual 47
ROCKY

6.2.2 Mesh generation

The Rocky DEM-CFD two-way coupling module uses a neighbor-cell


averaging procedure to avoid high-volume fractions in a certain cell
of the CFD domain, distributing the volume of the particles that are
positioned at the cell boundaries among the adjacent cells. Thus, this
methodology is suitable for dealing with a large number of particles
within a CFD cell and not for dealing with particles larger than the
CFD cell.
Therefore, it is recommended that the mesh size be larger than the
largest particle size throughout the simulation domain. Some regions
with detailed geometry and strong gradients can have smaller cells
but the coupling results in these regions will be less accurate.
Traditional CFD recommendations of mesh size for accurate
boundary-layer capturing can usually be ignored in this case. The
two-way coupling module will generally be applied to cases with
dense particle flows where the particle-fluid interaction will usually
be more significant than the boundary-layer effects.

6.2.3 CFD simulation setup

The CFD case file to be provided to Rocky for a two-way coupling


simulation is a standard Eulerian multiphase simulation setup, where
one of the phases should be defined as the particulate phase. Some
specific configurations for the model definitions and solver numerics
are presented below.

6.2.3.1 Multiphase model

As described above, the Eulerian approach is used within the CFD


model to perform the coupling between the two solvers. Hence, the
Eulerian Multiphase model should be selected in order to set the
coupled simulation. The number of Eulerian phases should be equal
to the number of fluid phases + 1, where the additional phase is the
particle phase. This means that, even in cases with only one fluid
phase, the simulation must be set as an Eulerian multiphase case,
since the particles will be represented by an additional phase.
The particle phase should be set as a secondary phase and the
momentum exchange terms between the fluids and particles are
calculated by Rocky. Therefore, the user does not need to set
momentum transfer coefficients between the particle and fluid phases.

© 2019, esss - all rights reserved


dem-cfd coupling manual 48
ROCKY

However, in the case of more than one fluid phase, the phase
interaction between fluid phases should be properly defined during
this step. The material used in the particle phase must be used only
for this phase. Phase characteristics, inter-phase transfer coefficients,
and material properties for the particles phase will be defined by the
Rocky coupling.

6.2.3.2 Boundary conditions and initialization

Since Rocky is responsible for the particulate phase solution, the solid
volume fraction and velocities information should come from the
DEM solver. Thereby, the particulate phase velocities and volume
fractions should be set equal to 0 in all boundaries.
Also, the particle volume fraction and velocities should be set to
0 during the case initialization. Particle volume fractions will be
updated during the coupling initialization process.
It is also possible to initialize the coupling using a Fluent data file
with an initial flow field. In this case also the volume fractions of
the particle phase must be 0 throughout the domain. When Mesh
Motion is used, the Reference Frame for initialization should be set to
Absolute.

6.2.3.3 Numerics

Phase Coupled SIMPLE method should be used as pressure-velocity


coupling method. First Order should be used as the transient scheme.
Be careful to choose a reasonable time step. Time step size will be
updated after coupling starts, in order to be an integer multiple of
the Rocky time step.

6.2.3.4 Solution data export

By default, Rocky does not export simulation data without specific


input from the user. In order to enable transient solution files, you
must specify that Rocky export the simulation data as usually done in
Fluent. Absolute paths should be used when exporting solution data.

6.2.4 Rocky simulation setup

© 2019, esss - all rights reserved


dem-cfd coupling manual 49
ROCKY

6.2.4.1 Particle size scale factor

When the Experimental (Beta) Features checkbox is enabled (located


on the Rocky Options | Preferences dialog), the Particle Size Scale Factor
becomes available. This setting enables the use of a particle equivalent
diameter this many times smaller to compute drag and non-drag
forces on particles, as well as to compute convective heat transfers
between particles and fluids.
In this way, the user can reduce computational costs by increasing
particle size on the DEM simulation (consequently reducing the
number of particles and contacts), while the interaction forces and
heat transfer from the CFD coupling are correctly computed for
particles with the original equivalent diameter. Volume and mass are
also respected when using this factor so that the number of particles
with smaller sizes are correctly calculated.

© 2019, esss - all rights reserved


dem-cfd coupling manual 50
ROCKY

7 Bibliography
ANSYS (2013a). CFX Solver Theory Guide. ANSYS, Inc., Canonsburg,
PA.

ANSYS (2013b). Fluent Theory Guide. ANSYS, Inc., Canonsburg, PA.

Batchelor, G. K. and O’Brien, R. W. (1977). Thermal or electrical


conduction through a granular material. Proceedings of the Royal
Society of London, 355:313–333.

Bejan, A. (2013). Convection Heat Transfer. Wiley.

Bergman, T., Incropera, F., DeWitt, D., and Lavine, A. (2011).


Fundamentals of Heat and Mass Transfer. Wiley.

Crowe, C., Schwarzkopf, J., Sommerfeld, M., and Tsuji, Y. (2011).


Multiphase Flows with Droplets and Particles, Second Edition. Taylor &
Francis.

Cundall, P. A. and Strack, O. D. L. (1979). A discrete numerical model


for granular assemblies. Geotechnique, 29:47–65.

DallaValle, J. M. (1948). Micromeritics: the technology of fine particles.


Pitman Pub. Corp.

Dandy, D. S. and Dwyer, H. A. (1990). A sphere in shear flow at finite


Reynolds number: effect of shear on particle lift, drag, and heat
transfer. J. Fluid Mech, 216:381–410.

Dennis, S., Singh, S., and Ingham, D. (1980). The steady flow due to a
rotating sphere at low and moderate Reynolds numbers. Journal of
Fluid Mechanics, 101:257–279.

Di Felice, R. (1994). The voidage function for fluid-particle interaction


systems. International Journal of Multiphase Flow, 20:153–159.

Drew, D. (1993). Mathematical modeling of two-phase flow. Annual


Review of Fluid Mechanics, 15:261–291.

E, M. (2006). Particles, Bubbles And Drops: Their Motion, Heat And Mass
Transfer. World Scientific Publishing Company.

© 2019, esss - all rights reserved


dem-cfd coupling manual 51
ROCKY

Feng, Y. and Yu, A. (2004). Assessment of model formulations in


the discrete particle simulation of gas-solid flow. Industrial &
Engineering Chemistry Research, 43:1713–1728.

Ganser, G. H. (1993). A rational approach to drag prediction of


spherical and nonspherical particles. Powder Technology, 77:143–152.

Gidaspow, D. (1994). Multiphase Flow and Fluidization. Academic Press,


San Diego.

Gidaspow, D. (2012). Multiphase Flow and Fluidization: Continuum and


Kinetic Theory Descriptions. Elsevier Science.

Gidaspow, D., Bezburuah, R., and Ding, J. (1993). Hydrodynamics of


circulating fluidized beds: Kinetic theory approach. In Engineering
foundation conference on fluidization, volume 7, pages 75–82, Brisbane,
Australia.

Goldschmidt, M., Beetstra, R., and Kuipers, J. (2002). Hydrodynamic


modelling of dense gas-fluidised beds: comparison of the kinetic
theory of granular flow with 3d hard-sphere discrete particle
simulations. Chemical Engineering Science, 57:2059–2075.

Gosman, A. D. and Ioannides, E. (1983). Aspects of computer


simulation of liquid-fueled combustors. Energy Journal, 7:482–490.

Gunn, D. (1978). Transfer of heat or mass to particles in fixed and


fluidised beds. Int. J. Heat Mass Transfer, 21:467–476.

Gunn, D. and De Souza, J. F. C. (1974). Heat transfer and axial


dispersion in packed beds. Chemical Engineering Science, 29:1363–
1371.

Haider, A. and Levenspiel, O. (1989). Drag coefficient and terminal


velocity of spherical and nonspherical particles. Powder Technology,
58:63–70.

Hoomans, B., Kuipers, J., Briels, W., and Van Swaaij, W. (1996).
Discrete particle simulation of bubble and slug formation in a two-
dimensional gas fluidised bed: a hard-sphere approach. Chemical
Engineering Science, 51:99–118.

Hoomans, B., Kuipers, J., and Van Swaaij, W. (2000). Granular


dynamics simulation of segregation phenomena in bubbling gas-
fluidised beds. Powder Technology, 109:41–48.

© 2019, esss - all rights reserved


dem-cfd coupling manual 52
ROCKY

Huilin, L. and Gidaspow, D. (2003). Hydrodynamics of binary fluidiza-


tion in a riser: Cfd simulation using two granular temperatures.
Chemical Engineering Science, 58:3777–3792.

Hutchinson, B. and Raithby, G. (1986). A multigrid method based on


the additive correction strategy. Numer. Heat Transfer, 9:511–537.

Ishii, M.; Mishima, K. (1984). Two-fluid model and hydrodynamic


constitutive relations. Nuclear Engineering and Design, 82:107–126.

Lewis, D. A. and Davidson, J. F. (1985). Pressure drop for bubbly


gas-liquid flow through orifice plates and nozzles. Chem. Eng. Res.
Des., 63:149–156.

Littman, H. and Silva, D. E. (1970). Gas-particle heat transfer


coefficient in packed beds at low Reynolds number. In Proceeding of
4th International Heat Transfer Conference, Versilles, France.

Mei, R. (1992). An approximate expression for the shear lift force on


a spherical particle at finite Reynolds number. International Journal
of Multiphase Flow, 18:145–147.

Miyauchi, T., Kikuchi, T., and Hsu, K.-H. (1976). Limiting sherwood
number of sphere packed beds by electrical method. Chemical
Engineering Science, 31:493 – 498.

Molerus, O. (1997). Heat transfer in moving beds with a stagnant


interstitial gas. International Journal of Heat and Mass Transfer, 40:4151–
4159.

Paladino, E. E. (2005). Estudo do escoamento multifásico em medidores de


vazão do tipo pressão diferencial. PhD thesis, Universidade Federal de
Santa Catarina, SC.

Patankar, S. (1980). Numerical Heat Transfer and Fluid Flow. Series in


computational methods in mechanics and thermal sciences. Taylor
& Francis.

Pritchard, P. J. (2010). Fox and McDonald’s Introduction to Fluid


Mechanics, 8th Edition. John Wiley & Sons.

Saffman, P. G. (1965). The lift on a small sphere in a slow shear flow.


J.Fluid Mech, 22:385–400.

Saffman, P. G. (1968). Corrigendum to: The lift on a small sphere in a


slow shear flow. J.Fluid Mech, 31:624.

© 2019, esss - all rights reserved


dem-cfd coupling manual 53
ROCKY

Tsuji, Y., Kawaguchi, T., and Tanaka, T. (1993). Discrete particle


simulation of two-dimensional fluidized bed. Powder Technology,
77:79–87.

Turner, G. A. and Otten, L. (1973). Values of thermal (and other)


parameters in packed beds. Ind. Eng. Chem. Proc. Des. Dev., 12:417–
424.

Vargas, W. L. and McCarthy, J. J. (2002). Stress effects on the


conductivity of particulate beds. Chemical Engineering Science,
57:3119–3131.

Vasquez, S. A. and Ivanov, V. (2000). A phase coupled method for


solving multiphase problems on unstructured meshes. In Fluids
Engineering Division Summer Meeting, Boston, Massachusetts. ASME.

Xu, B. and Yu, A. (1997). Numerical simulation of the gas-solid


flow in a fluidized bed by combining discrete particle method
with computational fluid dynamics. Chemical Engineering Science,
52:2785–2809.

Xu, B., Yu, A., Chew, S., and Zulli, P. (2000). Numerical simulation
of the gas-solid flow in a bed with lateral gas blasting. Powder
Technology, 109:13–26.

Ye, M., Van der Hoef, M., and Kuipers, J. (2004). A numerical study of
fluidization behavior of geldart a particles using a discrete particle
model. Powder Technology, 139:129–139.

Ye, M., Van der Hoef, M., and Kuipers, J. (2005). From discrete
particle model to a continuous model of geldart a particles. Chemical
Engineering Research and Design, 83:833–843.

Zabrodsky, S. S. (1966). Hydrodynamics and Heat Transfer in Fluidized


Beds. MIT Press, Cambridge, MA.

© 2019, esss - all rights reserved

You might also like