Tips and Tricks in

Multiphase Flows
14. 5 Release

Multiphase Flow Modeling

In ANSYS CFX
© 2013 ANSYS, Inc. 13-1 Release 14.5
 Overview
• Boundary Conditions
• Dense Particle Effects
• Turbulence in Multiphase Flows
• Particle Induced Turbulence
• Interfacial Area Modeling
• Convergence Issues
• Expert Parameters and CCL Settings

© 2013 ANSYS, Inc. 13-2 Release 14.5

 Boundary Conditions
• There are cases where one phase may be
Gas Head Space
continuous and the other is batch (i.e. for
a bubble column, the liquid is batch Liquid Level
while the gas bubbling through it flows
continuously)
Liquid
• For such systems, the liquid level could Continuous
take on any value depending upon the Phase

initial charge of liquid to the system

• What type of boundary conditions can
we use to preserve the initial charge of
liquid?

© 2013 ANSYS, Inc. 13-3 Release 14.5

 Degassing Boundary
• A degassing boundary is a special type of outlet
boundary condition. The dispersed phase sees it as
an outlet while the continuous phase sees it as a
free slip wall
Degassing
• Valid for both fluid and solid dispersed phases Condition

• No pressure level is set; the condition removes

whatever dispersed phase arrives at the boundary
• Normally the continuous phase is not removed at a
degassing boundary, but for an initial guess that has
zero volume fraction for the dispersed phase, some Computational
must be removed to make room for the entering Domain

dispersed phase
• The geometry for the simulation would be
terminated at the expected position of the free
surface where a degassing condition would be
applied
© 2013 ANSYS, Inc. 13-4 Release 14.5
 Headspace Condition
• Another approach is to incorporate the gas Outlet
headspace in the domain and set the initial Condition
condition for volume fraction to reflect the amount Gas Head Space
of the liquid initially charged
Liquid Level
• This type of boundary would be a normal outlet or
opening, typically with a specified pressure
• As long as the convergence is reasonable, no liquid
should leave the outlet and the initial charge of Liquid
Continuous
liquid should be preserved (see Workshop 1) Phase

• The initial condition/guess should have the

correct hydrostatic head for the specified
initial volume fraction distribution
• This approach does allow for the free surface
to be dynamic (i.e. not flat) and does impose
a pressure level on the simulation

© 2013 ANSYS, Inc. 13-5 Release 14.5

 Dense Particle Effects

• Drag laws are often obtained by correlation of data taken for single
particle or dilute particle systems
• As the volume fraction of the dispersed phase increases, the drag will
begin to be affected by concentration effects
– Small bubbles tend to rise more slowly at high void fractions due to an increase
in the mixture viscosity increased drag
– Large bubbles tend to rise more quickly at high void fractions due to being
dragged along by the wakes of other bubbles decreased drag

– Schiller Naumann:
Wen Yu (rc > 0.8):

Gidaspow :

– Volume Fraction Correction Exponent (Grace, Ishii Zuber):

• Small Bubbles: p < 0, p = -1, -0.5
• Large Bubbles: p > 0, p = 2, 4

© 2013 ANSYS, Inc. 13-6 Release 14.5

 Turbulence Modeling in Multiphase Flows
• For continuous-dispersed multiphase flows, a typical combination is:
– k-e (or SST) for the continuous phase
– Dispersed phase zero equation for the dispersed phase:

ntd = ntc/s mtd = (rd/rc) mtc/s

– where s is an eddy viscosity turbulent Prandtl number relating the dispersed phase
kinematic eddy viscosity to the continuous phase kinematic eddy viscosity (the default
value is 1, can be changed).
– Dispersed phases can also be laminar
• For continuous-continuous multiphase flows, all models available for single-phase
calculations are also available for continuous phases in multiphase calculations,
with the exception of the LES model
• In some cases, you may wish to solve for a homogeneous turbulence field for
an inhomogeneous multiphase flow (i.e. for any situation where the phases tend
to separate out or stratify). This is different than using the same turbulence
model for the two phases since that will solve for two turbulence fields instead of
a single one
© 2013 ANSYS, Inc. 13-7 Release 14.5
 Particle Induced Turbulence
• Sato successfully modeled Particle Induced Turbulence for
bubbly flows using:

mtc = mts + mtp

where mts = the normal shear induced viscosity and mtp is an additional
particle induced eddy viscosity defined as:

mtp = Cmp rc rd dP | uD – uC |

Cmp = 0.60 (default value, may be adjusted)

© 2013 ANSYS, Inc. 13-8 Release 14.5

 Interfacial Area Modeling: Particle Model
• For the Particle Model, the surface area per unit volume is then calculated by assuming that
phase b is present as spherical particles of Mean Diameter db:

Aab = 6rb/db

• This simple model is modified for robustness purposes in two ways:

– rb is clipped to a minimum value to ensure the area density does not go exactly to 0.
– For large rb when the assumption of b being dispersed is invalid, the area
density is decreased to reflect the fact that it should lead to zero as rb tends to 1.

• With these modifications, the area density for the particle model is implemented as

Aab = 6r~b/dbwhere: r~b=

• Default values of rmax and rmin for interfacial area are 0.8 and 10-7, respectively. In some
cases, it may be appropriate to use a different value for rmin; for example, increasing it to
10-3 provides a crude nucleation model for subcooled liquid boiling. rmin is controlled by the
CCL parameter Minimum Volume Fraction for Area Density, rmax by Minimum Volume
Fraction for Area Density (Fluid Pair Models).
© 2013 ANSYS, Inc. 13-9 Release 14.5
 Interfacial Area for Non-Drag Forces
• For non-drag forces, the solver uses a slightly different formulation of area density called the
Unclipped Interfacial Area Density.
• In this formulation, the area density is permitted to go to zero, i.e.,
rmin = 0 in the previous slide
• In addition, the area density is reduced more aggressively as the dispersed phase volume
fraction becomes large:

n is set by the CCL parameter Area Density Reduction Exponent (Fluid Pair Models)
© 2013 ANSYS, Inc. 13-10 Release 14.5
 Advanced Models for Interfacial Area
• CFX provides three built-in prescriptions for specific interfacial area:
– Particle Model (discussed) on previous slide
– Free Surface Model (computed from magnitude of the volume fraction gradient)
– Mixture Model: Aab = rarb/dab where dab is the interface length scale
• The mixture model is the method of choice when the continuous phase is
ill-defined or can switch via phase inversion. Note that dab can be an
expression!
• For example, suppose you have oil-water flow in a separator for which
you may have either water droplets of diameter da in continuous oil
(phase b), or oil droplets of diameter b in continuous water (phase a), in
the limits ra 0, rb 0, respectively. A simple model for interfacial area
density that has the correct behavior in these two limits is given by:

6 rarb
Aab = radb + rbda

which can be implemented by setting dab = (radb+radb)/6

© 2013 ANSYS, Inc. 13-11 Release 14.5

 User Defined Interfacial Area
• It is possible to directly prescribe the interfacial area which will take
precedence over whatever is set for the Interphase Transfer Model
• This option isn’t in the CFX-Pre GUI, but it can be done by adding the
appropriate CCL from the command editor or from the Solver Manager
definition file editor
2.

3.

1.

© 2013 ANSYS, Inc. 13-12 Release 14.5

 Bounding Interfacial Area
• It can be important to ensure that the
interphase drag does not go to zero when the
volume fraction goes to 0 or 1 to avoid
spurious velocities in the dilute phase.

• The clipping that’s performed for the particle

model was discussed on a previous slide

• The mixture model also has a 'minimum

volume fraction for area density‘

• The free surface model has a 'Maximum

Length Scale for Area Density'; the inverse of
this quantity is a minimum area density.

• It can be helpful to set both the Minimum

Volume Fraction for a phase as a parameter
to control the minimum area density, i.e. the
Minimum Volmume Fraction for Area Density
© 2013 ANSYS, Inc. 13-13 Release 14.5
 Convergence Issues
• Choose time step wisely for both steady and transient runs. Normally, this is the single
“switch” that we suggest users adjust to improve convergence
• Define Linearization Coefficients for Sources - For problems where source or sink terms are
significant, defining linearization coefficients can help greatly, even if they are not exact

Source Coefficient Added

© 2013 ANSYS, Inc. 13-14 Release 14.5

 Convergence Issues
• Consider boundary conditions and combinations carefully. Sometimes the artificial walls
that may be constructed to prevent reverse flow at an outlet may completely block the flow
and cause the solver to die prematurely. An opening may be a better choice for such
simulations, at least to get it past the early stages, at which
point the outlet condition could be restored

For example, consider a thermal phase change model where the initial condition is all liquid
except for a minimum volume fraction of gas set to 0.001 to provide a
“seeding” for thermal phase change to take place. If the liquid is initially subcooled,
that vapor may condense, which could tend to entrain flow from the outlet.

When this case was run with an outlet boundary condition, artificial walls blocked off all
flow and the solver crashed early on.

If the outlet boundary was changed to an opening, some flow was initially entrained, but as
time went on and vaporization began to occur, the flow became
universally outward at the exit and eventually the boundary could be switched
back to an outlet.

© 2013 ANSYS, Inc. 13-15 Release 14.5

 Expert Parameters and CCL Settings
• Turning off Redistribute in Rhie Chow (CCL)
redistribute in Rhie Chow = f

– Done at the individual force level under momentum transfer (i.e. turbulent
dispersion force, lift force, etc.) Normally, redistributing a force in Rhie
Chow helps avoid wiggles in the velocity field, but there are cases
(especially for the turbulence dispersion force) where it helps to turn it off as
in Workshop #4.

• Linear Solver Expert Parameters

solver target reduction fluids = 1e-4 (default value 0.10)

– The linear solver for the coupled mass-momentum system iterates until the
final RMS residual for the continuity equation is below this number times the
initial RMS continuity residual. It can sometimes be useful to drive the linear
solver to a tighter tolerance as in Workshop #3.

max solver its fluid = 50 (default value = 40 in CFX-12)

– Setting this parameter to force the linear equation solver to work harder can
be helpful as in Workshop #3.

© 2013 ANSYS, Inc. 13-16 Release 14.5

© 2013 ANSYS, Inc. 13-17 Release 14.5