WM 2011 Conference, February 27- March 3, 2011.

Phoenix, AZ

Numerical Simulations of Pulsed-Air Mixing Technology using Multiphase Computational

Fluid Dynamics Methods

Rinaldo G. Galdamez, Stephen Wood, Seckin Gokaltun

Florida International University, Applied Research Center, Miami, Florida 33174

ABSTRACT

COMSOL Multiphysics and OpenFOAM Computational Fluid Dynamics (CFD) methods are
used to create a computational model of a pulsed-air mixer. First, results are provided for a
benchmark problem with a single bubble rising due to buoyancy with density and viscosity ratios
of 10. The numerical results are verified using the bubble circularity and the terminal bubble
velocity at different meshing levels which is captured within 10% accuracy compared with the
benchmark simulation.

After the verification of the numerical methods, the flow characteristics created by a pulsed-air
mixer in a 1/12-scale tank based on Hanford double-shell tank dimensions are simulated using
the proposed CFD methods. This scaled experiment was carried out by Pacific Northwest
National Laboratory (PNNL) in 1996. The peak fluid velocities produced by the pulsed-air
mixing plate are compared against the PNNL experimental data at various locations away from
the plate. The simulations show that the proposed methods can predict the performance of the
pulsed-air system accurately and they can be used as computational tools for scaling up the
design of future pulsed-air mixing implementations at Department of Energy (DOE) waste tanks.

INTRODUCTION

As a result of atomic weapons production, millions of gallons of radioactive waste were

generated and stored in underground tanks at various U.S. Department of Energy (DOE) sites.
DOE is currently in the process of transferring the waste from single-shell tanks to double-shell
tanks. In order to decrease the probability of a plug occurring during the transfer process, several
tank mixing techniques have been devised and evaluated at these sites. One of these techniques,
pulsed-air mixing, consists of the injection of discrete pulses of air or inert gas by means of
accumulator plates located at the bottom of the tank. These pulses generate large bubbles that
rise due to buoyancy and create circulation in the surrounding fluid which contributes to mixing
of the contents in the tank. Pulsed-air mixers are operated by controlling the pulsing frequency,
pulse duration, type of accumulator plates and gas pressure. In comparison with other mixing
techniques, the main advantages of the pulsed-air mixers are the low cost, durability, and easy
maintenance and decontamination.

This technology is commercially available and its effectiveness has been demonstrated at Pacific
Northwest National Laboratory (PNNL) and Oak Ridge National Laboratory (ORNL). Various
scenarios and waste conditions can occur at DOE sites; hence, it is important to develop a
computational model of a typical pulsed-air mixing application that can serve as a tool for site
engineers to predict mixing performance and to optimize operational parameters.
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

In this paper, such a computational model was developed with COMSOL Multiphysics and
OpenFOAM CFD solvers using the phase field and volume-of-fluid multiphase methods,
respectively. In this study, simulations were performed for two cases. The first one was a
benchmark case for a single bubble rising for which the results are quantitatively analyzed and
compared to a reference solution using the bubble circularity and the bubble rise mean velocity.
The second test case was for the pulsed air mixer technology. In order to save computational
time, the simulations were carried out in a two dimensional domain where only one half of the
scaled tank was modeled. The parameters for the simulation, in terms of plate dimensions,
pressure values, injection time and other geometrical properties, were obtained from the PNNL
technical report by Powell et al. [1]. It was found that the peak fluid velocities obtained from the
simulation were within an average relative error of 19% compared to the experimental values
available in such report.

MATERIALS AND METHODS

The Phase-Field Method

The phase field method (PFM) is an approach based on free-energy for the modeling of
multiphase flow problems that was used in the COMSOL Multiphysics software. This method is
based on a Cahn-Hilliard equation, for which two second order partial differential equations are
decomposed and solved. The use of the Cahn-Hilliard equation ensures that the total energy of
the system diminishes correctly. The tracking of the interface between the two fluids is governed
by the so-called phase field variable φ [2], [3].

The free energy of a system of two immiscible fluids consists of mixing, bulk distortion and
anchoring energy. This type of energy is modeled as a function of the phase field variable φ.


 
2 2 ||2 
,  ,

1
(Eq.1)

Where ε is a measure of the interfacial thickness, controlled by the grid refinement parameter;
and ftot is the total free energy density of the system.

The evolution of the phase field variables is described by the following equation:








  .   .    . ,
 
(Eq.2)

Where u is the convective field and γ is a mobility parameter that serves to control the relaxation
time that minimizes the total free energy.

The free energy density of an isothermal mixture of two immiscible fluids is comprised of the
sum of the mixing energy and elastic energy. The mixing energy assumes the following form:

1 

 ,   ||2  2  1 ,
2
2 42
(Eq.3)
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

Where λ is the mixing energy density. These parameters, along with ε, are related to the surface
tension σ by the following expression:

2√2 
 , (Eq.4)
3 

The Cahn- Hilliard equation governing the phase field variable is


 .   . , (Eq.5)

Where G is the chemical potential defined by:

   1
      ,
(Eq.6)


For the solution process, COMSOL Multiphysics breaks down Eq.5 in two partial differential
equations:

 
 .   . , (Eq.7a)
2

  .        1, (Eq.7b)

For which ψ is called the phase field help variable. For laminar two-phase flow, the transport of
mass and momentum is governed by the incompressible Navier-Stokes equations including
surface tension (Eq. 8a-b).

  .  !"  . #$  ! %  &  '  (Eq.8a)

. ( (Eq.8b)

The Volume-of-Fluid Method

In addition to the COMSOL Multiphysics software, another multiphase CFD code called
OpenFOAM was utilized for comparison purposes. The multiphase solver used in OpenFOAM is
called InterFoam. This solver uses the Volume of Fluid Method (VOF) to compute multiphase
flows [4], [5].

One momentum equation and one continuity equation are solved for both fluid phases. The
physical properties of one fluid are calculated as weighted averages based on the volume fraction
of the two fluids in one cell. The momentum equation takes the form:


 .    . )  *  +  0, (Eq.9a)
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

.  0, (Eq.9b)

The volume of fluid in a cell is computed as Fvol= Vcell, where Vcell is the volume of a
computational cell and is the fluid fraction in a cell. The values of in a cell should range between
1 and 0. If the cell is completely filled with fluid then the value equals to one and if it is filled
with the other phase considered in the model then its value should be 0. At the interface, the
value is between 0 and 1. The scalar function can be computed from a separate transport
equation that takes the form:


 .    0, (Eq.10)

In OpenFOAM, the necessary compression of the surface is achieved by introducing an extra

artificial compression term into the VOF equation given as:


 .    . 1     0, (Eq.11)

Where ur is a velocity field suitable to compress the interface. This artificial term is active only
in the interface region due to the term γ(1-γ ). The density at any point in the domain is calculated
as a weighted averaged of the volume fraction of the two fluids as   1 . The
surface tension Fs is computed as 
, where n is a unit vector normal to the interface
that can be calculated by  || .


Turbulence Modeling: Large Eddy Simulation (LES)

A turbulence model was also explored for the CFD simulation of the pulsed air mixer. Large
eddy simulation (LES) is based on the computation of large energy-containing structures that are
resolved on the computational grid, where smaller, more isotropic, subgrid structures are also
modeled [6] [7]. This separation of scales is accomplished implicitly in the finite volume method
with low-pass filtering of the Navier-Stokes Equations. Hence, starting from the incompressible
Navier-Stokes equations:

-  . -⨂-  "  . /, (Eq.12)

. -  0, (Eq.13)

Where v is the velocity, p is the pressure, S = 2µD is the viscous stress tensor, where the rate-of-
1
strain tensor is expressed as D= 2 ሺ‫׏‬v+‫׏‬vT ሻ, and µ is the viscosity. The LES equations are
theoretically derived from Eq.12 by applying low-pass filtering, using a pre-defined filter kernel
function G = G(x, ∆), such that:

2  3,
-0  . -0⨂-0  "̅  . / (Eq.14)

. -0  0, (Eq.15)
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

Where overbars denote filtered quantities and commutation errors are not taken into account.
Eq.14 introduces one new term when compared to the unfiltered Eq.12: the unresolved transport
term ∇B, where,

000000  -0⨂-0,
3  -⨂- (Eq.16)

is the subgrid stress tensor. Following B can be exactly decomposed as

000000 5 %,
0 ⨂-0  -0 ⊗ -0  3
3  $- (Eq.17)

Where now only B  is modeled. For this paper, no subgrid modeling approach is applied. This
type of modeling is named implicit LES or ILES.
Regarding the wall treatment, LES required near-wall mesh refinement compared to the rest of
the free-stream flow mesh resolution in order to correctly and accurately solve for the energetic
structures. Since this procedure is computationally expensive, a logarithmic law function is used
along the wall which is implemented with an adjustment of the viscosity for the cells close to the
wall.

Important dimensionless parameters

For the benchmark case of a single bubble dynamics under gravity, three fundamental non-
dimensional numbers are used to describe the deformation of the bubble. These non-dimensional
numbers are quite useful, since it allows for a specific test case to be located in the bubble
deformation curve proposed by Clift et al [8]. Also, these non-dimensional numbers can serve as
a common ground of comparison between CFD methods such as Phase Field method and
Volume-of-Fluid Method. The Eötvös number (Eo), Morton number (M) and Reynolds number
(Re) can be obtained through the Buckingham Pi Theorem [9].

The Eötvös number (Eo) and Morton number (M) are defined as follows:

∆. *.   *)  ∆  = >

6  , 8 9: ;<  ,
   .   )
(Eq.18)

Where ∆ρ is the difference in density of the two phases, g is the gravitational acceleration, L is the
characteristic length, σ is the surface tension, µl is the viscosity of the surrounding fluid, ρl is the
density of the surrounding fluid, ρ1 is the density of the bubble, µ1 is the viscosity of the bubble,
and Ug is the velocity of the bubble given by ?2*@ where r is the initial radius of the bubble.

BENCHMARK COMPUTATIONS

COMSOL and OpenFOAM simulations were performed for a single bubble benchmark case
presented by Hysing et al. [10]. In this benchmark case, several academic codes (TP2D,
MooNMD and FreeLIFE) were used to simulate a well-defined problem and the results were
compared against commercially available codes such as CFX, Fluent and COMSOL
Multiphysics. The non-dimensional flow parameters for the benchmark test case are given in the
table below.
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

Table I. Parameters for the Validation Case

ρ1 ρ2 µ1 µ2 g σ Re Eo M ρ1/ρ2 µ1/µ2
1000 100 10 1 0.98 24.5 35 10 0.0006 10 10

The schematic of the non-dimensional geometry configuration is shown Fig.1 below.

Fig. 1. Domain configuration and boundaries for the benchmark test case.

Domains Ω1 and Ω2 are the domains for fluid 1 and fluid 2, respectively. Boundaries 1 and 3 are
set to wall type boundary with no slip condition. Boundaries 2 and 4 are set as symmetry (slip)
boundary type. Boundary 5 is the fluid interface. The initial diameter di of the bubble is 0.5.

In order to determine level of discretization error, a mesh convergence study is performed with
both COMSOL Multiphysics and OpenFOAM. The quantitative analysis of the results is
evaluated with two quantities: the bubble circularity and the bubble rise mean velocity [10].

The centroid of the bubble is defined as:

D E B
A  B , C   మ
, (Eq.19)
D 1 B
మ

For which Ω2 is the domain occupied by the bubble.

The circularity is given by the following expression:

G "<@HI< <@ J
FH@FK< LH M <NOHP9K<: 9@<9 R

F   ,
G "<@HI< <@ J
QOQQK< G
(Eq.20)
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

The mean rising velocity for the bubble is given by

D B
S  మ
, (Eq.21)
D 1 B
మ

(a) (b)

(c) (d)

Fig.2 : Circularity and bubble rise mean velocity using VOF and PFM. The time in this plot is non-dimensionalized
݀
by ܶ ൌ ‫ݐ‬ൗ‫ ݐ‬where ‫ݐ‬௥௘௙ ൌ ට ௜ൗ݃
௥௘௙

The shape of the bubble can be analyzed qualitatively by comparing the deformed interface with
the reference solution and the bubble shape diagram given by Clift et al [8]. Given the value of
the Eo, M and Re numbers, the expected bubble shape that corresponds to the flow conditions of
the test case can be located in such diagram that could be one of the following: spherical,
ellipsoidal, wobbling, dimpled ellipsoidal cap, skirted or spherical cap. In the case analyzed for
this paper, the shape would correspond to an ellipsoidal shaped deformation.
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

Fig.2 shows the quantitative analysis performed for this test case, where the influence of the
mesh size can be observed in regards to the convergence of the results compared to the available
published solution.

The circularity and the bubble rise mean velocity were analyzed using both the PFM and VOF
methods (Fig. 2a and 2c). Fig. 2d presents the rise velocity as a function of time. The bubble
velocity reaches a constant value after about T=2.5. These results agree within 10% error and
any discrepancy observed in both the circularity and velocity plots may be the result of the
difference between PFM and the level set method adopted in the reference solution [10], [11].

In regards to the results presented for the VOF method, it can be observed that the curve
tendency is similar to the one observed with PFM in Fig. 2a and Fig. 2c. It is interesting to point
out that the bubble takes more time to deform in the simulation with OpenFOAM using the VOF
method. In non-dimensional time, for PFM and TP2D, the bubble has reached its terminal shape
at around T=4.5, whereas at this time, VOF is still showing the highest deformation point in the
circularity plot and has not reached the terminal shape.

Similar comments can be made for the velocity plot (Fig. 2b) where the terminal velocity is
reached after T=4. The value for the terminal velocity for the VOF mesh of 200x400 is close to
the value given by the reference solution; however, the highest mesh resolution of 400x800
shows a higher value for the bubble mean rise velocity.

The differences observed between the results achieved with OpenFOAM and COMSOL
Multiphysics could be due to the variance of the numerical methods used in each these CFD
solvers. OpenFOAM uses finite volume whereas COMSOL is a finite element based software
package.

RESULTS

Parameters

The simulations are performed on a two dimensional domain, for which Fig. 3e shows the
different values of the dimensions considered. The hstandoff distance represents the vertical
distance that separates both accumulator plates. The Rtank is the radius of the tank, since only one
half of the PNNL 1/12 scaled experimental test tank is simulated. The value of hvel corresponds
to the location at which the velocity is measured for the different probe points in the experiment
and simulation. The distance Htank represents the height level of the water inside the tank. The
blue dotted line represents the tube connected to the accumulator plates and through which the
high pressure air is injected into the tank.

PNNL studied the velocities created by the expanding bubble around the accumulator plate
pulsed-air mixing technology by attaching an anemometer to a rail on top of the tank; this
anemometer was placed at 8, 11, 15 and 19 cm respectively from the plate centerline (middle of
blue dotted line in Fig. 3e).
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

During the experimental testing, PNNL used accumulator plates of several diameters (6.1 cm,
14.2 cm, 23.9 cm, and 36.6 cm) at different values of injection pressure (20 psi, 40 psi, 60 psi, 80
psi, 100 psi) and different gas pipe diameters.

(a) Sketch of bubble growth during pulsed-air mixing (b) Pulsed-air mixing experimental setup at PNNL

(c) Sketch showing mixed pattern induced by rising (d) High speed photo showing bubble growth in
bubble accumulator plate

(e) Schematics for the simulations performed in COMSOL and OpenFOAM

Fig..3.Description of the computational and experimental set-up for the testing of the pulsed-air mixing technology.
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

For this paper, one case is simulated for which the parameters of pressure, standoff distance and
the plate diameter are given in Table II.

According to the experimental data provided by PNNL, the fluid temperature varied between
13.2 ºC to 13.4º C, hence the viscosities, densities and surface tension for air and water, which
were the two fluids considered, were set accordingly.

Moreover, during the experimental setup, the pulse injection time was set to 0.4 s, from the
opening of the valve to the moment where the pulse is cut off.

Table II: Parameters for the Simulations for the Pulsed-Air Mixing

Standoff Plate
Pressure
Case # distance Diameter Gas pipe
(psig)
(cm) (cm)
1 20 0.635 6.1 1/8 S40

Mesh Convergence Study

In order to build the proper structured mesh for the turbulent simulation performed in
OpenFOAM using LES, a mesh convergence study was completed for which the results are
presented below in Fig. 4. This study corresponds to case 1 with an injection time of 0.4 seconds.
The water velocities at four sensor locations were plotted along with the phase fraction recorded
at each sensor location. The coarsest mesh contained 27,540 cells, the medium resolution mesh
contained 110,160 cells, and the finest mesh contained 440,640 cells. As the resolution of the
simulations increases, it becomes easier to discern the nearly linear increase of the water velocity
at each sensor location in advance of the arrival of the air-water interface. The low resolution
simulation results reveal significant numeric noise in the captured velocities as the air-water
interface nears each sensor. The larger cells in the low resolution simulation cause the interface
to be diffused and consequently phase fraction values well below 1 are reported. The air-water
interface is defined as α = 0.5 for all simulations.

Low resolution Medium Resolution High Resolution

Fig. 4: Convergence study results for the pulsed-air mixing simulation case.
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

Pulsed-air mixing simulations

The simulations for the pulsed-air mixing technology were performed using the PFM and LES
methods. First, a laminar flow simulation was performed using PFM in COMSOL Multiphysics.

(a) Radial Velocity vs. Distance from centerline for (b) Radial Velocity vs. Distance from centerline for
PFM LES

(c) Surface plot for LES alpha field (alpha=0 for air and (d) Surface plot for LES velocity field
alpha=1 for water)

Fig. 5: Simulation results for Pulsed-air Mixing.

The results for this model are shown in Fig.5a, where it can be observed that the velocities
calculated by PFM were very different from the ones gathered during the experiment. The
average relative error for this simulation is 755.82%, which is not acceptable in any type of CFD
simulation. However, several conclusions were drawn from this simulation which served to
accurately simulate Case 1 with a turbulence model. Since the velocities given by PFM were not
reasonable, it was inferred that PFM alone assuming an incompressible laminar flow inside the
pipe and beyond the accumulator plate was not sufficient for the given conditions of the model.
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

As previously stated, the effect of temperature was already taken into account in order to cross
out any source of discrepancy in regards to the fluid parameters that might be influencing the
laminar flow simulation.

Hence, a quick calculation on the flow velocity inside the pipe using Bernoulli’s equation was
performed and yielded that the velocity of water would be in the order of 16 m/s and the velocity
of air would be in the order of 462 m/s. Taking into account the gas pipe diameter, the Reynolds
number (Re) was calculated accordingly, which resulted in  2.94  10 and   
1.23  10 .

The calculation of the previous parameters made it clear that the flow inside the pipe and
therefore between and beyond the accumulator plates will also be highly turbulent. It was
decided to run this simulation with the LES turbulence model available in OpenFOAM.

The results for this turbulence simulation are shown in Fig. 5b where a clear difference can be
observed in comparison with the laminar flow simulation from Fig. 5b. The results agree with
the experimental data with an average relative error of 19%. For both plots in Fig.5, more probe
points were added in order to obtain a better velocity profile as a function of the distance from
the centerline.

Moreover, Fig.5c and 5d show the surface plots that are obtained for the alpha field in
OpenFOAM. As specified, the α field is equal to 0 for air and 1 for water. In Fig.5c, The
turbulent flow structures can be observed inside and beyond the accumulator plates. In Fig.5d,
the surface plot shows a considerably high value for the velocity field inside the gas pipe with a
maximum of 840 m/s.

From the observed plots, it is clear that the CFD simulation of the pulsed-air mixing technology
requires the implementation of an incompressible flow turbulence model in order to yield good
results. The discrepancies between LES and the experimental data could be due to initial values
assumed in the flow field for the LES model in OpenFOAM, which could be optimized to a
proper value in order to increase accuracy.

DISCUSSION AND CONCLUSION

The CFD capabilities of COMSOL Multiphysics and OpenFOAM were implemented for the
simulation of the single bubble benchmark study and the pulsed-air mixing technology. For the
first simulation, it was shown that both numerical solvers are able to accurately model the single
bubble validation study.

The PNNL experimental setup for the pulsed-air mixing technology was modeled in both
numerical solvers in a two-dimensional space and half of the fluid domain in order to efficiently
use the available computational resources. By observing the results of the simulation undertaken
with PFM in COMSOL Multiphysics, it can be concluded that a turbulence model is essential to
the accurate modeling of this type of mixing technology. The results provided by LES are in
agreement with the experimental data provided by PNNL within a low and reasonable margin of
WM 2011 Conference, February 27- March 3, 2011. Phoenix, AZ

error. Hence, it can also be concluded that simulating this application in a two-dimensional space
does not hinder the ability of achieving good results with the numerical solvers. This is of
primary importance for CFD modelers involved in this type of application since simulations in a
2D space are much less computationally expensive.
For future work, the turbulence capabilities of COMSOL Multiphysics will be investigated and
implemented using the available   and   turbulence models [12]. More importantly, the
effects of air pressure and plate diameter presented in the PNNL report [1] will also be simulated
with both OpenFOAM and COMSOL Multiphysics.

REFERENCES

[1] M R Powell and C R Hymas, "Retrieval Process Development and Enhancements FY96
Pulsed-Air Mixer Testing and Deployment Study," Pacific Northwest National Laboratory,
Richland, WA, PNNL-1120, UC-721, 1996.
[2] COMSOL AB, COMSOL User Guide- Chemical Engineering Module. Trondheit, Sweden,
2008.
[3] Pengtao Yue, Chungfeng Zhou, James J Feng, Carl F Olliver-Gooch, and Howard H Hu,
"Phase-Field simulations of interfacial dynamics in viscoelastic fluids using finite elements
with adaptive meshing," Journal of Computational Physics, vol. 219, no. 1, pp. 47-67,
November 2006.
[4] O Ubbink and R I Issa, "A Method for Capturing Sharp Fluid Interfaces on Arbitrary
Meshes," Journal of Computational Physics, vol. 153, pp. 26-50, 1999.
[5] Henrik Rusche, "Computational Fluid Dynamics of Dispersed Two-Phase Flows at High
Phase Fractions," University of London, Imperial College, London, Ph.D. Thesis 2002.
[6] Rickard E Bensow and Goran Bark, "Implicit LES Predictions of the Cavitating Flow on a
Propeller," Journal of Fluids Engineering, vol. 132, April 2010.
[7] Rickard E Bensow and Goran Bark, "Simulating Cavitating Flows with LES in
OpenFOAM," in V European Conference on Computational Fluid Dynamics , Lisbon,
Portugal , 2010.
[8] R Clift, J R Grace, and M E Weber, Bubbles, Drops and Particles. New York, USA:
Academic Press, 1978.
[9] Garrett Birkhoff, Hydrodinamics: A study in Logic, Fact and Similitude. Princeton:
Princeton University Press, 1960.
[10] S Hysing et al., "Quantitative benchmark computation of two-dimensional bubble
dynamics," International Journal for Numerical Method in Fluids, vol. 60, pp. 1259-1288,
2008.
[11] Elin Olsson and Gunilla Kreiss, "A Conservative level set method for two phase flow,"
Journal of Computational Physics, pp. 225-246, 2005.
[12] David C Wilcox, Turbulence Modeling for CFD.: DCW Industries, 2006.