JOM, Vol. 67, No.

7, 2015
DOI: 10.1007/s11837-015-1465-x
! 2015 The Minerals, Metals & Materials Society

Large Eddy Simulation of Bubbly Flow and Slag Layer Behavior

in Ladle with Discrete Phase Model (DPM)–Volume of Fluid
(VOF) Coupled Model

LINMIN LI,1 ZHONGQIU LIU,1 MAOXUE CAO,1 and BAOKUAN LI1,2

1.—School of Materials and Metallurgy, Northeastern University, Shenyang 110819, People’s

Republic of China. 2.—e-mail: libk@[Link]

In the ladle metallurgy process, the bubble movement and slag layer behavior
is very important to the refining process and steel quality. For the bubble–
liquid flow, bubble movement plays a significant role in the phase structure and
causes the unsteady complex turbulent flow pattern. This is one of the most
crucial shortcomings of the current two-fluid models. In the current work, a
one-third scale water model is established to investigate the bubble movement
and the slag open-eye formation. A new mathematical model using the large
eddy simulation (LES) is developed for the bubble–liquid-slag-air four-phase
flow in the ladle. The Eulerian volume of fluid (VOF) model is used for tracking
the liquid-slag-air free surfaces and the Lagrangian discrete phase model
(DPM) is used for describing the bubble movement. The turbulent liquid flow is
induced by bubble–liquid interactions and is solved by LES. The procedure of
bubble coming out of the liquid and getting into the air is modeled using a user-
defined function. The results show that the present LES–DPM–VOF coupled
model is good at predicting the unsteady bubble movement, slag eye formation,
interface fluctuation, and slag entrainment.

List of symbols
CD Drag force coefficient l Viscosity
CVM Virtual mass force coefficient lt Turbulent viscosity
CS Smagorinsky constant j Von Kármán constant
d Distance to the closest wall dij Dirac function
dp Bubble diameter
FD Momentum exchange coefficient
~VM
F Virtual mass force
~PG
F Pressure gradient force INTRODUCTION
~
g Gravitational acceleration
Ls Mixing length for subgrid scales Argon gas stirring is widely employed in the steel
n Number of bubbles refining to homogenize the chemical composition of
P Pressure alloy elements and temperature. It helps to remove
Q Gas flow rate inclusions and enhance the rates of refining reac-
Re Relative Reynolds number tions. The bubbles are successively formed at the exit
S Rate-of-strain tensor of the nozzle. Then, they rise upward and entrain the
t Time surrounding molten steel into their wakes and form a
u
~ Velocity turbulent bubble plume. With a high gas flow rate,
V Cell volume gas bubbles will break up the slag layer, cause fluc-
tuation, and strengthen the interfacial reactions. A
Greek letters strong interaction between slag and steel is needed to
a Volume fraction promote the efficiency of desulfurization, but the
q Density pick-up of oxygen and nitrogen from the atmosphere
s Subgrid-scale stress and the slag entrainment might happen. Thus, the

(Published online June 3, 2015) 1459

1460 L. Li, Liu, Cao, and B. Li

overall influence of slag eye formation and phase scheme is adopted for the gas bubble phase in the
interactions on the quality of steels is important. current work.
From the observation of the slag layer fluctuation On the other hand, the interface fluctuation is
and the slag open eye formation, a transient complex remarkable and the slag eye is irregular. Previous
turbulent multiphase phenomenon is found in which simulations are always based on the Reynolds-av-
a large number of eddies with a wide range of length eraged Navier–Stokes (RANS) equations, and it is
and time scales exist. The top free surface and slag hard for these models to describe the unsteady
eye format in a real ladle is shown in Fig. 1. phenomena including the slag eye formation and
Understanding the characteristics of bubble interface fluctuation caused by bubble movement
movement and interfacial behavior between the and lots of velocities in ladle. The large eddy simu-
molten steel and slag is important. Several cold lation (LES) resolves large eddies directly while
physical models1–5 and numerical simulations6–8 small eddies are modeled using the subgrid scale
have been taken to investigate the flow character- (SGS) model. Recently, several researchers11–13
istic and slag layer behavior in the gas-stirred ladle. have successfully taken the LES model into indus-
Both the Eulerian approach6,7 and Lagrangian trial applications. The results show that the LES
approach8–10 were used to deal with the gas bubble model can obtain more complex turbulent flow
phase. Li et al.6 developed a three-phase model by structures and predict more accurate flow patterns.
the Eulerian volume of fluid (VOF) method to study In the current work, the gas bubbles that trans-
the slag layer behavior using different gas flow port in the liquid are tracked using the DPM
rates. Llanos et al.7 also developed a water/oil/air because the bubbles are found dispersed in the liq-
system and employed the VOF model to simulate uid through the water model experiment. Free
the interaction among the phases. The operation surfaces among the liquid steel, slag, and air are
parameters including mixing time, lining refractory simulated using the VOF model. So the current
wear, and slag opening were studied for different mathematical model includes the four phases: liq-
cases. Guo and Irons8 used the Lagrangian–Eule- uid, slag, air, and gas bubbles. The procedure that
rian model to study the gas–liquid flow in a ladle, bubbles come out of the liquid and get into the air in
but the slag layer was ignored. Cloete et al.9 and Liu the current work is modeled by a user-defined
et al.10 employed the Lagrangian discrete phase function (UDF). The LES with the Smagorinsky–
model (DPM) to describe the bubble plume and the Lilly SGS model is used to describe the multiphase
Eulerian VOF model to track the free surfaces, but turbulent flow. The gas-bubble stirring progress and
the unsteady phenomena of the slag layer behavior the calculation approaches for different phases are
and the slag eye formation are not well predicted. In shown in Fig. 2.
the Eulerian–Lagrangian formulation, the liquid-
phase equations are solved in an Eulerian frame of
reference and the bubble-phase equations are solved
in a Lagrangian frame of reference. In ladles, the
gas phase is an unconfined plume with a free
boundary between bubbles and the melt. It is a
challenging task for the Eulerian formulation to
prevent the gas phase from numerically diffusing
into regions where no bubbles are present in real-
ity.8 Based on the consideration, the Lagrangian

Fig. 2. Schematic gas stirring process and calculation approaches

Fig. 1. The top free surface and slag eye format in a real ladle. for phases.
Large Eddy Simulation of Bubbly Flow and Slag Layer Behavior in Ladle 1461
with Discrete Phase Model (DPM)–Volume of Fluid (VOF) Coupled Model

and air. The tracking of the interfaces between

MODEL FORMULATION
phases is accomplished by the solution of a conti-
The calculation model is composed of the DPM nuity equation for the volume fraction. For the qth
and VOF models. The VOF model is used to track phase, the continuity equation is described in the
the free surfaces by solving a single set of momen- following form:
tum equations and tracking the volume fraction of % &
each fluid. The gas bubbles are tracked using the 1 @! " ! "
aq qq þ r $ aq qq u
~q ¼ 0; (5)
DPM, and the buoyancy, drag force, virtual mass qq @t
force, and pressure gradient force are included. The
LES turbulence model is taken to describe the tur- where the volume fraction aq is constrained by
bulence in the system. The subgrid scale modeling is Pn
aq ¼ 1. When the VOF model is used, it is as-
based on the Smagorinsky–Lilly kernel. The bub- q¼1
bles will disappear after coming out of the liquid and sumed that the velocity field is shared among the
the influence of bubbles on the top air layer is ig- phases and a single momentum equation is solved
nored, so a UDF is defined to delete the bubbles throughout the domain. The momentum equation is
arrived at the position that the volume fraction of as follows:
air is larger than 0.5.
@ ' ! "(
ðq~
uÞ þ r $ ðq~~Þ ¼ "rP þ r $ l r~
uu uT
u þ r~
Discrete Phase Model @t
þ q~ ~p ;
gþF ð6Þ
The gas bubble is treated as a discrete phase and
the trajectory is predicted by integrating the force
where the properties q and l used above are mix-
balance on it, which is written in a Lagrangian ~p represents
ture properties, P is the pressure, and F
reference frame. This force balance equates the
the forces that bubbles acting on the liquid.
bubble inertia with the forces acting on the bubble,
and it can be described as:
! " LES Model
d~up ! " ~g qp " q
¼ FD u " up þ
~ ~ ~
þ F; (1)
dt qp The LES model is the most widely known scale-
resolving simulation model, which resolves large
where F~ is the additional acceleration term, turbulent structures in space and time down to the
! "
FD u ~p is the drag force per unit bubble mass,
~" u grid limit everywhere in the flow. In the LES, the
and FD is written as: big three-dimensional eddies that are dictated by
the geometry and boundary conditions of the flow
18l CD Re involved are directly resolved, whereas the small
FD ¼ ; (2)
qp d2p 24 eddies that tend to be more isotropic and less
dependent on the geometry are modeled. The sub-
where u ~ is the fluid velocity, u
~p is bubble velocity, l is grid scale stresses sij resulting from the filtering
the molecular viscosity of the fluid, dp is the bubble operation are given by:
diameter, and Re is the relative Reynolds number.
In this work, the bubbles are assumed spherical 1
sij " skk dij ¼ "2lt S!ij ; (7)
throughout the domain. With this assumption, the 3
drag of a spherical object that determined by Liu
~ where lt is the subgrid-scale turbulent viscosity, dij
et al.14 is used. The additional acceleration term F
is equal to 1 if i = j and 0 otherwise, and S!ij is the
here includes the virtual mass force and the pres-
rate-of-strain tensor for the resolved scale that de-
sure gradient force. They can be described as follows:
# $ fined by:
q d~
up # $
~
FVM ¼ CVM u
~p r~u" (3) 1 @ u!i @ u!j
qp dt S!ij ¼ þ : (8)
2 @xj @xi

~PG ¼ q u
F u;
~p r~ (4)
The present work uses the Smagorinsky–Lilly
qp subgrid-scale model.15–17 In this model, the eddy-
viscosity is modeled by:
where CVM is the virtual mass factor with a value of ) )
0.5. lt ¼ qL2s )S!); (9)
) ) qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
VOF Model where )S!) ¼ 2S!ij S!ij and Ls is mixing length for
subgrid scales that computed using:
The VOF model can model two or more immiscible
fluids, and it is used in the current work for tracking + ,
the free surfaces among the liquid steel, slag layer, Ls ¼ min jd; CS V 1=3 ; (10)
1462 L. Li, Liu, Cao, and B. Li

Table I. Parameters of both experimental and

numerical simulation

Parameters Values
Bottom diameter 617 mm
Slope angle 2.44"
Liquid depth 700 mm
Slag layer thickness 50 mm
Porous plug diameter 44 mm
Plug radial position 0.67R
Water density 1000 kg m"3
Water viscosity 0.001 kg m"1 s"1
Oil density 900 kg m"3
Oil viscosity 0.006 kg m"1 s"1
Gas density 1.138 kg m"3 (25"C)
Gas viscosity 1.663 9 10"5 kg m"1 s"1
Bubble diameter 3 mm (70 L/h), 3.5 mm (110 L/h)
Interfacial tension 0.072 N/m (water/air)

where j is the von Kármán constant, d is the dis-

tance to the closest wall, V is the volume of the
computational cell, and CS is the Smagorinsky Fig. 3. The size, mesh, and boundary conditions.
constant equal to 0.1.

EXPERIMENTAL DETAILS
The calculation is carried out by the transient
A one-third scale water model is established with pressure-based solver. For using the LES model
a camera observing the slag open eye and a high- with the Smagorinsky–Lilly SGS model, the boun-
speed camera capturing the bubbles. The N2 (25"C, ded second-order implicit transient formulation and
1 atm) is chosen to simulate the argon gas, and the the bounded central-differencing scheme for
gas is injected into the liquid from a nozzle made of momentum are used. The conservation equations
the porous mullite that is the same as that used in are discretized using the control volume technique
the real ladle. An eccentric gas bubbling ladle is and the PISO scheme is used for the pressure–ve-
modeled and the radial position of nozzle is set at locity coupling. A physical time scale of 0.005 s is
0.67R. The water and oil are used to simulate adopted for the simulation. The gas flow rate and
the molten steel and the slag layer, respectively. bubble diameter are set according to the experi-
The open software ImageJ is used for analyzing the mental measurement. For calculations, the bubble
bubble diameter. It is observed that aggregation and aggregation and breakage are ignored because, as
breakage rarely happen during bubble floatation, so mentioned before, it is observed that aggregation
a uniform mean bubble diameter is calculated for and breakage rarely happen when bubbles are
each case. The details about the geometric param- floating. The number of injected bubbles is calcu-
eters and the material properties are shown in lated by the gas flow rate and mean diameter:
Table I.
6Q
n¼ ; (11)
NUMERICAL DETAILS pd!3p
The computation is performed using the com-
where Q is the gas flow rate and d!p is the mean
mercial computational fluid dynamics software
bubble diameter. The bubble diameter for each case
FLUENT (ANSYS, Inc., Canonsburg, PA). The
and other parameters of the computations are also
geometrical model for calculation is developed
shown in Table I.
according to the one-third scale water model for the
comparison. The grid densities of the three-dimen-
sional all-hexahedral element mesh are determined RESULTS AND DISCUSSION
as follows: The maximum mesh size of 10 mm and a
Bubble Distribution
stretching ratio of 1.1 is used near the wall. The
mesh size near the walls and phase interfaces is set In the water model experiment, the bubbles are
to 4 mm, and the mesh size of the inlet is also set to captured by a high-speed camera and a rule is set in
4 mm. The mesh and boundary conditions are the ladle near bubbles for bubble size analysis. The
shown in Fig. 3. The number of cells is approxi- bubble plume at the gas flow rate of 70 L/h is shown
mately 300,000. in Fig. 4a. It was already observed that the rise
Large Eddy Simulation of Bubbly Flow and Slag Layer Behavior in Ladle 1463
with Discrete Phase Model (DPM)–Volume of Fluid (VOF) Coupled Model

velocities of the bubbles in the plume have a Gaus- the isosurfaces where the volume fraction of slag is
sian radial distribution and bubbles will move away equal to 0.5 is displayed in Fig. 4b. It is found that
from the axis of the plume, causing the plume to the bubble distribution of the calculation fits well
spread laterally.18 The calculation result of the with the experiment.
bubble distribution with the slag layer showing by Figure 5 shows the bubble floating process at the
beginning of the stirring. Bubbles will stay for a
short time at the initial state, and thus a ‘‘mush-
room bubble plume’’ appears. When bubbles arrived
at the top of the liquid and break up the slag layer, a
relatively big slag open eye forms and quickly dis-
appears.

Interface Fluctuation and Slag Eye Formation

From the experimental observation, the slag eye
will form and collapse alternately after an ample
time stirring at a relatively small gas flow rate.
Figure 6 illustrates the slag layer fluctuation, slag
eye formation, and collapse of the experiment at the
gas flow rate of 70 L/h. The pictures are taken with
the help of the mirror at the 45" angle on the top of
the ladle. The pictures show a cycle of slag open eye
formation (Fig. 6a–c) and collapse (Fig. 6d–f). It is
found that the fluctuation of the top free surface is
not obvious; the slag–melt interface at the periphery
of the eye region is volatile. The slag might break up
into slag droplets and the ‘‘slag emulsification’’
Fig. 4. Bubble distribution of (a) experiment and (b) calculation. happens.

Fig. 5. The predicted bubble floating process at the beginning times: (a) 0.7 s (b) 1.1 s (c) 1.5 s (d) 2.2 s (e) 2.8 s (f) 3.5 s.
1464 L. Li, Liu, Cao, and B. Li

Fig. 6. Slag layer fluctuation and slag eye formation and collapse of experiment: (a) eye appear, (b) grow, (c) become biggest, (d) start closing,
(e) closing, and (f) closed.

The current work focuses on modeling the slag Figure 9 displays the slag volume fraction and the
layer behavior including the interface fluctuation, liquid velocity field of the central plane of the slag
slag eye formation, and collapse due to the gas layer (y = 0.725 m). It shows a collapse process of
bubbling. The liquid steel–slag–air interfaces and the slag eye. From the velocity field, we find that the
the slag open eye formation and collapse predicted velocity vector of the liquid around the slag eye
by the current model is shown in Fig. 7. The pro- points to the spout eye center, and the velocity
cess of slag eye formation and collapse is well vector of the liquid in the slag eye is pushing the
predicted and agrees with the experimental result. slag to the periphery. It illustrates how the slag eye
It is also found that the fluctuation of the top free becomes smaller and collapses. On the other hand, if
surface is not obvious, whereas the slag–melt the velocity in the slag eye pointing to the periphery
interface at the periphery of the eye region is is larger, then the slag eye will form and become
volatile. Figure 8 shows the predicted bubble dis- bigger.
tribution, slag layer, and flow field. It is shown
that bubbles describe curved paths in the liquid
Effect of Gas Flow Rate
and cause a lot of eddies in the liquid region near
the bubble plume. The result indicates that The gas flow will significantly influence the slag
the floatation of bubbles causes turbulence in the open formation and the interface fluctuation. Figure
liquid near the bubble plume and induces the 10 shows the melt–slag interface and the movement
interface fluctuation. of bubbles with displaying the bubble velocity of
Large Eddy Simulation of Bubbly Flow and Slag Layer Behavior in Ladle 1465
with Discrete Phase Model (DPM)–Volume of Fluid (VOF) Coupled Model

Fig. 7. Calculation results of the interface fluctuation and the slag eye formation and collapse: (a) eye appear (b) grow (c) become biggest (d)
start closing (e) closing (f) closed.

different gas flow rates. With a higher gas flow rate,

the bubbles are found moving faster and spread
less. The interface fluctuation is also greater. A
relatively larger slag open eye forms and exists for
most of the time, but the eye shape is irregular and
changes frequently. At a high gas flow rate, the slag
eye size changes alternately instead of the forma-
tion and collapse.
Furthermore, with a relatively high gas flow
rate, the melt–slag interface at the periphery of the
eye region might frequently break up into slag
droplets. The droplets generation can increase the
interfacial area for slag–melt reactions, but also it
can result in undesirable effects such as the slag
entrainment and influence the quality of steels.
Thus, the study on mechanisms of slag eye for-
mation and interface fluctuation, which is the
weakness of the traditional simulations using the
RANS equations, is important. Figure 11 compares
the slag entrainment between the calculation and
experiment at the gas flow rate of 110 L/h. Two
slag droplets are found from the calculation result.
A big slag eye is predicted and it agrees well
with the experiment. It shows that the calculation
model developed in this work can predict well
the slag layer behavior especially the slag
Fig. 8. The predicted bubble distribution and flow field. entrainment.
1466 L. Li, Liu, Cao, and B. Li

Fig. 9. Slag volume fraction and velocity on the slag layer center plane at the instants of (a) start closing (b) closing (c) closed.

Fig. 10. The bubble distribution and liquid steel–slag interface of different gas flow rates: (a) 70 L/h and (b)110 L/h.

Fig. 11. The slag entrainment of (a) calculation and (b) experiment.
Large Eddy Simulation of Bubbly Flow and Slag Layer Behavior in Ladle 1467
with Discrete Phase Model (DPM)–Volume of Fluid (VOF) Coupled Model

ACKNOWLEDGEMENTS
CONCLUSION
Authors are grateful to the National Natural
A new mathematical model based on the LES and
Science Foundation of China for support of this re-
the DPM-VOF coupled model is developed to simu-
search (Grant No. 51210007).
late the multiphase flow in the ladle with gas bub-
bling. The bubble phase is calculated using the DPM
while the liquid slag–air interfaces are tracked REFERENCES
using the VOF model. The procedure of bubbles 1. J.W. Han, S.H. Heo, D.H. Kam, B.D. You, J.J. Pak, and H.S.
getting into the air is achieved by a UDF. The model Song, ISIJ Int. 41, 1165 (2001).
2. K. Yonezawa and K. Schwerdtfeger, Metall. Mater. Trans. B
predicted results are compared with the water 30B, 411 (1999).
model experiment results. The conclusions can be 3. V. Sahajwalla, A.H. Castillejos, and J.K. Brimacombe, Me-
drawn as follows: tall. Mater. Trans. B 21B, 71 (1990).
4. D. Mazumdar and J.W. Evans, Metall. Mater. Trans. B 35B,
1. The bubbles are found moving in curved paths 400 (2004).
and inducing many eddies in the region near the 5. K. Krishnapisharody and G.A. Irons, ISIJ Int. 48, 1807 (2008).
bubble plume. The eddies make the melt–slag 6. B.K. Li, H.B. Yin, C.Q. Zhou, and F. Tsukihashi, ISIJ Int.
interface fluctuate and might break up the slag 48, 1704 (2008).
7. C.A. Llanos, S. Garcia, J.A. Ramos-Banderas, J.D.J. Bar-
into droplets. reto, and G. Solorio, ISIJ Int. 50, 396 (2010).
2. The shape and size of the slag eye are irregular. 8. D. Guo and G.A. Irons, Metall. Mater. Trans. B 31B, 1457
At a small gas flow rate, the slag eye form and (2000).
collapse alternately. And at a high gas flow rate, 9. S.W.P. Cloete, J.J. Eksteen, and S.M. Bradshaw, Prog.
Comput. Fluid Dynam. 9, 345 (2009).
the interface fluctuates more greatly and the slag 10. H. Liu, Z. Qi, and M. Xu, Steel Res. Int. 82, 440 (2011).
eye exists for most of the time, but the slag eye 11. Z.Q. Liu, B.K. Li, M.F. Jiang, and F. Tsukihashi, ISIJ Int.
size also changes. The formed slag eye size of the 53, 484 (2013).
present prediction qualitatively agrees well with 12. Z.Q. Liu, L.M. Li, B.K. Li, and M.F. Jiang, JOM 66, 1184
the experiment results. (2014).
13. F. Wang, B.K. Li, and F. Tsukihashi, ISIJ Int. 47, 568 (2007).
3. The slag entrainment happens frequently in the 14. A.B. Liu, D. Mather, and R.D. Reitz, Modeling the Effects of
experiment especially at high gas flow rates. Drop Drag and Breakup on Fuel Sprays, SAE Technical
Although the previous models can hardly predict Paper 930072 (Warrendale, PA: SAE, 1993).
the interface fluctuation and slag entrainment, 15. J. Smagorinsky, Month. Wea. Rev. 91, 99 (1963).
16. M. Germano, U. Piomelli, P. Moin, and W.H. Cabot, Phys.
the model developed in this study shows a good Fluids A 3, 1760 (1991).
ability and result for the slag entrainment pre- 17. D.K. Lilly, Phys. Fluids A 4, 633 (1992).
diction. 18. M.T. Dhotre and B.L. Smith, Chem. Eng. Sci. 62, 6615 (2007).