Continuous Casting Thesis
Continuous Casting Thesis
Continuous Casting Thesis
Modeling of Macrosegregation
in Direct Chill Casting
Doctoral Dissertation
Graduate School
University of Nova Gorica
2006
Miha Zalonik
Modeling of Macrosegregation
in Direct Chill Casting
Doctoral Dissertation
Graduate School
University of Nova Gorica
2006
Acknowledgements
My research life started when Professor Boidar arler established the Laboratory for Multiphase Processes. There I articulated my scientic curiosity,
began to gain autonomy in conducting research, and learned to swim when
thrown into water. I thank Professor arler for being my doctoral advisor,
for his initiative, perseverance, guidance and appreciation of my work.
Further I thank the committee members, Dr. Dominique Gobin, Professor
Matja Hriberek, and Professor Boris tok for their time, exibility and
rapid evaluation of the manuscript.
Special thanks go to Dominique Gobin for providing an invaluable working refuge at the FAST laboratory in Orsay, for his continuing encouragement, guidance, and inspiring suggestions. Further thanks to Benot Goyeau
for the several profound discussions, which claried the model development.
My thanks also go to Shihe Xin who provided the data of his natural convection computations used for comparison.
The working atmosphere was brightened up by Igor (always up for a
discussion and always helping me to get a more lucid viewpoint), Janez,
Robert, Jure, Marko, Simo, Henrik, Stanislav, Miran, and Gregor. Thank
you for all the stimulating discussions of our work, for sharing enthusiasm
and the (occasionally too good) pinch of sarcasm that was sometimes needed
to overcome temporary frustrations.
I would also like to thank everybody at Impol Aluminium Industry for
their friendliness and encouragement. Especially Edvard Slaek and Rajko
afhalter for supporting the project, and Marina Jelen, Vili Strnad and
Vukain Dragojevi for all discussions and their eorts in the attempts of
providing reliable experimental macrosegregation data.
The least direct but the most valuable way of support and stimulation
came from my family and friends who kept me sane in moments of deep
immersion into curiosity by leading me onto other warm and exciting paths.
This work was carried out during 2001-2006 at the Laboratory of Multiphase Processes of the University of Nova Gorica. The research was funded
by the Slovenian Government through the Young Researchers program and
by Impol Aluminium Industry.
Key words
Direct chill casting, macrosegregation, solidication, thermosolutal natural
convection, aluminum alloys
cije in transporta prosto plavajoih zrn. Ugotovljeno je, da tok vsled dvojno
difuzivne naravne konvekcije vpliva na makroizcejanje preko vpliva na tok
taline v poroznem obmoju strjevanja in preko transporta sestavin v kapljevinastem obmoju. Oba pojava, naravna konvekcija in krenje, bistveno vplivata na makroizcejno polje. Vpliv transporta prosto plavajoih zrn je tudiran
s poenostavljenim terminim modelom. Podane so ugotovitve o kvalitativnem
vplivu velikosti zrn in meje konsolidacije na negativno izcejno cono v srediu
ulitka.
Kljune besede
polkontinuirno ulivanje, makroizcejanje, strjevanje, dvojno difuzivna naravna
konvekcija, aluminijeve zlitine
Contents
List of Figures
iv
List of Tables
ix
List of Symbols
xiii
1 Introduction
1.1 Background and Rationale . . . . . . . . . . . . . . . .
1.2 Macrosegregation: The Underlying Physical Phenomena
1.2.1 Microsegregation . . . . . . . . . . . . . . . . .
1.2.2 Relative Motion of Solid and Liquid Phases . .
1.2.3 Macrosegregation in DC Casting . . . . . . . .
1.2.4 Numerical Modeling of Macrosegregation . . . .
2 Literature Review
2.1 Modeling of Alloy Solidication . .
2.2 Macrosegregation in DC Casting . .
2.2.1 Experimental Investigations
2.2.2 Model Investigations . . . .
2.2.3 Summary . . . . . . . . . .
2.3 Flow in Liquid Metals . . . . . . .
2.4 Numerical Modeling Issues . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
3
3
5
8
10
.
.
.
.
.
.
.
13
13
16
16
17
20
20
21
3 Objectives
25
3.1 Purpose of the Research . . . . . . . . . . . . . . . . . . . . . 25
3.2 Objectives and Methods . . . . . . . . . . . . . . . . . . . . . 25
4 Physical Model
4.1 Introduction . . . . . . . .
4.2 Local Transport Equations
4.3 Two-Phase Model . . . . .
4.3.1 Mass Conservation
.
.
.
.
.
.
.
.
i
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
29
29
31
32
32
ii
Contents
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
36
47
49
51
52
59
60
60
65
65
65
66
70
70
5 Numerical Method
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Finite Volume Method . . . . . . . . . . . . . . . . . . . . . .
5.2.1 Discretization of the Macroscopic Equations . . . . . .
5.3 Specic Issues . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Discretization Scheme for Advection . . . . . . . . . .
5.3.2 Viscous Term . . . . . . . . . . . . . . . . . . . . . . .
5.3.3 Phase Mass Fluxes . . . . . . . . . . . . . . . . . . . .
5.3.4 Flow Regime Function . . . . . . . . . . . . . . . . . .
5.3.5 Darcy Function . . . . . . . . . . . . . . . . . . . . . .
5.3.6 Pressure-Velocity Linkage With Collocated Variable Arrangement and Pressure Term Scaling . . . . . . . . . .
5.3.7 Iterative Solution Scheme . . . . . . . . . . . . . . . .
5.4 The Linear Equation System and Its Solution . . . . . . . . .
73
73
74
75
77
77
87
88
90
90
4.4
4.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
93
94
94
Contents
iii
6.3
iv
Contents
201
B Thermophysical Properties
207
C Experimental Measurements
209
Bibliography
212
List of Figures
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
4.1
4.2
4.3
4.4
4.5
4.6
5.1
5.2
5.3
5.4
5.5
5.6
5.7
. 30
. 39
. 40
. 54
. 54
. 59
. 74
. 81
. 83
. 85
. 86
. 88
. 91
vi
List of Figures
5.8
5.9
6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
6.9
6.10
6.11
6.12
6.13
6.14
6.15
6.16
6.17
6.18
6.19
Rectangular enclosure with imposed temperatures and concentrations at the vertical side walls, a prototypal case of natural
convection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Thermosolutal convection in the liquid pool of a DC casting. .
Illustration of relative sizes of thermal, solutal and velocity
boundary layers in thermosolutal natural convection in liquids.
Computation of ow in DC casting with two discretization
schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Grid types for the natural convection computations. . . . . . .
Case 1b: Concentration and ow elds at Sc = 1.0, and RaC =
108 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Case 1b: Grid convergence of velocity proles and comparison
with a reference solution. . . . . . . . . . . . . . . . . . . . . .
Case 2: Temperature, concentration and ow elds. . . . . . .
Case 2: Grid convergence of temperature and concentration
proles at cavity mid-width (x = 0.5). . . . . . . . . . . . . .
Case 2: Grid convergence of velocity proles. . . . . . . . . . .
Case 3: Evolution of temperature and ow elds in the cavity.
Case 3: Grid convergence of the FVM solution and comparison
with the SM solution. Time evolution of the Nusselt number. .
Case 3: Grid convergence of the FVM solution and comparison
with the SM solution. Velocity proles. . . . . . . . . . . . . .
Case 4: Concentration elds and streamlines in the enclosure
at local minima, maxima and inection points of a representative oscillation period. . . . . . . . . . . . . . . . . . . . . .
Case 4: Time evolution of the ow obtained on the 50 200
Gauss-Lobatto grid. . . . . . . . . . . . . . . . . . . . . . . . .
Case 5a: Temperature, concentration and streamlines in the
enclosure for the case of cooperating thermal and solutal forces
(Al-Cu). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Case 5a: Evolution of Sherwood and Nusselt numbers calculated on Gauss-Lobatto grids with 50200 and 100400 CV.
Case 5b: Temperature, concentration and streamlines in the
enclosure for the case of counteracting thermal and solutal
forces (Al-Mg). . . . . . . . . . . . . . . . . . . . . . . . . . .
Case 6: Grid convergence of the ow eld. . . . . . . . . . . .
109
113
114
116
119
122
122
125
126
126
128
129
130
134
135
137
138
139
141
List of Figures
vii
7.7
7.8
7.9
7.10
7.11
7.12
7.13
7.14
7.15
7.16
7.17
7.18
7.19
7.20
7.21
7.22
7.23
8.1
viii
List of Figures
8.2
8.3
8.4
8.5
8.6
8.7
8.8
8.9
8.10
8.11
8.12
8.13
8.14
8.15
8.16
8.17
8.18
8.19
8.20
List of Figures
8.21 Inuence of moving solid grains: Mixture velocity (~vm ) eld
and solid volume fraction in the billet at gsc = 0 (rigid solid
phase, Case G00). . . . . . . . . . . . . . . . . . . . . . . . .
8.22 Inuence of moving solid grains: Mixture velocity (~vm ) eld
and solid volume fraction in the billet at gsc = 0.30, dg = 75 m
(Case G30D075). . . . . . . . . . . . . . . . . . . . . . . . .
8.23 Inuence of grain size: Mixture velocity (~vm ) eld and solid
volume fraction in the billet at gsc = 0.30, dg = 25 m (Case
G30D025). . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.24 Inuence of grain size: Mixture velocity (~vm ) eld and solid
volume fraction in the billet at gsc = 0.30, dg = 150 m (Case
G30D150). . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.25 Inuence of consolidation solid fraction: Mixture velocity (~vm )
eld and solid volume fraction in the billet at gsc = 0.15, dg =
75 m (Case G15D075). . . . . . . . . . . . . . . . . . . . .
8.26 Inuence of consolidation solid fraction: Mixture velocity (~vm )
eld and solid volume fraction in the billet at gsc = 0.45, dg =
75 m (Case G45D075). . . . . . . . . . . . . . . . . . . . .
8.27 Longitudinal proles of solid volume fraction (gs ) along the
centerline. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix
. 189
. 189
. 190
. 190
. 191
. 191
. 192
List of Tables
4.1
4.2
Transport equations . . . . . . . . . . . . . . . . . . . . . . . 71
Constitutional laws and mixture relations . . . . . . . . . . . . 72
5.1
5.2
6.1
6.2
6.3
6.4
6.5
6.6
6.7
7.1
7.2
8.1
8.2
8.3
xii
List of Tables
B.2 Thermophysical properties of Al-5.00 wt%Mg. . . . . . . . . . 208
List of Symbols
Latin symbols
A
Ak
A
A
Ain
~b
cp
cR
cD
C
C0
Cref
C
dg
D
D
fexact
fi
fk
fsc
F
F~
F
Fs
~g , g
gk
gsc
gsD
GCI
xiv
h
hmix
H
I
I
jk
jkj
~jkd
~jkt
k
kp
K
K
K0
L
L
Lv
ll
ls
m
mL
mS
M
~i
M
~
M
k
~
M
k
~d
M
k
~p
M
k
~nk
ns
N
Nu
O( )
p
P
P
Pe
Pr
Q
List of Symbols
specic enthalpy, heat transfer coecient, grid distance
mixing enthalpy
cavity height
coecient for linear interpolation of CV face values
identity matrix
solute transport due to the movement of the phase interface
diusive solute ux across the interface
dispersion solute ux
tortuosity solute ux
thermal conductivity, interpolation coecient for ULTRA-QUICK
interpolation of CV face values
equilibrium partition coecient
permeability
permeability tensor
permeability constant
cavity width, macroscopic system scale
length scale of porosity variation
length scale of velocity variation
length scale of liquid microscopic structures
length scale of solid microscopic structures
mass
liquidus slope
solidus slope
second-order tensor used to represent ~vl
interfacial momentum source due to surface tension
momentum exchange due to phase change
momentum exchange due to interface stress
momentum exchange due to interface drag
momentum exchange due to interface pressure
normal vector of the interface pointing into phase k
number of grains per unit volume
buoyancy ratio (N = (lT T )/(lC C))
Nusselt number
order of leading error term of the discrete approximation
pressure, order of convergence
reduced pressure
dimensionless reduced pressure
Peclet number
Prandtl number
volumetric source term
xv
Q
Qw
Qk
Qqk
~qkd
~qkt
r
~r
r0
RaC
RaT
RDE
RDE band
R
~
S
SV
Sc
Sh
t
t
T
T0
Tcast
TL
Tref
TS
Tw
t
T
Tfg
u
u
U
v
~v
v
~v
vcast
~v il
xvi
vin
~vsys
V
V0
Vk
w
~i
~x
x
x
x
Xk
y
y
y
~z
~z0
List of Symbols
melt velocity at mold inlet
system velocity
vertical velocity scale
averaging volume
volume of phase k
velocity of the phase interface
position vector of the REV centroid
horizontal cartesian coordinate
dimensionless horizontal cartesian coordinate
discretization (grid) distance
phase function
vertical cartesian coordinate
dimensionless horizontal cartesian coordinate
discretization (grid) distance
position vector
reference level position vector
Greek symbols
k
kT
kC
ik
C
T
v
2
s
0
ref
k
ii
xvii
ij
Subscripts
AESex
C
D
E
e
EE
i
k
l
m
N
n
nb
NE
nee
NN
nn
nne
nnw
NW
P
ref
Rex
s
S
s
SE
see
xviii
SS
ss
SW
U
W
w
WW
List of Symbols
second south node in FVM discretization stencil
south CV face of CV S
southwest node in FVM discretization stencil
upstream node in FVM interpolation
west node in FVM discretization stencil
west CV face
second west node in FVM discretization stencil
primary solid phase
Superscripts
volume average
intrinsic volume average in phase k
deviation from intrinsic volume average
Acronyms
AES
AESex
CV
DC
DSC
FVM
LSRE
OES
OOC
Rex
SM
Introduction
1.1
The use of aluminum and its alloys has been rapidly expanding in the last
few decades. The annual world production of primary aluminum grew from
around 10 million tons in the 1970s to 15 million tons in 1994, reaching 23
million tons in 2005 [IAI, 2006]. The main applications of aluminum products
are found in transport (aerospace, rail, automotive, marine), building, packaging (foil, cans) and electrical power transmission. New applications (e.g.
the increasing use of aluminum in the automotive industry), more stringent
demands as regards quality, as well as growing demands relating to process
yield, energy eciency and recycling in production are posing new challenges
to production processes and require their further development and optimization.
Rolled and extruded aluminum products constitute the majority of all
aluminum alloys in use. In the production chain of rolled and extruded products the starting stock is a direct chill (DC) cast sheet ingot or an extrusion
billet, which gives a ne-grained material suitable for forming. In vertical DC
casting, shown in Figs. 1.1 and 1.2, molten metal is poured from the furnace
through a troughing system onto a mold table containing molds with the
cross-section of the casting. At the beginning of the batch process, a starting
head mounted to a hydraulic ram keeps the mold closed. When the metal lls
the water-cooled mold, the starting head is lowered at a controlled rate. As it
lowers, water jets below the mold spray water onto the casting (direct chill)
and complete the solidication. During DC casting many defects can occur
which are a direct consequence of the transport phenomena taking place in
the process. Severe defects can range from cracks, hot tears and warping as a
consequence of the residual stresses that build up due to uneven contractions
during solidication through to shrinkage-induced porosity.
A less serious but inevitable defect that occurs in solidifying alloys is
macrosegregation, an inhomogeneous macroscopic distribution of alloy components in the solidied casting. Chemical inhomogeneity can signicantly
1
Introduction
Molten metal
transfer trough
Mold
Liquid
Mushy zone
Cooling water spray
Solid
Ingot
Bottom block
Hydraulic ram
Figure 1.2: A DC billet casting machine in the Impol casthouse. Photo by Boidar arler.
1.2
1.2.1
During the solidication of a mixture the liquid phase is transformed into one
or more solid phases. The solubility of the mixture components in the solid
phases is usually not the same as in the liquid. In the case of a hypoeutectic
alloy, which is typical of most alloying elements used in wrought Al-alloys cast
by the DC process, the primary solid phase forms with a composition lower
than that of the liquid. This is shown in the equilibrium phase diagram in
Fig. 1.3, depicting equilibrium compositions of coexisting phases (denoted by
*) as a function of temperature. As the alloy of initial composition C1 = C2L
begins to solidify, the rst primary solid particle forms with composition
C2S
. Upon the formation of the low-concentration solid the excess solute
must be rejected into the liquid. We can assume that, during the course of
solidication, the interface of the growing solid particle is at thermodynamic
equilibrium [Kurz and Fisher, 1998], thus determining the local phase compositions from the equilibrium phase diagram. From the phase diagram we
can see that due to the gradual enrichment of the liquid with decreasing temperature the subsequently forming layers of solid have to have an increasing
solute concentration (shown in Fig. 1.4 on a simplied 1D model). As a consequence, the average phase concentrations are lower than the equilibrium
concentrations at the interface. In addition, a certain, although very small,
Introduction
Temperature [C]
650
2S*=2S
600
3S
2L*=2L
3L
3S*
4S*
3L*
4L
4L*
5
4S
550
500
450
0
0.1
0.2
0.3
Mass fraction Cu
0.4
Interface 4
Interface 3
Concentration
4L*
3L*
4L
3L
2L*
2 S*
0
3S *
3S
4 S*
4S
Figure 1.4: Evolution of the concentration profiles at the secondary dendrite arm scale.
The described processes of solute rejection take place on the scale of the
Figure 1.5: Coarse dendritic structure in a DC cast 2024 alloy billet. Courtesy of Impol d.d.
smallest structures of the growing solid. In an alloy the crystalline solid grows
in the form of dendritic grains, usually comprising two levels of branching
dendrites (Fig. 1.5). The representative structures for processes of local solute
partitioning are thus secondary dendrite arms with a typical spacing of the
order of 10 m.
1.2.2
Due to the solute rejection into the liquid on the scale of the growing grains
any macroscopically signicant relative motion of the solute-rich liquid and
the solute-lean solid grains causes the macroscopic transport of the solute
and thus macrosegregation in the solidied material.
Unlike pure materials which freeze at a xed melting temperature, alloys
solidify over a range of temperatures and phase compositions. In a solidifying system with moderate temperature gradients we can thus expect a
solidication region of a substantial size where the growing solid grains and
the surrounding liquid coexistthe mushy zone. If the solid grains in such
a region are grown together they form a porous solid skeleton (Fig. 1.6).
Given a sucient pressure head the liquid can ow through the porous solid
structure.
If the solid grains are not coalesced they can freely move with the ow
of liquid. Such a mushy zone behaves as a slurry of dispersed free-oating
grains (Fig. 1.7). Such grains can form through the fragmentation of den-
Introduction
Figure 1.6: Columnar mushy zone in a solidifying ammonium chloride solution [Huppert
and Worster, 1985]. Photo by Mark Hallworth, courtesy of M. Grae Worster, University
of Cambridge.
Figure 1.7: Equiaxed mushy zone in solidifying Al-20 wt%Cu, synchrotron X-ray radiography image [Mathiesen and Arnberg, 2006]. Courtesy of Ragnvald H. Mathiesen, SINTEF
Materials and Chemistry.
Introduction
transport the very slowly diusing solute (i.e. by advection) into thermally unstable regions and cause sharp, ne-scaled solutal structures
(due to the low diusivity) and complex interactions of the thermal and
solutal buoyancy forces [Hyun et al., 1995, Bergman and Hyun, 1996].
Motion of solid grains Free-oating grains in the slurry part of the mushy
zone move due to the density dierences between the solid and liquid
(settling or otation), due to drag forces exerted by the owing liquid,
and due to interactions with other grains.
Deformation of the solid skeleton The deformation of the solid skeleton causes the motion of the solid and the expulsion or suction of the
enriched interdendritic liquid. It can be caused by thermal stresses,
metalostatic head or external forces (as in compression from rolls in a
continuous steel slab caster)
1.2.3
Macrosegregation in DC Casting
extraction allowing the solid shell to reheat and partially remelt. In these
conditions the metalostatic head can force highly enriched liquid through
the permeable remelted shell and result in a thin layer of enriched metal at
the casting surface (exudation).
Liquid
(
us
f s=
uid
Porous
mushy zone
1)
So
Co
ns
ol
Grain
motion
lid
id
at
io
n
Slurry
us
(fs =
lim
it
(f
s=
f
Liq
Shrinkage
flow
Solid
10
Positive
macrosegregation
concentration
Introduction
Negative
macrosegregation
Cin
1.2.4
The diversity of the physical phenomena involved in the creation of macrosegregation and the complexity of their interactions makes numerical modeling
very challenging.
Multiple scales The transport phenomena leading to macrosegregation occur on dierent length and time scales. The most basic solute redistribution occurs on the length scale of dendritic structures ( 10 m) and
the time scale roughly identical with the solidication time, with the
solute partitioning originating in phase equilibrium relations valid at
the solid-liquid interface ( 1 nm [Kurz and Fisher, 1998]). The inter-
11
facial drag forces in the mushy zone originate in the uid dynamics at
the scale of the solid structures. Macroscopic heat, momentum and solute transport are all related to the length scale of the casting ( 1 m)
with the important observation that the largest gradients are found in
the mushy zone ( 1 cm). Another scale disparity is found in diusive
transport where the length scales of solute and momentum diusion
are much smaller than for heat diusion.
Different flow regimes Fluid ow occurs in the bulk liquid, the slurry, and
the porous regions. Regarding the system on the macroscopic continuum scale, dierent kinds of drag and viscous forces act on the liquid
in the dierent ow regimes.
Coupling of transport equations The transport equations are strongly
coupled via the buoyancy forces. The solidication dynamics are governed by all: heat, solute transport, and the uid ow.
Accuracy of numerical solution The accuracy of a solution is an important and dicult issue when solving predominantly advective transport
equations, which is the case for momentum and species transport in
metal alloys. More generally, accuracy is critical whenever steep gradients occur in an otherwise smooth eld. This occurs in the mushy
zone in a narrow region close to the liquidus, in the boundary layers,
and across the bulk liquid in the ow and species redistribution patterns arising from the unstable thermosolutal convection in liquid metal
mixtures.
The following techniques are used to cope with these challenges.
Volume averaging Due to the scale disparities of microscopic and macroscopic transport and the complex interface morphology of the dendritic structure it is impossible to solve the exact conservation equations on the microscopic scale for a macroscopic problem. To overcome
this problem we take advantage of the large-scale disparities. To model
the solidication region, averaged macroscopic equations are obtained
by averaging the microscopic equations over an averaging volume (representative elementary volumeREV) that is large compared to the
dendritic structure (i.e. large enough to make the average quantities
representative) and small compared to the scale of the system (i.e. small
enough to enable a continuum description of the average quantities). In
other words, ll , ls r0 L, where ll , ls , r0 , and L are the characteristic scales of the liquid and solid phase structures, the REV, and the
12
Introduction
macroscopic system, respectively. By averaging, microscopic phenomena (microscopic solute transport, drag forces) are no longer accounted
for. They have to be modeled using phenomenological models linked to
the averaged transport equations.
Literature Review
Be careful about reading health books. You may die of a misprint.
Mark Twain
2.1
14
Literature Review
15
needed to close the model. The rst of such models were developed by Beckermann and coworkers [Beckermann, 1987, Beckermann and Viskanta, 1988, Ni
and Beckermann, 1991,Beckermann and Viskanta, 1993] and by Ganesan and
Poirier [Ganesan and Poirier, 1990]. Their derivations are based on averaging
the usual conservation equations for mass, momentum and solute, which are
valid locally (on the microscopic scale of the phase structures) in each phase,
along with appropriate interface boundary conditions, over the continuum
scale of the macroscopic system. This is done by a formal volume averaging procedure [Drew, 1983]. This results in separate conservation equations
for each phase that have to be supplemented by constitutive models of the
interface interactions to close the model. The main advantage of the volumeaveraged two-phase models seems to be the more natural incorporation of
microscopic interphase phenomena and the clear connection of all equation
terms with the original local continuum equations. Many extended variants of
the aforementioned two-phase models were subsequently developed. Probably
the most prominent is the multiphase model by Wang and Beckermann [Wang
and Beckermann, 1993, Wang and Beckermann, 1996a, Wang and Beckermann, 1996b, Wang and Beckermann, 1996c]. It models the interdendritic
liquid as a separate phase, which enables the accounting of dierent solutediusion length scales in the solid, liquid and interdendritic liquid phases in
order to more rigorously model the eect of the dendritic grain structure.
Recently, Wu and coworkers [Ludwig and Wu, 2002, Wu et al., 2003] developed a version of the two-phase model that is simpler than the model by
Wang and Beckermann, but which includes a stochastic nucleation model to
account for the appearance of free-oating globular grains.
In the development of two-phase models many interfacial phase interactions, eective properties and other contributions (dispersion and tortuosity)
were modeled semi-empirically. Some of these simplications in the momentum equation were claried in a rigorous manner by Goyeau et al. [Goyeau
et al., 1997] and Bousquet-Melou et al. [Bousquet-Melou et al., 2002a]. They
rederived the equations using volume averaging, estimated the magnitudes
of individual terms by scaling analysis, and dened a local closure problem
for the uctuating variables (the dierence between the average and the local
value of the variable). The solution of the closure problem enables the determination of eective transport properties (permeability, inertia tensor etc.).
A similar task was performed by Neculae et al. [Neculae et al., 2002] and
Bousquet-Melou et al. [Bousquet-Melou et al., 2002b] for species transport.
16
2.2
2.2.1
Literature Review
Macrosegregation in DC Casting
Experimental Investigations
Macrosegregation in DC Casting
17
dendritic grains are those that were detached in a high cooling rate region,
suspended growth during transport to the bottom of the liquid pool and
continued to grow in situ. They hypothesized that the coarse grains in the
center grew in situ and coarsened due to the lower cooling rate.
In more recent investigations of the European EMPACT project, Joly
et al. [Joly et al., 2000], Lesoult et al. [Lesoult et al., 2001], and Daloz et
al. [Daloz et al., 2002] investigated macrosegregation and macrostructure in
grain rened and non-grain rened ingots using two dierent melt distributor
systems. They found that macrosegregation generally increases with grain
renement and linked this to the increased transport of free-oating coarse
(slowly growing) grains, formed either by the fragmentation of dendrites or
by nucleation on grain rener particles. In addition, they showed the strong
dependence of macrosegregation on the entry melt ow (distributor system).
This led them to contend from a detailed analysis of the macrosegregation
eld that, in addition to the eect of grain transport, there should be a
signicant impact of the interdendritic ow on the macrosegregation pattern,
possibly also aected by the inuence of the grain structure on the local
permeability of the porous mush [Daloz et al., 2002].
Eskin and coworkers [Eskin et al., 2004, Suyitno et al., 2004, Eskin et al.,
2005] very recently investigated the eects of alloy composition and dierent
casting parameters on the macrosegregation, structure, and casting defects
in DC cast billets in an attempt to complement their experimental investigations with somewhat simplied numerical simulations. They again conrmed
that casting speed increases macrosegregation in all regions across the casting cross section [Eskin et al., 2005]. The eect of casting temperature was
shown to aect macrosegregation only in the region close to the surface [Eskin et al., 2005], while the water ow rate did not exhibit a notable eect
on macrosegregation [Eskin et al., 2004]. Concerning the origin of negative
centerline segregation, they show, quite convincingly, that the structure of
coarse grains in the billet center can only be explained with their longer solidication trajectory in the presence of a slurry ow. However, they did not
nd a clear correlation between the fraction of coarse grains and the segregation intensity [Eskin et al., 2004, Eskin et al., 2005], which suggests that
the eect of the settling of solute-lean grains must be complemented by a
competing eect of the ow of the enriched melt.
2.2.2
Model Investigations
The rst to study macrosegregation in DC cast billets using a computational model were Flood, Katgerman and Voller [Flood et al., 1991] in 1991.
Their model assumed a rigid mushy zone moving at the casting speed and
18
Literature Review
Macrosegregation in DC Casting
19
tribution and that the packing fraction might not be uniform throughout
the two-phase region. This reveals a large source of uncertainty in an otherwise plausible model. Vreemans DC casting model was again later used by
Vreeman, Schloz and Krane [Vreeman et al., 2002]. They compared model
predictions to measurements on industrial-scale (450 mm diameter) DC cast
billets of an Al6 wt%Cu alloy trying to determine the value of the packing
fraction. While nding some agreement between the experimental and model
results they restated the problem concerning the uniform grain diameter and
packing fraction used in the model.
Jalanti [Jalanti, 2000, Jalanti et al., 2001] studied macrosegregation in
small (5 cm thick) and industrial-size (27.5 cm thick) Al-Mg ingots. Studying
the eects of thermal and solutal buoyancy and shrinkage ow, he contended
that the eect of shrinkage dominates and should be responsible for the almost whole macrosegregation eect in both small- and industrial-size ingots;
for the latter the results were compared to the EMPACT measurements [Joly
et al., 2000], showing good agreement. These conclusions should be treated
with some caution. In one case they were based on results obtained for small
geometries, where the ow induced by thermosolutal convection is relatively
weak and thus cannot penetrate the mushy zone and can indeed be outweighed by the shrinkage-induced ow. In the industrial-size case, the ow
was modeled using a viscosity increased by a factor of 100 (trying to mimic
turbulent ow) which importantly changes the fundamental ow characteristics (recall the discussion of natural convection in Section 1.2.2).
Very recently Du, Eskin and Katgerman [Du et al., 2006] again investigated the eects of buoyancy forces and shrinkage on macrosegregation in a
200 mm Al-4.5 wt%Cu billet, using a porous-mush model. Their results suggest that both the thermal and solutal components of the natural convection
ow have a signicant impact on the segregation prole, but the eect of
shrinkage ow competes with it and decisively alters the segregation in the
billet center (producing the negative segregation) and close to the surface
(inverse segregation in the subsurface region).
Gruber-Pretzler et al. [Gruber-Pretzler et al., 2006] made a very systematic investigation of the impacts of feeding ow, thermal and solutal convection, and shrinkage ow as well as of mushy zone permeability on macrosegregation in a DC cast bronze (Cu6 wt%Sn) billet, studying each eect
separately and in combination. Their results indicate that the shrinkage ow
decisively determines the macrosegregation pattern at low permeabilities of
the mush, when the convection ow cannot signicantly penetrate the mushy
zone; whereas at higher permeabilities it is combined in a more complex manner with forced and natural convection ows.
20
Literature Review
2.2.3
Summary
2.3
Thermal natural convection ow in low-Prandtl-number1 uids (as are liquid metals) is prone to instability, as was explained in the Introduction.
Experimental investigations are dicult since a visualization of the ow is
impossible in opaque metals. In numerical studies the physical complexity
is complemented with high requirements on the accuracy and stability of
numerical methods to obtain accurate solutions of the governing equations.
These may be the reasons that not many studies exist and their results are
not easily interpreted. Some information and basic ideas about the ow structures can be extracted from the early numerical work on natural convection in
low-P r uids by Mohamad and Viskanta [Mohamad and Viskanta, 1991] and
Refer to Section 6.3.1 for the definitions and explanation of dimensionless parameters
P r, Sc, RaT and RaC .
1
21
later by Cless and Prescott [Cless and Prescott, 1996a, Cless and Prescott,
1996b]. These works study natural convection in dierentially heated enclosures mostly at relatively low Rayleigh numbers (with one exception of
Ra = 105 in [Mohamad and Viskanta, 1991]) and very low Prandtl numbers
(down to P r = 103 ) and they also put a lot of emphasis on the impact of numerical methods on the results. Kamakura and Ozoe [Kamakura and Ozoe,
1996] later moved to higher Rayleigh numbers (up to Ra = 106 at P r = 102 )
and observed oscillatory and rotating multicellular ow structures.
Thermosolutal natural convection in the range of P r and Sc found in
liquid metal alloys (P r 102 , Sc 102 ) is not well explored. Numerical
investigations are scarce, in fact the only published numerical studies are
those of Hyun, Bergman and coworkers [Hyun et al., 1995, Bergman and
Hyun, 1996] and Shyy and Chen [Shyy and Chen, 1991]. While Hyun and
coworkers worked with liquid Al-Cu and Pb-Sn alloys, i.e. relevant P r and Sc,
their studies are limited to low Rayleigh numbers of only up to RaT = 5000.
The study of Shyy and Chen, on the other hand, reaches the interesting range
of Rayleigh numbers RaT 2106 , however it keeps a relatively low Schmidt
(Sc 10) and a relatively high Prandtl number (P r 101 in most cases).
In summary, there is no known investigation of thermosolutal ow in the
parameter range relevant for solidication during casting.
Our understanding of thermosolutal ow in high-P r uids is much more
profound. The work by Bennacer and Gobin [Bennacer and Gobin, 1996,
Gobin and Bennacer, 1996] investigates a range of parameters and studies
the eect on the ow structure and heat and mass transfer. It explains the
mechanism of thermal destabilization of solutally stratied layers and the
generation of multicellular thermosolutal ow structures. These studies can
be used as a basis for studying thermosolutal convection in low-P r uids,
although we have to keep in mind the fundamentally dierent relations in
thermal convection.
2.4
22
Literature Review
transport. The consequences are high local velocity gradients and thus potential numerical instability. In addition, the physical instability of the ow
can create unsteady ows, requiring relatively ne time discretization for a
proper resolution of the ow structure and its temporal evolution. All this
was already shown in the early work on thermal natural convection in liquid metals by Mohamad and Viskanta [Mohamad and Viskanta, 1991] and
Cless and Prescott [Cless and Prescott, 1996a, Cless and Prescott, 1996b].
They found the simulation results to be very sensitive to the accuracy of the
space (advection) and time discretization schemes employed, even aecting
the prediction of the qualitative ow structure (multicellular ow) and oscillatory ow. Although these ndings are not very recent it sometimes seems
that, in the context of casting simulations, they cannot be overemphasized
even today. The problem is accentuated by the lack of reference benchmark
solutions in the pertinent physical parameter range of natural convection.
Some recent activity on this topic [Bertrand et al., 1999, Hannoun et al.,
2003] seems set to raise awareness of the problems and hopefully boost the
development of a wider spectrum of relevant benchmark solutions. In their
development as well as in standalone testing of specic codes and solutions,
stringent verication and estimation of discretization errors is required using well developed verication techniques and concepts [Roache, 1998b, Roy,
2005, Oberkampf et al., 2003].
The most critical regions in any simulation are those with steep slopes
and fast changes of gradients. In solidication problems these typically occur
in the solidication region. At the transition from the fully liquid melt to
the mushy zone the velocities and concentrations experience fast changes. In
simulations of DC casting a common [Reddy and Beckermann, 1997, Vreeman and Incropera, 2000, Venneker and Katgerman, 2002, Vreeman et al.,
2002] and characteristic problem is an oscillatory concentration prole in
regions where the transitions of liquid-fraction are fast, i.e. close to the surface. This numerical eect was explained by Venneker et al. [Venneker and
Katgerman, 2000, Venneker, 2001]. A remedy was sought through the use
of high-resolution discretization schemes [Venneker, 2001], local mesh renement [Jalanti, 2000, Kaempfer, 2002] or mesh alignment with the liquidmush interface [Venneker and Katgerman, 2000], and recently both renement and alignment [Du et al., 2006, Gruber-Pretzler et al., 2006], mostly
using commercial CFD software [Jalanti, 2000, Venneker and Katgerman,
2000, Du et al., 2006, Gruber-Pretzler et al., 2006]. Fully automated selfadaptive grid renement has not yet been applied for macrosegregation simulations in DC casting. Nevertheless, the method developed in a preceding
project [Mencinger, 2001] would seem feasible since it combines local grid
renement in a relatively limited region with grid alignment to the steep
23
Objectives
3.1
3.2
26
Objectives
free-oating grains. To implement the model in a numerical code for
the simulation of DC casting.
To gain an understanding of ow patterns and heat and mass transfer
in thermosolutal natural convection in the liquid pool of DC castings.
To determine the accuracy of numerical solutions of ow and solute
transport relevant for the simulation of macrosegregation in industrialsize DC castings.
To investigate the inuence of the main DC casting parameters on
the resulting macrosegregation pattern. Here, the models predictive
capabilities are to be considered.
27
the mushy zone and their impact on macrosegregation were investigated. The eect of the presence of free-oating grains was qualitatively estimated by a parameter study of the implemented simplied
grain transport model.
Physical Model
Science may be described as the art of systematic over-simplification.
Karl R. Popper
4.1
Introduction
This chapter is dedicated to formulating the physical model and deriving the
corresponding governing model equations. The objective is to dene a onephase one-domain model capable of describing binary solidication during
DC casting. The model shall account for two distinct ow regimes in the
solidication zone of the casting, shown in Fig. 4.1. In the slurry regime the
solid is in the form of dispersed grains, which can move with the ow. The
ow behaves as a solid-liquid suspensiona slurry. In the porous regime the
solid phase is grown into a rigid matrix, moving with the system velocity
(e.g. the casting velocity in DC casting). There, the ow is regarded as a
one-phase liquid ow through a saturated porous medium. The purpose of
this modeling is to properly capture the eect the presence of a slurry zone
has on macrosegregation via the transport of solute-lean free-oating grains.
The model is constructed from the continuum description of the transport of mass, momentum, heat and species in the solidifying system and
of constitutive laws, which provide additional information on the diusion
transport, stresses, buoyancy forces, interfacial forces, thermodynamical interface conditions, species rejection, interphase heat and species transfer etc.
The constitutive laws are needed to provide the information that is needed
but not provided by the continuum equations themselves. Mathematically,
they are required to close the model.
Several alternative approaches can be used to derive a continuum model
of a multiphase ow. The mixture theory [Bennon and Incropera, 1987, Ni
and Incropera, 1995a], volume averaging [Drew, 1983, Ganesan and Poirier,
1990, Ni and Beckermann, 1991, Wang and Beckermann, 1993, Wang and
Beckermann, 1996a] and ensemble averaging [Furmaski, 1992,Banaszek and
29
30
Physical Model
Liquid
s(
0)
f s=
lim
(f
us
lid
id
at
io
So
ol
ns
Co
=1
Slurry
it
(
fs =
f
Liq
u
uid
Porous
Solid
Furmaski, 2005] were all used in the past to dene solidication models. In
this work the volume-averaging procedure was used because it provides, from
the authors viewpoint, a clearer relation between the original local equations
and the volume-averaged macroscopic equations. The rigorous procedure also
elucidates the phase interactions and the introduction of constitutive models.
To derive the continuum transport equations in the two-phase system, the
volume-averaging methodology is used for each of the interspersed solid and
liquid phases, resulting in a two-phase model. However, the fact that at the
start of its development the model implementation was based on the thermal
model code developed by Mencinger [Mencinger, 2002] presents historical
reasons for using a one-phase model. Hence, the objective of this work is to
dene a one-phase model and so the two-phase model is nally reshaped into
a one-phase model, dened in terms of mixture quantities. In the course of
this reformulation the interphase balances are taken into account and supplementary phenomenological (constitutive) models are formulated, accounting
for the interphase interactions.
31
4.2
The local continuum conservation equations for phase mass (continuum equation), heat (energy equation), momentum, and species are valid internally in
each phase. They are formulated on a scale that is large enough to accommodate the continuum concept, but is smaller than the scale of the phase
structures. Inside each phase these equations are identical with the classical transport equations for one continuous phase, which is modeled as a
(pseudo)uid.
The local continuum, momentum, energy, and species conservation equations, respectively, are
+ (~v ) = 0
(4.1)
t
(~v ) + (~v~v ) = p + + ~b
t
(4.2)
(4.3)
(4.4)
where is the density, ~v the velocity, p the pressure, the tangential stress
tensor, ~b the body force, h the specic enthalpy, k the thermal conductivity,
T the temperature, C the concentration (mass fraction of solute), and D the
diusion coecient.
The local heat and mass diusion uxes in both phases are modeled using
the Fourier and Fick laws, respectively. The tangential stress tensor requires
dierent approaches for the solid and liquid phases and therefore remains in
general notation here. All other assumptions about the constitutive laws are
left to be taken and justied during the course of derivation.
32
4.3
Physical Model
In this section the volume-averaging procedure, explained in detail in Appendix A, is performed on all governing local continuum equations to obtain
separate volume-averaged conservation equations for each phase, i.e. a twophase model. Supplementary constitutive models are introduced and, where
possible, the resulting governing equations are simplied based on an estimation of the scale relationships between dierent equation terms. Two
separate momentum equations are developed to account for the two dierent
ow regimes in the mushy zone, the slurry and porous regimes. The ow in
the bulk liquid (Navier-Stokes ow) and the motion of the bulk solid (uniform
casting velocity) are shown to be described by the slurry and porous regime
equations, respectively. The slurry-regime momentum equation reduces to
the Navier-Stokes equation at a zero solid fraction and the porous-regime
equation reduces to ~v = ~vcast at a solid fraction of unity. In this sense, the
four distinct ow regimes (liquid, slurry, porous and solid) are reduced to
two (slurry and porous). The other transport equations remain identical for
both regimes. Special care is taken with the two-phase equations, especially
with the momentum equations, to develop them in a form best adaptable to
the objective of combining them into a one-phase one-domain model, which
is done in Section 4.4.
4.3.1
Mass Conservation
The local continuity equation is constrained to phase k using the phase function Xk .
(
1 where ~x is in phase k at time t
(A.9)
Xk (~x, t) =
0 elsewhere
Xk
+ Xk (~v ) = 0
t
(4.5)
Volume-averaging of the equation is simply performed by using the summation rule of averaging (Eq. (A.4)) and applying the rules of volume-averaging
(Appendix A) on each term. hik denotes the intrinsic average of quantity
in phase k, V0 is the averaging volume, Ak the interface area of phase k,
w
~ i the phase interface velocity, ~nk the normal of the interface pointing into
phase k, gk the volume fraction of phase k, the local deviation of quantity
33
Two-Phase Model
from the average hik , and subscript k generally denotes phase k.
Z
1
k
Xk
=
gk hk i +
k w
~ i ~nk dA
t
t
V0
(4.6)
Ak
V0
k
k ~vk ~nk dA
Ak
D
Ek
+ gk k~vk
= gk hk i h~vk i
Z
1
k ~vk ~nk dA
V0
(4.7)
Ak
and thus the respective term in Eq. (4.7) is dropped. The last terms in
Eqs. (4.6) and (4.7) can be combined to give
Z
1
k (~vk w
~ i ) ~nk dA = ik ,
(4.9)
V0
Ak
4.3.2
Momentum Conservation
34
Physical Model
The local momentum equation for phase k is written using the phase
function Xk .
Xk
(4.11)
k
gk hk~vk i +
k~vk w
~ i ~nk dA
Xk (~v ) =
t
t
V0
Ak
(Eq.
0
D
E
*
k
k
k
gk k~vk
=
gk hik h~v ik +
t
t
Z
1
k~vk w
~ i ~nk dA
+
V0
Ak
Z
1
k
k
=
gk hk i h~vk i +
k~vk w
~ i ~nk dA.
t
V0
(4.8))
(4.12)
Ak
V0
k~vk~vk ~nk dA
Ak
vk
= gk hk i h~vk~vk i + k~vd
k~
Z
1
k~vk~vk ~nk dA
V0
k
Ak
Ek
D
Ek
+ gk hk i ~vk~vk
k
= gk hk i h~vk i h~vk i
(Eq. (4.8))
*
Ek 0 1 Z
D
d
+ gk k~v
vk
k~vk~vk ~nk dA
k~
V0
Ak
= gk hk i h~vk i h~vk i
Z
1
k~vk~vk ~nk dA.
V0
Ak
D
Ek
+ gk hk i ~vk~vk
(4.13)
35
Two-Phase Model
The pressure term is averaged as
hXk pi = gk hpk i
1
+
V0
pk~nk dA.
(4.14)
Ak
(4.15)
(4.16)
Ak
Ak
Ak
(4.17)
~
The interphase momentum transfer due to interfacial stress Mk is described
by the interfacial terms in Eqs. (4.14) and (4.15), the interfacial pressure
~p
force M
k
Z
1
~p
pk~nk dA = M
(4.18)
k
V0
Ak
~d
and the interfacial drag force M
k
Z
1
~ kd .
k ~nk dA = M
V0
(4.19)
Ak
1
V0
Ak
1
pk~nk dA
V0
Ak
1
k ~nk dA =
V0
Ak
~ p +M
~ kd = M
~ k
(pk I k )~nk dA = M
k
(4.20)
The additional dispersion term arising from the averaging of the velocity
uctuations is mathematically the variance of the velocity. The modeling of
36
Physical Model
the term is omitted for now and shall be discussed later. To compact the
notation it is customary to write it as a stress tk [Ni and Beckermann, 1991],
in analogy with the Reynolds turbulent stress in the modeling of turbulent
ow.
D
Ek
k
k
gk hk i ~vk~vk
= gk tk
(4.21)
gk hk ik h~vk ik + gk hk ik h~vk ik h~vk ik = gk hpk ik
t
D Ek
~+M
~ + gk t k (4.22)
+ gk h k ik + gk ~bk + M
k
k
k
~ k
M
1
=
V0
k~vk (w
~ i ~vk ) ~nk dA
Ak
~ k = M
~p+M
~ kd = 1
M
k
V0
4.3.3
t k
(pk I k ) ~nk dA
Ak
Ek
Ek
D
D
k
= hk i ~vk~vk k~vk~vk .
37
Two-Phase Model
Slurry Regime
In general, normal and tangential stresses are transferred inside the dispersed solid phase and stress gradients can exist. The stress state is described
by a stress tensor .
xx xy xz
(4.23)
= yx yy yz
zx zy zz
The average pressure in the solid phase hps is can be dened as part of this
tensor.
hps is
0
0
xx hps is
xy
xz
+ 0
hps is
0 (4.24)
yx
yy hps is
yz
=
s
0
0
hps is
zx
zy
zz hps i
(4.25)
(4.26)
(4.27)
In the liquid phase on the microscopic scale the general constitutive relation for the viscous stress tensor in a Newtonian uid is valid
2
l = l ~vl + (~vl )T l ( ~vl ) I,
(4.28)
3
38
Physical Model
(4.29)
where subscript m denotes the continuum mixture of solid particles and liquid, m is the dynamic viscosity of the slurry, and ~vm is the velocity of the
slurry mixture. The viscosity of the slurry depends on the volume fraction of
solid according to Thomas law, which is valid for gs 0.625 and for particle
sizes of 0.1 435 m [Kitanovski et al., 2005, Dhodapkar et al., 2006].
m = l 1 + 2.5gs + 10.05gs2 + 0.00273 exp(16.6gs )
(4.30)
Thomas law dependence of the mixture viscosity on the solid volume fraction
is shown in Fig. 4.2.
The body force term consists of the buoyancy force. The liquid buoyancy
force is
i
h
D El
~bl = ~g ref 1 T hTl il T ref C hCl il C ref ,
(4.31)
l
l
l
39
Two-Phase Model
m [Pa s]
0.2
0.1
0.05
0.02
0.01
0.005
0.002
0
gs
Figure 4.2: Mixture viscosity as a function of solid fraction according to Thomas law for
liquid viscosity of aluminum l = 1.3 103 Pa s.
T
C
where ref
s is the reference density of solid, and l and l the thermal and
solutal expansion coecients of solid, respectively. The momentum is trans~ ),
ferred between the phases via interfacial forces due to phase change (M
k
~ d ), and pressure (M
~ p ). The force balance at the solidinterfacial drag (M
k
k
liquid interface requires that
~ s + M
~ l + M
~ sd + M
~ ld + M
~ sp + M
~p+M
~i = 0
M
l
(4.33)
(4.35)
a result that will be used in the one-phase momentum equation for the slurry
regime.
The velocity of the solid phase that will be needed in the nal governing
equation system can be modeled in dierent ways, taking into account collisions of particles and other eects. For the initial model a simple approach
that was shown by Ni and Incropera [Ni and Incropera, 1995b] is chosen.
In their grain-velocity model they assumed that the predominant eect is
that of the buoyancy forces and interfacial momentum transfer (drag), and
neglected inertia eects, viscous eects and interfacial momentum transfer
40
Physical Model
0.1
0.2
0.3
0.4
g
0.5 s
Figure 4.3: Difference of velocities of solid grains and liquid according to Eq. (4.36) for
grain diameters of 150 m, 75 m, and 25 m (lines from top to bottom).
The grain size is modeled by a grain growth model that takes into account
the onset and growth of grains in aluminum melts in the presence of a grain
rener and is developed in Section 4.4.6.
Porous Regime
in the porous regime the solid phase forms a rigid and connected porous matrix which moves with a specied system velocity ~vsyst (casting velocity ~vcast in
the case of DC casting). To model the porous ow regime in the solidication
region, theoretically the same approach of adding the solid and liquid momentum equations to a one-phase model could be used. In this case, it would not
be possible to model stress terms in the solid ((gs hps is ) + (gs h s is ))
as a pseudo-uid. Instead, it would be necessary to use appropriate solidmechanics models to model the stresses. An alternative is to use only the
liquid-phase momentum equation and to model the interfacial forces using
principles used in the modeling of porous media [Bousquet-Melou et al.,
2002a]. In accordance with the assumption of a rigid solid matrix the solidphase momentum equation is at the same time simply reduced to
h~vs is = ~vsyst = const.
(4.37)
41
Two-Phase Model
(4.38)
The volume-averaged viscous stress term in Equation (4.22) can be expanded using the volume-averaging rules.
(gl h l il ) = h l i = hXl i
= l h(Xl~v )i + (Xl~v )T h~v (Xl )i h(Xl )~v i
2
I (h (Xl~v )i h~v Xl i)
3
h
iT 2
l
l
l
= l (gl h~vl i ) + (gl h~vl i ) I (gl h~vl i )
3
Z
Z
Z
1
1
2 1
l
~vl~nl dA +
~nl~vl dA I
~vl~nl dA (4.39)
V0
V0
3 V0
Al
Al
Al
To develop the interface integrals in the second divergence term in Eq. (4.39)
the velocity at the interface is expressed using Grays decomposition. For
position ~r at the interface it is dened ~vl |~r = h~vl il |~r + ~vl |~r . Here we are
confronted with the problem that the average quantities h~vl il |~r in the surface
integrals are evaluated for an averaging volume centered in ~r instead of ~x
like other averaged quantities. This can be overcome using a Taylor series
expansion of the quantity around the averaging volume centroid ~x [Goyeau
et al., 1997]. Goyeau et al. [Goyeau et al., 1997] showed that the higher order
geometrical moments in the integrated Taylor expansion are very small for
dendritic structures and can therefore be ignored. This means that
1
V0
Al
1
V0
Al
(4.40)
42
Physical Model
Al
Z
2 1
~nl~vl dA I
~vl~nl dA = h~vl il gl + gl h~vl il
3 V0
Al
Al
Z
2
1
2
l
I h~vl i gl +
dA. (4.41)
~vl~nl + ~nl~vl I ~vl ~nl
3
V0
3
1
~vl~nl dA +
V0
Al
Expression (4.41) is substituted into Eq. (4.39). The whole viscous term is
thus
2
k
T
(gl h l i ) = l (gl h~vl i) + [(gl h~vl i)] I (gl h~vl i )
3
2
l
l
l
l + h~vl i gl + gl h~vl i I h~vl i gl
3
Z
2
1
dA . (4.42)
~vl~nl + ~nl~vl I ~vl ~nl
l
V0
3
l
Al
Al
1
l ~nl dA =
V0
Z
2
T
l ~vl + (~vl ) l ( ~vl ) I ~nl dA
3
Al
(4.43)
Using Grays decomposition and a Taylor expansion for the interfacial velocities, as before, and again employing the justied approximation in Eq. (4.40)
after some manipulation we obtain
2
l
l
l
d
~
Ml = l h~vl i gl + gl h~vl i ( h~vl i )gl
3
Z
1
2
l
~vl ~nl + ~nl ~vl I( ~vl ) ~nl dA. (4.44)
V0
3
Al
Adding the bulk and interfacial viscous stress terms (Eqs. (4.42) and (4.44))
43
Two-Phase Model
after some arithmetical manipulation we obtain
1
l
l
2
d
~
(gl h l i ) + Ml = gl l h~vl i + ( h~vl i )
3
2
l
l
l T
+ gl h~vl i + ( h~vl i ) l ( h~vl i ) I (l )
3
Z
1
2
l
dA
~vl~nl + ~nl~vl I ~vl ~nl
V0
3
Al
Z
2
1
~vl ~nl + ~nl ~vl I( ~vl ) ~nl dA. (4.45)
l
V0
3
l
Al
1
=
V0
(4.46)
Pl I ~nl dA.
Al
Pl I ~nl dA.
i
h
D El
~bl = ~g ref 1 T hTl il T ref C hCl il C ref
l
l
l
(4.31)
~p= 1
M
l
V0
Al
1
hPl i I ~nl dA +
V0
l
Al
1
Pl I ~nl dA = hPl il gl +
V0
Al
(4.47)
The interfacial force originating in the mass transfer due to the phase
change is
Z
1
~
l~vl (w
~ i ~vl ) ~nl dA.
(4.48)
Ml =
V0
Al
44
Physical Model
gl ref
g
gl hl il h~vl il + gl hl il h~vl il h~vl il = gl hPl il
l ~
t
1
l
l
l
2
hPl i gl + gl l h~vl i + ( h~vl i )
3
2
l
l T
l
+ gl h~vl i + ( h~vl i ) ( h~vl i ) I (l )
3
i
h
l
l
l
ref
ref
C
T
hC
i
C
+
hP
hT
i
+ gl ref
~
g
l
l i gl
l
l
l
l
Z
Z
1
1
2
+
dA
~vl~nl + ~nl~vl I ~vl ~nl
Pl I ~nl dA l
V0
V0
3
Al
Al
Z
2
1
~vl ~nl + ~nl ~vl I( ~vl ) ~nl dA
l
V0
3
Al
Z
D El
1
l
(4.49)
l~vl (w
~ i ~vl ) ~nl dA gl hl i ~vl~vl
+
V0
Al
45
Two-Phase Model
viscous stress terms, which is consistent with the usual Boussinesq assumption used in buoyancy-driven ows. The following equation, identical to the
equation derived in [Bousquet-Melou et al., 2002a], as already mentioned, is
obtained in this way.
gl l h~vl il + gl l h~vl il h~vl il = gl hPl il + gl l 2 h~vl il
t
i
h
l
l
ref
ref
C
ref
T
+ l hCl i C
gl l ~g l hTl i T
Z
Z h
i
1
1
+
l~vl (w
~ i ~vl ) ~nl dA +
Pl I l ~vl ~nl dA
V0
V0
Al
A
| l
{z
}
Darcy-Forchheimer term
Z
D El
1
(4.50)
l
~vl~nl dA gl l ~vl~vl
V0
{z
}
|
A
|
{z l
}
Dispersion term
2nd and 3rd Brinkman correction terms
After the solution of the closure model the interface integral terms result in
the Darcy term with the Forchheimer correction, the 2nd and 3rd Brinkman
correction terms, and the dispersion term, all known from porous media ow
models. In the equation below we take into account that the solid porous
matrix moves with a uniform nonzero velocity h~vs is .
gl l h~vl il + gl l h~vl il h~vl il = gl hPl il + gl l 2 h~vl il
t
i
h
l
l
ref
ref
C
T
hC
i
C
hT
i
T
+
~g ref
l
l
l
l
l
Z
1
+
l~vl (w
~ i ~vl ) ~nl dA + gl2 l K 1 (I + F ) h~vl il h~vs is
V0
{z
}
|
Al
Darcy-Forchheimer term
+l h~vs is h~vl il gl + l 2 gl h~vs is h~vl il
{z
}
|
2nd and 3rd Brinkman correction terms
i
h
l l
l
s
s
T
gl l h~vl i h~vs i M M h~vl i h~vs i
(4.51)
|
{z
}
Dispersion term
46
Physical Model
length scale of porosity L and length scale of velocity Lv are similar. Based
on numerical results we can say that this is not the case in the high-fl region
of the mushy zone, where the velocity gradients are large (or L Lv in the
language of the scale estimates). In any case, this situation is not of much
interest in the two-regime model since the porous zone starts with lower fl .
In the lower-fl regions, on the other hand, where L Lv , the pressure and
Darcy terms will dominate by far. Thus the Brinkman terms are not expected
to have a notable inuence and will be neglected. The Forchheimer correction
term can be, as shown above, modeled as a modication of the permeability tensor. Bousquet-Melou et al. [Bousquet-Melou et al., 2002a] show that
the Forchheimer term is the dominant inertia term for moderate and small
porosity. However, from a rough estimation of the relative scales of K and F ,
based on formulae, found in [Whitaker, 1996] and [Dong and Dullen, 2006],
we get estimates of the order of magnitude kF k/kKk 104 . Thus, the
Forchheimer correction can clearly be neglected. Further, Bousquet-Melou et
al. [Bousquet-Melou et al., 2002a] also show that the dispersion term and
the inuence of interfacial momentum transfer due to phase change can be
neglected. Although the porous mushy zone is clearly not an isotropic porous
medium, at this stage of development and due to lack of detailed data, we
will model the mushy zone as an isotropic porous medium and simplify the
permeability tensor K to a scalar permeability K. Finally, we obtain the
simplied liquid momentum equation for the porous regime.
gl l h~vl il + gl l h~vl il h~vl il = gl hPl il + gl l h~vl il
t
i
h
l
l
l
s
2 l
ref
ref
C
T
ref
+ gl
+ l hCl i C
h~vl i h~vs i ~g l l hTl i T
K
(4.52)
47
Two-Phase Model
4.3.4
Energy Conservation
Z
1
k
gk hk hk i +
hk w
~ i ~nk dA
Xk (h) =
t
t
V0
Ak
(Eq. (4.8))
0
E
D
*
k
k
k
gk k hk
gk hk i hhk i +
=
t
t
(4.53)
Z
1
+
hk w
~ i ~nk dA
V0
Ak
Z
1
k
k
gk hk i hhk i +
hk w
~ i ~nk dA
=
t
V0
Ak
Z
1
k
hXk (h~v )i = gk hk hk~vk i
k hk~vk ~nk dA
V0
Ak
(Eq.
0
D
*
E
Ek
D
k
k
k
d
= gk hk i hhk~vk i
+ gk k h
vk
k~
(4.8))
V0
Ak
k hk~vk ~nk dA
D
Ek
k
k
k
k
= gk hk i hhk i h~vk i + gk hk i hk~vk
Z
1
k hk~vk ~nk dA
V0
Ak
(4.54)
48
Physical Model
(Eq.
0
*
D
E
k
k
k
dk
= gk hkk i hTk i + gk kk T
(4.8))
V0
kk Tk~nk dA
Ak
Z
1
Tk~nk dA
= gk hkk ik hTk ik hkk ik
V0
Ak
Z
1
kk Tk~nk dA
V0
Ak
(4.55)
t
+ Qk + Qk + ~qk + gk ~qkd
(4.56)
Qk
Qqk
~qkt
1
=
V0
Ak
1
=
V0
k hk (w
~ i ~vk ) ~nk dA
Z
kk Tk~nk dA
Ak
1
= hkk i
V0
k
Tk~nk dA
Ak
Ek
D
D
Ek
d k
k~vk .
k~vk k h
~qk = hk ik h
Qk is the energy exchange due to the movement of the phase interface, Qqk
is the diusive heat ux across the interface, ~qkt is the tortuosity ux, and
d k
~qk is the dispersion ux.
Two-Phase Model
4.3.5
49
Transport terms
The two least obvious terms to model in Eq. (4.56) are the terms involving
temperature, enthalpy and velocity uctuations; the tortuosity vector ~qkt and
the dispersion ux ~qkd . An insight into these terms is provided by the studies
of Neculae et al. [Neculae et al., 2002] and Bousquet-Melou et al. [BousquetMelou et al., 2002b]. There, the terms were investigated for solute transport
but the governing equation has the same mathematical form so the results
can also be used for the energy equation. They formulated a closure problem
which enables us to formulate the dispersion and tortuosity terms in terms of
the averaged macroscopic variable and a modication of the diusion tensor.
The numerical solution of the closure problem on schematic and real dendritic
structures then gave them the values of the modied diusion tensor. Using
their results it can be shown that for low Pclet numbers
h~vl il
(4.57)
Pe =
50
Physical Model
gradient. Moreover, the temperature is needed for the determination of possibly temperature-dependent thermophysical properties. Thus, a thermodynamic state model has to be provided which relates the temperature and the
enthalpy. In general, the enthalpy in a multiphase multicomponent system
is a function of the phase, temperature and the concentrations of the mixture components. If the thermodynamic reference state is dened to be the
primary solid phase at temperature Tref and concentration Cref = 0 (pure
solvent) the enthalpy of the primary solid phase at temperature T and
concentration C in a binary system is
h (T, C) = (1 C)href + Ch (Tref , C = 1) + hmix (Tref , C) +
ZT
cp (T, C)dT,
Tref
(4.60)
where href = h (Tref , Cref ) is the reference enthalpy, cp is the specic heat of
phase , and hmix is the mixing enthalpy. The enthalpy of the liquid phase
is
ZT
ZTf
cp (T, C = 0)dT + Lf + cpl (T, C = 0)dT
hl (T, C) = (1 C) href +
Tf
Tref
(4.61)
where Tf is the melting temperature of pure solvent and Lf the heat of fusion
of pure solvent. A common simplication can be applied, where the concentration dependence of the enthalpy is neglected. Thus the following simplied
relations for the enthalpies of solid and liquid phases are obtained.
hs (T ) = href +
ZT
(4.62)
cps (T )dT
Tref
hl (T ) = href +
ZTm
Tref
cps (T )dT + Lf +
ZT
cpl (T )dT
(4.63)
Tm
51
Two-Phase Model
4.3.6
Species Conservation
All terms in the species conservation equation are analogous to the terms in
the energy equation.
Z
1
k
k
gk hk i hCk i +
Ck w
~ i ~nk dA
Xk (C) =
t
t
V0
(4.64)
Ak
V0
+ gk hk i
Ck~vk
Ak
Ek
(4.65)
Z
1
hk ik hDk ik
Ck~nk dA
V0
Ak
Z
1
k Dk Ck~nk dA
V0
(4.66)
Ak
52
Physical Model
jk
jkj
1
=
V0
k Ck (w
~ i ~vk ) ~nk dA
Ak
1
=
V0
k Dk Ck~nk dA
Ak
~jkt = hk ik hDk ik 1
V0
Ck~nk dA
Ak
Ek
D
D
Ek
D Ek
~jkd = hk ik Ck~vk k Ck~vk .
jk is the solute transport due to the movement of the phase interface, jkj is
the diusive solute ux across the interface, ~jkt is the tortuosity ux, and
h~jkd ik is the dispersion ux.
4.3.7
(4.68)
(4.69)
We obtain an estimate of P e 20. Looking at the calculations by BousquetMelou et al. [Bousquet-Melou et al., 2002b] it is clear that the eect of dispersion will be very weak. When looking at the results in ref. [Bousquet-Melou
Two-Phase Model
53
et al., 2002b] we also have to note that the dispersion is modeled as a pseudodiusive ux via an augmentation of the macroscopic diusion coecient to
get an eective diusion-dispersion coecient. The eect of dispersion is assessed from the ratio of the augmented diusion-dispersion coecient and
the original diusion coecient. However, even with a strong augmentation
of the diusion-dispersion ux, i.e. more prominent dispersion at higher P e,
the impact of dispersion on the macroscopic species transport can still be insignicant compared to the undoubtedly dominant advection. The conclusion
of this discussion is that both dispersion and tortuosity can be neglected
D E
k
t
0
(4.70)
~jk + gk ~jkd
Microscopic Species Transport
To describe the relations of the local concentrations of individual phases
on the REV scale, which is needed in the macroscopic model, the species
transport at the scale of phase structures ( ) has to be modeled to capture
the pertinent transport processes. The structures relevant for microscopic
species redistribution are secondary dendrite arms. Diusion is the transport
mechanism responsible for species transport at this scale. To characterize
the microscopic species transport in dendritic microsegregation it is usually
analyzed on a small volume element, a control volume perpendicular to the
periodic dendrite arms [Kurz and Fisher, 1998], shown in Fig. 4.4. Based on
the geometry and scale of the control volume the following common assumptions are made:
The control volume is small enough to be in thermal equilibrium (see
thermal equilibrium assumption, Section 4.3.5)
The control volume is narrow; the concentration eld is assumed to be
one-dimensional
The geometry of the secondary dendrite arms is periodic and symmetry,
i.e. zero solute ux, is assumed at both ends of the control volume of
length 2 /2, where 2 is the secondary dendrite arm spacing
The phase interface is in thermodynamic equilibrium at solidication
rates encountered in normal casting and welding processes [Kurz and
Fisher, 1998]1
54
Physical Model
Liquid
l2
Solid
Interface 4
Interface 3
Figure 4.4: Control volume for microscopic species transport. 2 is the secondary dendrite
arm spacing.
Concentration
4 L*
3 L*
4L
3L
2 L*
2S*
0
3 S*
3S
4S*
4S
Figure 4.5: Evolution of the concentration profiles at the secondary dendrite arm scale.
55
Two-Phase Model
Mathematically the problem is formulated as a Stefan problem. The diffusion in the solid and liquid phases is modeled according to Ficks law of
diusion.
Ck
= Dk 2 Ck
(4.71)
t
At both ends of the control volume, i.e. the dendrite arm symmetry axes,
Neumann boundary conditions apply in either phase.
C
=0
(4.72)
x x=0
C
=0
x x= 2
(4.73)
Initially, the control volume is liquid, i.e. the initial position of the interface
is xi (t = 0) = 0 and the initial concentration of the liquid is uniform C(t =
0) = C0 . The equilibrium of both phases at the interface is described by the
condition
Cik = Ck (xi ) = Ck (T ),
(4.74)
where Ck (T ) is the equilibrium concentration of phase k at the local temperature T , denotes the thermodynamic equilibrium state, and subscript i the
solid-liquid interface. As the interface is advancing during solidication, the
velocity of the moving interface (vi = dxi /dt) is dened by the solute mass
balance at the interface
dx
C
dxi
C
i
s
l
= (Cl (T ) Cs (T ))
= Ds
(Cl (xi ) Cs (xi ))
D
,
l
dt
dt
x x=xi
x x=xi
(4.75)
which is often called the Stefan condition.
The microsegregation problem becomes more complex when eutectic reactions occur during solidication. Then the eutectic, a mixture of ne lamellar
primary and secondary phases, can be modeled as a pseudophase and equilibrium boundary conditions can be prescribed at all phase (and pseudophase)
interfaces. Such a complete description would thus model diusion in the liquid, growth of the eutectic at the liquid-eutectic interface and the dissolution
of the eutectic and back diusion in the primary phase at the eutectic-primary
interface. To the authors knowledge, a complete model like this has not been
developed yet as part of a coupled micro-macro solidication model. Instead,
In contrast to rapid solidification processes, where the interface is not in equilibrium
and solute trapping occurs. The range of validity of the equilibrium interface assumption
with respect to the solidification rate is analyzed in [Kurz and Fisher, 1998].
1
56
Physical Model
additional simplifying assumptions of the moving boundary problem (4.71)(4.75) are commonly taken to facilitate the consideration of the problem.
These assumptions include
The liquid diusivity is large enough (3-4 orders of magnitude larger
than that of the solid phases) to be able to assume a homogeneous
concentration in the liquid phase at normal solidication rates.
hCl il = Cil = Cl (T )
(4.76)
The back diusion in the primary solid phase after the onset of the
eutectic reaction, which causes dissolution of the eutectic, is not considered.
In the context of macroscopic solidication, where macroscopic relative
motion of solid and liquid phases is present, the additional eect of macrosegregation has to be taken into account. During the course of solidication
the interdendritic liquid is mixed with liquid owing from other regions of
the casting, which might be richer or leaner in solute. This causes an alteration of the relation between the averaged phase concentrations, compared
to solidication in the absence of macrosegregation. In terms of macroscopic
averaged variables this means a modied evolution of hCs il and T as a function of gs . If the rate of change of the overall concentration in the microscopic
control volume due to macrosegregation is dCm /dt, the general equation of
dendritic microsegregation is written [Zalonik and arler, 2003]
1 2Ds Cs
l
l
dCm = (1 fs ) dhCl i + (kp 1) hCl i dfs +
dt, (4.77)
{z
}
|
{z
}
|
gs 2 x x=xi
{z
}
|
change of solute
change of solute mass
mass in the liquid
where kp = Cs /Cl is the solute partition ratio, fs the solid mass fraction,
and gs the solid volume fraction. This equation attributes the whole solute
mass change in the control volume due to macrosegregation (dCm ), which
appears as a consequence of relative motion of segregated solid and liquid,
to the liquid phase. The solute content in the solid changes only due to its
growth from the liquid at the solid-liquid interface (phase change) and due
to diusion of solute from the interface, where the solid concentration is
highest, into the solid phase (back diusion). This means that the inuence
of a moving solid phase, as it is modeled in the slurry ow regime of the
model developed here, is not accounted for. Equation 4.77, exact in the case
57
Two-Phase Model
(4.78)
and integrated
1
(1 kp )
l
hC
Z li
d hCl il
C0
hCl i
Zfs
0
dfs
,
(1 fs )
(4.79)
to give
hCl il = C0 (1 fs )(kp 1)
(4.80)
C0 (1 fs ) hCl il
.
hCs i =
fs
s
hCl il
it+t
0
h
it [C ]t+t [C ]t
m
m
= hCl il +
,
t
1 [fs ]
(4.81)
hCl il
it+t
= hCl il
it+t
0
1 [fs ]t+t
1 [fs ]t
!(kp 1)
(4.82)
58
Physical Model
Local EquilibriumLever rule Another common, rather grave simplication is to assume local thermodynamic equilibrium of the solid and liquid
phases (equivalent to innitely fast diusion in both phases). This eliminates the need of modeling microscopic solute transport and requires only
thermodynamic equilibrium data.
hCs is = Cs (T )
hCl il = Cl (T )
(4.83)
The model study by Vuanovi [Vuanovi, 2001] claims that the eect of
the equilibrium assumption, if compared to the Scheil assumption, on the
macrosegregation prediction is not signicant. It thus presents a justication
for the equilibrium approach. However, the eect may be dierent for dierent
alloys (a larger partition ratio will increase the dierences) or applications
and should thus be further investigated to arm the evidence and its general
applicability.
In the present model the equilibrium assumption is employed for the
sake of simplicity. In the presence of moving solid grains, a model that can
handle the tracking of the solid phase should be included if a more detailed
microsegregation model (with nite or zero diusion in the solid) is used.
The present model is, however, implemented in the macroscopic model in a
way that enables a simple replacement with more complex microsegregation
models later on.
Thermodynamic state relations
The relations between state variables of a multiphase system in thermodynamic equilibrium are described by the equilibrium phase diagram. In the
modeling of macrosegregation this information is needed to describe the conditions at the phase interface, which is in thermodynamic equilibrium, as
shown before. A binary eutectic phase diagram, which is an adequate model,
suitable for all Al-rich binary aluminum alloys of interest (Cu, Mg, Si, Zn,
Mn, Fe and Ni), is shown in Fig. 4.6(a). The equilibrium (interface) and average phase concentrations at a temperature are shown schematically. The
same situation is shown in Fig. 4.6(b) on an enthalpy phase diagram, which
makes it easier to understand the enthalpy-concentration coupling used in the
model developed here. In most binary phase diagrams of binary aluminum
alloys the liquidus and solidus curves are close to straight lines. Therefore,
the common linearization of the phase diagram can be used and the solidus
and liquidus lines are given by
TS = Tf mS Cs
(4.84)
59
1100
2S*=2S
2L*=2L
3L
3*
3S S
4S*
600
3L*
4L
4L*
5
4S
550
Temperature [C]
650
1000
500
900
800
700 2S*=2S
1
2L*=2L
3L
3L*
4L
4L*
600
3S*
3S 4S*
4S
500
5
400
300
200
450
100
0
0.1
0.2
0.3
Mass fraction Cu
0.4
(a) Temperature-concentration.
0.1
0.2
0.3
Mass fraction Cu
0.4
(b) Enthalpy-concentration.
(4.85)
TL = Tf mL Cl ,
where TS and TL are the solidus and liquidus temperatures, mS is the solidus
line slope and mL the liquidus line slope. Thus in conjunction with Eq. 4.83
Cs = kp Cl ,
kp =
mL
.
mS
(4.86)
Combeau and Mo [Combeau and Mo, 1997] have demonstrated that the
linearization of the solidus and liquidus lines can have a noticeable eect
on the resulting solidication path. However, compared to the eect of the
much more crude assumption of local equilibrium, which is used in the present
model, the eect of the simplication of the phase diagram will be minor.
In any case, these days the most feasible way to model phase equilibria,
especially in multicomponent mixtures, is to use CALPHAD thermodynamic
models [Kattner, 1997], which is left for the future extension of the model
developed here.
4.4
The one-phase model is to be developed from the two-phase model and the
supplementary constitutional laws. It is formulated in terms of mixture quantities and properties, which are dened in terms of phase quantities and phase
60
Physical Model
fractions
m = gs s + gl l
hm = fs hs + fl hl
km = gs ks + gl kl
(4.87)
where subscript m generally denotes the solid-liquid mixture. The phase mass
and volume fractions (fk and gk respectively) are dened
mk
fk = P
k mk
Vk
gk = P
k Vk
gk =
m
fk
k
(4.88)
4.4.1
gs + gl = 1
(4.89)
Mass Conservation
we obtain
4.4.2
il + is = 0,
(4.90)
m
+ (m~vm ) = 0.
t
(4.91)
Momentum Conservation
Slurry Regime
The most convenient way of deriving the one phase- momentum equation
in the slurry regime appears to be the addition of the phase equations,
Eq. (4.22). In this way the rheological considerations are included in the
most natural way, as already mentioned in Section 4.3.3, and the need for
modeling of interfacial transfer terms is partly omitted. The interfacial force
balance, as already given in Section 4.3.3, is
s.
~ s + M
~ l + M
~ sd + M
~ ld + M
~ sp + M
~ p = i g
M
l
(4.35)
Further, phase pressure relations (Eq. (4.27)), the mixture viscous stress term
(Eq. (4.29)), and the Boussinesq buoyancy forces (Eqs. (4.31) and (4.32)) are
61
T
+
hC
i
C
l
l
l
l
T
ref
ref
gs~g s s T T
+ sC hCs is C ref
Es
D El
D
l
s
. (4.92)
gl hl i ~vl~vl
gs hs i ~vs~vs
(4.94)
where ~vcast is the system velocity with which the connected rigid solid phase
moves. In the DC casting process this is the casting velocity. The development
62
Physical Model
of all terms is quite straightforward, with the exception of the viscous term,
for which we obtain
l m
m
l
(~vm ) (fl )
gl l h~vl i =l (~vm ) 2
l
fl l
(4.95)
l m
+ 2 (~vcast ~vm )(2 fl 2fl fl ).
fl l
The scales of the three terms can be estimated using characteristic length
scales for velocity and porosity variation, Lv and L , respectively.
vm
m
(~vm ) l 2
(4.96)
l
Lv
vm 2Lv
l m
(~vm ) (fl ) l 2
(4.97)
2
fl l
Lv fl L
2
l m
vm
~vcast
Lv
2
(~vcast ~vm )( fl 2fl fl ) l 2
1
(4.98)
fl2 l
Lv fl2 L2
vm
l
If Lv < L and fl 1, which is true at the dendrite tips, the two latter ((4.97)
and (4.98)) will be smaller than the rst term and can be neglected. Deeper
in the mushy zone, however, the scales can be dierent (Lv L , fl 1) and
these terms can be important in relation to the rst viscous term. However,
the scales of these terms should be compared to the scales of all other terms
in the equation and we can say, that in the low-fl regions of the mushy zone,
the pressure, buoyancy and Darcy terms clearly dominate. Using the KozenyCarman model for the permeability of the porous mush [Dong and Dullen,
2006]
gl3
,
(4.99)
K = K0
(1 gl )2
they scale as
gl hPlhil
gl~g ref
l
gl
m
l
lT
l
K
T T
ref
(~vm ~vcast )
lC
hCl i C
ref
i
1 gl
gl
2
l
(vm vcast ),
K0
(4.100)
where the permeability constant K0 scales as K0 1010 . The terms (4.97)
and (4.98) will therefore be negligible in comparison with the viscous term
in the high-fl region and in comparison with the Darcy term in the low-fl
region of the mushy zone.
Keeping in mind the objective of combining momentum equations of both
regimes into a one-domain equation, the viscous term is reshaped to comply
63
with the form of the slurry-regime equation. There, the viscosity had to be
kept inside the divergence operator to account for viscosity variations which
occur with varying solid fraction. Keeping only the rst term on the RHS of
Eq. (4.95) as justied by the foregoing discussion, we reshape Eq. (4.95).
m
s
m
l ~vm l 1
(~vm ) (fl ) (4.101)
l (~vm ) =
l
l
l
The two terms on the RHS of Eq. (4.101) are scaled as
vm
m
l ~vm l 2
l
Lv
s
vm
Lv
l 1
(~vm ) (fl ) l 2
.
l
Lv 10L
(4.102)
(4.103)
Based on the dierence of length scales (Lv < L ) for fl 1 and on the
scale comparison with the pressure, buoyancy and Darcy terms for fl 1,
as before, the second term in Eq. (4.101) can be neglected.
After these considerations the following momentum equation is obtained
for the porous regime.
m
l
(m~vm ) + (m~vm~vm ) = gl hPl i +
l ~vm
t
l
h
i
m l
l
T
ref
C
ref
+ gl
(~vm ~vcast ) ~g ref
T
+
hC
i
C
l
l
l
l
l K
i
+ [ (m~vm~vm ) (gs hs is h~vs is h~vs is ) gl hl il h~vl il h~vl il
+
D
(m fs~vcast ) (4.104)
Dt
D
~ s
(m fs~vcast ) = ~vcast (gs s ) + ~vcast (gs s~vs ) = ~vcast is = M
Dt
t
(4.105)
64
Physical Model
This can be neglected using the same scaling argument as used previously
~ . The nal simplied one-phase equation for the porous regime is then
for M
l
m
l
(m~vm ) + (m~vm~vm ) = gl hPl i +
l ~vm
t
l
h
i
m l
l
T
ref
C
ref
+ gl
(~vm ~vcast ) ~g ref
T
+
hC
i
C
l
l
l
l
K
h l
i
+ (m~vm~vm ) (gs hs is h~vs is h~vs is ) gl hl il h~vl il h~vl il .
(4.107)
One-domain equation
To obtain a one-domain momentum equation, valid in the liquid, slurry and
porous ow regimes, the momentum equations for the slurry and porous
regimes are cast into one equation valid all over the domain by the penalization of the diering terms, an idea previously proposed by arler et al. [arler
et al., 2003,arler and Pepper, 2004]. For this purpose, the ow regime function is introduced
(
1 gs (~x, t) > gsc
R(~x, t) =
,
(4.108)
0 gs (~x, t) gsc
where gsc is the consolidation solid volume fraction, dening the transition
between slurry and porous ow regimes. In implementation of the numerical
model the zone function has to be represented by some smooth function,
which will be discussed in Chapter 5. The one-domain momentum equation
is then
T
+
~g ref
l
l
l
l
h
i
s
s
s
l
l
l
+ (m~vm~vm ) (gs hs i h~vs i h~vs i ) gl hl i h~vl i h~vl i
m l
l
(~vm ~vcast )
+ R gs hPl i + gl
l K
ref
+ (1 R) m (~vm )T + gs~g ref
s l
T
gs~g ref
s T T ref + sC hCs is C ref
. (4.109)
s
eff = R
m
l + (1 R)m .
l
(4.110)
4.4.3
65
Energy Conservation
The energy equation of the one-phase model is obtained by adding the solid
and liquid phase equations and taking into account the interfacial heat balance.
Qs + Ql + Qqs + Qql = 0
(4.111)
To facilitate the numerical solution, the diusive term is expressed in terms
of the leading variable hm . The correction needed in the solidication and
liquid zones is expressed as a correction term on the rhs of the equation.
km
km
hm
hm km T
(4.112)
(km T ) =
cps
cps
The nal equation is
km
km
(m hm ) + (m hm~vm ) =
hm
hm km T
t
cps
cps
h
i
s
s
s
l
l
l
+ (m hm~vm ) (gs hs i hhs i h~vs i ) gl hl i hhl i h~vl i .
(4.113)
4.4.4
Species Conservation
The species equation of the one-phase model is obtained by adding the solid
and liquid phase equations and taking into account the interfacial solute mass
balance.
(4.114)
js + jl + jsj + jlj = 0
(m Cm ) + (m Cm~vm ) = (m Dm Cm )
t
i
h
+ m Cm~vm gs hs is hCs is h~vs is gl hl il hCl il h~vl il
h
i
s
s
l
l
+ m Dm Cm gs hs i Ds hCs i gl hl i Dl hCl i
(4.115)
4.4.5
Microsegregation Model
The phase quantities (fs , fl , hhs is , hhl il , hCs is , hCl il ) and the temperature
T , which still appear in the mixture transport equations (4.91), (4.107),
(4.113), and (4.115) are determined from the leading quantities (hm and
Cm ) by the supplementary microsegregation model. The model consists of
descriptions of phase equilibria at the solid-liquid interface, thermodynamic
66
Physical Model
ZT
ZT
hm (T, fl ) = href +
cps (T )dT + fl Lf + (cpl (T ) cps (T ))dT (4.116)
Tref
Tf
hCl il (Cm , fl ) =
Cm
kp + fl (1 kp )
TL (hCl il ) = Tf + mL hCl il
(4.117)
(4.118)
If we assume that the specic heats cps and cpl do not depend on temperature
but are constant, a quadratic equation for fl is obtained by substitution of
variables in Eqs. (4.116)-(4.118)
Afl2 + Bfl + C = 0
(4.119)
A = (kp 1)Lf
B = (1 kp )(hm hfS ) kp Lf + (cpl cps )(Tf TL )
C = kp (hm hfS ) + cps (Tf TL )
hfS = href + cps (Tf Tref ),
where hfS is the enthalpy of the pure solid phase at the melting temperature
Tf . This equation is solved analytically for fl . Then the remaining unknowns
fs , hhs is , and hhl il can be directly determined from fl and T using Eqs. (4.89),
(4.62), and (4.63). The solid concentration hCs is is determined by the partitioning that follows from the inverse lever rule.
hCs is = kp hCl il
4.4.6
(4.120)
67
particles (e.g. TiB2 particles in Al-Ti-B reners, which consist of TiB2 and
Al3 Ti phases) into the melt. These particles act as nucleation sites (seed
crystals) for -Al grains, resulting in a uniformly ne, equiaxed cast microstructure.
Formerly, the onset of grain growth on inoculant particles was understood
as a stochastic nucleation process. The classical spherical-cap model of heterogeneous nucleation [Kurz and Fisher, 1998] was used. According to this
model, nucleation of a grain occurs when the crystal cluster (embryo) that
forms by stochastic atomic additions on the inoculant particle (which is an
energetically favorable site) in the form of a spherical cap, exceeds the critical
radius corresponding to the momentary undercooling. Once the stochastically
formed embryo is larger than the critical radius of the momentary undercooling, further growth leads to a decrease of the overall Gibbs free energy and so
the crystallite can freely grow. This kind of stochastic formation of embryos
is modeled with dierent models [Hecht et al., 2004].
A few years ago, Greer et al. [Greer et al., 2000] showed that this growth
onset model is not applicable for grain-rened melts. They developed the freegrowth model, which assumes that the embryo is formed on an inoculant
particle at very low undercooling. It can continue to grow as long as its
radius of curvature exceeds the critical radius. Since during its growth on the
inoculant particle the spherically growing crystal has to go through a radius
minimum, equal to the particle radius, the critical condition for free growth
of the crystal is then the size of the particle. This means that the free growth
of a crystallite on the inoculant particle begins as soon as the undercooling
is reached, where the corresponding critical radius is equal to the particle
radius. For usual grain rener particle sizes this happens much faster and at
much smaller undercoolings than the stochastic process can form embryos
large enough for further growth. The growth onset is thus deterministic and
is not controlled by the probability of stochastic formation of suciently large
nuclei, but by the free-growth barrier. The thermodynamics of this barrier
was investigated by Quested [Quested, 2004] and was shown to be controlled
mainly by the size of inoculant particles.
When modeling the growth of the initiated crystal grains, Greer et al. [Greer et al., 2000] justied the simplifying assumption of a locally isothermal
melt and showed that the growth is controlled by the solute diusion eld in
front of the solid-liquid interface. The solute-diusion controlled growth rate
of the spherical grain of radius r is
2 Ds
dr
= s ,
dt
2r
(4.121)
68
Physical Model
The relations in the solutal diusion layer producing the undercooling and
controlling the growth are shown in Figs. 4.5 and 4.6(a).
Using the free-growth model, Greer et al. [Greer et al., 2000] modeled
the performance of the grain rener by using an experimentally determined
distribution of inoculant TiB2 particles. The smaller particles need a larger
undercooling to initiate free growth. In their calculations they show that soon
after the growth onset on the largest inoculant particles (with diameters of
about 3 6 m, representing around 1% of all TiB2 particles), which starts
at undercoolings of the order of magnitude
Tfg =
4i
0.1 K,
SV dg
(4.123)
the latent heat release causes a temperature increase (recalescence) and thus
entirely blocks the activation of smaller particles. In Eq. (4.123) i is the
solid-liquid interfacial energy, SV is the entropy of fusion per unit volume,
and dg is the inoculant particle diameter.
A Simplified Treatment
From these ndings a rst approximation of a model of grain onset and
growth can be developed, whose primary aim is to model the grain size
needed in the solid-velocity model (Eq. (4.36)). Two simplifying assumptions
are extracted from the discussion of the free-growth model.
The range of inoculant particle size, which eectively initiates growth,
is rather limited and therefore the corresponding undercoolings needed
for free growth are in a relatively narrow range. As the order of magnitude of these undercoolings is small Tfg 0.1 K, they are thermally
certainly negligible. However, if the curvature undercooling and thus
the free-growth barrier are not modeled, we cannot model the activation of the particles using the inoculant particle size distribution as
in Greers model. Instead, I introduce a parameter of active particle
density. This is the number of particles per unit volume that actually
initiate growth. All active particles are assumed to initiate growth at
the local equilibrium liquidus temperature. The active particle density
can be determined experimentally from the grain density in the casting or by a simulation by Greers model, using the expected cooling
rate and inoculant particle size distribution of the grain rener. The
69
(4.125)
This means that in the framework of the present one-phase model the
grain growth could be modeled by a microsegregation model that includes a microscale solute transport model accounting for solute diusion in both phases. If the model is kept in the present simplied form,
adapted for the equilibrium assumption (Section 4.4.5), the solutal undercooling could also be modeled by modied solidus and liquidus lines
in the binary phase diagram, resp. a corrected solute partition coecient kp , such that
hCs is = kp hCl il ,
(4.126)
as proposed by Ni and Incropera [Ni and Incropera, 1995b]. Such a corrected phase diagram would then give the solid fraction from Eq. (4.124)
corresponding to the grain growth rate given by Eq. (4.121). Although
this is not entirely consistent with the grain growth mechanism, I do
not take into account the solutal undercooling in this rst version of the
model, but rather keep the local equilibrium assumption (Eq. (4.83)).
Later, this can be improved by a more intricate microsegregation model.
Following from these two simplifying assumptions, the local grain size dg
is simply determined from the local grain density ns (local density of active
particles), and the solid mass fraction fs .
s
6m fs
dg = 3
(4.127)
s ns
The solid mass fraction fs is determined by the microsegregation model,
described in Section 4.4.5. The local grain density ns is determined by the
grain transport model which uses the active particle density as a parameter
and is given in the following section.
70
4.4.7
Physical Model
Transport of Grains
(4.128)
where ns is the grain densitythe number of grains per unit volume. The
grain velocity ~vs is deduced from Eqs. (4.36), (4.37) and (4.87)
fl2 m (l s )d2g~g
~vs = R~vcast + (1 R) ~vm
,
(4.129)
18l m
where the viscosity of the slurry is as given before (Section 4.3.3).
m = l 1 + 2.5gs + 10.05gs2 + 0.00273 exp(16.6gs )
(4.30)
At the melt inlet the grain density is prescribed at the active particle density
napd . There is no generation of grains, only activation, which proceeds via
the solid fraction. Once fs exceeds zero, the grain diameter denes the grain
velocity, which is dierent for dierent grain diameters and thus dierent
across the slurry zone. In the liquid region, on the other hand, the virtual
grain density remains uniform as the solid velocity is equal to the liquid
velocity at a zero grain diameter.
4.5
Summary
71
Summary
(4.91)
l
(m~vm ) + (m~vm~vm ) = hPl i + (eff ~vm )
t
h
i
l
lT T T ref + lC hCl i C ref
gl~g ref
l
h
i
s
s
s
l
l
l
+ (m~vm~vm ) (gs hs i h~vs i h~vs i ) gl hl i h~vl i h~vl i
m l
l
(~vm ~vcast ) + (1 R) m (~vm )T
+ R gs hPl i + gl
l K
T
s
ref
ref
+gs~g s l gs~g ref
s T T ref + sC hCs i C ref
(4.109)
s
km
km
(m hm ) + (m hm~vm ) =
hm
hm km T
t
cps
cps
h
i
s
s
s
l
l
l
+ (m hm~vm ) (gs hs i hhs i h~vs i ) gl hl i hhl i h~vl i
(m Cm ) + (m Cm~vm ) = (m Dm Cm )
t
h
i
s
s
s
l
l
l
+ m Cm~vm gs hs i hCs i h~vs i gl hl i hCl i h~vl i
h
i
s
s
l
l
+ m Dm Cm gs hs i Ds hCs i gl hl i Dl hCl i
ns
+ (ns~vs ) = 0,
t
(4.113)
(4.115)
(4.128)
72
Physical Model
(4.87)
m
fk
l
(4.88)
fs + fl = 1
(4.89)
(
1 gs (~x, t) > gsc
R(~x, t) =
0 gs (~x, t) gsc
(4.108)
gk =
f 2 m (l s )d2g~g
~vs = R~vcast + (1 R) ~vm l
18l m
s
6m fs
dg = 3
s ns
(4.129)
(4.127)
m = l 1 + 2.5gs + 10.05gs2 + 0.00273 exp(16.6gs )
(4.30)
(4.110)
eff = Rl + (1 R)m
K = K0
gl3
(1 gl )2
hs (T ) = href +
ZT
(4.99)
(4.62)
cps (T )dT
Tref
hl (T ) = href +
ZTm
cps (T )dT + Lf +
hm (T, fl ) = href +
Tref
(4.63)
cpl (T )dT
Tm
Tref
ZT
ZT
cps (T )dT + fl Lf +
Cl (Cm , fl ) =
ZT
Tf
Cm
kp + fl (1 kp )
T (Cl ) = Tf + mL Cl
l
hCs i = kp hCl i
(4.116)
(4.117)
(4.118)
(4.120)
Numerical Method
n
So far I have paid piece rates for the operation x2 + y2 of about 18
pence per coordinate point, n being the number of digits. The chief
trouble to the computers has been the intermixture of plus and minus
signs. As to the rate of working, one of the quickest boys averaged
2000 operations x2 + y2 per week, for three digits, those done wrong
discounted.
Lewis F. Richardson
5.1
Introduction
In this chapter I describe the method of numerically solving the model equations developed in the previous chapter. To solve the continuum transport
equations the well-established nite volume method was used. I will only explain the basic features of the method and focus primarily on the specics of
the implementation of the method in the DC casting model. After a general
explanation of the method and the discretization technique, I discuss in detail
the discretization of the advection and viscous terms, the computation of the
phase velocities and phase mass uxes, the implementation of the ow regime
function, and the introduction of the Darcy function which enables the solution of the momentum equation in the porous regime at high gs . Further,
I comment on the specics of the pressure-velocity coupling scheme with a
collocated variable arrangement and a scaled pressure term and specify in
detail the SIMPLE-based iterative solution scheme for the coupled model
equations. Finally, I explain the approach to the solution of the linear equation system given by the discretized equations and give suggestions for its
improvement.
73
74
Numerical Method
Sn
Sw
n
P
wP
Se
Dy
s
Ss
Dx
Figure 5.1: Finite volume grid. Control volumes, control volume nodes () and boundary
~i ), grid distances (x, y)
nodes (), faces of the control volume P in compass notation (S
and interpolation stencil (yellow+blue).
5.2
The main principle of the nite volume method is the subdivision of the
problem domain into small polygonal control volumes (CVs) over which the
equations are integrated (Fig. 5.1). The integrations over each CV are approximated by numerical integration formulae using eld values represented
in computational nodes. This nally gives the discretized linear equation system that is then solved by an appropriate linear system solver. The volume
integrals of terms that contain a divergence term are converted to surface
integrals, using the divergence theorem (Gausss theorem). This means that
these terms are evaluated as uxes over the surfaces of each control volume.
Because the ux entering a given volume is identical to that leaving the
adjacent volume, the method is locally and globally conservative.
The method will only be described for orthogonal structured grids with
rectangular CVs, which are used in the developed numerical model. Moreover,
only specic approaches for the approximation of the integrations, gradients,
function interpolation etc., that were used in the developed numerical code,
are to be explained. The method can be extended to unstructured grids,
75
curved coordinate systems, general polyhedral CVs, higher-order approximation functions etc. More on these topics can be found in the literature
(e.g. [Ferziger and Peri, 1996, Peyret, 2000, Fletcher et al., 1991, Roache,
1998a]).
5.2.1
Transient term
()
d +
t
+ (~v ) = () +
| {z }
| {z }
Advection term
(~v )d =
Diffusion term
Q
|{z}
Source term
() d +
Q d
(5.2)
(5.1)
~,
F~ dS
(5.3)
the diusion and advection term are transformed into surface integrals over
the surface of the integration volume and we obtain
Z
I
I
Z
~
~
d + ~v dS = dS + Q d.
(5.4)
t
The transient term is integrated over the control volume using a secondorder approximation of the mean value in the CV by the value at the CV
center node. The time derivative can be approximated in dierent ways. We
use the implicit rst-order (implicit Euler) approximation with the time step
t.
Z
old
(P P old
P P )
+ O (x)2 + O(t)
d =
(5.5)
t
t
The overbar denotes the discrete approximation of the term for CV P, the
superscript old denotes values from the preceding timestep, is the CV
volume, x is the representative discretization length, and t is the timestep.
The advection term is integrated over the surface, i.e. the four faces of the
rectangular CV, usually denoted in compass notation as north (n), south
76
Numerical Method
(s), east (e) and west (w) faces. To perform the integration we need to
know the value of (~v ) everywhere on the CV face. Since this is not available
we have to use some kind of approximation. The rst level of approximation
is to approximate the ux integral in terms of variable values along the CV
face. The simplest for this purpose is the midpoint rule, where the integral
is approximated as the product of the integrand at the face center and the
face area, which is a second-order accurate approximation.
I
~ = (~v ) A
~ e + (~v ) A
~ w + (~v ) A
~ n + (~v ) A
~s
~v dS
e
w
n
s
+ O (x)2
(5.6)
where A is the area of the respective CV face. The mass uxes across the
CV surfaces are assumed to be known from the velocity eld solution and we
denote them
Fw
Fe
Fs
Fn
= (u)w Aw
= (u)e Ae
= (v)s As
= (v)n An .
Thus we write
I
~ = Fe e Fw w + Fn n Fs s + O (x)2 .
~v dS
(5.7a)
(5.7b)
(5.7c)
(5.7d)
(5.8)
Since the values of at the CV faces are not available, a second level of
approximation has to be used to obtain them by interpolation of the node
values. Numerous interpolation methods can be used. When the inuence of
advection in the transport problem is large compared to diusion, the choice
of the interpolation function has a key impact on the overall accuracy of the
FVM approximation. This is a very important issue in the problems tackled
with the model developed in this dissertation and will be explained in more
detail later on. For now, we leave the advective term as it is at the rst
approximation level.
77
Specific Issues
The diusion term is approximated using the midpoint rule and the central dierence scheme (CDS) approximation of the normal derivative.
I
~
Ae
Aw +
An
dS =
x e
x w
y n
S
As + O (x)2
y s
(5.9)
E P
P W
N P
= e
Ae w
Aw + n
An
xE xP
xP xW
yN yP
P S
As + O x(r 1) + (x)2
s
yP yS
{z
}
|
O((x)2 )
5.3
5.3.1
Specific Issues
Discretization Scheme for Advection
Numerical Method
78
um
eff
m l
R gl
um
l K
vm
eff
hm
m
cps
Cm
ns
Surface
force
p
x
Correction Flux
T
C
~g ref
(T
T
)
+
(C
C
)
ref
l
ref
l
l
l
ref
ref
+(1 R) gs~g s l
T
gs~gref
(T Tref ) + sC (C
C
)
s
ref
s s
m l
+R gl
(vm vcast )
l K
p
y
m Dm
79
Specific Issues
and momentum. Due to very low species diusion coecients and viscosities
in liquid metals and relatively large ow velocities it strongly dominates over
diusion. To be able to obtain a reasonably accurate numerical result we
therefore rely on sucient accuracy of the advection term approximation.
This is critical because the rate of grid convergence of the numerical result
towards the exact solution depends on the order of accuracy of the discrete
approximation. The numerical solution fi of a p-th order method on grid
spacing xi relates to the exact solution fexact by
fi = fexact + O ((xi )p ) .
(5.11)
With grid renement from grid spacing xi to xi+1 and the corresponding
solutions fi and fi+1 the discretization error then evolves as
p
xi+1
fi+1 fexact
=
.
(5.12)
fi fexact
xi
Eq. (5.12) clearly shows why the order of accuracy of the discretization
method is crucial for obtaining reasonable accuracy on reasonable grid sizes.
Generally, at least second-order accuracy is desired. The problem with secondand higher-order discretizations of the advection term is that they are only
conditionally stable. Godunovs monotonicity theorem [Murthy, 2002] proves
that linear monotonicity-preserving (not prone to oscillations) discretization
schemes are at best rst-order accurate.
The interpolation is always shown for the east face e. It is analogous for
face n.
Central Difference Scheme
The most straightforward way is to use the linear interpolation of two adjacent nodal values, which implies a locally linear variation of (Fig. 5.2(a))
e = Ie E + (1 Ie )P + O (x)2
(5.13)
xe xP
,
(5.14)
xE xP
where Ie is the linear interpolation coecient for CV face e. This is termed
the central dierence scheme (CDS). The Taylor series expansion can show
that its truncation error is O((x)2 ) (second-order accurate). The secondorder accuracy means a good accuracy performance of the CDS scheme and
makes it a good choice. The problem is that in the case of a strongly predominant advection the CDS scheme becomes numerically unstable and produces
oscillations in the solution. We thus seek a remedy. In accordance with Godunovs monotonicity theorem the most robust solution to the problem of
numerical instability is hence the use of a rst-order scheme.
Ie =
80
Numerical Method
(5.15)
(5.16)
81
Specific Issues
Fe<0
Fe<0
P?D
E
Fe>0
Fe>0
E
e
E?C
Fe>0
E?D
Fe<0
P?C
W?U
EE?U
EE
(c) QUICK
Figure 5.2: Interpolation of the CV e face value in the advection term with different
interpolation schemes.
(x) =
(5.18)
2c
2d
+ [sgn(Fe ), 0] Q2u
e EE + Qe E + Qe P
(5.19)
82
Numerical Method
2d
+ P Q1c
max[F
,
0]
+
Q
max[F
,
0]
Ae
e
e
e
e
1d
2c
+ E Qe max[Fe , 0] + Qe max[Fe , 0] Ae
+ EE Q2u
e max[Fe , 0]Ae ,
(5.20)
(xe xE )(xe xP )
(xE xW )(xP xW )
(5.21)
Q1c
e =
(xe + xE )(xe xW )
(xE xP )(xP xW )
(5.22)
Q1d
e =
(xe xP )(xe xW )
(xE xP )(xE xW )
(5.23)
1c
1d
(Q1u
e + Qe + Qe = 1)
Q2u
e =
(xe + xE )(xe xP )
(xE xEE )(xEE xP )
(5.24)
Q2c
e =
(5.25)
Q2d
e =
(5.26)
2c
2d
(Q2u
e + Qe + Qe = 1)
83
Specific Issues
NN
(i,j+2)
N
(i,j+1)
WW
(i-2,j)
w
(i-1,j)
Pn
(i,j)
S
E
e
(i+1,j)
EE
(i+2,j)
(i,j-1)
SS
(i,j-2)
Figure 5.3: Discretization stencil of the QUICK advection interpolation scheme in compass and index notation.
84
Numerical Method
been dened, dierent methods rely upon dierent sucient criteria. One
such philosophy is expressed by the ULTRA-SHARP ux limiter [Leonard
and Mokhtari, 1990]. This limiter function is used in conjunction with higher
order (usually polynomial) interpolation schemes with the objective of preserving the monotonicity of the original interpolation function, which may
cause overshoots. ULTRA-SHARP is an extension of the convective boundedness criterion (CBC) on which the well-known SMART scheme [Gaskell
and Lau, 1988], one of the earliest bounded high-order schemes, is based.
The ULTRA-SHARP limiter is best dened in the normalized variable diagram (NVD). This diagram, shown in Fig. 5.4(a), depicts the value of the
interpolated CV face variable in terms of three adjacent node values, an upstream (U), a central (C) and a downstream (D) node (Fig. 5.2(c)). The
normalized variable is dened as
U
=
D U
(5.27)
(5.28)
(5.29)
(5.30)
C e 1
(5.31)
e 0+
(5.32)
85
Specific Issues
~
fe
ru
CK
I
QU
de
2 nd
-or
CD
pw
ind
s
-ot
rd
er
up
w
in
d
~
fe
0
0
~
fC
~
fC
1
e = 1 + (C 1),
2
(5.33)
3
C .
2
(5.34)
The stable range dened by the ULTRA-SHARP limiter is shown in the NVD
in Fig. 5.4(a).
We apply the ULTRA-SHARP limiter to the QUICK interpolation scheme,
which is expressed in terms of normalized variables as
e = QD + QC C .
(5.35)
C 0
C
QD
0 < C SQ
S C
C
QD
D
e = QD + QC C SQ
,
(5.36)
< C 1Q
QC
C
1QD
< C 1
1
QC
C
1 < C
86
Numerical Method
f~
~
fe
1
~
S fC
LT
R
CK
I
QU
ASH
AR
lim
ite
~f C
K
b
a+ UIC
Q
AR
LT
~
fC
f~
Figure 5.5: The ULTRA-QUICK scheme in the normalized variable diagram (NVD).
where QC and QD are the QUICK interpolation factors from Eq. (5.21). The
ULTRA-QUICK scheme is shown in the NVD in Fig. 5.5.
In the developed code the ULTRA-QUICK interpolation is implemented
via interpolation factors, such that
e = kU U + kC C + kD D .
(5.37)
1 S
kU = 1 QD QC
0
kD = QD
C 0
QD
0 < C SQ
C
QD
C 1QD
<
SQC
QC
1QD
C 1
<
QC
1 < C
C 0
QD
0 < C SQ
C
QD
D
< C 1Q
SQC
QC
1QD
< C 1
QC
1 < C
(5.38)
(5.39)
87
Specific Issues
S
kC = QC
C 0
QD
0 < C SQ
C
QD
C 1QD
<
SQC
QC
1QD
C 1
<
QC
1 < C
(5.40)
The direction of the ow across CV face e is taken into account in the same
fashion as in the original QUICK scheme (Fig. 5.2(c)).
5.3.2
Viscous Term
The nonstandard part of the Newtonian viscous term m (~vm )T appears in Eq. (4.109) for the slurry ow regime. It was taken into account,
although it nally appeared to be signicantly smaller than (eff (~vm )).
The integration over the CV gives
u
u
I
~ e Ae x w Aw x
m (~vm )T dS
u
u
S
y
y
e
w
v
v
x
x
+ n An
v s As v
y s
y n
(5.41)
v
x
In
In
2
In
v
v
+ (1 In )
x N
x
P
v
v
v
v
(1 In )
+
+
+
x nne
x nnw
2
x e
x w
vNE vN
vN vNW
(1 In ) vE vP
vP vW
+
+
+
xNE xN xN xNW
2
xE xP xP xW
(5.42)
88
Numerical Method
NW nnw N nne NE
n
nee
W
w
SW
E
e
see
SE
Figure 5.6: Stencil for the discretization of (~vm )T in the viscous term.
u
y
Ie
Ie
2
Ie
u
u
+ (1 Ie )
y E
y
P
u
u
u
u
(1 Ie )
+
+
+
y nee
y see
2
y n
y s
(1 Ie ) uN uP uP uS
uNE uE uE uSE
+
+
+
yNE yE
yE ySE
2
yN yP
yP yS
(5.43)
This part of the viscous term is small compared to the other equation terms
and can be computed from the known preceding-iteration values of u and v
and combined with the source terms.
5.3.3
The advection terms in the transport equations involve phase quantities for
the solid and liquid phases: velocity, enthalpy, and concentration. They are
determined from the mixture quantities by the respective constitutive laws.
In addition, the phase advection terms involve mass uxes of the solid and
~l . Basically, their determination
liquid phases across the CV faces, F~s and F
from the mixture velocity or mass ux has to be based on the constitutive
law for the velocity of solid
fl2 m (l s )d2g~g
,
(4.129)
~vs = R~vcast + (1 R) ~vm
18l m
and the denition of the mixture velocity
~vm = fs~vs + fl~vl
(4.87)
89
Specific Issues
(5.44)
The solid face velocity is given by Eq. (4.129), evaluated at face n, and
depends on fl , m , dg , m , and nally ~vm . Expressing the relations between
mass and volume phase fractions
gk =
m
fk ,
k
(4.88)
(4.87)
(4.127)
(4.30)
(5.45)
we can see that we can reduce the problem of the determination of phase
uxes to the interpolation of gs and ns . The evaluation of the solid ux
follows the sequence
(5.46)
gs,n (gs )n
m,n
ns,n (ns )n
(5.47)
(5.48)
i
2
= l 1 + 2.5(gs )n + 10.05(gs )n + 0.00273 exp(16.6(gs )n )
s
6(gs )n
dg,n = 3
(ns )n
h
(5.49)
(5.50)
90
Numerical Method
i
h
Fs,n = R s (gs )n~vcast An
#
"
(gs )n (1 (gs )n )2 l s (l s )d2g,n k~g k
s (gs )n
Fm,n
An . (5.51)
+ (1 R)
m,n
18m,n m,n
The liquid mass ux is then simply
(5.52)
5.3.4
Fs,e = um s (gs )e
(5.53)
(5.54)
The momentum equation includes some terms that are only valid in one of
the two modeled ow regimes. To switch between these terms the ow regime
function R was introduced in Section 4.4.2.
(
1 gs (~x, t) > gsc
R(~x, t) =
(4.108)
0 gs (~x, t) gsc
A discontinuous step function could cause diculties during the numerical
solution, e.g. step cycling during the iterations. To preclude such problems,
we use a smooth but sharp transition function for the ow regime function.
It is shown in Fig. 5.7.
R(gs ) =
1
(1 tanh [cR (gsc gs )]) ,
2
cR = 100
(5.55)
The parameter cR sets the slope of the tanh function at the trigger value gsc .
5.3.5
Darcy Function
As the solid fraction gs approaches and nally reaches unity, all its terms
approach zero and the one-domain momentum equation, Eq. (4.109), degenerates. At gl = 0, the buoyancy, pressure, and Darcy terms, which are
multiplied by gl , become zero. The slurry-regime terms are zero, turned o
by the regime function multiplier (1 R). Further, all velocity gradients and
derivatives are zero in the solid phase, turning o the advective, viscous and
91
Specific Issues
R
1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
gs
+~g ref
l
lT
lC
92
Numerical Method
regime function.
(
gs
D(~x, t) =
0
(5.57)
1
1 tanh cD (gsD gs ) gs ,
2
gsD = 0.95,
cD = 500
(5.58)
D
1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
gs
(T
T
)
+
93
Specific Issues
multiplier
1
0.8
R
(1 R)
Darcy term
Buoyancy term
Pressure term
0.6
0.4
0.2
0.2
0.4
0.6
0.8
gs
Figure 5.9: Penalization functions for individual terms in the momentum equation,
Eq. (5.59), assembled of the regime and Darcy functions.
5.3.6
Pressure-Velocity Linkage With Collocated Variable Arrangement and Pressure Term Scaling
The momentum equation, an extension of the Navier-Stokes equation, involves a pressure gradient term that is not known in advance, but depends
on the velocity eld. For incompressible ows there is no separate equation directly describing the pressure eld. Nevertheless, the whole system of
model equations involves the same number of equations and unknowns. This
enables us to calculate the pressure eld indirectly, from the requirement of
mass conservation expressed by the continuity equation.
A standard way of doing this is by inserting the discretized expression
for the velocity from the momentum equation with the velocity eld from
the momentum equation solution into the continuity equation. The pressure
term is kept separate. In this way a pressure correction equation is obtained
giving the correction of the pressure that is necessary to adapt the velocities to fulll the continuity requirement. The solutions of the momentum
and pressure correction equation are iterated until the convergence of both
equations, when the velocity and pressure together fulll both the momentum and continuity equations. In the development of the pressure correction
equation I had to take into account that the pressure gradient term in the
momentum equation (4.109) is multiplied by (1 Rgs + D) and to keep the
scaling accordingly.
In the developed numerical code this kind of pressure-velocity coupling
is done with the SIMPLE method for a collocated arrangement of velocity
and pressure. Rhie-Chow momentum interpolation is used to cure potential stability problems appearing when using the collocated variable arrange-
94
Numerical Method
ment [Ferziger and Peri, 1996]. The reason for using collocated grids is that
the development of the numerical code was based on the DC casting code by
Mencinger [Mencinger, 2002], which was developed as a simplied version of
his adaptive-grid code [Mencinger, 2001]. The pressure-correction algorithm
was already described in detail by Mencinger [Mencinger, 2001], generalized
for r-adaptive curvilinear grids, and will not be repeated here.
5.3.7
The implicit iterative scheme for the solution of the coupled model equation
is based on the SIMPLE scheme. The owchart is shown in Fig. 5.10.
The iterative loop within each timestep is iterated until the convergence
criterion is fullled.
P
aP P nb anb nb bP
P
< errmax
(5.60)
max
ij
aP P + nb anb nb + bP
5.4
When we assemble all discretized terms from Eqs. (5.5), (5.8), (5.9), and
(5.10) for CV centered in point P and generalize the interpolation of CVface values of in the advection term and the approximation of gradients of
in the diusion term as being expressed in terms of the neighboring CV
node values (denoted nb), according to the interpolation stencil, we obtain
X
aP P
anb nb bP = 0.
(5.61)
nb
This is the general discretized form of the transport equation for the CV
centered in point P. It was obtained by integrating the transport equation
over the CV and applying dierent approximations to express the volume and
surface integrals. As mentioned before we use an implicit time integration.
This is done to stabilize the time integration of the strongly coupled equation
system within the model. The implicit formulation necessitates the solution
of a system of linear Eqs. (5.61). When we assemble the linear discretized
equations for all, say m n, CVs, we obtain a system of m n linear equations
Ab=0
A Rmnmn
Rmn
b Rmn
(5.62)
95
converg.
Y
converg.
Y
converg.
Y
converg.
Y
convergence
(Eq. (5.60))
err < errmax
t=t+Dt
96
Numerical Method
Table 5.2: Interpolation schemes used for discretization of the advection term in the
transport equations
Transport equation
Variable
Momentum (4.109)
um , v m
Energy (4.113)
hm
Species (4.115)
Cm
Continuity (Pressure) (4.109), (4.91) p
Grain density (4.128)
ns
Discretization scheme
QUICK
CDS
QUICK
CDS
ULTRA-QUICK
The best solution method depends on the structure of the matrix A. The
most eective methods often heavily rely on the known matrix structure.
The matrix structure in turn depends on which nb neighbors are included
in the discretized equations. Within the developed FVM formulation this now
depends exclusively on the interpolation functions used in the discretization
of the advection terms.
With the central-dierence scheme, which is used for the energy equation,
we obtain linear equations of the form
i,j
i1,j
i+1,j
i,j1
i,j+1
ai,j
ai,j
ai,j
ai,j
ai,j
bi,j
P
W
E
S
N
P = 0,
(5.63)
where on the structured orthogonal grid the neighbors are denoted in index
and compass notation (i, j) P, (i 1, j) W, (i + 1, j) E, (i, j 1) S,
(i, j +1) N (also shown in Fig. 5.3, where both compass and index notation
are shown). This results in a ve-band matrix with a nonzero central diagonal,
one nonzero band directly above and below the diagonal and two o-diagonal
nonzero bands one above and one below the diagonal.
With the QUICK scheme used for the momentum and species equations,
eight neighbors are included in the discretized equation, as shown in Fig. 5.3.
i,j
i2,j
i1,j
i+1,j
i+2,j
ai,j
ai,j
ai,j
ai,j
ai,j
P
WW
W
E
EE
i,j2
ai,j
SS
i,j1
ai,j
S
i,j+1
ai,j
N
i,j+2
ai,j
NN
(5.64)
bi,j
P
= 0,
(5.65)
This results in a nine-band matrix with four more o-diagonal nonzero bands,
two above and two below the diagonal.
A standard and very eective method which exploits the matrix structure of the CDS matrix is to use the iterative alternating-direction-implicit
tridiagonal-matrix-algorithm (ADI-TDMA). The basic TDMA (or Thomas
algorithm) is a direct method, basically a Gaussian elimination, specialized
for tridiagonal matrices, where it is very eective as it leaves out the operations involving the zero elements and makes a very compact scheme. The
97
TDMA is often used to directly solve the 1D diusion problem (or convectiondiusion, provided that a 3-cell scheme is used). To solve the CDS matrix
by TDMA the matrix has to be recast into tridiagonal form. This is done
by transforming the 2D problem on the structured grid into a stack of 1D
problems. If the 1D rows of CVs are running along, say, the x direction, the
terms involving the N and S nodes are added to the source term.
bi,j = bi,j + ai,j i,j1 + ai,j i,j+1
x, P
P
S
N
(5.66)
(5.67)
(5.69)
(5.70)
for the x- and y-direction sweeps, respectively. PDMA is less commonly encountered than TDMA; it is described for example in [Fletcher et al., 1991],
while a more optimized version was implemented in the DC casting code.
6.1
Introduction
6.1.1
Verication
6.1.2
An understanding of the ow structures induced by thermal and solutal natural convection is an important piece of the puzzle of modeling macrosegregation in DC castings. It is all the more important within the present work as
I hypothesize a signicant impact of ow structure in the liquid sump on the
macrosegregation patterns in DC castings (Ref. [Zalonik and arler, 2005a]
and Chapter 7). Thermosolutal natural convection ow and solute and heat
Introduction
101
6.1.3
Numerical Methods
6.2
6.2.1
Methods of Verification
Aims of Verication
103
Methods of Verification
were also used. Only the methods used in the present work are described
hereafter. More details and a more general and extensive overview of verication methods can be found in the two main references for verication
methods used in this text, namely the book by Roache [Roache, 1998b] and
the review paper by Roy [Roy, 2005].
6.2.2
where h3 , h2 , and h1 are the grid spacings from the coarsest to the nest,
and f3 , f2 , and f1 are the respective numerical solutions. Strictly speaking, the formal order of accuracy of the QUICK nite-volume discretization
scheme, which is used for advection-diusion of momentum and species, is
second-order, due to some minor simplications that were made by omitting
fi fexact
.
fexact
(6.3)
where fi is the solution on the ner grid, fi+1 the solution on the coarser
grid, r is the grid renement factor from fi+1 to fi and p is the observed
order of convergence. Fs = 3 is a safety factor based on the convergence of
a second-order method at a grid renement factor of 2 as a reference. The
GCI thus gives the relative error |(fi+1 fi )/fi | that would be obtained by a
second-order method with r = 2 if the estimated error of the ne-grid solution
(GCIi ) is the same. A third alternative of error estimation shows up when
using a benchmark solution or an extrapolated solution as a reference. In
this case fexact in Eq. (6.3) can be replaced by the benchmark solution fbench
or the extrapolated solution fRex , but in this case an error band of fbench
or fRex has to be added to the error estimate. Using classical Richardson
extrapolation the extrapolated solution is
fRex = fi +
fi fi+1
rp 1
(6.5)
105
Methods of Verification
Richardson-extrapolated solution fRex by
1 fi fRex
GCIRex = p
r 1 fRex
r=
hi+1
,
hi
(6.6)
where r is the grid renement factor between the coarse- and ne-grid solutions used in the Richardson extrapolation. Error bands of benchmark solutions are not always given but they can be estimated from the grid convergence data, according to Eqs. (6.2), (6.5) and (6.6).
Code-to-code comparison
A code-to-code comparison, which is performed for some cases, could also be
classied as code verication. However, it should be used with much caution
for this purpose [Trucano et al., 2003] if none of the codes can be strictly
dened as a reference and the reference code is not rigorously veried (as
for example in [Vertnik et al., 2006]). Thus, this step is regarded more as an
initial condence boost rather than a formal verication method.
6.2.3
0 h h1
(6.7)
107
(6.9)
Roy [Roy, 2005] shows that the dierence between the values of the GCI and
a respective RDE band are small and are also ordered in h. They can thus be
compared consistently.
The leading error term in oscillatory convergence does not behave in accordance with Eq. (6.1) (C could change sign or the higher-order error terms
are also signicant) and Eq. (6.2) cannot be used to determine the OOC.
Therefore, I tried to estimate the OOC using a power-law approximation
where the dependence of the relative discretization error RDEi (Eq. (6.3),
approximating fexact fAESex ) on grid distance h is approximated by a power
law.
RDEi = Chp
(6.10)
The OOC is then estimated by the power p in the least-squares t of the
power-law function.
6.3
6.3.1
(6.11)
gT T H 3
Pr =
(6.15)
,
D
(6.16)
gC CH 3
D
Sc =
109
C2
T1
C2
T1
Tinit=T1
Tinit=T1
DT=T1-T2
C1
DC=C1-C2
Cinit=C1
T2
Cinit=C1
C1
T2
d
N<0
N>0
Figure 6.1: Rectangular enclosure with imposed temperatures and concentrations at the
vertical side walls, a prototypal case of natural convection in a closed domain.
heated (and spiced) rectangular cavity with insulated top and bottom walls
(Fig. 6.1). The scaling analysis within the domain of the boundary layer and
horizontal discharge jet, shown in Fig. 6.1 relies on the basic scales
x T
yH
T T
C C.
(6.18)
The scales of velocity and time then stem from estimates of velocity and
boundary layer growth time and are deduced by scaling analysis. They are
dierent for dierent heat transfer regimes, i.e. conductive or convective,
and Prandtl number ranges, i.e. P r 1 or P r 1 1 . The fact that in the
literature they are not always used consistently with the parameter range of
interest is marginal but it somewhat obscures the relations of the terms in
the equations. The scales of velocity V , time and thermal boundary layer
thickness T in the dierent regimes are the following. In the conduction
regime
#
"
# "
V A2
U A
Pr 1
Pr 1
H
H
(6.19)
41
41
H2
H
1
(Ra
P
r)
Ra
1
T
T
T
2A
4A2
Note, that the discussion of the velocity and time scales here is confined to the purely
thermal situation. Its scales remain valid if the heat transfer is controlled by the thermal
buoyancy flow. The scales found in thermosolutal flows are elaborated, for example, in
reference [Bennacer and Gobin, 1996] for P r 1.
1
RaT4
H
1
H 2 21
RaT
T HRaT 4
1
1
U (RaT P r) 4
V (RaT P r) 2
H
H
1
1
H2
(RaT P r) 2
T H(RaT P r) 4
RaT2
H
"
"
#
Pr 1
(6.20)
#
Pr 1
(6.21)
RaT 4 1
(RaT P r) 4 1
(6.22)
~v = 0,
momentum equations
1 p
u
u
u
2
+
u
+
v
=
(Ra
P
r)
+ Pr
T
t
x
y
x
p
v
v
v
+u
+v
= + Pr
t
x
y
y
2
2 u
21 u
+ (RaT P r)
x2
y 2
(6.23)
2
2v
21 v
+ (RaT P r)
x2
y 2
( + N )
(6.24)
energy equation
2
2
21
+
u
+
v
=
+
(Ra
P
r)
,
T
t
x
y
x2
y 2
(6.25)
P r 2
21
+u
+v
=
+ (RaT P r)
,
t
x
y
Sc x2
y 2
(6.26)
C Pr
. As the scaling is consistent with
where the buoyancy ratio is N = Ra
Sc RaT
the heat and mass transfer regime, all dimensionless quantities in the above
equations are of the order of magnitude of unity. The relations between the
111
various transport terms can thus be inferred from P r, RaT , Sc, and RaC
alone, which will be used in Section 6.3.2. An exception, although unimportant here, is the species equation, where the pertinent diusion length is
much smaller C T and the scales of the dimensionless term thus deviate
from unity. The general relation with very small diusion terms remains clear
however.
Although the problems of main interest in this work fall in the convective
regime with P r 1, the scale estimates of Eq. (6.20) will be used for the
nondimensionalization of variables. Further, all dimensions are normalized by
the same lengthH. This is done simply to conform to the scales often used
in the literature and to retain the same scales throughout the text. Thus, the
dimensionless variables will be dened
(x, y)
H
1
1
= (RaT ) 2
(x , y ) =
6.3.2
(RaT ) 2 t
2
H
T T0
=
T
t =
1
H
(RaT ) 2 ~v
C C0
=
.
(6.27)
C
~v =
From estimates of known parameters of the DC casting problem and properties of metal melts an evaluation of the parameter range of RaT , RaC ,
P r, and Sc, that is of interest in investigating the ow in DC casting, can
be made. The evaluation is based on the estimates given below, which were
taken for an Al-Cu alloy (at around 5 wt.%Cu). The thermophysical properties of Al-Cu alloys give the most severe and thus the most conservative
estimates of RaT and RaC since they combine high alloying element concentrations (up to 5 wt.%Cu) with strong solute partitioning and the largest
solutal expansion coecient of all prominent alloying elements. On the other
hand, the estimates of P r and Sc can be considered safe for all Al-alloys as
they depend on properties that do not vary substantially for dierent alloys.
Also, considering the geometry and boundary conditions of the relevant liquid pool in the DC casting the estimates of RaT and RaC given hereafter are
safe 2 . In the liquid pool, the high temperature is imposed at the top and the
low temperature at the inclined liquidus front (Fig. 6.2). This arrangement
represents a conguration that should be considerably more stable than the
prototypal enclosure heated at the side walls, similarly to the eect caused
by inclining an enclosure as lucidly explained by Bejan [Bejan, 2004].
To refer to parameter estimates as safe or conservative in this context means that
they define an equally or more complex physical situation than the realistic parameters.
These cannot always be determined exactly or they float within a range, depending on the
casting parameters.
2
(6.28)
(6.29)
The implications of the low Prandtl number can be seen from the relations of the terms in Eq. (6.24), which is properly scaled for P r 1. The
buoyancy forces driving the ow are balanced by the inertial forces. The viscous eect of the wall is conned to a very thin momentum boundary layer.
This situation, occurring in low-P r metals, is fundamentally dierent from
the relations in high-P r uids, which is shown in Fig. 6.3. In high-P r uids
the driving buoyant force is conned inside the viscous layer and the frictioninertia force balance outside the driving heated layer causes a transition into
a stagnant zone. Contrarily, in low-P r uids the inuence of the driving force
extends with the heated layer beyond the near-wall region where the inuence of viscous friction is still important. This means that most regions of
the ow are governed by a balance of buoyancy and inertia. Intuitively, we
113
C0 Tcast
Liquid
Mushy zone
0)
f s=
lid
u
So
TLIQ
Cl > C0
s(
s(
fs =
1)
Liq
u
uid
Solid
can understand that such a ow is less stable than a high-P r ow. Mathematically, this means that the non-linearity of the governing equation system
becomes stronger with the relative amplication of the non-linear advection
term in the momentum equation.
The high thermal Rayleigh number implies a predominantly convective
heat transfer. Special attention is given since the inherent instability of lowP r ows amplies with larger RaT . Instabilities, triggering ow structures
with multiple ow cells and an oscillatory ow are known from experimental
and numerical studies to already occur at moderate RaT at low P r [Mohamad and Viskanta, 1991, Janssen and Henkes, 1993, Cless and Prescott,
1996a, Cless and Prescott, 1996b, Kamakura and Ozoe, 1996, Le Qur and
Gobin, 1999, Hannoun et al., 2003]. In their numerical study and review of
experimental results, Lage and Bejan [Lage and Bejan, 1991] recommend the
relation RaT 109 P r as an empirical threshold for the transition of laminar to turbulent ows in natural convection. From their numerical results
and theoretical analysis a lower threshold-RaT for the onset of the transition
would actually follow. But these lower values apparently (clearly in the case
Pr>1
Pr<1
v
v
dC
dT
dv
dC dn
dT
Figure 6.3: Illustration of relative sizes of thermal, solutal and velocity boundary layers
in thermosolutal natural convection in liquids.
Numerical Issues in Modeling the Fluid Flow in Liquid Metal Mixtures 115
limited to low Rayleigh numbers of only up to RaT = 5000. The study of
Shyy and Chen, on the other hand, reaches the interesting range of Rayleigh
numbers RaT 2106 , however it keeps a relatively low Schmidt (Sc 10)
and a relatively high Prandtl number (P r 101 in most cases). In summary, there is no known investigation of thermosolutal ow in the parameter
range relevant for solidication during casting.
The complexity of the physical phenomena in the buoyancy-driven ow
during casting, as is suggested by the foregoing scaling analysis, discussion
and literature review, shows that it could be dicult to analyze the thermosolutal ow in the parameter range of interest (Eqs. (6.28)(6.29)). It seems
more reasonable to break up the convection problem with respect to thermal
and solutal eects and the range of the physical dimensionless parameters
into simplied subproblems and to study the eects of the individual transport phenomena on these simplied cases before exploring the thermosolutal
convection problem for the nal relevant parameters. This approach might
help us understand more deeply the ow structures in the nal parameter
range and the nal DC casting problem. At the same time, it is believed
to clarify the code and calculation verication procedure. The stepwise approach of simultaneous ow studies and code/solution verication is dened
in detail in Section 6.5.
6.4
-0.1
-0.1
-0.1
-0.1
-0.2
-0.3
0.00
-0.2
0.05
0.10
0.15
-0.3
0.00
0.20
z [m]
z [m]
z [m]
z [m]
-0.2
0.05
r [m]
0.10
0.15
-0.3
0.00
0.20
-0.2
0.05
r [m]
0.10
0.15
-0.3
0.00
0.20
0.05
r [m]
0.10
0.15
0.20
r [m]
(a) First-order upwind scheme (mesh sizes, from left to right: 2777, 52152, 102302, 270502).
0
-0.1
-0.1
-0.1
-0.1
-0.2
-0.3
0.00
-0.2
0.05
0.10
r [m]
0.15
0.20
-0.3
0.00
z [m]
z [m]
z [m]
z [m]
-0.2
0.05
0.10
r [m]
0.15
0.20
-0.3
0.00
-0.2
0.05
0.10
r [m]
0.15
0.20
-0.3
0.00
0.05
0.10
0.15
0.20
r [m]
(b) ULTRA-QUICK scheme (mesh sizes, from left to right: 2777, 52152, 102302 (periodic flow at two
different times)).
Figure 6.4: Computation of flow in DC casting with two discretization schemes. Streamlines of relative velocity (~v ~vcast ) are shown.
excellent results [Zalonik and arler, 2005b]. The scheme is based on the
QUICK scheme and stabilized by the ULTRA-SHARP limiter [Leonard and
Mokhtari, 1990], uniting higher-order accuracy and stability. After some exploration it was later found that the original QUICK scheme remains stable at
moderately dense grids. The mistaken belief that a linear higher-order scheme
(QUICK) would not be stable at grid Pclet numbers far beyond the stability limit [Zalonik et al., 2004] was based on the case of a one-dimensional
convection-diusion with Dirichlet boundary conditions [Patankar, 1980]. It
was later corrected by the observation that the two-dimensionality of the
problem and the presence of source terms stabilizes the numerical method.
As the QUICK scheme was found to be more eective (faster and easier to
code) than the nonlinear-limiter ULTRA-QUICK scheme, it was used afterwards for the discretization of the momentum and species equations.
117
6.5
We want to devise a testing procedure that will satisfy two aims, which were
pointed out before:
a systematic and thorough code and solution verication
a comprehensive study of ow structures and heat and solute transport
in thermosolutal natural convection in casting-relevant situations.
Both aims are pursued in parallel on the same test cases. The cases consider
thermal, solutal and thermosolutal natural convection in rectangular cavities
with imposed temperatures and concentrations at both vertical walls. The
test cases are specied below and are arranged by increasing the complexity
of the ow physics and demands to the numerical code. The ndings gained
in the tests can be used on the way as we approach the nal parameters
of thermosolutal ow. All explored cases, their parameters and verication
objectives are summarized in tables 6.1 and 6.2.
case
1a
1b
2
3
4
5a
5b
6
Pr
Sc
RaT
RaC
- 0.71
1
7 700
0.0137
100
0.0137 100
0.0154 187
0.0184
-
105
108
7 108
10
10
1011
2.85 1010
-
1
1
1
4
4
4
4
7 105
2.81 105
2.81 105
2.92 105
2.5 105
1
2
steady ow/
transient ow
steady
steady
steady
transient
transient
transient
transient
steady
Table 6.1: Summary of test cases for natural convection flow: Physical parameters.
verication type
code verication
code verication
code verication
solution & code verication
solution verication
solution verication
solution verication
solution verication
verication methods
Bench, OOC, ErrEst
Bench, OOC, ErrEst, CtCC
OOC, ErrEst, CtCC (limited)
OOC, ErrEst, CtCC
OGC
OGC
OGC
Table 6.2: Summary of test cases for natural convection flow: Verification methods and
objectives.
where xi is the position of the i-th node and m is the total number of nodes in
the x direction along length L. An analogous formula applies to the vertical
(y) direction. Both the uniform and wall-rened grids are shown in Fig. 6.5.
In the subsequent gures and tables the uniform and wall-rened grids are
denoted by u and c respectively.
Case 1: Basic code verification for thermal or solutal buoyancydriven flow
With the aim of code verication two benchmark cases, namely the original
study of de Vahl Davis [de Vahl Davis, 1983] and the later extension of Wan,
Patnaik and Wei [Wan et al., 2001] which correspond to thermal natural
convection of air and a similar uid in a square cavity, were considered. The
code was tested for cases of purely thermal and purely solutal convection
by assessing the discretization error and observed order of grid convergence.
Only the results of the solutal convection tests are shown here since Case
3 is later presented as a much more discriminative initial test for thermal
natural convection. In any case, the parameter range of the benchmark cases
(P r 1) is not relevant to thermal natural convection and the test was
used merely as a rst check. The parameters of the two benchmark cases
are Sc = 0.71, RaC = 105 , A = 1 [de Vahl Davis, 1983] and Sc = 1.0,
RaC = 108 , A = 1 [Wan et al., 2001]. 3 In the purely solutal convection
The results in ref. [Wan et al., 2001] should be used with caution because the paper
does not explicitly state the Prandtl number that was used (P r = 1.0), while the comparison with de Vahl Davis results for P r = 0.71 gives the misleading impression that the
3
119
Figure 6.5: Grid types for the natural convection computations, shown for a grid of
50100 CVs on a A = 4 domain.
121
6.6
Two numerical benchmark cases are considered for a basic verication of the
code. The rst is the classical benchmark dened by de Vahl Davis [de Vahl
Davis, 1983] and later rened by Hortmann, Peri, and Scheurer [Hortmann
et al., 1990], which is used to check the solute transport model with Sc = 0.71
and RaC = 105 . Three uniform grids were calculated, with 5050, 100100,
and 200 200 CV. The grid convergence is monotonous and achieves an
observed order of convergence slightly above 2, uniformly distributed across
the whole domain for all relevant variables (u, v, , and C). The maximum
error in the eld values of these variables is estimated as GCIRex 0.2% and
the function values agree with the benchmark results and estimated error
0.75
0.75
y*
y*
bands given in refs. [de Vahl Davis, 1983] and [Hortmann et al., 1990].
0.5
0.25
0.5
0.25
0.25
0.5
0.75
0.25
x*
0.5
0.75
x*
Figure 6.6: Case 1b: Concentration and flow fields at Sc = 1.0, and RaC = 108 . Solution
by the FVM DC casting code on a 200200 Gauss-Lobatto grid.
1
0.2
DSC
FVM 5252c
FVM 102102c
FVM 202202c
0.75
DSC
FVM 5252c
FVM 102102c
FVM 202202c
v*mid
y*
0.1
0.5
-0.1
0.25
-0.2
0
-0.03
-0.02
-0.01
0.01
0.02
0.0
u*mid
0.25
0.5
0.75
x*
Figure 6.7: Case 1b: Grid convergence of velocity profiles and comparison with the DSC
solution by Wan, Patnaik and Wei [Wan et al., 2001].
123
The second case is a little more demanding and reaches RaC = 108 , with
P r = 1.0. Some more exploration of the numerical results was undertaken
since the benchmark results claimed by Wan, Patnaik and Wei [Wan et al.,
2001] are less clear and do not explicitly specify error estimates. The problem
was calculated on uniform and Gauss-Lobatto grids with 5050, 100100,
and 200200 CV to nd out the dierence in the performance of the two grid
types in the case of relatively high RaC and moderate Sc (i.e. thin boundary
layers). The results might give some useful conclusions on the best meshing
of the subsequent cases, which move toward the nal parameter range of
interest, Sc = 102 and RaC = 1010 . . . 1011 . The convergence over the three
uniform grids was found to be oscillatory and the approximate error spline
(AES) method had to be used to extrapolate the results and compute the
errors based on the extrapolated solution. The observed order of convergence
(OOC) could not be clearly determined by the power-law approximation of
p, the convergence eld values nor from the functionals umax , vmax , and max .
Based on some exploration it is felt, however, that the OOC can be assumed
to be p 2 for uniform grids. Based on p = 2 the error band of the AESextrapolated solution on a uniform grid is estimated as GCIAESex 3%. On
the Gauss-Lobatto grids the convergence was found to be oscillatory as well.
The OOC however, determined by the power-law approximation, is clear and
mostly p 2 across all elds. The error band of the extrapolated solution on
the Gauss-Lobatto grid is GCIAESex 1%.
From these rst tests we can conclude the very good performance of the
code at moderate P r and RaT and respectively Sc and RaC , and the clearly
better performance of locally rened grids in cases with local sharp gradients
as they occur in thin boundary layers at high Ra.
6.7
The problem that was tackled next corresponds to the case characterized by
A = 1, P r = 7, Le = 100, GrT = 105 , and N = 10 in the detailed study of
Gobin and Bennacer [Gobin and Bennacer, 1996]. This problem represents
the steady multicellular regime of thermosolutal ow in an aqueous solution.
Gobin and Bennacer found out that this regime occurs in the transition between the extremes of thermally and solutally dominated monocellular ow
regimes. If the solutal buoyancy dominates over the thermal buoyancy force,
stratied zones appear at the top and bottom of the cavity, while the core is
solutally homogeneous, thermally stratied and driven by the thermal buoy-
max(GCIRex )
3%
4
10 %
0.1 %
0.1 %
Table 6.3: Case 2: Field maxima of the error band estimate for the Richardsonextrapolated solution (GCIRex ).
Grid
100100 c
200200 c
400400 c
pRex
fRex (p)
fRex (p = 2)
GCIRex (p)
GCIRex (p = 2)
umax
9.7569 102
9.4965 102
9.4482 102
2.4
9.4372 102
9.4321 102
0.03 %
0.04 %
vmax
1.7781 101
1.7690 101
1.7681 101
3.4
1.7680 101
1.7678 101
5 104 %
2 103 %
| |max
1.4283 102
1.4178 102
1.4173 102
4.2
1.4172 102
1.4171 102
104 %
103 %
Nu
5.4507
5.4107
5.4053
3.0
5.4045
5.4035
2 103 %
5 103 %
Sh
57.461
57.449
57.445
1.7
57.444
57.444
1 103 %
1 103 %
Table 6.4: Case 2: Observed order of convergence pRex , extrapolated solution fRex and
error band estimates GCIRex based on functionals.
The numerical tests were all performed on Gauss-Lobatto grids since the
preceding case showed that they perform better in terms of grid convergence.
The solution is again studied on three grids, 100100, 200200, and 400400
CV. Finer grids as before were chosen since ner solutal eld structures are
expected to occur at the high Sc = 700 and sharper boundary layers are
anticipated at the RaC = 7 108 . The convergence of all variables of interest
(u, v, , T , and C) is monotonous across most of the domain, thus the
classical Richardson extrapolation can be used. The OOC is clearly p = 2
125
0.75
0.75
0.5
In this basic test for double-diusive convection the numerical code suc-
0.25
0.5
0.25
0.25
0.5
0.75
0.25
0.5
*
0.75
y*
0.75
0.5
0.25
0.25
0.5
0.75
x*
Figure 6.8: Case 2: Temperature, concentration and flow fields. Solution by the FVM
DC casting code on a 400400 Gauss-Lobatto grid.
102102c
202202c
402402c
102102c
202202c
402402c
0.75
y*
0.75
0.5
0.25
0.5
0.25
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.1
0.2
0.3
mid
0.4
0.5
0.6
0.7
0.8
0.9
mid
(a) Temperature .
(b) Concentration .
Figure 6.9: Case 2: Grid convergence of temperature and concentration profiles at cavity
mid-width (x = 0.5).
0.20
0.15
102102c
202202c
402402c
0.75
102102c
202202c
402402c
0.10
v*mid
y*
0.05
0.5
0.00
-0.05
0.25
-0.10
-0.15
0
-0.1
-0.05
0.05
0.1
u*mid
-0.20
0.25
0.5
0.75
x*
cessfully reproduced the known multicellular ow structure [Gobin and Bennacer, 1996]. Excellent accuracy of the results was shown and was particularly
remarkable with respect to the concentration eld. The high accuracy is at-
127
6.8
(a)
t = 12.29
(b)
t = 32.77
(c)
t = 53.24
(d)
t = 73.72
(e)
t = 821.61
(f)
t = 827.34
Figure 6.11: Case 3: Evolution of temperature and flow fields in the cavity. The first four
times are from the initial transient, the last two correspond to oscillation extrema in the
final oscillatory state. Solution by the FVM DC casting code on a 100400 uniform grid.
( = 0.05, = 5 104 )
129
NuL/2
Nu0
Nu0
Nu0
Nu0
Nu0
Nu0
50
100
(Spectral)
(FVM 13x50u)
(FVM 25x100u)
(FVM 50x200u)
(FVM 100x400u)
150
NuL/2
NuL/2
NuL/2
NuL/2
NuL/2
200
50
100
(Spectral)
(FVM 13x50u)
(FVM 25x100u)
(FVM 50x200u)
(FVM 100x400u)
150
200
t*
0.8
0.8
0.6
0.6
NuL
NuL
NuL
NuL
NuL
50
100
t*
(Spectral)
(FVM 13x50u)
(FVM 25x100u)
(FVM 50x200u)
(FVM 100x400u)
150
200
NuL
0.4
0.4
0.2
0.2
0
0
0.1
0.2
0
0
0.1
0.2
(d) Temperature and flow fields in the cavity: comparison of the 100 400 grid FVM solution with the SM
solution ( = 0.1, = 5 104 )
Figure 6.12: Case 3: Grid convergence of the FVM solution and comparison with the SM
solution. Time evolution of the Nusselt number.
1
Spectral
FVM 1552u
FVM 2752u
FVM 52202u
FVM 102402u
0.01
v*mid
y*
0.75
0.5
-0.01
0.25
Spectral
FVM 1552u
FVM 2752u
FVM 52202u
FVM 102402u
-0.05
0.05
u*mid
-0.02
0.05
0.1
0.15
0.2
0.2
x*
Figure 6.13: Case 3: Grid convergence of the FVM solution and comparison with the SM
solution. Velocity profiles.
in detail [Zalonik et al., 2005] to the solution obtained by Shihe Xin, using
the spectral Chebyshev-collocation method (SM) [Xin and Le Qur, 2002].
The SM solution is veried to be grid independent, however, error estimates
are not specied. This comparison could be regarded as a code-verication
activity, but not in a very rigorous manner. Despite this, the belief in the
good accuracy of the SM solution gives substantial condence in a correct
result of the FVM code. An additional comparison was attempted with a
local meshless method [Divo and Kassab, 2006]. While the latter comparison
gave a very good qualitative match, and thus some additional condence, the
quantitative comparison is lacking and is not shown here in detail.
The time evolution of the Nusselt numbers at both walls of the cavity
and at the vertical mid-plane is shown in Fig. 6.12 for the FVM solutions
on the four grids and for the SM solution. The dierences in the evolution
between the FVM on the densest (100400) grid and the SM solution are
very small. Also, the nal oscillation periods match to within 1 % and the
oscillation amplitudes are close. The SM solution should have a much better
time accuracy as it employs a second-order time-stepping scheme and uses a
much smaller timestep. This should have some impact on the small discrepancies. The numerical parameters used in both codes are given in Table 6.5.
A comparison of the instantaneous temperature eld and streamlines in the
nal oscillatory state is shown in Fig. 6.12(d). The results in Fig. 6.12(d)
131
umax
1.060
1.019
1.032
1.038
1.040
1.7
0.1 %
vmax
3.051 102
2.940 102
2.989 102
3.023 102
3.039 102
1.3
0.3 %
| |max
3.93017 102
3.91031 102
3.99413 102
4.00588 102
4.00448 102
3.0
0.005 %
Table 6.6: Case 3: Observed order of convergence pAESex , extrapolated solution AESex,
and error band estimates GCIAESex based on functionals.
6.9
The high-RaC ow of a high-Sc uid in this case is close to the limit of the
steady ow given by Lage and Bejan for uids with Sc 1 and cavities of
an aspect ratio A < 50 as RaC 108 Sc2 . An objective of this test case is
to determine the possible occurrence of instabilities in the solutal convection
ow. It was found out that the small oscillatory ow structures are conned
to regions close to the vertical walls (Fig. 6.14). They form small near-wall
ow cells which appear due to instabilities of the boundary layer. It was
shown before [Bergholz, 1977,Le Qur and Gobin, 1999] that large-Sc uids
in slender enclosures or slots are prone to dierent kinds of instabilities at
high enough RaC , however, a detailed discussion of the exact mechanism
exceeds the scope of this work.
Looking at the most accurate numerical solution, obtained on a 100400
Gauss-Lobatto grid, the oscillation amplitude of max reaches about 15 % of
its value while the vertical velocity uctuations (vmax ) are only about 3 %.
The horizontal velocity (umax ) and Sh uctuations are even much smaller. We
can conclude that the solution of the problem with RaC = 1010 is oscillatory,
but with relatively small oscillation amplitudes and long oscillation periods
(of the order of 103 s). Also, the periodicity of oscillations could not be clearly
determined as there are several lower-frequency oscillations superimposed
onto the base frequency. The boundary layer instability occurring in this
ow does not necessarily present implications for the double-diusive ow
representative for solidication since there we can expect a dominant eect
of the thermal buoyancy in those regions where the solutal boundary layer
instability occurs in the solutal ow.
The numerical solution of the solutal-buoyancy driven ow at Sc = 100
133
and RaC = 1010 is strongly dependent on the right choice of grid type and
density. As shown in Table 6.7, a steady solution is obtained with coarse
grids, apparently due to a damping eect of the discretization error. With
very coarse grids the solution of the high-Pclet ow quickly becomes numerically unstable. Based on the ne-grid solutions I concluded that the ow is
oscillatory. We can also conrm that the Gauss-Lobatto grid, rened at the
walls, is better suitable in this case for the numerical approximation to capture the eects triggering the instabilities. Obviously, however, the inuence
of the instabilities on the overall species transport at the wall (Sh) is not
important here and the eect is minor even locally.
6.10
6.10.1
After the onset of cooling and imposed concentration gradient the thermally
induced convection builds up quickly and is not aected by the solutal eect.
The latter remains conned to the very thin solutal boundary layer at the
right wall and a shallow accumulation of virtually stagnant solute rich liquid
at the bottom of the enclosure (Fig. 6.16(a)). The thermal convection results
in a multicellular ow pattern, qualitatively similar to the one found with
identical parameters for pure thermal convection in section 6.8. With time
the density of the upper of the two main thermal ow cells is increased by the
advective solute transport from the solutal boundary layer. The cell moves
downwards and merges with the lower cell. At the same time, a new thermal
cell is generated at the top of the enclosure. This phenomenon is observed repeatedly. The lower cell interacts with the solutal layer and attempts to carry
solute into the center of the enclosure. Due to the vigorous thermally-induced
ow and consequently eective mixing this does not establish a closed solutal
loop between the thermal cells. Later, the stagnant stratied layer continues
to grow and the concentration in the top part slowly increases (Fig. 6.16(b)).
A solute-lean stratied layer starts to grow at the top of the cavity, while
(a)
t = 5629.0
(b)
t = 5807.3
(c)
t = 5946.6
(d)
t = 6154.1
(e)
t = 6268.0
(f)
t = 6599.8
Figure 6.14: Case 4: Concentration fields and streamlines in the enclosure at local minima, maxima and inflection points of a representative oscillation period (oscillations shown
in Fig. 6.15) Solution by the FVM DC casting code on the 50 200 Gauss-Lobatto grid.
( = 0.02, = 104 ).
135
10.47
2.9x10-03
10.46
2.8x10-03
10.45
max*
Shmid
2.7x10-03
2.6x10-03
10.44
2.5x10-03
10.43
2.4x10-03
2.3x10
-03
2000
4000
6000
8000
10000
t*
12000
10.42
2000
4000
6000
8000
10000
12000
t*
Figure 6.15: Case 4: Time evolution of the flow obtained on the 50200 Gauss-Lobatto
grid.
the bottom layer is somewhat reduced in height (Fig. 6.16(c)). The stratication of the top layer is slightly weaker than that of the bottom layer, which
remained stable at its maximum height (Fig. 6.16(b)). Consequently, the
lateral temperature gradient nally destabilizes the stratication at the top
(Fig. 6.16(d)). A pulsating ow continues to evolve with transitions changing from a structure with two solutally uniform thermal cells conned by a
sharp solutal gradient (Fig. 6.16(e)) to monocellular ow between two solutally stratied layers (Fig. 6.16(f)).
The evolution of the ow seems to depend on the buoyancy ratio (N = 31
in this case) and on the stability of the thermally-induced ow. The thermal force resulting from the lateral temperature gradient inside the solutally
stratied layer is strong enough to destabilize the stable solutal stratication
as the layer grows; a pulsating multicellular thermosolutal ow structure
evolves. Such structures are also shown in Section 6.7. They were studied by
Bennacer and Gobin [Gobin and Bennacer, 1996] in high-P r uids, where at
moderate N several thermal ow cells, internally of uniform concentration,
were observed which were contained in the outer solutal cells, with the solutal eect conned to the wall boundary layers and the high-concentrationgradient layers between the thermal cells. Bennacer and Gobin showed that
the multicellular regime is promoted by higher Rayleigh numbers and higher
geometry aspect ratios A = H/L. The buoyancy ratio N , where a similar
transition should occur at lower P r, is not obvious. Although not very extensive in this respect, the results of Shyy and Chen [Shyy and Chen, 1991]
6.10.2
137
(a)
t = 32.77
(b)
t = 1555.56
(c)
t = 3305.26
(d)
t = 4047.41
(e)
t = 7618.90
(f) t
11533.61
Figure 6.16: Case 5a: Temperature, concentration and streamlines in the enclosure for the
case of cooperating thermal and solutal forces (Al-Cu). Solution obtained on the 50 200
Gauss-Lobatto grid. ( = 0.05, = 0.02, = 3 104 )
180
100400
50200
50200
100400
160
6
140
Nu0
Sh0
120
100
80
60
4
40
20
0
1000
2000
3000
4000
1000
2000
3000
4000
t*
460
440
100400
50200
420
50200
100400
400
6
380
360
320
NuL
ShL
340
300
280
260
240
220
200
180
160
1000
2000
3000
4000
1000
t*
2000
3000
4000
t*
2000
7
50200
100400
1500
1000
Numid
Shmid
500
-500
-1000
4
100400
50200
-1500
-2000
250
500
750
t*
(e) Sh at mid-plane (x = 0)
1000
1000
2000
3000
4000
t*
(f) N u at mid-plane (x = 0)
Figure 6.17: Case 5a: Evolution of Sherwood and Nusselt numbers calculated on GaussLobatto grids with 50200 and 100400 CV.
139
(a)
t = 29.17
(b)
t = 41.00
(c)
t = 406.07
(d)
t = 2539.69
(e)
t = 6205.33
(f) t
12237.19
Figure 6.18: Case 5b: Temperature, concentration and streamlines in the enclosure for
the case of counteracting thermal and solutal forces (Al-Mg). Solution obtained on the
50 200 Gauss-Lobatto grid. ( = 0.05, = 0.02, = 3 104 )
6.11
The test is mostly numerical since a purely thermal ow is not really relevant
for alloy casting and the ow in pure aluminum castings is not of interest in
this work. The properties of the metal are dened with xed solidus and liquidus temperatures approximating the solidication range of an Al-5.25 %Cu
alloy (TS = 827 K, TL = 920 K). The liquid fraction in the mushy zone is assumed to vary as
fl =
T TS
.
TL TS
(6.31)
The results of a grid convergence study are presented in Figs. 6.19 and 6.20.
Note that the whole computational domain extended vertically to 1.0 m and
the discretizations across the liquid pool are 2573, 50146, 100109, and
200219 CV, respectively. The ow eld shows an evolution of the solution
into a multicellular ow with grid renement. The ow structure is practically
grid independent at the nal grid density and the velocity proles, shown in
Fig. 6.20, clearly converge with only small dierences left. The accuracy of
the results for the two nest grids can thus be regarded as satisfactory, being
cognizant of the fast grid convergence of the numerical code. The result lives
up to the expectations given by the results of the thermal convection case as
the grid extending over the liquid pool is of a comparable size at very similar
parameters (P r and RaT ).
141
-0.05
us
1)
(f l=
=0
)
id
liqu
us
1)
(f l=
-0.10
so
so
l
id
lid
us
us
(f
(f
-0.10
id
liqu
=0
)
-0.05
z [m]
0.00
z [m]
0.00
0.05
r [m]
0.10
-0.15
0.00
0.05
(a) 25100
-0.05
z [m]
-0.05
1)
(f l=
=0
)
z [m]
0.00
us
id
liqu
us
1)
(f l=
-0.10
-0.15
0.00
0.05
so
so
lid
lid
us
us
(f
(f
-0.10
0.10
(b) 50200
0.00
id
liqu
r [m]
=0
)
-0.15
0.00
r [m]
(c) 100400
0.10
-0.15
0.00
0.05
r [m]
0.10
(d) 200800
Figure 6.19: Case 6: Grid convergence of the flow field. Streamlines of relative velocity
(~v ~vcast ) are shown as obtained on four grids.
0.008
0.006
0.002
0.004
0.002
0
vmid [m/s]
vaxe [m/s]
0.001
-0.001
-0.002
-0.002
-0.004
-0.003
-0.004
-0.15
25100
50200
100400
200800
-0.006
-0.1
-0.05
z [m]
-0.008
-0.15
25100
50200
100400
200800
-0.1
-0.05
z [m]
6.12
Summary
Physical aspects
The range of parameters of thermosolutal natural convection relevant for the
ow in the liquid pool of a DC casting was estimated. The ow in the pertinent parameter range was studied on test cases of natural convection in
a rectangular cavity. Instabilities of the thermal convection in low-P r liquid metals were found to be decisive for the ow structure and consequently
the mass transfer along the walls. We could use the good understanding of
ow structures in thermosolutal convection in high-P r uids to partly explain the occurrence of thermal ow cells, solutally stratied layers, and the
destabilization of the stratication. Further, we could explain the implications of the fundamental dierence in the mechanisms of thermal convection
between high- and low-P r uids for the thermosolutal ow structures. The
main conclusions are the following.
With cooperating thermal and solutal buoyancy thermally-driven ow
cells and solutally-stratied layers are generated. The ow structure
is important for macrosegregation during solidication because mass
transfer from the wall (representative of the liquid-mush interface) is
enhanced in the regions of thermal ow and smaller at the stratied
layers. The ow is very unstable primarily due to the instability of the
Summary
143
7.1
Introduction
146
Casting parameter
casting size
casting velocity
casting temperature
mold cooling
7.2
Direct eect
sump size (H, L)
sump depth (H)
char. temp. dierence (T )
sump depth (H)
Aected parameters
RaT , RaC , A
RaT , RaC , A
RaT
RaT , RaC , A
The casting parameters to investigate are those we expect would most strongly
impact on the macrosegregation. One way is to determine them based on experimental evidence or industrial experience. Another way is to identify the
most relevant casting parameters based on the impact they have on the solute transport phenomena (Fig. 1.8). The shrinkage-induced ow depends on
the shape of the mushy zone. A deeper sump (larger slope of the sump) promotes the impact of shrinkage. The thermosolutal ow is best characterized
by the natural convection parameters (RaT , RaC and A) as this helps to
see where a signicant change in the ow structure could be triggered. The
anticipated eects are summarized in Table 7.1. In this way, I identied the
size of the casting, the casting velocity, as well as the mold cooling (mold
type) having the strongest impact on macrosegregation. Another, generally
important casting parameter, the water ow rate, should only have a very
small eect on the thermal eld in the liquid sump, which is inferred from experience [Volak et al., 2005]. Consequently, it should not impact on the ow
and thus should have only a negligible eect on macrosegregation. The latter
was also shown experimentally [Dorward and Beerntsen, 1990, Eskin et al.,
2004]. The impact of parameters on the eect of grain transport is less clear;
it should be aected by the sump depth and the mushy zone thickness. A
more detailed investigation of the grain transport is presented in Section 8.2.
7.2.1
The model equations presented in Section 4.5 are solved for an entirely porous
mushy zone, i.e. equations deduced from R = 1 (or gsc = 0). A minor difference with the original equations is in the momentum equation, where the
Darcy equation terms are reshaped as previously by Prescott, Incropera and
Bennon [Prescott et al., 1991], being equivalent to setting D = gs (or gsD = 0).
147
(7.2)
Slip boundary conditions are prescribed at the right boundary for the contact
of the liquid and the mold. The heat transfer to the mold in the non-insulated
part was modeled by a heat transfer coecient, varying between 400 W/m2 K
and 5000 W/m2 K, depending on the liquid fraction at the contact surface.
At the insulation ring the heat transfer coecient is approximated to be
zero. The secondary cooling is modeled using the Weckman-Niessen correlation [Weckman and Niessen, 1982] for the heat transfer coecient at a vertical surface cooled by a subcooled-nucleate-boiling falling-water lm. The
heat transfer coecient is thus
148
hchill
hgraphite = 0 W/m2 K
z > zgraphite
(1fl )
fl
Qw [m /s]
zmold > z
D
[m]
The water entry temperature is 300 K and a heat balance is used to calculate the increase of the water temperature Tw along the billet surface. The
bottom of the computational domain is assumed to be adiabatic. While this
assumption is not exact due to the short domain length of 0.6 m, it should
not signicantly aect the ow nor the macrosegregation. All boundaries,
except the inow boundary, are impermeable for species transport. The left
boundary (billet centerline) is a symmetry boundary. These boundary conditions are shown schematically in Fig. 7.1. The thermophysical properties of
Al-5.25 wt%Cu used in the simulations are given in Table B.1 in Appendix B
together with their sources. They were obtained from the materials properties
modeling software JMatPro [Saunders et al., 2001] and from reference [Vreeman and Incropera, 2000]. All process parameters, which vary from case to
case and are not explicitly given here, are summarized in Table 7.2.
7.3
7.3.1
The numerical method is described in detail in Chapter 5. Only some features, specic to the computations performed in this part of the work, are
given here. The set of macroscopic transport equations (Eqs. (4.91), (7.1),
(4.113), and (4.115)) was solved in axisymmetric geometry and on an orthogonal structured grid. A grid of 100356 control volumes (CVs) with additional
nodes at the domain boundaries (thus a total of 102358 nodes) was used in
all cases and it was strongly biased in the top part (the height of the liquid
and mushy regions). The top part (0.15 m in the large billet and 0.12 m in
the small billet) was discretized into 100200 equal CVs and the remaining
bottom part into 100256 CVs with a grid expansion factor of 1.0175. The
149
grid is shown in Fig 7.2. For the discretization of advective uxes the central
dierence scheme was used in the energy equation and the rst-order upwind
scheme in the species and momentum equations. The upwind scheme was
used to cure the severe problems with the iterative convergence of the coupled equation system which were encountered upon the use of higher-order
schemes for momentum and species advection. One should note that this is
only a preliminary computational framework that works eciently. It is used
before the stable full implementation of more accurate numerical schemes,
which have been shown to be necessary [Zalonik and arler, 2005b], is completed.
7.3.2
Solution Procedure
The implicit iterative solution of the strongly coupled equations can be relatively unstable at the beginning of the calculation. The development of a
strong thermosolutal convective ow in the liquid pool and its interaction
with the mushy zone can severely slow down the convergence if the solution
150
H1
y =
H1
n1
yi1 = ryi
H2
H2 =
Pn2
i=1
yi =
yn2 (1r n2 )
(1r)
D
x =
D
m
Figure 7.2: A biased 102358-node finite volume grid for DC casting calculations.
151
of the full problem is started right from the beginning with an unfavorable
initial state. Therefore, a special procedure was used to speed up the computations.
1. First, only an approximation of the temperature, liquid fraction and
ow elds was obtained by solving the energy and momentum equations
with both thermal and solutal expansion coecients (T , C ) set to
zero.
2. Then the realistic expansion coecients and the solute conservation
equation were invoked, with species diusion coecients initially increased by 100 times. The latter was done to somewhat relax the gradients of the solutal buoyancy body force driving the ow, thus easing
convergence during the initial redistribution of copper in the mushy
and liquid zones.
3. After a calmed down ow was obtained, the diusion coecients were
set to their realistic values and the computations were continued until
a quasi-steady radial macrosegregation prole was obtained.
A steady state of the system was not found since slight oscillations of
the ow persisted even long after the quasi-steady segregation prole was
reached. This is not surprising; strong instabilities of the thermosolutal natural convection ow were already observed in Chapter 6. Also, a physically
oscillatory ow was reported before in computations of thermal convection
in DC casting [Zalonik and arler, 2005b]. However, it is important to note
that in the cases treated here the ow oscillations were weak and had only
a negligible eect on the ow structure and the macrosegregation pattern.
Previously, Vreeman and Incropera [Vreeman and Incropera, 2000], who reported an unsteady ow in some of their macrosegregation computations,
presumed a numerical problem. Even though some unphysical oscillations
are present in the concentration eld it is believed that they are not an important source of the ow oscillations in the present computations, and their
origin is physical.
Severe oscillations are present in the subsurface part of the segregation
prole (Figs. 7.107.10). Their origin is numerical and was analyzed before
in detail [Venneker and Katgerman, 2000]. The eect is stronger with a small
mushy zone thickness (larger fl -gradient). A remedy is the alignment of the
grid with the liquid-mush interface and local grid renement.
152
Case Diameter
vcast
Tcast
[mm] [mm/s]
[K]
1
282
1.25 973.15
2
218
1.43 973.15
3
218
1.14 973.15
4
218
1.43 953.15
5
218
1.43 973.15
7.4
7.4.1
Mold type
hot-top
hot-top
hot-top
hot-top
water cooled
Qcool
[m3 /s]
1.75 103
1.56 103
1.56 103
1.56 103
1.56 103
Inlet
[mm]
172
140
140
140
140
Five dierent cases are studied. They represent variations of the billet diameter, casting velocity (vcast ), casting temperature (Tcast ), and mold type. As
reasoned in Section 7.2, these casting parameters are expected to have the
strongest inuence on macrosegregation. The specications of the ve cases
are given in Table 7.2. Case 2 represents the basic case, also used to study
the impact of billet size, when compared to Case 1. The casting velocity is
decreased by 20 % in Case 3 and the casting temperature by 20 K in Case 4.
In Case 5 a change of mold type causing a signicant change of heat transfer
in the mold is simulated. Case 1 is observed in more detail to explain the ow
structures and species transport in the sump and on the way they result in
the nal macrosegregation pattern. The other cases are discussed afterwards,
focusing on the mechanism of the eect of the dierent parameters on the
macrosegregation prole. The main distinguishing features of the ow and
species transport are explained.
7.4.2
To make a clear presentation of the origins of the thermosolutal buoyancydriven ow in the liquid sump, Fig. 7.5 shows the ow eld together with
the liquid-density eld (calculated according to the Boussinesq approximation used in the momentum equationEq. (7.1)). As can be seen in the
plot, the liquid is relatively quiescent at the bottom and in the center of the
liquid sump, which is caused by the counteracting eect of thermal buoyancy and the stratication tendency of the heavy solute-rich liquid. Next to
the liquid-mush interface a complex ow structure consisting of small vortices forms. Five vertically arranged ow cells can be identied, delimited by
sharp density gradients. Each cell contains one or two vortices of thermoso-
153
lutal origin. The solutal buoyancy eect also causes a slight vertical density
bump inside each thermosolutal cell and raises the splitting of the ow into
two vortices. As explained in Section 6.10.1, such a multicellular structure
occurs in thermosolutal ows when a stable solutal stratication is destabilized by a large enough lateral temperature gradient. In contrast to test
Case 5a in Section 6.10.1, however, the ow in the liquid pool of the casting
is relatively stable. Despite the small oscillations that persist which prevent
the development of a steady state, the structure and arrangement of the ow
cells are retained throughout. In the solidication zone (0 < fl < 1) the liquid is enriched due to segregation of the solute (Cl > Cm ; Cs < Cm ). Since
copper has a larger density than aluminum, a solutal downward ow of enriched liquid can be observed in the high-fl portion of the mushy zone (along
the liquidus front). This ow carries enriched liquid away from the top-outer
part (top right in Fig. 7.4) of the mushy zone, replacing it with solute-lean
liquid, which penetrates into the mush from the bulk liquid region. Solute
is carried into the bulk liquid and the heavy, enriched liquid ows down the
slope of the mushy zone front, accumulating at the bottom of the sump.
Along the liquidus front the thermosolutal circulations additionally carry
high-concentration liquid out of the mushy zone into the bulk-liquid sump.
In the low-fl regions of the mushy zone, the permeability of the porous mush
rapidly decreases; consequently, drag forces dominate over buoyancy and inertia. The ow is driven mostly by solidication shrinkage, as the liquid lls
up the space left by the shrunken solidied metal. The liquid in the lowfl mushy zone is highly segregated, thus even the small shrinkage-induced
relative velocities can result in considerable net solute transport. This can
be clearly seen along the whole mushy zone. The direction of the shrinkageinduced ow is opposite to the liquid fraction gradient (Fig. 7.5). As the
enriched liquid ows towards the solidus front it is replaced by the leaner
liquid from regions of the mushy zone with higher fl . The enriched liquid
accumulates at the front and the concentration rapidly increases in the direction opposite to the fl -gradient. Because of the characteristic sump shape
and the direction of the shrinkage-induced solute transport, this ow tends to
decrease the concentration in the billet center and increase the concentration
in the outer part of the billet. This is also the mechanism causing typical
positive subsurface segregation.
154
T [K]
fl = 1
969
961
953
fl = 0
-0.05
945
937
z [m]
929
921
913
-0.10
-0.15
0.00
0.05
r [m]
0.10
CCu [%]
fl = 1
6.3
6.1
5.9
fl = 0
-0.05
5.7
5.5
z [m]
5.3
5.1
4.9
4.7
-0.10
4.5
4.3
4.1
3.9
-0.15
0.00
3.7
0.05
r [m]
0.10
Figure 7.4: Mixture Cu concentration (Cm ) field in the billet for Case 1.
155
l
2500
2496
2492
-0.05
2488
fl = 1
2484
z [m]
fl = 0
2480
2476
-0.10
2472
Shrinkage flow direction
2468
2464
2460
-0.15
0.00
0.05
r [m]
0.10
Figure 7.5: Liquid density (l ) and relative velocity (~v ) fields in the billet for Case 1. The
density field is shown only in the liquid and part of the mushy zone (fl > 0.1).
0.00
fl
T
Cm
m
um
z [m]
-0.05
-0.10
1.0
1000
2540
0.0
-1.0
-2.0
900
800
700
0.065
0.060
2520
2500
0.03
0.02
-3.0
600
0.055
2480
0.01
-4.0
500
-5.0
400
0.050
2460
0.00
-6.0 fl
300 T [K]
0.045 C m
3
2440 m [kg/m ]
um [m/s]
Figure 7.6: Vertical profiles of liquid mass fraction (fl ), temperature (T ), mixture concentration (Cm ), mixture density (m ), and horizontal mixture velocity component (um )
at (r = 0.051465 m).
156
7.4.3
157
T [K]
969
961
953
z [m]
-0.05
945
fl = 1
937
fl = 0
929
-0.10
921
913
0.00
0.05
0.10
r [m]
CCu [%]
fl = 1
5.8
5.7
5.6
5.5
5.4
fl = 0
z [m]
-0.05
5.3
5.2
5.1
5.0
4.9
4.8
4.7
-0.10
4.6
4.5
4.4
0.00
0.05
r [m]
0.10
Figure 7.8: Mixture Cu concentration (Cm ) field in the billet for Case 2.
158
0.00
l
2488
2484
fl = 1
2480
fl = 0
-0.05
z [m]
2476
2472
2468
2464
-0.10
2460
0.00
0.05
0.10
r [m]
Figure 7.9: Liquid density (l ) and relative velocity (~v ) fields in the billet for Case 2. The
density field is shown only in the liquid and part of the mushy zone (fl > 0.1).
6.5
CCu [wt%]
6.0
5.5
initial concentration
5.0
4.5
4.0
3.5
0.00
Map 5
Case1
Case2
Case1
Case2
0.25
0.50
0.75
1.00
r/R
Figure 7.10: Comparison of macrosegregation profiles as a function of the normalized
radius for different billet sizes.
159
T [K]
969
961
953
z [m]
-0.05
fl=1
945
fl=0
937
929
-0.10
921
913
0.00
0.05
r [m]
0.10
CCu [%]
fl = 1
5.8
5.7
5.6
5.5
5.4
-0.05
5.3
z [m]
fl = 0
5.2
5.1
5.0
4.9
4.8
4.7
-0.10
4.6
4.5
4.4
0.00
0.05
r [m]
0.10
Figure 7.12: Mixture Cu concentration (Cm ) field in the billet for Case 3.
160
0.00
l
2488
2484
fl = 1
2480
fl = 0
-0.05
z [m]
2476
2472
2468
2464
-0.10
2460
0.00
0.05
0.10
r [m]
Figure 7.13: Liquid density (l ) and relative velocity (~v ) fields in the billet for Case 3.
The density field is shown only in the liquid and part of the mushy zone (fl > 0.1).
5.8
CCu [wt%]
5.6
5.4
initial concentration
5.2
initial concentration
Case 2
Case 3
Case 2
Case 3
5.0
4.8
4.6
0.00
0.02
0.04
0.06
0.08
0.10
r [m]
Figure 7.14: Prediction of the influence of casting velocity on macrosegregation.
161
T [K]
969
961
953
z [m]
-0.05
945
937
fl=1
fl=0
929
-0.10
921
913
0.00
0.05
0.10
r [m]
CCu [%]
fl = 1
5.8
5.7
5.6
5.5
5.4
fl = 0
z [m]
-0.05
5.3
5.2
5.1
5.0
4.9
4.8
4.7
-0.10
4.6
4.5
4.4
0.00
0.05
r [m]
0.10
Figure 7.16: Mixture Cu concentration (Cm ) field in the billet for Case 4.
162
0.00
l
2488
2484
fl = 1
2480
fl = 0
-0.05
z [m]
2476
2472
2468
2464
-0.10
2460
0.00
0.05
0.10
r [m]
Figure 7.17: Liquid density (l ) and relative velocity (~v ) fields in the billet for Case 4.
The density field is shown only in the liquid and part of the mushy zone (fl > 0.1).
5.8
CCu [wt%]
5.6
5.4
initial concentration
5.2
initial concentration
Case 2
Case 4
Case 2
Case 4
5.0
4.8
4.6
0.00
0.02
0.04
0.06
0.08
0.10
r [m]
Figure 7.18: Prediction of the influence of casting temperature on macrosegregation.
Summary
163
7.5
Summary
164
T [K]
969
961
953
z [m]
-0.05
945
937
929
fl=1
fl=0
-0.10
921
913
0.00
0.05
0.10
r [m]
CCu [%]
5.8
5.7
5.6
5.5
fl = 1
5.4
-0.05
z [m]
5.3
fl = 0
5.2
5.1
5.0
4.9
4.8
4.7
-0.10
4.6
4.5
4.4
0.00
0.05
r [m]
0.10
Figure 7.20: Mixture Cu concentration (Cm ) field in the billet for Case 5.
165
Summary
0.00
l
2488
2484
2480
-0.05
z [m]
2476
fl = 1
2472
fl = 0
2468
2464
-0.10
2460
0.00
0.05
0.10
r [m]
Figure 7.21: Liquid density (l ) and relative velocity (~v ) fields in the billet for Case 5.
The density field is shown only in the liquid and part of the mushy zone (fl > 0.1).
5.8
CCu [wt%]
5.6
5.4
initial concentration
5.2
initial concentration
Case 2
Case 5
Case 2
Case 5
5.0
4.8
4.6
0.00
0.02
0.04
0.06
0.08
0.10
r [m]
Figure 7.22: Prediction of the influence of mold cooling on macrosegregation.
166
z [m]
-0.05
-0.10
Case 2
Case 4
Case 2
Case 3
-0.15
0.00
0.05
r [m]
(a) casting velocity
0.10
0.00
0.05
Case 2
Case 5
0.10
r [m]
(b) casting temperature
0.00
0.05
0.10
r [m]
(c) mold cooling
Figure 7.23: Impact of process parameters on the sump shape. The solidus (fl = 0) and
liquidus (fl = 1) fronts are shown.
thermal ow along the liquidus front and due to solutal stratication, particularly pronounced in the Al-Cu alloy. The solute transport is intensied
when instability causes the emergence of smaller ow cells which make the
transport of solute-rich liquid from the mushy zone even more eective. It is
generally intensied by a deeper liquid pool and by a higher casting temperature. The opposite, or better said, alleviating eect is that of the shrinkage
ow which transports solute-rich interdendritic liquid away from the centerline. Its magnitude depends primarily on the shape of the mushy zone, i.e.
locally on its slope and globally on its depth. A more detailed investigation
of the interaction of convection and shrinkage ow is necessary to estimate
the interaction of both solute transport modes and is made in Chapter 8.
Higher casting velocity and higher casting temperature both promote macrosegregation through their eect on the natural convection in the sump. The
rst occurs through an increase of the liquid pool depth and the latter by an
increase of the temperature dierence driving the thermal natural convection
in the liquid pool.
The estimations of the eects of the casting parameters show that, although the porous ow regime model is incapable of properly predicting
macrosegregation patterns, it can be used to predict trends caused by variations of the parameters. Most importantly, the eect on the thermosolutal
ow is shown. The results indicate that the ow structure plays an important
role in macrosegregation formation. These trends, in the sense of changes of
the global degree of macrosegregation, show some relation to experimentally
and industrially observed trends. Macrosegregation is generally observed to
Summary
167
increase with casting velocity [Dorward and Beerntsen, 1990, Eskin et al.,
2004]. The AirSlipTM mold cooling is known from industrial practice to produce less macrosegregation than conventional water-cooled molds, especially
in the subsurface layer. The eect of casting temperature seems to be more
complex since dierent eects were observed: a reversal from negative to
positive centerline segregation with a temperature increase [Dorward and
Beerntsen, 1990], and no signicant inuence of the casting temperature except close to the surface [Eskin et al., 2005]. It remains clear that an extension
of the model to account for grain transport is necessary to move closer to the
correct prediction of local macrosegregation deviations.
Interaction of Flow
Phenomena Causing
Macrosegregation in DC
Casting
170
iterative solution strategy 5.10 could possibly alleviate these problems, but
such developments were out of scope of the present work. Facing these problems, the decision was taken to try to extract some conclusions on the eect
of the presence of a slurry ow regime in part of the mushy zone on macrosegregation from a simplied model, assuming a constant concentration (i.e.
no solute transport) and presuming a xed-grain diameter. Constant concentration implies that the liquid mass fraction depends solely on temperature
according to the inverse-lever rule.
TS (C0 ) T
kp
(8.1)
fl =
1 kp
T Tf
By presuming a constant grain diameter the grain density becomes an unnecessary parameter, making the grain transport equation (Eq. 4.128) redundant.
8.1
8.1.1
Six cases are studied by comparing computations of macrosegregation, accounting for shrinkage only with computations taking into account both thermosolutal convection and shrinkage, for three dierent castings.
171
Case
5
5sh
6
6sh
7
7sh
Case
5, 5sh
6, 6sh
7, 7sh
Alloy composition
Al-5.25 wt%Cu
Al-5.25 wt%Cu
Al-5.00 wt%Mg
Al-5.00 wt%Mg
Al-5.25 wt%Cu
Al-5.25 wt%Cu
kp Geometry
0.17 billet
0.17 billet
0.44 billet
0.44 billet
0.17 ingot
0.17 ingot
Flow
N
TSC + Sh >0
Sh
TSC + Sh <0
Sh
TSC + Sh >0
Sh
-
Diameter/
vcast
Tcast
Width [mm] [mm/s]
[K]
218
1.43 973.15
218
1.43 963.15
218
1.25 973.15
Mold type
water cooled
water cooled
water cooled
Qcool
[m3 /s]
1.56 103
1.56 103
1.56 103
Inlet
[mm]
140
140
140
172
8.1.2
The velocity eld and its impact on macrosegregation for Case 5 (Fig. 8.1)
was discussed in more detail in Section 7.4.2. Recall that the thermallydriven ow extracts some solute (Cu) from the mushy zone. The solute is
transported with a strong current (additional eect of solutal buoyancy) to
the bottom of the sump. The stratication tendency of the solute-rich liquid
creates an enriched region in the center, causing strong positive centerline
segregation.
The result for Case 5sh is shown in Fig. 8.2. The velocity eld is set up by
the mold feeding geometry and the shrinkage-induced ow. The ow enters
the mold and expands into an outward radial direction. The cooperation of
the inow geometry and the shrinkage feeding generates a ow component,
owing outward-parallel to the isotherms in the high-fl part of the mushy
zone. This is shown for Case 7sh in a close up of the ow eld near the liquidus
front in Figs. 8.11(b) and 8.10(b). While it is true that the shrinkage ow is
driven by the increase of density, which is directed opposite to the fl gradient,
this demonstrates that the ow in the porous region is determined by the ow
in the bulk liquid and the pressure relations in the porous mush, as the ow in
the high-fl part of the porous mush is governed by the momentum equation
which can in the low-fl regions be approximated by the Darcy equation.
h~vl il ~vcast =
1K
hPl il
gl l
(8.2)
Comparing the situation for Al-Mg, the picture in the case of the shrinkagedriven ow (Case 6sh , shown in Fig. 8.4) is very similar to the Al-Cu Case 5sh .
Case 6 with thermosolutal ow and shrinkage, on the other hand, shows a
picture very dierent from that of the Al-Cu Case 5. In a similar way, driven
by thermal convection and the inlet ow, solute (Mg) is transported into
the bulk liquid. Unlike in the Al-Cu case, however, where the heavy copper
ows downwards and straties at the sump bottom, magnesium creates a
solutal buoyancy force, cooperating with the thermal buoyancy eect and
driving the ow of enriched liquid upwards. Part of the horizontal ow deects
downwards to the sump bottom, carrying enriched liquid to the liquidus front.
This is the cause of the positive segregation in the central region. At the
centerline a downward owing current, originating from the inow, remains
unaected by the solutal eect. It carries solute-lean liquid to the sump
bottom and creates the relatively sharp negative segregation in the center.
The smaller overall Mg-segregation in Al-Mg is attributed to two factors: less
pronounced partitioning (larger kp ) and the absence of solutal stratication,
which causes a more uniform composition in the liquid sump.
173
CCu [%]
5.8
5.7
5.6
5.5
5.4
-0.05
z [m]
5.3
fl = 0
5.2
5.1
5.0
fl = 1
4.9
4.8
4.7
-0.10
4.6
4.5
4.4
0.00
0.05
0.10
r [m]
Figure 8.1: Mixture Cu concentration (Cm ) and velocity (~vm ) fields in the billet for
Case 5.
0.00
CCu [%]
5.8
5.7
5.6
5.5
5.4
-0.05
z [m]
5.3
fl = 0
5.2
5.1
5.0
4.9
fl = 1
4.8
4.7
-0.10
4.6
4.5
4.4
0.00
0.05
r [m]
0.10
Figure 8.2: Mixture Cu concentration (Cm ) and velocity (~vm ) fields in the billet for
Case 5sh .
174
CMg [%]
5.2
5.1
5.0
4.9
4.8
4.7
z [m]
-0.05
fl = 0
4.6
fl = 1
-0.10
0.00
0.05
r [m]
0.10
Figure 8.3: Mixture Mg concentration (Cm ) and velocity (~vm ) fields in the billet for
Case 6.
0.00
CMg [%]
5.2
5.1
5.0
4.9
4.8
4.7
z [m]
-0.05
fl = 0
4.6
fl = 1
-0.10
0.00
0.05
r [m]
0.10
Figure 8.4: Mixture Mg concentration (Cm ) and velocity (~vm ) fields in the billet for
Case 6sh .
175
0.00
CCu [%]
-0.05
z [m]
fl = 0
fl = 1
-0.10
5.8
5.7
5.6
5.5
5.4
5.3
5.2
5.1
5.0
4.9
4.8
4.7
4.6
4.5
4.4
-0.15
0.00
0.05
0.10
x [m]
Figure 8.5: Mixture Cu concentration (Cm ) and velocity (~vm ) fields in the ingot for
Case 7.
176
0.00
CCu [%]
-0.05
z [m]
fl = 0
fl = 1
-0.10
5.8
5.7
5.6
5.5
5.4
5.3
5.2
5.1
5.0
4.9
4.8
4.7
4.6
4.5
4.4
-0.15
0.00
0.05
0.10
x [m]
Figure 8.6: Mixture Cu concentration (Cm ) and velocity (~vm ) fields in the ingot for
Case 7sh .
177
CCu [wt%]
5.4
initial concentration
5.2
5.0
initial concentration
Case 5
Case 5sh
Case 5
Case 5sh
4.8
4.6
4.4
0.00
0.02
0.04
0.06
0.08
0.10
r [m]
Figure 8.7: Radial Cu macrosegregation profiles for Cases 5 and 5sh .
CMg [wt%]
5.2
initial concentration
5.0
4.8
initial concentration
Case 6
Case 6sh
Case 6
Case 6sh
4.6
4.4
0.00
0.02
0.04
0.06
0.08
0.10
r [m]
Figure 8.8: Radial Mg macrosegregation profiles for Cases 6 and 6sh .
178
CCu [wt%]
5.4
5.2
initial concentration
5.0
initial concentration
Case 7
Case 7sh
Case 7
Case 7sh
4.8
4.6
4.4
0.00
0.02
0.04
0.06
0.08
0.10
r [m]
Figure 8.9: Radial Cu macrosegregation profiles for Cases 7 and 7sh .
fraction gl
gl
hCl il
1
1 kp
(h~vl il h~vs is ) T
1+
DT /Dt
gl
hCl il
s l
s
Its integration in the absence of an eutectic reaction gives
Zgl
Zgl
1 kp
dgl
1
gs hCs is = kC0 exp
dgl
1
1 + gl
=
(8.3)
(8.4)
(8.5)
(~vl ~vs ) T
DT /Dt
(8.6)
(8.7)
and states
<1
positive macrosegregation
(8.8)
>1
negative macrosegregation,
(8.9)
179
CCu [%]
CCu [%]
5.8
5.8
5.7
5.7
5.6
5.6
5.5
5.5
5.4
5.4
5.3
5.3
5.2
5.2
5.1
5.1
5.0
5.0
4.9
4.9
4.8
4.7
4.6
-0.15
z [m]
z [m]
4.8
4.7
4.6
-0.15
4.5
4.5
4.4
4.4
0.00
0.00
x [m]
x [m]
Figure 8.10: Mixture Cu concentration (Cm ) field and relative velocity (~vl ~vs ) streamlines in the mushy zone at the ingot center.
z [m];y [m]
CCu [%]
0.00
CCu [%]
5.8
5.8
5.7
5.7
5.6
5.6
5.5
5.5
5.4
5.4
5.3
5.3
5.2
5.2
5.1
5.1
5.0
5.0
4.9
4.9
4.8
4.7
z [m]
0.00
4.8
4.7
4.6
4.6
4.5
4.5
4.4
4.4
-0.05
-0.05
0.10
x [m]
0.10
x [m]
Figure 8.11: Mixture Cu concentration (Cm ) field and relative velocity (~vl ~vs ) streamlines in the mushy zone at the ingot surface.
180
0.00
1.05
0.00
1.05
1.00
1.00
0.95
0.95
0.00
0.00
-0.05
z [m]
z [m]
-0.05
-0.10
-0.10
-0.15
-0.15
0.00
0.05
0.10
0.00
x [m]
0.05
0.10
x [m]
Figure 8.12: Macrosegregation criterion in the mushy zone of the Al-Cu billet.
0.00
1.05
0.00
1.05
1.00
1.00
0.95
0.95
0.00
0.00
-0.05
z [m]
z [m]
-0.05
-0.10
-0.10
-0.15
-0.15
0.00
0.05
0.10
x [m]
0.00
0.05
0.10
x [m]
Figure 8.13: Macrosegregation criterion in the mushy zone of the Al-Mg billet.
181
1.05
0.00
1.05
1.00
1.00
0.95
0.95
0.00
0.00
-0.05
z [m]
z [m]
-0.05
-0.10
-0.10
-0.15
-0.15
0.00
0.05
0.10
x [m]
0.00
0.05
0.10
x [m]
Figure 8.14: Macrosegregation criterion in the mushy zone of the Al-Cu ingot.
where positive and negative refers to increased and decreased nal concentration with respect to the initial concentration C0 .
The criterion is applied for all six cases investigated in this chapter. The
maps are shown in Figs. 8.12-8.14. We can see that for all cases with
shrinkage-driven ow only (Figs. 8.12(b), 8.13(b) and 8.14(b)), the criterion
clearly predicts positive segregation at the surface and negative segregation
in the center and in the region next to the surface. What is interesting is
the comparison to the corresponding cases with both thermosolutal ow and
shrinkage, shown in Figs. 8.12(a), 8.13(a) and 8.14(a). As we would expect,
the general situation in the deeper (i.e. low-fl ) region of the porous mushy
zone is retained. We can even observe a more pronounced tendency to negative segregation in the center. In the center, in the high-fl part of the mushy
zone, just below the liquidus front, the situation changes drastically due to
the changed path of liquid ow. In Cases 5 and 7, the path of relative liquid
motion out of the mushy zone (see Figs. 8.10(a) and 8.11(a) for Case 7) creates a tendency to positive segregation in the center (Figs. 8.12 and 8.14). In
Case 6 these eects are less pronounced (Fig.8.13), but they clearly create
zones with a tendency to negative segregation at points of ow penetrating
the mushy zone and a tendency to positive segregation at around r = 0.06 m,
where the ow detaches from the liquidus front (Fig.8.13(a)). In all three
cases (5, 6 and 7), we observe a thin region of negative segregation tendency
182
just below the liquidus front. It starts close to the surface and extends towards the center. Correlating it with the segregation proles, it becomes clear
that these zones must be causing the strong reversal of segregation, predicted
at the surface. At the surface the concentration of the liquid at the liquidus
front (the jet coming from the inlet) is close to the nominal concentration
in both cases (shrinkage only and combined thermosolutal convection and
shrinkage) so the dierence cannot originate in dierent initial melt concentrations but must be exclusively due to the relative motion of liquid and
solid in the mushy zone. Moving on to the center with the analysis, at around
mid-radius/mid-width the -map still shows a more pronounced negativesegregation tendency for cases 5 and 7 than for cases 5sh and 7sh , respectively.
Despite this, we observe a clear reversal to positive segregation in cases 5 and
7 (Figs. 8.7 and 8.9). This implies that the enrichment of the melt in the liquid region next to the liquidus front competes with the segregation tendency
in the mushy zone and outweighs the negative-segregation tendency of the
relative solid-liquid motion. A similar eect is observed in Case 6 at around
r = 0.02 . . . 0.05 m, where segregation is strongly positive despite the clearly
more negative segregation tendency in the mushy zone; clearly an impact of
the enriched melt penetrating the porous mush (the composition actually
decreases across the mushy zone but remains above the nominal due to the
high melt concentration at the onset of solidication at the liquidus front
Fig. 8.3). In the center, the eects of melt-concentration (stratication of
heavy solute-rich liquid in the Al-Cu cases and the solute-lean jet in the
Al-Mg case) and segregation tendency in the high-fl part of the mushy-zone
mostly cooperate. Thus it is not easy to say which eect is stronger, but
already from the discussion of mid-radius/mid-width segregation it is clear
that the bulk-melt composition at the liquidus plays an important role.
To summarize, the thermosolutal ow aects both the transport of solute
in the bulk liquid pool and the ow in the porous mushy zone (alas, only in
the high-fl part). The impact of thermosolutal convection on macrosegregation is a combined eect of both the altered segregation tendencies due to
the changed relative liquid-solid motion and the solute transport in the bulk
liquid region caused by natural (thermosolutal) and forced convection. The
eect of the mushy-zone ow is most obvious in the subsurface region where
a thin region with changed ow conditions drastically reverses the relative
segregation. The eect of bulk liquid concentration is most pronounced in
the region between the center and surface, where the eects of bulk-liquid
concentration and mush ow compete and the situation is quite delicate (
is close to unity). At the centerline we have seen (in the studied cases) that
both eects cooperate to some degree.
8.2
183
8.2.1
The ranges of of gsc and dg were chosen based on values used in previous model
studies [Vreeman and Incropera, 2000, Wu et al., 2003, Vreeman et al., 2002],
and those determined in experimental investigations [Daloz et al., 2002,Eskin
et al., 2004]. The model studies use grain diameters between 25 and 100 m.
The experiments show grain sizes in the solidied cast structure of around
150 to 250 m in billets [Eskin et al., 2004] and between 300 and 500 m
for non-grain rened [Daloz et al., 2002] and between 120 and 270 m for
grain rened ingots [Daloz et al., 2002]. The grains in the slurry zone will
be considerably smaller than in the cast structure. As we do not expect the
nucleation of new grains but only further growth (recall the discussion on
grain initiation and growth in Section 4.4.6), we can roughly estimate the
grain size by dg dgsc /2 25 . . . 150 m. In laboratory experiments the
consolidation solid fraction is usually observed at around 30 % of solid and
should not vary a lot. Vreeman, Schloz and Krane [Vreeman et al., 2002]
surmised from comparisons of numerical and experimental results that gsc
could be somewhere around 20 % to 25 % in grain-rened DC cast billets.
These estimations remain relatively speculative, however, and I decided to
study the variation between 15 % and 45 % solid volume fraction. In short,
to study the impact of the consolidation fraction gsc and the grain size dg ,
six calculations with varying gsc and dg are compared. Their parameters are
summarized in Table 8.3.
The casting parameters, thermophysical properties, boundary conditions
and computational grids are identical to Case 1 in Chapter 7. For the solution of the momentum equation the QUICK discretization scheme was used,
ensuring high (second-order) accuracy (also see Section 6.11).
184
8.2.2
The results of the simulations are presented in Figs. 8.15-8.20, showing the
solid fraction in the sump and the streamlines of the solid phase, and additionally in Figs. 8.21-8.26, showing the mixture velocity eld. In all cases we
observe a strong ow of free-oating grains down along the liquidus front.
Once they reach the grain accumulation at the bottom of the sump their
path depends primarily on the grain size, which more strongly aects the
relative velocity of grains. Smaller grains are carried by the liquid, driven
by thermal convection in the bulk liquid region and in the slurry zone an
important portion is transported upwards, where they remelt. With a large
grain size, on the other hand, the settling tendency is more pronounced as is
also evident from Fig. 8.27 where the proles of gs along the centerline are
shown. Despite the eect of uid ow, most solid grains end up attached to
the porous matrix. In Case G30D150 (Fig. 8.18) the solid that forms in the
slurry region is observed to settle to the bottom, an eect caused by the large
grain size. This is probably not realistic since the grains in the high-fl part
of the slurry should be much smaller than average in reality. It is also interesting to see that larger grains eectively coalesce at a lower solid fraction,
their further path, once gs gets close to gsc , being drastically shorter than for
smaller grains (compare gs. 8.17 and 8.18).
The variation in the consolidation solid fraction does not fundamentally
change the path of the solid in the slurry region but does strongly aect the
extent and shape of the mushy zone. A higher gsc enhances the transport of
the solid to the bottom of the sump and thus the adverse transport of latent
heat, resulting in a colder sump. At higher gsc a larger fraction of the solid
in the center of the solidied casting will originate from free-oating grains
and this zone will be wider.
The solid streamlines show that the solid in the center forms from solid
grains transported from the outer regions of the billet. The grains slide down
with the ow along the liquidus front and reach deep into the center of the
Table 8.3: Summary of cases: Effect of moving solid grains
Case
G00
G30D075
G30D025
G30D150
G15D075
G45D075
gsc
0
0.30
0.30
0.31
0.15
0.45
dg
75 m
25 m
150 m
75 m
75 m
185
casting. The grains higher in the mostly isothermal slurry region are partly
a result of the portion of grains originating from the periphery being deected by the ow upwards, partly remelting completely and partly traveling
downwards and attaching to the porous solid region.
It becomes evident that the solid in the center originates in most part from
the solid grains that start to grow at the periphery and grow isothermally in a
thin layer close to the liquidus before accumulating at the bottom. Due to the
settling tendency of the solute-lean solid and adverse ow of liquid upwards
in the slurry zone this should cause a solutal stratication tendency, with a
positive (upward) gradient of concentration (in terms of the mixtureCm ).
This should also cause a more isothermal slurry region, with the solid fraction
gradients being more aected by the compositional stratication than by
temperature. In the Al-Cu case this implies that the solid phase, which ends
up in the center of the solidied casting, does not start growth from a strongly
enriched liquid (accumulated at the bottom due to stratication) but from a
solute lean solid and lower-concentration liquid.
In the Al-Mg case we have seen that the solutal buoyancy force can drive
enriched interdendritic liquid extracted from the mushy zone away from the
liquidus front. There is no solutal stratication and liquid enrichment in the
high-fl regions at the sump bottom is thus less pronounced. In the presence of
moving grains the grains should grow at a low concentration, with the solutal
buoyancy being weaker than the sedimentation tendency, which should cause
the downward ow of the lean solid and additional drainage of solute-rich
liquid into the bulk liquid region. The eect in the billet center will be the
same as for Al-Cu, although the dierence from the prediction of the porous
model should be smaller as there is no solutal stratication eect in AlMg and the overall segregation is smaller also due to the less severe solute
partitioning.
In summary, as a consequence of solid transport we expect the formation
of a negative or lower concentration region at the centerline. A larger grain
size should cause a narrower centerline segregation band. A larger consolidation fraction should cause more segregation with a wider distribution. The
region close to the surface should not be signicantly aected. Conservation
of the solute dictates a positive segregation zone between the surface and
center, but any more detailed speculation on the patterns in this zone is
impossible as the interplay of the governing phenomena is very delicate.
186
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.15: Influence of moving solid grains: Streamlines of solid and solid volume
fraction in the billet at gsc = 0 (rigid solid phase, Case G00).
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.16: Influence of moving solid grains: Streamlines of solid and solid volume
fraction in the billet at gsc = 0.30, dg = 75 m (Case G30D075).
187
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.17: Influence of grain size: Streamlines of solid and solid volume fraction in the
billet at gsc = 0.30, dg = 25 m (Case G30D025).
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.18: Influence of grain size: Streamlines of solid and solid volume fraction in the
billet at gsc = 0.30, dg = 150 m (Case G30D150).
188
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.19: Influence of consolidation solid fraction: Streamlines of solid and solid volume fraction in the billet at gsc = 0.15, dg = 75 m (Case G15D075).
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.20: Influence of consolidation solid fraction: Streamlines of solid and solid volume fraction in the billet at gsc = 0.45, dg = 75 m (Case G45D075).
189
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.21: Influence of moving solid grains: Mixture velocity (~vm ) field and solid volume
fraction in the billet at gsc = 0 (rigid solid phase, Case G00).
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.22: Influence of moving solid grains: Mixture velocity (~vm ) field and solid volume
fraction in the billet at gsc = 0.30, dg = 75 m (Case G30D075).
190
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.23: Influence of grain size: Mixture velocity (~vm ) field and solid volume fraction
in the billet at gsc = 0.30, dg = 25 m (Case G30D025).
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.24: Influence of grain size: Mixture velocity (~vm ) field and solid volume fraction
in the billet at gsc = 0.30, dg = 150 m (Case G30D150).
191
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.25: Influence of consolidation solid fraction: Mixture velocity (~vm ) field and
solid volume fraction in the billet at gsc = 0.15, dg = 75 m (Case G15D075).
0.00
gs
1.00
0.45
0.30
0.15
0.00
z [m]
-0.05
-0.10
-0.15
0.00
0.05
r [m]
0.10
Figure 8.26: Influence of consolidation solid fraction: Mixture velocity (~vm ) field and
solid volume fraction in the billet at gsc = 0.45, dg = 75 m (Case G45D075).
192
-0.04
G00
G30D075
G30D025
G30D150
G15D075
G45D075
-0.05
-0.06
-0.07
z [m]
-0.08
-0.09
-0.10
-0.11
-0.12
-0.13
-0.14
0.2
0.4
0.6
0.8
gs
Figure 8.27: Longitudinal profiles of solid volume fraction (gs ) along the centerline. The
respective consolidation solid fraction (gsc ) is marked by a cross.
9.1
194
195
196
9.2
Conclusions
Conclusions
197
198
9.3
199
Industrial Relevance
In a broader context the present work relates to the development of integrated process modeling. This idea was previously propagated, e.g. by the
VIR[*] projects (VIR[CAST], VIR[FAB] and VIR[FORM]) [VIR[*], 2004],
aimed at modeling an aluminum production chain from casting operations
through to nished products. The model developed by the author ts into
the similar, alas much smaller-scale, context of aluminum production modeling initiated at the Laboratory for Multiphase Processes, University of Nova
Gorica [arler et al., 2006]. The idea is to couple models of casting (macrosegregation), microsegregation, and homogenization. The local solidication
path of solid determined by the casting model should be used as an input
for a decoupled microsegregation model [Vuanovi, 2004, Vuanovi et al.,
2005] using thermodynamic model data [JMatPro, 2004] which can estimate
the microscopic solute distribution in primary Al dendrites and the quantities
and compositions of other solid phases in the cast structure. This information
is to be used as input data for the homogenization model of solid-solid phase
transformations [Kovaevi et al., 2005b, Kovaevi et al., 2005a, Kovaevi
and arler, 2005]. Given properly dened requirements for the homogenized
microstructure this model chain should help optimize the production chain
for extruded products.
Further relevance of this work for industry lies in its exploitation of the
understanding of the formation of macrosegregation to develop technical measures for macrosegregation optimization. Details of such developments lie
beyond the scope of the present work as they require a prior denition of
requirements for macrosegregation tolerances and a survey of the possibilities to redesign or supplement the installed casting equipment, both of which
must be dened in close cooperation with process engineers.
Volume Averaging
Volume averaging is a formal procedure used to derive macroscopic continuum equations for individual phases in a multiphase system. The procedure
is performed by averaging the governing equations over the pertinent phase
inside a volume element, signicantly larger than the scale of the phase structures. This volume element is usually called the representative elementary
volume (REV) or the averaging volume. The volume averaging concept relies
upon scale separation among the microscopic continuum scale m , the scale of
the primary solid phase structures , and the macroscopic continuum scale
L (m L). The main advantage of volume averaging over alternative theories used to derive macroscopic equations is that volume averaging
clearly denes the relations between the microscopic and the volume-averaged
macroscopic variables and thus also elucidates the origins of the individual
volume-averaged terms in the macroscopic equations [Drew, 1983]. The procedures of volume averaging employed here are extracted mainly from the
review paper by Drew [Drew, 1983].
The volume average of a quantity is dened as
Z
1
dV,
(A.1)
hi =
V0
V0
where V0 is the volume of the averaging volume (REV). Accordingly, the local
value of a variable is
= hi + ,
(A.2)
where hi is the average over an averaging volume centered in the point of
interest and is the local deviation from the average in the point of interest.
Accordingly
D E
= 0
(A.3)
The averaging process is assumed to satisfy the following properties
h + i = hi + hi
201
(A.4)
202
Volume Averaging
hhi i = hi hi
(A.5)
hconsti = const
hi
=
t
t
hi
=
xi
xi
(A.6)
(A.7)
(A.8)
Xk
= ~nk (~x ~xi )
n
(A.11)
Xk
n
1
=
V0
V0
(A.12)
Ak
1
Xk
dV =
n
V0
Ak
dA =
Ai
= Si ,
V0
(A.13)
where Ai is the phase interface area and Si the volumetric interface density.
The volume integral of the eld quantity multiplied by (Xk /n) thus
picks out the interface and equals the integral of over the phase interface.
The volume average of (Xk /n) is equal to the interfacial area per unit
203
volume Si . Further, Drew [Drew, 1983] shows that the phase function obeys
the relationship
Xk
+w
~ i Xk = 0,
t
(A.14)
where w
~ i is the velocity of the phase interface.
For use in the derivation that follow, we here further dene the phase
quantity averages and the properties of their product and derivatives. The
volume average of quantity in phase k over the volume V0 is
1
hk i = hXk i =
V0
Xk dV
(A.15)
V0
The intrinsic volume average, i.e. the average value of the quantity inside
phase k is
Z
1
k
k
Xk dV = gk hk i
(A.16)
hk i = hXk i =
Vk
V0
The local deviation of the phase quantity is dened via the denition
k = hk ik + k
(A.17)
= hk i hk i + hk i k + hk i k + k k
E
D Ek D
D Ek
k
= hk i hk i + hi k + hi k + k k
D
E
= hk i hk ik + k k
(A.18)
204
Volume Averaging
Xk
Xk =
(Xk )
=
t k
t
t
t
=
(Xk ) + h w
~ i Xk i
t
hk i
Xk
=
+ w
~ i ~nk
t
n
Z
1
hk i
+
k w
~ i ~nk dA
=
t
V0
Ak
Z
1
k
gk hk i +
k w
~ i ~nk dA
=
t
V0
(A.19)
Ak
(A.20)
Ak
~nk dA
k
Z
1
k
= gk hk i
hk ik + k ~nk dA
V0
Ak
Z
1
Xk
k
k
k
k~nk dA
= gk hk i + hk i gk hk i ~nk
n
V0
Ak
Z
1
= gk hk ik + hk ik gk hk ik gk
k~nk dA
V0
Ak
Z
1
= gk hk ik
k~nk dA
V0
Ak
(A.21)
205
Finally, to summarize we write
Z
Z
1
1
k
k~nk dA = gk hk i
k~nk dA
hk i = hk i
V0
V0
Ak
(A.22)
Ak
E D
E
D
E D
E
~
~ = Xk
~
~ Xk
= Xk
k
Z
D E
1
~ k ~nk dA
~k
=
V0
A
D E k
Z
k
1
~
~ k ~nk dA
= gk k
V0
Ak
Z
D Ek D Ek
1
~
~ k ~nk dA
~
= gk k + k gk
V0
Ak
(A.23)
Thermophysical Properties
Solid density
Liquid density
Solid specic heat
Liquid specic heat
Solid thermal conductivity
Liquid thermal conductivity
Solid diusivity
Liquid diusivity
Latent heat at eutectic
Liquid viscosity
Thermal expansion coecient
Solutal expansion coecient
Reference temperature
Reference concentration
Reference density
Permeability constant
Pure Al melting temperature
Al-Cu eutectic temperature
Eutectic solidication interval
Eutectic concentration
Cu solubility in Al at Teut
(Partition coecient)
[JMatPro, 2004],
s
l
cps
cpl
ks
kl
Ds
Dl
Leut
T
C
T0
C0
0
K0
Tf
Teut
Teut
Ceut
C,eut
(kp )
[kg/m3 ]
[kg/m3 ]
[J/kgK]
[J/kgK]
[W/mK]
[W/mK]
[m2 /s]
[m2 /s]
[J/kg]
[Pa s]
[K1 ]
[]
[K]
[]
[kg/m3 ]
[m2 ]
[K]
[K]
[K]
[wt% Cu]
[wt% Cu]
[]
207
2750
2460
1030
1130
180
80
12
5 10
5 109
3.77 105
1.3 103
1.17 104
0.73
973.15
0.0525
2460
6.67 1011
933.6
821.2
1
33.21
5.74
(0.173)
208
Thermophysical Properties
Solid density
Liquid density
Solid specic heat
Liquid specic heat
Solid thermal conductivity
Liquid thermal conductivity
Solid diusivity
Liquid diusivity
Latent heat at eutectic
Liquid viscosity
Thermal expansion coecient
Solutal expansion coecient
Reference temperature
Reference concentration
Reference density
Permeability constant
Pure Al melting temperature
Al-Mg eutectic temperature
Eutectic solidication interval
Eutectic concentration
Mg solubility in Al at Teut
(Partition coecient)
[JMatPro, 2004],
s
l
cps
cpl
ks
kl
Ds
Dl
Leut
T
C
T0
C0
0
K0
Tf
Teut
Teut
Ceut
C,eut
(kp )
[kg/m3 ]
[kg/m3 ]
[J/kgK]
[J/kgK]
[W/mK]
[W/mK]
[m2 /s]
[m2 /s]
[J/kg]
[Pa s]
[K1 ]
[]
[K]
[]
[kg/m3 ]
[m2 ]
[K]
[K]
[K]
[wt% Mg]
[wt% Mg]
[]
2510
2340
1080
1180
165
85
12
1 10
3 109
3.30 105
1.3 103
1.22 104
0.49
963.15
0.0500
2340
6.67 1011
933.6
723.44
1
34.15
16.19
(0.474)
Experimental Measurements
This section presents an attempt of confrontation of model results with industrial experimental data. While it is not within the scope of the present
work to perform an extensive experimental program or to calibrate the developed numerical model, it is indeed desirable to show and discuss the relation
of model results to measured data.
Macrosegregation was measured in a DC cast non-grain-rened binary Al4.6 wt%Cu billet, cast by the WagstaTM AirSlipTM system, presenting practically identical process conditions as Case 1 in Chapter 7, with exception of
the alloy composition (Al-5.25 wt%Cu in Case 1). Two transversal sections
of the billet were made and in each section the chemical composition was
measured with optical emission spectrometry (OES) in a line across the diameter (sample shown in Fig. C.1). The number of measurement points was
26 on the rst section (Prole 1 in Fig. C.2) and 22 on the second section
(Prole 2 in Fig. C.2). The results are shown in Fig. C.2.
Figure C.1: Optical emission spectrometry sample of the first billet section (Profile 1 in
Fig. C.2).
210
Experimental Measurements
5.2
Profile 1
Profile 2
C0
5.1
CCu [wt%]
4.9
4.8
4.7
4.6
4.5
4.4
-0.1
0.1
r [m]
Figure C.2: Measured concentration of copper over two transversal sections of a 282 mmdiameter Al-4.6 wt%Cu billet.
develop with the AirSlip mold technology), this interpretation seems somewhat preconceived in light of the results of Prole 2. The latter shows smaller
deviations, which are mostly of opposite sign. Further, it shows a very clear
(over three measurement points) strong peak at r = 0.02 . . . 0.06 m, clearly
distorting the expected symmetry of the prole.
Due to the unclear repeatability, these experimental results cannot be
directly used to validate the numerical model. More samples should be taken
at several billet cross sections before we can arm the experimental observation. An attempt to get more detailed experimental data was made at
Impol Aluminium Industry, by taking OES samples at transversal planes of
a bar cut across the diameter of the transversal billet section. By gradually
grinding away layers of the bar in the transversal direction this technique
allows to take samples at smaller distances (i.e. more samples across the
diameter). The results showed good resolution and repeatability (the measurements were made before and after homogenizing heat treatment and
showed identical trends). However, due to an unidentied systematic error in
the measurements [Dragojevi, 2006], they could not be used. The employed
measurement technique was thus recognized as ecient, but the problems
211
arisen during the measurements should be identied and the experiments
should be repeated to gather reliable data. Unfortunately, this experimental
program was not completed before the nalization of the present work.
Despite that, the results of the rst experiment can be used to dene the
total deviation of composition. We see in Fig. C.2 that it can be estimated at
0.7 wt%, which is a maximum relative deviation of (C Cu C0Cu )/C0Cu 15 %.
The numerical results of Case 1 in Chapter 7, which has comparable conditions, show a relative deviation of about 50 % on the other hand (Fig. 7.10).
Given the uncertain interpretation and reliability of the gathered experimental data, the model results were further confronted with experimental
results from the literature. A relevant source are the measurements by Eskin
et al. [Eskin et al., 2004]. They cast 200 mm-diameter Al-4.5 wt%Cu billets
on a laboratory DC caster equipped with hot-top molds. These castings can
be roughly compared with Case 2 in Chapter 7. We see that the measured
maximum relative deviation is around 10 % [Eskin et al., 2004], while the
model predicts around 20 % (Fig. 7.10).
The large discrepancies between numerical predictions and experimental
results in terms of estimated maximum macrosegregation deviations cannot be completely claried at this point. It was clear already at the beginning, however, that a quantitatively correct segregation map can hardly be
expected from the porous-mush model (Chapter 7) as even the qualitative
segregation proles are not realistic. All discussions on DC casting macrosegregation mechanisms, made previously in this chapter, and those already
presented in the Literature review (Chapter 2), point to the tentative conclusion that the transport of free-oating grains is a phenomenon that decisively
changes the local and global segregation deviations. The porous-mush model
is thus not directly applicable for segregation prediction but is a tool for
studying the possible ow patterns in the liquid pool and the interaction of
ow and mushy zone on the periphery of the solidication zone.
Bibliography
AIAA (1998). AIAA Guide for the Verication and Validation of Computational Fluid Dynamics Simulations (G-077-1998). AIAA Standards
Series. AIAA, Reston (VA), USA.
Banaszek, J. and Furmaski, P. (2005). Momentum equation in macroscopic
modelling of interdendritic ow in the columnar mushy zone. In Biaecki,
R. A. and Wecel, G., editors, Eurotherm Seminar 82: Numerical Heat
Transfer 2005, pages 397404, Gliwice, Poland. Institute of Thermal
Technology, Silesian University of Technology.
Barnes, H. A., Hutton, J. F., and Walters, K. (1998). An Introduction to
Rheology. Elsevier, Amsterdam.
Beckermann, C. (1987). Melting and Solidication of Binary Mixtures with
Double-Diusive Convection in the Melt. PhD thesis, School of Mechanical Engineering, Purdue University, West Lafayette (IN), USA.
Beckermann, C. and Viskanta, R. (1988). Double-diusive convection during
dendritic solidication of a binary mixture. Physico Chemical Hydrodynamics, 10:195213.
Beckermann, C. and Viskanta, R. (1993). Mathematical modeling of transport phenomena during alloy solidication. Applied Mechanical Review,
46:127.
Bejan, A. (2004). Convection Heat Transfer. John Wiley and Sons, Hoboken
(NJ), USA, 3rd edition.
Bennacer, R. and Gobin, D. (1996). Cooperating thermosolutal convection
in enclosuresI. scale analysis and mass transfer. International Journal
of Heat and Mass Transfer, 39(13):26712681.
Bennon, W. D. and Incropera, F. P. (1987). A continuum model for momentum, heat and species transport in binary solid-liquid phase change
213
214
BIBLIOGRAPHY
systemsI. model formulation. International Journal of Heat and Mass
Transfer, 30(10):21612170.
BIBLIOGRAPHY
215
216
BIBLIOGRAPHY
Drikakis, D. and Rider, W. (2005). High-Resolution Methods For Incompressible And Low-Speed Flows. Computational Fluid and Solid Mechanics.
Springer-Verlag, Berlin.
Du, Q., Eskin, D., and Katgerman, L. (2006). Modeling macrosegregation
during DC casting of a binary aluminium alloy. In Gandin, C.-A. and
Bellet, M., editors, Modeling of Casting, Welding and Advanced Solidication Processes XI, pages 235242, Warrendale (PA), USA. TMS.
Eskin, D. G., Savran, V. I., and Katgerman, L. (2005). Eects of melt temperature and casting speed on the structure and defect formation during
direct-chill casting of an Al-Cu alloy. Metallurgical and Materials Transactions A, 36A:19651976.
Eskin, D. G., Zuidema Jr., J., Savran, V. I., and Katgerman, L. (2004). Structure formation and macrosegregation under dierent process conditions
during DC casting. Materials Science and Engineering A, 384:232244.
Ferziger, J. H. and Peri, M. (1996). Computational Methods for Fluid Dynamics. Springer-Verlag, Berlin, 2nd edition.
Flemings, M. C., Mehrabian, R., and Nereo, G. E. (1967). Macrosegregation:
Part II. Transactions TMS-AIME, 242:4149.
Flemings, M. C. and Nereo, G. E. (1967a). Macrosegregation: Part I. Transactions TMS-AIME, 239:14491461.
Flemings, M. C. and Nereo, G. E. (1967b). Macrosegregation: Part III. Transactions TMS-AIME, 242:5055.
Fletcher, C. A. J., Orszag, S. A., Holt, M., and Glowinski, R. (1991). Computational Techniques for Fluid Dynamics 1: Fundamental and General
Techniques. Springer-Verlag, Berlin, Heidelberg, 1st edition.
Flood, S. C., Katgerman, L., and Voller, V. R. (1991). The calculation of
macrosegregation and heat and uid ows in the DC casting of aluminum
alloys. In Rappaz, M., Ozgu, M. R., and Mahin, K. W., editors, Modeling
of Casting, Welding and Advanced Solidication Processes V, pages 683
690, Warrendale (PA), USA. TMS.
Furmaski, P. (1992). Eective macroscopic description for heat conduction
in heterogeneous materials. Interantional Journal of Heat and Mass
Transfer, 35(11):30473058.
BIBLIOGRAPHY
217
218
BIBLIOGRAPHY
Hortmann, M., Peri, M., and Scheuerer, G. (1990). Finite volume multigrid
prediction of laminar natural convection: Bench-mark solutions. International Journal for Numerical Methods in Fluids, 11(2):189207.
Huppert, H. E. and Worster, M. G. (1985). Dynamic solidication of a binary
melt. Nature, 314:703707.
Hyun, M. T., Kuo, D. C., Bergman, T. L., and Ball, K. S. (1995). Direct
simulation of double diusion in low Prandtl number liquids. Numerical
Heat Transfer A, 27:639650.
IAI (2006). Statistics of the International Aluminium Institute (IAI).
http://www.world-aluminium.org.
Iida, T. and Guthrie, R. I. (1993). The Physical Properties of Liquid Metals.
Clarendon Press, Oxford Science Publications, Oxford.
Ishii, M. (1975). Thermo-Fluid Dynamic Theory of Two-Phase Flow. Eyerolles, Paris.
Jalanti, T. (2000). tude et modlisation de la macrosgrgation dans la
coule semi-continue des alliages daluminium. PhD thesis, cole polytechnique fdrale de Lausanne, Lausanne.
Jalanti, T., Swierkosz, M., Gremaud, M., and Rappaz, M. (2001). Modelling
of macrosegregation in continuous casting of aluminium. In Ehrke, K.
and Schneider, W., editors, Continuous Casting, pages 191198, Weinheim, Germany. DGM Publ., Wiley VCH.
Janssen, R. J. A. and Henkes, R. A. W. M. (1993). Accuracy of nite-volume
discretizations for the bifurcationg natural-convection ow in a square
cavity. Numerical Heat Transfer A, 24:191207.
JMatPro (2004). JMatPro the materials property simulation package, version
3.0, Sente Software Ltd., Guildford, UK. Computer program.
Joly, A., Grn, G.-U., Daloz, D., Combeau, H., and Lesoult, G. (2000). Eect
of grain renement on macrosegregation in direct chill semi-continous
cast aluminium sheet ingot. Materials Science Forum, 329-330:111120.
Kaempfer, T. U. (2002). Modeling of Macrosegregation Using an Adaptive Domain Decomposition Method. PhD thesis, cole polytechnique
fdrale de Lausanne, Lausanne.
BIBLIOGRAPHY
219
Kamakura, K. and Ozoe, H. (1996). Oscillatory phenomena of low-Prandtlnumber uids in a rectangular cavity. Numerical Heat Transfer A,
30:427437.
Kattner, U. R. (1997). The thermodynamic modeling of multicomponent
phase equilibria. JOM, 49(12):1419.
Kitanovski, A., Vuarnoz, D., Ata-Caesar, D., Egolf, P. W., Hansen, T. M.,
and Doetsch, C. (2005). The uid dynamics of ice slurry. International
Journal of Refrigeration, 28:3750.
Kovaevi, I. and arler, B. (2005). Solution of a phase-eld model for dissolution of primary particles in binary aluminum alloys by an r-adaptive
mesh-free method. Materials Science and Engineering A, 413-414:423
428.
Kovaevi, I., arler, B., and Kores, S. (2005a). Numerino modeliranje
procesa homogenizacije, drugi del: Mikroskopski model, zlitine: D50,
D60, AC30, AC41, AC42. Technical report, Politehnika Nova Gorica,
Nova Gorica, Slovenia.
Kovaevi, I., Vertnik, R., and arler, B. (2005b). Numerino modeliranje
procesa homogenizacije, prvi del: Makroskopski model, zlitine: D50, D60,
AC30, AC41, AC42. Technical report, Politehnika Nova Gorica, Nova
Gorica, Slovenia.
Krane, M. J. M. (2004). Macrosegregation development during solidication
of a multicomponent alloy with free-oating solid particles. Applied
Mathematical Modelling, 28:95107.
Krane, M. J. M., Incropera, F. P., and Gaskell, D. R. (1997). Solidication
of ternary metal alloysI. model development. International Journal of
Heat and Mass Transfer, 40(16):38273835.
Kurz, W. and Fisher, D. J. (1998). Fundamentals of Solidication. Trans
Tech Publications, Aedermannsdorf, Switzerland, 4th edition.
Lage, J. L. and Bejan, A. (1991). The Ra-Pr domain of laminar natural
convection in an enclosure heated from the side. Numerical Heat Transfer
A, 19:2141.
Larson, R. G. (1998). The Structure and Rheology of Complex Fluids. Oxford
University Press, New York.
220
BIBLIOGRAPHY
BIBLIOGRAPHY
221
222
BIBLIOGRAPHY
Reddy, A. V. and Beckermann, C. (1995). Simulation of the eects of thermosolutal convection, shrinkage induced ow and solid transport on macrosegregation and equiaxed grain size distribution in a DC continuous
cast Al-Cu round ingot. In Voller, V. R., Marsh, S. P., and El-Kaddah,
N., editors, Materials Processing in the Computer Age II, pages 89102.
TMS.
Reddy, A. V. and Beckermann, C. (1997). Modeling of macrosegregation due
to thermosolutal convection and contraction-driven ow in direct chill
continuous casting of an Al-Cu round ingot. Metallurgical and Materials
Transactions B, 28B(3):479489.
Ridder, S. D., Kou, S., and Mehrabian, R. (1981). Eect of uid ow on
macrosegregation in axi-symmetric ingots. Metallurgical and Materials
Transactions B, 12B:435447.
Roache, P. J. (1998a). Fundamentals of Computational Fluid Dynamics.
Hermosa, Albuquerque (NM), USA.
Roache, P. J. (1998b). Verication and Validation in Computational Science
and Engineering. Hermosa, Albuquerque (NM), USA.
Roy, C. J. (2005). Review of code and solution verication procedures for
computational simulation. Journal of Computational Physics, 205:131
156.
arler, B., Kovaevi, I., Vertnik, R., Hartman, S., Vuanovi, I., Zalonik,
M., afhalter, R., Slaek, E., Dragojevi, V., Jelen, M., Strnad, V., and
Robi, A. (2006). Integrated numerical simulation aproach in Impol Aluminium Industry casthouse. In International Conference on Aluminium,
21-22 September, 2006, Essen, Germany. DGM.
arler, B. and Pepper, D. (2004). Momentum transport modeling including
solid phase movement and nucleation for direct chill casting processes. In
Bergles, A. E., Golobi, I., Amon, C. H., and Bejan, A., editors, Thermal
Sciences 2004: Proceedings of the ASME - ZSIS International Thermal
Science Seminar ITSS II, Ljubljana. ZSIS.
arler, B., Zalonik, M., and Vertnik, R. (2003). Razvoj makroizcejnih modelov za polkontinuirno in trano ulivanje aluminijevih zlitin (in slovenian:
Development of models of macrosegregation in direct-chill and twin-roll
strip casting of aluminum alloys). Technical Report PNG-DP-046, Politehnika Nova Gorica, Nova Gorica, Slovenia.
BIBLIOGRAPHY
223
Saunders, N., Li, X., Miodownik, A. P., and Schill, J.-P. (2001). Computer
modeling of materials properties. In Zhao, J.-C., Fahrmann, M., and
Pollock, T. M., editors, Materials Design Approaches and Experiences,
Warrendale (PA), USA. TMS.
Scheil, E. (1942). Zeitschrift fr Metallkunde, 34(70-72).
Shyy, W. and Chen, M.-H. (1991). Double-diusive ow in enclosures.
Physics of Fluids A, 3(11):25922607.
Suyitno, Eskin, D. G., Savran, V. I., and Katgerman, L. (2004). Eects
of alloy composition and casting speed on structure formation and hot
tearing during direct-chill casting of Al-Cu alloys. Metallurgical and
Materials Transactions A, 35A:35513561.
Trucano, T. G., Pilch, M., and Oberkampf, W. L. (2003). On the role of code
comparisons in verication and validation. Technical Report SAND20032752, Sandia National Laboratories, Albuquerque (NM), USA.
Venneker, B. C. H. (2001). A quest for the best discretisation scheme to
prevent numerical diusion and dispersion at high Pclet number ows.
Technical Report p.01.4.023, Netherlands Institute for Metals Research,
Delft, The Netherlands.
Venneker, B. C. H. and Katgerman, L. (2000). Macrosegregation during DC
casting of aluminum alloys: Numerical issues and the eect of metal
entry. In Sahm, P. R., Hansen, P. N., and Conley, J. G., editors, Modeling of Casting, Welding and Advanced Solidication Processes IX, pages
680686, Aachen, Germany. Shaker Verlag.
Venneker, B. C. H. and Katgerman, L. (2002). Modelling issues in macrosegregation predictions in direct chill castings. Journal of Light Metals,
2(3):149159.
Vertnik, R., Zalonik, M., and arler, B. (2006). Solution of transient directchill aluminium billet casting problem with simultaneous material and
interphase moving boundaries by a meshless method. Engineering Analysis with Boundary Elements, 30(10):847855.
VIR[*] (2004). VIR[*] conference 2004. Aluminium, 80(6):550752.
Volak, D., Zalonik, M., and arler, B. (2005). Izraun regulacijskaih koecientov polkontinuirnega ulivanja drogov na livni napravi Wagsta za
spekter aluminijevih zlitin in spekter formatov. Technical report, Politehnika Nova Gorica, Nova Gorica, Slovenia.
224
BIBLIOGRAPHY
BIBLIOGRAPHY
225
226
BIBLIOGRAPHY