Beruflich Dokumente
Kultur Dokumente
genehmigte Dissertation
vorgelegt von
Diplom-Physiker, M.A.
Georg Ehlen
Wrselen
Band 45
Georg Ehlen
Shaker Verlag
Aachen 2004
Bibliographic information published by Die Deutsche Bibliothek
Die Deutsche Bibliothek lists this publication in the Deutsche
Nationalbibliografie; detailed bibliographic data is available in
the internet at http://dnb.ddb.de.
Printed in Germany.
ISBN 3-8322-2886-1
ISSN 1435-6198
Nach ber zehn Jahren am Gieerei-Institut, in denen ich mit vielen Leuten zusammengearbeitet
und -gelebt habe, mchte ich mich bedanken bei...
After ten years of work at the Foundry Institute where I met and worked with many people I want to
say thank you to...
...Prof. Dr.-Ing. Dr.-Ing. E.h. Peter R. Sahm und Prof. Dr. rer. nat. Andreas Ludwig fr die lang-
jhrige Betreuung und die Freiheit, die Ausrichtung meiner Forschungen sehr selbstndig gestalten
zu knnen,
...Prof. Dr.-Ing. Andreas Bhrig-Polaczek fr den notwendigen Fertigstellungsdruck gegen Ende
der Arbeit und die Finanzierung der letzten Monate,
...der Deutschen Forschungsgemeinschaft (DFG) fr sechs Jahre kontinuierliche Frderung im
Rahmen der Forschergruppe Werkstoffbezogene numerische Simulation thermischer Prozesse in
der Produktionstechnik unter Sa 335/30,
...the European Commission for support through IHP grant number HPRI-1999-CT-00026 (the
TRACS programme at EPCC),
...der Besatzung von Cafeteria Raum 403, besonders Angela Goldbach, Joachim Wittich, Jrgen
Wolf, Monika Wirth, Roland Scheer und allen, die hin und wieder stndig zum Kaffeetrinken,
reden, reden und reden vorbeikamen, fr alles, was das Leben am GI so lebenswert macht,
...Peter Otten, Martin Feller, Nils Warnken, Toni Ivas, Menghuai Wu und nochmals Andreas
Ludwig und Jrgen Wolf fr hilfreiche, ausufernde, laute und leise Diskussionen ber Gott, die
Welt und ... SoliCon1.0,
...Andi Schweizer fr die beste Erfindung seit Windows 95: Bild2D,
...Detlef Kube fr sechs Wochen Sand stampfen und die Betreuung im Kindergarten fr Alumini-
umgu,
...den Ingenieuren Bjrn Pustal und Exequiel Garcia fr berzeugende Hilfe bei meinen allerletzten
Experimenten,
...den guten Geistern im Sekretariat, der Metallographie, dem Fotolabor, in der Analytik, in der
mechanischen Werkstatt und in der Giehalle, sowie allen Kollegen am GI und bei ACCESS fr
eine gute Zusammenarbeit und ausgezeichnete Atmosphre,
...Margaret Mearns for special support to do my PhD in five weeks and for checking and correct-
ing my awful English,
...the TRACS team at the Edinburgh Parallel Computing Centre, especially Mario Antonioletti,
Mark Bull, J-C Desplat and Catherine Inglis, for a warm welcome and excellent support during all
my stays in Scotland, and to TRACS visitor Arturo for a big birthday trip to the end of the world,
...Prof. Dr. Wolfgang Mller und Markus Gross fr fortgesetzte Gastfreundschaft auf meinen Dien-
streisen nach Edinburgh,
...Marc Schneider for important support in a difficult starting time,
...zuletzt, aber nicht zuletzt (an besonderer Stelle) meinen Eltern fr langjhrige selbstlose Unter-
sttzung!
TRANSIENT NUMERICAL SIMULATION OF COMPLEX CONVECTION
EFFECTS DURING SOLIDIFICATION IN
CASTING AND WELDING
Georg Ehlen
Abstract
In casting and welding processes the properties of the finally solidified ingot or the resolidified
weld are governed by convection effects via the transport of mass, heat or alloying elements. Sub-
ject of this work is to provide a substantial contribution to research in the area of complex convec-
tion effects which require the coupled modelling of many single aspects.
Development of new basic models and algorithms
In order to simulate complex convection effects a number of innovative basic models describing
single aspects of casting and welding processes had to be developed:
Modified VOF-model to describe the solidification of moving free surfaces with partial solidifi-
cation of surface cells.
Split Solid Model to describe the transport of floating equiaxed grains and to prevent premature
solidification of the moving free surface.
Porosity model for the simultaneous but separate prediction of hydrogen and shrinkage poros-
ities, as well as internal shrinkage cavities.
Solidification algorithm to describe eutectic and peritectic solidification of binary alloys in sys-
tems with strong convective solute and mass transport.
Coupling of new basic models with state of the art models
The new basic models have been integrated together with state of the art models in order to form a
framework which allows the coupled simulation of thermosolutal convection, buoyancy driven free
surface movement, shrinkage flow, Marangoni flow, and flow due to electromagnetic forces and has
been used to simulate the following complex convection effects:
Influence of the formation of a of deep external shrinkage cavity on the final macrosegregation
pattern in cast ingots
Influence of the sedimentation of globulitic crystals on the flow field, macrosegregations and
the shape of the external shrinkage cavity
Solute redistribution and pool shape formation in laser and GTA spot welding processes.
Application of the integrated models
The integrated models have been applied to the following scientific problems:
Prediction of deep shrinkage cavities in ingot casting of unalloyed steel,
Simulation of shrinkage cavity formation in a cylindrical Al-7wt%Si ingot,
Simultaneous but separate prediction of hydrogen and shrinkage porosities during casting of Al-
7wt%Si; experimental validation for pure shrinkage porosity formation
Transient effects during pool shape formation in laser and GTA spot welding processes; Maran-
goni flow influences the pool shape as a function of welding power and duration
Solute redistribution during laser spot welding processes.
TRANSIENTE NUMERISCHE SIMULATION KOMPLEXER KONVEKTI-
ONSEFFEKTE WHREND DER ERSTARRUNG BEIM
GIESSEN UND SCHWEISSEN
Georg Ehlen
Zusammenfassung
Bei Gie- und Schweiprozessen werden die Eigenschaften des fertigen Gustcks sowie der
Schweinaht durch Konvektionseffekte whrend des Prozesses ber den Transport von Masse,
Wrme und Legierungselementen bestimmt. Ziel dieser Arbeit ist es, einen wesentlichen Beitrag
zur Forschung auf dem Gebiet komplexer Konvektionseffekte zu leisten, die die gekoppelte Model-
lierung vieler Einzelaspekte erfordern.
Entwicklung neuer Einzelmodelle und -algorithmen
Um komplexe Konvektionseffekte simulieren zu knnen, musste eine Reihe innovativer Modelle
fr einzelne Aspekte von Gie- und Schweiprozessen entwickelt werden:
Modifiziertes VOF-Modell zur Beschreibung der Erstarrung beweglicher freier Oberflchen mit
partieller Erstarrung von Oberflchenzellen.
Split Solid Modell zur Beschreibung des Transport schwimmender globulitischer Kristalle und
zur Verhinderung eines zu frhen Einfrierens der beweglichen freien Oberflche
Porosittsmodell zur gleichzeitigen aber separaten Vorhersage von Wasserstoff- und Schrump-
fungsporositt sowie von internen Schrumpfungslunkern.
Erstarrungsalgorithmus zur Beschreibung eutektischer und peritektischer Erstarrung binrer
Legierungen in Systemen mit starkem konvektivem Konzentrations- und Massentransport
Kopplung der neuen Einzelmodelle mit etablierten (Stand der Technik) Modellen
Die neuen Einzelmodelle wurden kombiniert mit etablierten Modellen, um ein System zur gekop-
pelten Simulation thermosolutaler Konvektion, schwerkraftgetriebener Oberflchenbewegung,
Schrumpfungsstrmung, sowie Marangoni- und Lorentzkonvektion zu erhalten. Dieses wurde ver-
wendet, um folgende komplexen Konvektionseffekte zu simulieren:
Einfluss der Bildung tiefer Oberflchenlunker auf Makroseigerungen in Gublcken,
Einfluss der Sedimentation globulitischer Kristalle auf das Strmungsfeld, Makroseigerungen,
und die Form des Oberflchenlunkers,
Konzentrationsumverteilung und Entwicklung der Schweibadform beim Laser- und WIG-
Punktschweien.
Anwendung der gekoppelten Modelle
Die gekoppelten Modelle wurden auf folgende wissenschaftliche Probleme angewandt:
Vorhersage tiefer Oberflchenlunker beim Blockgu von unlegiertem Stahl,
Simulation des Oberflchenlunkers in einem zylindrischen Al-7Gew%Si Block,
Gleichzeitige aber separate Vorhersage von Wasserstoff- und Schrumpfungsporositt beim Gie-
en von Al-7Gew%Si; experimentelle Validierung fr reine Schrumpfungsporositt,
Transiente Effekte bei der Schweibadausbildung beim Laser- und WIG-Punktschweien; Ma-
rangonikonvektion bestimmt die Badform als Funktion von Schweileistung und -dauer,
Konzentrationsumverteilung beim Laserschweien.
Table of contents
0 Executive summary ...................................................................................................................I
0.1 Introduction ...........................................................................................................................I
0.2 Modelling of convection effects in casting and welding processes......................................IV
0.3 State of the art.....................................................................................................................IV
0.4 Mathematical model (SoliCon) ............................................................................................ V
0.5 Algorithms..........................................................................................................................XI
0.6 Applications: Simulations and experiments ..................................................................... XIX
0.7 Conclusions and outlook...................................................................................................LIII
1 Introduction ..............................................................................................................................1
1.1 Subject of this work ..............................................................................................................3
1.1.1 Development of new basic models................................................................................3
1.1.2 Modelling of complex convection effects......................................................................4
1.1.3 Application of the models .............................................................................................4
2 Modelling of convection effects in casting and welding processes
Physical phenomena / State of the art / Contribution by this work ............................................6
2.1 Macrosegregations................................................................................................................6
2.2 Formation of shrinkage cavities and porosities ................................................................... 18
2.3 Movement of floating globulitic grains / twophase flow ................................................... 30
2.4 Solidification and fluid flow: Thermal and solutal effects................................................... 39
2.5 Complex integrated models for casting and welding........................................................... 44
2.5.1 Casting models............................................................................................................44
2.5.2 Welding models .......................................................................................................... 45
2.6 Application: Convection effects in welding processes ........................................................ 48
3 Theory .................................................................................................................................... 58
3.1 Mathematical model ........................................................................................................... 58
3.2 Further assumptions and definitions.................................................................................... 62
3.2.1 Phase diagrams used ................................................................................................... 63
3.3 Equations............................................................................................................................ 63
3.3.1 Continuity equation..................................................................................................... 64
3.3.2 Momentum equations.................................................................................................. 65
3.3.3 Energy equation .......................................................................................................... 73
3.3.4 Species conservation equations ................................................................................... 76
3.3.5 Modified free surface (VOF) equation ........................................................................ 77
3.3.6 Split solid equations.................................................................................................... 79
3.3.7 Porosity formation equations....................................................................................... 84
4 Algorithms.............................................................................................................................. 93
4.1 General properties of the code ............................................................................................ 93
4.2 Flow charts of SoliCon2.0 .................................................................................................. 93
4.3 Discretisation of differential equations................................................................................ 95
4.3.1 Grid definitions ........................................................................................................... 95
4.3.2 Power law for convection diffusion term..................................................................... 96
4.3.3 Explicit and implicit discretisation of continuity equation........................................... 97
4.3.4 Semi-explicit discretisation of momentum equations .................................................. 99
4.3.5 Explicit discretisation of the energy equation............................................................ 104
4.3.6 Explicit discretisation of the liquid concentration conservation equation .................. 107
4.4 Semi-explicit version of the SIMPLER algorithm............................................................. 110
4.4.1 Semi-explicit momentum equations .......................................................................... 111
4.4.2 Implicit pressure equation ......................................................................................... 111
4.5 Stable phase change algorithm.......................................................................................... 113
4.5.1 Basic idea.................................................................................................................. 113
4.5.2 Solution strategy ....................................................................................................... 115
4.5.3 Solidification modules .............................................................................................. 117
4.5.4 Guessing strategy ...................................................................................................... 122
4.5.5 Checking and correction rules................................................................................... 126
4.6 Algorithm for solidification of moving free surfaces......................................................... 132
4.6.1 Discretisation of the modified VOF advection equation............................................ 132
4.6.2 Surface reconstruction algorithm............................................................................... 135
4.6.3 Boundary conditions at the free surface..................................................................... 135
4.6.4 Correction algorithms................................................................................................ 137
4.6.5 Interpretation of the modified VOF advection equation ............................................ 138
4.7 Split Solid Algorithm........................................................................................................ 144
4.8 Porosity formation algorithm ............................................................................................ 145
4.8.1 Prediction of hydrogen porosity formation................................................................ 145
4.8.2 Prediction of shrinkage porosity formation ............................................................... 146
4.8.3 Criterion determining whether microporosities or macroscopic cavities form ........... 146
5 Implementation Details......................................................................................................... 148
5.1 Interpolation of volume fractions to cell faces .................................................................. 148
5.1.1 Interpolation of phase fractions ................................................................................. 148
5.1.2 Interpolation of localised quantities........................................................................... 149
5.1.3 Interpolation of volume averaged quantities.............................................................. 150
5.1.4 Interpolation of densities: Upwind model.................................................................. 152
5.2 Adaptive time step control ................................................................................................ 152
5.3 Convergence control structures......................................................................................... 153
5.4 Parallelisation ................................................................................................................... 154
5.5 Image processing software................................................................................................ 156
6 Applications: Simulations and experiments, Casting processes............................................. 158
6.1 Test problems ................................................................................................................... 158
6.1.1 Heat conduction benchmark...................................................................................... 158
6.1.2 Sloshing tank benchmark .......................................................................................... 160
6.1.3 Test of the phase change algorithm ........................................................................... 162
6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot .................................. 165
6.2.1 Results from /Bhmer 97/ ......................................................................................... 166
6.2.2 Simulations with new extended model ...................................................................... 168
6.3 Formation of shrinkage cavity in Al-7wt%Si ingot ........................................................... 182
6.3.1 Experiment................................................................................................................ 182
6.3.2 Simulation................................................................................................................. 182
6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys ................. 185
6.4.1 Simulation of shrinkage porosities ............................................................................ 185
6.4.2 Experimental validation of shrinkage porosity formation.......................................... 188
6.4.3 Simulation of hydrogen porosities............................................................................. 189
6.4.4 Coupled calculation of hydrogen and shrinkage porosities........................................ 191
6.4.5 Conclusions............................................................................................................... 194
7 Applications: Simulations and experiments, Welding processes ........................................... 196
7.1 Systematical investigation of pool shape formation in welding......................................... 196
7.1.1 Numerical set-up....................................................................................................... 196
7.1.2 Pool shapes with constant Marangoni coefficients (laser welding)............................ 197
7.1.3 Pool shapes with temperature dependent Marangoni coefficients (laser welding) ..... 212
7.1.4 Power dependent pool shape evolution in GTA welding........................................... 230
7.2 Solute redistribution in weld pools (laser welding) ........................................................... 234
7.2.1 Investigation of numerical concentration errors......................................................... 235
7.2.2 Formation of macrosegregations in no viscosity case................................................ 236
7.2.3 Comparison viscosity / no viscosity .......................................................................... 238
7.2.4 Discussion of redistribution model............................................................................ 240
8 Conclusions and Outlook...................................................................................................... 243
8.1 Conclusions ...................................................................................................................... 243
8.2 Outlook............................................................................................................................. 246
9 Appendices ........................................................................................................................... 249
9.1 Material data..................................................................................................................... 249
9.1.1 Density...................................................................................................................... 249
9.1.2 Heat conductivity ...................................................................................................... 250
9.1.3 Alloys used for the applications ................................................................................ 250
9.2 Remarks on the genesis / history of the model .................................................................. 253
10 Nomenclature and Abbreviations.......................................................................................... 255
11 References ............................................................................................................................ 263
0 Executive summary
0.1 Introduction
In casting and welding processes many of the properties of the finally solidified ingot or the re-
solidified weld are governed or at least influenced by convection, i.e. fluid flow phenomena which
have an impact via the transport of mass, heat or alloying elements. In general the term convection
is defined as:
A transfer of heat or mass that occurs when a fluid flows over a solid body or inside a channel
while temperatures or concentrations of the fluid and the boundary are different; transfer occurs
within the fluid as a consequence of the motion within the fluid relative to the flow boundary
/Academic 96/
In this work the term convection will be used to describe all kinds of fluid flow which occur in
molten alloy systems. The different types of convection can be classified by their governing forces.
The types of convection considered here are:
Thermosolutal convection (buoyancy flow due to combined temperature and concentration de-
pendent density gradients)
Shrinkage flow (driven by air pressure and initiated by temperature dependent density and vol-
ume losses during phase change)
Gravity induced surface movement (surface waves and drop of the plane liquid surface)
Marangoni convection (surface flow due to surface tension gradients, very important for weld-
ing and crystal growth)
Electromagnetic convection (GTA welding)
Mould filling and other forced types of convection, like flow due to stirring, will not be considered.
Numerical simulation has proven to be a powerful tool which can be used to understand and govern
the influence of convection on solidification processes. Even though intensive research has been
performed for more than two decades, there are still many important tasks left. Especially complex
convection effects which require the coupling of many single effects make high demands on ma-
thematical and numerical models.
The subject of this work is to provide a substantial contribution to the research on numerical simu-
lation of complex convection effects which occur in casting and welding processes. The work
focuses on three tasks:
1. Developing a number of new innovative basic models and algorithms which describe single
aspects of casting and welding processes.
II Chapter 0 Executive summary
2. Coupling these new models together with state of the art models in order to form a framework
of interacting models which can be used to simulate higly complex convection effects.
3. Applying the complex models to real scientific problems and thus to demonstrate their applica-
bility and value.
a)
b)
Fig. 0.1: Main convection effects which occur during a) casting process of steel ingot with
riser, modelled in software SoliCon; b) laser spot welding process with vertical incidence of
laser beam, modelled in EL2D. For details on the two software packages, cf. App. 9.2.
Influence of the formation of a of deep external shrinkage cavity on the final macrosegregation
pattern in cast ingots
Influence of the sedimentation of globulitic crystals on macrosegregations, the flow field and
the shape of the external shrinkage cavity
Solute redistribution and pool shape formation in laser and GTA welding processes.
Transient effects during pool shape formation in laser and GTA welding processes (EL2D);
Marangoni flow influences pool depth and width as a function of welding power and duration.
Solute redistribution during laser spot welding processes (EL2D), Fig. 0.1 b). Leading scientists
have declared this task to be one of the research issues [that] must be resolved if there are to be
further advances in our fundamental understanding of the complex phenomena that occur during
welding, /Zacharia 95/.
it is principally possible. The influence of hydrogen porosities on the amount of inverse segrega-
tions has already been determined.
What concerns the application of welding simulations, systematic parameter studies are rare. In
particular the investigation of transient phenomena requires further effort.
Even if the influence of convection effects on solute redistribution has been recognised as one of the
research issues [that] must be resolved if there are to be further advances in our fundamental un-
derstanding of the complex phenomena that occur during welding, /Zacharia 95/, there are almost
no publications covering this.
In the strict thermodynamic sense there are only three (alloy) phases: Liquid, -solid and - solid
(in the case of peritectic Fe-C solidification, for eutectic Al-Si solidification it would be liquid, solid
(Al)-phase and solid (Si)-phase. For the SSM each of the solid alloy phases is split into a mobile
and a stationary part. The remaining four phases are: The two porosity types shrinkage and hy-
drogen porosities; obstacles, which are massive blocked out regions used to describe mould mate-
rial; void, which corresponds to areas outside the cast part and mould, usually filled with air, but
here considered to be empty.
VI Chapter 0 Executive summary
For each of these phases the following physical quantities have to be determined:
Velocities in x-, y- [and z-] direction: uq, vq[, wq] Concentration of solute species i: C i
This results in a total of 45 + 9nelements variables where nelements is the number of alloying elements.
In this work only binary alloys are considered, therefore nelements is always 1 and i is always the bi-
nary alloy element (C in case of Fe-C and Si in case of Al-Si). The high number of 54 variables can
be reduced by a number of assumptions. Thermal equilibrium between all phases and pressure
equilibrium between all mobile phases and the porosities yields
T := Tq p := pl = p move = p move = p pore shrink = p pore hydro p dead = p dead = pobst = 0
Mobile and stationary part of the same solid phase are assumed to be in solutal equilibrium:
C i := C i move = C i dead
C i := C i move = C i dead
C ipore shrink = C iporehydro = Cobst
i
= Cvoid
i
=0
The 54 variables have been reduced to 16: u, v, T, p, Cl, C , C , and the nine phase fraction vari-
ables q. In the next sections the 16 equations are presented which are used to solve this problem.
The quantities are volume-averaged. To simplify the representation the following definitions are
used:
move dead :=
/ q where move l move move dead dead dead obst = , , = , ,
q= move / dead
mat : = q
q void
mat eff mat = ( q q )
q void
1. Continuity equation:
v
( moveeff move dead eff dead ) + [ move eff move vl ] = 0
v
q q =
+ (0.1)
t
v
u v
[
l move eff move l l
+ v
v
= move
vu]
p
move eff move
t
l v
v
[
move eff move l l
v v]
v v
+ v
[
eff move move l
v +
u]
move v v
v move v v
u
v v
l + ( move l ) eff move l + ( move l )
[
eff move move l v]
eff move
x +
x
y y
(
move eff move dead + dead ) 2
q q + C (C qi C qi ref ) g
v
+
q
vvl (0.2)
q = l , move ,
move
K 0 move 2
Consequence of the SSM is the presence of the mobile solid phases in all terms that are usually a
function of liquid fraction only.
4. Energy equation:
The energy equation was developed from a mixed enthalpy equation considering enthalpies as lin-
ear functions of temperature, hq = h0 q + c P q (T T0 q ) . The complete equation is:
[( move eff move c P eff move + dead eff dead c P eff dead )T ] v v
+ ( move eff movec P eff movevvl T mat eff mat T )
t
( ) v
[ *q ( q qvl )] q ( q Sexternal q )
qq v
(0.3)
*
= q + +
q l t q = move
move
where *q = (h0 l c P l T0 l ) (h0 q c P qT0 q ) = const. and Sexternal q are external heat sources for phase q.
Consequences of the SSM are the presence of the mobile solid phase fractions in the convective
term and the new term on the right side of the equation which describes the transfer of latent heat by
convection of mobile solid.
l l
i
(0.4b)
The liquid concentration equation was developed from a volume-averaged mixture concentration
equation:
( C ) + v ( vv C ) = v ( D~ v C )
i i i i
t l l l l l l l l l l l
~
t [( q move + q dead )qCqi ] ( q moveqvl Cqi ) + ( q dead q Dqi deadCqi )
v v v v
+ (0.5)
q= ,
VIII Chapter 0 Executive summary
Main consequence of the SSM is the convective term on the right side of the equation which ac-
counts for the solute transport by mobile solid. There is no diffusive term for the mobile solid
phases because they are assumed not to be in contact.
There are still nine equations necessary in order to determine the nine phase fractions.
move
where Fmove :=
open
A way to determine the shrinkage porosity formation rate is to calculate the diminished vapour
pressure and to limit the minimum local pressure to this value. The momentum equations using the
reduced pressure gradients yield smaller velocities and in the divergence terms of the continuity
equation an additional mass sink appears. The continuity equation is a mass conservation equation.
This is why the the term
t
( pore shrink pore shrink )
with pore shrink = 0 (by definition, pore shrink 0 would not help much either...) does not have direct
impact on the equation. The impact comes through the displacement of the mobile phases. The time
dependent term for the mass of liquid and mobile solid must be modified to become
t
[
( move pore shrink ) eff move ]
to account for the reduced fraction of mobile phases. In the continuity equation it equilibrates the
velocity divergence created by the reduced pressure gradients. So the equation to determine the rate
of shrinkage porosity formation becomes
( ( v v v
pore shrink eff move ) = move eff move + dead eff dead ) + [ move eff move v l (p limited )] (0.9)
t t
Other than the model by /Pequet 00/ who treat mushy zone flow and solidification of completely
separated areas (liquid pockets) as different cases, the current model does not need this differen-
tiation. The porosity formation rate only depends on whether the volume losses due to shrinkage
can be compensated or not. Whether the feeding path is hindered or completely blocked does not
make any difference.
The fraction of hydrogen porosities pore hydro is determined as a function of the fraction of newly
formed solid and local solidification conditions
p0 T s
pore hydro = p (V V ) dt (0.10)
T0 t s H H max
To determine the overall solid fractions a concept is used which has been described in /Prakash 89/.
One of the two equations is obtained from the phase diagram:
T (C ) = T l
i
liq
(0.11)
X Chapter 0 Executive summary
Unfortunately this equation does not contain any of the phase fractions in demand, whereas the
temperature and liquid concentration equations do:
T = T (
, ) C = C (
i
l
i
l
, )
But as they are already used to calculate the temperature and liquid concentration respectively, they
give no additional information about the phase fractions. The missing equation is obtained from the
conditions arising from the current solidification type. Four different types are considered:
primary -solidification : Only liquid and solid is present = 0 (0.12a)
primary -solidification : Only liquid and solid is present = 0 (0.12b)
eutectic / peritectic solidfication : Liquid, solid and solid is present, The missing equation
arises from the condition, that the solidification is linked to the eutectic / peritectic plateau.
T ( , ) = Teut peri
/
(0.12c)
Solid state transformation / : No liquid is present = alloy (0.12d)
a) b)
Fig. 0.2: a) Comparison of three possible Split Solid Functions (SSF). b) Fraction of mobile
solid as a function of total solid fraction for three types of SSF.
0.5 Algorithms XI
stationary. - and - solid are both split up into mobile part and stationary part in the same way.
Split solid equations. These rules lead to the following set of equations:
move = SSF ( + ) (0.13a)
move = SSF ( + )
(0.13b)
The Split Solid Function (SSF) can be chosen in different ways. So far three function types have
been tested, Fig. 0.2. For the present calculations the sinus type function was used.
Columnar Growth Model. Close to mould walls or areas with high solid fraction solidification in
many cases is not equiaxed but columnar. To describe this phenomenon a Columnar Growth Model
(CGM) has been developed. The current version of this model switches off the SSM if inside a ra-
dius RCGM there is either a solid wall, an obstacle or solid fractions above a limit CGM.
T 1 + K seg a s T
The energy equation has got additional models for heat input by a Gauss shaped laser or GTA
welding source, and heat losses by radiation, air convection and vaporization.
0.5 Algorithms
this work often referred to as solidification algorithm) used to model solidification and remelting.
The task of this algorithm is to find a set of phase fractions
, , l with + + l = alloy
alloy is the total volume fraction covered by alloy material and thus taking part in the solidification
process. The total volume balance for each cell is:
alloy + pore shrink+ pore hydro + obst + void = 1
During the application of the solidification algorithm alloy is considered to be constant.
One key problem of solidification algorithms is how to treat the onset and end of solidification. For
cases which do not consider concentration changes this is a one-dimensional problem dealing with
vertical movements through the phase diagram. A more difficult situation arises if changes of the
local mixture concentration also occur. In particular casting problems including shrinkage induced
feeding flow, and welding problems considering Marangoni convection, involve high flow veloci-
ties, strong convective solute transport and, thus, strong horizontal movements through the phase
diagram. Fig. 0.3 shows a typical (C , T ) path for a cell of Fe-0.42wt%C which has been partially
molten and resolidifies as Fe-0.14wt%C. The new algorithm was designed to be unconditionally
stable for arbitrary (C , T ) paths inside the phase diagram.
(C , T ) depends on
the amount of heat, latent heat and solute advected or removed by diffusion, convection and the
transport of mobile solid,
the amount of latent heat released or consumed by solidification or remelting,
the amount of solute transferred from one phase into another by solidification or remelting.
The latter two points imply the knowledge of the amount of phase change occuring during the time
step. Therefore the aim of the solidification algorithm must be to find out what kind of solidification
or remelting process occurs in the special situation defined by the terms of the differential equations
and the phase diagram.
One way to treat the problem is to solve the differential equations without considering the phase
change terms and, if this leads to a solution inside a two- or three-phase-area, to determine the cor-
rect phase fractions by iterative adaption, e.g. /Laschet 98/.
For this work a different approach has been chosen. It also uses the result of the differential equa-
tions without the phase change terms, but mainly as information that helps to guess which solidifi-
cation type will most probably occur. The principle is the following:
For each cell the temperature resulting from the diffusive and convective heat transport terms of
1
the energy equation is modified by the amount of phase change occuring. The range of possible
phase changes yields two limiting temperatures: One, Tmax , for the case of maximum release of
latent heat possible and one, Tmin , for the case of maximum consumption of latent heat. If the cor-
rect solidification type has been found, the final temperature can only lie between these two tem-
peratures.
The same holds for the concentration equations. The diffusive and convective terms of the liquid
concentration equation yield a liquid concentration which can be modified by solidification or
2
remelting only in a range between Cl max for the case of maximum solute release and C l min for the
case of maximum solute consumption.
The task of the solidification algorithm is to check for this area Tmax , Tmin , Cl max , Cl min of the phase
diagram, which of the possible solidification types yields
a target point (C , T ) inside one of the phase areas
a set of phase fractions that is reasonable for the same phase area.
For each of the possible solidification types the temperature and concentration equations are used
together with additional equations resulting from the phase diagram, to form algebraic equations
which allow us to directly calculate the phase fractions, temperature and concentration of the cell,
1
As explicit temperature and concentration equations are used, the problem can be solved for each cell separately.
2
The solid concentration equations are directly linked to it by the lever rule.
XIV Chapter 0 Executive summary
provided that the solidification type is the correct one to apply in the current situation. In these so-
lidification modules iteration is only used to find the null of the algebraic equations.
The concept which is used in all three solidification modules is basically described in /Prakash 89/.
Fig. 0.4 shows how the triple of quantities T, Cl and the non-zero solid fraction s in the case of
primary solidification are determined using the three equations
T = T ( ) s
C = C ( )
l l s
T = T (C )liq l
As already stated in section 0.5.2.1 the solution of the differential equations for T and Cl leaves one
degree of freedom, s. The final values depend on how much solid actually forms or remelts. Solidi-
fication leads to the relaease of latent heat and results in a higher equilibrium temperature. At the
same time the rejection of solute by the newly formed solid results in a higher equilibrium liquid
concentration. Remelting leads to lower temperatures due to consumption of latent heat and at the
same time lower Cl due to remolten depleted solid. Therefore in Fig. 0.4 variation of s results in the
point (T(s), Cl(s) ) following the bold curve. The point where the curve cuts the liquidus line of the
phase diagram is described by the third equation T = T (C ) and indicates how much solid actually
liq l
forms or remelts. The modules for solid state transformation and eutectic or peritectic solidification
work similarly.
A guessing algorithm has been designed to reduce the number of trials necessary to find this correct
solidification type. The next section gives a more detailed overview of the solution strategy.
2. The preliminary position (C*, T * ) determined by solving the energy and liquid concentration
equations of the new time step without considering the influence of solidification or remelting.
The guessing algorithm uses a set of rules which are based on probabilities as to which solidifica-
tion type to expect, to determine in which order the solidification modules are tried out. A trial con-
sists of
assuming a certain solidification type for the current situation,
applying the corresponding solidification module,
checking the resulting set of phase fractions, temperature and concentrations for consistency.
In some cases even an unsuccessful application of a solidification module can be a valid result,
helping to dermine the correct values of the phase fractions using physical considerations. This
is especially the case for the end of solidification or remelting, when the system steps out of a
two-phase area into a single phase area.
If the algorithm declares that the right solidification type has been found, the set of new phase
a) b)
Fig. 0.5: Development of solid and liquid phases for the solidification of a binary Fe-C alloy with
constant mixture concentration C and uniform cooling. a) C = 0.12 wt % C (C < C < C ); the cri-
teria for the end of peritectic solidification / remelting are l < 0 / < 0; b) C = 0.25 wt % C (C <
C < Cl ); the criteria are < 0 / < 0.
XVI Chapter 0 Executive summary
a) b)
Fig. 0.6: Development of solid and liquid phases for the solidification of a binary Al-Si alloy with
constant mixture concentration C and uniform cooling. a) hypoeutectic composition with C = 7 wt
% Si (C < Cl ); the criteria for the end of eutectic solidification / remelting are l < 0 / < 0; b)
hypereutectic composition with C = 17 wt % Si (Cl < C ); the criteria are l < 0 / < 0.
fractions is returned to the main algorithm and the differential equations are now solved using the
correct new phase fraction values.
Fig. 0.5 and 0.6 show four examples of peritectic and eutectic solidification processes, calculated
using the new phase change algorithm.
where
( eff move )e
Vx = and CF = max([ F FAD ] V x [ F FDonor ]( x )Donor ,0 )
( eff move )P (ul )e t
0.5 Algorithms XVII
with
F = max (FDonor , FDM , 0.1)
Here AD is a cell which can either be the acceptor cell or the donor cell, depending on the fluid con-
figuration. DM is the cell upstream of the donor cell. The redefined VOF fluxes (CVOF) are intro-
duced into the discretised VOF advection equation. Using the definition
( open )P (Fmove )P = ( move )VOF
P (0.16)
the total change due to the VOF algorithm
P ( move )P = ( open )P (Fmove )P ( open )P (Fmove )P
VOF
( move )P = ( move )VOF 0
becomes
( move )VOF
P = ( open )P (Fmove )P ( open )P (Fmove )0P
( )0 ( ) ( ) ( )0 ( )0
= ( open )0P eff move P ( open )P (Fmove )0P closed P eff closed P closed P eff closed P (0.17)
( eff move )P ( eff move )P
+
1
x
[( open )w CVOF
( )w ( open )e (CVOF )e ] +
1
y
[( open )s CVOF
( )s ( open )n (CVOF )n ]
After each iteration of the pressure solver all cells where the pressure has dropped under the critical
pressure pcrit pore shrink are set back to this pressure and the cells are marked for pore formation being
in progress. When the pressure field has converged, the velocity equations are solved. As the pres-
sure drop which would have been necessary to ensure mass conservation has been limited inside the
porosity formation cells, there will be more outflow than inflow in these cells.
The resulting mass loss is calculated using the implicit continuity equation on the normal grid and
transformed into a porosity formation rate. The fraction pore shrink of newly generated shrinkage
porosities is calculated as
q (qdiscretised )P
( pore shrink )P = ( (0.18)
eff move )P x y z
To get the mass balance for the cells right again, this fraction is subtracted from the new fraction of
liquid l and added to the fraction of shrinkage porosities pore shrink . This is done before solving the
differential equations for temperature and concentrations.
is calculated at the beginning of each new time step. As the hydrogen porosities push the mobile
phases away, the hydrogen porosity formation rate porehydro = ( porehydro )P ( porehydro ) 0P is
introduced into the discretised / pressure equation as a transient term which reduces the phase frac-
tion of the mobile phases.
0.6 Applications: Simulations and experiments XIX
0.5.5 Parallelisation
The software was parallelised using the shared memory standard OpenMP. Typical welding simu-
lations without moving free surfaces take up to 10 days of calculation time on a 180 MHz R10000
processor. For casting simulations the grid refinements and the increasing complexity of the cou-
pled models increase the calculation times. The decision to use shared memory parallelisation
OpenMP was based on a detailed analysis of the serial code, typical applications and the hardware
available.
a) b)
Fig. 0.7: a) Geometry of cast part; the figure shows which faces of the ingot are in contact with the
sand mould (white + bottom), insulation (dark grey) and air (light grey); b) polished cut image of
the resulting cast part.
SSM: 0 mv dd = 0.3
mv dd = 0.2
CGM: RCGM = 0.01 m
CGM = 0.6
corresponds to the one of the cast alloy GS60 used by /Bhmer 97/. All material data, if not other-
wise mentioned are those described in section 9.1.3.2. If shrinkage flow is considered, temperature
dependent liquid density is used and a constant and higher solid density.
Before simulating the full problem a number of preliminary investigations have been made to study
the influence and importance of different submodels like SSM, CGM, thermosolutal convection and
shrinkage flow on the formation of the macroscopic external shrinkage cavity.
solid fraction []
t = 600 s t = 1200 s
vmax = 0.00139 m/s 0.0120 m/s vmax = 0.00081 m/s 0.0049 m/s
Fig. 0.12: Importance of thermosolutal convection for the formation of deep shrinkage cavities;
Comparison of calculations with SSM and CGM for case with shrinkage flow only (left half) and
shrinkage flow with thermosolutal convection (right half); solid fraction and velocities; the two
maximum velocity values are for the left and right half of the picture.
0.6.1.2.4 Complete calculation with SSM, CGM, shrinkage and thermosolutal convection
When the SSM is completed by the CGM and shrinkage and thermosolutal flow is considered, the
simulation of a deep shrinkage cavity becomes possible. Fig. 0.13 shows the results of a calculation
using all models apart from the porosity formation model. The calculation considers macroscopic
diffusion in the liquid and an isotropic permeability model (K0 = 810 m ).
-10 2
The left half of each picture shows the development of the velocity field and the total fraction of
solid. The right half shows the formation of the macrosegregation pattern, described by the percent-
age of carbon concentration deviation from the initial concentration.
The development of the flow field shows three distinct phases:
1. 20 100 s: Thermal convection
2. 100 600 s: Oscillating flows at the bottom due to sedimentation of cool melt and mobile solid
XXIV Chapter 0 Executive summary
Fig. 0.13 (part 2): Shrinkage cavity formation of Fe-0.42wt%C ingot; simulation with SSM, CGM,
shrinkage and thermosolutal convection; development of solid fraction and flow field (left half)
and macrosegregation pattern (right half).
carbon concentration
deviation [%]
a) b) c)
Temperature [K] Fig. 0.15: Formation of shrink-
age cavity during solidification
728 745 762 779 796 813 830 846 864 881 900 of Al-7wt%Si ingot; simulated
a) temperature and solid frac-
Solid fraction s [] tion, b) velocities + mushy zone
contours (0.020.98) and de-
0.05 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.95 1 viation of Si-concentration from
Si concentration deviation [%] initial concentration after 80 s;
c) Polished cut image of real
cast part.
-4,5 3,5 -2,5 1,5 -0,5 0,5 1,5 2,5 3,5 4,5 5,5
Maximum hydrogen
contents in the solid: VH max = 0.1 cm / 100 g3
fraction of
shrinkage
porosities
[]
t = 10 s t = 10.5 s t = 12 s t = 13 s t = 14 s t = 15 s
Fig. 0.17: Prediction of shrinkage porosity formation in cast Al-7wt%Si ingot; development of main
shrinkage porosity area; volume fraction of shrinkage porosities in grey scale representation (top)
and porosity representation (bottom); the area of black dots is proportional to the area of shrinkage
porosities; proportionality factor: 3.
path has completely closed. It starts beneath the point where feeding is blocked and then develops
downwards. New porosity only forms at exactly one point, more exactly at one height of the cast
part. Only when all liquid at this height has either solidified or dropped out of the cell will porosity
formation continue at lower levels. This is an effect of gravity. When one point reaches the critical
pressure for pore formation, the pressure is fixed to this point and porosity formation starts. Every-
where beneath this point total pressure is higher due to the hydrostatic pressure and no porosities
can form. Shrinkage will continue to pull the liquid out of the area while the emptied shrinkage po-
rosity area remains fixed to the critical pressure. This is why the porosities grow downwards.
Fig. 0.17 shows the development of the main shrinkage porosity area in two representations, once in
a grey scale and once in a stochastic porosity visualisation.
a) c)
porosity area is smaller than expected. For the horizontal size this can be explained by the low grid
resolution, but this explanation does not hold for the difference in vertical direction. It may be a
consequence of the primitive pore formation models. The shape of the porosity area indeed com-
pares well. This indicates that the model describes the formation mechanism of the porosity area
correctly.
bined.
The distribution of shrinkage porosities is similar to the case of shrinkage porosity formation only.
The amount is much smaller (maximum pore shrink = 0.10 instead of pore shrink = 0.24). The total pore
3
The top part of the casting did not solidify completely due to the adiabatic boundary conditions.
XXX Chapter 0 Executive summary
fraction of fraction of
porosities solid
[] []
a) b) c)
Fig. 0.19: Simultaneous prediction of shrinkage and hydrogen porosities for the case of zero vis-
cosity; final distribution of a) shrinkage porosities b) hydrogen porosities c) total amount of porosi-
ties; volume fraction of porosities with grey scale (left); fraction of solid and stochastic porosity
visualisation (left); the area of black dots is proportional to the area of porosities; proportionality
factor: 3.
area is 6.9 mm instead of 22,8 mm . This is due to the fact that the hydrogen pores are additional
2 2
0.6.3.4 Conclusions
Many publications have been written on the question if it is possible to distinguish hydrogen poro-
sities from shrinkage porosities or not. The results of the coupled calculation give room for specu-
0.6 Applications: Simulations and experiments XXXI
lation. They predict a situation where hydrogen porosities form which may look like shrinkage po-
rosities because they form inside an area with already high solid fraction.
The new model presented in this work is based on the postulate, that two completely different for-
mation mechanisms are responsible for the formation of hydrogen and shrinkage porosities. There-
fore, even if in many practical cases it is not possible to distinguish between them, it is possible to
separately predict them by numerical simulation. This is how numerical simulation could help to
identify the porosity type in real castings.
Table 0.1: Additional parameters and physical constants used for all welding calculations in this
chapter
Av constant for evaporation model (iron, evaporation 2.52
model)
Hv specific evaporation heat (evaporation model) 6259.5 kJ kg -1
b emissivity 0.2
b Stefan-Boltzmann constant 5.6710 W m K )
-8 -2 -4
In addition a number of preliminary calculations have been done to investigate the influence of nu-
merical parameters like grid resolution, the maximum number of SIMPLER iterations or the use of
adaptive time step control. The results of these calculations have been used to choose the following
basic set of numerical parameters which is a compromise between numerical accuracy and passable
calculation times: 2D-axisymmetric grid of 60x40 cells, 10 SIMPLER iterations per time step, use
of adaptive time step control. If not otherwise mentioned the calculations were performed using the
welding parameters P = 2000 W, rlaser = 4 mm and welding duration 1.5 s.
Fig. 0.21 shows the temperature scale which is valid for all temperature plots of this section. As the
different initial alloy compositions have different liquidus and solidus temperatures which deter-
mine the shape of the mushy zone, these values are given in the caption of the corresponding figure.
sponding liquidus and solidus temperatures which have been read from the equilibrium Fe-C dia-
gram. Fig. 0.22 a) c) shows the temperature distribution after 1.5 s for carbon concentrations
0.1wt % and 0.8 wt % for each of the three MC . 4
The figure shows that the influence of composition is rather limited. Observable effects with in-
creasing carbon concentration are:
The weld pools become slightly larger due to the decreased liquidus temperature.
The solidification time after switching off the laser at t = 1.5 s becomes longer because the ma-
terial needs to cool down more to reach the lower solidus temperatures.
The mushy zone becomes larger. This is related to the variation of the solid/liquid interval.
The results of the composition study lead to the conclusion that the choice of the weld material for
the studies is not critical and can follow practical considerations. The first calculations were made
with Fe-0.1wt%C. Later during the investigation it turned out that for Fe-0.2wt%C the numerical
stability is higher and concentration losses are smaller.
0.23 shows the development of the pool shapes. The rounded pool of the calculation without
4
For negative MC the calculation with 0.8 wt % C crashed after 1.2 s due to instabilities of the old solidification algo-
rithm. For the same reason there are no solidification times available for calculations with positive MC and concentra-
tions higher than 0.4 wt % C.
XXXIV Chapter 0 Executive summary
a) Fe-0.1wt%C
Fe-0.8wt%C
b) Fe-0.1wt%C Fe-0.8wt%C
c) Fe-0.1wt%C Fe-0.7wt%C
Fig. 0.22: Influence of weld metal composition on the pool shape during laser welding. Temperature
distributions after 1.5 s for different constant MC a) +10 N m K , b) 0, c) a) 10 N m K ; grey
-4 -1 -1 -4 -1 -1
area: Solid/liquid interval of initial composition, determined from the binary Fe-C equilibrium phase
diagram. Isotherms inside and outside the weld pool have the same values as in Fig. 0.21. d)
Solid/liquid intervals of the used alloys.
0.6 Applications: Simulations and experiments XXXV
Marangoni convection becomes a deep V-shape for positive MC and a wide flat shape for negative
MC. The diagrams in Fig. 0.24 show interesting details: Already small positive values of the MC
have a large influence on the pool depth. For values > 0.1 N m K the increase slows down con-
-1 -1
siderably. For this reason 0.1 N m K was chosen to be the MC of the basic set of welding pa-
-1 -1
rameters.
radii of the laser source were made. The radius values are rlaser = 0.5, 1, 2, 3, 4, 5, 6 and 7 mm. The
results require some interpretation. For small radii the pools get smaller. This is not consistent with
MC = +410 -4
MC = +210 -4
MC = +510 -5
MC = +110 -5
MC = 0 MC = 110 -5
MC = 510 -5
MC = 210 -4
MC = 410 -4
Fig. 0.23: Influence of different values of constant Marangoni coefficients on the shape of the weld
pool for welding of Fe-0.1wt%C with P = 2000 W; temperature distribution after 1.5 s; scale cf.
Fig. 0.21; Tliq = 1801 K, Tsol = 1766 K.
XXXVI Chapter 0 Executive summary
a) b)
Fig. 0.24: Influence of different constant Maran-
goni coefficients on the shape of the weld pool:
a) maximum horizontal (umax) and vertical (vmax)
velocity; b) pool radius and central depth; c)
weld pool volume, all after 1.5 s welding time.
c)
marked with (V) are not realistic, because for them the formation of a vapour capillary must be ex-
pected.
0.6 Applications: Simulations and experiments XXXVII
marked with (V) are not realistic, because for them the formation of a vapour capillary is expected.
a) b)
c) d)
Fig. 0.28: Influence of laser radius and surface vaporization model on the pool shape for welding of
Fe-0.1wt%C with P = 2000 W; a)c) pool radius and central depth calculated with and without sur-
face temperature limit for MC = a) +110 N m K , b) 0 c), 110 N m K ; d) weld pool vol-
-4 -1 -1 -4 -1 -1
experimental observations. An explanation for this numerical effect can be found in the type of the
surface vaporization model used and in the absence of a boiling model. Two different surface
vaporization models, one with limited surface temperature, one with unlimited temperature, have
been tried in order to understand the effect and to find the best description. Fig. 0.25 0.27 show
the variation of the pool shapes with the laser radius for different MC. The calculations were made
using the surface vaporization model with limited surface temperature. The cases of the missing
vapour capillary are shown, but marked with (V). The diagrams in Fig. 0.28 show clearly the
point where the two surface vaporization models start to make a difference. For rlaser > 3 mm the
results are almost identical, for smaller values the pool shapes start to differ.
and different laser powers P = 500, 1000, 1500, 2000, 2500 and 3000 W were made. Fig. 0.29
0.31 show the development for the different MC. The power of 500 W is not sufficient to reach the
melting temperature during the welding duration of 1.5 s. For higher powers the molten volume
increases almost linearly with the input power, Fig. 0.32 b).
varying viscosity has been observed in the resulting macrosegregation patterns. It will be described
later.
K, Tsol = 1761 K.
0.6 Applications: Simulations and experiments XXXIX
temperature distribu-
tion after 1.5 s welding
P = 3000 W time; scale cf. Fig.
0.21; Tliq = 1793 K,
Tsol = 1761 K.
a) b)
Fig. 0.32: Influence of welding power on the pool shape for welding of Fe-0.2wt%C; a) pool radius
and central depth for MC = 110 N m K and MC = 0; b) weld pool volume, all after 1.5 s
-4 -1 -1
welding time.
XL Chapter 0 Executive summary
kg m s ; velocity vectors and extension of mushy zone, i.e. area with fraction solid values between
-1 -1
0.6.4.3 Pool shapes with temperature dependent Marangoni coefficients (laser welding)
The second study considers the MC to depend on temperature and (constant) contents of surface
active elements like sulphur or oxygen. The pool shapes shown in this section, especially V- and W-
shapes, have been experimentally observed and are well known in literature, e.g. /Heiple 85/,
/Winkler 98/, and /Pavlyk 01/, but in general the formation of the different shapes are explained and
interpreted as a function of the concentration of surface active elements. The study describes them
as a result of different welding powers and welding durations if the concentration of surface active
elements is kept constant. Special attention is paid to the description of transient effects which in
general are neglected by stationary calculations and which can significantly influence the final pool
shape. These effects should be considered during the interpretation of experimentally observed weld
pools.
The study does not want to give quantitative recipes for specific materials or configurations. The
aim is rather to give an overview of the different types of weld pool dynamics that can occur in la-
ser welding systems and to provide a deep understanding of the evolution of weld pools under dif-
ferent welding conditions, how certain pool shapes form, and which effects influence the time de-
pendent and final pool shape.
Table 0.2 shows the parameters which have been used to calculate the temperature dependent MC.
They were determined by /Choo 92b/ for the industrial steel AISI 304.
0.6 Applications: Simulations and experiments XLI
Table 0.2: Parameters used to calculate the temperature dependent MC using eq. 0.14; data for in-
dustrial steel AISI304 by /Choo 92b/
s surface excess of the solute species (S) at saturation 1.3 10 mol m
-8 -2
O R3l /
s
= 0.53 , R R3 =
/
s
0.74
XLII Chapter 0 Executive summary
Fig. 0.34 shows the temperature scale which is valid for all temperature plots of this section. The
grey area with temperatures above the critical temperature of the MC, 2287 K, represents the area
with negative MC, the grey area with temperatures between 1720 K and 1775 K is the mushy zone
for the initial composition Fe-0.42wt%C.
a) b)
c) d)
Fig. 0.37: Laser welding of Fe-0.42wt%C with temperature dependent MC; time evolution of char-
acteristic numbers for different welding powers: a) normalised central depth Dc / R3s; b) normalised
lateral depth Dl / R3s; c) normalised offset of CTP Ol / R3s d) normalised radius R / R3s
a)
b)
Fig. 0.38: Laser welding of Fe-0.42wt%C with temperature dependent MC; characteristic numbers
of weld pools after a) 1 s and b) 3 s as a function of welding power
a)
b)
Fig. 0.40: Laser welding of Fe-0.42wt%C with temperature dependent MC; laser is switched off; a)
time evolution of normalised central depth Dc / R3s; b) principle mechanism explaining the increase
of the central pool depth Dc after switching off the HS. The quick surface cooling makes the surface
temperature drop under the CT and changes the Marangoni flow distribution from the case that has
created the W-shape to the case that would create a V-shape pool.
a) b)
Fig. 0.41: Laser welding of Fe-0.42wt%C with temperature dependent MC and P = 2000 W; a) De-
velopment of the pool shape between the time when the HS is switched off and the time of the
maximum central pool depth for welding duration 2 s; black: additionally molten parts of the pool;
grey: parts resolidified in the same time; b) deviation in % of central pool depths which would be
measured from micrographs and real depths at the moment when the HS is switched off.
observed that first the pool radius grows much quicker than the depth, forming a flat shape. The
lateral deepening starts developing only when the CTP reaches an equilibrium position after 0.6 s.
to develop W-shape behaviour and later turns into a V-shape. Fig. 0.43 shows two calculations with
powers within the critical range. Both start developing a W-shape configuration but end up with
different pool shape types.
a) b)
Fig. 0.44: GTA welding of Fe-0.42wt%C with Marangoni and EM forces; time evolution of all
characteristic numbers for a) P = 1550 W; b) P = 2000 W
L Chapter 0 Executive summary
P = 2000 W as two examples for extreme transient behaviour. The representation of the four char-
acteristic numbers in one plot demonstrates the strongly changing pool shape. Especially noticeable
points are
the sudden overturning of the W-shape after about 1.8 s for P = 1550 W, signalled by the sud-
den reduction of Ol / R1.4s and the steep increase and joining of the two depth numbers Dc / R1.4s
and Dl / R1.4s
the instability observed after 3.5 s for P = 2000 W.
Both effects as well as the complex interaction of different convection cells can be studied in the
temperature and stream function plots in Fig. 0.45.
The observations made in this study can be summarized as follows: Due to the fact that the EM
forces are body forces and Marangoni forces only act on the surface, the influence of EM forces
rises with rising pool volume. For low power calculations the early stages all seem to be governed
by Marangoni convection and later EM forces take over and form a V-shape pool. For higher pow-
ers the predominance of Marangoni convection lasts longer and longer. For high powers it must still
be found out, if the EM forces can gain predominance, or if in these cases the Marangoni forces
win.
5
Even if the length scales considered are very small, the term macrosegregation is correct because it describes solute
redistribution on the scale of the whole weld pool.
0.6 Applications: Simulations and experiments LI
The reliability of this kind of solute redistribution predictions depends a lot on the quality of the
mushy zone models. In future the description of the mushy zone can be improved by studying pol-
ished cut images of resolidified weld pools. Fig. 0.47 shows the bottom part of the former molten
area of a weld in Al-7wt%Si-0.3wt%Mg. Three areas can be distinguished:
1. the unmodified base material with coarse dendritic structure and coarse eutectic (bottom part);
2. the area which has been completely molten and is quickly resolidified; this has led to the forma-
tion of a fine dendritic structure and very fine interdendritic eutectic (top part);
LII Chapter 0 Executive summary
carbon
concentration
deviation [%]
a)
b)
c)
d)
Fig. 0.46: Comparison of final macrosegregation patterns in laser weld pools of Fe-0.42wt%C with
temperature dependent MC; with zero viscosity (left) and constant realistic viscosities (right) for MC
= a) +110-4 N m-1K-1, b) 0, c) 110-4 N m-1K-1 d) temperature dependent MC with H0 = 1.88108
J/mol; deviation of carbon concentration from initial concentration.
0.7 Conclusions and outlook LIII
3. the area which has obviously only been partially molten; here the coarse dendritic structure of
the base material has been conserved, but the interdendritic eutectic seems to have been molten
and to have rapidly resolidified as very fine eutectic structure (middle part); this area is consid-
ered to represent the mushy zone at the time when the heat source has been switched off. The
extension of this area is about 200 250 m.
The mushy zone looks very dense and blocked by the coarse dendritic structure. high permeability
and thus strong fluid flow can only be expected in parts which are close to the fully liquid area. Ob-
viously the shape of the dendritic network varies with the alloy composition. Alloys with lower
fractions of primary phase might give a structure which is more open.
Porosity model for the simultaneous but separate prediction of hydrogen and shrinkage porosi-
ties, as well as internal shrinkage cavities.
To the authors knowledge the model implemented in (SoliCon) is the first one which couples a
mobile solid model with a solidifying free surface in order to simulate the formation of deep shrink-
age cavities in a full NavierStokes formulation. The model has been tested and then applied to a
large number of scientific problems. The evaluation of all results leads to the following final con-
clusions:
The new model (SoliCon), which has been designed for the simulation of casting processes is
able to qualitatively model an immense number of complex phenomena related to convection
effects. The interaction between the submodels works well. As some of them are relatively sim-
ple, quantitatively correct results cannot be expected so far. In the present state the model can be
used to study principal mechanisms and interactions. Due to the modular structure and the well
defined interfaces the simple submodels can easily be replaced in the future without jeopardis-
ing the entire model.
The welding model (EL2D), which is a compilation of state of the art algorithms, completed by
a stable solidification algorithm, has shown to be a reliable tool for carrying out systematic in-
vestigations of welding processes. It can be used as a means to designing welding experiments.
As mentioned before, some of the models implemented in the mathematical and numerical frame-
work are rather preliminary. The scope of this work was the creation of a modular framework of
basic models which co-operate via well defined interfaces. The next steps are
Refinement of the existing models by comparison to experiments,
Replacement of simple models by more sophisticated, physical ones,
Addition of important new submodels and features which so far have been neglected,
Rigorous testing, benchmarking and experimental validation of the models and complex inter-
actions.
The potential of the new model is almost infinite. And so is the amount of work remaining to do...
1 Introduction
In casting and welding processes many of the properties of the finally solidified ingot or the re-
solidified weld are governed or at least influenced by convection, i.e. fluid flow phenomena which
have an impact via the transport of mass, heat or alloying elements. In general the term convection
is defined as:
A transfer of heat or mass that occurs when a fluid flows over a solid body or inside a channel
while temperatures or concentrations of the fluid and the boundary are different; transfer occurs
within the fluid as a consequence of the motion within the fluid relative to the flow boundary
/Academic 96/
In this work the term convection will be used to describe all kinds of fluid flow which occur in
molten alloy systems. The different types of convection can be classified by their governing forces.
The types of convection considered here are:
Thermosolutal convection (buoyancy flow due to combined temperature and concentration de-
pendent density gradients)
Shrinkage flow (due to air pressure and initiated by temperature dependent density and volume
losses during phase change)
Gravity induced surface movement (surface waves and drop of the plane liquid surface)
Marangoni convection (surface flow due to surface tension gradients, very important for weld-
ing and crystal growth)
Electromagnetic convection (GTA welding)
Mould filling and other forced types of convection like flow due to stirring will not be considered.
Numerical simulation has proven to be a powerful tool which can be used to understand and govern
the influence of convection on solidification processes. Even though intensive research has been
performed for more than two decades, there are still many important tasks left. Especially complex
convection effects which require the coupling of many single effects make high demands on ma-
thematical and numerical models.
Fig. 1.1 shows two examples of industrial processes where complex convection effects play a major
role and determine the final properties of the workpiece.
The casting process starts with mould filling which causes strong forced convection of the liquid
cast metal in the mould. The metal which has been poured first starts to cool down before the end of
the filling. Even if the mould filling process can in some cases have a big influence on the initial
temperature distribution of melt and mould and thus on the course of solidification, it will not be
considered here.
2 Chapter 1 Introduction
a)
b)
Fig. 1.1: Main convection effects which occur during a) casting process of steel ingot with riser,
modelled in software SoliCon; b) laser spot welding process with vertical incidence of laser beam,
modelled in EL2D. For details of the two software packages, cf. App. 9.2.
When mould filling is finished the forced convection decays and is superposed by thermosolutal
convection. This flow keeps the temperature in the fluid domain more uniform than if the fluid was
stagnant. In addition to thermosolutal flow the density changes create a shrinkage flow towards the
cooler areas.
When solidification starts the situation becomes extremely complex. The density difference be-
tween liquid and solid phases intensifies the shrinkage flow, now directed towards the solidifying
areas. If solidification is dendritic there will be columnar growth of depleted dendrites from the
walls and formation of enriched interstitial liquid. Thermosolutal and especially shrinkage flow can
enter between the dendrite arms and change the local liquid concentration. Depleted globulitic
grains which form in the liquid without contact with the wall settle and leave enriched melt behind.
Both effects lead to the formation of macrosegregations. If the hydrogen content is high, the phase
change from liquid to solid causes the hydrogen which is dissolved in the liquid to be rejected and
to form bubbles which either rise to the surface or are built into the solid as hydrogen porosities.
The shell-like solidification and the shrinkage make the liquid level go down. The shape of the re-
maining solid shell forms the shrinkage cavity.
In later stages of solidification, when the permeability of the mushy zone becomes very small,
feeding may become impossible in some areas and shrinkage porosities form. The solute which has
been washed out of the mushy zone by thermosolutal flow has gathered at the last point to solidify
and forms a highly enriched area.
1.1 Subject of this work 3
During spot welding similar convection effects occur, but due to the difference of the processes the
importance of the different effects changes. Some effects almost disappear, others that are almost
negligible in casting become the governing effects.
Obviously there is no mould filling in spot welding, but the process starts with a completely solid
domain. The heat input of the laser or GTA heat source starts to remelt the material and to form a
liquid weld pool surrounded by a mushy zone (the extension and character of this mushy zone is
subject to ongoing discussion and research). The forces which influence the flow pattern in the
molten area are of different orders of magnitude. Thermosolutal convection does not play any major
role. The governing force in laser welding and in early stages of GTA welding is the surface tension
driven Marangoni convection. In GTA welding the body forces of the electromagnetic fields (Lor-
entz forces) become more important with growing pool volume.
On the one hand these strong flows with velocities up to 1 m/s carry large amounts of heat and de-
termine the areas to be molten and thus the development of the pool shape. On the other hand they
wash out the mushy zone and create positive macrosegregations in the middle of the pool. This ef-
fect is difficult to measure because of the strong solute losses due to evaporation and the smoothing
of concentration profiles caused by quick solid state diffusion (e.g. of carbon).
Porosity model for the simultaneous but separate prediction of hydrogen and shrinkage porosi-
ties, as well as internal shrinkage cavities.
Solidification algorithm to describe eutectic and peritectic solidification of binary alloys in sys-
tems with strong convective solute and mass transport.
washing out of the mushy zone, because the concentration difference between the interdendritic
liquid and the replacing liquid from the reservoir diminishes.
The local average mixture concentration is the concentration that determines the (C, T ) position
inside the phase diagram. From this it follows that the shift of the local average composition due to
macrosegregation causes a horizontal shift inside the phase diagram. Therefore in the case of hy-
poeutectic alloys (e.g. Al-7wt%Si) depletion leads to an increase of the local liquidus temperature
and thus isothermal solidification, whereas local enrichment lowers the liquidus temperature and
leads to isothermal remelting. As the formation of the macrosegregations is a continuous process,
there may be no remelting, but a slowing down of solidification, or, in the case of depletion an ac-
celerated solidification. The effect of decreasing liquidus temperature in case of enrichment is of
high importance, because it keeps areas with high concentrations longer in the liquid state. This
means that the position of the last point to solidify is not necessarily the position of highest tem-
perature, but in a solidification process with very flat and homogeneous temperature profiles it is
the point with the strongest posotive macrosegregations.
6
Of course this is only valid if the solid density is higher than the liquid density, which is not always the case.
2.1 Macrosegregations 9
segregation. If the new liquid had the composition C0 , the composition would remain constant,
depleted liquid would lead to the formation of negative segregation. Of course the dashed line is
only an auxiliary means to explain the mechanism. In reality the process is continuous and the
amount of macrosegregation will continuoulsy grow from the dendrite tips down to the bases.
a) b)
2.1 Macrosegregations 11
of dendritic or globular floating crystals to the front of columnar crystals, Fig. 2.7. Detailed discus-
sions can be found in /Hultgren 73/, /Flemings 76/ and /Olsson 86/.
2.1.2.1 Experiments
This section reviews pure experimental publications. Of course most of the analytical and numerical
models come along with experimental verifications. They will be presented in the following sec-
tions.
/Hultgren 73/, /Flemings 76/, and /Olsson 86/ investigate the casting of large steel ingots and de-
velop theories to explain typical macrosegregation effects like A- and V-segregates, banding, and
the formation of the negative segregation cone at the bottom of the casting.
/Kato 85/ compare inverse segregation in directionally solidified ingots of Al-Cu-Ti alloys with
theoretical predictions. They find better agreement in columnar grained ingots than in equiaxed
14 Chapter 2 Modelling of convection effects in casting and welding processes
grained ingots. Their explanation is incomplete feeding which is consistent with the observation of
an increased amount of shrinkage porosity in the equiaxed case.
/Doudoux 98/ investigate the strong macrosegregation effects for the case of squeeze-casting, when
the feeding flow is enhanced by high external pressures.
dendrite erosion and the formation of channel type A-segregates and verify the results by comparing
them to qualitative experimental observations. Some years later /Krane 95/ apply the model to ex-
amine the effects of solidification shrinkage on the solute redistribution in a Pb12.2%Sn alloy
which is convectively cooled at a side wall. They find that buoyancy flow is the governing effect
determining the final macrosegregation pattern. Shrinkage induced flow influences the macrosegre-
gation pattern close to the chilled wall if solidification rates are high, but the overall effect on ma-
grosegregation is relatively small. This work on scaling analysis of different macrosegregation phe-
nomena is continued in /Krane 96/.
Tsai and co-workers modify the continuum equations developed by Bennon and Incropera to in-
clude shrinkage induced flow, /Chiang 92a, b/. In /Chen 93/ and /Diao 93a,b/ they apply the model
and simulate inverse segregation of Al-Cu alloys cooled from the bottom. They find, that the redis-
tribution of solute in the solidifying alloy is caused by the flow of solute-rich liquid in the mushy
zone due to solidification shrinkage.The severity of inverse segregation can therefore be reduced by
increasing the heat extraction at the bottom and, as a result, decreasing the size of the mushy zone.
The formation of negative- and positive-segregated bands is explained as a consequence of fluctu-
ating heat-extraction rates during solidification. In /Diao 94/ they study the formation of negative
under-riser segregations in a rectangular cavity with reduced cross section from the casting to the
riser.
/Chang 96/ develop a continuum model considering thermosolutal convection and shrinkage driven
flow and investigate macrosegregation effects in Al-Cu alloys with three different casting geome-
tries. They find, that solidification contraction has a stronger influence in the mushy zone, than
thermosolutal convection. /Singh 96/ use a continuum formulation to study the role of double diffu-
sive convection resulting from the solutal rejection in the evolution of macrosegregation in an Fe-
1wt%C system. /Vannier 98/ couple a 2D code describing heat transfer, fluid flow and chemical
composition changes in multicomponent alloys with the Suzuki criterion function to predict A-
segregations in heavy steel ingots.
/Beckermann 88/ and /Ni 91/ develop a volume-averaged two-phase model for transport phenomena
occuring during solidification of multicomponent mixtures, which has since been widely used. The
macroscopic transport equations for each phase are derived from microscopic equations using the
technique of volumetric averaging. A good review of this model can be found in /Beckermann 93/.
The model is numerically realized and applied to the solidification of a Pb-20wt%Sn alloy in a
square cavity cooled from one side and fed by a rectangular riser, /Schneider 95a/. The authors in-
troduce a sophisticated anisotropic permeability model and study the influence of different perme-
ability functions on the formation of segregation channels. /Schneider 95b,c/, /Bhmer 97/ and
/Schneider 98/ apply the model to the solidification of small ingots of multicomponent steel. /Gu
99/ use the model to predict macrosegregation defects in a large industrial steel ingot. They even
consider the formation of a flat shrinkage cavity at the surface. The influence of the sedimentation
of equiaxed grains is neglected.
16 Chapter 2 Modelling of convection effects in casting and welding processes
This shortcome is tackled by Wang and Beckermann in a series of publications. They investigate
dual scale volume-averaging which integrates the modelling of micro- and macroscopic phenomena
in one continuum theory, /Wang 93/. This work results in a complete volume-averaged two-phase-
model to describe dendritic alloy solidification including columnar-to-equiaxed transition and the
movement of floating equiaxed grains, /Wang 94a/, /Wang 94b/, /Wang 95a/, /Beckermann 95/.
/Beckermann 96b/ predict how sedimentation of globulitic grains influences the final macrosegre-
gation pattern and grain size distribution in Al-4wt%Cu.
A high grid resolution is crucial for good numerical macrosegregation predictions. /Kaempfer 00/
find a solution for the problem that grid resolution in equidistant grids must always be a compro-
mise between accuracy and tolerable calculation times. They introduce dynamic grid remapping
which allows to refine the grid in areas where the mushy zone currently is and to re-coarsen it
where solidification is finished.
interfacial drag in both globular and dendritic equiaxed solidification. They develop a drag correla-
tion which is valid over the full range of solid volume fractions.
Most permeability models assume a regular development of the mushy zone. In reality, strong
forces due to shrinkage induced flow and solid shrinkage can deform and tear the dendritic and
equiaxed network especially in early stages of solidification. The deformation can cause cracks
(hot-tears) which increase the permeability of the area. /Engler 73/ measure the tensile strength of
Al-Si alloys in different stages of solidification. /Shen 02/ measure the deformation of a columnar
dendritic mushy zone in a transparent succinonitrile-acetone (SCN-ACE) alloy. /Martin 98/ use a
Sn-Pb model system to investigate the mechanical properties of a high-solid-fraction mushy zone
under compression and tension conditions. /Dahle 99/ model the stress development in a coherent
dendritic network due to shrinkage-induced interdendritic fluid flow. They compare the calculated
stresses to the measured shear strength of equiaxed mushy zones and find that reorientation or col-
lapse of the dendritic network is possible. /Farup 00/ and /Mo 00/ develop a two-phase model of an
isotropic mushy zone to calculate strength parameters and stresses due to shrinkage-induced flow
and solid shrinkage. /Braccini 00/ examine the mushy zone rheology in Al-Cu alloys to formulate a
hot-tearing criterion. Not only mechanical forces can change the mushy zone morphology. /Diepers
98/ use a phase field method to model the convection-induced coarsening of a binary alloy mushy
zone.
Application of the macrosegregation model to spot welding processes. Due to the high flow
velocities which occur during welding processes, the mushy zone is washed out and positive
segregations should form in the middle of the weld pool, section 7.2.2.
Density changes due to phase change. In most alloys the solid phase is more dense than the liq-
uid phase.
Density changes lead to local volume losses which are balanced by shrinkage flows. In contrast to
thermosolutal flow which is driven by gravity only, the driving forces for shrinkage flow are very
high, because it is driven by air pressure.
The most important phenomena which lead to the formation of shrinkage defects are:
Movement of the solidifying free surface (external shrinkage cavities)
interruption of feeding flow (internal shrinkage cavities, shrinkage porosities)
precipitation of gas bubbles (hydrogen porosities)
different shrinkage velocities of the instable solid network in the mushy zone (hot-tearing)
obstruction of solid shrinkage by the mould (hot-tearing).
One of the basic assumptions of this work is, that the density of all solid phases is kept constant.
This is due to the enormous additional difficulties which arise from modelling the movement of the
solid network, especially in the framework of a control volume method with regular grid. From this
follows that a number of technically important shrinkage defects will not be considered:
Linear contraction / cubic shrinkage
Hot tearing due to internal stresses
Hot tearing due to obstruction by the mould
All shrinkage defects considereded here are caused by convection phenomena. In the following the
most important mechanisms will be described.
the surface line of the shrinkage cavity always indicates the height of the liquid level at the moment
of solidification.
The importance of external shrinkage cavities goes beyond the question of surface shapes. Due to
the mechanism of their formation, they are very often associated to the last point to solidify in a
casting or riser and therefore linked to areas of strong macrosegregations. By control of the forma-
tion mechanisms of external shrinkage cavities it is possible to control the position and type of mac-
rosegregations.
sure is low, the hydrogen pressure that is generated by the phase change, finds less resistance, and
therefore more and bigger pores form. This fact is used for some methods of hydrogen detection
and hydrogen contents measurement (density samples, Straube-Pfeiffer test which is an industrial
standard).
Unfortunately the numerical modelling of linear contraction is a very tricky task, especially in the
framework of a fixed grid technique. For this reason the deformation of stationary solid has not
been considered in the present work. The interpretation of some of the results must take this fact
into account.
umes in their liquid state, using a hierarchic set of rules. For pure aluminium they manage to simu-
late the typical deep cavities. They cite some related work, but most of them have appeared in con-
ference proceedings and are not easily available. The ones which were available only treated inter-
nal shrinkage cavities. The FDM / CVM casting simulation software MAGMAsoft /MAGMAsoft
02/, /Schneider 03/ uses a sophisticated but unfortunately unpublished feeding algorithm to model
the formation of external shrinkage cavities. It evaluates information on temperature distribution,
mass balance, density changes and several feeding parameters to predict the formation of external
and internal shrinkage cavities. The algorithm decides where mass has to be taken away to replace
shrinkage losses in other areas. It does not use fluid flow calculations. The module MAGMAiron
uses the same feeding algorithm, but in addition takes into account the fact that in cast iron the for-
mation of graphite precipitations can lead to expansion. The FLOW-3D web page, /FLOW-3D
02/, shows animations of a shrinkage cavity formation calculated with a rapid solidification shrink-
age model, based on heat transfer only, /Barkhudarov 95a,b/, / Barkhudarov 97/, /Beech 98/, Fig.
2.10. In this feeding algorithm no fluid flow is considered. The level of the free liquid surface,
which is assumed to be flat, is readapted in each time step to match the volume losses due to shrink-
age. The shape of the shrinkage cavity is generated because the area of the liquid surface which is
adapted, changes between the time steps due to solidification.
The number of full Navier-Stokes models is quite restricted, to be honest, the only published model
known to the author is /Ehlen 00b,c, 02b, 03a/. A reason might be that working models in commer-
cial software are a valuable secret. The FLOW-3D web page, /FLOW-3D 02/, announces a second
solidification shrinkage model based on first principles using the full system of Navier-Stokes
equations, but the web page does not show any results. The MAGMAsoft module MAGMAsteel,
/MAGMAsoft 02/, uses a Navier-Stokes approach, which describes convection effects and mushy
zone flow and includes the surface shape result of the feeding algorithm as a boundary condition,
but the feeding algorithm does not use information of the fluid flow. Work seems to be in progress
to do the complete coupling in a full NavierStokes model. /Gu 99/ simulate melt convection in
casting of a large steel ingot, Fig. 2.11. The model presented in /Schneider 95b/ is coupled with a
feeding algorithm which sums all shrinkage losses. Each time the sum exceeds the volume of a grid
cell that contains steel, has a solid fraction less than 0.05 and is closest to the top and centre line,
that cell is emptied.
Moving free surfaces are in this work defined as surfaces between alloy and air where the alloy is
liquid or has been liquid at any time, and where the surface shape is determined by the free move-
ment due to physical forces. Only a few publications deal with solidification of moving free sur-
faces. There may also be commercial reasons behind this fact. An excellent algorithm is used by
Mostaghimi, Pasandideh-Fard and co-workers, /Bussmann 99/, /Pasandideh-Fard 00/. They simu-
late impact and solidification of a tin droplet on a steel plate /Pasandideh-Fard 98/ and compare the
results with experiments finding excellent agreement. Other publications model the sequential
2.2 Formation of shrinkage cavities and porosities 25
impact and solidification of two molten droplets on a solid surface, /Pasandideh-Fard 99a/, or the
surface cooling by an impinging water drop, /Pasandideh-Fard 99b/. Considering the highly dy-
namic processes modelled, the Volume of Fluid (VOF) based algorithm seems to be uncondition-
ally stable. Unfortunately the authors do not give detailed information about the algorithm used.
There is not enough information available either to decide, which model is used by FLOW-3D,
/FLOW-3D 02/, to model the solidifying free surface in their full Navier-Stokes model. /Kuznetsov
97, 98, 01/ simulates the formation of macrosegregations in a strip casting process and considers a
free surface flow with solidification. As in this process mainly the steady state behaviour is impor-
tant, the surface configuration is assumed to be to be fixed. The papers do not give much informa-
tion about how the steady state free surface shape is found.
2.2.2.2.1 Experiments
Many publications try to correlate porosity formation to alloy composition and casting parameters
like riser and casting geometry, temperature gradients, solidification times, pouring temperatures
etc. Examples are /Michels 89/, who examine the influence of composition, modification, grain re-
finement and casting geometry on feeding kinetics for pure Al and different Al-Si alloys, /Kao 95/,
who study the porosity formation in long-freezing-range A206 alloy plate and taper casting, and
who suggest a relationship of volume ratio and freezing ratio for the production of sound castings,
and /Beech 97/, who investigate the formation of macroscopic internal shrinkage cavities in T-shape
castings of pure Al and Al-Cu alloys.
An early investigation and overview of the mechanisms of hydrogen porosity formation including
diffusion, solubility, nucleation is given by /Ransley 55/. The authors measure the temperature de-
pendent diffusivities of hydrogen in solid aluminium, copper and nickel, estimate the diffusivity of
hydrogen in liquid aluminium, measure the solubility of hydrogen in the melt and discuss pore for-
mation mechanisms. X.-G. Chen and S. Engler have led systematic experimental investigations on
the correlation between hydrogen content and porosity in Al-Si alloys, /Chen 91a/, and AlMg alloys
/Chen 91b/ under consideration of solidification conditions and melt treatment. In /Chen 94/ they
investigate the mechanisms of bubble formation and pore morphology in solidifying Al castings.
/Chen 96/ consider the influence of melt cleanliness on pore formation in Al-Si alloys. They find,
that, at the same hydrogen level and with increased inclusion concentration, the density of reduced
pressure samples decreases, and the amount of porosities and the number of pores increase. /Lee 97/
perform in situ observation of hydrogen porosity formation in directional solidified Al-Cu alloys.
They use a temperature gradient stage and real time micro-focus radiography and characterise pore
morphology both in the final structure and as a function of temperature during solidification, pro-
viding a qualitative insight into the relative importance of the competing physical processes.
2.2 Formation of shrinkage cavities and porosities 27
/Anson 99/ use image analysis data to examine micropororosities in Al-7wt%Si castings and find
that there is a simple means to distinguish between shrinkage and gas porosity. On a metallographic
section shrinkage pores are found in groups, while gas pores are found as isolated entities. Applying
their method they find that the two predominant types of porosity are gas porosity and combined
gas-shrinkage porosity. Pure shrinkage porosity occurs only at very low gas levels.
percentage of porosity, but not the distribution of pore sizes. A well known example for this ap-
proach is /Kubo 85/, who calculate the distribution, amount and size of porosity formed in Al-
4.5wt%Cu plate castings, and suggest, that the formation of shrinkage and gas porosities happens in
the stage of interdendritic feeding. Other models of this type are /Zhu 90/, /Ampuero 91/, and /Suri
93/. /Rousset 95/ examine the influence of porosity formation on the amount of inverse segregation
in directionally solidified aluminium alloys. /Bounds 98/ distinguish between internal and surface
connected porosities which occur when the local pressure drop in the liquid is sufficient to draw air
into the casting through the permeable mould. /Pequet 00/ use dynamical grid refinement in the
mushy zone to allow an accurate calculation of gas and shrinkage porosity. /Wendt 00/ considers
Darcy-law and partial gas pressures to formulate a criterion where porosity can be expected in steel
castings.
Gas diffusion models suggest, that pore formation is governed by hydrogen-diffusion-controlled
growth. The deterministic approach by /Atwood 00a/ predicts the growth of a representative pore
caused by the diffusion of hydrogen from its source (the partitioning at the growing solid phase) to
its sink, the pore. The model assumes that the pores nucleate all at the same time, distributed in a
regular periodic array. The model does neither account for the stochastic nature of pore nucleation,
nor for the shrinkage driven growth of pores.
Continuum-stochastic models couple a continuum model for the calculation of general flow proper-
ties and diffusion processes with stochastic nucleation and growth models. A 2D continuum-
stochastic model is /Huang 98a,b/, who predict grain size, pore size, pore morphology, and location.
The model couples hydrogen gas evolution and microshrinkage pore formation mechanisms with a
grain growth simulation model. Nucleation and grain growth are modelled with a probabilistic
method that uses the information from a macroscale heat transfer simulation. The model can even
predict the influence of Sr modifier on pore size and morphology for A356 aluminium alloy. /Lee
01b/ model the influence of hydrogen diffusion on the formation of porosity during the solidifica-
tion of aluminium alloys by combining a continuum model of the diffusion equations with a sto-
chastic nucleation model. They simulate the growth of each individual pore including its interaction
with the developing dendritic structure. /Atwood 00b/ couple a cellular automaton (CA) to a finite
difference (FD) solution to simulate the diffusion of both silicon and hydrogen in an aluminium
alloy (CA-FD model). The results show good qualitative but poor quantitative correlation to ex-
perimental observations of the nucleation and growth of porosity in an Al-7wt%Si alloy.
Other authors add the category of thermodynamic models, which assume, that microporosity can
form whenever it is thermodynamically possible. /Sigworth 97/ propose a thermodynamic model
which includes the effect of surface tension of the bubbles, but adds the requirement, that the pore
must fit into the space available between individual solid grains or between the solid arms of den-
drites. /Poirier 87b/ present a thermodynamic model and use it to predict the formation of interden-
dritic porosity in aluminium-rich Al-Cu castings. They assume, that in these alloys interdendritic
porosity arises, because due to solidification hydrogen is expelled into the interdendritic liquid.
2.2 Formation of shrinkage cavities and porosities 29
The micromodel by /Sasikumar 01/ describes the growth of pores in a dendritic network. They pre-
dict the pore shape under different solidification conditions and postulate that pure shrinkage cavi-
ties exist only in very special situations. They state that most observed porosities, even those which,
due to their shape, are named shrinkage porosities are in fact gas porosities.
Many publications cannot be classified easily, because they couple several of the models. /Katzarov
96/ couple a darcy-law model with a solid contraction model. They describe the evolution of pres-
sure and the formation of porosity defects in hot spots by the simultaneous treatment of thermoelas-
tic, crystallisation and porosity growth in axially symmetric castings. /Kuznetsov 95/ develop a
three phase model (liquid, solid and gas) of the mushy zone for the investigation of gas porosity
formation in solidifying castings. The third phase is treated as an empty volume fraction which
yields an additional term in the continuity equation. The important new feature is, that in this ap-
proach the formation of hydrogen porosities influences the local pressure drop. In /Kuznetsov 96/
they compare numerical results of this three phase model with results of a two phase model by
/Poirier 87b/ which assumes that porosity formation does not influence the local pressure and flow,
and an analytical solution based on their three phase model. They find significant differences in the
pressure fields and porosity distributions.
and is based on the assumption that the formation mechanisms for hydrogen and shrinkage po-
rosities are completely different but can both be joined in an integrated model. The model can
be used as a tool to decide if porosities observed in real castings are dues to shrinkage or hydro-
gen.
The model has been applied to a test geometry which was designed to provoke the formation of
both shrinkage and hydrogen porosities. A first experiment for validation has been performed.
Fig. 2.12: Formation of grain structure in chill cast ingot, /Ohno 87/. a) Equiaxed crystals form at
the wall, separate and are transported into the middle of the ingot by convection. b) Crystals close
to the wall form a fine equiaxed structure and some become columnar grains when convection de-
cays (ECT). Crystals in the middle precipitate and c) form the equiaxed zone in front of the colum-
nar zone (CET).
for dendritic grains with fourfold symmetry, while globulitic describes spherical grains. In practice
this distinction is not really made. On the other hand the term dendritic grain structure is some-
times used to describe columnar and equiaxed structures. In micrographs columnar growth is char-
acterised by arrays of long, parallel grains, while equiaxed areas show fine, compact grains.
There are many theories dealing with the formation of equiaxed grains, including nucleation, sepa-
ration, growth, and their contribution to the formation of macrosegregations due to sedimentation.
Areas of special interest are the transition from columnar to equiaxed growth (Columnar-to-
Equiaxed Transition, CET ) and, less frequent, the transition from equiaxed to columnar growth
(Equiaxed-to-Columnar Transition, ECT) close to the wall in chill casting.
Fig. 2.14 c) shows a typical grain structure found in a cylindrical ingot of an aluminium-based alloy
cast in a steel mould. Three regions can be distinguished: Fine equiaxed crystals close to the chill
wall which have formed in early stages of solidification. At a certain distance from the chill this fine
grained structure is replaced by large columnar grains (ECT). At the bottom there is a sedimentation
cone of fine equiaxed crystals resulting from a crystal precipitation mechanism (German: Kristall-
schauer). Fig. 2.12 shows the principal formation history of this kind of grain structure as de-
scribed by /Ohno 87/. Already during the filling process equiaxed grains form close to the wall.
Due to strong convection and remelting effects they separate and some of them are transported into
the bulk melt in the middle of the ingot, Fig. 2.12 a). Some crystals are captured by the mould walls
and form a fine equiaxed structure. When convection decays a stable solid shell forms and columnar
32 Chapter 2 Modelling of convection effects in casting and welding processes
crystals begin to grow, Fig. 2.12 b). The crystals floating in the bulk melt grow and precipitate,
stopping the growth of the columnar zone and forming the equiaxed zone (CET). If the rules are
known, where in the cast part these transitions occur, the casting conditions might be adapted to
yield completely equiaxed or completely columnar structures.
2.3.1.2 Two-phase-flow
If floating equiaxed grains form in the melt, they are subject to different forces. Gravity tries to
drive them towards the bottom of the mould, and the surrounding fluid interacts with them. The
interaction or drag force increases with
increasing viscosity of the melt
increasing relative velocity between melt and grain
decreasing grain radius
Combining Stokes law with Newtons law the acceleration av of a grain of radius R, density hav-
ing a relative velocity vv compared to the surrounding fluid with dynamic viscosity can be calcu-
lated as
v
9 v
av =
2 R2
This has the following consequences: For very small particles the drag force quickly drives the
relative velocity to zero. Small grains move with the velocity of the fluid. In this case grains and
fluid form a mixture with averaged density. Fluid with a high number of small grains is heavier than
the pure fluid and is subject to gravity forces. This leads to sedimentation effects even if the grains
do not have a separate velocity. For larger grains inertia forces and gravity lead to relative veloci-
ties between grain and fluid. The grains tend to settle and form depleted areas while the pure fluid
tends to rise and to form enriched areas.
a) b) c)
Fig. 2.13: Influence of solidification morphology on the formation of external shrinkage cavities;
position of solidification front for a) planar growth, b) columnar growth, c) columnar and equiaxed
growth. Floating equiaxed grains do not participate in the shaping of the free surface but follow the
dropping fluid level.
ahead of them equiaxed grains are floating in the melt? In this case the position of the surface is not
defined by the innermost solid forming in the melt, but by the innermost stationary solid, i.e. by the
dendrite tips again, Fig. 2.13 c). The floating solid grains do not participate in the shaping of the
free surface but move with the melt and follow the dropping fluid level. The floating equiaxed
grains only start influencing the surface shape when they reach the point of coherency.
Sand castings of unalloyed steel tend to form deep shrinkage cavities, even if the alloy has a large
liquid-solid-interval (up to 100 K) and thus a long solidification range. Sand cast cylinders of Al-
7wt%Si do not form a shrinkage cavity at all. In a very early stage of solidification a dendritic net-
work forms in the whole cast part and volume losses due to shrinkage become microporosities.
Grain size can be increased by melt cleaning with filters, by directional solidification, grain se-
lectors, slow solidification. The extreme case is the casting of single crystal turbine blades using
grain selectors and a Bridgeman furnace to yield very slow unidirectional solidification.
a) b) c)
Fig. 2.14: 2D-CAFE-(Cellular Automaton - Finite Element) Simulation of grain structure in a cast
Al-7wt%Si ingot, /Gandin 98/; influence of grain movement on the resulting grain structure;
simulation results a) without, b) with convection and convective transport of floating equiaxed
grains. c) Longitudinal section of an aluminium-base alloy cast in a steel mould (height 120 mm,
60 mm). A sedimentation cone of fine equiaxed crystals and a coarse columnar structure are clearly
seen at the bottom and top parts of the ingot, respectively. The external shrinkage cavity has not
been modelled.
consist of equiaxed dendritic crystals with interdendritic liquid between their arms, together swim-
ming in extradendritic liquid. The mechanical behaviour is described by two sets of momentum
equations for solid and extradendritic liquid. The increasing interaction forces between the grains,
from the free particle regime to the packed bed regime, are described by a unified approach com-
bining several analytical and empirical solutions, /Wang 95b/. The separate treatment of the inter-
dendritic liquid is used to model micro- and macrosegregation phenomena. For some of the sub-
models the dendrite envelope is replaced by the envelope volume equivalent sphere. In
/Beckermann 96a/ they compare simulation results of the solidification of a NH4Cl-70wt%H2O so-
lution with experimental observations. They find qualitative agreement of the flow patterns and the
formation of a sedimented bed of NH4Cl crystals at the bottom of the enclosure. /Beckermann 96b/
use a more simple model assuming spherical grains to model sedimentation effects in solidification
of an Al-4wt%Cu alloy. /Ludwig 00/ and /Ludwig 02/ consider the growing solid as a separate con-
tinuous phase. Complete sets of conservation equations for each phase allow the modelling of sedi-
mentation and floating of solid during solidification. The interaction between liquid and solid is
36 Chapter 2 Modelling of convection effects in casting and welding processes
modelled using the Blake-Kozeny equation for an isotropic permeability of the mushy zone, and
choosing the solid viscosity as the empirically known viscosity of a liquid / granular mixture. A
conservation equation for grain density allows to track grain size evolution. The continuum model
by /Ohnaka 00/ uses a Darcy law approach for high solid fractions and a single particle drag force
for low ones. It can be used for liquid / solid and gas / solid systems, i.e. for equiaxed solidification
or core shooting / vacuum sand moulding processes. The authors attach great importance to the cor-
rect numerical solution of their equations.
Some approaches try to avoid the additional problems arising from the solution of a second set of
momentum equations for the solid phase. They use volume averaging continuum models, but try to
implement the properties of the mobile solid phase by introducing the concept of an effective mix-
ture velocity into the liquid momentum equations. /Voller 89/ investigate the question, how two-
phase-models considering the motion of liquid and solid phases can be reduced to form one-phase
models without loosing too much of the information yielded by the second set of momentum equa-
tions. They assume equal velocity for solid and liquid, which is valid for a highly dispersed solid
phase and model further solidification by increasing the viscosity of the mixture. /Oldenburg 92/
present a hybrid model for continuum phase change systems. They consider three types of flow re-
gions: Concentrated mushy zone, dilute mushy zone and single-phase liquid. For high solid frac-
tions a darcy-law is used, for low solid fractions an enhanced viscosity model, and for single-phase
liquid pure liquid viscosity. arctangent switching functions assure a smooth transition between the
models. /Chang 96/ use a two stage concept to describe the different phases of the solidification
process. In the first stage, the liquid containing freely moving equiaxed grains is described through
the relative viscosity concept. In the second stage, when a fixed dendritic network has formed, the
mushy zone is treated as a porous medium. /Yang 00/ use the same approach, but add a grain
tracking model which calculates the transport of volumetric grain density and grain density distri-
bution function. /Flood 91/ introduce a consolidation factor specifying the relationship between
solid and liquid velocities. This factor varies linearly from 1 (solid velocity equals liquid velocity)
for zero solid fraction to 0 (no solid velocity) for a chosen solid fraction where all solid is assumed
to be stationary.
Grains nucleating at the mould wall do not necessary become columnar grains. The separation the-
ory by /Ohno 87/ which is based on experimental observations, predicts, that in early phases of a
casting process, when thermosolutal or forced convection is strong, grains which have nucleated at
the wall separate due to mechanical forces or thermosolutal remelting and add to the formation of
the sedimentation cone and the equiaxed zone close to the chill. Other grains form close to the sur-
face or nucleate at surface impurities and sink down.
To the authors knowledge there is no publication dealing with the movement of free surfaces con-
taining floating equiaxed grains. Most models considering a movement of the liquid level, e.g. /Gu
99/, use a solid fraction limit. In cells with solid fractions under the limit the level can change, in
cells with fractions above the limit, the surface is fixed. /Suri 93/ think about the critical solid frac-
tion above which liquid loses its fluidity, but the movement of the surface is then realized by sub-
tracting cell layers from the top of the pool. So far no publication seems to consider the movement
of a surface consisting of liquid and floating equiaxed grains.
The model can describe the formation of deep shrinkage cavities, because it allows the partially
solidified surface to drop, the sedimentation of solid (together with the liquid), and the transport of
heat, concentration and momentum by the solid particles.
So far the model cannot describe the separation of solid and liquid, e.g. the formation of macroseg-
regations due to the sedimentation of equiaxed grains, section 2.1.1.3. Work is in progress to find an
averaging formulation for this effect in the framework of the SSM.
A big restriction of the model is the fact that only equilibrium solidification without undercooling
effects can be considered so far.
2.4.1.1 Horizontal and vertical movements through the binary phase diagram
In many simulations of casting and welding processes macrosegregation phenomena are neglected.
The alloy composition is considered to be constant. In these cases a vertical cut through the binary
phase diagram can be used. In cases where concentration transport cannot be neglected, the local
average composition of the alloy becomes time dependent, horizontal movements through the phase
diagram occur. This causes considerable problems for the solidification algorithm. Examples are:
Isothermal remelting or solidification due to solute transport
If the system is inside a two phase area, e.g. inside a dendritic mushy zone, sometimes isother-
mal flows of alloy with a concentration which is different from the concentration of the inter-
dendritic liquid go through the mushy zone. This is a typical event which leads to the formation
of normal segregations, section 2.1.1.1. The solute changes the local average composition,
which, for equilibrium processes, determines the position inside the phase diagram. A horizontal
40 Chapter 2 Modelling of convection effects in casting and welding processes
shift of the (C , T ) point results. Fig. 2.15 a) shows the situation inside a liquid-solid area of an
alloy with segregation coefficient ls =Cs /Cl < 1.
The equilibrium phase diagram prescribes the concentrations of solid and liquid phase at the
phase boundary at the current temperature. If the new liquid concentration is higher than the old
one, the local equilibrium between solid and liquid is violated. Due to the solute transport the
liquid concentration has become too high. The solution of the problem: Some of the depleted
solid remelts until the average composition of the liquid has come down to the composition pre-
scribed by the phase diagram.
Of course the inverse effect occurs if the advected melt is depleted or if the segregation coeffi-
cient ls > 1. In both cases isothermal solidification happens. In the first case because the for-
mation of depleted solid releases solute which fills up the depleted melt, in the second case,
because the solid is already enriched, and the high concentration level of the liquid can only be
lowered by the formation of more enriched solute.
Remelting or solidification due to convective heat transport
A comparable effect which is easier comprehensible for the common sense, is the remelting
which occurs if the local temperature rises due to the advection of heat, Fig. 2.15 b). This is the
effect which is mainly responsible for the formation of very different weld pool shapes in
welding processes. Molten material which has been heated by a welding source is transported
into cooler areas. Here it accelerates the remelting process and opens the way for further advec-
tion of hot melt. Due to this process the shape of the molten area is to a great extent defined by
the flow field. For more details about the influence on welding processes, cf. section 2.6.
Of course in general both effects do not occur separately but at the same time as parts of very com-
plex solidification and remelting processes. Fig. 2.16 shows a typical simulated (C , T ) path
through a binary phase diagram which occurs during a welding process inside the mushy zone. As
soon as the system reaches the two-phase-area, convection changes the mixture concentration and
2.4 Solidification and fluid flow: Thermal and solutal effects 41
leads to mixed horizontal and vertical movements through the phase diagram. The simulation of
these arbitrary paths requires a high stability of the solidification algorithm.
The authors give numerous examples for all three groups. A detailed review of available fixed grid
techniques is given by /Voller 90/. An important subject when dealing with convection-diffusion
phase change in fixed grid formulation is the treatment of velocities inside the grid cells that contain
the phase boundary. Depending on the nature of solidification a cell can contain
A smooth continuous front (water, rapid solidification of pure metals)
A mushy zone consisting of liquid and dendritic or floating equiaxed grains (most metal alloys)
A continuous transition between liquid and solid without distinct phase boundary (wax, poly-
mers or glass)
A way to treat especially the mushy zone and continuous type is the enthalpy-porosity technique,
/Brent 88/. It considers the cell to be filled with a porous medium. The flow through this medium is
inhibited by drag forces which depend on flow velocity and permeability of the porous medium.
The permeability is determined by the fraction of solid and the morphology. The application of iso-
tropic or anisotropic permeability models allows to account for local variations of the solidification
structure, /Schneider 95a/
The modelling of solidification including the influence of concentration changes due to convective
solute transport is difficult and not yet very common in industrial applications. Many models,
among which are most welding models describing weld pool formation, flow phenomena and so-
lidification, do not take concentration changes into account at all. They only consider vertical cuts
through the phase diagram containing fixed liquidus and solidus temperatures, and they include
solidification characteristics by the definition of latent heat and temperature dependent solid frac-
tions, e.g. using the apparent heat capacity method in the FEM casting simulation software package
CASTS, /Laschet 98/. A pioneering work in the field of solidification including convective solute
transport is /Prakash 89/, which is based on earlier publications by Bennon and Incropera, /Bennon
87a/, /Bennon 87b/. The authors use relations derived from the binary phase diagram to calculate
the fraction of solid as a function of local temperature and mixture concentration. The resulting
solid fractions are reinserted into the temperature and concentration equations to yield a better guess
of temperatures and concentrations. This procedure is iteratively repeated until equilibrium is found.
Start and stop of solidification or remelting are triggered by limiting the solid fraction value to an
interval between 0 and 1. If temperature reaches eutectic temperature the model switches to differ-
ent phase diagram relations using the fixed eutectic temperature.
These basic procedures are refined and extended to the simulation of PbSn, /Schneider 93/ and
multicomponent unalloyed steel, /Schneider 95b/, adding Scheil and backdiffusion model and peri-
tectic solidification. They use different subroutines for the calculation of the different solidification
types, but the decision which subroutine to use is based on the temperature and concentrations at the
beginning of the time step. This makes especially the transitions between different solidification
types difficult.
2.4 Solidification and fluid flow: Thermal and solutal effects 43
It can be extended for Scheil- and backdiffusion models. For this purpose a differentiation be-
tween the average concentrations and the concentrations at the phase boundaries has to be intro-
duced (not the easiest task of all ...).
culation of binary alloy macrosegregations, but there is no evidence that both mechanisms have
been coupled yet. Complex models considering the interaction between macrosegregation and sur-
face movement have been developed in the area of continuous casting. /Kuznetsov 97, 98, 01/
simulates the formation of macrosegregations in a strip casting process and considers a steady state
free surface flow with solidification. /Thevik 98, 99/ develop a complex model to describe the for-
mation of surface segregation in aluminium direct chill casting.
To the authors knowledge this is the first model which couples a mobile solid model with a solidi-
fying free surface in order to simulate the formation of deep shrinkage cavities in a full Navier
Stokes formulation.
Together with differential equations for temperature, concentration in the liquid, pressure and ve-
locities the models have been integrated to form a framework with complex interactions. In the pre-
sent stage of development little effort has been put on the use of complicated physical models. The
philosophy of the project is to create a modular framework of simple models which in the future can
be replaced by more sophisticated models without jeopardising the complex interactions.
Table 2.1: Overview over important GTA welding models of the last two decades
Publications Characteristics of the model Special features
/Kou 85/ GTA, 2D, CVM, steady state, spot Mushy zone, 6061 Al-alloy
/Kou 86/ GTA, 3D, CVM, steady state, line Mushy zone, 6061 Al-alloy
/Oreper 86/ GTA, 2D-axisymmetric, FDM, tran- Dendrite arm spacing, carbon steel
sient, spot
/Zacharia 89/, GTA, 3D, DEA, transient, spot Moving free surface, temperature de-
/Zacharia 91/ pendent material properties and va-
porization (thermal), AISI 304 stainless
steel
/Ramanan 90/ GTA, 2D-axisymmetric, FDM, steady Mushy zone, multigrid methods, di-
state, spot mensionless, Al and steel
/Choo 92a/, GTA, 3D, CVM, transient, spot, AISI 304 stainless steel, vaporization
/Choo 92b/ PHOENICS (thermal)
/Hong 95, 98/, GTA, 2D-axisymmetric, FEM, steady Turbulence, deformed free surface,
/Weckman 98/ state,spot AISI 304 stainless steel, 6061 Al-alloy
/Clarke 97/ GTA, 3D, FEM, steady state, line Turbulence, columnar to equiaxed tran-
sition (CET), mushy zone, dynamic
mesh remapping, Al-4.5wt%Cu
/Winkler 98/ GTA, 2D-axisymmetric, FEM, tran- Experimentally observed W-shape
sient, spot, FEMLEGO pools, AISI 304 stainless steel
/Dilthey 98/, GTA, 2D-axisymmetric, FEM, steady Deformed free surface, excellent ex-
/Pavlyk 99/, state, spot, FIDAP perimental verification, AISI 304
/Pavlyk 01/ stainless steel and S460M steel.
/Wang 01/ GTA, 2D-axisymmetric, CVM, tran- Excellent study, AISI 304 stainless
sient, spot steel
/Aidun 02/ GTA, 3D, DEA, transient, spot and Moving free surface, shrinkage effects,
line, WELDER vaporization, arc pressure, 6061 Al-
alloy
/Hughes 02/ GTA, 3D and 2D-axisymmetric, Application to T-junction weld, time
FVM, transient, line and spot, dependent surface deformation, steel,
PHYSICA 6061 Al-alloy
48 Chapter 2 Modelling of convection effects in casting and welding processes
Table 2.2: Overview over important laser welding models of the last two decades
a) b)
Fig. 2.17: a) Surface tension is due to in-
tramolecular forces. b) Top view onto a liq-
uid/air surface with a boundary between two
liquids with different surface tensions. The
varying surface tension creates a resulting force
along the surface, perpendicular to the boundary
between the liquids. c) The quick surface flow
that is created by Marangoni forces couples into
the melt due to viscosity and causes the forma-
c) tion of convection cells.
2.6 Application: Convection effects in welding processes 51
The variety of pool shapes can be explained by variations of surface tension behaviour. The gov-
erning quantity is the Marangoni coefficient (MC) T , i.e. the temperature dependence of sur-
face tension. As the direction of Marangoni flow is always towards the area of higher surface ten-
sion, the MC determines this direction. In most welding problems there are high temperatures in the
middle of the pool and cooler temperatures at the borders. Fig. 2.19 shows the situation for the (as-
sumed) cases of constant negative and positive MC. A negative MC means that the surface tension
is highest at the borders. The Marangoni flow will be directed outwards, hot melt will be pushed
against the borders of the pool and the pool will be enlarged, Fig. 2.19 a). If the MC is positive, the
hot region in the middle is the area of highest surface tension. The Marangoni induced flows from
all directions meet in the middle of the pool, and, for reasons of mass conservation, are redirected
towards the bottom of the pool. The flow of hot melt deepens the pool considerably and forms a V-
shape pool, Fig. 2.19 b).
This explanation is correct for the assumption of constant MC and helps to understand some of the
principal mechanisms. In reality, surface tension and therefore the MC is a complex function of
temperature and concentration of surface active elements. A famous semi-empirical model proposed
by /Sahoo 88/, section 3.3.2.9, predicts that the MC of steel changes its sign from + to for high
temperatures. Therefore there must be a critical temperature (CT) where the MC becomes zero.
Cooler areas have a positive MC, and thus a flow towards the middle of the pool, hotter areas have
a negative MC, and thus a flow towards the borders of the pool. This mixed flow field can lead to
the formation of W-shape pools as shown in Fig. 2.19 c). Pools of this shape have been
52 Chapter 2 Modelling of convection effects in casting and welding processes
a) b)
Fig. 2.19: Influence of MC on the shape of the
weld pool. a) Flat weld pool due to outwards
directed Marangoni flow for constant MC < 0.
b) Deep (V-shape) weld pool due to inward
Marangoni flow which is redirected towards
the bottom for constant MC > 0. c) W-shape
weld pool due to mixed in- and outward flow
for temperature dependent MC using the for-
mula by /Sahoo 88/.
c)
a) b)
Fig. 2.20: Experimental W-shape pools, observed by /Winkler 98/ in a GTA spot welding process
with P = 690 W and rHS = 2 mm; sulphur content: 0.0005 wt%; welding duration: a) 1 s; b) 3 s
2.6 Application: Convection effects in welding processes 53
Experimentally observed, but thery are not as common as to be expected from theory. Fig. 2.20
shows an example of a W-shape pool observed by /Winkler 98/.
2.6.2.1.1 Experiments
Heiple, Roper and Burgardt have done basic experimental research to determine the influence of
small additions of surface active elements like S and O on weld penetration. The elements were
added in form of Al / Al2O3 / FeS impurities in holes drilled into the base material, /Heiple 82/, or
by adding small amounts of SO2 and O2 to the shielding gas, /Heiple 85/, cf. Fig. 2.18 a). The
authors describe the principal mechanisms leading to flat or deep weld pools respectively. /Xiao
97/, cf. Fig. 2.18 b), measure the surface tension of liquid metals and alloys under arc welding con-
ditions by evaluating pool oscillations and find the link between pool shape and the sign of the MC
confirmed. Very recently /Subramaniam 01/ have measured the surface tension of molten steel
droplets in a gas-metal-arc welding plasma with varying shielding gas compositions. A very im-
portant question in predicting the pool shape formation is the magnitude of the Marangoni driven
surface velocity. Measurements are extremely difficult due to the high temperatures (extremely
bright light) involved and the small dimensions of the weld pools. /Ogawa 00/ manages to observe
Marangoni driven surface flow during GTA welding of AISI 304 stainless steel under microgravity
54 Chapter 2 Modelling of convection effects in casting and welding processes
and determines the mean velocity of the flow to be 0.15 m/s, the flow just below the arc being much
faster.
2.6.2.1.2 Simulations
Many simulations use steady state models. This is very reasonable for line welding, as the interest-
ing part of most of these processes is the equilibrium state. A steady state in spot welding can be
reached if the welding plate is cooled.
/Kou 85/ investigates the relative importance of the influence of the three driving forces buoyancy,
Marangoni and electromagnetic forces. /Weckman 98/ states that turbulence modelling is essential
for correctly predicting weld penetration. He finds that a simulation yielding a deep pool with a
laminar model may yield a flat pool if a turbulence model is used.
Due to the constant increase of computer power transient simulations which require much longer
calculation times become easier accessible. Nonetheless one publication investigating the transient
GTA weld pool evolution with zero, negative and positive MC is from the mid-eighties, /Oreper
85/. Even earlier is the simulation of transient laser weld pool behaviour for negative MC by /Chan
84/. /Choo 92a/ and /Choo 92b/ investigate the interactions between buoyancy, Marangoni and
electromagnetic forces and their influence on the pool shape. Many publications show transient axi-
symmetric 2D, /Lei 95/, /Pericleous 95/, /Hughes 02/, /Taylor 02/ and 3D /Aidun 02/ calculations of
pool shapes with constant negative and positive MC resulting in flat and deep pools. An excellent
work is /Wang 01/, who present a parametric study of transient GTA spot welding processes that
shows how the interaction of Marangoni and electromagnetic forces lead to the formation of differ-
ent numbers of vortices / convection cells. They study the influence of temperature and sulphur
concentration and find, that for given welding conditions, a minimum threshold sulphur concentra-
tion is required to create a single vortex for deep penetration.
Models using constant values for the MC cannot model W-shape pools, but even many models that
use the Sahoo formula for temperature and concentration dependent MC from /Sahoo 88/ do not
observe this kind of weld pool behaviour, e.g. /Wang 01/, /Weckman 98/. Explanation for this may
be the fact that W-shape pools can only be observed under special welding conditions. Only a few
publications show experimental results and simulations of W-shape pools. Excellent works are
/Winkler 98/, who present systematic studies of transient GTA spot welding simulations with dif-
ferent sulphur contents and welding powers and compare them to experiments; /Dilthey 98/, /Pavlyk
99/ and /Pavlyk 01/, who compare steady-state experiments of GTA spot welding on steel plates
cooled from the bottom with simulations including temperature dependent MC and temperature
dependent material properties and find excellent agreement.
/Robert 01/ aims at the explanation of different pool shapes observed in stationary laser welds of
different materials for different welding conditions. They compare experimentally-determined and
numerically-computed weld pool geometries in materials that do not contain any surface active
2.6 Application: Convection effects in welding processes 55
elements like Ga, Al, 5182-Al alloy, Fe, steel, Ti, and NaNO3 and try to extract general rules of
weld pool behaviour using dimensionless numbers. They find a W-shape pool for NaNO3, but the
simulation shows this as a result not of two competing vortices, but of a single vortex, and the
shallow central depth is explained by reduced heat conductivity.
In most publications the formation of the different shapes is explained and interpreted as a function
of the concentration of surface active elements like sulphur or oxygen, but there is theoretical evi-
dence, that the pool shapes must depend on the maximum temperature, and therefore on the welding
power, welding duration and beam radius /Ehlen 02a, 03b/. Unpublished experimental work by
/Pitscheneder 96/ investigates the question why the splitting into flat pools for low sulphur contents
and deep pools for high sulphur contents happens at high welding powers and not at low ones. He
states that the reason is competition between conductive and convective heat transfer as well as the
temperature dependence of the MC.
2.6.2.2.1 Experiments
The importance of concentration changes in the weld pool is subject to ongoing discussions. There
is some experimental evidence, that concentration changes occur, but due to the numerical and ex-
perimental difficulties, there is not much evidence on the exact proceses and the relative importance
of competing effects.
Quite assured is the influence of vaporization on the composition of the weld metal. /Block-Bolten
84/ measure the vaporization of alloy elements from aluminium and stainless steel in order to test a
vaporization model. They find, that quantitative predictions are difficult due to the imperfect
knowledge of surface temperature distributions and condensation effects, but they identify the
dominant metal vapors. /Collur 89/ relate the rates of vaporization of alloying elements from the
weld pool to the emission spectra of the plasma during pulsed laser welding of AISI 201 stainless
steel inder various welding conditions. They investigate the increase of the iron vaporization rate
due to the addition of surface active elements. /Zhao 01/ present an experimental and theoretical
study on the change of weld metal composition during conduction mode laser welding of alumini-
mum alloy 5182. They develop a model for the vaporization rate and weld metal composition
change and compare their simulation results to experiments.
/Benilov 01/ publish a theoretical work on the vaporization of a solid surface in an ambient gas.
Their aim is to estimate the effect of vaporization cooling on the energy balance of an arc cathode,
but the results are general enough to be used, with some precaution, for the prediction of vaporiza-
tion from a weld pool, too.
56 Chapter 2 Modelling of convection effects in casting and welding processes
/Indacochea 85/ investigate the compositional changes of the weld metal during submerged arc flux
welding. They find that the concentrations of C, P, S and possibly Si tend to be larger in the molten
weld metal than in the base plate.
So far there is no experimental evidence of concentration changes in weld pools which could be
compared to macrosegregations in castings. The reason might be that the vaporization effects are so
much stronger than the solute redistribution effects, that the latter are completely superimposed.
An experimental report that might give a hint on convective macrosegregation effects in the mushy
zone is /Munitz 85/, who states a copper depletion on the fusion line between unmolten region and
dendritically resolidified area in a laser molten Al-4.5wt%Cu sample.
2.6.2.2.2 Simulations
The modelling of concentration changes during welding processes is only beginning. In the conclu-
sion of their review of the current state of the art in welding simulation, /Zacharia 95/, a number of
the leading researchers in the area of welding simulation regard convection effects on solute re-
distribution as one of the research issues [that] must be resolved if there are to be further ad-
vances in our fundamental understanding of the complex phenomena that occur during welding.
Nonetheless the numerical modelling of concentration changes is not yet very common. Most nu-
merical welding models do not even consider concentration changes. Even if vaporization of alloy
elements is modelled, the aim is mostly to predict the influence of vaporization on the surface tem-
perature. The composition change in the weld is not calculated. Examples for this kind of models
are /Zacharia 89/, /Choo 92a/, /Pavlyk 01/, /Wang 01/ and /Aidun 02/.
/Chan 84/ at least compare the maximum melt velocity in the pool of a laser surface alloying proc-
ess to the scanning speed and find out that it is one or two orders of magnitude higher. They con-
clude that if any solute is added, it should be redistributed uniformly. /Chakraborty 01/ attempt to
investigate the role of the species diffusion coefficient in laser surface alloying. /Pirch 00/ numeri-
cally study the interaction between mushy zone and Marangoni driven flow during laser surface
alloying. They calculate flow and solute redistribution inside an approximate 3D FEM model of the
transition area between directionally grown dendrities and fully liquid area.
Some work dealing with solute redistribution in laser spot welding, /Ehlen 98/, and in a 2D laser
line welding process, /Ehlen 00a/, has been published by the author. Section 7.2 continues the task.
A recent publication simulating the spatial distribution of dissolved nitrogen is /Palmer 01/. The
nitrogen concentration is calculated using complete time dependent transport equations considering
absorption, diffusion and convection.
2.6 Application: Convection effects in welding processes 57
shrinkage porosities; obstacles, which are massive blocked out regions used to describe mould ma-
terial; void, which corresponds to areas outside the cast part and mould, usually filled with air, but
here considered to be empty.
For each phase the following quantities have to be determined:
The mobile solid phases are assumed to have the same velocities as the liquid phase. All other
phases have velocity zero.
v v v v v v v v v
v move = v move = v l v dead = v dead = v pore shrink = v pore hydro = vobst = v void = 0
(3.1)
Remaining velocity variables: ul, vl (2)
It can be assumed that the local pressure is balanced immediately between all phases. The sta-
tionary phases, obstacles and void do not add to the acceleration of the mobile phases. There-
fores their pressure can be defined to be zero.
p l = p move = p move = p pore shrink = p pore hydro = p p dead = p dead = p obst = pvoid = 0 (3.2)
Thermodynamically the mobile and stationary fractions of - and -phase are considered to be
one single phase. They have identical concentrations:
60 Chapter 3 Theory
Remaining phase variables: l, move, move, dead, dead, obst, pore shrink, pore hydro, void (9)
For convenience reasons sometimes the additional variables
:= move+ dead := move + dead
are used instead of dead, dead and pore shrink, pore hydro are replaced by
pore := pore shrink + pore hydro
Thus the 45 + 9 nelements variables reduce to be 13 + 3 nelements. The set of equations which can be
So far we have 4 + 3 nelements equations. This equation array is sufficient to determine velocities,
pressure, temperature and concentrations assuming that the phase fractions are known. Now 9 more
equations have to be found which are suitable for determining the phase fractions:
The fraction of the obstacle phase is constant and results from initial conditions.
obst = const. (equation 2)
The fraction of shrinkage porosities pore shrink is determined by a modified continuity equation
where the minimum pressure is limited to
p = pcrit pore shrink (equation 3)
The fraction of hydrogen porosities pore hydro is determined as a function of the fraction of newly
formed solid and local solidification conditions
3.1 Mathematical model 61
p0 T s
pore hydro = p s (VH VH max ) dt (equation 4)
T0 t
The fraction of void results from a modified volume of fluid (VOF) equation
void = 1 vofm (equation 5)
The fractions of mobile solid phases move and move are determined as functions of the over
all solid fractions and . (equations 6, 7)
To determine the over all solid fractions and a concept is used that has basically been de-
scribed in /Prakash 89/. One of the two missing equations is obtained from the phase diagram:
T (C ) = T
i
l liq
(equation 8)
Unfortunately this equation does not contain any of the phase fractions in demand, whereas the
temperature and liquid concentration equations do:
T = T ( , ) C = C (
i
l l
i
, )
But as we already need them to calculate the temperature and liquid concentration respectively,
they give no additional information about the phase fractions. This information is obtained from
the conditions arising from the actual solidification type. We consider three different types:
1. primary delta solidification: Only liquid and solid is present, the missing equation be-
comes = 0. (equation 9)
2. primary gamma solidification: Only liquid and solid is present, the missing equation be-
comes = 0. (equation 9)
3. eutectic / peritectic solidfication: liquid, solid and solid is present, The missing equa-
tion arises from the condition, that the solidification is linked to the eutectic / peritectic pla-
teau. There are two possibilities for the equation:
T ( , ) = T
or C ( , ) = C
eut / peri
i
l
(equation 9)
i
l eut / peri
The other equation is then always identical with the phase diagram relation, equation 8.
4. For the case of solid state transformation - the equation for C ( , ) does not yield i
any information because = 0. The equation is replaced by the mixture concentration equa-
l
T (C ) = T
i
(equation 8)
62 Chapter 3 Theory
Now there is a complete system of equations to solve the problem. The next section will give a de-
tailed description of these equations.
1. Equal velocity for all mobile phases, zero velocity for all other phases (eq. 3.1)
2. Pressure equilibrium between mobile phases, zero pressure for all other phases (eq. 3.2)
3. Thermal equilibrium between the phases (eq. 3.3)
4. Concentration equilibrium between mobile and stationary part of same phase (eq. 3.4)
some more assumptions and definitions were made to simplify the complex problem to be solved:
5. Heat capacities of all phases are assumed to be constant and equal for the mobile and stationary
part of the same phase.
c P q = const (3.5)
c P move = c P dead =: c P c P move = c P dead =: c P
c P obst 0 c P pore shrink = c P pore hydro = c P void = 0
7. The mobile phases are hydrodynamically considered as a fluid having an effective viscosity
eff move.
8. The densities of the gas phases void, pore shrink and pore hydro are assumed to be exactly zero.
Some definitions help to simplify the equations. In some cases all mobile phases
q = move = l, move, move
or all stationary phases
q = dead = dead, dead, obst
3.3 Equations 63
are considered as a mixed phase. For these mixed phases the following phase fractions and effective
properties are defined:
move dead = q
/
(3.7)
q =move / dead
( q q ) ( q q )
q=move / dead q=move / dead
eff move / dead = = (3.8)
q move / dead
q=move / dead
( q q cP q ) q=movedead
( q q cP q )
c P eff move dead = q=move dead
/
= /
(3.9)
( q q ) eff move dead move dead
/
/ /
q =move / dead
3.3 Equations
This section gives a detailed description of all basic equations introduced in section 3.1. For the
conservation equations we always start with general single phase conservation equations and de-
velop the multi-phase equations by introducing the model assumptions.
64 Chapter 3 Theory
a) b)
Fig. 3.1: Schematic equilibrium phase diagrams used by SoliCon and EL2D; a) eutectic b) peritec-
tic; the numbers in circles denote the phase areas, the numbers without circles the phase lines that
correspond to each other and that are used by the program.
v
( ) + ( q q vvq ) = p q = : q (3.13)
t q q p
As for arbitrary phases p q = q p , these terms disappear when being summed over all phases
q p p q = q q = 0
This equation can be considered as a version of the continuity equation and can be written as:
v
q q = q ( q q ) + ( q q vq ) = 0
v
(3.14)
t
The terms concerning the weightless phases void , pore hydro and pore shrink have disappeared,
because for them the we assume q = 0.
3.3 Equations 65
Adding the momentum equations for all phases yields a single vector equation which is able to de-
scribe the movement of the mixture of liquid and mobile solid phase fractions.
vv ) + v ( vv vv ) = ( v p ) + v t + ( gv ) + Mv + ( S )
t (
q
qq q
q
q q q q
q
q
q
q
q
qq
qp
qp
q
q external q
(3.18)
Now we are going to work on the single terms. On the one hand they should be expressed in the
formulation of the general differential equation, on the other hand they should be simplified by in-
troducing the model assumptions, that there is only one velocity, vv , for the liquid and the mobile l
solid phases. This velocity is the , i.e. the real velocity which is observed inside a po-
pore velocity
v v uquq u q v q
( )
v v
q q vq vq = q q
q q vq u q vq v q
( q ququq ) +
( q q u q vq )
(
v
q v q u q )
v
x y
=
v ( q
q q v q v q )
= v
q
x
( q q vq u q ) +
( q q vq vq )
q
With introduction of the model assumptions all terms, that contain the velocities of the stationary
solid phases, obstacles, pores and void phase, disappear. In all remaining terms the velocities are
replaced by the velocity of the liquid phase. The convection term becomes:
v
v
[( l l + move move + move move )vvl ul ] = vv [ move eff move vvl ul ]
[( l l l + move move + move move )vvl vl ] [ move eff move vvl vl ]
The phase fractions occurring in this term must not be interpreted as each phase adding its own par-
tial pressure to obtain the total pressure. It can rather be assumed that the pressure is locally equili-
brated between the phases instantaneously. For the processes considered this assumption is a very
good approximation. Multiplication with the corresponding phase fractions must be interpreted as a
reduction of the area the pressure is acting on. If the whole domain is liquid or contains liquid and
mobile solid phase, the pressure acts onto the whole area, the phase fractions sum up to 1. If a cer-
tain amount of the domain is covered by stationary solid, then the area, by which the pressure can
cause acceleration, is reduced.
uq q q v q q
v v q q +2 u q + q uq + q + v q
( q q u q ) x x x y y x x
= v v +
q ( q q v q ) q uq q q v q q
q q + uq + v q + q q +2 v q
x y
y y x y y
l move move 0
+ l + move + move
u
x y y y l
move move ul v
+
l
y y
l
+ move
y
+ move
y
2v
( )
+ l l + move move + move move v l
l
l move move vl
+ l + move + move
0
y x x x
Only assuming that all mobile phases have a common equal viscosity eff move and using eq. 3.7 and
3.8, the dissipative term reduces to
v v
[
eff move move ul
v v
]
[
eff move move v l ]
move 2ul v move 0
+ eff move
+ eff move move ul +
eff move
x x vl x y u l
move ul v move v l
+ eff move
2v + eff move move vl +
eff move
y y l y x 0
represents the influence of gravity on the development of the velocity and pressure fields. A more
detailed form is:
)+ (
v
q q
q
(
T ,C
q ref
q
C
q
)
C q ref g
Cq
In this form the term can be used to describe the following buoyancy effects:
To model this term in the context of the Split Solid Model, the phases which help generating the
local hydrostatic pressure have to be identified. One condition for the model is that buoyancy term
and pressure term must fulfil the hydrostatic paradoxon. The terms for pores and void disappear
because their density is assumed to be zero. Further it is assumed that the stationary solid phases do
not participate in the pressure build-up, but the liquid and the mobile solid phases do. The solutal
expansion coefficients are assumed to be equal for all phases. Thus the thermal buoyancy term be-
comes
q q (
T ,C q ref )+ ( Cq
v
g
C q ref )
q =l , move ,
C
move
For the pure hydrostatic case with zero velocities the momentum equations reduce to
)+ (
v
0 = ( l ) ( )
v
+ move + move p +
q =l , move,
q
q T ,C q ref
C
Cq
g
C q ref
move
p
q T ,C q ref +
q
C
C C
q (
q ref ) ( )
q =l , move ,
x = move 0
p (
l + move + move )
g
y
q q ( ) ( )
T ,C
q ref + C
q C q ref
q =l , move,
C
p p move
0
x
=
y
=
( l + move + move ) g
In the case of disappearing solutal expansion coefficient and equal density for all phases a pres-
sure gradient results, that depends only on the depth and in no way on the distribution of the phase
fractions.
p
= g
y
Mv l l Mv l move
Mv l move
Mv l dead
Mv l dead
Mv l obst Mv l pore sh Mv l pore hy Mv l void
Mv movel
v
Mv move move
v
Mv move move
v
Mv move dead
v
Mv move dead
v
Mv moveobst
v
Mv move pore sh
v
Mv move pore hy
w
Mv movevoid
v
Mv movel
Mv move move
Mv move move
Mv move dead
Mv move dead
Mv moveobst Mv move pore sh
Mv move pore hy
Mv movevoid
Mv dead l
Mv dead move
Mv dead move
Mv dead dead
Mv dead dead
Mv dead obst
Mv dead pore sh
Mv dead pore hy
Mv dead void
Mv dead l
Mv dead move
Mv dead move
Mv dead dead
Mv dead dead
Mv dead obst
Mv dead pore sh
Mv dead pore hy
Mv dead void
5. The mobile solid phases are assumed to be completely mixed with the liquid phase and to be
moving with the same velocity. Therefore the interaction terms between mobile solid phases and
liquid phase are zero (i.e. considered in the dissipative term; 6 terms).
6. The stationary solid phases do not influence each other among themselves (2 terms).
7. With all now remaining terms a stationary solid phase is involved. Normally the relation
Mv qp = Mv pq should apply and the interactions would eliminate each other due to momentum
conservation. However, a stationary fixed phase represents a momentum sink, since the mo-
mentum transferred by friction is removed from the closed system via the mould wall. This can
be described by setting one of the symmetrical terms in each case to zero (6 terms).
0
0 0 Mv l dead vMv l dead
0 0 0 0
0 0 0 Mv move dead Mv move dead
v
0 0 0 0
0 0 0 M move dead M move dead
0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
Six terms remain, which all describe the momentum loss by interaction of one mobile phase with
one stationary solid phase. If flow inside a dendritic network is considered, all solid phases together
and all mobile phases together can be regarded as one phase. The momentum loss of a flow through
a porous medium can thereby be written using the law by Darcy:
v v
M q p = M Darcy = ( l + + move )( l l + move move + move move )K
v
1
2
vl
( )
move
qp
(3.19)
cos 2 sin 2 1 1
+ sin cos +
K K K K
with K2
(3.20)
1
( )
=
1
sin cos
1
cos 2
sin 2
+
K K
K K
K and are the permeabilities parallel and perpendicular to the dendrite axes and were chosen
K
to be functions of the primary and secondary dendrite arm spaces as well as the volume fraction of
mobile phases according to the model presented in /Schneider 95a/.
In the case of isotropic permeability K = K = K0 . In this case the permeability tensor degenerates
to be
3.3 Equations 71
K2
( )
1
=
1 1 0
with K = K0
( l + move + move )
3
(3.21)
K 0 1 ( dead + dead )
2
q q + C (Cq Cq ref ) g
v
+
dead
vvl
q = l move
, , K ( l +
0 move + move )
2
move
(
+ l S external l + move S external dead + move S external dead ) (3.22)
where u is the radial velocity component parallel to the surface, r and z are the coordinates parallel
and perpendicular to the surface, and a is the thermodynamic activity of alloy element i. As gener-
i
ally solutal Marangoni convection is considered to be much smaller than thermal convection, the
72 Chapter 3 Theory
solutal effect has been disregarded in this work by assuming the solutal Marangoni coefficients
/ a = 0 . In fact this does not mean that the thermal Marangoni coefficient / T does not
i
depend on the actual amount of surface active elements. A semi-empirical equation set up in /Sahoo
88/ for binary metal-solute systems gives the following expressions for surface tension and thermal
Marangoni coefficient / T :
(T , a s ) = m
0
A(T Tm ) RT s ln (1 + K seg a s )
K seg a s s H 0
= A Rs ln (1 + K seg a s ) with K seg = S l e H
( 0 /( RT ) )
(3.24)
T 1 + K seg a s T
The effective radius b of the current distribution has been chosen to be equal to the effective radius
rHS of the heat distribution. Now the EM force can be expressed as
v v
J B = B (J zv J rv )
r z
(3.27)
where
r ) exp ( z 2 b 2 / 12) d
I
Jz =
2 0 J0(
r ) exp ( z 2 b 2 / 12) d (3.28)
I
2 0
Jr = J1 (
m I
B = 0 J1 ( r ) exp( z 2b 2 / 12) d
2
3.3 Equations 73
is the third independent variable in the cylindrical co-ordinate system ( r, z, ). J0 and J1 are the
Bessel functions of the first kind and of the zero and first order, respectively.
If the energy conservation equations for all phases are summed up, the relation Hpq = Hqp leads to
the identical elimination of the exchange terms between the phases. The resulting equation is
v v v
( q hq ) + ( q q v q hq ) = ( q q ) + ( q S external q )
v
q (3.30)
q
t q q
This mixed enthalpy equation is a function of q different enthalpies. If thermal equilibrium between
all phases is assumed, eq. 3.3, and the relations
h q (T ) v v v
= c P q (T ) T hq (T ) = c P q (T )T and v
qq = q q T (3.31)
t t
hold, the enthalpy equation can be transformed into a temperature equation. To achieve this the
LHS terms are derived using the product rule, and the heat conduction terms are replaced:
hq v v v v
+ [ q q v q hq ] = ( q q )T
q q q t q
q
v
q hq t ( q q ) + ( q q vq ) + q ( q S external q )
v
The spatial and temporal derivatives of the enthalpies are replaced by the corresponding functions
of T.
T v
v v
q [ q q cP q ] t + q [ q q cP q v q ] T = q ( q q )T
v
v v
hq ( q q ) + ( q q v q ) + ( q S external q )
q t q
In order to discretise the convection-diffusion-term, all factors on the LHS are reintegrated into the
operators. In addition we assume constant heat capacities, eq. 3.5:
74 Chapter 3 Theory
v v v
( c )T + ( q q c P q v q )T = ( q q )T
v
t q q q Pq q q
v v
+ (c P qT hq ) ( q q ) + ( q q v q ) + ( q S external q )
q t q
If we insert the general continuity equation, eq. 3.14, which has been resolved for
v v
( ) + ( v ) , we get:
l l l l l
t
v
( q q c P q )T + ( q q c P q vq )T
v
t q q
v v v v
[ ]
= ( q q )T + (hl hq ) (c P l c P q )T ( q q ) + ( q q v q ) + ( q S external q )
q q l t q
+ [(hl
hq ) T (c P l cP q ) ] (qt q )
q l
v
{[(hl hq ) T (c P l ]
c P q ) ( q q vl )}+ ( q S external q )
v
+
q = move q
move
If we assume that enthalpy is a linear function of temperature, eq. 3.6, the equation simplifies to
v v v v
( q q c P q )T + ( q q c P q ) v l T = ( q q ) T
t q q = move q
+ [([h0 l
c P l T0 l ] [h q
0 c P qT0 q ])] (qt q )
(3.32a)
q l
v
{[(h0 l c P l T0 l ) (h0 q ]
c P qT0 q ) ( q q vl )}+ ( q S external q )
v
+
q = move q
move
In fact this is a major simplification as it causes the major part of the RHS source terms neither to
depend on temperature nor on enthalpy, but only on the amount of solid phase that has been created
or exchanged with neighbour cells. Using the definitions eq. 3.7 3.12 the equation can be written
as:
[( move eff move c P eff move + dead eff dead c P eff dead )T ] v v
+ ( move eff move c P eff move vl T mat eff mat T )
v
t
(3.32b)
* ( q q )
v
= q
+
t q = move q
[
*
(
q q lv
v
)] + (
q q external q
S )
q l
move
3.3 Equations 75
The heat source that is used for laser and GTA welding problems is assumed to be Gaussian and can
be described as:
2
2r
qheat source = qmax exp 2
(3.34)
rHS
where r is the distance from the middle of the heat source, rHS is defined as the radius where the
heat flux has decreased to 1 / e 2 of the maximum, and qmax is numerically determined in a way that
the integral of the heat flux over the whole surface is exactly normalised to the effective input
power. This already includes the heat input efficiency , or assumes = 100 %.
Heat losses by radiation and air convection are calculated using the well known relations
qradiation = b b (T 4 Ta4 ) (3.35)
qconvection = convection (T Ta ) (3.36)
3.3.3.2.3 Vaporization
In welding processes with high surface temperatures the high vapour pressure plays a major role. To
calculate the heat loss by evaporation a two-step model was implemented: Beneath the boiling tem-
perature the heat loss is linked to the temperature dependent vapour pressure. The model has been
described in /Zacharia 91/. The evaporation heat flux is calculated based on an over all vaporization
model from /Choi 87/ given by
qevaporation = WH v (3.37)
where Hv is the heat of evaporation. The equation given in /Dushman 62/
log W = Av + log p 0 0.5 log T (3.38)
is used to calculate the evaporation rate W. Av is a constant that depends slightly on the main com-
ponents of the alloy used. For the case of a binary Fe-C alloy the values of pure iron were taken.
The evaporation rate is valid for the case of evaporation into vacuum. In the present case evapora-
tion into an atmosphere of 1 bar is considered. The real evaporation rate will therefore be smaller
76 Chapter 3 Theory
than the one assumed here. Test calculations showed, that the qualitative behaviour of the pool
shapes does not change. To calculate the vapour pressure an expression given by /Kim 75/ for
stainless steel is used:
18 836
log p 0 = 6.1210 (3.39)
T
If the temperature is about to exceed the boiling point, the temperature is limited to this value, be-
cause it is assumed that all excess heat is directly lost by evaporation. The mass loss and the high
surface pressures are not considered.
The concentration equation for alloying element i in phase q reads (units: [kg/(sm3)]):
( C i ) + v ( vv C i ) = v ( D~ i v C i ) + J i + J i j + S (3.40)
t q q q q q q q q q q q q q q external q
Following /Ni 91/ the concentration transfer by phase change can be modelled as
J = C q
i
q
i
qi
(3.41)
where the phase change rate q corresponds to the single phase mass conservation equation, eq.
3.13
( q q ) v
( q vq ) .
v
q = + q
t
Here it is to be noted that in this equation the solidification rate q is the crucial quantity. If in a cell
both solidification occurs and solid state is removed by convection, these two effects compensate
each other, sometimes even completely. Therefore the quantity
( q q )
t
has no direct meaning any more, since with complete compensation of solidification and removal
by convection the phase portions remain unchanged, the term becomes zero. In this case the solidi-
fication rate p is equal to the rate of the mass removed by convection. The concentration of the
mass formed by solidification must however be equal to the concentration at the solid / liquid inter-
face, C . Thus also the convection term is multiplied by this interface concentration. Whereas the
i
qi
concentration removed by convection from the cell is already considered in the convection term on
the LHS of the equation. Here the concentrations at the cell boundary respectively the adjoining cell
occur.
According to the continuity equation for species i
q J = 0
i
q
J
i
q
= J p q
i
p
3.3 Equations 77
which can be written as a sum of the single phase species conservation equations
~
q t ( q qCqi )+ ( q q vqCqi ) ( q q Dqi Cqi ) q Sexternal q = 0
v v v v
Neglecting the back diffusion terms and using
q J = 0 q
i
In the current version of the model the lever rule is the only microsegregation model. This means
that the concentrations of the solid phases do not need to be determined by solving a differential
equation, but can be directly calculated as
C = C i
l
(3.43a)
i
and
C = C
i
l
i
l
(3.43b)
The tracking of the moving free surface is performed using a modified VOF (Volume Of Fluid)
algorithm based on the algorithms SOLA-VOF /Nichols 80/ and NASA-VOF2D /Torrey 85/. Even
if there are more sophisticated algorithms available nowadays (e.g. SLIC, PLIC, CIP), this algo-
rithm has been chosen for various reasons. VOF algorithms and their derivatives have proved to be
stable and reliable in various numerical applications. The algorithms are easy to implement and
easy to modify. As the fluid dynamics problem to solve does not require a high performance algo-
rithm (the flow velocities are small and the behaviour of the free surface is quite boring compared
to e.g. mould filling problems), the properties of the VOF algorithm are satisfactory. As the free
78 Chapter 3 Theory
surface calculation is implemented as separate modules, it is easy to add more sophisticated models
later on.
The main modifications of the algorithm are the extension to variable densities to describe shrink-
age effects, the introduction of time dependence to the partial cell treatment for modelling solidifi-
cation inside free surface cells and the adaption to the split solid model.
The original VOF algorithm describes the free surface by introducing a function F(x, y, t) which is
defined to be unity at any point occupied by fluid and zero elsewhere. The governing differential
equation for this function is
F + vv v F = 0
t
To adapt this equation for use with partially blocked cells this equation is combined with the conti-
nuity equation. We use a similar procedure to generalise the VOF equation for use with variable
liquid densities and solidification.
We start by introducing the function Fmove(x, y, t) which is defined in a similar way as F. Integrated
over a finite volume it denotes the fraction of the area open to flow which is filled by mobile
phases.
The governing equation for this function is
Fmove + vv v F = 0 (3.44)
t l move
and
l l + move move + move move + pore shrink pore shrink + pore hydro pore hydro + void void
eff open =
l + move + move + pore shrink + pore hydro + void
l l + move move + move move
=
open
dead dead + dead dead + obst obst dead dead + dead dead + obst obst
eff closed = = (3.46a,b)
dead + dead + obst closed
with
open eff open = l l + move move + move move = move eff move
closed eff closed = dead dead + dead dead + obst obst = dead eff dead
3.3 Equations 79
t
( open eff open ) + Fmove ( open eff openvvl ) = Fmove
t
( closed eff closed )
and the VOF equation by open eff open :
( F )
t open eff open move
+ ( open eff openvvl Fmove ) = Fmove
t
( closed eff closed ) (3.47)
The discretisation of this equation requires special care to maintain the step character of the func-
tion and to prevent the function Fmove from being smeared. This is important to obtain a neat de-
scription of the interface between castpart and void. In contrast to the original algorithm it is not the
function Fmove alone which is used for the surface reconstruction, but the volume of all phases which
belong to the cast part, including porosities. The discretisation technique and the surface recon-
struction algorithm are described in section 4.6.
One of the most important parts we need to describe the formation of the shrinkage cavity is the
Split Solid Model (SSM). It is a model which defines the repartition of the solid phases into a mo-
bile part and a stationary or dead part (the first name of the model was dead solid model).
The objective of the model is the following: The assumption that all solid forming is stationary is
far from reality. It is a good approximation for columnar growth only. This would happen for alloys
with small solidification range like pure metals or eutectic alloys. For alloys with long solidification
range there is a columnar-to-equiaxed transition with equiaxed crystals forming inside the melt and
not being attached to any wall.
A model which considers all solid as stationary will not be able to describe the formation of deep
shrinkage cavities as they occur in the riser of some steel castings. Due to the long solidification
range of unalloyed steel (solid liquid interval: 30 50 K) some solid forms early at the surface. In
reality this solid would not be attached to any wall but float and be able to sink down when the fluid
level drops. In a model with stationary solid only these small amounts of solid will be fixed in a
very early stage and inhibit the further surface development.
80 Chapter 3 Theory
To solve this problem the SSM has been developed. The basic idea is that small amounts of solid
which are not attached to the walls or in contact with regions of high solid fraction are considered to
be mobile and hydrodynamically treated as liquid. Obvious limitation of this model is the fact that
no relative velocities between solid and liquid can be considered. Nevertheless some sedimentation
effects have been observed during simulations. The proper mechanisms of these effects have still to
be investigated.
The core parts of the model are the Split Solid Function (SSF) and the Columnar Growth Model
(CGM). The SSF is a repartition function which defines how much of the locally existing solid is
considered to be mobile. The model assumes that the repartition is the same for - and -solid.
Thus the split solid equations become
move SSF (
= + )
(3.48a,b)
move = SSF ( + )
The (CGM) decides in which areas the SSF is applied and in which all solid is considered to be sta-
tionary, i. e. where the SSF is set equal to zero.
Three types of functions have been examined which meet the above criteria
The sinus type function is composed of two constant parts for high and low solid fractions and a
sinus function for the transition area from all mobile to all stationary. Using the definitions
= 0 mv dd 12 mv dd
:
+ := 0 mv dd + 12 mv dd
it is defined as
3.3 Equations 81
1
for + [0 , ]
1 sin
( + 0 mv dd )
SSF sin
( + ) =
mv dd for + ] , + [ (3.49)
2
0 for + [ + , 1]
with
2 6 +
a = 3 2 2
3
c = 3 2
2 3
3 + + 3 + + 3 + + 3 + +
b =
3( + +)
d
= 1 +
(
2
3 +
3
)
3
3 +
2
3 2 3 3 2 2
3 + + 3 + +
3
+ + +
SSFFD ( + ) =
1
+ [ 0 , 1] (3.51)
for
( + 0 mv dd ) +
10
exp 1
mv dd
82 Chapter 3 Theory
a)
b)
c)
3.3 Equations 83
= 0.2. The derivatives are needed for the numerical solidification algorithm. The sinus and spline
functions are almost identical, only the sinus is a little bit steeper, whereas the Fermi-Dirac function
shows a clearly different behaviour. The Fermi-Dirac function shows a much smoother onset of the
transition area. Even its derivative is continuously differentiable. The kinks in the derivatives of the
sinus and spline functions show that their curvatures are not continuous. This would be an argument
to use the Fermi-Dirac function, but it has a big disadvantage: For the points and + it is not ac-
tually 1 and 0, but, in this example, 0.9933 and 0.0067 respectively. As objective of the split solid
model is to prevent the formation of stationary solid in early stages of solidification this disadvan-
tage is essential. However, there are numerical remedies that make this function type applicable.
The choice whether to use the sinus or the spline function merely depends on reasons of numerical
efficiency.
a) b)
Fig. 3.3: Assumptions of the Columnar Growth Model (CGM); a) New solid forming inside a radius
RCGM from the wall is assumed to stick to the wall and add to the forming solid shell. b) Solid
forming close to an area where the local solid fraction exceeds the critical fraction CGM (in this
work chosen to be 0.6) is assumed to grow on top of the solid already present.
forms, the total fraction of solid exceeds CGM (in this work chosen to be 0.6), the solid is assumed
to grow on top of the solid already present. If the solid fraction of the environment is lower, again
floating grains form.
Of course this model is a rough simplification of the multiple effects which have been observed in
reality and which are described by complex theories of CET. Nevertheless it allows to simultane-
ously model columnar and equiaxed growth and can therefore be used to simulate the formation of
deep shrinkage cavities. The performance and validity of the model can certainly be improved by
adaption of the empirical parameters RCGM and CGM to the specific situation using experimental
observations and parameter studies. For other purposes the simple model can, in the present frame-
work, be replaced by more complex models without too much effort.
phases away because they exert a high pressure. Shrinkage porosities form inside an existing den-
dritic network by vaporization of liquid alloy components if the local pressure drops so far that their
partial pressure can initiate pore nucleation against the remaining local pressure. Internal macro-
scopic shrinkage cavities are formed by the same mechanism as shrinkage porosities. The differ-
ence is that they form, if no dendritic network or solid phase is present at the moment when pore
nucleation is initiated.
Other than the model by /Pequet 00/ who treat mushy zone flow and solidification of completely
separated areas (liquid pockets) as different cases, the model presented in this section does not
need this differentiation. The porosity formation rate only depends on whether the volume losses
due to shrinkage can be compensated or not. Whether the feeding path is hindered or completely
blocked does not make any difference.
Even if the model pretends to yield quantitative results, due to some simple submodels used the
current version cannot yield quantitatively correct results yet. So far it can give a description of the
mechanisms leading to the formation of the different porosity types, and explain why they occur in
different areas of the cast part.
In the following sections the basic assumptions and laws used in the model are presented.
element dx dy dz becomes
VH rejected 0 s m
3
= s (VH VH max )dx dy dz
t t s
The term s (VH VH max ) has no units, but can be understood as m3 excess hydrogen per m3 alloy
material for the case that the whole alloy was solidifying during the time dt. Integrated over the vol-
ume element dx dy dz and multiplied by the real volume fraction of the alloy solidifying during the
time dt, it becomes the rate of hydrogen volume rejected inside the volume element during the time
dt.
86 Chapter 3 Theory
The value of VH is valid for ambient pressure p0 = 1.013 bar and temperature T0 = 273.15 K. To
predict how much volume the rejected hydrogen will occupy we use the constitutive equation for
ideal gases pV T = p0V0 T0 :
VH rejected p0 T s m
3
= s (VH VH max ) dx dy dz
t p T0 t s
Here T is the local temperature and p is the local pressure formed as a sum of air pressure and hy-
drostatic head. This is the only force the liquid and mobile solid can object against the pore forma-
tion. Higher local pressures can only reduce the pore volume, but not completely prevent the for-
mation. If a certain mass of hydrogen gas is rejected from the alloy, it must occupy at least some
volume. The smaller the volume is, the higher the pressure inside the volume which works against
the external pressure trying to compress the gas. The gas will get into an equilibrium state where its
internal pressure is equal to the local fluid pressure, Fig. 3.4. An important consequence of this fact
is the prediction, that higher amounts (volumes) of hydrogen porosities should form in areas with
low local pressure like areas close to shrinkage porosity formation. If this happens inside an area
where there is a dendritic network already present, large amounts of pores should appear which look
like shrinkage pores but which are indeed low pressure hydrogen pores.
The formation rate for the fraction of hydrogen porosities becomes:
pore hydro p0 T s
t
= (V VH max )
p T0 t s H
[s ]
1
(3.52)
3.3 Equations 87
Of course this formation rate is limited by the condition that the local liquid and mobile solid frac-
tion move > 0. Hydrogen pores can only form where there is liquid and mobile solid to be pushed
away.
The formation of shrinkage porosities and internal shrinkage cavities can be modelled by adding a
transient term to the pressure equation. The pressure equation consists of the general mixed conti-
nuity equation, eq. 3.16, with the velocity divergences being replaced by the respective velocity
differential equations. It assures that the local pressure is adjusted in a way that volume losses due
to shrinkage are compensated by inflows from neighbouring regions due to pressure gradients. If
there are shrinkage losses, the local pressure must become lower than the pressure in the surround-
ing areas. If the shrinkage losses increase, or the flow resistance of the surrounding areas increases,
which inhibits the inflow, the local pressure must decrease more. The reason why porosities form is,
that the local pressure cannot decrease infinitely. At a certain point the vapour pressure of the base
metal or the alloying elements diminished by the surface tension pressure needed for the pore nu-
cleation becomes higher than the local hydrodynamic plus hydrostatic pressure and bubbles of alloy
vapour form. As the reason for the local pressure drop is the wish to attract liquid and mobile solid
to compensate for the shrinkage induced volume losses, the volume of the newly formed bubbles
(which will later become pores or internal shrinkage cavities) helps to fulfill the continuity equa-
tion. The size of the bubbles is determined by the following mechanism: As long as the diminished
vapour pressure is higher than the local pressure, the bubbles grow, decrease the violation of the
continuity equation and allow the local pressure to rise (i.e. the local pressure drop to decrease),
because the need to attract mobile phases from the neighbouring regions decreases. The bubble
growth stops exactly when the local pressure is equal to the diminished vapour pressure. In this
moment the volume of liquid and mobile solid attracted from neighbour areas and the newly formed
pore volume exactly compensate the shrinkage induced volume losses.
A way to determine the shrinkage porosity formation rate is to calculate the diminished vapour
pressure and to limit the minimum local pressure to this value. The momentum equations using the
reduced pressure gradients yield smaller velocities and in the divergence terms of the continuity
equation an additional mass sink appears. The continuity equation is a mass conservation equation.
This is why the the term
t
( pore shrink pore shrink )
with pore shrink = 0 (by definition, pore shrink 0 would not help much either...) does not have direct
impact onto the equation. The impact comes through the displacement of the mobile phases. The
time dependent term for the mass of liquid and mobile solid must be modified to become
88 Chapter 3 Theory
t
[
( move pore shrink ) eff move ]
to account for the reduced fraction of mobile phases. In the continuity equation it equilibrates the
velocity divergence created by the reduced pressure gradients. So the equation to determine the rate
of shrinkage porosity formation becomes
( ) = ( +
v v
) + [ move eff move vl (p imited )] (3.53)
poreshrink eff move move eff move dead eff dead
v l
t t
The local pressure equilibrium which determines when shrinkage pore nucleation starts, can be
written as
p0 + phydrodynamic + phydrostatic < pvapour psurface tension . (3.54)
The vapour pressure of the alloy components and the pressure created by the surface tension of the
forming bubble are complex functions and have not been modelled in the present work.
They have been combined to form the critical pressure for shrinkage pore formation
Pcrit pore shrink = pvapour psurface tension = const. (3.55)
If there is a dendritic network or solid present at the time when the vapour bubbles form, the shape
of the resulting pores is determined by the interdendritic space available. If possible, the bubbles
displace the mobile phases and fill the space between the dendritic skeleton. This is mostly the case
for alloys with long solidification range (Al-7wt%Si, Fe-0.4wt%C). If the pressure limit is reached
in an all liquid area, the bubbles stay together and form continuous internal shrinkage cavities. This
happens mostly in pure metals or eutectic alloys with plane solidification front (Al99.99, Al-
12.6wt%Si).
As the interactions of the different mechanisms are not easy to understand, the following example
demonstrates the relations between shrinkage losses, feeding flow, drag forces and local pressure. It
shows, how models of the mechanisms can be used to determine the shrinkage porosity formation
rate.
First a general view on the nature of pressure shall be given. Each velocity field in a fluid is accom-
panied by a pressure field which mirrors all velocity changes within time and space. Fluid accelera-
tion is driven by a negative pressure gradient, i.e. pressure drop along the acceleration direction. If
we consider a flow which is deflected in a rectangular elbow of a tube we see, that it is redirected
by a high pressure zone in the edge. The x-component of the incoming flow is decreased by a posi-
tive pressure gradient and then the y-component increased into the perpendicular direction by a
negative one, Fig. 3.5. The pressure gradient is always in equilibrium with the inertia forces of the
accelerated fluid. If pressure gradients exist in stationary situations without velocity change, their
function is to equilibrate body forces. An example is the hydrostatic pressure where a negative pres-
3.3 Equations 89
sure gradient from bottom to top compensates for gravity acceleration. A second example is the
drag force which acts on a flow through a porous medium or mushy zone. A porous medium has a
specific velocity dependent resistance which depends on the pore fraction and the geometry of the
medium. To maintain a constant velocity flow through a porous medium, a constant pressure gradi-
ent is needed. Integrating the pressure gradient along the porous medium yields a discrete pressure
drop. The pressure drop needed to maintain a certain velocity depends on the properties of the po-
rous medium like fraction of porosity and permeability tensor. If considered the other way round,
for a specific porous medium the maximum velocity achievable is given by the maximum pressure
drop available.
Provided with this knowledge, we examine a simple model of a hot spot. To eliminate the compli-
cations induced by gravity the geometry is a horizontal liquid reservoir closed at the left end by a
fixed wall and closed at the right end by a mobile piston exerting air pressure p0. As the model is
considered to be one-dimensional, there are no effects in vertical (y-)direction. At the beginning the
liquid is in rest and the whole reservoir has constant pressure p0 , Fig. 3.6 a).
If now solidification starts at the left end and the solidifying liquid contracts, the volume loss must
be compensated by a flow from the right to the left. The velocity of this flow is determined by the
volume contraction rate. The fluid is accelerated by a pressure gradient. The magnitude of this pres-
sure gradient is determined by the time the system has to accelerate from zero to end velocity v. The
higher the acceleration is, the higher must be the gradient. As the pressure at the right end is fixed to
air pressure, the local pressure sinks along the liquid reservoir, Fig. 3.6 b). The difference between
air pressure p0 and the pressure p1 pressure at the left end is the pressure drop needed to accelerate
the whole liquid reservoir during the time step t from 0 to to end velocity v. The pressure occuring
90 Chapter 3 Theory
a) e)
b)
f)
c)
d) g)
Fig. 3.6: Principal mechanisms during shrinkage porosity formation in a hot spot.
3.3 Equations 91
during this time is hydrodynamic pressure. It disappears once a constant volume contraction rate has
been established and the system is in a stationary state again. The pressure becomes p0 again, even
if now there is a constant velocity, Fig. 3.6 c). Interesting is the case if the acceleration time t is
extremely small (this should not occur during casting processes). In this case the pressure p1 would
become negative. In reality the pressure at the left end has a minimum value, namely the vapour
pressure of the liquid. This limits the possible pressure drop and thus the possible acceleration. The
air pressure is not able to accelerate the fluid enough to compensate for the volume losses, and so
the liquid boils and creates a vapour filled cavity. This effect is known as cavitation and can oc-
cur on the back of ships propeller blades or when liquid is pulled up against gravity.
As a next step solidification starts in the middle of the liquid reservoir and a porous mushy zone
forms. Now a force becomes necessary to maintain the constant mass flux, which equilibrates the
drag forces due to friction. This leads to the formation of a pressure gradient along the mushy zone.
As mentioned before the gradient depends on the mushy zone properties, especially liquid fraction
and permeability tensor. The total pressure drop depends on the gradient and the zone lenght. Fig.
3.6 d) shows the pressure field for two different lengths.
If the mushy zone becomes too long, Fig. 3.6 e), and the pressure p1 drops under the vapour pressur
of the liquid, we have the first case of shrinkage porosity formation. In this case the alloys starts to
evaporate and raises the pressure back to the vapour pressure. As the whole left part has the same
pressure, the porosities can form everywhere in this area. If the sample geometry was not horizontal
but vertical, gravity would add an additional gradient and vapour pressure would be reached only at
exactly one point. This is the point where porosity formation starts in this case, Fig. 3.6 f).
The porosity volume formation rate can be easily determined from the difference between the ve-
locity needed to compensate for the volume losses due to shrinkage and the maximum velocity
achievable with the reduced pressure drop p0 p1. It can be written as
V pore shrink
= [vshrinkage v feed p p ]A (3.56)
t
( 0 1)
where A is the cross-sectional area of the liquid reservoir. The velocities are superficial velocities,
i.e. the velocities observed outside the mushy zone. The pore velocity inside the zone is
v feed ( p0 p1 )
v pore ( p0 p1 ) = (3.57)
move
The second case where shrinkage porosities form occurs, when the mushy zone is not too long, but
becomes more and more closed. If the solid fraction of the mushy zone increases, the velocities in-
side the mushy zone must increase to maintain the constant mass flux. The higher velocities and the
additionally increased drag forces require an increase of the pressure gradient. Therefore the pres-
sure curve becomes steeper and, at a certain point, the pressure p reaches the vapour pressure of the
1
liquid phase, Fig. 3.6 g). Again porosities can form inside the constant pressure area on the left. The
porosity volume formation rate is the same as in the first case.
92 Chapter 3 Theory
Fig. 3.7 shows the principal pressure evolution inside the hot spot for the case of constant volume
contraction rate (feeding mass flux) when the fraction of stationary solid inside the mushy zone
increases. The solid curve is the real pressure inside the hot spot. pressure drop between air pressure
p and the local pressure p . The dashed line shows the pressure that would be necessary to com-
0 1
pletely compensate the shrinkage volume losses. As the pressure in fact is limited, the truncated part
of the pressure is linked to the porosity formation rate.
Two versions of the model have been implemented in the framework of a 2D control volume
(CVM) code in FORTRAN 90. Details on the differences between the software packages EL2D9.1
and SoliCon2.0 can be found in App. 9.2. The code can be classified as a single domain volume
averaging multiphase model. Most of the partial differential equations have been discretised using
an explicit or semi-explicit time discretisation scheme, the pressure equation using an implicit one.
The coupling of the complex equation system is realized using a modified SIMPLER iteration loop.
The calculation domain is always rectangular. This is generally useful for welding processes where
the liquid is always contained by the solid that has not been molten. For casting processes different
mould shapes can be defined by using obstacle cells. These are areas which are filled with rigid,
non fusible solid.
Fig. 4.1 shows the flow charts of the program SoliCon2.0. The main part consists of a number of
nested loops. The initialisation routine (init_calculation), which is used to identify the di-
rectory structure, read in program parameters, geometry, material data, set initial conditions and
allocate and initialise data arrays is followed by the main time step loop.
The time step loop performs the main calculations which advance the system in time and writes the
results into output files when this is wanted. The main calculation is wrapped into a second loop
that allows to redo the time step if by any reason a wrong time stepwidth has been chosen. This can
happen if e.g. the maximum velocity becomes too high. In this case a new, smaller time stepwidth is
calculated and the time step calculation repeated. In general this loop is only executed once.
The calculation for each time step starts with an initialisation step (init_new_time_step) that
determines the ideal time step width (time_step, if adaptive time step control is activated),
stores the values of the old time step and resets the arrays for the new one (store_reset_var).
It determines the new surface heat flux distribution if there is a time dependent (moving / pulsed)
welding source and calculates those terms of the differential equations which depend on the values
of the old time step only (explicit terms, which have not to be iterated).
Follows the outer (SIMPLER) iteration loop which is responsible for the solution of the system of
15 coupled equations. The basic procedure is as follows:
1. Iteration 0 only solves the energy equation using the old values of all phase fractions. This pro-
vides a first guess of the temperatures and gives information to the solidification algorithm
which solidification type is to be expected.
94 Chapter 4 Algorithms
5. Iteration 2, ... Repeat steps 2. 4. Several times, always using the new VOF configuration, but
still starting with old values of the other quantities.
6. Execute phase_frac_and_bound again to get the boundary conditions right for output.
The following sections will give a more detailed view on the different parts of this solution algo-
rithm.
Most of the partial differential equations have been discretised using an explicit or semi-explicit
time discretisation scheme, the pressure equation using an implicit one. As there are some special
problems and assumptions, the discretisations will be described in detail. For reasons of compara-
bility the discretised equations will be cast into the formalism used in /Patankar 80/.
A three dimensional formulation has been used to avoid unit problems (like densities with units
[kg/m2]). There are no dependencies in the third direction and the integration of the third space co-
ordinate is always integrated from 0 to 1 m. Therefore z = 1 m does not change the results, but
results in correct 3D-units.
where the coefficients a , a , a , a and a describe the influence of the variable at the corre-
W E S N
,
P
sponding grid locations and b is the source term. All equations are formulated using the grid loca-
tion names of the normal grid as defined in 4.2 a). This holds also for equations on the staggered
grids, i.e. the general differential equations read
96 Chapter 4 Algorithms
a) b)
Fig. 4.2: Grid definitions used for discretisa-
tion. a) Normal grid, b) x-staggered grid for
velocity component u, c) y-staggered grid for
velocity component v. The outlined cell al-
ways indicates the control volume (i,j) of the
grid. To show the spatial relation between the
different grids the control volume of the nor-
mal grid is always hatched.
c)
a w w = a ww ww + a e e + a Sw Sw + a Nw Nw + b (x-staggered grid)
a s s = a ss ss + a n n + aWs Ws + a Es Es + b (y-staggered grid)
The decision to notate all differential equations in relation to the normal grid was taken to avoid
inconsistencies when equations that are discretised on different grids are joined, e.g. to form the
pressure equation.
A( P ) = [
0, 1( 0.1 P )
5
] (4.1)
where
P= F
D
is the equation dependent Peclet-number which is always defined as the ratio between flow rate F
and conductivity D of the problem.
For discretisation we start with the continuity equation in its most general form, eq. 3.14:
v kg
q q = q ( q q ) + ( q qvq ) = 0 units : m 3 s
v
t
We define
v (Fq ) kg
Fq = (F )x = q q vvq units :
q y
m 2 s
and get
v v
q q = q ( q q ) + Fq = 0
q dx dy dz q dt units : kg ]
[
x y z t (4.2)
= {[( q q )P ( q q )0P ] x y z + [(Fq )e (Fq )w + (Fq )n (Fq )s ] t }=: (qdiscretised )P = 0
q q
98 Chapter 4 Algorithms
where
(F ) = (F ) dy dz = ( u ) y z kg
q e/w q x q q q e/ w units :
s
e/ w
(F ) = (F ) dx dz = ( v ) x z
q n/s q y q q q n/s
n/s
q dx dy dz q dt
x y z t (4.3)
= {[( q q )P ( q q )0P ] x y z + [(Fq )0e (Fq )0w + (Fq )0n (Fq )0s ] t } = : (qdiscretised )P = 0
0
q q
where
(F ) 0
q e/ w = ( q q u q )e w y z
0
/
(F ) 0
q n/s = ( q q v q )n s x z
0
If effective quantities are used, eq. 3.16, the explicit continuity equation can be written as
[( +
move eff move ) ( move eff move + dead eff dead )P ] x y z
dead eff dead P
0
(4.4)
+ ((F ) (F
eff move e
0
eff move )w + (Feff move )n (Feff move ) ) = 0
0 0 0
s
t
where
(Feff move )e / w = ( move eff moveul )e / w y z
0 0
(Feff move )n / s = ( move eff moveul )n /
0 0
s
xz
q dx dy dz q dt
x y z
( ) t (4.5)
= {[( q q )w ( q q )0w ]( x ) y z + [(Fq )0P (Fq )W0 + (Fq )0nw (Fq )0sw ] t }=: (qdiscretised )w = 0
0
q q
where
(Fq )P W = ( q q uq )P W y z
0
/
0
/
(Fq )nw sw = ( q q vq )nw sw ( x ) z
0
/
0
For y the equation is similarly integrated over the y-staggered control volume x ( y ) z .
Special care has to be taken with the definition of the velocities. As the grid locations P, W, nw, sw
are no genuine velocity locations, the velocities there have to be determined by suitable interpola-
tions.
4.3 Discretisation of differential equations 99
g
+ move eff move x ul + ( l S external l + move S external move + move S external move )
K 0 move 2
The most difficult problem that has to be solved during discretisation is the treatment of the con-
vection-diffusion-term. We define:
units : N2
v (J u )x v v
Ju = = move eff move v l ul move eff move ul
m
(J )
u y
After integration over the control volume of the staggered grid it becomes
t + t
dx dy dz t
v
v
[
J u dt = (J u )P
0
0
(J u )W +
0
( J u )nw
0
( J u )sw ] t
( x y z
)
with
0 0 0 0
( J u )P / W = (J
P /W
u
) x dy dz ( J u )nw / sw =
nw /
(J
sw
u
) y dx dz units : [N ]
dx dy dz t
t
( )
move eff moveul dt = move eff move [( )w (ul )w ( move eff move )w (ul )w ]( x ) y z
0 0
( x y z
)
We get the complete LHS of the equation by subtracting the explicit discretised continuity equation
for the x-staggered grid, eq. 4.5, multiplied by (u )0 l w
q ( )
discretised 0
0
(ul )w q w
=0
100 Chapter 4 Algorithms
q move
[ 0 0 0 0
] [
(J u )P (u l )w (Feff move )P (J u )W (u l )w (Feff move )W +
0 0
]
+ t
[
0 0 0 0 0
] [ 0
(J u )nw (u l )w (Feff move )nw (J u )sw (u l )w (Feff move )sw ]
with
(Feff move )P / W 0
= (Fl )P / W +
0
(F move )P / W + (F move )P / W
0
0
= ( move eff move ul )dy dz
P /W
(Feff move )nw / sw = (Fl )nw / sw + (F move )nw / sw + (F move )nw / sw = ( move eff move vl )dx dz
0 0
0
0
units :
kg
s
nw / sw
Now let us have a look at the source terms. The pressure gradient term is integrated using an im-
plicit scheme:
t + t
p ( p )P ( p )W
(
d x d y d z t move x dt = ( move )w
x yz
)
( x )
( x ) y z t
The dissipative term is split into four parts for (explicit) discretisation. The first part becomes:
t + t
d x d y d z t x move eff moveul dt
( x yz
)
x
0
( x ) y z t
0
) (
0 0
) (
0 0
= ( eff move )0P (ul )0P move e move w ( eff move )W0 (ul )W0 move w move ww ( x ) y z t
( ) ( )
x ( x ) ( x ) ( x )
0 0 0 0
0 ( u ) ( moveul )w 0 ( u ) ( move ul )ww
= ( eff move )P move l e ( eff move )W move l w ( x ) y z t
x ( x ) x ( x )
t + t
move
y z d x d y d z t y y eff move ul dt
( x )
0
( move )0w 0
( move )0Sw
= ( eff move )nw
( move )Nw ( move )w
0 0
(u l )nw ( eff move )0sw (ul )0sw ( x ) y z t
( y )+ y ( y ) y
4.3 Discretisation of differential equations 101
t + t
( move v l )
(
x yz
)
d xd yd z t y
eff move
x
dt
0 0 0 0
0 ( v ) ( move v l )nW 0 ( v ) ( move vl )sW
= ( eff move )nw move l n ( eff move )sw move l s ( x ) y z t
( x ) y ( x ) y
The buoyancy term is integrated using an implicit scheme because the new liquid configuration
must be in equilibrium at the end of the time step:
t + t
The same holds for the Darcy term, here only shown for the case of isotropic permeability:
t + t
(
move eff move dead dead ) u dt
2
dx dy dz
+
l
( x ) y z t K 0 move2
( move eff move )w [( 0
) (
0
+ )]
0 2
=
dead w
02
dead w
u
( l )w (
0
x ) y z t
K 0 ( move )w
The external source term becomes:
t+ t
(
dx dy dz t S external dt
x y z
)
t
[J
( u )P
0
(u l )w (Feff move )P
0 0
] [J ( u )W
0
(u l )w (Feff move )W
0 0
]
+
+
[J
( u )nw
0
(u l )w (Feff move
0
) ] [ J
0
nw
( u )sw
0
(u l )w (Feff move )sw
0 0
]
[(qdiscretised )w (ul )w ] t
( p )P ( p )W 1 0
= ( move )w ( x ) y z +
0
( x ) q move
) ( move )w
0 0
) ( move )ww
0 0
u ) ( move ul )w
0 0
u ) ( moveul )ww
0 0
v ) (
0 0
v ) ( move vl )sW
0 0
( x ) y ( x ) y
( )0 [( )0 + ( dead )0w ] 2 0
+ ( move eff move )0w g x ( x ) y z move eff move w dead 0w2 (ul )w ( x ) y z
K 0 ( move )w
with
aw = a w0 S P ( x ) y z b = (a w0 ae0 a ww
0
a Nw
0
a Sw0 )(u l )0w + S C ( x ) y z
where
(Du )0P / W = y z ( move eff move )P / W ( Du )nw0 / sw = ( x ) z ( move eff move )nw / sw
( x )P / W ( y )+ /
(Du )P / W / nw / w
0
s
( move )e
0
( move )0w ( move )w
0
( move )0ww
+ ( eff move )0P (ul )0P ( eff move )W0 (ul )W0
x ( x ) x ) ( x )
(
0
u ( move ul )0w
0
u ( moveul )0ww
+ ( eff move )P0 ( move l )e
( eff move )W 0 ( move l )w
x ( x ) x ( x )
( move )Nw
0 0
( ) ( move )Sw
0 0
+ ( eff move )0nw (ul )nw ( eff move )0sw (ul )0sw move w
0 ( )
move w
( y )+ y ( y ) y
+ [(
q move
discretised 0
q )(u )
w l w
0
]x ( )
1
y z t
4.3 Discretisation of differential equations 103
+ [(
q move
q ) u
discretised 0
( )
w l w
0
]x
( )
1
y z t
( move eff move )w [( 0
) + (
0
)]
0 2
K0 ( move )w 02
S P = ( l S P external l + S
move P external move + S
move P external move w )
Now the discretised -momentum equation can be derived by simple symmetry considerations. The
y
replace indices by , by , by , by , by , by , by ,
n e e n s w w s N E E N S W W by .
S
double indices follow the same rules as single indices, i.e. nE becomes eN
x becomes y, y becomes x.
g x becomes gy.
Marangoni convection has been implemented as a momentum source term in the cells at the (rigid)
surface. The shear stress (thermal Marangoni convection only; solutal part is neglected and set to
zero)
u T N
Mara = =
z T r m2
is integrated over the cell face touching the surface and added as a momentum source term to C. As S
in Patankar formalism all source terms are integrated over the cell volume, the discretised Maran-
goni term must be divided by the cell height . The source term becomes: y
T 0
( )P ( )W T 0
1 N
S C Mara =
T r y m3
The Marangoni source term is only added to the momentum equation when at least the top two
rows of cells are at least partially molten (l of second cell row > 0.3). If this condition is not
applied and only one cell row is molten, the application of the Marangoni boundary condition
builds up a high pressure field along the surface, because no back flow is possible and no con-
vection roll can form. Convergence errors of the pressure field lead to velocity errors and create
strong concentration errors in the top surface layer.
For the treatment of different boundary types two types of velocity boundary conditions have been
implemented:
No slip condition: The tangential velocities at a wall or obstacle are set to zero. The fluid sticks
to the wall and forms a boundary layer. This boundary condition is applied to solid walls and
obstacles
u (i,1) = u ( i , m1 ) = 0
v (1,j) = v ( l1 , j ) = 0
Free slip condition: The tangential stresses at the boundary are set to zero by setting the tangen-
tial velocities equal to the first velocity inside the fluid. This boundary condition is applied to
open surfaces and symmetry lines.
u (i,1) = u (i,2)
u ( i , m1 ) = u ( i , m2 )
v ( 1, j ) = v (2,j)
v ( l1, j ) = v ( l2, j )
As no in- or out-flow has been implemented, all normal velocities are set to zero:
u (2,j) = u ( l1, j ) = 0
v (i,2) = v ( i , m1 ) = 0
units : W2
v (J T )x v v
JT = = move eff move c P eff move vl T mat eff mat T
m
(J )
T y
with
0 0 0 0
( J T )e / w = (J
e/ w
T
) x dy dz ( J T )n / s =
n/
(J
s
T
) y dx dz units : [ W]
(
move eff move c P eff move + dead eff dead c P eff dead P TP )
x y z
( 0
move eff move c P eff move + dead eff dead c P eff dead P TP
0
)
and have all ingredients for the complete discretisation of the LHS of the equation. We add the time
dependent term to the convection-diffusion-term and subtract the discretised continuity equation,
4.4, multiplied by the term (c P eff move ) 0P TP0 . The LHS becomes
[(c )
P eff move P P T (cP eff move )0P TP0 ]( move eff move )P x y z
[( ) ( dead eff dead )P ](cP eff move )P TP
dead eff dead P
0 0
0
+ x y z
+ ( dead eff dead )P (cP eff dead )P TP ( dead eff dead )P (cP eff dead )P TP 0 0 0
[(J T )e (Feff move )e (cP eff move )P TP ] [(J T )w (Feff move )w (cP eff move )P TP ]
0 0 0 0 0 0 0 0
+ t
+ [(J T )n (Feff move )n (cP eff move )P TP ] [(J T )s (Feff move )s (cP eff move )P TP ]
0 0 0 0 0 0 0 0
([ ][ ][ ][
+ (J T )e0 (Feff* move )e TP0 (J T )0w (Feff* move )w TP0 + (J T )0n (Feff* move )n TP0 (J T )0s (Feff* move )s TP0 t
0 0 0 0
])
Using new definitions for the diffusion terms
DT )e0 / w = (DT )0x dy dz = ( mat eff mat )e / w dy dz
1 0
(
e/w ( x )e / w e/w
W
DT )n / s = (DT ) y dx dz = ( mat eff mat )n / s dx dz
0 0 1 0
(
( y )n / s n / s K
units :
n/ s
(PT* )e / w / n / s =
0
units : [ ]
( DT )0e / w / n / s
Now the source terms are integrated:
t + t
( q q ) t + t
v
dx dy dz *q dt + dx dy dz [*q ( q q vvl )] dt
x y z t q l t x y z t q = move
move
t + t
+ dx dy dz t q ( q S external q ) dt
x y z
= {*q [( q q )P ( q q )0P ] x y z}+ {*q [(Fq )0e (Fq )0w + (Fq )0n (Fq )0s ] t}
ql q = move
move
[(
c P eff move TP (c P eff move )0P TP0 ]( move eff move )P x y z
)
P t
[( )0
( ) ](c )0
T
0
x y z
+ dead eff dead P dead eff dead P P eff move P P0
+ ( dead eff dead )P (c P eff dead )P TP ( dead eff dead )P (c P eff dead )P TP0 t
0
+ [(J T )e0 (Feff* move )e0 TP0 ] [(J T )0w (Feff* move )0w TP0 ]+ [( J T )0n (Feff* move )0n TP0 ] [( J T )0s (Feff* move )0s TP0 ]
= *q [( q q )P ( q q )0P ] x y z + {*q [(Fq )0e (Fq )0w + (Fq )0n (Fq )0s ]}
q l t q= move
move
+ [( q S C external q )P + ( q S P external q )P TP ] x y z
0
( 4 .7 )
q
4.3 Discretisation of differential equations 107
with
aP = a P0* S P x y z
b = (a P0 a E0 aW0 a N0 a S0 )TP0 + S C x y z
[
aE0 = (DT )e0 A (PT* )e + (Feff* move )e ,0
0
0
]
0
[
aW0 = (DT )0w A (PT* )w + (Feff* move )w ,0
0
]
A (P ) + [ (F
* 0
) ,0 ] A (P ) + [(F ) ,0 ]
0 * 0 0
aN0 = (DT )0n T n
*
eff move n aS0 = (DT )0s T s
*
eff move s
T0
SC = ( q SC external q )0P + {( dead eff dead )0P (cP eff dead )0P + [( dead eff dead )P ( dead eff dead )0P ](cP eff move )0P } P
q t
[ 0 1
+ *q ( q q )P ( q q )P ] *
[
+ q (Fq )e (Fq )w + (Fq )n (Fq )s
t q = move
0 0 0 0 1
x y z
]
q l
move
S P = ( q S P external q )P
q
The solute conservation equation is here written as a liquid concentration equation with the solid
concentrations introduced as source terms. As the lever-rule is the only microsegregation model
used here, the equations for the solid phases reduce to
C = C
i
l
i
C = C
i
l
i
t l l l l l l l l l l l l l external l
108 Chapter 4 Algorithms
with
v ~ v
( )
q dead q Dqi dead C qi + q move S external q move + q dead S external q dead
S li = v
q =
t
( q move
, + q dead ) [
C
q q
i
(
v i
v
q move q l q C ) ]
We start defining:
v (J Ci )x v i ~i v kg
J Ci =
(J Ci )y = l l vl Cl l l Dl Cl
l i
units :
m s
l
2
l
[( ) (J Ci )w (J Ci )n (J Ci ) ]
t + t v vi
t ( ) i 0 0 0 0
x y z
dx dy dz J C dt = J C e
l l
l
+
l
l s
t
with
(J ) Cl
i
0
e/ w
= (J ) dy dz
0 i
Cl
x
(J ) i
Cl
0
n/ s
= (J ) dx dz
0 i
Cl
y
units :
kg
s
e / w n/ s
t
l l P l P l l P l P
x y z
We obtain the LHS of the equation by subtracting the explicit discretised continuity equation, eq.
4.3, multiplied by (Cqi )P :
0
( l l )P [(C ) i
l P ]
(C li )P x y z (Cli )P (qdiscretised )P
0 0
ql
0
{[
+ (J Ci l
)e (Cli )P (Fl )e
0 0 0
] [(J i 0
C w
l
) (Cli )P (Fl )w ]+ [(J Ci )n (Cli )P (Fl )n ] [(J Ci )s (Cli )P (Fl ) ]}
0 0
l
0 0 0
l
0 0 0
s
t
dx dy dz (S l + l S external l )dt
i i
x y z t
= v
q = ,
t
[( q move + q dead ) C
q q
i
] (
v i
v
q move q l q C )
(DCi q )0 (C qi )0 + (DCi q )0 (C qi )0 + (DCi q )0 (C qi )0 + (DCi q )0 (C qi )0
= e E w W n N s S
t
[ ]
l l l l
q = dead (D i ) + (D i ) + (D i ) + (D i ) (C i )
0 0 0 0 0
dead C q e C q w C q n C q s q P
[ ]
l l l l
+ ( q S C external q )P + ( q S P external q )P (C q )P x y z t
i 0 i i
q pore
obst , void
4.3 Discretisation of differential equations 109
{( q q )P (Cqi )P ( q q )P (Cqi )P } x y z
0
0
q = move, dead ,
move, dead
{(Fq )e0 (Cqi )e (Fq )0w (Cqi )w + (Fq )0n (Cqi )n (Fq )0s (Cqi )s } t
0 0 0 0
q = move
move
with
(D ) i
0
= ( D~ ) i
0 y z (D )0 i
= ( D~ ) i
0 xz kg
x) y) s
units :
Cl q
e/ w
q q q
e/ w
( Cl q
n/ s
q q q
n/ s
(
e/ w n/ s
l )P
( l [( Cl P
i
) (Cli )P ] x ty z 0
+ {[( )e (Cli )P (Fl )e ] [(J Ci )w (Cli )P (Fl )w ]+ [(J Ci )n (Cli )P (Fl )n ] [(J Ci )s (Cli )P (Fl )s ]}
JC
i
l
0 0 0
l
0 0 0
l
0 0 0
l
0 0 0
= e E w W n N s S
q = dead [(D i ) + (D i ) + (D i ) + (D i ) ](C i )
l l l l
0 0 0 0 0
dead
C q e C q w l
C q n C q s q P l
l l
+ [( q S C external q )P + ( q S P external q )P (C q )P ] x y z
i i i 0
q pore
obst , void
+ {( ) [( ) ( ) ] ( ) [( ) ( ) ]}
q = move , dead ,
q
i 0
q P Cl P Cq
i
P q
0 i 0
q P Cl P Cq
i 0
P
x y
t
z
move , dead
+ {( ) [( ) ( ) ] ( ) [( ) ( ) ]+ ( ) [( ) ( ) ] ( ) [(
q = move
Fq
0
e
Cl P
i 0
Cq
i 0
e
Fq
0
w
i 0
Cl P Cq
i 0
w
Fq
0
n
i 0
Cl P Cq
i 0
n
Fq
0
s
) (C qi )s ]}
i 0
Cl P
0
( 4.8 )
move
with
a P = (a P0 S P x y z )
where
(P ) i
0
=
F )0
( l n/ e/ s / w
Cl
n/ e/ s / w
(D )0 i
Cl l
n/ e/ s / w
110 Chapter 4 Algorithms
SC = [( S
q pore
i 0
q C external q P ) ]+ {( ) [(C )
q = move , dead ,
q q P
i 0
l P ]
(C qi )P ( q q )0P (Cli )P (C qi )P [ 0 0
]}1t
obst , void move , dead
dead (D i ) + (D i ) + (D i ) + (D i ) (C i )
0 0
n
0
N
0
l s S 1
x y z
l
]
l
dead C q e C q w l
C q n C q s l
q P l l
+ {(F ) [(C )
q = move
q e
0 i 0
l P (C i 0
q e ) ] (Fq )w [(C ) 0 i 0
l P (C ) ] + (F ) [(C )
i 0
q w q n
0 i 0
l P ]
(C qi )n (Fq )0s (Cli )P (C qi )s
0
[ 0 0
]} x 1y z
move
SP = [( S
q pore
q
i
P external q P )]
obst , void
q lq
)(C ) we obtain:
i
x yz
aP0 = ( l l )P
t
l
0
[
+ (Fl )s ,
0
0 ]
SC = [( S
q pore
i 0
q C external q P + )]
{[( ) ( ) (1 ( ) )](C ) } 1t
q = move , dead ,
q q P q q P
0
lq P
0 i 0
l P
obst , void move , dead
(D ) [( ) (C ) ( ) (C ) ] (D ) [( ) (C ) ( ) (C ) ]
i 0 0 i 0
l E
0 i 0
l P +
i 0 0 i 0
l W
0 i 0
+
q = dead +
C q e
i 0
lq E
(D ) [( ) (C ) ( ) (C ) ] (D ) [( ) (C ) ( ) (C ) ]
l
0 i 0
lq P
0 i 0
C q w
l
i 0
lq W
0 i 0
lq P
0 i 0
l P
1
x yz
dead C q n
l
lq N l N lq P l P + C q s
l
lq S l S lq P l P
(F ) [(C ) ( ) (C ) ] (F ) [(C ) ( ) (C ) ]
0 i 0 0 i 0 0 i 0 0 i 0
+
q = move +
q e
0
l P
(F ) [(C ) ( ) (C ) ] (F ) [(C ) ( ) (C ) ]
i 0 x
1
lq e
y z0
l e
i 0
q w
0
l P
i 0
lq w
0
l w
i 0
move q n l P lq n l n q s l P lq s l s
SP = [( S
q pore
i
q P external q P )]
q q P
q = move , dead ,
( ) ( l q )P 1
t
obst , void move , dead
rithm using old pressure gradients to calculate the new velocities would yield large errors and mass
violations. This is why the pressure equation has been chosen to be implicit. Thereby some terms of
the velocity equation, namely the source terms like gravity and drag forces must become implicit,
too. The resulting equation is called semi-explicit, cf. section 4.3.4.
with
a nb e w (ul )nb e w + (b)e w
nb e / w
/
0
/ /
( move )e / w
(u
l )
e/w = (d )e / w = yz
(a P )e / w (a P )e / w
(a P )e / w = (a P )e w (S P )e w ( x ) y z
0
/ /
(b )e / w =
(a P )e w anb e w (ul )e w + (SC )e w ( x ) y z
0
/ /
0
/ /
nb e w /
(v l )n / s = (vl )n s + (d )n/ / s [( p )P / S ( p )N / P]
with
anb n s (vl )nb n s + (b)n s
nb n / s
/
0
/ /
( move )n / s
(v l )n / s = (d )n / s = xz
(a P )n / s (a P )n / s
(a P )n / s = (a P )n s (S P )n s x ( y ) z
0
/ /
(b )n / s =
(a P )n s anb n s (vl )n s + (SC )n
0
/ /
0
/ / s
x ( y ) z
nb n s /
(u l )e = (ul )e + (d )e [( p )P ( p )E ]
(u l )w = (u l )w + (d )w [( p )W ( p )P ]
(v l )n = (v l )n + (d )n [( p )P ( p )N ]
(v l )s = (vl )s + (d )s [( p )S ( p )P ]
a P ( p )P = a E ( p )E + aW ( p )W + a N ( p )N + a S ( p )S + b
where
aP = a E + aW + a N + a S
The adapted steps of the pressure velocity coupling procedure follows the steps of the original
SIMPLER algorithm:
0
v (ul )
1. Start with the old velocity field 0
(v l ) =
(v )0
of the last time step
l
2. Calculate the coefficients for the momentum equations and hence calculate the pressureless
velocities (ul )e w and (v ) .
/ l n/ s
3. Calculate the coefficients for the implicit pressure equation and solve it to obtain the new pres-
sure field.
4. Add the new pressure gradients to the momentum equations and solve them to obtain the new
velocities.
5. No mass source should occur and no pressure correction / velocity correction is necessary.
6. Solve the discretisation equations for the other quantities (C , T) l
7. Return to step 2.
nb
anb ( p )nb + b
*
*
( p )P = ( p )P + ( p )P (4.10)
aP
where ( p )*P is the pressure value of the last iteration and > 1 is the overrelaxation factor. The
ideal value of has to be determined empirically. A common choice is = 1.7. For the problems
presented in this work higher relaxation factors up to = 1.9 were used. This may be possible due
to the flow characteristics without dramatic changes.
The latter two points imply the knowledge of the amount of phase change occuring during the time
step. Therefore the aim of the solidification algorithm must be to find out what kind of solidification
or remelting process occurs in the special situation defined by the terms of the differential equations
and the phase diagram.
One way to treat the problem is to solve the differential equations without considering the phase
change terms and, if this leads to a solution inside a two- or three-phase-area, to determine the cor-
rect phase fractions by iterative adaption, e.g. /Laschet 98/.
For this work a different approach has been chosen. It also uses the result of the differential equa-
tions without the phase change terms, but mainly as information that helps to guess which solidifi-
cation type will most probably occur. The principle is the following:
For each cell7 the temperature resulting from the diffusive and convective heat transport terms of
the energy equation is modified by the amount of phase change occuring. The range of possible
phase changes yields two limiting temperatures: One, Tmax , for the case of maximum release of
latent heat possible and one, Tmin , for the case of maximum consumption of latent heat. If the cor-
rect solidification type has been found, the final temperature can only lie between these two tem-
peratures.
The same holds for the concentration equations. The diffusive and convective terms of the liquid
concentration equation8 yield a liquid concentration which can be modified by solidification or
remelting only in a range between Cl max for the case of maximum solute release and C l min for the
case of maximum solute consumption.
The task of the solidification algorithm is to check for this area Tmax , Tmin , Cl max , Cl min of the phase
diagram, which of the possible solidification types yields
a target point (T ,C ) inside one of the phase areas
a set of phase fractions which is reasonable for the same phase area.
For each of the possible solidification types the temperature and concentration equations have been
used together with additional equations resulting from the phase diagram, to form algebraic equa-
7
As explicit temperature and concentration equations are used, the problem can be solved for each cell separately.
8
The solid concentration equations are directly linked to it by the lever rule.
4.5 Stable phase change algorithm 115
tions which allow to directly calculate the phase fractions, temperature and concentration of the cell,
provided that the solidification type is the correct one to apply in the current situation. In these so-
lidification modules iteration is only used to find the null of the algebraic equations. A guessing
algorithm has been designed to reduce the number of trials necessary to find this correct solidifica-
tion type. The next section gives a more detailed overview of the solution strategy.
Fig. 4.3: Flow charts of the new solidification algorithm; main subroutine calcu-
late_phase_fractions and core routine solidification_type_new.
4.5 Stable phase change algorithm 117
C = C ( )
l l s
T = T (C )liq l
118 Chapter 4 Algorithms
As already stated in section 4.5.1 the solution of the differential equations for T and Cl leaves one
degree of freedom, s. The final values depend on how much solid actually forms or remelts. Solidi-
fication leads to the relaease of latent heat and results in a higher equilibrium temperature. At the
same time the rejection of solute by the newly formed solid results in a higher equilibrium liquid
concentration. Remelting leads to lower temperatures due to consumption of latent heat and at the
same time lower Cl due to remolten depleted solid. Therefore in Fig. 4.4 variation of s results in the
point (T(s), Cl(s) ) following the bold curve. The point where the curve cuts the liquidus line of the
phase diagram is described by the third equation T = T (C ) and indicates how much solid actually liq l
forms or remelts. The modules for solid state transformation and eutectic or peritectic solidification
work similarly.
For the case of primary solidification the two differential equations can be written as:
(a P
0*
S P x y z )TP = a
nb
0
nbTnb
0
+ a P0
a
nb
0
nb TP
0
+ SC x y z (energy)
nb nb
All terms are split into one part that contains the new phase fractions and in one part that does not.
An additional index T / Cl indicates if they are part of the energy or liquid concentration equation.
E.g. the constant source term of the energy equations is split into
S C = SCT + S CT ( , )
T
a nb TP + (S C + S C ( , )) x y z
T
a 0
nbTnb
0
+ a PT 0 + a PT 0 ( , ) a nb0 TP0 + (SCT + S CT ( , )) x y z
TP = nb nb
(4.12)
(a PT 0* + a TP 0* ( , ) (S PT + S PT ( , )) x y z )
The liquid concentration equation becomes
[ PC
a l
0
+ a PC 0 (
l
, ) (S PC + S PC (
l l
, )) x y z ](Cli )P
= a nb0 (Cli )nb + a PC 0 + a PC 0 (
0
l l
, ) a nb (Cli )P + (S CC + S CC (
0 0
l l
, )) x y z
nb nb
(Cli )P = nb nb
(4.13)
[a PC + a PC (
l
0 l
0
, ) (S PC + S PC ( , )) x y z ]
l l
The third and fourth equation used to calculate the phase fractions in primary or solidifica-
tion are the phase diagram relation
T [(C )P ] = TP
liq l
i
(4.14)
and the condition resulting from the solidification type, = 0 for primary solidification, and
= 0 for primary solidification.
The equation to be solved is thus:
f ( ) = T [(C ) ( )] T ( ) = 0
/ liq l
i
P / P /
(4.15)
f
=
f [( ) ] [( ) ]
/ 1 f / 0
/ ( ) ( )
/ 1 / 0
with indices and denoting the values of the last and the actual iteration, respectively. In this
0 1
way by each iteration a better value for ( )1 can be found for f [( )1 ] = 0 , i.e. the point where
/ /
( ) = ( )
f [( ) ]
/ 0
(4.16)
/ 1 / 0 f
f ( ) T (C i
) ( ) T ( )
/
=
liq l P / P /
(4.17)
/ C l
i
Cl
i
/ /
The derivation of the concentration and temperature terms is quite lengthy, but easy to do and will
be omitted here. The equations are implemented as subroutine primary_solidification and
yield / = 0, / , , for each single cell with indices i, j. Former versions of the algorithm fix
T Cl
the temperature values of the energy differential equation inside the mushy zone to the temperatures
which result from the solidification modules. The new algorithm does not need this, as solving the
energy equation using the new phase fractions yields exactly these temperatures anyway.
120 Chapter 4 Algorithms
(for = 0) has an equilibrium distribution , of the two solid phase fractions. The -phase can
l
incorporate more of the binary solute (e.g. C in Fe-C, Si in Al-Si) than the -phase. As no long
range diffusion transport is considered, the total amount of solute remains constant. If the maximum
solubility of the -phase decreases by a decrease of temperature, some solute will be rejected. As
the -phase does not want to absorb more solute than necessary, the only way out is, that a part of
the -phase absorbs rejected solute and transforms into -phase.
all
In order to calculate the quantities T, C , C , and the following equations are used:
T = ( )
T ,
C ( ,
C C ,
, )= C0 = const
T = T ( )
C
+ = alloy
C = C
The mixture concentration conservation, eq. 4.18, together with the last two relations can be used to
get an equation for :
C
1 + alloy
C = C0
+ alloy
(4.19a)
As the solid densities are assumed to be equal the equation reduces to
4.5 Stable phase change algorithm 121
1
C ( ) = C0
(4.19b)
alloy
(1 ) +
1
= 0
f [ ]
0
where the indices 0 and 1 denote the values of the last and the actual iteration, respectively, and
f T C T ( , ) T ( , )
=
C C
T alloy (1 ) C0 T ( , ) T ( , )
=
+ (4.21)
C C
[ (1 ) + alloy ]
2
In case of eutectic / peritectic solidification two independent solid phases have to be calculated. In
order to calculate the four quantities and the following equations are used:
T, Cl ,
T = ( )
T ,
Cl = Cl ( )
,
Cl = Cl eut/peri
The latter two equations are deduced from the special situation of the eutectic / peritectic plateau.
The eutectic / peritectic reaction runs at fixed temperature and with fixed liquid species con- Teut/peri
centration Cl .
eut/peri
The equations are implemented as subroutine eutec_peritec and yield , , T = T eut / peri ,C =
l
evaluating the energy and liquid concentration equations of the new time step without considering
the influence of solidification or remelting. C and C are always mixture concentrations, while the
0 *
concentrations resulting from the application of the solidification modules are always liquid con-
centrations.
Fig. 4.5 shows an example situation for the case of a peritectic system cooling down in mid of pri-
mary -solidification at the moment just before the primary solidification is finished and peritectic
4.5 Stable phase change algorithm 123
solidification starts. The straight line between the two position describes the path along which the
system will more or less evolve. This is not the necessarily the exact path, as solidification or
remelting processes can cause slight changes, but the prediction is good enough to use it as a basis
for the choice of the solidification module to apply first.
One safe statement can be made: The distance of the two positions is the maximum distance the
system can cover during this time step , as it is proportional to the local heat loss or gain. Each so-
9
lidification or remelting process happening on the way will prevent the system from reaching the
predicted position (C , T ). If, as in our example, solidification occurs during cooling, the release
* *
of latent heat will slow down the cooling process. If, during heating, remelting occurs, the con-
sumption of latent heat will equally slow down the heating process and thus equally prevent the
system from reaching the position (C , T ). * *
The question to answer now is, how far will the system procede on the path from (C , T ) to 0 0
Start from (C , T ) and apply the modules for all solidification types which occur on the path,
0 0
until a reasonable set of phase fractions, temperature and liquid concentration occurs. If all possi-
9
This is valid for non-pathological cases. A system stepping over the peritectic plateau and landing in a phase area with
extremely small heat capacity could give a different result.
124 Chapter 4 Algorithms
bilities suggested by the guessing strategy have been tried without success, assume that a melting or
solidification process has just finished and proceed to the application of the correction rules.
For the example, Fig. 4.5, the first choice on the path is to continue primary -solidification. If
(C , T ) is far enough from the peritectic plateau the release of latent heat could cause the final
* *
state (C , T ) of the system to be above the peritectic plateau, case (1). In this case the application of
the solidification module primary_solidification(delta) yields a reasonable set of phase
fractions = 0 and 0 < < . If (C , T ) is closer to the plateau, the -solidification could be
alloy
* *
finished and the peritectic solidification start, case (2). In this case application of the module pri-
mary_solidification(delta) would yield no reasonable result, i.e. a liquid phase concentra-
tion bigger than C . Application of the module eutec_peritec would give a set of phase
l eut/peri
fractions that fulfill the conditions 0 < < , 0 < < , 0 < + < . The last possi-
alloy alloy alloy
bility, case (3) is not likely to happen, because in general it takes some time steps for the peritectic
solidification to finish. It could become important at very right edge of the peritectic plateau. In this
case only application of the module primary_solidification(gamma) would yield a reason-
able set of phase fractions = 0 and 0 < < . alloy
This example is one out of a big number of possible situations. The task of the guessing algorithm
is, to classify the situation and then to propose a sequence of actions to arrive quickly at a reason-
able solution.
The system used to classify the situation is based on the two phase areas of the starting point
(C , T ), referred to as starting phase area PA , and the preliminary new point (C , T ), target
0 0 0 * *
phase area PA . For each reasonable pair of starting phase areas PA and target phase areas PA the
* 0 *
algorithm has a fixed sequence of modules to apply. After each application it checks for success. If
a reasonable result has been found, the sequence is stopped and the new phase fractions and (C , T ) l
we assume, that the system has just stepped out of one of the two-phase or three-phase areas and
finished a solidification or remelting process. In this case we proceed to the correction algorithm.
Whereas if PA is one of the two-phase areas + l, l + , + , or the three phase area eut / peri,
*
there must be an error in the system. This situation could occur very rarely due to roundoff errors or
for very pathological situations with strong convective concentration transport at points of the phase
diagram where three phase areas meet. To stop a situation like this from killing the whole calcula-
tion, we have implemented a security measure. In similar cases we assume that the solidification
type of the last time step is continued and then hope, that the correction algorithm, that is applied
afterwards manages to keep the system on track.
Table 4.1 gives a complete overview of the sequences to follow for the different combinations of
PA and PA .
0 *
4.5 Stable phase change algorithm 125
Table 4.1: Guessing algorithm to determine the solidification type. Overview of the sequences to
follow for the different combinations of PA and PA . Nomenclature: Application of solidification
0 *
PA : l
*
EXIT PA :
*
EXIT
+l p ? EXIT l+ p ? EXIT
p ? EXIT l p ? EXIT
+l p ? e/p ? p ? EXIT + _ ? EXIT
p ? EXIT _ ? EXIT
+ e/p ? p ? p ? EXIT +l p ? _ ? e/p? p ? EXIT
eut / peri e/p ? p ? p ? EXIT eut / peri e/p ? EXIT
PA :
0
PA : l +
0
PA :
*
EXIT PA : l +
*
p ? EXIT
+l p ? EXIT l p ? EXIT
l p ? EXIT p ? EXIT
+ _ ? EXIT +l p ? e/p ? p ? EXIT
_ ? EXIT + p ? e/p ? _ ? EXIT
eut /peri e/p ? p ? _ ? EXIT eut / peri p ? e/p ? EXIT
PA : + l
0
PA : +
0
PA : + l
*
p ? EXIT PA : +
*
_ ? EXIT
l p ? EXIT _ ? EXIT
p ? EXIT _ ? EXIT
p ? e/p ? EXIT eut / peri _ ? e/p ? EXIT
l+ p ? e/p ? p ? EXIT l _ ? e/p ?
+ p ? e/p ? _ ? EXIT if(C < C
l l eut/peri )p? EXIT
eut / peri p ? e/p ? EXIT if(C C
l l eut/peri )p? EXIT
l+ _ ? e/p ? p ? EXIT
+l _ ? e/p ? p ? EXIT
126 Chapter 4 Algorithms
If in contrast the heat or concentration transfer into or out of the cell causes the system to step over
one of the phase lines surrounding the + liquid area, further application of the module pri-
mary_solidification(delta) leads to inconsistencies.
1. If the system heats up, -solid transforms into liquid. The liquidus line is the place where the
phase fraction exactly becomes zero. If the system steps over the liquidus line, the solidifica-
tion module extrapolates the phase fractions, i.e. it assumes that the remelting process continues
and yields negative values for (and values > for ).
alloy l
2. Something similar happens if the system cools down and steps over the -solidus line. During
the solidification process liquid transforms into -solid. rises and becomes when
alloy
reaching the solidus line. If the system steps over the line the solidification module extrapolates
the phase fractions, i.e. it assumes that the solidification process continues and yields values
> (and negative values for ).
alloy l
3. If the system cools down and steps over the peritectic plateau the situation is different. When
reaching the plateau, the system is not yet completely solidified, < . The solidification
alloy
module just assumes that the + liquid area continues and calculates a reasonable set of 0 <
< , T and C . The inconsistency in this case is, that the point (C , T ) is now outside the +
alloy l l
diagram is not a reasonable tool to explain the behaviour of the system and to identify rules when
the peritectic reaction is finished.
In case of the two-phase processes the transformation is finished when one of the two phases has
disappeared or the temperature has dropped down to the peritectic temperature. In case of the iso-
thermal three-phase transformation other arguments have to be found. Fig. 4.7 shows the develop-
ment of phase fractions for the solidification of a binary Fe-C alloy with constant mixture concen-
tration C and uniform cooling for the cases a) C = 0.12 wt % C and b) C = 0.25 wt % C. In both
cases the system starts from the liquid and goes through primary solidification and peritectic
transformation. Due to the lower mixture concentration case a) reaches the peritectic plateau with
higher fraction of -solid and there is not enough liquid left to transform all -solid into -solid.
The system will continue by starting the solid state transformation + . At the end of the peritectic
reaction in case b) all -solid has been transformed into -solid and the system continues with pri-
mary -solidification. In both cases the system ends up in the all phase area.
One observation which is common for all parts of the solidification with liquid phase present is: The
sum of both solid fractions is continuously raising. Single solid phase fractions can in some cases
increase and decrease during the same process. On the peritectic plateau the concentrations of the
three phases involved are constant: C < C < C . The heat loss of the system causes solid phase to
l
form. If new -solid with C < C formed from the liquid, concentration conservation would make
l
the liquid concentration rise. This is not allowed. The same holds for the formation of -solid alone.
The solution is the peritectic transformation: -solid with C < C forms from the liquid. Instead of
l
increasing the concentration of the liquid the released solute is absorbed by some -solid which
transforms into -solid with C > C .
If the start of a peritectic reaction is governed by reaching the peritectic temperature, the end can be
formulated as a function of the phase fractions involved. Peritectic remelting is the easiest to de-
scribe: The peritectic reaction is finished when all -solid has disappeared and only -solid and
liquid remain. For peritectic solidification the mixture concentration decides if either the liquid is
used up first case a), the system ends up in the + area or the -solid, case b), with the system
ending up in the + liquid area.
As the equations solved in solidification module eutec_peritec all assume constant tempera-
ture and concentrations the system is not able to step over any phase lines. What happens is that the
system steps over the phase fraction values that signalise the end of the peritectic reaction. The ex-
trapolation leads to negative for peritectic remelting continued too far, negative for peritectic
solidification continued too far and mixture concentration C > C and negative for peritectic
l
a) b)
Fig. 4.7: Development of solid and liquid phases for the solidification of a binary Fe-C alloy with
constant mixture concentration C and uniform cooling. a) C = 0.12 wt % C (C < C < C ); the cri-
teria for the end of peritectic solidification / remelting are < 0 / < 0; b) C = 0.25 wt % C (C <
l
a) b)
Fig. 4.8: Development of solid and liquid phases for the solidification of a binary Al-Si alloy with
constant mixture concentration C and uniform cooling. a) hypoeutectic composition with C = 7 wt
% Si (C < C ); the criteria for the end of eutectic solidification / remelting are < 0 / < 0; b)
l l
Table 4.2: Checking rules for the successful application of the solidification modules.
primary_solidification(delta) primary_solidification(gamma)
0 alloy 0 alloy
C C
l l eut/peri C Cl l eut/peri
solid_state_trans eutec_peritec
0 alloy 0
Rule valid for both cases: T T eut/peri valid for eutectic and peritectic case
decreases by transformation into -solid, the onset of eutectic solidification causes none of the solid
fractions to decrease. If the system reaches the eutectic plateau the remaining liquid with constant
concentration C = C solidifies splitting into the two solid phases with constant concentrations C
l l eut
eutectic solidification continued too far results in negative values for . Eutectic remelting is fin-
l
ished if one of the two solid phase has disappeared completely. Thus eutectic remelting continued
too far results in negative values for if C > C and in negative values for if C < C . Fig.
l eut l eut
4.8 shows the development of phase fractions for the solidification of a binary Al-Si alloy with con-
stant mixtureconcentration C and uniform cooling for the cases a) hypoeutectic composition with C
= 7 wt % Si and b) hypereutectic composition with C = 17 wt % C.
Table 4.3: Complete set of correction rules for entering a single phase area.
primary_solidification(delta) eutec_peritec
< 0 = 0 < 0 = + ; = 0
> alloy = alloy > alloy = alloy ; = 0
solid_state_trans < 0 = + ; = 0
< 0
= 0 > alloy = alloy ; = 0
> alloy = alloy + > exited to all area via +
alloy
> =
=
alloy alloy
alloy
The physical background why this is reasonable is: The new phase fractions are needed to calculate
those terms of the energy equation which describe the amount of latent heat released or consumed
during the time step and those terms of the concentration equations that describe the amount of sol-
ute released or consumed by the newly solidifying or remelting material. As not more solid can
remelt than exists and and not more liquid can solidifiy than exists, and the terms of the differential
equation only depend on the change of phase fractions since the last time step, the correction gives
the exact result.
The cases of
the system leaving a two-phase area and entering a three-phase area,
the system leaving a three-phase area and entering a two-phase area
do not need to be considered, as in these cases the corresponding solidification module is called
anyway and no correction needed. There are only corrections necessary if the system enters a phase
area for which no solidification module exists, i.e. the single-phase areas.
As a security measure the correction rules are also applied in cases where the solidification algo-
rithm does not find a defined solution.
The correction rules for the last case to consider, if the system leaves a three-phase area and enters a
single-phase area, are difficult to formulate. They should only be needed close to the eutectic or
peritectic point or in extreme cases of convective concentration transport, i.e. horizontal movements
through the phase diagram. So far only a complex preliminary solution has been found which is not
exact, but seems to work well enough.
Table 4.3 shows the complete correction rules for entering a single phase area.
132 Chapter 4 Algorithms
calised quantity which can be quantified on each cell face separately, but inside this cast part area
the different phase fractions (l, move, move, dead, dead, pore shrink, pore hydro) are volume
averaged. For the VOF algorithm it does not matter, whether the fraction of the face which is open
to flow is a porous baffle or it consists of massive solid and an open channel. If the fraction and the
velocity on this face are the same, the flow of VOF will be the same. The characteristics of the cell
face are only mirrored by the drag term of the momentum equations.
The VOF algorithm consists of
the discretised advection equation, which uses the velocity results of the SIMPLER algorithm to
calculate the new filling states of each cell and
the surface reconstruction algorithm which uses the new filling states to determine the exact
position of the surface line in each cell.
t
( open eff open Fmove ) + ( open eff open vvl Fmove ) = Fmove
t
( closed eff closed )
1 ( ) ( )
( open )w (ul Fmove )w t ( open )
eff move w eff move e
+ (ul Fmove )e
x
(eff move )P e (
eff move )P
t
10
For details on the definition of localised and volume averaged quantities, cf. section 5.1
4.6 Algorithm for solidification of moving free surfaces 133
1 ( )
eff move s (eff move )n
+ ( )
open s (
(vl Fmove )s t open ) (vl Fmove )n (4.25)
y
( )
eff move P n (eff move )P t
the term Fmove
t
( closed eff closed ) which accounts for the phase change inside the cell during
solidification.
The equation can now be cast into the well known VOF donoracceptor scheme in which both the
fluid velocities and the Fmove values of the cell boundaries are redefined. For the right-hand cell
boundary the amount of fluxed fluid becomes
( eff move )e
( eff move )P (ul Fmove )e t = sgn[(ul )e ]min[FAD Vx + CF , FDonor ( x )Donor ] = (CVOF )e (4.26)
:
where
( eff move )e
Vx = (4.27)
( eff move )P (ul )e t
and
CF = max([ F FAD ]Vx [ F FDonor ] x Donor , 0) (4.28)
with
F = max (FDonor , FDM , 0.1) (4.29)
Here AD is a cell which can either be the acceptor cell or the donor cell, depending on the fluid con-
figuration. DM is the cell upstream of the donor cell. The redefinition of Vx has some interesting
consequences. So far the volume fluxes had to be calculated only once for each cell face. Now the
density factors require to determine the fluxes for incoming as well as for outgoing flow, i.e. a cer-
134 Chapter 4 Algorithms
tain amount of mass which leaves one cell and enters a neighbour cell can have different volumes if
the two cells have different densities (e.g. due to different temperatures). Therefore the algorithms
which assure that cells are not over-filled or over-emptied have to use different Vx for the two cells
which share each single cell face.
The redefined VOF fluxes (CVOF) are introduced into eq. 4.25. Using the definition
( open )P (Fmove )P = ( move )VOF
P (4.30)
( eff move )P
0
( closed )P ( eff closed )P ( closed )P ( eff closed )P
0 0
VOF
( move )P = ( open )P (Fmove )P = ( open )0P ( Fmove )P
0
( eff move )P ( eff move )P
+
1
x
[( open )w CVOF w ( open )e CVOF e ] + y [( open )s CVOF s ( open )n CVOF n ]
( ) ( )
1
( ) ( ) (4.31)
The new index VOF demonstrates one of the difficulties of the VOF algorithm. The new velocities
and the new phase fractions including ( move )P have been calculated using the old filling state
(Fmove )P . If these quantities are now used to calculate the new filling state (Fmove )P , the fraction of
0
mobile phases is changed again. This makes a newly defined quantity, ( move )VOF P , necessary, which
denotes the new fraction of mobile phases after application of the VOF algorithm while
( move )P signifies the new fraction before the application, but after application of the solidification
algorithm.
The total change due to the VOF algorithm,
move )VOF P ( move )P = ( open )P (Fmove )P ( open )P (Fmove )P
= ( move )VOF 0
( P
is therefore calculated as
( move )VOF
P = ( open )P (Fmove )P ( open )P (Fmove )0P
( )0 ( ) ( ) ( )0 ( )0
= ( open )0P eff move P ( open )P (Fmove )0P closed P eff closed P closed P eff closed P (4.31a)
( eff move )P ( eff move )P
+
1
x
[( open )w CVOF
( )w ( open )e (CVOF )e ] +
y
1
[( open )s CVOF
( )s ( open )n (CVOF )n ]
It is important to notice, that this equation is applied to all cells of the calculation domain, not only
to surface cells. The application also to completely filled cells is done, because the algorithm does
not know in advance, which cell will become a surface cell in the next time step.
4.6 Algorithm for solidification of moving free surfaces 135
tained from linear interpolation (or extrapolation) between the surface pressure ps, in this case only
the ambient pressure p0, and a neighbouring pressure pn inside the fluid in a direction most nearly
perpendicular to the free surface:
p = (1 ) p + p
i, j n s
(4.32)
136 Chapter 4 Algorithms
where = dc / d is the ratio of the distance between the cell centres and the distance between the
free surface and the centre of the interpolation neighbour cell, Fig. 4.9. The boundary condition is
set as a part of each solver iteration of the implicit pressure equation. The solver runs through all
free surface cells which contain some liquid. If the cell is a free surface cell, the boundary condition
is set, if not, the pressure equation is solved.
The decision, which cell is considered a free surface cell, must be taken very carefully. The condi-
tion from the all liquid VOF algorithm, /Torrey 85/, does not work any more. It says, that
A free surface or interface cell (i,j) is defined as a cell containing a nonzero value of F and having
at least one neighbouring cell (i 1, j) or (i, j 1) that contains a zero value of F.
In the new algorithm the solidifying free surface adds new situations. Fig. 4.10 shows an example
which is quite common during ingot solidification. A cell at the surface which is only partially filled
but completely solidified, cell (1), fulfills the original VOF condition, but it must not be considered
a free surface cell. Its pressure value is defined to be zero. If the condition defines free surface cells
as cells which contain at least some mobile phase and have at least one neighbouring cell which
contains no mobile phase, cell (2) would misleadingly be recognised as a free surface cell. The de-
cisive condition is: The cell must be in direct contact with void area.
Therefore the complete new condition for mobile free surface cells for which the pressure boundary
condition is set reads as follows:
A mobile free surface cell is defined as a cell containing a nonzero volume of mobile phases, and
having at least one neighbouring cell (i 1, j) or (i, j 1) that contains no mobile phases and having
at least one interface towards a neighbouring cell that is not completely covered with material.
4.6 Algorithm for solidification of moving free surfaces 137
course the bigger the volume errors are the coarser must be the limit 1 - VOF.
Especially if shrinkage pocesses are simulated very strict convergence criteria are needed to keep
volume errors small. This increases the iteration numbers considerably. To break the link between
the accuracy of the velocity calculations and the performance of the free surface algorithm the vol-
ume source of the continuity equation is inserted as an additional term into the VOF equation. Fol-
lowing eq. 4.3 the mass source in case of non-convergence of the momentum equations is
Rmass = (qdiscretised )P
0
unit : kg ]
[ (4.33)
q
If the wrong velocities are used to calculate the new filling state of an internal cell which should
always be completely filled, the fraction of mobile phases is either irregularly reduced or increased
instead of staying constant or developing according to the solidification process going on in the cell.
The cell will be partially emptied or overfilled, with sometimes dramatic consequences. Fortunately
the mass error Rmass is exactly this amount of emptying or overfilling and can be used to correct the
138 Chapter 4 Algorithms
VOF equation. Dividing it by the effective density of the mobile phases of the new time step and
normalising it to the cell volume yields the volume error
q (qdiscretised )
0
Rvol = (4.34)
( eff move )P x y z unit : [ ]
If this term is added to the RHS of the VOF equation, even a strongly non-conservative velocity
field would not lead to a violation of the filling function. As a consequence the accuracy of the ve-
locity calculations can be chosen as a function of the overall accuracy required and is not enforced
by the threat of the free surface algorithm collapsing.
The new terms of the modified VOF equation are dues to the extended features of the model:
The fraction of blocked out areas becomes time dependent due to solidification.
The fluid consists of liquid and mobile solid phases.
The density of the fluid depends on the temperature dependent density of the liquid phase and
the changing amount of mobile solid phases.
The new VOF equation has been designed to calculate the volume change of the fluid during one
time step considering all these effects. For interpretation we use the absolute form, eq. 4.31. The
new fraction of mobile phases after application of the new modified VOF algorithm is calculated
using the following terms:
( open )P (Fmove )P is the old value ( move )P at the beginning of the time step.
0 0 0
(eff move )P
0
denotes the gain of stationary solid multiplied by a density factor which accounts for the volume
loss due to phase change. For the case of solidification [( closed )P ( closed )0P ] is positive and
140 Chapter 4 Algorithms
( eff closed )P
( eff move )P is greater than one. The product is subtracted from the fraction of mobile phases.
This all means that the fluid loses more volume by solidification than the stationary solid gains.
The third and fourth term are the new versions of the classical advection terms. They consist of
one subterm for each cell face and denote the amount of mobile phases entering or leaving the
cell through this face. Even if it would lead too far to explain how the coefficients (CVOF )n / e / s / w
work in the myriads of possible cases and fluid configurations, some general remarks can help
to understand at least some of the principles. In the form
( eff move )e
C
( VOF )e =
( eff move )P (ul Fmove )e ,
t
eq. 4.26, the coefficient can be interpreted as a length denoting how far the corresponding in-
flow penetrates the cell. Assuming that the whole cell face is open to flow, this stands for a cer-
tain inflow volume. This volume can be reduced or increased by two factors: Reduced, if not
the whole cell face which is open to flow is covered by fluid, i.e. (Fmove )n / e / s / w < 1 ; reduced or
increased if the density of the incoming fluid is not identical to the density inside the cell. If the
coefficient is used to describe outflow, the density ration is equal to one due to the upwind
model used, cf. section 5.1.4. The density value on the cell face is always set to the value of the
cell centre lying upstream. As mentioned before this leads to the new situation that the volume
leaving one cell is different from the volume entering the adjacent cell. Only the transferred
masses are identical.
When the coefficients are inserted into the terms
1
x
[( ) C
open w ( VOF )w ( open )e (CVOF )e ] +
1
y
[( ) C
open s ( VOF )s ( open )n (CVOF )n ] ,
the corresponding transport terms are another two times modified: The inverse cell widths,
which belong to the directions perpendicular to the face, i.e. parallel to the corresponding stag-
gered velocity, normalize the penetration lengths to become fractions of the cell lengths. The
factors ( )open
reduce the fluxes through the faces if not the whole corresponding cell face
n/ e/ s/ w
is open to flow. A natural result of these factors is the fact that there is no flow possible through
faces which are completely covered with stationary solid or obstacle.
Fig. 4.12 shows the algorithm at work and illustrates the single steps for a case of combined solidi-
fication, shrinkage and outflow during one time step. The configuration represents a standard situa-
tion in a surface cell of a solidifying ingot including solidification front, liquid11 and void. A part of
the liquid solidifies, another part leaves the cell via the bottom cell face.
11
To simplify the representation the fluid is considered to consist entirely of liquid phase (SSM switched off).
4.6 Algorithm for solidification of moving free surfaces 141
a) b)
Fig. 4.12: The new VOF al-
gorithm at work in a typical
solidifying free surface cell;
modification of the filling
function due to; a) shrinkage
of the liquid phase; b) solidi-
fication; c) outflow.
c)
Starting point is a representation where the solid existing at the end of the last time step is (virtu-
ally) piled up in a solid block on the left while the liquid has (virtually) settled into the remaining
space. The solidification algorithm has already determined the new solid fractions and the SIM-
PLER algorithm has calculated converged temperature, concentration and velocity fields. The new
VOF algorithm uses this information to perform the following steps which modify the volume frac-
tion of the liquid:
1. The liquid cools down and shrinks due to the temperature dependence of the density. Therefore
the (virtual) liquid level drops which represents the new (intermediate) filling state (Fmove )Step
P ,
1
more volume than the solid gains. The grey area represents the volume lost by the liquid. It is
larger than the volume of the new solid layer. The mass is equal. The reduced liquid volume
virtually settles into the reduced area open to flow and determines the new intermediate filling
state (Fmove )Step
P
2
.
3. The fluxes of liquid which is advected or removed via the four cell faces are summed up, Fig.
4.12 c). In the current example the only flux which is not equal to zero is the one through the
southern cell face. The figure demonstrates the different reasons why fluxes can become zero:
The western cell face is completely covered with solid, i.e. the fraction of the face is completely
closed to the flow. The northern cell face contains no liquid at all. Even if the velocity on the
face is not necessarily equal to zero (due to the velocity boundary condition which drives the
shear stress to zero), the advective flux is zero. On the eastern cell face the momentum equation
has yielded a zero velocity. On the southern cell face the velocity is negative (i.e. outflow), the
cell face is not completely closed and it is at least partially covered by liquid. The conditions for
a non-zero advective flux are fulfilled. The outflow determines the final filling state (Fmove )P .
For the cases of cells containing no free surface, the new VOF algorithm fulfills the condition, that
the sum of all advective fluxes exactly compensates the volume losses due to shrinkage induced by
cooling and phase change.
It is important to notice, that the fractions of the cell faces which are either open to the flow, or cov-
ered with stationary solid or covered with fluid, are in most cases not identical to the corresponding
quantities which can be derived from the virtual configuration with the solid forming a vertical
block and the liquid settling into the remaining space. The real cell face fractions cannot be deter-
mined from the situation in a single cell, but also depend on the configuration in the neighbour
cells. Fig. 4.13 shows an example, how in the present case the immediate neighbourhood of the so-
lidifying surface cell determines the values of the cell face fractions. Fig. 4.13 a) shows the real
phase distribution, Fig. 4.13 b) the virtual phase distribution used by the modified VOF advection
algorithm. The bold lines denote the surface position as calculated by the surface reconstruction
algorithm using the complete non-void area. Fig. 4.13 c) shows the distribution of stationary and
mobile phases used by SoliCon. The surface reconstruction algorithm has determined the areas
which are covered by alloy. Inside these areas the volume averaged phases are equally distributed
corresponding to their relative fraction in the alloy. The numbers denote four cell faces which can
be used to demonstrate how the algorithm determines phase fractions on cell faces.
For each cell face two questions have to be answered: How much of the face is covered by alloy on
each side of the face? How is the ratio between the alloy phases on each side of the face? These
informations are used in a complex interpolation process to calculate the phase fractions on the face.
It is important to remember that the interpolations are done using the harmonic mean instead of the
arithmetic mean. As a result a cell face is already completely closed to the flow if this face is closed
4.6 Algorithm for solidification of moving free surfaces 143
a) b)
Fig. 4.13: Typical neighbourhood of a solidify-
ing free surface cell (middle); a) real phase dis-
tribution; b) virtual phase distribution used by
the modified VOF advection algorithm; bold
lines: Surface position as calculated by the sur-
face reconstruction algorithm using the complete
non-void area; c) distribution of stationary (dark
grey) and mobile (light grey) phases used by
SoliCon. The surface reconstruction algorithm
has determined the areas which are covered by
alloy. Inside these areas the volume averaged
phases are equally distributed corresponding to
their relative fraction in the alloy. The treatment
c) of the phase fractions on the numbered cell faces
is explained in the text.
in only one of the adjacent cells. The definition of the filling state is:
( move ) face
F = if ( open ) face > 0
( move ) face
( open ) face
F
( move ) face =1 if ( open ) face = 0
For more details cf. the section about the interpolation of volume fractions to cell faces, section 5.1.
For the four numbered faces the situation is as follows:
144 Chapter 4 Algorithms
C ell face 1: The easiest case has two completely filled cells as neighbours. The right cell is 100 %
liquid12, the left one contains 25 % liquid and 75 % solid. Interpolation of the liquid
fraction using the harmonic mean leads to ( l ) face = 0.4 , ( s ) face = 0.6 .
l
( open ) face = ( l ) face = 0.4 For two completely filled adjacent cells both containing at
least some liquid (Fmove ) face = 1 . Flow through this face is possible.
Cell face 2: Again both cells are completely filled. Even if the cell face is partially open from the
right side, the complete closure of the left cell dominates. The values are:
( l ) face = 0 , ( open ) = 0 , (Fmove ) face = 1 , (u l ) face = 0 .
face
Cell face 3: The mean coverage of the cell face is 0.875 (0.9 from the left, 0.85 from the right). The
alloy in the right cell consists of 73 % solid and 27 % liquid. For the right side of the
cell face this means that 73 % of 0.875 = 0.64 is covered by solid, 27 % of 0.875 =
0.24 covered by liquid. The rest of the cell face is void. Interpolation using the har-
monic mean leads to ( l ) face = 0 , while ( open ) face = 0.125 0 . The liquid fraction is
important for the solution of the momentum equation. Therefore in this case the mo-
mentum equation does not allow flow across the cell face, i.e. (u l ) face = 0 , even if it is
not completely closed. (Fmove ) face = 0 .
Cell face 4: The mean coverage of this cell face is 0.2 (0.15 from the left, 0.25 from the right).
From cell face 3 we know the phase distribution inside the alloy. For the left side of
face 4 this results in a coverage with solid of 0.15 and with liquid of 0.05. The part of
the right side which is covered by alloy is completely liquid. Harmonic mean interpo-
lation leads to the values: ( l ) face = 0.07 ( s ) face = 0.13 , the rest is void.
,
( open ) face = 0.87 . (Fmove ) face = 0.08 , flow through the face is possible.
12
Again the terms liquid and solid are used instead of mobile phases and stationary phases. Therefore
( move ) face = ( l ) face
4.8 Porosity formation algorithm 145
The solid motion adds a lot of complexity to the solidification algorithm. Many tricky situations
can arise due to the interactions between velocity fields, heat and mass flows, solidification,
split solid equations, columnar growth model, free surface algorithm etc. For some situations
numerical remedies have been implemented, but it would lead too far to describe them here. For
some situations remedies have still to be found.
max move p0 T s
pore hydro = , s (VH VH max ) t
[ ] (4.35)
p T0 t
As the hydrogen porosities push the mobile phases away, the hydrogen porosity formation rate
pore hydro = ( pore hydro )P ( pore hydro )0P is introduced into the discretised / pressure equation as a tran-
sient term which reduces the phase fraction of the mobile phases. The phase fractions on the cell
faces are assumed to be constant during the whole time step. The new source term of the pressure
equation becomes
[( move pore hydro ) eff move ] P [( move pore hydro ) eff move ] P
0
x yz
b =
+ ( dead eff dead ) P ( dead eff dead )P
0
t
(4.36)
[( move eff move )w (ul )w ( move eff move )e (ul )e ] y z
+
+ [( move eff move )s (vl )s ( move eff move )n (vl )n ] x z
146 Chapter 4 Algorithms
is calculated. Currently this pressure is set to a constant value. In later versions this can easily be
replaced by a complex function taking into account temperatures, composition, vapour pressures of
alloying elements etc.
After each iteration of the pressure solver, consisting of one sweep over all cells which contain mo-
bile phases, all cells are identified where the pressure has dropped under the critical pressure
pcrit pore shrink . The pressure in these cells is set back to pcrit pore shrink and the cell is marked for pore
formation being in progress. After each solver iteration the normalised residuum of the whole area
containing mobile phases is calculated. As due to the modified pressure there is a continuity viola-
tion in the cells with porosity formation, the marked cells are excluded from the residuum calcula-
tion. Otherwise they would prevent convergence. This procedure makes sense, because the cells
have already reached their physically correct pressure value.
When the pressure field has converged, the velocity equations are solved. As the pressure drop
which would have been necessary to ensure mass conservation has been limited in the porosity for-
mation cells, there will be more outflow than inflow in these cells.
The resulting mass loss is calculated using the implicit continuity equation on the normal grid, 4.2,
and transformed into a porosity formation rate. The fraction pore shrink of newly generated shrink-
age porosities is calculated as
q (qdiscretised )P
( pore shrink )P = ( ) x yz
(4.37)
eff move P
To get the mass balance for the cells right again, this fraction is subtracted from the new fraction of
liquid l and added to the fraction of shrinkage porosities pore shrink . This is done before solving the
differential equations for temperature and concentrations.
A neighbouring cell which contains no stationary solid is a good indicator that the current cell
contains a more or less planar front. Mobile solid is assumed to be floating with the liquid.
Therefore, if a cell where shrinkage defects are forming has at least one neighbour cell without
stationary solid, the formation of an internal shrinkage cavity is assumed. This is the model used
in the current software. The porosity type is stored in an integer array.
Another possible technique is to use the local composition and temperature distribution to cal-
culate liquid solid interval, temperature gradient and thus the local extension of the mushy zone.
If the mushy zone is smaller than the local grid size, the formation of a cavity can be assumed.
5 Implementation Details
In this chapter some important details on the implementation of models and algorithms will be
given.
Inside a cell localised phase fractions behave like a massive block. Each parts of a cell is either
completely covered with the phase or it is not. It can be clearly determined where inside the cell the
phase is located and which parts of which cell faces are covered. Volume averaged phase fractions
behave more like a gas. They fill all room they are allowed to fill, but mix homogeneously with all
other volume averaged phase fractions.
Fig. 5.1 shows a situation where localised and averaged phase fractions are both present in a single
cell. The localised phase fractions are obst, void and vofm. They occupy different parts of the cell.
The phase fractions on the cell faces can be higher or lower than the corresponding value of the
whole cell. e.g. half of the cell is filled by obstacle phase, but as the obstacle is localised in the left
half of the cell, the fraction of obstacle on the left cell face is 1 and on the right face 0.
The localised phase fraction vofm has been defined for easier handling of the volume averaged phase
fractions. It defines the area of the cast part including porosities and is defined as
5.1 Interpolation of volume fractions to cell faces 149
The reason why porosities are included is the special use in free surface calculations. These calcu-
lations are not alloy based but volume based. A cell which is in the middle of the fluid / solid, but
which contains porosities has to be considered as a full cell, not a surface cell. Therefore the gov-
erning volume fraction is vofm.
vofm defines the averaging volume of all volume averaged phase fractions. The textured area in Fig.
5.1 denotes the cast part which consists of several homogeneously distributed phases (e.g. and
or and l). If there are only two phases with equal phase phase fractions of 0.15, their values on
the cell faces are 0.3 each on the eastern face and 0.25 each on the southern face.
For reasons of convenience another phase fraction has been defined, this time a volume averaged
one. alloy is defined as the volume fraction covered by alloy phased, not including porosities. i.e.
It is used for all thermodynamic calculations like phase transformations which only take into ac-
count the alloy phases.
Fig. 5.2: Interpolation of localised phase fraction vofm for the case that the surface segments of
neighbour cells do not form a continuous line (which is normal, but especially the case if the cur-
vature of the surface is high or the grid resolution relatively low). The figure shows the definitions
of the quantities for the two vertical (west and east) faces of cell i.
but a series of surface segments which not necessary meet at the cell face. This regularly leads to
cases where the right and the left or the top and the bottom value of the filling state on a cell face
are different. Fig. 5.2 shows a typical example of several adjacent surface cells with different area
fractions on the cell faces between them. The task is to interpolate these values. One special diffi-
culty is the treatment of cells which have obstacle cells as neighbours. The procedure currently ap-
plied in SoliCon2.0 works as follows; it allows to use complete obstacle cells ( obst = 1) only:
For all cell faces with no obstacle neighbors vofm is the average of the two area fractions
For all cell faces with two obstacle neighbors vofm = 0
For all cell faces with exactly one obstacle neighbor vofm takes the area fraction of the non ob-
stacle cell. In this case this would be the contact area between the obstacle and the cast part.
holds, the same must hold for the area fractions. Fig. 5.3 shows the situation in a free surface cell
with non horizontal surface line. The area fractions vofm on the left cell face and vofm
+
on the right
face are not identical to the volume fraction vofm of the whole cell. Therefore the volume averaged
quantities q (here only two phases l and chosen as examples) have to be adapted and yield
, , , (in 1D model) with
+ +
l l
5.1 Interpolation of volume fractions to cell faces 151
Fig. 5.3: Projection of volume averaged phase fractions l and s (example) to area fractions of the
localised phase vofm.
+ q + q
q = vofm q = vofm
vofm vofm
The interpolation between the area fractions of two neighbour cells requires some additional care.
As the phase fractions on cell faces are mainly used for the calculation of fluid flow, the interpola-
tion cannot be done by simple averaging. This becomes obvious if a simple example of two com-
pletely filled (vofm = 1) neighbour cells is considered: If the cell left of the cell face is completely
solidified and the cell right of it is completely liquid, simple averaging would yield a liquid fraction
l = 0.5 on the cell face and pretend the cell face being open to flow. Therefore an interpolation
formula is needed which yields zero values of the mobile phases and porosities on the cell face if
either of the two cells is completely closed.
An interpolation scheme with this property is the harmonic mean. Therefore the following relations
are used to calculate the phase fractions on the cell faces:
2 x q+ (i ) q (i + 1)
( ) ( ) =
*
q e i
x(i + 1) q+ (i ) + x(i ) q (i + 1)
for q = l , move, move, pore hydro, pore shrink
( ) ( ) = 1
* [ ][
2 x 1 q+ (i ) 1 q (i + 1) ]
q e i
x (i + 1)[1 q+ (i )]+ x (i )[1 q (i + 1)]
for q = dead, dead
152 Chapter 5 Implementation Details
The asterisk signifies that even after applying these formulas the work is not done yet. We must still
fulfill the condition that the sum of all area fractions of volume averaged quantities must be equal to
the area fraction of vofm:
!
( vofm )e = ( q )e with q = l , move, move, dead, dead, pore hydro, pore shrink
q
This would already be the case if simple mean value averagin had been used, but it does not hold in
case of the averaging using the harmonic mean. Therefore the area fractions have to be normalised
again:
( q )e
*
( q )e = ( vofm )e
p ( p )e *
t new =
min dt max ,
adaptfactor 0 5
. x
(5.1)
( move u l )max
5.3 Convergence control structures 153
where (move ul ) max is the highest staggered superficial velocity of the whole calculation domain and
x is the cell width of the two adjacent cells. move is the averaged fraction of mobile phases in the
two adjacent cells. To prevent that time steps have to be redone too often, the factor adaptfac-
tor has been introduced. It should be chosen to be smaller than 1. If this is the case, the time step
width is reduced before the stability criterion has been violated. Therefore the criterion will never
be violated if the rise of velocity is not too fast. The fact that the criterion is checked after each time
step has another important effect: It allows to raise the time step again when the velocities are de-
creasing.
Fig. 5.4 shows a typical time step width evolution during a welding simulation with Marangoni
convection. The melting starts after 0.28 s. When the maximum flow velocity becomes so high that
it violates the stability criterion (0.45 s), the time step width is reduced. After about 0.65 s the ac-
celeration of the flow decreases, the decrease of the time step width slows down. After 1.5 s the
laser source is switched off. Viscosity damps the velocity of the flow quickly and the initial time
step width is restored after 1.62 s. The single spots are time steps with shorter step width which
were inserted to exactly hit saving times (e.g. time step width 0.0450 s; last time step: 1.2380 s.;
next time step to be saved: 1.24 s, time step width of inserted step: 0.002 s).
nb a nb nb + b a P P
cells
<R (5.2)
a P P
cells
A higher-ranking criterion is the maximum allowed mass source. Even if both other criteria are ful-
filled at the end of a SIMPLER iteration, a violation of this one means non-convergence.
5.4 Parallelisation
Typical welding simulations without moving free surfaces take up to 10 days of calculation time on
a 180 MHz R10000 processor. If free surfaces are used, the high velocities of Marangoni flow re-
quire a dramatical decrease of the time step due to the restrictions of the VOF model. For casting
simulations the grid refinements and the increasing complexity of the coupled models increase the
calculation times.
This is why the software was parallelised. The decision to use the shared memory parallelisation
standard OpenMP was based on a detailed analysis of the serial code, typical applications and the
hardware available. The main reasons for the choice were:
OpenMP allows an incremental parallelisation, loop by loop, subroutine by subroutine, of ex-
isting code in an easy way.
As OpenMP mainly uses compiler directives that are ignored by compilers without OpenMP
capabilitiy the parallelisation does not affect the performance of the serial code.
The compiler directives do not reduce too much the transparency, readability and maintainabil-
ity of the complex code.
High capacities of shared memory at RWTH Aachen.
The parallelisation required the following steps:
Migration to Fortran 90 (to save time by dynamical allocation of the main arrays)
Further optimisation of the serial code
Isolation of subroutines by reducing global variables and restructuring of the code
Incremental parallelisation
5.4 Parallelisation 155
a) b)
Fig. 5.5: Performance of the parallelisation using the shared memory standard OpenMP; a) speed-
up; b) parallelisation efficiency; the speedup is defined as the factor how much faster a parallel pro-
gram runs if more than one processor is used. The ideal speed-up on 4 processors would be 4. This
would correspond to a parallelisation efficiency of 1 (or 100 %).
The main body of the program consists of two nested loops, one for the time steps and one for the
pressure / velocity coupling and for solving the non-linear equations, Fig. 4.1. None of these can be
parallelised using OpenMP parallel do loops because the results of the former time step / iteration
are needed. Thus the parallelisation effort had to concentrate on the subroutines inside the inner-
most loop which use approximately 80 % of the total calculation time. Tests showed that the most
effective parallelisation structure is the use of parallel do loops. Parallel sections turned out to be
less effective. In addition they reduce the scalability of the code. Some parts are not worth while
parallelising because their portion of the total execution time are too small. Some other parts cannot
be parallelised easily because they have to be executed in a special order. Special care has to be
taken in parallelising the iterative solvers. The parallel execution can be very effective, but the links
between neighbour cells can lead to results which depend on run time behaviour. For production
runs this is not a problem, because the converged results should be the same inside the error toler-
ances, but especially for debugging, when a deterministic behaviour is important, these loop paral-
lelisations should be switched off.
Due to the fine grained structure of the problems (domain sizes of typically 1023 to 4060 cells)
the parallelisation efficiency is not very high. Fig. 5.5. For several reasons the efficiency values can
only be given for special cases. The load distribution between different modules changes during the
calculation. At the beginning of a welding simulation, for example, there is no flow calculation and
no phase change. When the material melts and the weld pool grows, the amount of work required
for the velocity and pressure equations and for the phase change routines increase dramatically. A
casting problem instead starts with a completely liquid area, but without phase change. After some
time it goes through a stage with increasing number of solidifying cells, but decreasing flow area.
156 Chapter 5 Implementation Details
In addition, a shrinkage problem needs much more time for the pressure / velocity coupling than a
welding problem.
The analytical solution meeting the initial and boundary conditions is:
x
T = T0 + (TS T0 ) erfc (/Baehr 98/, p. 158)
2 at
The problem was simulated in pure 1-D heat conduction mode (all special features like free sur-
faces, fluid flow etc. switched off) and using a fictitious material with material parameters listed in
Table 6.1. The calculation domain is an equidistant 500 by 5 grid of L = 1 m by H = 0.02 m, which
makes a spatial resolution of x = 0,002 m Time resolution of the simulation is t = 0.01 s. The
initial conditions are listed in Table 6.2.
Fig. 6.1 a) compares the resulting temperature profiles T(x,tn ) along the bar for different times tn
with the analytical solution, Fig. 6.1 b) does the same for the resulting time evolution of tempera-
tures T(xn ,t ) for different positions xn.
The curves, especially the time evolutions T(xn ,t ) for different positions xn agree very well. The
temperature profiles T(x,tn ) along the bar for different times tn show very small deviations, which
may be due to the steep temperature gradients at x = 0 and could probably be corrected by using an
even finer grid.
6.1 Test problems 159
a)
b)
Fig. 6.1: Comparison of simulated a) temperature profiles T(x,tn ) along the bar for different times tn
and b) time evolution of temperatures T(xn ,t ) for different positions xn with the analytical solution
from /Baehr 98/. In order to visualize simulation and analytical solution in the same diagram the
choice of simulation data points was guided by the rule to have a minumum distance from each
other along the curves. So the time steps shown vary from curve to curve.
160 Chapter 6 Applications: Simulations and experiments, Casting processes
Table 6.1: Material parameters used for the heat conduction benchmark
= 1000 kg/m3 constant density of the fluid phase (whole simulation in fluid state)
cP = 1000 J/(kgK) constant specific heat capacity of the fluid phase
= 100 W/(Km) constant heat conductivity of the fluid phase
() = 0.0001 m2/s thermal diffusivity (results from other quantities)
Table 6.2: Initial conditions used for the heat conduction benchmark
T0 = 1600 K Initial temperature of the calculation domain
TS = 2000 K Fixed boundary temperature at one end of the domain
The gravitational acceleration g = 9.81 m/s2 starts the fluid sloshing back and forth across the tank,
and the surface shape always remains proportional to the cosine term in the initial condition. The
slosh period is given by
P= 2
= 0.374s
g tanh H
L L
Fig. 6.2: Sloshing tank benchmark (/Torrey 85/). Velocity vectors and free surface configuration
for some time time steps during the first two oscillation periods t = 0 0.75 s. The different shades
of grey show the filling state of the corresponding cell 0 < Fmove < 1.
Fig 6.2 shows plots of the velocity vectors and free surface configuration for some time steps during
the first two oscillation periods. The oscillation period of the simulated wave determined from two
oscillations is P = 0.375 s. The shape after two oscillations is little disturbed. If the model is applied
to casting problems the small deviations are tolerable as the dynamics is much smaller than in the
sloshing tank problem. For application to welding problems which are highly dynamic, the algo-
162 Chapter 6 Applications: Simulations and experiments, Casting processes
rithm needs to be refined or replaced by an algorithm based on the more modern PLIC or CIP
method.
cooled from one side by setting the boundary to the fixed temperature . One cell was chosen for
Tfix
the observation of the development of temperatures, phase fractions and concentrations. The calcu-
lation was continued until short after the observed cell had completely solidified.
Aim of the procedure was to test if the algorithm is able to model equilibrium solidification of
eutectic and peritectic alloys as well as pure substances. The following sections shows the results
for the three cases using the eutectic system Al-Si, the peritectic system Fe-C and the (technically)
pure substance Al99.99.
wt % (right part of peritectic plateau, > ) were solidified. Fig. 6.5 shows the evolution of
C C peri
temperatures, phase fractions and concentrations. The curves exactly reproduce the evolution to be
expected from the analytical solution.
6.1 Test problems 163
the description of the influence of the macroscopic shrinkage cavity on the final macrosegregation
pattern.
6.2.1.1 Experiment
A 45 kg ingot of unalloyed13 multicomponent steel GS60 (DIN 1681, composition in Table 6.3) was
cast into a sand mould with insulating material at the two ends to obtain quasi-2D conditions in the
cross section. Fig. 6.7 a) shows the geometry which consists of a simple rectangular ingot with rec-
tangular riser, Fig. 6.7 b) a polished cut image of the resulting cast part. The final macrosegregation
pattern was determined by spark spectrum analysis. The results for carbon are shown in Fig. 6.8 c).
Table 6.3:
Alloying element C Mn Si S P Ni Cr Cu Mo
wt % 0.420 0.800 0.600 0.014 0.015 0.100 0.100 0.100 0.020
a) b)
Fig. 6.7: a) Geometry of cast part; the figure shows which faces of the ingot are in contact with the
sand mould (white + bottom), insulation (dark grey) and air (light grey); b) polished cut image of
the resulting cast part.
13
total sum of all alloying elements < 5 wt%
6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 167
6.2.1.2 Simulation
2D-CVM simulations on a grid of 40x60 cells were performed and the results compared to the ex-
periment. Fig. 6.8 shows the final macrosegregation pattern of two calculations using an isotropic,
Fig. 6.8 a), and an anisotropic permeability model, Fig. 6.8 b). When compared to the experimental
pattern, Fig. 6.8 c), similarities and differences can be observed.
The calculation with isotropic model yields very weak macrosegregations. Only the area around the
internal edge and at the very top of the riser show strong positive segregations. The calculation re-
sults obtained with the anisotropic model manages to reproduce the order of magnitude of the mac-
rosegregations, the strong negative segregations in the edges, and the positive segregated area in the
left part, which is probably a result of a segregation channel.
As the simulations do not consider the movement of the free surface, neither of them is able to re-
produce the correct position of the last point to solidify which is positioned beneath the deep macro-
scopic shrinkage cavity. The reason why the correct prediction of this point is so important, is the
fact that it contains strong positive macrosegregations and is not situated in the riser, but inside the
proper cast part. The maximum carbon concentration deviation of +129 % reveals that at the
boundary between riser and cast part there is an area which almost consists of cast iron.
SSM: 0 mv dd = 0.3
mv dd = 0.2
CGM: RCGM = 0.01 m
CGM = 0.6
was one cell column open to flow). In order to demonstrate the formation process of the shrinkage
cavity higher constant solid densities of 7900 and 8344 kg/m3 were chosen for the calculations in
this section. In the future parameter studies with different grid configurations, solid densities and
SSM parameters have to be done.
t = 800 s t = 1200 s
vmax = 0.00002 m/s 0.0139 m/s vmax = 0.00081 m/s 0.0049 m/s
Fig. 6.12: Importance of thermosolutal convection for the formation of deep shrinkage cavities;
Comparison of calculations with SSM and CGM for case with shrinkage flow only (left) and
shrinkage flow with thermosolutal convection (right); solid fraction and velocities; the two maxi-
mum velocity values are for the left and right half of the picture.
frozen too early. Even if the SSM is switched on, and the solid can move, the shrinkage flow is too
slow to remove the mobile solid from the narrowed feeding path. There is a high solid fraction close
to the surface while the bulk liquid underneath is almost free of solid. Before the bottleneck has
been passed by the surface, the fraction solid exceeds the limit, the solid starts to become stationary,
and the feeding path is blocked. If thermosolutal convection is switched on, right half of the pictures
in Fig. 6.12, the mobile solid is removed from the bottleneck by flow and sedimentation. The
total solid fraction never reaches the critical fraction when the first solid becomes stationary, and
the surface can pass the constricted opening. In addition the mixing of the fluid creates a more uni-
form temperature field in the area open to flow. This leads to lower temperatures in the bulk liquid
and higher temperatures in the riser, resulting in retarded solidification, Fig. 6.13.
6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 173
6.2.2.5 Complete calculation with SSM, CGM, shrinkage and thermosolutal convection
When the SSM is completed by the CGM and shrinkage and thermosolutal flow is considered, the
simulation of a deep shrinkage cavity becomes possible. Fig. 6.14 shows the results of a calculation
with all models apart from the porosity formation model switched on. The calculation considers
macroscopic diffusion in the liquid and an isotropic permeability model (K0 = 810-10 m2).
The left half of each picture shows the development of the velocity field and the total fraction of
solid. The right half shows the formation of the macrosegregation pattern, described by the percent-
age of carbon concentration deviation from the initial concentration.
The development of the flow field shows three distinct phases:
1. 20 100 s: Thermal convection leads to two downward flows along the cooled walls. The top
one enters the mould cavity of the cast part and disturbs the formation of a real convection cell
in the bottom part
2. 100 600 s: The sedimentation of cool melt and mobile solid leads to oscillating flows in the
bottom part
3. 600 4620 s: Different influences lead to complex flow patterns. The thermal downward flow
along the areas of stationary solid (e.g. 700 s) is often superposed by the flows due to sedimen-
tation of mobile solid from the top part of the riser. These solid-containing flows which are
comparatively strong due to the high density difference compared with the liquid create quickly
changing, complex velocity distributions (e.g. 700 s, 1400 s, 2000 s). The upward flow close to
the internal edge (500 700 s) could be due to the rise of enriched liquid from the mushy zone
or just a compensating flow for the strong thermal downward flow which has been dislocated
from the wall by the solid area above.
174 Chapter 6 Applications: Simulations and experiments, Casting processes
The shape of the external shrinkage cavity is governed by the heat extraction distribution on the
riser wall. There is strong cooling at the top and less cooling at the bottom part, because the internal
edge of the sand mould is heated from two sides, from the riser and the cast part. Therefore it can-
not remove as much heat as the top part of the riser or the left wall of the cast part and remains hot
for a long time (Sandkanteneffekt). At the beginning of solidification columnar solid grows from
the wall, while the shrinkage flow slowly lowers the level of the liquid surface. As the shrinkage
flow remains about constant but the cross section of the free surface decreases, the speed at which
the surface drops increases, creating a curved solid surface. When the liquid level has passed the
point of highest cooling, the solid wall draws back again, the liquid level sinks slower. The end
phase is characterised by the solid closing down from all sides. The solid fraction everywhere
reaches the point of all solid being stationary and the surface is deadlocked. In this stage of the
simulation the porosity formation model would start working, but it was not yet available at the
time.
The final size of the shrinkage cavity is too big compared to the experiment, Fig. 6.15. One of the
current problems is the low grid resolution. Even if the shrinkage cavity looks very wide, the effec-
tive width of one half of the geometry (due to the symmetric calculation) at the bottleneck is two
cells. This is one reason why the high value for the constant solid density has been chosen. The re-
sulting higher shrinkage speed makes the liquid level pass the narrow part before solidification can
close it down to one cell, which causes big numerical trouble. When calculations with higher grid
resolution, at least in the riser, become possible, the constant solid density can be reduced and a less
oversized shrinkage cavity will form.
The observation of the solid fraction evolution shows some interesting effects. At the very begin-
ning of solidification all solid which forms grows as columnar solid fixed to the wall, as postulated
by the CGM. When the solid shell has grown so far that the wall adhesion rule of the CGM is no
longer valid, the solid splitting rules of the SSM start to apply. According to the current model pa-
rameters, solid fractions < 0.2 are considered to be completely mobile, solid fractions > 0.4 com-
pletely stationary. New solid forming close to the columnar front becomes mobile and drops to-
gether with the thermal flow, accelerating it due to its high density (400 500 s). The sedimentation
of mobile solid forms a fraction solid gradient in the open flow area (500 1000 s). This is a crucial
effect concerning the formation of the deep shrinkage cavity, because it keeps the solid fraction in
the bottleneck low enough to prevent the formation of stationary solid and the blocking of the
moving free surface, cf. section 6.2.2.4. The end of solidification is little spectacular. In agreement
with experimental observations the last point to solidify is located beneath the external shrinkage
cavity.
The solute redistribution is closely linked to the flow field and solid fraction evolution. The princi-
pal effect which can be observed during the whole solidification process is the normal segregation,
i.e. the depletion of the mushy zone due to thermosolutal flows passing it. The flows remove en-
riched interdendritic liquid, which leads to successive enrichment of the open flow area.
6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 175
Fig. 6.14 (part 3): Shrinkage cavity formation of Fe-0.42wt%C ingot; simulation with SSM, CGM,
shrinkage and thermosolutal convection; development of solid fraction with flow field (left half)
and macrosegregation pattern (right half).
In addition to this governing mechanism some minor effects can be observed. One of them is the
formation of a large positive segregation channel due to shrinkage flow. While solidification in the
edges of the cast part continues, a horizontal channel filled with enriched melt remains open for
quite a long time (1000 2000 s). The channel is self-sustaining, because the slow but steady trans-
port of enriched melt keeps the solidus temperature low and prevents the channel from freezing. It
has formed in the middle because this was the hottest part, had the least solid fraction and therefore
the easiest flow conditions. Another effect is the formation of enriched areas close to the free
178 Chapter 6 Applications: Simulations and experiments, Casting processes
carbon concentration
deviation [%]
surface due to the sedimentation of depleted mobile solid (1400 s). It is an interesting question, how
this works in a model which does not explicitly allow relative movement between liquid and mobil
solid phases. It must be an effect of the volume averaging model.
The temperature field is not spectacular, and a detailed description can be omitted without loosing
important information.
Fig. 6.15 compares the final shape of the shrinkage cavity and the final macrosegregation pattern
with the results from /Bhmer 97/, using their concentration deviation scale. In some aspects both
the shape of the shrinkage cavity and the final macrosegregation pattern are in accordance with the
experimental data, in others they show large discrepancies. The depth of the cavity is comparable as
well as the position of the last point to solidify.
The discrepancies can be explained by different arguments:
The current model assumes constant solid density s. The choice of this density determines the
amount and speed of shrinkage and by that the size and shape of the shrinkage cavity.
The cooling has been realized by applying MAGMAsoft cooling curves as boundary conditions.
The MAGMAsoft calculation did not consider fluid flow and convective heat transfer.
6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 179
A parameter study has to be done to investigate the influence of the parameters of the SSM and
the CGM on the shape of the shrinkage cavity.
An anisotropic permeability model instead of the iso tropic one used so far will have great
influence on the final macrosegregation pattern.
Fig. 6.16: Shrinkage cavity formation of Fe-0.42wt%C ingot; comparison of solid fractions (left
half), velocity fields and macrosegregations (right half) after 540 s for grid resolutions a) 10x23
cells, b) 20x46 cells, c) 40x96 cells
When phase fractions have values of about 10-5, and there is still in- or outflow and free surface
movement, sometimes the solidification algorithm cannot cope any longer with the situation. A
temporary workaround has been to increase the filling state limit under which a cell is consid-
ered to be empty, and therefore is made empty. Consequence of this workaround are small mass
losses. The calculations presented in this chapter have been made using a maximum limit of 10-4
The mass losses due to this error were acceptable. Another reason for this crash may be small
surface waves, when a cell is emptied and refilled in turns. The problem gets worse when the
grid size diminishes, because the surface waves become bigger compared to the cell size. In
these cases even higher limits could not finally prevent the crash.
A wrong definiton which cells are considered to be free surface cells and which are moving free
surface cells led to heavier mass losses (up to 6 %) in some special cases of surface configura-
tions. The proper problem has been eliminated, but the changes had difficult consequences and
the new problems have not yet been solved.
6.3.1 Experiment
A cylindrical Al7wt%Si ingot (H = 107 mm, R = 40 mm) was cast in a cast iron chill standing
freely on a steel plate, Fig. 6.19 a). The temperature in the furnace just before the casting was
990 K. Fig. 6.20 e) shows the resulting surface geometry after total solidification. Beneath the mid-
dle of the shrinkage cavity there are two hot crackings which are not considered in the current
model, and an area of shrinkage porosities. The cavity in the middle of the block is due to gas for-
mation at one of the thermo couples.
6.3.2 Simulation
For the simulation of the process an early version of the free surface model was used which is part
of the software package EL2D. Due to problems with the free surface boundary conditions concen-
tration conservation is not fulfilled. Fig. 6.19 b) shows the calculation grid and the boundary condi-
tions. The temperature boundary condition at the contact zone between casting and chill are time
dependent temperatures taken from a MAGMAsoft solidification calculation (without mould fill-
ing). The heat transfer coefficients for the MAGMAsoft calculation were adapted to yield a total
solidification time similar to the one which was experimentally observed. Table 6.4 shows some of
the calculation parameters used for this simulation. All other physical data are those listed in App.
9.1.3.1. The initial temperature was set to 900 K, taking into account the heat loss of the alloy on its
way from the furnace to the chill and during the mould filling. The time stepwidth was fixed and set
to 0.1 s with a total solidification time of 120 s. The calculation took about 24 hours using an
R10000 Processor.
6.3 Formation of shrinkage cavity in Al-7wt%Si ingot 183
b)
a)
Fig. 6.19: a) Experimental setup for the casting of a cylindrical Al7wt%Si ingot. b) Corre-
sponding calculation domain and boundary conditions for the numerical simulation.
The calculation results presented in Fig. 6.20 show the typical behaviour of an alloy with long
freezing range. The mushy zone grows quickly and fills the whole casting after 60 s. After 80 s the
shrinkage cavity at the free surface is fully developed and the calculated shape shows a promising
agreement with the experimental results, Fig. 6.20 e). The surface slope depends on the remaining
surface area open to the liquid. With beginning solidification the open surface area is reduced and
the sinking velocity of the surface increases. This mechanism results in a flat outer area and a steep
slope in the centre. These characteristics of the real casting are reproduced by the simulation very
well.
184 Chapter 6 Applications: Simulations and experiments, Casting processes
a) 10 s
b) 30 s
d) 80 s
c) 60 s
vmax (t = 60 s) = 0,0025 m/s
Temperature [K]
e) Experiment
728 745 762 779 796 813 830 846 864 881 900
Fig. 6.20: Formation of shrinkage cavity dur-
Solid fraction s [] ing solidification of Al-7wt%Si ingot; a) d)
simulated temperature, solid fraction, veloci-
0.05 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.95 1 ties + mushy zone contours (0.020.98) and
Si concentration deviation [%] deviation of Si-concentration from initial con-
centration (from left to right); e) Polished cut
image of real cast part.
-4,5 3,5 2,5 1,5 0,5 0,5 1,5 2,5 3,5 4,5 5,5
6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys 185
The time evolution of the solute redistribution shows interesting effects. After 10 s a small area in
the bottom edge has completely solidified. The solid cells are slightly enriched. A certain area of the
mushy zone in front of the solidified cells is slightly depleted. This effect of inverse segregation is
caused by the shrinkage induced flow. In principle the initial concentration C0 (mass fraction) inside
a cell is not changed by density changes. If a cell contracts, the mass proportions remain the same.
If a constant volume is considered instead (cells in a rigid grid), the volume losses have to be com-
pensated by a flux from the neighbour cells. In the considered case the neighbour cells are partially
solidified and contain (in the lever-rule model) enriched liquid of concentration Cl > C0. Therefore
the cell gains enriched liquid and will also be enriched after total solidification. In the neighbour
cells the inverse mechanism works: They lose highly enriched liquid and gain liquid from cells
which have a smaller solid fraction and thus a smaller liquid concentration down to C0. With pro-
ceeding solidification the completely solidified cells get less enriched. The quantitative correctness
of the predicted macrosegregations has to be tested. Due to numerical problems the presented cal-
culation does not yet consider thermosolutal convection, temperature dependent liquid densitiy and
macroscopic diffusion in the liquid. When these models are added, the resulting species distribution
will be changed significantly.
New porosity only forms at exactly one point, more exactly at one height of the cast part. Only
when all liquid at this height has either solidified or dropped out of the cell will porosity forma-
tion continue at lower levels. This is an effect of gravity. When one point reaches the critical
pressure for pore formation, the pressure is fixed to this point and porosity formation starts.
Everywhere beneath this point total pressure is higher due to the hydrostatic pressure and no po-
rosities can form. Shrinkage will continue to pull the liquid out of the area while the emptied
shrinkage porosity area remains fixed to the critical pressure. This is why the porosities grow
downwards. Fig. 6.22 shows the development of the main shrinkage porosity area in two repre-
sentations, once in a grey scale and once in a stochastic porosity visualisation.
At the time when the feeding path is closed, the pressure in the closed off area drops almost
instantaneously, Fig. 6.23 a). The rapid pressure drop is difficult to manage for the iterative
pressure solver. The iteration numbers necessary to reach convergence increase rapidly, Fig.
6.23 b). When the critical pressure for pore formation pcrit pore shrink is reached by the first cell, its
pressure is fixed to this value. As now the pressure solver does no longer need to pull the pres-
sure down, the iteration numbers drop instantaneously to a level which is more than a factor of
100 lower than the peak. The same calculation without shrinkage porosity formation model
would yield a pressure which would drop to infinity hitting the iteration number limit in each
time step.
fraction of
shrinkage
porosities
[]
Fig. 6.22: Prediction of shrinkage porosity formation in cast Al-7wt%Si ingot; development of main
shrinkage porosity area; volume fraction of shrinkage porosities in grey scale representation (top)
and porosity representation (bottom); the area of black dots is proportional to the area of shrinkage
porosities; proportionality factor: 3; as the current model does not give any information about pore
numbers, the size of the single dots has been chosen arbitrarily.
188 Chapter 6 Applications: Simulations and experiments, Casting processes
Fig. 6.23: Initial phase of shrinkage porosity formation; a) pressure development in a cell of the hot
spot, three cells beneath the cell where shrinkage porosity formation starts; b) development of pres-
sure solver iteration numbers necessary to reach convergence.
a) c)
Fig. 6.25 shows a cut polished section of the Al-Si ingot in different magnifications. Due to the
preliminary nature of the experiment the results differ in many points, but there are some promising
accordances, too, which motivate further experimental research. The position of the porosity area
compared to the chill position is lower than expected. This may be a result of the different heat ex-
traction conditions. The size of the porosity area is smaller than expected. For the horizontal size
this can be explained by the low grid resolution, but this explanation does not hold for the differ-
ence in vertical direction. It may be a consequence of the primitive pore formation models. The
shape of the porosity area indeed compares well. This indicates that the model describes the forma-
tion mechanism of the porosity area correctly.
fraction of
solid []
fraction of
hydrogen
porosities
[]
a) b)
vmax = 0.0001 m/s vmax = 0.0011 m/s
Fig. 6.26: Prediction of hydrogen porosity formation in cast Al-7wt%Si ingot; situation after 6 s for
cases a) without shrinkage flow; b) with shrinkage flow; solid fraction distribution and velocity
field (left), volume fraction of hydrogen porosity as grey scale representation plus stochastic poros-
ity visualisation; the area of black dots is proportional to the area of hydrogen porosities; propor-
tionality factor: 3.
additional space. What is not physically correct, and explains why this effect in general is not ob-
served, is the assumption of no shrinkage. The alloy used has a quite high shrinkage factor. In real-
ity, the volume reduction due to shrinkage is bigger than the volume gain due to hydrogen pore
formation. In addition, in real solidification processes some of the hydrogen bubbles rise to the sur-
face and leave the cast part.
6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys 191
The completely solidified areas show negative macrosegregations. The liquid which is squeezed out
of the cell by the forming bubbles is enriched, as the process happens inside the liquid-solid area.
Therefore the average concentration of the cell which contains the bubbles decreases.
The porosities are quite uniformly distributed over the whole solidified area. The volume fractions
vary from 0.0025 close to the chills to 0.0031. These variations must be a consequence of the dif-
ferent solidification conditions.
If shrinkage is switched on, the net flow, i.e. the sum of shrinkage flow and hydrogen expansion
flow is a flow towards the forming solid, but it is slower than the flow would be without hydrogen
porosity formation. As a result, inverse segregations which form close to the chill at the bottom are
reduced, because less enriched melt is attracted. This effect has been experimentally observed by
/Rousset 95/. They examine the influence of porosity formation on the amount of inverse segrega-
tion in directionally solidified aluminium alloys and find, that the amount of inverse segregation is
significantly lowered if hydrogen porosities form. Fig. 6.26 b) shows the flow field and solid frac-
tion distribution after 6 s. As in the case without shrinkage the porosities are quite uniformly dis-
tributed over the whole solidified area and show a similar pattern.
Neither of the two calculations was terminated, because without shrinkage porosity formation
model the hot spot leads to unphysical result in later stages of solidification.
14
The top part of the casting did not solidify completely due to the adiabatic boundary conditions.
192 Chapter 6 Applications: Simulations and experiments, Casting processes
fraction of fraction of
porosities solid
[] []
a) b) c)
Fig. 6.27: Simultaneous prediction of shrinkage and hydrogen porosities for case of zero viscosity;
final distribution of a) shrinkage porosities b) hydrogen porosities c) total amount of porosities;
volume fraction of porosities with grey scale (left); fraction of solid and stochastic porosity visuali-
sation (left); the area of black dots is proportional to the area of porosities; proportionality factor: 3.
As expected the whole cast part shows almost equally distributed hydrogen porosities. This can be
explained by the feature of the model that per unit volume the same mass of hydrogen is rejected
and forms pores. As the pores are assumed to form during solidification, the temperature intervall is
restricted to the liquid solid interval of the alloy. As a reminder: The mass of hydrogen is constant,
the final volume depends on local pressure conditions during solidification and therefore on the
local temperature, too. The small spatial differences can be explained by slightly different solidifi-
cation conditions. Hydrostatic pressure leads to slightly smaller amounts in the bottom part of the
casting.
Less expected, but likewise easy to explain is the high hydrogen porosity volume inside the area
which is still liquid when the first shrinkage porosities start to form. When the hot spot is separated
from the feeding reservoir, the pressure inside the whole area starts to drop until it reaches the criti-
cal pressure of pore formation. Due to the reduced resistance the hydrogen pores forming from now
on are much bigger than the ones before. As they only form in solidifying areas, the area with high
hydrogen porosity fraction grows in a ring-shape towards the middle of the hot spot, Fig. 6.28.
The results of second case, the calculation with small non-zero viscosity, Fig. 6.29, look very simi-
lar, as much as the strong shrinkage porosities in the middle and the hydrogen porosity distribution
are concerned. New are very fine shrinkage porosities almost all over the cast part which occur as
6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys 193
large grey areas in the grey scale image of the total amount of porosity. It looks strange that they do
not occur in the grey scale image of the shrinkage porosities. The explanation is, that only added to
the equally distributed hydrogen porosities they become visible in this scale. The fine porosities
occur because the increased viscosity increases the drag forces inside the solidifying cells. The cells
need to compensate the shrinkage losses of the solidifying liquid and attract liquid from neighbour
cells. If the cell is almost completely solidified, the drag forces become so high, that the pressure
limit for shrinkage porosity formation is reached, the feeding flow is blocked, and microporosities
form. These porosities can be interpreted as the microscopic porosities which form between den-
drite arms due to the rupture of interdendritic feeding.
t=9s t = 10 s t = 11 s
fraction of
porosities []
fraction of
solid []
t = 13 s t = 15 s
Fig. 6.28: Simultaneous prediction of shrinkage and hydrogen porosities for case of zero viscosity;
development of total porosity fraction (left) in comparison to solid fraction (right) with time; visu-
alisation of porosities by black dots (right) uses proportionality factor 3.
194 Chapter 6 Applications: Simulations and experiments, Casting processes
The third case which increases the drag forces in two ways, by increasing the viscosity and by de-
creasing the permeability constant of the mushy zone, yields qualitatively the same result, but with
stronger micro-shrinkage porosities.
6.4.5 Conclusions
Many publications have been written on the question if it is possible to distinguish hydrogen poro-
sities from shrinkage porosities or not. Of course the primitive assumptions used in the present
work to model the pore formation do not allow any predictions of pore size or shape, but the results
of the coupled calculation give room for speculation. Typical hydrogen porosities have rounded,
compact shapes because they form in areas with comparably low solid fractions. The large amount
of hydrogen pores occuring inside the low pressure area of the present calculation forms inside an
area with solid fractions of about 0.5. If the pores form inside a stable dendritic network, would it
not be possible that they look like typical shrinkage porosities and that they cannot be distinguished
from the shrinkage porosities in the same area? Certainly there are many cases where porosities can
be definitely identified as hydrogen porosities, and there are many other cases where there is a high
fraction of fraction of
porosities solid
[] []
a) b) c)
Fig. 6.29: Simultaneous prediction of shrinkage and hydrogen porosities for the case of small
nonzero viscosity = 1.12510-3 and isotropic permeability with K0 = 810-10 m2; final distribution
of a) shrinkage porosities b) hydrogen porosities c) total amount of porosities; volume fraction of
porosities with grey scale (left); fraction of solid and stochastic porosity visualisation (left); the area
of black dots is proportional to the area of porosities; proportionality factor: 3.
6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys 195
probability for porosities being the result of shrinkage. But certainly there are also many
cases,where shape and topology do not give last evidence whether porosities are dues to hydrogen
precipitation or to shrinkage, or maybe to both at the same time.
The new model presented in this work is based on the postulate, that two completely different for-
mation mechanisms are responsible for the formation of hydrogen and shrinkage porosities. There-
fore, even if in many practical cases it is not possible to distinguish between them, it is possible to
separately predict them by numerical simulation. This is how numerical simulation could help to
identify the porosity type in real castings. Of course a lot of modelling effort is still necessary be-
fore the model can give quantitatively correct results, but this section has shown, that the principle
works and that the basis for further development has been laid.
7 Applications: Simulations and experiments, Welding processes
7.1 Systematical investigation of pool shape formation in welding
A main reason for the development and application of weld pool simulations is the wish to under-
stand the influence of the large number of welding parameters. Final aim is it to become able to
predict the shape and to control it by a deliberate choice of the parameters.
Even if many works on the simulation of weld pool formation have been published, section
2.6.2.1.2, the number of large systematical studies is small. One example is /Wang 01/ who system-
atically study the influence of the sulphur contents on the pool shape formation and the number of
vortices in the pool during GTA welding.
The numerical model developed in the context of this work is able to describe the influence of many
of the parameters for the cases of both laser and GTA welding and has therefore been used to lead
three systematic investigations. The first one assumes the MC to be constant and therefore allows a
good coverage of some important parameters of laser welding processes. The second one, still on
laser welding, assumes the MC to depend on temperature and initial sulphur content of the base
material. In this case the parameter space becomes much more complex and the number of possible
variations so huge, that only special cases have been studied. The weight in this study has been put
on transient effects which are a consequence of the temperature dependent MC. The third one de-
scribes the transient interaction between Marangoni forces and electromagnetic forces in GTA
welding processes.
boundary. The initial time step width is 0.001 s. Due to the high Marangoni flow velocities this
value sometimes had to be reduced by the adaptive time step control down to 0.00005 s.
Table 7.1: Additional parameters and physical constants used for all welding calculations in this
chapter
Av constant for evaporation model (iron, evaporation 2.52
model)
Hv specific evaporation heat (evaporation model) 6259.5 kJ / kg
R universal gas constant 8314.3 J/(molK)
Ta ambient temperature 300 K
Tmax maximum surface temperature 2800 K
convection convective heat transfer coefficient 6.4 W/(m2K)
b emissivity 0.2
b Stefan-Boltzmann constant 5.6710-8 W/(m2K4)
198 Chapter 7 Applications: Simulations and experiments, Welding processes
1. Find a set of numerical parameters which is a good compromise between accuracy and reason-
able calculation times. This was done by comparing a number of preliminary calculations with
different settings.
2. Find one or more representative parameter sets which can be used as a base for parameter
variations.
Important parameters which have been studied here are:
Composition of the welded material
Laser radius
Welding power (laser power and efficiency are integrated into one parameter)
(constant) Marangoni coefficient
Viscosity of the melt
Fig. 7.2 shows the temperature scale which is valid for all temperature plots of this section. As the
different initial alloy compositions have different liquidus and solidus temperatures which deter-
mine the shape of the mushy zone, these values are given in the caption of the corresponding figure.
a)
b)
c)
simulation must have a high grid resolution at least inside the surface layer. /Choo 92a,b/ and
/Wang 01/ use a grid spacing comparable to the lowest resolution used in this work (60x40 cells for
an area of 9x6 mm2). They have performed extensive grid sensitivity trials and found out that the
accuracy of the results does not increase much with further refinement. To check these assumptions,
calculations with equidistant grids of 60x40, 90x60 and 120x80 cells were made. Fig. 7.3 shows the
200 Chapter 7 Applications: Simulations and experiments, Welding processes
resulting pool shapes of a laser welding of Fe-0.1wt%C for P = 2000 W, rlaser = 4 mm after 1.4 s
together with the corresponding grid resolutions.
Strange to say, only the calculations with positive MC depend strongly on the grid resolution. They
grow deeper with refined grid. This can be explained by the increase of the maximum vertical ve-
locity (60x40: 0. 3247 m/s, 90x60: 0.4012 m/s, 120x80: 0. 4542 m/s) which is driven by the
increased maximum surface velocity (60x40: 0.2197 m/s, 90x60: 0.2782 m/s, 120x80: 0.3180 m/s).
The calculations with negative MC only show a refinement of the curve describing the solidification
front, but no real enlargement of the pool. An enlargement would be expected, because the maxi-
mum surface velocity increases considerably (60x40: 0.136 m/s, 90x60: 0.189 m/s, 120x80:
0.219 m/s). For the calculations without Marangoni convection no significant change was due.
7.1.2.1.4 Conclusions
After carefully considering the results of the foregoing sections, the following set of numerical pa-
rameters was chosen to be the basic numerical set-up: 2D-axisymmetric grid of 60x40 cells, 10
SIMPLER iterations per time step, and use of adaptive time step control. The decision for the quite
coarse grid was taken to obtain acceptable calculation times. A calculation with positive Marangoni
coefficient takes more than 20 days if a resolution of 120x80 cells and adaptive time step control is
chosen, 6 days for 90x60 cells and only 20 hours for 60x40 cells. Using equidistant time steps in-
stead reduces calculation times by a factor of four, but the error occuring here is due to wrong
physics, while the grid error is a problem of resolution. This is why the coarse grid was chosen, but
a correct time step control. The decision for the low number of SIMPLER iterations was easy taken,
because the time gain is big and the loss in accuracy seems to be negligible.
In addition to these settings a maximum pressure residual of 10-5, a maximum pressure and pressure
correction solver iteration number of 2000 per SIMPLER iteration, SIMPLER convergence criteria
of 10-7 for velocities, temperatures and concentrations, 10-10 for pressure correction, 10-3 for pres-
sure and underrelaxation factors of 0.6 for velocities and temperature were chosen for various rea-
sons not to be discussed here. If not otherwise mentioned the anisotropic permeability model is
used.
The choice of the basic set of welding parameters, weld material, welding power, laser radius and
Marangoni coefficient developed slowly during the different parameter studies. The composition
study showed, that the composition of the weld material is not a critical parameter. This is why
early calculations with Fe-0.1wt%C or Fe-0.42wt%C were not redone when it turned out that cal-
culations with Fe-0.2wt%C are numerically more stable and yield smaller numerical concentration
losses. The parameter combination laser radius rlaser = 4 mm, welding power P = 2000 W and MC =
110-4 are close to the largest possible pool extensions before the pool explodes due to evaporation
or penetrates the metal plate, but is still safe.
202 Chapter 7 Applications: Simulations and experiments, Welding processes
The following sections show the results of the parameter studies in detail. If not otherwise men-
tioned the calculations were performed using P = 2000 W, rlaser = 4 mm and welding duration 1.5 s
and grid resolution 60x40 cells.
7.1.2.2 Influence of weld material composition
The influence of the composition of the welded material was studied for different binary Fe-C al-
loys and constant MC of 10-4 and 0. Fig. 7.5 d) shows the alloys tested together with the corre-
sponding liquidus and solidus temperatures which have been read from the equilibrium Fe-C dia-
gram. Fig. 7.5 a) c) shows the temperature distribution after 1.5 s for carbon concentrations
0.1wt % and 0.8 wt % for each of the three MC15.
The figure shows that the influence of composition is rather limited. This is why not all results are
presented. Observable effects with increasing carbon concentration are:
The weld pools become slightly larger. This effect can be explained by the decrease of the liq-
uidus temperature with increasing concentration. If the positions of the isotherms remain con-
stant which is approximately the case in the calculations with negative and zero MC , the
solidus front moves towards lower temperatures. In the calculations with positive MC the iso-
therms at the bottom do not remain constant and the pool is considerably deeper. This is a result
of a higher maximum vertical velocity (0.330 m/s for Fe-0.8wt%C compared to 0.325 m/s for
Fe-0.1wt%C). Why the velocity is higher could not yet be determined.
The solidification time after switching off the laser at t = 1.5 s becomes longer because the ma-
terial needs to cool down more to reach the lower solidus temperatures. For MC = 0 the time in-
creases from 0.402 s to 0.456 s for 0.1 to 0.8 wt % C (12.6 %), for negative MC it increases
from 0.339 s to 0.396 s for 0.1 to 0.7 wt % C (15.5 %), and for positive MC it increases from
0.507 to 0.529 for 0.1 to 0.4 wt % C (4.2 %).
The mushy zone becomes larger. This is related to the variation of the solid/liquid interval.
The results of the composition study lead to the conclusion that the choice of the weld material for
the studies is not critical and can follow practical considerations. The first calculations were made
with Fe-0.1wt%C. Later during the investigation it turned out that for Fe-0.2wt%C the numerical
stability is higher and concentration losses are smaller.
15
For negative MC the calculation with 0.8 wt % C crashed after 1.2 s due to instabilities of the old solidification algo-
rithm. For the same reason there are no solidification times available for calculations with positive MC and concentra-
tions higher than 0.4 wt % C.
7.1 Systematical investigation of pool shape formation in welding 203
development of the pool shapes. The rounded pool of the calculation without Marangoni convection
becomes a deep V-shape for positive MC and a wide flat shape for negative MC. The diagrams in
Fig. 7.7 show interesting details: Already small positive values of the MC have a large influence on
the pool depth. For values > 0.1 N m-1K-1 the increase slows down considerably. For this reason
0.1 Nm-1K-1 was chosen to be the MC of the basic set of welding parameters.
For negative MC the surface flow drives hot melt away from the centre of the pool. Therefore the
temperature in the centre decreases slightly with increasing negative MC and so does the pool
depth. As the hot melt is driven towards the outer areas of the pool, the radius increases with in-
creasing negative MC and increasing negative surface velocities.
204 Chapter 7 Applications: Simulations and experiments, Welding processes
MC = +110-5
MC = 0 MC = 110-5
a) b)
Fig. 7.7: Influence of different constant Maran-
goni coefficients on the shape of the weld pool
for welding of Fe-0.1wt%C with P = 2000 W: a)
maximum horizontal (umax) and vertical (vmax)
velocity; b) pool radius and central depth; c)
weld pool volume, all after 1.5 s welding time.
c)
temperature and thus the radiation losses. In the case of positive MC the heat is removed from the
surface and transported downwards, which as well reduces surface temperature and heat losses. In
both cases the net heat input and thus the total molten volume is higher.
b)
a)
c)
Fig. 7.8: Influence of surface vaporization model on the shape of the weld pool for welding of Fe-
0.1wt%C with P = 2000 W in the case of rlaser = 0.5 mm and MC = a) +110-4 , b) 0, c) 110-4; left:
surface temperature limited to 2800 K; right: surface temperature unlimited (maxima > 5000 K);
temperature distribution after 1.5 s welding time; scale cf. Fig. 7.2; Tliq = 1801 K, Tsol = 1766 K;
black area: temperatures above 2800 K.
rlaser = 5 mm rlaser = 7 mm
7.1 Systematical investigation of pool shape formation in welding 207
1. The heat loss at the surface is governed by the temperature dependent vapour pressure of the
alloy, /Zacharia 91/. Maximum temperature is the boiling temperature. If the temperature
reaches the boiling point, it is fixed to this value, cf. section 3.3.3.2.3.
2. The heat loss at the surface is governed by the temperature dependent vapour pressure of the
alloy, but the temperature is not limited.
All calculations of this section were made with both models. Fig. 7.8 shows the big difference of the
results for the smallest radius rlaser = 0.5 mm. Both models neglect the most important phenomenon,
namely the formation of a vapour capillary and the transition from conduction mode welding to
keyhole welding, and therefore cannot accurately describe the pool shape formation for cases where
the surface temperature reaches the boiling point.
Fig. 7.9 7.11 show the variation of the pool shapes with the laser radius for different MC. The
calculations were made using the surface vaporization model with limited surface temperature. The
208 Chapter 7 Applications: Simulations and experiments, Welding processes
a) b)
c) d)
Fig. 7.12: Influence of laser radius and surface vaporization model on the pool shape for welding of
Fe-0.1wt%C with P = 2000 W; a)c) pool radius and central depth calculated with and without sur-
face temperature limit for MC = a) +110-4 N m-1K-1, b) 0 c), 110-4 N m-1K-1; d) weld pool vol-
ume, all after 1.5 s welding time.
cases of the missing vapour capillary are shown, but marked with (V). The diagrams in Fig. 7.12
show clearly the point where the two surface vaporization models start to make a difference. For
rlaser > 3 mm the results are almost identical, for smaller values the pool shapes start to differ. The
temperature limit creates an almost constant temperature profile along the surface which only de-
cays in the outer parts of the pool. As only surface areas with steep temperature gradients add mo-
mentum to the Marangoni flow, the maximum surface velocities are in this case much smaller than
in the case of unlimited surface temperatures, where the whole surface has a Gaussian temperature
profile and works as a momentum source for Marangoni flow. Reduced Marangoni velocities yield
7.1 Systematical investigation of pool shape formation in welding 209
the same effects as a reduced MC would. The pool depth reduces in the case of positive MC, the
pool radius in the case of negative MC.
The behaviour of the weld pool volume, Fig. 7.12 d), for large and small radii can be explained by
two different effects. For large radii the temperature rise is slower. Therefore there is more time for
heat to be abducted into the welded plate by conduction. The maximum surface temperature is
lower. For small radii the surface either reaches the maximum temperature and the heat input is
limited, or the surface temperature rises very high and heat is lost by radiation. An important obser-
vation is the independence of the molten volume from the pool shape and flow pattern for large
radii. In this case the molten volume is exactly equal for all types of Marangoni flow.
flat pools in case of negative MC does not increase at all. Instead of that a second convection cell
forms and transports hot melt towards the bottom of the pool. This leads to an increase of the cen-
tral pool depth of about 40 % (1.54 mm instead of 1.11 mm). In the case of positive MC a second
convection cell forms too, but only during the resolidification phase after the laser has been
7.1 Systematical investigation of pool shape formation in welding 211
a) b)
Fig. 7.16: Influence of welding power on the pool shape for welding of Fe-0.2wt%C; a) pool radius
and central depth for MC = 110-4 N m-1K-1 and MC = 0; b) weld pool volume, all after 1.5 s
welding time.
switched off, Fig. 7.38. Another interesting effect is the decreasing pool in the case of zero MC. The
reason has not yet been established, but it certainly has something to do with the increased veloci-
ties inside the mushy zone. A possible explanation could be, that the mushy zone is washed out
more and the local solidus temperature rises.
212 Chapter 7 Applications: Simulations and experiments, Welding processes
The main effect of varying viscosity has been observed in the resulting macrosegregation patterns.
This subject will be discussed in section 7.2.3.
7.1.3 Pool shapes with temperature dependent Marangoni coefficients (laser welding)
If the MC is no longer considered to be constant but assumed to vary with temperature and (con-
stant) contents of surface active elements like sulphur or oxygen, the complexity of the weld pool
behaviour increases considerably. The variations of the pool shapes become so complex, that a
comprehensive analysis is almost impossible. This chapter, which is partially based on some early
work (this explains the use of a different basic parameter set), investigates some general properties
and effects occuring if temperature dependent MC is assumed.
The pool shapes shown in this section, especially V- and W-shapes, have been experimentally ob-
served and are well known in literature, e.g. /Heiple 85/, /Winkler 98/, and /Pavlyk 01/, but in gen-
eral the formation of the different shapes are explained and interpreted as a function of the concen-
tration of surface active elements. /Pitscheneder 96/ investigate the question why the splitting into
flat pools for low sulphur contents and deep pools for high sulphur contents happens at high weld-
ing powers and not at low ones. He states that the reason is competition between conductive and
convective heat transfer as well as the temperature dependence of surface tension coefficients.
The numerical study in this section systematically investigates the influence of welding power and
welding duration on the formation of pool shapes during laser welding processes using the model
for temperature dependent surface tension gradients by /Sahoo 88/, section 3.3.2.9. Special attention
is paid to the description of transient effects which in general are neglected by stationary calcula-
tions and which can significantly influence the final pool shape. These effects should be considered
during the interpretation of experimentally observed weld pools.
The study does not want to give quantitative recipes for specific materials or configurations. The
aim is rather to give an overview of the different types of weld pool dynamics that can occur in la-
ser welding systems and to provide a deep understanding of the evolution of weld pools under dif-
ferent welding conditions, how certain pool shapes form, and which effects influence the time de-
pendent and final pool shape. Stress is laid on establishing general rules of melt pool behaviour,
without taking into account that the parameters might not always represent realistic welding condi-
tions for the specific system described. In many cases it may not be possible to realize all the ef-
fects.
The behaviour and effects observed with the system investigated in this section (Fe-0.42wt%C with
0.014 wt% S) are so general that they should occur in many systems. The study gives an idea how
variations of parameters such as welding power, welding duration, amount of surface active ele-
ments or even imperfect knowledge of physical constants such as the standard heat of adsorption
H 0 can influence the welding results. On the other hand it becomes obvious, which parameters
7.1 Systematical investigation of pool shape formation in welding 213
Fig. 7.18: a) Dependence of Marangoni coefficient / T on sulphur activity and standard heat of
adsorption. b) Dependence of critical temperature on sulphur activity
must vary to reach a certain result. If the general rules are to be applied to a specific system it will
be necessary to carry out systematic experimental studies to find out material data, critical tem-
peratures etc. The study gives the theoretical background on how to do the experiments, and what to
search for.
In fact the amount of surface active elements as well as the standard heat of adsorption H 0 have a
high influence on the value of the Marangoni coefficient. Fig. 7.18 a) shows the temperature de-
pendent Marangoni coefficient for different values of H 0 and aS, Fig. 7.18 b), how the critical
temperature (CT), defined as the temperature where the Marangoni coefficient changes its sign,
depends on the sulphur activity aS. It becomes evident, that the correct knowledge of these parame-
ters is crucial even for qualitatively correct calculations. With increasing sulphur concentration the
CT rises, but qualitatively the temperature dependence of / T does not change very much. For
this reason the calculations in this work which use temperature dependent MC have been limited to
one fix pair of the parameters H 0 1.88108 J/mol and aS = 0.014. The qualitative results can be
transferred to cases with different sulphur activities by adapting the welding powers.
Table 7.2: Parameters used to calculate the temperature dependent MC using eq. 3.24; data for in-
dustrial steel AISI304 by /Choo 92b/
s surface excess of the solute species (S) at saturation 1.3 10-8 mol m-2
/ T for pure iron 4.3 10-4 N m-1K-1
aS constant activity / concentration of surface active element S 0.014 wt%
H 0 standard heat of adsorption 1.88108 J mol-1
S1 constant, related to the entropy of segregation 3.18 10-3
214 Chapter 7 Applications: Simulations and experiments, Welding processes
Table 7.2 shows the parameters which have been used to calculate the temperature dependent MC.
They were determined by /Choo 92b/ for the industrial steel AISI 304.
Fig. 7.19 shows the temperature scale which is valid for all temperature plots of this section. The
grey area with temperatures above the critical temperature of the MC, 2287 K, represents the area
with negative MC, the grey area with temperatures between 1720 K and 1775 K is the mushy zone
for the initial composition Fe-0.42wt%C.
The calculations were all run on four out of 26 processors of a SUN HPC 3500/6500 UltraSPARC-
II system at the Edinburgh Parallel Computing Centre (EPCC, Scotland), and took between one and
six days to finish, depending on the molten area to be calculated.
O R3
l /
s
= 0.53 , R R3 =
/
s
0.74
a) b)
Fig. 7.21: Time development of normalised characteristic numbers for a) V-shape pool, P = 1000
W, R3s = 0,0027 m; b) W-shape pool, P = 2000 W, R3s = 0,0044 m
without considering the value of the offset of the critical temperature point. The lateral depth
only gives additional information if the offset is different from zero.
When the laser is switched off, the critical temperature point shifts quickly towards the middle
of the pool and then disappears. Nevertheless, the second deepening does not disappear that
fast, but follows slowly. In this case the lateral depth is defined as the maximum depth of the
second deepening until it disappears and joins the central deepening.
Fig. 7.21 shows typical time developments of the characteristic numbers for V-shape and W-shape
pools. For P = 1000 W there is no offset Ol of the CTP and therefores the lateral depth is defined to
be equal to the central depth Dc. In the case of P = 2000 W the CTP directly moves outwards and Ol
almost equals the radius. Dl and Dc develop in different ways. It is obvious that the W-shape only
becomes visible when Dl becomes bigger than Dc. The inverted situation at the beginning of the
melting process (t < 0.7 s) and the maximum occuring in the graph of Ol are due to an interesting
transient effect which will be described in section 7.1.3.3.2. V-shape pools can be recognised from
the central depth Dc growing quickly and often after some time depassing the radius. This would
never happen to a W-shape pool.
Fig. 7.22 shows the pool shape development for comparable welding processes leading to V- and
W-shape pool formation respectively.
7.1.3.2.1 Simulations
For this study laser welding calculations of Fe-0.42wt%C with temperature dependent MC and
welding powers of P = 1000 W, 1200 W, 1300 W, 1395 W, 1500 W, 2000 W and 4000 W are taken
into account. Fig. 7.23 gives some examples for the weld pool shape, the temperature distribution
and the flow field after 2.5 s welding time for different welding powers. The plot of all characteris-
tic numbers for all calculations, Fig. 7.24, gives a more complete and detailed view.
If the initial phase of pool formation is ignored, the calculations split into two groups forming V-
shape or W-shape pools respectively. For low powers (1000 W 1395 W) a stable V-shape pool is
formed. For high powers (1500 W 4000 W) it is a W-shape. This result can be qualitatively cor-
related to unpublished experimental work by /Pavlyk 03/, Fig. 7.26. The GTA weld pools show flat
shapes for powers P < 2000 W and W-shape pools for higher powers. For very high powers proba-
bly boiling leads to the formation of an additional deepening and porosities in the middle. Fig. 7.25
shows the characteristic numbers for different welding times as a function of welding power. The
data marked as estimated are extrapolations which have been determined from the most probable
pool development for calculations which did not reach the time steps 1 s or 3 s. The two calcula-
tions with 1395 W and 1500 W (as well as another calculation with 1400 W that is not shown here)
7.1 Systematical investigation of pool shape formation in welding 217
a) P = 1200W b) P = 2000W
a) b)
c) d)
Fig. 7.24: Laser welding of Fe-0.42wt%C with temperature dependent MC; time evolution of char-
acteristic numbers for different welding powers: a) normalised central depth Dc / R3s; b) normalised
lateral depth Dl / R3s; c) normalised offset of CTP Ol / R3s d) normalised radius R / R3s.
crashed due to numerical instabilities: 1395 W after 1.05 s, 1500 W after 0.75 s. The exact reasons
for the crashes are still to be found, but a closer look at the initial pool development shows that be-
tween the power regions that lead to a stable V-shape or W-shape formation, there is a critical
power region (~ 1400 W 1500 W), where the pool development shows a mixed, transient behav-
iour. The decision to use and extrapolate the data of these calculations is due to the fact that at the
time of the crash they seem to have reached a state of stable and predictable pool development.
The careful analysis of the time evolution of characteristic numbers yields a number of interesting
observations:
220 Chapter 7 Applications: Simulations and experiments, Welding processes
a)
b)
Fig. 7.25 : Laser welding of Fe-0.42wt%C with temperature dependent MC; characteristic numbers
of weld pools after a) 1 s and b) 3 s as a function of welding power
There is no smooth transition area between the areas with V-shape and W-shape behaviour. In-
stead with rising power the normalised central depth Dc / R3s grows. The transition happens sud-
denly in the critical power area, Fig. 7.25. It is interesting to observe that this is not defined by
the power where the surface first reaches the CT. In some calculations with stable V-shape for-
mation an area with temperatures above CT appears, but it is not able to move far enough from
the middle of the pool to help the pool to establish a W-shape. The transport of hot melt towards
outer regions of the pool is rather prevented by the strong inward flow of cool melt. From
P = 1200 W, where the surface reaches the CT only from time to time over P = 1300 W, where
a small region with temperatures above CT is established up to P = 1395 W, where the hot melt
tries an outbreak, all calculations show a predominance of the areas with positive MC. The first
calculation that probably manages to establish a W-shape is the one with P = 1500 W. The cal-
culations with higher powers show a CTP that is quickly shifted outwards to reach a maximum
offset Ol and to be shifted back when the outer convection roll grows.
The graphs of the central and lateral depths Dc / R3s and Dl / R3s show clearly the two groups of
pool shape types and the undecided behaviour of the calculations with P = 1395 W and
P = 1500 W. This shows up especially in the graphs of Dc / R3s for P = 1395 W, that takes about
0.8 s to take its place in the series from P = 1000 W to P = 1300 W, and Dl / R3s for
P = 1500 W, that begins as a V-shape and then starts to switch to W-shape behaviour.
normalised radius R / R3s: With increasing power the pool formation starts earlier and the radius
grows quicker towards the maximum radius.
7.1 Systematical investigation of pool shape formation in welding 221
The search for W-shape pools in laser welding is an important task on the way to understanding the
principles of Marangoni convection. It can help to understand better temperature and concentration
dependent Marangoni coefficients.
7.1.3.3.1 Increasing central pool depth when heat source is switched off
In W-shape pools the molten area at the moment when the heat source (HS) is switched off is not
identical with the total molten area. In fact the central pool depth Dc increases considerably when
the heating stops and the pool starts to resolidify. The occurence of the effect is not limited to a
certain welding duration. Calculations with P = 2000 W and durations of 0.5 s, 1.0 s, 1.5 s, 2.0 s,
2.5 s and 3.0 s were made.
Fig. 7.27 a) shows the time evolution of the normalised central depth Dc / R3s for the different
welding durations. All calculations that had enough time to form a W-shape pool showed the same
effect. Only the calculation where the HS was switched off after 0.5 s, during the stage of the flat
shaped pool, did not grow deeper. The time evolution of the normalised radius R / R3s,
Fig. 7.27 b), shows that after 0.5 s the pool has not yet reached the stage of stable, regular W-shape
growth.
The effect of increasing central depth is easy to understand. Fig. 7.28 shows the situation a short
time before and after switching off the HS. The temperature distribution on the surface has been
creating a stable W-shape pool. When the heating stops, the surface starts to cool down quickly.
The CTP shifts towards the middle of the pool and then disappears. Even if the temperature sinks,
the temperature distribution approximately keeps a Gauss-shape. The MC, that is now positive on
the whole surface, creates a Marangoni flow towards the middle of the pool, trying to transform the
pools W-shape into a V-shape. This process is stopped by solidification.
224 Chapter 7 Applications: Simulations and experiments, Welding processes
a) b)
Fig. 7.27: Laser welding of Fe-0.42wt%C with temperature dependent MC; laser is switched off;
time evolution of a) normalised central depth Dc / R3s, b) normalised radius R / R3s for different
welding durations with P = 2000 K.
Fig. 7.29 shows the flow patterns during the first 0.3 s after the HS has been switched off. Of course
the effect does not occur in V-shape pools because here the Marangoni flow is already directed to-
wards the middle of the pool and when the heating stops the pool will only shrink.
Of course the volume of the melt pool is not increased by this effect. When the HS is switched off
the molten area starts losing energy by conduction, radiation and air convection. This energy loss
results in an decreasing amount of liquid phase. So the melting process in the middle is paid by an
increased solidification rate in the outer parts of the pool. Fig. 7.30 shows the change of the pool
shapes between the time when the HS is switched off and the time of the maximum central pool
depth for different welding durations. The black areas are the additionally molten parts of the pool
while the grey areas represent the parts resolidified in the same time.
7.1 Systematical investigation of pool shape formation in welding 225
Fig. 7.29: Laser welding of Fe-0.42wt%C with temperature dependent MC and P = 2000 W; flow
patterns which lead to an increased pool depth after switching off the HS. Left: Temperatures, scale
cf. Fig. 7.19; Right: Streamlines; the scales of different time steps are not comparable.
226 Chapter 7 Applications: Simulations and experiments, Welding processes
pool and the 1500 W speed up the pool enlargement showing the early stage of a typical W-shape
formation, compare Fig. 7.31.
Of course the calculations presented in this paper have not been continued until reaching a
steady-state (which would be impossible anyway, as no artificial cooling was applied). The pa-
rameter study with different welding durations and the obvious explanation of the mechanism
implies, that this effect will occur also in equilibrium conditions of a steady-state experiment /
calculation, when the HS is switched off.
A quantitative knowledge of the process of W-shape formation could be used in the design of
welding processes by determining minimum or maximum times for specific depth / width ratios
or pool geometries wanted. If the characteristic times are well known for the material actually
used, they could be used to determine the time, when the laser has to be switched off.
On the first glance, the instable behaviour of the pool shapes in the critical power region seems
to be of minor importance, as the transient effect is quite small. One feature of this instable pool
development is the formation of quite compact pool shapes in the initial phase. Of immense im-
portance however is the quantitative knowledge of the position of the critical power region as a
function of the concentration of surface active elements. A control by direct regulation of con-
centration and welding power could also be used in the design of specific welding processes.
7.1.3.4 Conclusions
The numerical study of a special welding system has led to a set of general rules and remarks which
can be used to understand and predict the dynamics and behaviour of weld pools in different sys-
tems.
1. If the concentration of surface active elements allows the formation of a CTP on the surface, the
welding power range splits into three regions with a) pure V-shape behaviour for the low pow-
ers b) instable mixed V-W-shape for a critical power region and c) pure W-shape behaviour for
high powers. If these pool shapes are not observed in reality, reasons can be:
The CT is too high to be reached, so only the behaviour of the lower power range is ob-
served
The power is too high for the W-shape to develop. The pool would explode by evaporation
(especially for Al alloys)
230 Chapter 7 Applications: Simulations and experiments, Welding processes
2. As for rising amounts of surface active elements the CT shifts to higher values, the position of
the critical power range in this case rises too. The common observation that, for constant weld-
ing powers, the pool shape changes from flat pools at low sulphur concentrations to deep pools
at high concentrations is consistent with the fact that the sulphur concentration determines the
value of the CT. For high sulphur activities a power could be in the V-shape region, while for
low sulphur contents it would be part of the W-shape region. A flat shape in this case can be due
to a short welding duration preventing the W-shape from forming. The position of the critical
power region is further influenced by the value of the standard heat of adsorption, H0.
3. The experimental central pool depth Dl which can be determined by evaluation of micrographs
is, in case of a W-shape pool, not identical with the central pool depth in the moment when the
HS is switched off. This is due to the transient effect of a changing flow pattern, when the
cooling changes the distribution of the MC on the surface. If steady-state calculations are com-
pared to steady-state experiments, the error induced by this effect can be as big as 20 % of the
observed depth.
4. The time evolution of W-shape pools shows two distinct phases, the quick formation of a flat
pool in the early stage, and the development of the full W-shape later. A quantitative knowledge
of this effect could help in the active design of welding processes by choosing ideal welding du-
rations to obtain a special weld shape.
5. The weld pool dynamics in the critical power region during the initial phases leads to compact
pool shapes which could also be used for weld pool design. A problem could be, that the exact
position of this region might be too difficult to determine and too instable for practical applica-
tions.
The transferability of the results is of course limited by the assumption of a homogeneous and con-
stant distribution of surface active elements. /Winkler 98/ gives evidence for the distribution being
strongly affected by the surface flow. This will result in a non constant value of CT and thus in a
more complex flow pattern.
a) b)
c) d)
Fig. 7.34: GTA welding of Fe-0.42wt%C with Marangoni and EM forces, time evolution of char-
acteristic numbers for different welding powers: a) normalised central depth Dc / R1.4s; b) normal-
ised lateral depth Dl / R1.4s; c) normalised offset of CTP Ol / R1.4s d) normalised radius R / R1.4s
If both Marangoni and EM forces are considered, the fact that both forces are of similar order of
magnitude gives rise to interactions resulting in complex transient flow patterns. In the context of
this work it is not possible to systematically investigate real GTA welding processes. The aim is
more to show the importance of the effects and to give an idea how to control them.
Transient effects are most likely to occur in the power range where Marangoni and EM forces push
into different directions. At low welding powers, when V-shape pools form in the case of pure Ma-
rangoni convection, the addition of EM forces will mainly encourage the development of the
232 Chapter 7 Applications: Simulations and experiments, Welding processes
a) b)
Fig. 7.35: GTA welding of Fe-0.42wt%C with Marangoni and EM forces; time evolution of all
characteristic numbers for a) P = 1550 W; b) P = 2000 W
V-shape. Much more interesting are cases of original W-shape behaviour. Here the EM forces com-
pete with Marangoni forces, and the time dependent balance between them can create remarkable
transient effects.
Five calculations were made with P = 1400 W, 1500 W, 1550 W, 1600 W and 2000 W. The
evaluation of the characteristic numbers yields a systematic behaviour. All calculations start by
forming a W-shape pool. With growing pool depth and thus increasing influence of the EM forces
most of them seem to flip to become a V-shape. This happens early in the cases of P = 1400 W and
P = 1500 W, but in a quite spectacular way for P = 1550 W, where the W-shape develops for a long
time before it suddenly transforms (after 1.8 s). In the case of P = 2000 W the Marangoni forces
first seem to be strong enough to establish the W-shape, but also here the EM forces interfere. In
contrast to the cases with lower powers the pool does not reach a V-shape during the calculation
time, but remains in some kind of U-shape with complex flow patterns due to the competing forces.
The calculation with 1600 W has crashed too early to allow a decision, which of the two behaviours
it will follow. The details of the time evolution for all calculations can be studied in the graphs of
the central depth Dc, Fig. 7.34 a), and the lateral depth Dl , Fig. 7.34 b). Here also the special role of
the calculation with P = 2000 W becomes obvious. After 3.5 s it starts showing instable behaviour.
The time evolution of Ol / R1.4s , Fig. 7.34 c), shows that with increasing power the zone with nega-
tive MC, that is necessary for the formation of a W-shape pool is kept up longer and longer. In the
cases of 1600 W and 2000 W it can not be decided, if it would collapse when the calculations were
continued.
7.1 Systematical investigation of pool shape formation in welding 233
a) b)
Fig. 7.36: GTA welding of Fe-0.42wt%C with Marangoni and EM forces; time evolution of weld
pool shape for a) P = 1550 W; b) P = 2000 W; Left: Temperatures, scale cf. Fig. 7.19; Right:
Streamlines; the scales of different time steps are not comparable.
234 Chapter 7 Applications: Simulations and experiments, Welding processes
As two examples for extreme transient behaviour the time evolutions of the two calculations with
P = 1550 W and P = 2000 W are presented. The representation of the four characteristic numbers in
one plot, Fig. 7.35, shows the strongly changing pool shape. Especially noticeable points are
the sudden overturning of the W-shape after about 1.8 s for P = 1550 W, signalled by the sud-
den reduction of Ol / R1.4s and the steep increase and joining of the two depth numbers Dc / R1.4s
and Dl / R1.4s
the instability observed after 3.5 s for P = 2000 W.
Both effects as well as the complex interaction of different convection cells can be studied in the
temperature and stream function plots in Fig. 7.36.
The observations made in this study can be summarized as follows: Due to the fact that the EM
forces are body forces and Marangoni forces only act on the surface, the influence of EM forces
rises with rising pool volume. For low power calculations the early stages all seem to be governed
by Marangoni convection and later EM forces take over and form a V-shape pool. For higher pow-
ers the predominance of Marangoni convection lasts longer and longer. For high powers it must still
be found out, if the EM forces can gain predominance, or if in these cases the Marangoni forces
win.
16
Even if the length scales considered are very small, the term macrosegregation is correct because it describes solute
redistribution on the scale of the whole weld pool.
7.2 Solute redistribution in weld pools (laser welding) 235
The concentration errors seem not to have a great impact on the qualitative accuracy of the results.
Sometimes calculations with very similar parameters show different errors, but look very similar.
For presentation in this section calculations were chosen with as small concentration errors as pos-
sible.
carbon
concentration
deviation [%]
t = 1.5 s t = 1.6 s
a) t = 1.7 s t = 2.05 s
t = 1.5 s t = 1.6 s
b) t = 1.7 s t = 2.0 s
t = 1.5 s t = 1.6 s
c) t = 1.7 s t = 2.0 s
Fig. 7.38: Development of macrosegregations during the resolidification phase of laser weld pools
of Fe-0.42wt%C with zero viscosity and constant MC = a) +110- N m-1K-1, b) 0, c) 110-4
Nm-1K-1; deviation of carbon concentration from initial concentration (left); streamlines of the ve-
locity field (right); the scales of different time steps are not comparable.
238 Chapter 7 Applications: Simulations and experiments, Welding processes
t = 1.5 s t = 1.6 s
t = 1.7 s t = 2.0 s
Fig. 7.39: Development of macrosegregations during the resolidification phase of laser weld pools
of Fe-0.42wt%C with zero viscosity and temperature dependent MC with H0 = 1.88108 J/mol;
deviation of carbon concentration from initial concentration (left), scale cf. Fig. 7.38; streamlines of
the velocity field (right); the scales of different time steps are not comparable.
depleted dendrites and enriched interstitial liquid. As viscosity is zero, the slowly decaying flow in
the pool continues to wash out the newly forming mushy zone and to gradually enrich the remaining
liquid. With raising concentration level in the all liquid area the newly forming solid becomes less
depleted and at a certain point (when Cl = C0/ ls) the new solid has the initial concentration C0.
From now on all solid is enriched and if there is still fluid flow, the enrichment process continues
and extremely high concentration values occur at the last point to solidify.
The irregular shape of the enriched area in the case of positive MC is due to the existence of two
convection cells turning in opposite directions. This effect is a pure consequence of the zero viscos-
ity, cf. section 7.1.2.6. The same happens in the case of negative MC, but the effect on the final
macrosegregation pattern is not as obvious.
carbon
concentration
deviation [%]
a)
b)
c)
d)
Fig. 7.40: Comparison of final macrosegregation patterns in laser weld pools of Fe-0.42wt%C with
zero viscosity (left) and constant realistic viscosities (right) for MC = a) +110-4 N m-1K-1, b) 0, c)
110-4 N m-1K-1 d) temperature dependent MC with H0 = 1.88108 J/mol; deviation of carbon con-
centration from initial concentration.
240 Chapter 7 Applications: Simulations and experiments, Welding processes
forces, the flow through the mushy zone can only be slowed down, but not completely stopped by
drag forces. The viscosity value determines the amount of segregations. An important topic is the
grid dependency during this phase. If the mushy zone is smaller than the cell size, a smaller grid
must yield smaller segregations, because only the volume of partially molten cells adds to the rised
concentration level in the pool. All fully molten cells only contribute melt with initial concentration.
The second phase starts when the heat source is switched off. Now only inertia forces maintain the
flow and the amount of macrosegregations forming depends on the permeability of the mushy zone,
i.e. on how fast the flow is stopped by drag forces. Viscosity is only one parameter to control this
property and the two cases shown for demonstration are two limiting cases of high and low perme-
ability. Other parameters influencing the permeability tensor may compensate effects of high or low
viscosity values and lead to macrosegregation patterns somewhere between the two limiting cases
shown. Test calculations have shown that already small drag forces can stop the flow quickly and
therefore the results of the real viscosity case seem to be the the most probable pattern.
always parallel to the temperature gradient. For casting processes this is an acceptable approxima-
tion, because the solidification is the first process and determines the dendritic structure. Eventual
remelting processes happen later in this same structure.
For welding processes this model is not valid, because they start by remelting the cast structure of
the base material and resolidify later, forming a new structure with new scaling and alignment of
the dendritic structure.
Therefore during remelting the initial dendritic structure which is a function of the casting and
maybe rolling process, must be taken into account. The question of the dendrite growth direction
during resolidification can also not be answered in such a simple way. Of course the dendrite
growth direction depends on the temperature gradient, but also on the orientation of the nuceleus
which has been the origin of the dendritic grain and the flow direction. If dendritic structures of
weld pools are considered we do not find a radial alignment of the dendrites towards the centre of
the pool.
To find out if the final concentration distribution in the weld pool is significantly influenced by
these effects, calculations should be made which use different models for melting and resolidifica-
tion:
242 Chapter 7 Applications: Simulations and experiments, Welding processes
Within the scope of this work a complex framework of models has been built which can be used to
simulate the influence of different types of convection on solidification and remelting processes in
casting and welding.
A complex system of coupled mathematical equations has been designed which yields flow ve-
locites, pressures, temperatures, concentrations and volume fractions of nine phases. The equa-
tions have been discretised using implicit and explicit schemes and a complex solution strategy has
been developed which allows to solve the equation system.
Modified VOF-model to describe the solidification of moving free surfaces with partial solidifi-
cation of surface cells.
Split Solid Model to describe the convective transport of floating equiaxed grains and to prevent
premature solidification of the moving free surface.
Solidification algorithm to describe eutectic and peritectic solidification of binary alloys in sys-
tems with strong convective solute and mass transport.
Porosity model for the simultaneous but separate prediction of hydrogen and shrinkage porosi-
ties, as well as internal shrinkage cavities.
The submodels have been integrated to form a complex framework which is now able to model the
following convection effects:
244 Chapter 8 Conclusions and Outlook
Macrosegregations:
Gravity segregation due to solutal convection
Normal (negative) segregation due to thermosolutal mushy zone flow
Inverse (positive) segregation due to shrinkage flow inside the mushy zone
Positive and negative segregations due to settling of equiaxed grains
Influence of external shrinkage cavity on final macrosegregation pattern
Influence of micro porosity on amount of inverse segregations
Formation of channel segregates
Shrinkage defects:
Formation of macroscopic external shrinkage cavities
Formation of hydrogen and shrinkage porosities, as well as internal shrinkage cavities
Welding:
Dependence of pool shape types as functions of different welding parameters
Concentration changes in weld pools
To the authors knowledge the model implemented in (SoliCon) is the first one which couples a
mobile solid model with a solidifying free surface in order to simulate the formation of deep shrink-
age cavities in a full NavierStokes formulation. The model has been tested and then applied to a
large number of scientific problems:
ence of the grid resolution and numerical errors were discussed. Comparison to experimental results
showed some important aaccordances. The total depth of the external shrinkage cavity and the po-
sition of the area with maximum enrichment, which corresponds to the last point to solidify, could
be reproduced well. The final macrosegregation patterns showed some similarities, but equally large
quantitative and also qualitative deviations. The same holds for the total size of the shrinkage cav-
ity.
The experimental verification of the simulation of an Al-7wt%Si ingot (EL2D) showed excellent
agreement of the predicted surface shape. The macrosegregation pattern shows interesting effects,
but cannot be expected to be correct, as thermosolutal convection and the limited diffusivity of Si in
the solid phase were neglected.
The porosity formation model was applied to a special casting process which had been designed to
provoke the formation of both shrinkage and hydrogen porosities. Simulations predicting hydrogen
porosity and shrinkage porosity formation separately and in a coupled calculation showed that the
model is able to reproduce typical porosity distributions, even if very simple pore nucleation models
are used. The coupled calculation additionaly demonstrated, that some areas, which in a real cast
part look like shrinkage porosities, in fact are dues to hydrogen precipitation. The model can there-
fore be used to remedy the old problem, that shrinkage and hydrogen porosities can not always be
told apart. The final porosity distribution in the case of pure shrinkage porosity formation was com-
pared to experimental results. Even if the model and the experiment are rather preliminary, there are
some promising accordances concerning the general shape of the porosity area. This is a sign, that
the formation mechanism has probably been correctly modelled.
An additional study has treated the question of solute redistibution in welding, a subject which has
been neglected so far, buth which has recently been recognised as important by leading scientists.
Macrosegregation patterns for different viscosity values have been calculated and the reliability of
the results was discussed.
The evaluation of all results leads to the following final conclusions:
The new model (SoliCon), which has been designed for the simulation of casting processes is
able to qualitatively model an immense number of complex phenomena related to convection
effects. The interaction between the submodels works well. As some of them are relatively sim-
ple, quantitatively correct results cannot be expected so far. In the present state the model can be
used to study principal mechanisms and interactions. Due to the modular structure and the well
defined interfaces the simple submodels can easily be replaced in the future without jeopardis-
ing the entire model.
The welding model (EL2D), which is a compilation of state of the art algorithms, completed by
a stable solidification algorithm, has shown to be a reliable tool for carrying out systematic in-
vestigations of welding processes. It can be used as a means to designing welding experiments
while hunting the famous W-shaped weld pool...
8.2 Outlook
As mentioned before, some of the models implemented in the mathematical and numerical frame-
work are rather preliminary. The scope of this work was the creation of a modular framework of
basic models which co-operate via well defined interfaces. The next steps are
Refinement of the existing models by comparison to experiments,
replacement of simple models by more sophisticated, physical ones,
addition of important new submodels and features which so far have been neglected,
rigorous testing, benchmarking and experimental validation of the models and complex interac-
tions.
Areas with specific development needs are:
Solidification algorithm
Modelling of undercooling effects
Reintroduction of Scheil- and Backdiffusion model
Reintroduction of multiple alloy elements
Generalisation of the solidification algorithm for use with complex binary phase diagrams
Macrosegregation modelling
Development of new permeability and segregation models for planar solidification fronts and
small mushy zones
Welding studies
The systematic investigations of welding processes should be continued by considering additional
parameters. Interesting parameters are:
The intensity distribution of the heat source; while simulations very often assume Gaussian
shape, real industrial lasers can have very different shapes like hat profiles or very irregular, os-
cillating ones.
The welding current I and voltage U in GTA welding. So far only the power has been studied
without considering that the power is always the product P = U I and the electromagnetic forces
depend strongly on the current.
The investigations aimed at finding W-shape pools in laser welding should be continued. A second
series of experiments should be performed. The experiment design has already been prepared.
The potential of the new model is almost infinite. And so is the amount of work remaining to do...
9 Appendices
9.1 Material data
The type of models implemented in the program requires a different kind of thermodynamic data as
usual. Many casting simulation tools assume a constant concentration of the binary alloying element
and thus only consider a vertical cut through the phase diagrams with constant liquidus and solidus
temperature. As the macrosegregation models allow concentration distributions that vary in space
an time, the liquidus and solidus temperatures are no longer constant but are taken from the com-
plete equilibrium phase diagram.
Physical properties like density, heat conduction and enthalpy are in reality complex functions of
temperature and composition. As many of these data are unavailable and a complete consideration
of all dependencies would have an impact on the stability and performance of the software, the fol-
lowing simplifications was made: The values are assumed to be constant over each cell and are de-
fined as the average of the single phase properties weighted with the corresponding phase fractions.
The next sections give more detailed information on the single properties.
9.1.1 Density
The density of all solid phases including obstacles are assumed to be constant and equal:
s := move = move = dead = dead = obst = const
2. constant and different from solid density: = const (shrinkage flow, no thermal convec-
l s
tion)
3. temperature dependent: = (T ) (shrinkage flow and thermal convection)
l l
4. constant for continuity equation, temperature and / or concentration dependent for buoyancy
terms of momentum equations (Boussinesq approximation):
lcontinuity = const lbuoyancy = lbuoyancy (T , C ) (only thermosolutal convection)
The influence of concentration on densities is neglected for all solid phases and for the liquid phase
only implemented in the framework of the Boussinesq approximation into the solutal buoyancy
250 Chapter 9 Appendices
term of the momentum equations. The continuity equation / shrinkage flow does not consider any
concentration terms.
9.1.2.1 Enthalpy
Enthalpy is assumed to be a linear function of temperature
hq = h0 q + c P q (T T0 q )
with
cP q = const
c P move = c P dead = cP
c P move = cP dead = c P
9.1.3.1 Al-Si
T pure Al
l = 933.6 K Cl eutectic = 12.6 wt%
Teutectic = 850.15 K C eutectic = 1.5 wt%
C eutectic = 99.994 wt%
9.1 Material data 251
Density
s = 2552 kg/m3 corresponds to (T) at Teutectic = 850.15 K
l(T) = 2737 0.3338 T [kg m ]
-3
2453 kg/m at Teutectic = 850.15 K
3
Heat conductivity
l = 75 W K m -1 -1
= = 165 W K m
-1 -1
Enthalpy
q T0q [K] h0q [J kg ]
-1
cPq [J kg K ]
-1 -1
Dynamic viscosity
l = 0.01125 kg m-1s-1
252 Chapter 9 Appendices
Emission coefficient
b = 0.2
Additional remarks
The diffusion of silicon in Al-Si is very slow. Thus an accurate description of the solidification of
Al-Si would require the use of a Scheil type microsegregation model. This model has not yet been
implemented. This has to be taken into account when interpreting the results obtained with the
equilibrium phase diagram using the lever rule.
9.1.3.2 Fe-C
Density
s = 7344 kg m-3 corresponds to (T) at Tperitectic
l(T) = 8547 0.835 T [kg m ]
-3
7071 kg m at Tperitectic
-3
Heat conductivity
l = 27 W K-1m-1
Enthalpy
q T0q [K] h0q [J/kg] cPq [J/(kgK)]
l 1766.167 1369272 789.657
1766.167 1098107 706.394
1766.167 1084263 675.417
/Bhmer 97/
Dynamic viscosity
l = 0.006 kg/(ms)
Emission coefficient
b = 0.2 0.7 depending on experiment
This is why EL2D nowadays is still used for the simulation of spot welding processes, while Soli-
Con is used for the simulation of casting processes.
EL2D (Development 1993-2000, 19000 lines FORTRAN 90, current version el2d9.1)
Main applications: Spot welding processes, calculation of weld pool shapes, solute redistribution
Models:
transient coupled calculation of u, v, p, T, Cl, C , C ,
stable solidification algorithm suitable for strong convective solute transport
anisotropic permeability of the Mushy zone
Thermosolutal convection, Marangoni convection, shrinkage flow
Numerics: Implicit discretisation
SoliCon (Development 2000-2003, 24500 lines FORTRAN 90, current version solicon2.0)
Main Applications: Casting Processes
Indices
A system of indices has been developed especially for the description of the differential equations
in continuous and discretised form. For a physical quantity X the following indices are possible:
(X )
b d
a c
a) Phase
In addition to the single phases
l: liquid pore shrink: shrinkage porosities
move: mobile solid - phase pore hydro: shrinkage porosities
move: mobile solid - phase obst: obstacles
dead: stationary solid - phase void: void
dead: stationary solid - phase
the following phase indices have been defined for reasons of convenience:
,: move + dead , move + dead
c) Spatial position of the quantity (discretised equation only): (face, nb). The meaning is:
P: cell centre of the current cell
N,E,S,W: cell centre of the top (North) / right (East) / bottom (South) / left (West)
neighbour cell
n , e, s, w : top (north), right (east), bottom (south), left (west) cell face
x, y : horizontal (x) / vertical (y) direction
face: any of the cell faces
nb: any of the neighbour cell centres
combinations are also possible, e.g. nW, sE etc., cf. Fig. 4.2 a)
All indices and the brackets are optional and can be missing in certain cases. The meaning of the
symbols X is explained in the table in the next section. The table contains also symbols where the
indices are used in a different way. If a symbol cannot be found in the list, it is only locally used
and its meaning is explained in the text nearby.
Main symbols
kg s ]
-1
T/C 0
a P
T/C 0
,a P
l l
neighbour coefficients of energy and liquid concentration equations used
T 0* T 0*
a P , a P by the phase change algorithm; the coefficients a P0 and a P0* are split to
separate the term depending on the amount of solid phase fractions:
T / Cl
aP
0
= a T / Cl
P
0
+a T /
P
C l
0
( , )
257
A(|P|) function used to discretise the convection diffusion term (power law),
defined in eq. 4.1 /Patankar 80/ []
Av constant for evaporation model []
b constant term of the discretised differential equation, /Patankar 80/ [mo-
mentum: N; pressure kg s ; energy: W; species: kg s ]
-1 -1
cies (Cl): kg m s ]
-2 -1
(D u / v / T / Cl
)
n/ e/ s / w integrated diffusive flux in diffusion convection term of discretised differ-
ential equation, /Patankar 80/[momentum (u,v): kg s ; energy (T ): W K ;
-1 -1
D~ diffusion coefficient [m s ]
2 -1
258 Nomenclature and Abbreviations
(Feff move )*n / e / s / w modified convective flux of mobile phases used in discretised energy
equation [W K ]-1
v
FEM Electromagnetic body forces during GTA welding [N]
v
g , gx/y gravitational acceleration vector and components [m s ] -2
h specific enthalpy [J kg ]
-1
J q i
concentration transfer due to phase change [kg m s ]
-3 -1
v
J u / v / T / Cl total flux of diffusion convection term of differential equation,
(J u / v / T / Cl
) x / y
/Patankar 80/ [momentum (u,v): N m-2; energy (T ): W m-2; species (C ): l
kg m-2 s-1]
(J u / v / T / Cl
) n/ e/ s / w integrated total flux of diffusion convection term of discretised differen-
tial equation, /Patankar 80/ [momentum (u,v): N; energy (T ): W; species
(C ): kg s-1]
l
K permeability tensor []
1
2
( )
C / P
+S T /
C / P
C l
( , )
t time [s]
t time step width [s]
T temperature [K]
Ta , T0 ambient temperature [K]
T0q reference temperature for calculating temperature dependent specific en-
thalpy; h0q = hq(T0q) [K]
Tfix fixed boundary temperature in heat conduction benchmark [K]
Tmax maximum allowed surface temperature [K] (welding)
Tliq, Tsol liquidus / solidus temperature [K]
Teut, Tperi eutectic / peritectic temperature [K]
U Welding voltage (GTA welding model) [V]
u l ,v l pressureless x- and y-velocity of implicit pressure equation (SIMPLER
algorithm) [m s-1]
v
v , u, v, w velocity vector and components in x-, y- and z-direction [m s-1]
vmax maximum absolute velocity in vector plots [m s-1]
v
v sup erficial superficial velocity [m s-1]
v
v pore pore velocity [m s-1]
VH initial hydrogen contents in the liquid phase [m3 kg-1]
VH max Maximum hydrogen contents in the solid phase [m3 kg-1]
VH rejected volume of hydrogen rejected by the solid [m3]
VH rejected 0 volume of hydrogen rejected by the solid at ambient pressure p0 and am-
bient temperature [m3]
Vpore shrink volume of shrinkage porosity [m3]
x, y, z cartesian co-ordinates (SoliCon, EL2D); in 2D model always: z = 1 m
x, y, z cell widths; in 2D model always z = 1 m [m]
261
pq
Abbreviations
CA Cellular Automaton GTA Gas Tungsten Arc
CET Columnar-to-Equiaxed Transition HS Heat Source
CGM Columnar Growth Model LHS Left Hand Side (of equation)
CT Critical Temperature MC Marangoni Coefficient
CTP Critical Temperature Point PA Phase Area
CVM Control Volume Method RHS Right Hand Side (of equation)
EB Electron Beam SSF Split Solid Function
ECT Equiaxed-to-Columnar Transition SSM Split Solid Model
FDM Finite Difference Method VOF Volume Of Fluid
FEM / FE Finite Element Method
11 References
/Academic 96/ Academic Press Dictionary of Science and Technology,
www.harcourt.com/dictionary, Academic Press Inc. (1996)
/Aidun 02/ D. K. Aidun, J. J. Domey, and G. Ahmadi, Digital Simulations of a Station-
ary and a Linear Weld Metall. Mater. Trans. B, 33B, (2002), 101-110.
/Ampuero 91/ J. Ampuero, A.F.A. Hoadley, M. Rappaz, Numerical and Experimental Study
of Microporosity Evolution during the Solidification of Metallic Alloys,
Modeling of Casting, Welding and Advanced Solidification Processes
(MCWASP) V ed. M. Rappaz, M. R. zg, and K.W. Mahin, (Warrendale,
PA: TMS, 1991), 683-690.
/Anson 99/ J.P. Anson, J.E. Gruzleski, Effect of Hydrogen Content on Relative Shrink-
age and Gas Microporosity in Al-7% Si Casting, ASF Transactions 135, 26
(1999), 135-142.
/Atwood 00a/ R.C. Atwood, S. Sridhar, W. Zhang, and P.D. Lee, Diffusion-Controlled
Growth of Hydrogen Pores in Aluminium-Silicon Castings: In-Situ-
Observation and Modelling, Acta Mater., 48-2 (2000), 405-417.
/Atwood 00b/ R.C. Atwood, P.D. Lee, A Combined Cellular Automaton and Diffusion
Model for the Prediction of Porosity Formation during Solidification,
MCWASP IX, ed. P. R. Sahm, P. N. Hansen, J. G. Conley, (Aachen: Shaker,
2000), 2-9.
/Baehr 98/ H.D. Baehr, K. Stephan, Wrme- und Stoffbertragung, Springer-Verlag,
Berlin, Heidelberg etc. (1998).
/Barkhudarov M.R. Barkhudarov, Computer Modelling of Solidification of Pure Metals and
95a/ Alloys, PhD Thesis, The University of Sheffield, Sheffield, (1995).
/Barkhudarov M.R. Barkhudarov, Enhancements to Heat Transfer and Solidification
95b/ Shrinkage Models in FLOW-3D(R), Flow Science Technical Note FSI-95-
TN43, July (1995).
/Barkhudarov 97/ M.R. Barkhudarov, Is Fluid Flow Important for Predicting Solidification?,
Proceedings of the Solidification Processing 97 Conference, Sheffield, U.K.,
July 7-10 (1997), 181-184.
/Beckermann 88/ C. Beckermann, and R. Viskanta, Double-Diffusive Convection during Den-
dritic Solidification of a Binary Mixture, PHC Physico Chemical Hydrody-
namics, 10 (1988), 195-213.
264 References
616.
/Bhat 96/ M. S. Bhat, D. R. Poirier, J. C. Heinrich, Permeability for Cross Flow through
Columnar-Dendritic Alloys, Metall. Mater. Trans. B, 26B, October (1996),
1049-1056.
/Block-Bolten 84/ A. Block-Bolten, and T.W. Eagar, Metal Vaporization from Weld Pools,
Metall. Trans. B, 15B, September (1984), 461-469.
/Bhmer 95/ W.F.A. Bhmer, Numerische Simulation und Experimentelle Validierung
Makroskopischer Seigerungen im Stahlgu, PhD Thesis, Gieerei-Institut,
RWTH Aachen, (1995).
/Bhmer 97/ W.F.A. Bhmer, and P.R. Sahm, Numerische Simulation und experimentelle
Validierung makroskopischer Seigerungen im Stahlgu, Giessereiforschung,
49-1 (1997) 1-10.
/Bounds 98/ S. M. Bounds, G. J. Moran, K. A. Pericleous, and M. Cross, Finite Volume
Modeling of Shrinkage Porosity Formation in Metal Casting, MCWASP VIII,
ed. B. Thomas, C. Beckermann (Warrendale, PA: TMS, 1998), 857-864.
/Braccini 00/ M. Braccini, C.L. Martin, M. Sury, and Y. Brchet, Relation between Mushy
Zone Rheology and Hot Tearing Phenomena in Al-Cu Alloys, MCWASP IX,
ed. P. R. Sahm, P. N. Hansen, J. G. Conley, (Aachen: Shaker, 2000), 18-24.
/Brent 88/ A.D. Brent, V.R. Voller, K.J. Reid, Enthalpy-Porosity Technique for Model-
ing Convection-Diffusion Phase Change: Application to the Melting of a Pure
Metal Numerical Heat Transfer, 13 (1988), 297-318.
/Bussmann 99/ M. Bussmann. J. Mostaghimi and S. Chandra, On a Three-Dimensional Vol-
ume Tracking Model of Droplet Impact, Physics of Fluids, vol. 11 (1999),
1406-1417.
/Celentano 98/ D. Celentano, Shrinkage Prediction in the Thermomechanical Analysis of
Castings, MCWASP VIII, ed. B. Thomas, C. Beckermann (Warrendale, PA:
TMS, 1998), 819-826.
/Chai 95/ G. Chai, L. Bckerud, and L. Arnberg, Study of Dendrite Coherency in Al-Si
Alloys during Equiaxed Dendritic Solidification, Z. Metallkd., 86 (1995), 54-
59.
/Chakraborty 01/ S. Chakraborty, S. Sarkar, and P. Dutta, Effect of Constitutional Supercooling
on the Numerical Solution of Species Concentration Distribution in Laser Sur-
face Alloying, Metall. Mater. Trans. B, 32B, October (2001), 969-972.
/Chan 84/ C. Chan, J. Mazumder, M. M. Chen, A Two-Dimensional Transient Model for
Convection in Laser Melted Pools, Metall. Trans. A, 15A, December (1984),
266 References
2175-2184.
/Chang 96/ S. Chang, D. M. Stefanescu, A Model for Macrosegregation and its Applica-
tion to Al-Cu Castings, Metall. Mater. Trans. A, 27A, September (1996),
2708-2721.
/Chen 91a/ X.G. Chen , and S. Engler, Wasserstoff und Porositt in Aluminium-Silicium-
und Aluminium-Magnesium-Legierungen / Teil I: Aluminium-Silicium-
Legierungen, Metall, 45-10, (1991), 995-1000.
/Chen 91b/ X.G. Chen , and S. Engler, Wasserstoff und Porositt in Aluminium-Silicium-
und Aluminium-Magnesium-Legierungen / Teil II: Aluminium-Magnesium-
Legierungen und Diskussion ber die Porenbildung, Metall, 45-12, (1991),
1225-1231.
/Chen 93/ J.H. Chen, and H.L. Tsai, Inverse Segregation for a Unidirectional Solidifica-
tion of Aluminum-Copper Alloys, Int. J. Heat Mass Transfer, Vol. 36 (1993),
3069-3075.
/Chen 94/ X.G. Chen , and S. Engler, Blasenbildung und Porenmorphologie in erstar-
renden Aluminiumgustcken, Metall, 48-3, (1994), 276-283.
/Chen 96/ X.G. Chen, and J.E. Gruzleski, Influence of Melt Cleanliness on Pore Forma-
tion in Aluminium-Silicon Alloys, Int. J. Cast Metals Res., 9, (1996), 17-26.
/Chiang 92a/ K.C. Chiang, and H.L. Tsai, Shrinkage-Induced Fluid Flow and Domain
Change in Two-Dimensional Alloy Solidification, Int. J. Heat Mass Transfer,
Vol. 35 (1992), 1763-1770.
/Chiang 92b/ K.C. Chiang, and H.L. Tsai, Interaction between Shrinkage-Induced Fluid
Flow and Natural Convection during Alloy Solidification, Int. J. Heat Mass
Transfer, Vol. 35 (1992), 1771-1778.
/Choi 87/ Choi, R. Grief, and M. Salcuden, Numer. Heat Transf., vol. 11 (1987), 477.
/Choo 92a/ R.T.C. Choo, J. Szekely, Vaporization Kinetics and Surface Temerature in a
Mutually Coupled Spot Gas Tungsten Arc Weld and Weld Pool Welding Re-
search Supplement, March (1992), 77s-93s.
/Choo 92b/ R.T.C. Choo, J. Szekely, and S.A. David, On the Calculation of the Free Sur-
face Temperature of Gas-Tungsten-Arc Weld Pools from First Principles: Part
II. Modeling the Weld Pool and Comparison with Experiments, Metall. Trans.
B, 23B, June (1992), 371-384.
/Clarke 97/ J. Clarke, D.C. Weckman, and H.W. Kerr, Predicting the Columnar-To-
Equiaxed Grain Transition in Gas Tungsten Arc Welds in Aluminium-Copper
Alloys, (Presented to the: 36th Annual Conference of Metallurgists of CIM,
267
1998), 305-312.
/Dupouy 98/ M.D. Dupouy, D. Camel, F. Botalla, J. Abadie, and J.J. Favier, Experimental
Study of the Columnar to Equiaxed Transition in a refined Al-4wt%Cu Alloy
under Diffusive Transport Conditions, MCWASP VIII, ed. B. Thomas, C.
Beckermann (Warrendale, PA: TMS, 1998), 415-423.
/Dushman 62/ S. Dushman, Scientific Foundations of Vacuum Technique, (New York: John
Wiley & Sons, 1962)
/Ehlen 98/ G. Ehlen, A. Schweizer, A. Ludwig, P.R. Sahm, Macroscopic Modeling of
Marangoni Flow and Solute Redistribution during Laser Welding of Steel
MCWASP VIII, ed. B. Thomas, C. Beckermann (Warrendale, PA: TMS,
1998), 289-296.
/Ehlen 00a/ G. Ehlen, A. Ludwig, P.R. Sahm, Influence of Convection and Surface Effects
on Macrosegregations in Eutectic and Peritectic Systems Mater. Sci. Forum,
Vols. 329-330 (2000), 105-110.
/Ehlen 00b/ G. Ehlen, A. Schweizer, A. Ludwig, P.R. Sahm, Free Surface Model to Pre-
dict the Influence of Shrinkage Cavities on the Solute Redistribution in Cast-
ings, MCWASP IX, ed. P. R. Sahm, P. N. Hansen, J. G. Conley, (Aachen:
Shaker, 2000), 632-639.
/Ehlen 00c/ G. Ehlen, A. Schweizer, A. Ludwig, P.R. Sahm, Formation of Shrinkage
Cavity and Solute Redistribution during the Casting of an Al-7wt%Si Ingot,
Materials Week 2000 Proceedings CD-ROM, (Mnchen, 2000), available via
www.materialsweek.org
/Ehlen 02a/ G. Ehlen, A. Ludwig, P.R. Sahm, Influence of Welding Power and Welding
Duration on the Shape of the Weld Pool: Transient Effects , Mathematical
Modeling of Weld Phenomena 6, ed. H. Cerjak, IOM Communications, Lon-
don (2002), 161-191.
/Ehlen 02b/ G. Ehlen, A. Ludwig, P.R. Sahm, A. Bhrig-Polaczek, Simplified Two-Phase-
Flow Model for the Simulation of Convection Effects during Solidification
with Free Surfaces, Materials Week 2002 Proceedings CD-ROM, (Mnchen,
2002), available via www.materialsweek.org
/Ehlen 03a/ G. Ehlen, A. Ludwig, P.R. Sahm, A. Bhrig-Polaczek, Split-Solid-Model to
Simulate the Formation of Shrinkage Cavities and Macrosegregations in Steel
Casting, MCWASP X., ed. D.M. Stefanescu, J.A.Warren, M.R. Jolly, and
M.J.M. Krane, (Destin, Florida: TMS, 2003), 285-292.
/Ehlen 03b/ G. Ehlen, A. Ludwig, and P.R. Sahm, Simulation of Time-Dependent Pool
Shape during Laser Spot Welding: Transient Effects, Metall. Mater. Trans. A,
269
/Gandin 98/ Ch.-A. Gandin, T. Jalanti, and M. Rappaz, Modeling of Dendritic Grain
Structures, MCWASP VIII, ed. B. Thomas, C. Beckermann (Warrendale, PA:
TMS, 1998), 363-374.
/Gandin 00a/ C.A. Gandin, From Constrained to Unconstrained Growth during Directional
Soldification Acta mater., 48 (2000), 2483-2501.
/Gandin 00b/ Ch.-A Gandin, From Constrained to Unconstrained Growth during Directional
Solidification of Dendritic alloys, MCWASP IX, ed. P.R. Sahm, P.N. Hansen,
J. G. Conley, (Aachen: Shaker, 2000), 473-480.
/Gao 98/ J.W. Gao, and C.Y. Wang, Effects of Grain Transport on Columnar to Equi-
axed Transition in Dendritic Alloy Solidification, MCWASP VIII, ed. B.
Thomas, C. Beckermann (Warrendale, PA: TMS, 1998), 425-432.
/Ghosh 01/ A. Ghosh, Segregation in Cast Products, Sdhan, Vol. 26, parts 1 & 2, Feb-
ruary & April (2001), 5-24.
/Giamei 70/ A.F. Giamei, and B.H. Kear, Metall. Trans. vol. 1 (1970), 2185-2192.
/Goyeau 98/ B. Goyeau, T. Benihaddadene, D.Gobin, and M. Quintard, Numerical Deter-
mination of the Permeability Tensor in a Columnar Dendritic Mushy Zone,
MCWASP VIII, ed. B. Thomas, C. Beckermann (Warrendale, PA: TMS,
1998), 353-360.
/Goyeau 99/ B. Goyeau, T. Benihaddadene, D. Gobin, and M. Quintard, Numerical Calcu-
lation of the Permeability in a Dendritic Mushy Zone, Metall. Mater. Trans.
B, 30B, August (1999), 613-622. ERRATUM: Metall. Mater. Trans. B, 30B,
December (1999), 1107.
/Gu 97/ J.P. Gu, C. Beckermann, and A.F. Giamei, Motion and Remelting of Dendrite
Fragments during Directional Solidification of a Nickel-Base Superalloy,
Metall. Mater. Trans. A, 28A, July (1997), 1533-1542.
/Gu 99/ J.P. Gu and C. Beckermann, Simulation of Convection and Macrosegregation
in a Large Steel Ingot, Met. Mater. Trans. A, 30A May (1999), 1357-1366.
/Guan 91/ J. Guan, and P.R. Sahm, Berechnung der Restspannungen abkhlender
Gustcke mit Hilfe der Finiten Elemente Methode, Gieereiforschung, 43
(1991), 10-17.
/Guan 92/ J. Guan, and P.R. Sahm, Numerische Untersuchung der Thermischen Span-
nungen in realen 3D-Gubauteilen, Gieerei, 79 (1992), 318-322.
/Guan 94/ J. Guan, Entwicklung eines 3D-FEM-Programmoduls zur Simulation Ther-
mischer Eigenspannungen in Gieprozessen sowie seine Anwendung fr die
Optimierung von Gubauteilen, PhD Thesis, Gieerei-Institut, RWTH
271
2000), 640-647.
/Kao 95/ S.T. Kao, E. Chang, and S.C. Horng, An Experimental Study of Porosity
Formation in Long-Freezing-Range A206 Alloy Plate and Taper Castings,
Cast Metals, 8, 1 (1995), 21-26.
/Kato 85/ H. Kato, and J.R. Cahoon, Inverse Segregation in Directionally Solidified Al-
Cu-Ti Alloys with Equiaxed Grains, Metall. Trans. A, 16A, April (1985),
579-587.
/Katzarov 96/ I.H. Katzarov, and J.R. Popov, Pore Formation in Hot Spots, Int. J. Heat
Mass Transfer, vol. 39, 14, September (1996), 2861-2867.
/Kim 75/ C.S. Kim, Thermophysical Properties of Stainless Steels, Argonne National
Laboratory, Report No. ANL-75-55 (1975)
/Ko 95/ Y.K. Ko, V. Sahai, J.T. Berry, and R.A. Overfelt, Prediction of Porosity in
Cast Equiaxed Alloy 718, MCWASP VII, ed. M. Cross, J. Campbell (War-
rendale, PA: TMS, 1995), 731-738.
/Kou 85/ S. Kou, and D. K. Sun, Fluid Flow and Weld Penetration in Stationary Arc
Welds, Metall. Trans. A, 16A, February (1985), 203-213.
/Kou 86/ S. Kou, and Y. H. Wang, Computer Simulation of Convection in Moving Arc
Weld Pools, Metall. Trans. A, 17A, December (1986), 2271-2277.
/Krane 95/ M.J.M. Krane, F.P. Incropera, Analysis of the Effect of Shrinkage on Macro-
segregation in Alloy Segregation Met. Mater. Trans. A, 26A September
(1995), 2329-2339.
/Krane 96/ M.J.M. Krane, F. P. and Incropera, A scaling analysis of the unidirectional
solidification of a binary alloy, Int. J. Heat Mass Transfer 39 (17) (1996),
3567-3579.
/Krane 97a/ M. J. M. Krane, F.P. Incropera, D.R. Gaskell, Solidification of ternary metal
alloysI. Model development, Int. J. Heat Mass Transfer 40 (16) (1997),
3827-3835.
/Krane 97b/ M. J. M. Krane, F. P. Incropera, Solidification of ternary metal alloys II.
Predictions of convective phenomena and solidification behavior in Pb-Sb-Sn
alloys Int. J. Heat Mass Transfer 40 (16) (1997), 3837-3847.
/Kreutz 99/ W. Kreutz, Lehrstuhl fr Lasertechnik, RWTH Aachen, private communication
/Kubo 85/ K. Kubo, and R.D. Pehlke, Mathematical Modeling of Porosity Formation in
Solidification, Metall. Trans. B, vol. 16B (1985), 359-366.
/Kuznetsov 95/ A.V. Kuznetsov, K. Vafai, Development and Investigation of Three-Phase
273
/Ohno 87/ A. Ohno, Solidification, The Separation Theory and its Practical Applica-
tions, (New York: Springer, 1987)
/Oldenburg 92/ C.M. Oldenburg, and F.J. Spera, Hybrid Model for Solidification and Con-
vection, Numer. Heat Transfer part B, Vol 21, pp. 217-229, 1992.
/Olsson 86/ A. Olsson, R. West, R. Fredriksson, Macrosegregation in Ingots Scand. J.
Met., vol. 15 (1986), 104-112.
/Oreper 86/ G.M. Oreper, J. Szekely, T.W. Eagar, The Role of Transient Convection in
the Melting and Soldification in Arc Weldpools Metall. Trans. B, 17B (1986),
735-744.
/Palmer 01/ T.A. Palmer, and T. DebRoy, Modelling of Nitrogen Dissolution during the
GTA Welding of Iron and Steels, Mathematical Modeling of Weld Phenom-
ena 5, ed. H. Cerjak, (London: IOM Communications, 2001), 95-122.
/Pasandideh-Fard M. Pasandideh-Fard, R. Bhola, S. Chandra and J. Mostaghimi, Deposition of
98/ Tin Droplets on a Steel Plate: Simulations and Experiments, Int. J. Heat Mass
Transfer, Vol. 41, (1998), 2929-2945.
/Pasandideh-Fard M. Pasandideh-Fard, J. Mostaghimi and S. Chandra, Modeling Sequential
99a/ Impact of Two Molten Droplets on a Solid Surface, The ILASS-Americas
12th Annual Conference on Liquid Atomization and Spray, Indianapolis, IN,
May 16-19 (1999).
/Pasandideh-Fard M. Pasandideh-Fard, S.D. Aziz, S. Chandra, and J. Mostaghimi, Surface
99b/ Cooling by an Impinging Water Drop, paper HTD99-89, The 33rd National
Heat Transfer Conference, Albuquerque, NM, August 15-17 (1999).
/Pasandideh-Fard M. Pasandideh-Fard, M. Bussmann, S. Chandra, and J. Mostaghimi, On a 3D
00/ Computational Model of Free Surface Flows including Internal Obstacles,
Proceedings of the 8th Annual Conference of the CFD Society of Canada,
Montreal (2000), 603-610.
/Patankar 80/ S. V. Patankar, Numerical Heat Tranfer and Fluid Flow, (Washington, D.C.:
Hemisphere, 1980)
/Patterson 62/ W. Patterson, and S. Engler, Zusammenhnge zwischen Volumenfehlern und
Erstarrungstypen bei Gukrpern aus Aluminium-Silicium-Legierungen,
Aluminium, 38-8, (1962), 507-519.
/Pavlyk 99/ V. Pavlyk, and U. Dilthey, Using FIDAP for Temperature and Flow Simula-
tion in Weld Pool during Arc Welding with the Emphasis of Free Surface Phe-
nomena, Report presented at the FLUENT User Meeting, Viernheim, Sep-
tember 20.-21. (1999)
276 References
/Weckman 98/ D.C. Weckman, Modeling Thermofluids Phenomena in Arc Welds, (Keynote
Presentation to: 5th International Conference on Trends in Welding Research,
Pine Mountain, Georgia, USA, June 1998).
/Wendt 00/ J. Wendt, R. Arola, M. Verho, J.Rintamki, and I. Rytky, A Criterion for Po-
rosity Formation in Steel Castings, MCWASP IX, ed. P.R. Sahm, P.N. Han-
sen, J. G. Conley, (Aachen: Shaker, 2000), 110-117.
/Windelberg 00/ D. Windelberg, and S. Ramm, Evaluation of Shrinkage, MCWASP IX, ed.
P.R. Sahm, P.N. Hansen, J. G. Conley, (Aachen: Shaker, 2000), 118-125.
/Winkler 98/ C. Winkler, G. Amberg, H. Inoue, T. Koseki, A Numerical and Experimental
Investigation of Quantitatively Different Weld Pool Shapes, Mathematical
Modeling of Weld Phenomena 4, ed. H. Cerjak, IOM Communications, Lon-
don (1998), 37-69.
/Xiao 97/ Y.H. Xiao, and G. den Ouden, Measurement of Surface Tension of Liquid
Metals and Alloys under Arc Welding Conditions, Materials Science and
Technology, 13, September (1997), 791-794.
/Xu 91/ D. Xu, and Q. Li, Numerical Method for Solution of Strongly Coupled Binary
Alloy Solidification Problems, Num. Heat Transfer A, Vol. 20 (1991), 181-
201.
/Xu 01/ D. Xu, A Unified Microscale-Parameter Approach to Solidification-Transport
Phenomena-Based Macrosegregation Modeling for Dendritic Solidification:
Part I. Mixture Average-Based Analysis, Metall. Mater. Trans. B, 32B, De-
cember (2001), 1129-1141.
/Yang 00/ B. J. Yang, and D. M. Stefanescu, Solidification Modelling with Grain
Tracking, MCWASP IX, ed. P.R. Sahm, P.N. Hansen, J. G. Conley, (Aachen:
Shaker, 2000), 582-589.
/Yang 01/ W. Yang, W. Chen, K. Chang, S. Mannan, and J. de Barbadillo, Freckle Crite-
ria for the Upward Directional Soldification of Alloys Metall. Mater. Trans.
A, 32A (2001), 397-406.
/Zacharia 89/ T. Zacharia, A. H. Eraslan, D. K. Aidun, and S.A. David, Three-Dimensional
Transient Model for Arc Welding Process, Metall. Trans. B, 20B, October
(1989), 645-659.
/Zacharia 91/ T. Zacharia, S.A. David, and J.M. Vitek, Effect of Evaporation and Tem-
perature Dependent Material Properties on Weld Pool Development, Metall.
Trans. B, 22B, (1991), 233-241.
/Zacharia 95/ T. Zacharia, J.M. Vitek, J.A. Goldak, T.A. DebRoy, M. Rappaz, and H. K. D.
282 References
School education
1974 1978 : Primary school St. Balbina, Wrselen
1978 1987 : Stdtisches Gymnasium Wrselen
Graduation : Abitur
Studies
1988 1995 : Studies of physics at Aachen University, Germany
Main focus : Solid state physics, Fluid mechanics, Cluster physics
Graduation : Diploma
1991 1992 : Studies in Grenoble, France (Physics, Italian)
1989 1999 : Studies of romance literature and linguistics at Aachen University
Main focus : Andr Malraux, Dino Buzzati, french 19th cent. short novel and novel
Graduation : Magister Artium
Professional Experience
1993 1997 : Student worker / graduate assistant, Foundry-Institute
Aachen University
1997 2003 : Research associate, Foundry-Institute
Main focus : Development of numerical simulation software for casting and
welding; automatical optimization of casting processes
1999 / 2000 : Research visit Edinburgh Parallel Computing Centre
(EPCC) and Heriott Watt University, Edinburgh, Schottland
since 2003 : R&D department for numerical simulation and process development
at SCHOTT AG, Mainz, Germany
Work area : Model based control and optimization
PhD
1997 2003 : Foundry-Institute, Aachen University
12.02.2004 : Examination summa cum laude
LEBENSLAUF
Name: Georg Ehlen
Geburtsdatum: 20.05.1968
Geburtsort: Wrselen
Schulausbildung
1974 1978 : Grundschule St. Balbina, Wrselen
1978 1987 : Stdtisches Gymnasium Wrselen
Schulabschluss : Abitur
Studium
1988 1995 : Studium der Physik an der RWTH Aachen
Schwerpunkte : Festkrperphysik, Strmungslehre, Clusterphysik
Abschluss : Diplom
1991 1992 : Auslandsstudium in Grenoble, Frankreich (Physik, Italienisch)
1989 1999 : Studium der Romanischen Literatur- und Sprachwissenschaft an der
RWTH Aachen
Schwerpunkte : Andr Malraux, Dino Buzzati, Novelle und Roman im franz. 19. Jh.
Abschluss : Magister Artium
Berufliche Ttigkeiten
1993 1997 : Studentische / Wissenschaftliche Hilfskraft am Gieerei-Institut der
RWTH Aachen
1997 2003 : Wissenschaftlicher Mitarbeiter am Gieerei-Institut
Schwerpunkte : Simulationsentwicklung Gieen und Schweien, Automatische
Optimierung von Gieprozessen
1999 / 2000 : Forschungsaufenthalte am Edinburgh Parallel Computing Centre
(EPCC) und der Heriott Watt University, Edinburgh, Schottland
seit 2003 : Wissenschaftlicher Referent Mathematische Simulation und Ver-
fahrensentwicklung bei SCHOTT AG, Mainz
Arbeitsgebiet : Modellbasierte Regelung und Optimierung
Promotion
1997 2003 : Gieerei-Institut der RWTH-Aachen
12.02.2004 : Prfung zum Doktor der Naturwissenschaften mit Auszeichung