Entdecken Sie eBooks

Kategorien

Entdecken Sie Hörbücher

Kategorien

Entdecken Sie Zeitschriften

Kategorien

Entdecken Sie Dokumente

Kategorien

0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

78 Ansichten352 Seitene-book

© © All Rights Reserved

PDF, TXT oder online auf Scribd lesen

e-book

© All Rights Reserved

Als PDF, TXT **herunterladen** oder online auf Scribd lesen

0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

78 Ansichten352 Seitene-book

© All Rights Reserved

Als PDF, TXT **herunterladen** oder online auf Scribd lesen

Sie sind auf Seite 1von 352

IN CASTING AND WELDING

Rheinisch-Westflischen Technischen Hochschule Aachen

genehmigte Dissertation

vorgelegt von

Diplom-Physiker, M.A.

Georg Ehlen

Wrselen

Univ.-Prof. Dr.rer.nat. Andreas Ludwig

Univ.-Prof. Dr.-Ing. Andreas Bhrig-Polaczek

Gieerei-Institut: Forschung, Entwicklung, Ergebnisse

Band 45

Georg Ehlen

Complex Convection Effects during Solidification

in Casting and Welding

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.

All rights reserved. No part of this publication may be reproduced, stored

in a retrieval system, or transmitted, in any form or by any means, electronic,

mechanical, photocopying, recording or otherwise, without the prior

permission of the publishers.

Printed in Germany.

ISBN 3-8322-2886-1

ISSN 1435-6198

Phone: 0049/2407/9596-0 Telefax: 0049/2407/9596-9

Internet: www.shaker.de eMail: info@shaker.de

Kasimir Malewitsch (18781935): Schwarzes Quadrat,

um 19141915, l auf Leinwand, 79,5 x 79,5 cm, Moskau, Tretjakow Galerie

dass die Wahrheit sich selber Bahn breche?

Ernst Bloch (1885 1977)

doch Streit ist der Weg.

Jrgen Habermas (*1929)

Danksagung / Acknowledgement

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.

In order to simulate the complex convection effects which are the main subject of this work, a num-

ber of new basic models describing single aspects had to be developed. The main developments are:

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 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.

The new basic models have been integrated together with state of the art models based on the vol-

ume averaging model by /Ni 91/, describing fluid flow and convective and diffusive transport phe-

nomena. Two 2D software packages have been developed which contain different combinations of

coupled models:

EL2D (current version el2d9.1). The software is originally based on work by /Schneider 95a/

and /Bhmer 97/ and was used for the calculation of macrosegregation phenomena in casting. It

was extended by the author by early versions of the new solidification algorithm and the solidi-

fication of moving free surfaces as well as by sophisticated models for laser and GTA welding.

It is mainly used for welding simulations in cylindrical co-ordinates.

SoliCon (current version solicon2.0), which is based on completely new concepts and sets of

equations. It contains all new models, i.e. the algorithm for solidification of moving free sur-

faces, split solid model, porosity formation model, and new solidification algorithm. It is mainly

used for casting simulations in cartesian co-ordinates.

A more detailed description of the two software packages can be found in App. 9.2.

The numerical models allow the coupled simulation of thermosolutal convection, buoyancy driven

free surface movement, shrinkage flow, Marangoni flow, and flow due to electromagnetic forces

and have been used to simulate the following complex convection effects:

0.1 Introduction III

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.

The integrated models have been applied to the following scientific problems

Deep shrinkage cavities in ingot casting have a large impact on the final macrosegregation pat-

tern. The correct prediction of this influence can help to improve the criteria for correct riser

dimensioning. The model was applied to the casting of an unalloyed steel ingot with riser (Soli-

Con), Fig. 0.1 a), and the results compared to experimental results from /Bhmer 97/.

Prediction and experimental validation of shrinkage cavity formation in a cylindrical Al-7wt%Si

ingot (EL2D); the model allows to consider the impact of the alloy composition on the surface

shape.

Simultaneous but separate prediction of hydrogen and shrinkage porosities during casting of Al-

7wt%Si (SoliCon); experimental validation for pure shrinkage porosity formation; The new

model can be a tool to determine the type of experimentally observed porosities.

IV Chapter 0 Executive summary

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/.

As the phenomena related to convection are various and often complex, a great effort has been

made to describe and explain the different mechanisms. Detailed overviews have been written on

macrosegregation types like normal and inverse segregation, segregation due to gravity and

sedimentation of floating globulitic grains, channel segregates etc.

shrinkage defects like external and internal shrinkage cavities, hydrogen and shrinkage porosi-

ties, linear contraction

formation and movement of floating globulitic grains and two-phase flow

influence of fluid flow on solidification and remelting

the formation of different weld pool shapes as functions of different parameters

concentration changes in weld pools

Alongside the description of the convection phenomena are comprehensive reviews of the state of

the art in experimental investigation and mathematical and numerical modelling for each of these

areas. In every case this is followed by a brief summary of this works contribution in this area.

The review of numerical modelling has shown that for most subjects there is an almost infinite

number of high level publications available, as far as the modelling of single convection effects is

concerned. Integrated models are highly developed in the areas of macrosegregation modelling in

casting, porosity prediction and welding. Models with an even higher integration level, e.g. com-

bining segregation and porosity formation, are very rare. Most models with promising approaches

are commercial models, and therefore not really well documented. For the target casting process

presented in this work, i.e. the formation of deep shrinkage cavities in steel casting and the calcula-

tion of their influence on the final macrosegregation pattern, no satisfactory approach has been

published yet. Even commercial solutions seem to be unavailable. The same seems to be true for the

simultaneous but separate prediction of hydrogen and shrinkage porosities or internal shrinkage

cavities, and especially for the full coupling of porosity prediction, formation of shrinkage cavities

and formation of macrosegregations. In the context of this work this coupling has not yet been

demonstrated by examples because the porosity formation model is a very recent development, but

0.4 Mathematical model (SoliCon) V

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.

This section describes the framework of mathematical and physical models developed for use in the

casting simulation code SoliCon. The models used in the welding simulation code EL2D are basi-

cally those described in /Schneider 93,95b/. They were extended by a laser and GTA welding model

(Marangoni + Lorentz convection), cylindrical co-ordinates and by early versions of the new solidi-

fication algorithm and the algorithm for the solidification of moving free surfaces.

Except of the momentum equation, which has been written in 2D formulation, all equations have

been developed in vector notation and can therefore be used for full 3D treatment. The extension of

the momentum equation to 3D is some work, but should be easy enough to do.

One central idea of the model is the introduction of mobile solid fractions without introducing addi-

tional solid momentum equations. This is achieved by treating the mobile solid fraction thermody-

namically as solid, but hydrodynamically as liquid with solid density. The concept will be called

Split Solid Model (SSM).

The mathematical model considers nine phases:

liquid: q=l shrinkage porosities: q = pore shrink

mobile solid - phase: q = move hydrogen porosities: q = pore hydro

mobile solid - phase: q = move obstacles: q = obst

stationary solid - phase: q = dead void: q = void

stationary solid - phase: q = dead

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

v v v v v v v v v v

v := v move = v move = vl v dead = v dead = v pore shrink = v pore shrink = vobst = vvoid = 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

q = move / dead

) c P eff move / dead move / dead eff move / dead := ( q q cP q

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

The equation has been developed using the assumption of a common constant effective viscosity

eff move for all mobile phases. The complete equation is:

0.4 Mathematical model (SoliCon) VII

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.

The model uses the lever-rule, i.e. it assumes complete solutal equilibrium inside each phase and

uses equilibrium phase diagrams. This leads to solid concentration equations

Ci = li Cli (0.4a)

C = C

i i

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.

Obviously the sum over all phases must be 1:

l = 1 move move dead dead obst pore shrink pore hydro void (0.6)

and the fraction of the obstacle phase is constant and results from the initial conditions:

obst = const. (0.7)

The well known VOF equation with partial cell treatment, /Torrey 85/, has been extended to the

case of solidification (time dependent obstacles) and time dependent liquid densities. These modifi-

cations allow us to use the equation to describe complex interactions of shrinkage flow and the so-

lidification of moving free surfaces.

v

t

( open eff open Fmove ) + ( open eff open vl Fmove ) = Fmove ( closed eff closed )

v

t

(0.8)

move

where Fmove :=

open

A new model has been developed which is able to quantitatively predict the formation of hydrogen

and shrinkage porosities as well as internal shrinkage cavities. Of course the prediction hydrogen

porosities only makes sense for aluminium alloys.

There are two different mechanisms responsible for the formation of hydrogen porosities on the one

hand and shrinkage induced porosities and cavities on the other. Hydrogen porosities form by re-

jection of hydrogen bubbles during solidification and cooling due to the reduced hydrogen solubil-

ity in the solid phase. If hydrogen bubbles form they push the local liquid and mobile solid phases

away because they exert a high pressure. Shrinkage porosities form inside an existing dendritic

network by vaporization of liquid alloy components if the local pressure drops so far that their par-

tial pressure can initiate pore nucleation against the remaining local pressure. Internal macroscopic

shrinkage cavities are formed by the same mechanism as shrinkage porosities. The difference is that

they form if no dendritic network or solid phase is present at the moment when pore nucleation is

initiated.

0.4 Mathematical model (SoliCon) IX

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

Two additional phase fractions are defined for reasons of simplicity. As already stated, mobile and

stationary solid are thermodynamically treated as one single phase, therefore

: = move + dead : = move + dead

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)

The last two equations determine how the solid that forms is split into the mobile and stationary

part. The Split Solid Model (SSM) has been developed to prevent small amounts of stationary solid

which are forming close to the free surface in early stages of solidification from blocking the further

surface movement and the formation of a deep shrinkage cavity. For small overall solid fractions

( + ) all solid is mobile, for high solid fractions all solid is stationary. For solid fractions of

0 mv dd mv dd , cf. Fig. 0.2, the fraction of mobile solid varies smoothly from all mobile to all

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.

A number of source terms has been added for the description of welding processes. The momentum

equations have been extended by models for surface tension driven Marangoni convection and

electromagnetic forces for GTA welding. The temperature dependent Marangoni coefficient is cal-

culated using the semi-empirical formula by /Sahoo 88/:

K a H 0

= A Rs ln (1 + K seg a s ) seg s s with K seg = S l e H RT

(

(0.14)

0 /( ))

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

The model has been implemented in the framework of a 2D control volume (CVM) code. The code

can be classified as a single domain volume averaging multiphase model. Most of the partial differ-

ential 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 real-

ized using a modified SIMPLER iteration loop.

The aim to have a program which is especially suitable for the calculation of convective solute

transport in casting and welding processes makes great demands on the phase change algorithm (in

XII Chapter 0 Executive summary

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.

The problem to be solved by the solidification algorithm can be described as follows: For each new

time step the discretised differential equations for temperature and solute concentration start from a

point (C , T ) in the phase diagram and yield a point (C , T ). The position of this new point

0 0

(C , T ) depends on

tational cell in a calculation of a stationary

welding process with Fe-0.42wt%C. When the

cell starts to melt, solute is removed by convec-

tion and the mean concentration is reduced.

During the primary -melting the laser source is

switched off and the cell starts to cool down. At

the end of the peritectic resolidification the mean

concentration of the cell has been reduced so far

that some -phase remains in the solid. The new

solidification algorithm now correctly treats the

(equilibrium-) solid state transformation inside

the + -phase area.

0.5 Algorithms XIII

0 0

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.

module for primary -solidification

based on /Prakash 89/. The other solidifi-

cation modules work similarly.

0.5 Algorithms XV

The initial step of the solution strategy is to find out the region of the phase diagram where the final

solution may lie and the direction in which the system is likely to develop. This information is

given by two positions of the alloy system inside the phase diagram:

1. The position (C , T ) at the end of the last time step.

0 0

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.

The VOF algorithm consists of an advection equation, which uses the velocity results of the SIM-

PLER algorithm to calculate the new filling states of each cell and the surface reconstruction algo-

rithm which uses the new filling states to determine the exact position of the surface line in each

cell. The modified VOF advection equation has been discretised and cast into the well known VOF

donoracceptor scheme, /Torrey 85/, 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 (0.15)

:

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 ]

The surface reconstruction algorithm determines the new position of the surface line inside each

cell, after the advection algorithm has calculated the new filling states. The currently used surface

reconstruction algorithm is similar to the one described in /Nichols 80/. The only important differ-

ence is the fact that the new algorithm takes the volume of liquid, solid and porosities, i.e. the vol-

ume of the cast part, as filled volume. This is necessary to describe liquid and solid surfaces using

one and the same algorithm.

The new porosity formation algorithm is capable of predicting the formation of hydrogen and

shrinkage porosities as well as internal shrinkage cavities.

The shrinkage porosity calculation has been implemented into the iterative solver. At the beginning

of each time step the critical pressure for shrinkage pore formation

pcrit pore shrink = pvapour psurface tension

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.

XVIII Chapter 0 Executive summary

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.

In a volume averaging model it is not trivial to define a criterion to detemine whether microporosi-

ties or macroscopic shrinkage cavities form. As the shrinkage defects in general form in cells where

there is already solid present, it must be decided whether the solid has a planar front or forms a den-

dritic network. Observation of the immediate neighbourhood can help.

A neighbouring cell which contains no stationary solid is a good indicator that the current cell con-

tains 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 soft-

ware. The porosity type is stored in an integer array.

The fraction of newly formed shrinkage porosities

pore hydro = , (0.19)

p T0 t

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.

Correct riser dimensioning in casting is a matter of great economical impact. Risers are responsible

for the soundness of the casting, but they should be kept as small as possible, because all material of

the riser which is not used for feeding is waste material. So far the main criterion for good riser de-

sign has been to make them as small as possible, but large enough that the external shrinkage cavity

which occurs due to feeding does not enter the proper cast part. Experimental and numerical work

by /Bhmer 97/, who have cast and analysed a steel ingot with riser, has demonstrated that this cri-

terion is not sufficient. Spark spectrum analysis of the final macrosegregation pattern shows se-

verely carbon enriched areas directly beneath the shrinkage cavity, which penetrate the proper cast

part. The enriched areas contain up to 130 % more carbon than the initial mixture. The result is a

cast iron area inside a steel part. Apart from the problem that this unwanted modification of local

mechanical properties can lead to failure of the component, the cast part can already be damaged

during the production process when the riser is cut off.

Fig. 0.7 a) shows the geometry which consists of a simple rectangular ingot with rectangular riser,

Fig. 0.7 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. 0.8 c).

The present work continues the numerical investigation by /Bhmer 97/ who have simulated the

solidification of the steel ingot, but not considered the formation of the deep shrinkage cavity, Fig.

0.8 a), b). The current simulations use the new integrated model which allows the description of the

influence of the macroscopic shrinkage cavity on the final macrosegregation pattern.

For most of the calculations the grid configuration, boundary conditions and model parameters

shown in Fig. 0.9 were used. The cast material was a binary Fe-0.42wtC alloy. The carbon content

XX Chapter 0 Executive summary

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.

Fig. 0.8: Results from /Bhmer 97/; comparison of simulated carbon concentration deviation of

GS60 unalloyed steel using a) isotropic, b) anisotropic permeability model with experimental re-

sults from spark spectrum analysis, c); (x) strong macrosegregations at the last point to solidify.

0.6 Applications: Simulations and experiments XXI

tions and parameters of SSM and CGM for steel

ingot with macroscopic shrinkage cavity.

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.

A calculation with solidification of free surface, shrinkage flow and thermosolutal convection, but

without SSM, Fig. 0.10 shows the importance of this model for the simulation of a deep shrinkage

cavity. Without SSM small amounts of stationary solid form close to the surface in early stages of

solidification and block the movement of the free surface. The free surface model and the pressure

solver get into trouble and the calculation crashes.

A simulation with SSM alone, without CGM, is also not able to model the formation of deep

shrinkage cavities. Fig. 0.11 shows a calculation with SSM and thermosolutal convection. All

forming solid is considered to be mobile and sinks down with the fluid. Instead of a solid shell

along all walls a sedimentation zone forms at the bottom. The free surface always remains flat, be-

cause it never contains stationary solid. To form a deep shrinkage cavity the CGM must make some

of the solid stick to the walls

XXII Chapter 0 Executive summary

for the split solid

model: This calcula-

tion crashed after

880 seconds due to a

pressure failure,

when the free sur-

face model tried to

empty a cell which

already contained

small amounts of

stationary solid; a)

solid fraction; light-

est grey: 0 < s 0.1;

b) velocities; c) pres-

a) b) vmax = 0.0018 m/s c) sure; arrow: Position

of pressure failure;

lightest grey: high

pressure values.

vmax = 0.018 m/s vmax = 0.022 m/s vmax = 0.027 m/s vmax = 0.053 m/s

Fig. 0.11: Simulation of steel ingot with SSM, but without CGM; velocity vectors and development

of the solid fraction.

0.6 Applications: Simulations and experiments XXIII

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.

Thermosolutal convection seems to be important for preventing the premature blockage of the

feeding path. If the SSM takes care that the solid close to the surface remains mobile, thermosolutal

convection is responsible for removing this solid by sedimentation. Otherwise the mobile solid re-

mains in the top part of the riser until the solid fraction becomes high enough such that stationary

solid forms. In this case the feeding path is blocked early and the surface is deadlocked. Fig. 0.12

compares two calculation, both with SSM and CGM, but one with (too slow) shrinkage flow only

(left half) and one with thermosolutal convection (right half).

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 1, continued on following page)

0.6 Applications: Simulations and experiments XXV

deviation [%]

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 [%]

Fig. 0.14: Shrinkage cavity formation of Fe-0.42wt%C ingot; comparison of final macrosegregation

patterns of a) the new results of the simulation with SSM, CGM, shrinkage and thermosolutal con-

vection with b) the experimental results by /Bhmer 97/.

XXVI Chapter 0 Executive summary

At the very beginning all solid which forms grows as columnar solid fixed to the wall, as postulat-

edby the CGM. When mobile solid starts forming in the liquid, sedimentation creates a fraction

solid gradient in the open flow area. In agreement with experimental observations the last point to

solidify is located beneath the external shrinkage cavity.

The principal solute redistribution effect which can be observed during the whole solidification pro-

cess is the normal segregation, i.e. the depletion of the mushy zone due to thermosolutal flows

passing it. The flows remove enriched interdendritic liquid, which leads to successive enrichment of

the open flow area. In addition the formation of a large positive segregation channel due to shrink-

age flow can be observed (1000 s). Sedimentation of depleted mobile solid leads to the formation of

enriched areas close to the free surface (1400 s).

Fig. 0.14 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.

As a second example the formation of an external shrinkage cavity during solidification of a cylin-

drical Al7wt%Si ingot was simulated using an early version of the free surface algorithm which is

part of the software package EL2D. The surface shape compares excellently to the one which has

been experimentally observed, Fig. 0.15.

To test the porosity formation model which is able to simultaneously but separately predict the for-

mation of hydrogen and shrinkage porosities, a 2D numerical setup was designed which provokes

the formation of both hydrogen and shrinkage porosities. Fig. 0.16 shows the configuration, bound-

ary conditions and some important parameters used for all calculations. A tall ingot of Al-7wt%Si

(physical data from section 9.1.3.1) is cooled uniformly from the bottom to initiate upward direc-

tional solidification and a shrinkage flow towards the bottom. At the same time the ingot is cooled

from the sides by two chills in order to slowly close the feeding path by creating a mushy zone with

increasing solid fraction and to interrupt the shrinkage flow.

A test calculation was made to check and demonstrate some basic properties of the shrinkage po-

rosity formation algorithm. As viscosity is set to zero porosity formation starts just after the feeding

0.6 Applications: Simulations and experiments XXVII

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

tions and some important parameters used for all

porosity calculations.

tents in the liquid: VH = 0.3 cm / 100 g 3

Maximum hydrogen

contents in the solid: VH max = 0.1 cm / 100 g3

shrinkage pore for-

mation: pcrit pore shrink = 0.1 p0

= 1.01310 Pa 4

XXVIII Chapter 0 Executive summary

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.

An experiment was designed to validate the simulation results. Fig. 0.18 shows a cut polished sec-

tion 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 extraction conditions. The size of the

0.6 Applications: Simulations and experiments XXIX

of shrinkage porosity formation in

absence of hydrogen; a) complete

ingot showing the positions of the

three cast iron chills; b) horizontal

clipping, showing the main porosity

b)

area (height: 8 mm); c) large mag-

nification of the main porosity area;

the horizontal size of the ingot is 48

mm.

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.

The coupled calculations with both models switched on gave fascinating results. Calculations with

different permeability conditions in the mushy zone were made (zero viscosity, small viscosity and

large isotropic permeability, realistic viscosity and high isotropic permeability)

Fig. 0.19 shows the porosity distributions for the zero viscosity case after complete solidification of

the bottom part of the casting separately for hydrogen and shrinkage porosities, and both com-

3

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

As expected the whole cast part shows almost equally distributed hydrogen porosities. Less ex-

pected, but likewise easy to explain, is the high hydrogen content 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 critical 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 fraction of

hydrogen porosity grows in a ring-shape towards the middle of the hot spot. If the permeability of

the mushy zone is reduced, small amounts of microporosity are added to the porosity distribution of

the zero viscosity case. They result from the rupture of interdendritic feeding.

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.

Welding processes are processes which are equally governed by convection effects. A number of

comprehensive studies have been carried out to demonstrate the applicability of the models devel-

oped in the context of this work. Three of the studies, two in laser welding, one in GTA welding,

investigate, how convection effects govern the formation of different weld pool shapes. The fourth

study is one of the first investigations on solute redistribution in laser weld pools. The studies pro-

vide data which help to understand basic phenomena and which can be used in the design of future

welding processes.

The set-up used for the calculations, Fig. 0.20, is a stationary welding process on an infinite 6 mm

steel plate. A cylindrical section of this plate with a radius of 9 mm is simulated. The plate is melted

by a stationary heat source with vertical incidence and Gauss-shaped intensity profile. The calcula-

tion domain is an equidistant fixed grid with mesh resolutions between 60x40 and 120x80 square

cells (constant axial and radial grid spacing between 0.15 mm and 0.075 mm) with the cylinder axis

on the right boundary. If not otherwise mentioned the material parameters are those of the binary

Fe-C system, section 9.1.3.2. Table 0.1 shows additional parameters and physical constants used for

all welding calculations in this chapter if not otherwise mentioned. 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.

In the first study the MC is assumed to be constant. This allows a good coverage of some important

parameters of laser welding processes. 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

XXXII Chapter 0 Executive summary

welding calculations; boundary and

initial conditions.

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

-1 -1

Tmax maximum surface temperature 2800 K

convection convective heat transfer coefficient 6.4 W m K )

-2 -1

b emissivity 0.2

b Stefan-Boltzmann constant 5.6710 W m K )

-8 -2 -4

plots in this section. The grey area shows the

mushy zone for the initial composition. The liq-

uidus and solidus temperatures are always given

in the caption of the corresponding figure. In this

case, Fe-0.1wt%C, Tliq = 1801 K, Tsol = 1766 K;

the other isolines are identical for all composi-

tions.

0.6 Applications: Simulations and experiments XXXIII

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.

The influence of the composition of the welded material was studied for different binary Fe-C al-

loys and constant MC of 10 and 0. Fig. 0.22 d) shows the alloys tested together with the corre-

-4

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.

A series of welding simulations of Fe-0.1wt%C with different constant Marangoni coefficients The

MC values were 0, 110 , 210 , 510 , 110 , 1.510 , 210 and 410 N m K . Fig.

-5 -5 -5 -4 -4 -4 -4 -1 -1

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

sol wt % C T [K] liq T [K]

sol

0.2 1792.93 1760.67 0.6 1761.99 1687.44

0.3 1784.82 1742.37 0.7 1756.03 1669.14

d) 0.4 1776.71 1724.06 0.8 1750.07 1650.83

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.

A series of welding simulations of Fe-0.1wt%C with MC = 110 N m K and 0 and different

-4 -1 -1

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)

Fig. 0.25: Influence of laser radius on pool shape for welding of Fe-0.1wt%C with P = 2000 W and

positive MC +110 N m K ; scale cf. Fig. 0.21; Tliq = 1801 K, Tsol = 1766 K; the pool shapes

-4 -1 -1

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

Fig. 0.26: Influence of laser radius on pool shape for welding of Fe-0.1wt%C with P = 2000 W and

no Marangoni convection; scale cf. Fig. 0.21; Tliq = 1801 K, Tsol = 1766 K; the pool shapes marked

with (V) are not realistic, because for them the formation of a vapour capillary is expected.

Fig. 0.27: Influence of laser radius on pool shape for welding of Fe-0.1wt%C with P = 2000 W and

negative MC 110 N m K ; scale cf. Fig. 0.21; Tliq = 1801 K, Tsol = 1766 K; the pool shapes

-4 -1 -1

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

XXXVIII Chapter 0 Executive summary

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.

A series of welding simulations of Fe-0.2wt%C, rlaser = 4 mm with MC = 110 N m K and 0

-4 -1 -1

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).

Viscosity differences influence the pool shape via different effects. The most interesting one is the

formation of a second convection cell in the case of negative MC and zero viscosity. Fig. 0.33

shows the different flow fields and pool shapes for the cases of zero viscosity (left) and constant

realistic viscosity = 610 kg m s (right) for MC = 110 N m K and 0. The main effect of

-3 -1 -1 -4 -1 -1

varying viscosity has been observed in the resulting macrosegregation patterns. It will be described

later.

Fig. 0.29: Influence of welding power on the pool shape for welding of Fe-0.2wt%C with MC =

+110 N m K ; temperature distribution after 1.5 s welding time; scale cf. Fig. 0.21; Tliq = 1793

-4 -1 -1

K, Tsol = 1761 K.

0.6 Applications: Simulations and experiments XXXIX

Fig. 0.30: Influence of welding power on the pool shape for welding of Fe-0.2wt%C with MC = 0;

temperature distribution after 1.5 s welding time; scale cf. Fig. 0.21; Tliq = 1793 K, Tsol = 1761 K.

welding power on the

P = 1000 W P = 1500 W P = 2000 W pool shape for welding

of Fe-0.2wt%C with

MC = 110 N m K ;

-4 -1 -1

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

c) vmax = 0.147 m/s vmax = 0.136 m/s

Fig. 0.33 : Influence of viscosity on the shape of the weld pool for welding of Fe-0.2wt%C with P =

2000 W and different MC a) +110 N m K ; b) 0; c) 110 N m K ; left: = 0; right: = 610

-4 -1 -1 -4 -1 -1 -3

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

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

temperature plots in this section. The

grey area with temperatures above

the critical temperature of the MC,

2287 K (H 0 =1.88108 J/mol),

represents the area with negative

MC, the grey area with temperatures

between 1720 K and 1775 K is the

mushy zone for the initial composi-

tion Fe-0.42wt%C.

ties used to calculate the characteris-

tic numbers for W-shape weld pools.

The example would yield the char-

acteristic numbers

D R3c /

s

= 0.25, D R3 =

l /

s

0.47

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.

In order to describe the development of the pool shapes, especially W-shape pools, some new

quantities have been introduced. Corresponding to the usual habit of describing the weld pool shape

by the depth/width ratio, the W-shape weld pool, and as a special case also the V-shape weld pool is

characterised here by a set of four numbers, Fig. 0.35. All of them are geometric quantities normal-

ised to a fixed reference radius R ref, e.g. the maximum pool radius of the calculation. The four

quantities are:

The normalised central depth Dc / R ref (in the middle of the pool).

The normalised lateral depth Dl / R ref (the depth of the second deepening, in general beneath the

critical temperature point on the surface).

The normalised offset Ol / R ref of the critical temperature point.

The normalised radius R / R ref.

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. 0.36 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, Fig. 0.37 shows the devel-

opment of all characteristic numbers for all powers.

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. Fig. 0.38 shows the characteristic

numbers for different welding times as a function of welding power. A closer look at the initial pool

development shows that between the power regions which directly lead to 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 behaviour and where the decision whether a V-shape or a W-shape is

formed takes some time.

0.6 Applications: Simulations and experiments XLIII

b) 1300 W

Fig. 0.36: Laser welding of Fe-0.42wt%C with temperature dependent MC; dependence of weld

pool shape and depth on the welding power. Left: temperatures after 2.5 s; scale cf. Fig. 0.34;

Right: Velocities after 2.5 s; grey: Mushy zone.

XLIV Chapter 0 Executive summary

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

Unpublished work by /Pavlyk 03/, Fig. 0.39, gives a great support to the theoretical predictions of

power dependent pool shapes and especially to the predictions of Fig. 0.38. Even if the results have

been obtained with a GTA welding set-up and show effects of electromagnetic forces, there is

strong evidence, that the following main observations can be made which agree with the predictions

derived from simulations in the foregoing sections:

Three power regions can be distinguished: V-shape region for low powers (1000W, 1600 W),

(modified-)W-shape-region for high powers (2200 W, 2900 W, 3400 W), transition region for

powers in between (1900 W)

In the V-shape region the pool depth increases with increasing power

0.6 Applications: Simulations and experiments XLV

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

welding power on shape of

the weld pool in GTA

welding processes. Un-

published work, /Pavlyk

03/.

In the W-shape region the pool radius increases with increasing power

Of course the absolute powers and details of the pool shapes differ due to the GTA welding condi-

tions and different welding materials used, and of course this experiment would not be sufficient to

create a theory, but it definitely supports the new theory developed in this work.

XLVI Chapter 0 Executive summary

Most welding applications include moving heat sources. For these applications numerical steady-

state solutions yield the most interesting information. For welding problems with a stationary heat

source the numerical steady-state solutions only give a part of the information which might be use-

ful. Some recent works investigate the transient behaviour of weld pools, but do not take into ac-

count the transient processes after switching off the heat source.

Careful consideration of the series of calculations made for this study gives strong evidence that a

number of transient effects play a major role during the development of the molten area. Another

important effect occurs when the HS has been switched off and the calculation is continued until

complete resolidification. Taking the case of 2000 W welding power and welding duration 3 s as an

example, mainly three effects can be observed:

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. 0.40 a) shows the time evolution of the normalised central depth Dc / R3s for the different

welding durations. All calculations which had enough time to form a W-shape pool showed the

same effect.

The effect of increasing central depth is easy to understand. Fig. 0.40 b) 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, which is now positive on

the whole surface, creates a Marangoni flow towards the middle of the pool, trying to transform the

pool's W-shape into a V-shape. This process is stopped by solidification.

The black area in Fig. 0.41 a) shows the additionally molten parts of the pool for the case of 2 s

welding duration, while the grey area represent the part resolidified in the same time. Fig. 0.41 b)

shows the error which would be made if measured pool depths from micrographs were compared

with results of steady state simulations giving the status at the moment when the HS is switched off.

The characteristic W-shape does not develop from the beginning, but in a later stage of the melting

process. Fig. 0.42 shows the time evolution of pool shape and characteristic numbers. It can be

0.6 Applications: Simulations and experiments XLVII

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.

There is a critical power range between the two extreme behaviours of full W-shape formation for

high powers and full V-shape formation (never reaching CT) for low powers, where the pool starts

XLVIII Chapter 0 Executive summary

a)

Fig. 0.42: Laser welding of Fe-0.42wt%C with

temperature dependent MC and P = 2000 W; a)

early stage in the time evolution of W-shape

pool; growth of flat pool between 0 and 0.6 s,

formation of lateral deepening when CTP

reaches equilibrium after 0.6 s; Left: Tempera-

tures, scale cf. Fig. 0.34; Right: Streamlines; the

scales of different time steps are not comparable;

b) time development of all characteristic num-

b) bers; R3s = 0,0044 m

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.

The third study investigates GTA welding processes. The range of transient effects gets still wider

when electromagnetic (EM) forces are considered. In the case where no Marangoni forces are con-

sidered, it is well known, that Lorentz forces create an inwards directed flow, which results in some

kind of V- or U-shape pool. 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. Five calculations were made with P = 1400 W, 1500 W, 1550 W, 1600 W and

2000 W. Fig. 0.44 shows the time evolution for the two calculations with P = 1550 W and

0.6 Applications: Simulations and experiments XLIX

1395 W 1500 W

Fig. 0.43: Laser welding of Fe-0.42wt%C with temperature dependent MC; time evolution of weld

pool shape, temperature and flow field for initial phase of calculations in the critical power region:

a) P = 1395 W; b) P = 1500 W; both calculations start with comparable W-type pool shapes, but

then develop into two different directions, V-shape type for 1395 W and flat initial phase of W-

shape formation for 1500 W; Left: Temperatures, scale cf. Fig. 0.34; Right: Streamlines; the scales

of different time steps are not comparable.

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.

The fourth and last welding study deals with the subject of solute redistribution in weld pools. The

composition of the welded area is crucial for the mechanical properties of the joint. While species

losses due to vaporization have been described by several researchers, the formation of macrosegre-

gations has not been investigated5, even if /Zacharia 95/ regard convection effects on solute redis-

tribution as 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.

The mushy zone flow model which has been applied to the casting processes has been used to pre-

dict concentration redistribution on the scale of the weld pool. Fig. 0.46 compares the final macro-

segregation patterns for zero viscosity and realistic viscosity values for different MC. The strong

macrosegregations in the non viscosity cases is a result of fast mushy zone flows which wash out

the interdendritic melt. In the case of realistic viscosities, the mushy zone flow is stopped immedi-

ately after the HS has been switched off. This happens due to the drag forces inside the mushy zone

(which do not exist in the zero viscosity case) when the driving Marangoni forces diminish. In this

case the result is that the solute distribution of the moment when the HS is switched off, just freezes

and not changes any more.

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

a) b)

Fig. 0.45: 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. 0.34; Right:

Streamlines; the scales of different time steps are not comparable.

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

phology of mushy zone of spot

welded Al-7wt%Si-0.3wt%Mg

alloy; welding duration: 8 s.

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.

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 number of new concepts have been developed and integrated into the model:

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 globulitic 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.

LIV Chapter 0 Executive summary

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).

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.

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.

In order to simulate the complex convection effects which are the main subject of this work, a num-

ber of new basic models describing single aspects had to be developed. The main developments are:

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.

4 Chapter 1 Introduction

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.

The new basic models have been integrated together with state of the art models based on the vol-

ume averaging model by /Ni 91/, describing fluid flow and convective and diffusive transport phe-

nomena. Two 2D software packages have been developed which contain different combinations of

coupled models:

EL2D (current version el2d9.1). The software is originally based on work by /Schneider 95a/

and /Bhmer 97/ and was used for the calculation of macrosegregation phenomena in casting. It

was extended by the author by early versions of the new solidification algorithm and the solidi-

fication of moving free surfaces as well as by sophisticated models for laser and GTA welding.

It is mainly used for welding simulations in cylindrical co-ordinates.

SoliCon (current version solicon2.0), which is based on completely new concepts and sets of

equations. It contains all new models, i.e. the algorithm for solidification of moving free sur-

faces, split solid model, porosity formation model, and new solidification algorithm. It is mainly

used for casting simulations in cartesian co-ordinates.

A more detailed description of the two software packages can be found in App. 9.2.

The numerical models allow the coupled simulation of thermosolutal convection, buoyancy driven

free surface movement, shrinkage flow, Marangoni flow, and flow due to electromagnetic forces

and have 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 macrosegregations, the flow field and

the shape of the external shrinkage cavity

Solute redistribution and pool shape formation in laser and GTA welding processes.

The integrated models have been applied to the following scientific problems

Deep shrinkage cavities in ingot casting have a large impact on the final macrosegregation pat-

tern. The correct prediction of this influence can help to improve the criteria for correct riser

dimensioning. The model was applied to the casting of an unalloyed steel ingot with riser

(SoliCon), Fig. 1.1 a), and the results compared to experimental results from /Bhmer 97/.

1.1 Subject of this work 5

ingot (EL2D); the model allows to consider the impact of the alloy composition on the surface

shape.

Simultaneous but separate prediction of hydrogen and shrinkage porosities during casting of Al-

7wt%Si (SoliCon); experimental validation for pure shrinkage porosity formation; The new

model can be a tool to determine the type of experimentally observed porosities.

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. 1.1 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/.

The following chapter will give detailed descriptions and explanations of the most important con-

vection effects, an overview of the state of the art in experimental investigation and mathematical

and numerical modelling of each area and a short description of the particular contribution by this

work.

2 Modelling of convection effects in casting and welding processes

Physical phenomena / State of the art / Contribution by this work

2.1 Macrosegregations

Macrosegregations are concentration inhomogeneities on the macroscopic scale of a whole cast part

or weld. As many important properties of castings depend on the exact alloy composition, unpre-

dictable composition fluctuations can reduce the productivity of a casting process or even be dan-

gerous if the damage is not noticed.

Steel manufacturers who produce large steel ingots weighing many tons have problems to fulfill

the industrial standards of alloy composition

Cast parts may contain weak areas (cast iron instead of steel due to extreme carbon enrichment)

and fail under load or during the removal of risers or gating systems

In welding concentration redistribution can have a bad influence on the material properties of

the weld seam

This is why macrosegregation is a phenomenon that has great practical importance for the casting

industry. The aim of industrial casting must be to produce castings with completely homogeneous

composition.

While it is easy to produce a homogeneous liquid mixture of the desired composition, the alloy con-

stituents tend to segregate, i.e. to separate, during solidification. Main reason for this behaviour is

the different solubility of the alloy elements in liquid and solid phases. Most alloy elements have a

lower solubility in the solid than in the liquid phase. During solidification the solid that forms is

depleted and rejects the excess solute into the interdendritic liquid. This effect alone would not lead

to macro- but to microsegregations. After complete solidification there would be concentration

variations on the length scale of the dendrite arm spacing. The phenomenon of macrosegregation is

mainly created by the relative movement of liquid and solid phase in the solidifying areas. Macro-

segregations occur if the interdendritic liquid is removed by fluid flowing through the dendritic

network and replaced by liquid with a different composition.

The most important phenomena that lead to macrosegregation effects are:

Shrinkage flow in general leads to positive (inverse) segregations

Thermosolutal convection

Forced convection e.g. by pouring, electromagnetic stirring, vibration

Movement of floating globulitic grains

2.1 Macrosegregations 7

In the following the most important mechanisms of macrosegregation formation will be described.

Thermosolutal or forced fluid flow through the mushy zone can lead to the formation of negative

macrosegregations in the zone and positive ones in the liquid reservoir outside. Fig. 2.1 shows the

mechanism for the case of partitioning coefficient ls < 1. Microsegregation during solidification

leads to the formation of a depleted dendritic network with enriched interdendritic liquid, Fig. 2.1

a). If no convection or long range liquid diffusion occurs, the average concentration on a

mesoscopic scale, i.e. in a volume which contains at least some dendrites, will remain constant (C0).

Macrosegregations form if material with concentration that is different from the concentration of

the interdendritic liquid is transported into the mushy zone. Fig. 2.1 b) shows a case where liquid

from a reservoir with initial concentration C0 flows into the mushy zone and replaces the enriched

interdendritic liquid. This makes the average concentration of the mushy zone decrease. The liquid

that has been washed out into the reservoir forms an enriched area. Very often, especially at the

beginning of the solidification process, the amount of solid that forms is small compared to the total

amount of liquid. So the washed out interdendritic liquid is strongly diluted by convection and does

not rise much the concentration level of the liquid reservoir. This leads to the phenomenon that

solid which forms at the beginning is strongly depleted, while the concentration of the liquid re-

mains almost constant. With proceeding solidification the amount of solute driven out of the mushy

zone accumulates in the decreasing amount of liquid phase. This leads to an accelerated enrichment

of the liquid. On the other hand the rising concentration level of the liquid reservoir reduces the

thermosolutal or forced convection through the

mushy zone. a) A dendritic mushy zone forms in-

side a liquid reservoir with initial concentration C0 .

The forming solid is depleted (C) and rejects the

excess solute into the interdendritic liquid (C+).

Still the average concentration is C0 . b) fluid flow

crosses the mushy zone and replaces the interden-

dritic liquid by liquid of initial concentration C0 .

The mushy zone becomes depleted and the area

which receives the washed-out interdendritic liquid

a) b) becomes enriched.

8 Chapter 2 Modelling of convection effects in casting and welding processes

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.

In contrast to normal segregation inverse segregation leads to the formation of enriched areas in

parts of the casting which solidify in early stages of the solidification process. This macrosegrega-

tion type is caused by shrinkage flow due to temperature dependent liquid and solid density and

especially to the density difference between liquid and solid phase. Fig. 2.2 shows the principle

mechanism. Fig. 2.2 a) shows a liquid boundary layer which is about to solidify. During solidifica-

tion, Fig. 2.2 b), the volume of the liquid layer decreases6. For continuity reasons this volume loss

has to be compensated by advection of liquid phase. This shrinkage flow is driven by a pressure

drop in the solidifying area.

Fig. 2.3 shows how the shrinkage flow can lead to the formation of positive segregations in the so-

lidifying area, in this case at the base of an array of columnar dendrites. During columnar dendritic

solidification the continued rejection of solute by the depleted solid leads to the formation of a con-

centration gradient with low concentration close to the dendrite tips and higher concentration close

to the dendrite base, Fig. 2.3 a). If there is no flow (no macroscopic diffusion in the liquid as-

sumed), the solute rejection from the solid only leads to microsegregation and the average concen-

tration of solid and liquid together in the area beneath the dashed line remains constant, C0 . If the

continued solidification, Fig. 2.3 b), initiates a shrinkage flow across the dashed line, new liquid

enters. The composition of this liquid determines which kind of macrosegregation occurs. In the

present case the liquid comes from inside the mushy zone / dendritic network and is enriched itself

(C+). This leads to an increase of the average concentration and formation of positive / inverse

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

sity difference between liquid and solid phase. a)

The hatched area signifies a liquid layer that is

about to solidify. b) The hatched area has solidified

and shrunken. The volume difference (dashed line)

must be replaced by new liquid, so shrinkage flow

a) b) starts.

shrinkage flow. a) Columnar dendritic solidifica-

tion leads to the formation of a concentration gra-

dient along the dendrite axes. Without long range

diffusion and convection the average concentration

of solid and liquid together above and beneath the

dashed line remain constantly C0 . b) If a new layer

of solid forms, the volume loss causes a shrinkage

flow towards the dendrite bases. This flow trans-

ports enriched solute across the dashed line and

a) b) thus increases the average concentration of the

bottom area.

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.

Globulitic grains are small crystals which nucleate in the liquid without contact with any wall. The

solute which is rejected by the solid forms a thin boundary layer of enriched melt around the grain,

Fig. 2.4 a). According to the density difference between liquid and solid phase the grain starts to

move down- or upwards by gravity. The grain leaves the enriched boundary layer behind and enters

an area with non-enriched melt, Fig. 2.4 b). This leads to an increase of the average concentration,

i.e. positive macrosegregation in the area the grain has left and negative segregation in the area the

grain has entered. A well known macroscopic effect of this mechanism is the existence of a very

fine grained area with negative segregations at the bottom of cast ingots. It is caused because the

globulitic grains which form in the liquid settle like a rain shower.

10 Chapter 2 Modelling of convection effects in casting and welding processes

tation of globulitic grains; a) The solute that is rejected

by the solidifying grain forms an enriched solute

boundary layer. The average concentration in the top

and bottom areas is both C0 . b) The solid moves

downward due to gravity and leaves the enriched melt

behind. Its place in the top area is filled with melt of

concentration C0 . Now the average concentration of the

top area is higher than before. In the lower area melt of

concentration C0 has been replaced by depleted solid.

a) b) The average concentration of the bottom area has thus

lowered.

Gravity segregation is caused by solutal convection. During the solidification process microsegre-

gation occurs and creates local density inhomogeneities, Fig. 2.5 a). If the rejected solute is much

lighter than the matrix material, which is the case in alloy systems like Fe-C and Al-Mg, the en-

riched melt causes a solutal flow towards the surface. This results in positive segregations close to

the surface, Fig. 2.5 b). Of course the same happens if the solute is heavier than the matrix. In this

case the solutal convection leads towards the bottom of the cast part. If the density difference be-

tween matrix and solute is not very big, solutal convection is superposed by thermal convection.

The effect is not always separable from other convection effects.

mation of macrosegregations due to solutal flow. a)

Columnar solidification leads to the formation of

an enriched liquid layer close to the solid (sche-

matic picture for the case of planar solidification

front, for dendritic solidification the enriched liq-

uid would be inside the mushy zone). b) In this

case the solute is much lighter than the matrix and

initiates a solutal convection flow towards the sur-

face, where positive segregations occur.

a) b)

2.1 Macrosegregations 11

Freckles or channel segregates are chains of small globulitic grains in areas with otherwise colum-

nar structure. They emerge from a complex sequence of events. Typically freckles occur in upward

directional solidification, e.g. Bridgeman casting of turbine blades.

They are initiated by flow instabilities. In upward directional solidification the casting is cooled

from the bottom. This creates stable thermal layers with heavy, cold melt at the bottom and light hot

melt on top. When solidification starts, light solute is rejected into the melt in a non-uniform distri-

bution. These small density inhomogeneities can lead to a local solutal upward flow that is stabi-

lised by the feeding flow of enriched melt from neighbour regions, Fig. 2.6 a). The incremental en-

richment locally reduces the liquidus temperature so that isothermal remelting can occur, 2.6 b).

The continuous supply of enriched melt leads to the formation of a self-sustaining segregation

channel. When the feeding flow stops because the neighbour areas are almost solidified, the channel

itself starts to solidify. After complete solidification the channel appears as a strongly enriched area

filled with a chain of globulitic grains. Different theories exist for the origin of these grains. It

seems to be common sense that the grains are secondary or tertiary dendrite arms which have sepa-

rated from their trunks in the channel during solidification. The separation is thought to be caused

by local remelting and coarsening.

Due to convective instabilities a local

enrichment of the liquid occurs in the

mushy zone. b) The enriched liquid

that is moving upwards due to solutal

convection remelts a part of the

mushy zone. Due to continuous sup-

ply with enriched liquid from neigh-

bour areas a stable flow channel

forms that is kept open by the en-

riched melt. c) At the end the channel

solidifies filled with globulitic grains.

In large steel ingots two types of characteristic macrosegregation patterns form which are known as

V-segregates and A-segregates (other names for the A type are -segregates, ghosts or the German

term Schattenstreifen). In spite of many years of investigations and discussion there is still no com-

plete agreement about the mechanisms of their formation. They seem to be linked to the attachment

12 Chapter 2 Modelling of convection effects in casting and welding processes

volved in the formation of

macrosegregation defects

like A-segregations: At-

tachment of a) dendritic

floating crystals, b) globular

floating crystals to the front

of columnar crystals.

a) b)

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/.

The macrosegregation effects described in general appear in combination. The final macrosegrega-

tion pattern of a big ingot is determined by the complex interaction and superposition of all these

flow effects. Fig. 2.8 shows a typical example of a 10 ton steel ingot with negative macrosegrega-

tions due to thermosolutal convection and sedimentation of globulitic grains, positive inverse segre-

gations, as well as A- and V-segregates.

segregation patterns of a

10 ton steel ingot; a)

macroetched section

from /Hultgren 73/; b)

sketch from /Flemings

76/

2.1 Macrosegregations 13

Due to the low diffusivity of the alloy components the long range macrosegregation defects cannot

be removed once solidification is complete. Heat treatment can only smooth microsegregations.

Therefore the aim must be to inhibit the formation of macrosegregation effects during the solidifi-

cation process. So far many attempts have been made to prevent or at least weaken the defects.

Most of them try to prevent the relative movement between solid and segregated liquid. Possible

concepts are, /Beckermann 00/:

change of cooling conditions or alloy composition to reduce the driving forces of thermosolutal

flow

change of riser and mould design to influence the thermal convection patterns

influencing the flow by application of nozzles, baffles, porous materials, centrifugal forces or

electromagnetic fields

change of effective viscosity by addition of inert particles

controlled deformation of the solid to squeeze enriched liquid away

modifications to the grain structure (grain refiners) to influence the flow resistance of the mushy

zone

Still macrosegregation defects are a big problem for all producers of cast metals and large castings.

In order to find effective solutions the correct prediction of macrosegregation effects by simulation

is an important task.

In consequence of the intensive research on the domain of macrosegregation, there are detailed re-

views on the state of the art available. Recent ones are: /Beckermann 95/, /Prescott 96/, /Combeau

98/, /Beckermann 00/, /Beckermann 01/, and /Ghosh 01/. The overview given here presents publi-

cations related to ingot casting only. Continuous casting processes will not be considered.

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.

Pioneering work for understanding and predicting macrosegregations has been done by Flemings

and co-workers. /Flemings 67/ derive analytical expressions for inverse segregation and centre line

segregation. Al-4.5wt%Cu alloy is used to give numerical examples. /Flemings 68a/ apply these

analytical expressions to predict technical macrosegregation effects like solute-rich and solute-poor

bands, composition changes at abrupt changes of ingot cross section, and under-riser segregations.

/Flemings 68b/ at last reports experiments to test the analyses quantitatively and qualitatively for

laboratory-size ingots of Al-4.5wt%Cu. All experiments show quantitatively good agreement.

Recent analytical publications include /Sundarraj 95/, who examine the impact of the microscale

phenomena and the choice of the density model on the resulting inverse segregation in uni-

directional casting of Al-Cu alloys, /Voller 98/, who incorporates a new semi-analytical microseg-

regation model which includes coarsening effects of the dendritic structure, into a macrosegregation

model for the prediction of inverse segregation, and /Haug 98/, who derive analytical expressions

for inverse segregation which include finite diffusion in the solid phase. /Xu 01/ develops a com-

plex macrosegregation model for dendritic solidification which takes different dendrite growth

morphologies into account.

It is difficult to derive analytical models for describing transient effects like the formation of normal

segregations due to thermosolutal convection. Inverse segregation is the result of a directed flow

covering a short distance which can be calculated. Normal segregation can be the result of a long

process of fluid flow crossing the mushy zone transverse to the dendrite growth direction. In large

steel castings this can take several hours, during which the liquid covers a way of many meters. It is

difficult to derive valid analytical concentration balances from this type of process. For the same

reason the application of criterion functions in numerical simulation is not very common.

The most promising models for predicting all coupled macrosegregation effects are numerical mod-

els which describe fluid flow, solidification, heat and mass transport and interactions between the

different phases.

A famous early model being able to describe momentum, heat and species transport in binary solid-

liquid phase change systems is presented in /Bennon 87a/. In /Bennon 87b/ and /Bennon 87c/ the

model is applied to the solidification process of an aqueous solution of ammonium chloride

(NH4Cl-H2O) in a rectangular cavity cooled from the bottom. The authors predict mushy region

2.1 Macrosegregations 15

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.

As macrosegregations generally occur due to flow phenomena inside the liquid/solid area of a cast-

ing, the precise knowledge of the mushy zone properties are of crucial importance for the quantita-

tively correct prediction of macrosegregation. Most publications in this area try to predict the flow

resistance of the mushy zone function of solid fraction, alloy composition and morphology. In po-

rous media with high solid fractions usually Darcy-terms are used to calculate the flow resistance as

a function of solid fraction and velocity:

v

M Darcy = l [K ]1 vv

where K is the permeability tensor. In a columnar dendritic mushy zone the permeability tensor

takes the fact into account, that the permeability for flow parallel to the dendrite axes is different

from the permeability for flow perpendicular to the axes. In areas where the solid fraction is small

(s < 0.3) the Darcy-law fails. This situation defines two main aims of research: One the one hand to

obtain permeability data to be used with Darcy-law models, on the other hand, to develop models

which can deal with situations where Darcy-law formulations fail.

/Poirier 87a/ summarizes experimental permeability data for the flow of interdendritic liquid in par-

tially solid Pb-Sn and borneol-paraffin columnar-alloys. He uses regression analysis and simple

flow models to arrive at relationships between permeability and the morphology of the solid den-

drites. A numerical way to determine permeabilities is to model the dendritic microstructure as ob-

stacles in a flow field and to use a CVM or FEM Navier-Stokes-Solver to calculate the pressure

drop caused by this liquid / solid configuration. This is how /Bhat 96/ determine permeabilities

down to solid fractions of s = 0.02. /Goyeau 98/ and /Goyeau 99/ use a digitalised image of a real

dendritic structure to perform a numerical determination of the parallel and transverse permeability

by solving the closure problem associated with the derivation of the macroscopic momentum equa-

tion using a volume averaging technique. /Wang 95b/ propose a physical model for the solid / liquid

2.1 Macrosegregations 17

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.

Freckle formation, which is not explicitly subject of this work, has been experimentally investigated

and explained in a pioneer work by /Giamei 70/ and later by /Pollock 96/. Many important publica-

tions on the numerical simulation of freckles have treated the formation of the segregation channels,

plumes and volcanoes in multicomponent alloys in 2D, /Neilson 91/, /Schneider 97/, and 3D,

/Felicelli 98/, or the formation and motion of equiaxed grains, /Gu 97/. Other works try to predict

the occurrence of freckles using criterion functions based on the evaluation of the Rayleigh number,

which is the ratio between the driving force for the convective flow and the drag force to resist the

flow, /Beckermann 00a/, /Yang 01/.

The macrosegregation modelling in the present work is based on the volume averaged mushy zone

model by /Ni 91/. New contributions have been added in the following areas:

The macrosegregation model has been implemented into the framework which allows to simu-

late the formation of deep external shrinkage cavities in steel casting and to model their influ-

ence on the final macrosegregation pattern, section 6.2.

18 Chapter 2 Modelling of convection effects in casting and welding processes

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.

Shrinkage defects are all negative consequences of temperature or phase dependent densities that

occur during solidification and cooling processes.

For casting and welding industry they are one of the most important issues.

Porosities downgrade important material properties like toughness and yield strenght and can

lead to failure or reduce the life cycle of cast components.

Porosities can ruin pressure-tight components.

Porosities which are hidden in mechanically important areas of the cast part can lead to failure

under load.

The correct prediction of external shrinkage cavities is important for economic riser dimen-

sioning.

External shrinkage cavities of ingots or risers can influence the macrosegregation distribution

inside the cast part and lead to unwanted local material properties

Movements of the solidifying free surface have an influence on the quality of the weld seam in

spot and line welding (problem: Ripples), and on the surface quality in continuous strip casting.

Hot tears which occur in semi-manufactured products like ingots increase the amount of waste

material and reduce the productivity.

The principal reason for the formation of shrinkage defects are the inhomogeneity of cooling and

solidification processes. If cooling of a casting was completely homogeneous, if solidification hap-

pened at a single moment, and if there was no obstruction by the mould, the only effect would be

cubic shrinkage, i.e. the cast part would be an exact downsized copy of the mould cavity. Shrinkage

defects only form if local density differences occur, which is the case in virtually every casting and

cannot be completely inhibited by any means.

Local density differences occur mainly for three reasons:

Temperature inhomogeneities which are needed to drive the cooling and solidification process

Concentration inhomogeneities which occur due to microsegregation phenomena during solidi-

fication

2.2 Formation of shrinkage cavities and porosities 19

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.

External shrinkage cavities are surface deformations which are generated by a complex interaction

of the formation of a solid shell, shrinkage flow and movement of the remaining liquid surface.

They are not always considered as defects, because they mostly occur on top of risers or ingates,

which are removed when solidification is finished. If temperature and solidification control is not

good they can also occur in the cast part, e.g. during horizontal casting of slabs.

External shrinkage cavities form in areas with a free liquid surface, if the cooling by the mould is

stronger than the cooling of the liquid surface by radiation and air convection, e.g. in permanent

mould casting. In this case the strong cooling leads to the formation of a solid shell in contact with

the mould, while the liquid surface remains open, 2.9 a). As the solid shell has a higher density than

the remaining liquid, the solidification initiates a shrinkage flow towards the solid. Gravity tries to

keep the surface of the remaining liquid even so the liquid level is lowered. Due to this mechanism

20 Chapter 2 Modelling of convection effects in casting and welding processes

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.

Internal shrinkage cavities and shrinkage porosities are defects which are caused by interruption of

the feeding flow that is responsible for replacing the volume losses due to shrinkage. If solidifica-

tion leads to a situation where a liquid reservoir is completely surrounded by solid and thus cut off

from all feeding possibilities, Fig. 2.9 b), internal shrinkage cavities or shrinkage porosities form.

The pressure in the closed off area which so far has been the hydrostatic pressure plus the air pres-

sure at the surface, descends if shrinkage due to solidification and cooling continues. In general the

closing of the area does not happen suddenly, but gradually. To the same degree as the entry to the

area is closed, the pressure drop needed to maintain the feeding flow rises. So the pressure in the

area that tries to attract more liquid drops until it reaches the critical pressure for pore nucleation.

Now the question if internal shrinkage cavities or shrinkage porosities form, is answered by the so-

lidification morphology of the alloy. If the area is still completely liquid, as in the case of alloys

with planar solidification front like Al-12.6wt%Si or Al99.99, a coherent internal shrinkage cavity

forms on top of the liquid area. If the area is already part of the mushy zone, and covered by a den-

dritic network, as in the case of alloys with large solidification interval like Al-7wt%Si, shrinkage

porosities form in the interdendritic liquid which are confined by the shape of the dendrite skele-

tons.

Hydrogen porosities are microporosities that are caused by the changing solubility of hydrogen

during the solidification of aluminium alloys. In the liquid phase of aluminium about twenty times

more hydrogen can be dissolved than in the solid phase. So if the hydrogen contents of the liquid is

higher than the maximum possible contents of the solid, the excess hydrogen is rejected during so-

lidification and hydrogen bubbles form at the solidification front. Some of them rise to the surface,

and some of them are trapped in the solidifying front and form hydrogen porosities. The probability

for hydrogen bubbles to be trapped is of course very high if they form inside a dendritic mushy

zone. The easiest way to prevent the formation of hydrogen porosities is degassing the melt before

casting, i.e. lowering the hydrogen contents to be smaller than the maximum contents allowed by

the solid. For industrial casting the limit of hydrogen allowed in the melt is about 0.1 cm3 H2 / 100 g

Al. This porosity type is not caused by a local pressure drop, but it benefits from it. If the local pres-

2.2 Formation of shrinkage cavities and porosities 21

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).

acteristics on formation of shrink-

age cavities in cast ingots: a)

Strong heat extraction in steel

mould leads to the formation of a

solid shell while the surface re-

mains liquid. The lowering liquid

level during solidification creates

the characteristic surface shape. b)

In a sand mould the heat extraction

is equal to all sides. A closed solid

shell forms. Volume losses due to

shrinkage lead to the formation of

a) b) an internal shrinkage cavity or

shrinkage porosities.

Linear contraction is the linear change of the dimensions of the whole cast part. It is responsible for

the fact that at the end of solidification and cooling, the cast part is smaller than the mould cavity.

The reason is obvious: When solidification is finished, the cast part is still hot and shrinks with

continued cooling. If this happens in a homogeneous way, the problem can just be solved by over-

sizing the mould. In general, of course neither solidification nor cooling is homogeneous and so in

some areas the cast part is solid and shrinks, while in others it is still liquid and fills all available

space. These processes lead to many important defects like hot tearing, residual stresses and prob-

lems with dimensional accuracy. Effects that are not so obvious occur inside the cast part. The con-

traction of dendritic networks can may be stronger than the shrinkage inside the network due to

phase change and thus invert and eliminate the other shrinkage effects. There is strong evidence that

especially inverse segregation can be suppressed by linear contraction. Instead of attracting melt the

shrinking network pushes it away. A well known effect is the formation of surface segregations in

continuous casting. Here due to the mechanical compression of the solid shell which is still porous,

the enriched interdendritic liquid is squeezed out and forms a layer of highly enriched solid on the

outer surface.

22 Chapter 2 Modelling of convection effects in casting and welding processes

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.

Like macrosegregation defects also shrinkage defects cannot be removed once solidification is ter-

minated. Therefore all methods to prevent the formation must be applied to the melt or to the solidi-

fication process. Typical measures are

Usage of filters to slow down the filling flow and to prevent gas entrapment

Usage of risers. Risers are reservoirs of molten material that are positioned in order to replace

local volume losses due to shrinkage.

Correct choice of the mould materials. Solidification can be directed by use of intelligent com-

binations of sand, steel parts and insulators. Insulators can be used to keep risers liquid so that

their size can be reduced

Correct choice of the mould geometry, gating system, ingate and risers to control the solidifica-

tion sequence. Risers should contain liquid material until the end of solidification, so they work

as hot spots and heat their surrounding area

Influencing of directional solidification by application of heating or cooling devices in the

mould

Degassing of the melt. Aluminium melt has to be treated to reduce the hydrogen content

Oversizing of the mould to take the linear contraction into account if the casting has to fulfill

size requirements

Pressurised feeding. Pressure is applied at the ingate to reduce the formation of gas bubbles in

the melt during solidification. Special case: Squeeze casting, where pressures of thousands of

bar are applied to reduce shrinkage porosities.

Use of special cast materials like spheroidal cast iron; these alloys do not cause much trouble,

because there is not much difference in the densities of liquid and solid phase.

As experimental investigation of shrinkage effects is costly, and their economic impact is consider-

able, the prediction of porosities and shrinkage cavities is one of the most accepted applications of

numerical simulation. So far most commercial software packages use criterion functions based on

cooling characteristics to determine the most probable areas of porosity formation. Commercial

simulations based on flow phenomena exist, but they still need to be developed to become reliable

tools for the prediction of shrinkage defects.

2.2 Formation of shrinkage cavities and porosities 23

The great technical importance, the large number of mechanisms and the even larger number of

explanations for the formation of porosities, internal and external shrinkage cavities have initiated

an enormous number of experimental studies and mathematical and numerical models.

Systematic experimental investigations on the alloy dependent formation of external shrinkage

cavities can be found in an extensive study on the correlation between shrinkage defects and solidi-

fication morphologies in Al-Si alloys, /Patterson 62/.

Not many publications deal with the prediction of external macroscropic shrinkage cavities so far.

Therefore a classification of models is difficult. This section will review a number of approaches

dealing with the flow phenomena on top of the casting:

Treatment of shrinkage losses by inflow boundaries and variable cell size.

Modelling of external shrinkage cavities using feeding algorithms.

Modelling of external shrinkage cavities using Navier-Stokes approaches.

Models generally dealing with the solidification of moving free surfaces.

Many publications describe the influence of shrinkage induced flow on macrosegregation. All of

them have to deal with the total volume loss occuring on top of the casting. One of the common

methods is to use inflow boundaries. The top of the casting, which very often is designed as a thin

riser, is treated as a liquid reservoir, which feeds as much liquid volume as the casting loses due to

shrinkage, e.g. /Krane 95/, /Chang 96/. The result is a casting with constant volume but increasing

mass. A method with mass conservation is the use of simple front tracking methods with flat sur-

face and variable cell size on top of the casting, e.g. /Neilson 91/, /Schneider 95a/. The vertical size

of the topmost cell layer is adapted by subtracting the summed volume losses of the whole casting.

As the surface is assumed to be flat and all top cells are equally modified, this model is not able to

describe solidification effects like the formation of external shrinkage cavities.

/Suri 93/ use a feeding algorithm to predict the formation of macroscopic external shrinkage cavities

(shrinkage pipes) of a simple Al-7wt%Si casting and a ductile iron automotive piston. They cal-

culate the net sum of volume losses in separated pools and subtract a cell layer from the corre-

sponding pool surface if the total volume loss is big enough. The pool surface is defined as the sur-

face area where the solid fraction is smaller than 0.74 assuming the solid to consist of spherical

grains. /Trovant 94/ and /Trovant 95/ calculate the formation of external shrinkage cavities

(shrinkage profiles) for cylindrical castings of pure aluminium and aluminium alloys under dif-

ferent conditions of metal viscosity and thermal diffusivity, and with addition of chill walls. They

couple heat transfer and fluid flow equations with a feeding algorithm which sums up the effective

shrinkage of all control volumes and subtracts the according amount of material from control vol-

24 Chapter 2 Modelling of convection effects in casting and welding processes

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

Fig. 2.10: Formation of deep external shrinkage cavity in a cylindrical iron casting, cast in a grey

iron mould; calculated with rapid solidification shrinkage model, (feeding algorithm, based on heat

transfer only). a) d) development of solidification front position and free surface shape; e) com-

parison with experiment, /Beech 98/.

of flat external shrink-

age cavity during so-

lidification of large

steel ingot, simulated

by coupling Navier-

Stokes solver with

feeding algorithm;

height of mould cav-

ity: 2.55 m; a) d)

simulated velocity

vectors (max. veloc-

ity: 0.025 m/s) and

a) t = 500 s b) t = 3000 s c) t = 9000 s d) t = 15000 s

solid fraction contours

(in 20 % increments),

/Gu 99/.

26 Chapter 2 Modelling of convection effects in casting and welding processes

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.

/Lee 01a/ review the models describing porosity development in aluminium alloys. They categorise

the published models according to their approach:

Analytic solutions

Criterion functions

Computational models using Darcys law

Gas diffusion models

Continuum-stochastic models

They find that none of the models reviewed provides accurate predictions over the complete range

of conditions found in industrial shape castings, where both shrinkage and gas evolution are major

driving forces.

Analytical models try to predict the final porosity distribution in a casting from global or local proc-

ess parameters like geometry, pouring temperature, feeding range etc. Transient effects which may

occur during the process can only be taken into account in form of averaging assumptions. An ex-

ample is /Piwonka 66/, who predict pressure drop and pore formation in cylindrical castings of pure

aluminium and aluminium alloys. They take into account fluid flow through a partially solid mushy

zone, dissolved gas and surface tension.

Criterion functions are semi-empirical formulae which use local macroscopic information about the

cooling and solidification process like the solidification front velocity (R) or cooling rate ( T& ) and

the local temperature gradient (G) and combine them to form general relations which allow to iden-

tify areas with high potential for porosity formation. In general they can be applied as postprocess-

ing after the simulation has been finished, using the temperature data of the simulation. The most

famous of these criterion functions is the Niyama criterion /Niyama 82/ for the prediction of centre

line porosities in steel casting. Good overviews on criterion functions are given by /Hansen 93/ and

/Taylor 98/. /Ko 95/ apply criterion functions to predict porosities in a complex shaped casting of

equiaxed superalloy 718. /MAGMAsoft 02/ uses different criterion functions to model the forma-

tion of centre line shrinkage defects and microporosities. /Windelberg 00/ combine different criteria

for relative, absolute and maximum shrinkage to form a Vector of quality.

Darcy-law models assume that porosity is governed by shrinkage-driven growth with the influence

of dissolved gases added into the final pressure balance. These models are able to predict the final

28 Chapter 2 Modelling of convection effects in casting and welding processes

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.

The numerical fixed grid framework presented in this work does not consider thermomechanical

effects. For this reason the modelling of linear contraction could not be modelled. Examples for

publications dealing with this problem using FEM models are /Guan 91/, /Guan 92/, /Guan 94/ and

/Fackeldey 98/, who couple solidification modelling with thermomechanical models and describe

solid shrinkage and elastic and plastic deformation. The model is used to calculate residual stresses

and to describe the influence of air gap formation between cast part and mould on the solidification

process. /Celentano 98/ describes the formation of internal and external (surface) shrinkage for

shaped castings of cast iron.

A number of new concepts have been developed which allow the prediction of most of the impor-

tant shrinkage defects linked to convection phenomena:

A new mathematical model and algorithm to simulate the solidification of moving free surfaces

has been developed. It reformulates the well-known volume of fluid (VOF) method to be used

with solidifying surface cells and temperature dependent liquid densities, sections 3.3.5 and 4.6.

The model has been applied to simulate the formation of deep shrinkage cavities in steel and Al-

Si castings, sections 6.2 and 6.3.

A new concept for the simultaneous but separate calculation of hydrogen and shrinkage porosi-

ties as well as internal shrinkage cavities is presented. It uses a full Navier-Stokes formulation

30 Chapter 2 Modelling of convection effects in casting and welding processes

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.

An important convection effect is the solid transport by fluid flow. If globulitic grains form in the

melt and move due to fluid flow or gravity, they carry mass, heat contents and concentration that is

different from the liquid phase. In addition they carry a cristallographic orientation.

Research takes a great interest in predicting the alloy dependent solidification morphology and

mechanisms. One of the basic problems is the question, when, how and why the columnar-

equiaxed-transition (CET) happens, i.e. which alloys solidify in columnar or equiaxed or mixed

morphology, and under which circumstances. Investigation of this topic includes intense research

on nucleation, grain growth, wall effects, growth instabilities.

The knowledge of these mechanisms can help with important problems of industrial casting:

macrosegregation phenomena, e.g. negative segregation cone at the bottom of large steel ingots

due to the sedimentation of globulitic grains

Inhomogeneous grain size distribution throughout the whole cast part (fine grain structure in

regions where globulitic grains settle, rough, columnar structure in other regions)

Closure of feeding paths. Alloy composition and cooling conditions determine when the feeding

path is blocked by stationary solid.

The formation of mobile or stationary solid also determines the development of the free surface

shape. If stationary solid forms close to the surface in early stages of the solidification process, a

deep external shrinkage cavity cannot form. The following sections describe some effects linked to

formation and movement of floating globulitic grains.

Two important growth types can be distinguished in solidification processes: Columnar and equi-

axed or globulitic growth. The growth is called columnar if the solid is fixed to a wall and grows

into the melt. The grain shape can be dendritic or cellular. The growth is called equiaxed or globu-

litic, if the solid nucleates in the melt and grows while it is swimming. The term equiaxed is used

2.3 Movement of floating globulitic grains / twophase flow 31

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.

Fig. 2.13 shows the influence of growth morphology on the solidification of the surface. The sur-

face shape is always determined by the position of the solidification front. The question, where this

solidification front effectively is, has different answers different growth morphologies. For solidifi-

cation of pure substances or eutectic alloys the front is planar and easy to locate, Fig. 2.13 a). In

case of columnar dendritic solidification, where all formed solid is stationary because it is fixed to

the wall, the front is defined by the position of the innermost solid, the dendrite tips, Fig. 2.13 b). If

the liquid level drops, the interdendritic liquid may leave the dendritic network in general it is

kept inside by capillary forces but the position of the surface is defined by the remaining dendritic

skeleton. What happens if solidification is not columnar or if there are columnar dendrites and

2.3 Movement of floating globulitic grains / twophase flow 33

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.

Solidification morphology can be influenced by many techniques

Grain size can be decreased by use of grain refiners, i.e. addition of nucleation cores (e.g. TiB).

Grain size can be decreased by faster cooling, e.g. by use of chills or quenching

Grain size can be decreased by mechanical methods like vibration, ultrasound, electromagnetic

stirring, etc.

34 Chapter 2 Modelling of convection effects in casting and welding processes

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.

As the number of publications treating the subject of formation and convective transport of floating

equiaxed grains is large, only papers dealing with two main areas which are of special interest for

the present work will be reviewed. The first area deals with the question of how the grain transport

is modelled, the second area investigates, how the repartition between mobile and stationary solid

can be described.

The approaches which describe the movement of solid can be classified as follows

Direct particle tracking: The nucleated grains are identified as single entities and the movement

of each of them is calculated separately.

Volume averaged two-phase-flow: Solid and liquid phases are each described by a separate set

of momentum equations. The interaction between them is modelled by momentum exchange

terms. The formation of a stationary solid network is achieved by increasing the effective mix-

ture viscosity.

Volume averaging with effective mixture velocity. Only one set of momentum equations is

solved. The influence of the mobile solid phase is considered by modifying the properties of the

liquid like density, viscosity, heat conductivity etc.

Especially researchers who are interested in predicting the final grain size distribution use the la-

grangian method of direct particle tracking, combined with continuum Euler-formulations for the

treatment of the liquid phase. A famous approach is the 2D CAFE (Cellular Automaton Finite

Element) model by Gandin, Rappaz and co-workers, /Gandin 98/. The authors use the FE solution

of average conservation equations for mass, momentum and energy as input for a CA model which

describes the growth of single grains or grain clusters (moving region). The grain transport is

modelled by recalculating the positions of the moving regions in each time step using the FE veloc-

ity field, gravity forces and a drag coefficient model describing the interaction between moving

regions and surrounding liquid. Even if the authors state that the presented version of the model was

still too crude to give quantitative results, comparison to experiments shows that the modelling of

grain movement is the governing factor for accurately predicting the grain structure, Fig. 2.14.

A comprehensive volume averaged two-phase-flow model has been developed by Wang and Beck-

ermann, /Wang 93, 94a,b, 95a,b, 96a,b/, reviewed in /Beckermann 95/. The model is able to de-

scribe the complete solidification process of equiaxed alloys. The forming solid phase is assumed to

2.3 Movement of floating globulitic grains / twophase flow 35

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.

The most important phenomenon linked to the repartition of solid into mobile and stationary frac-

tions, is the columnar-to-equiaxed transition (CET). An overview of the state of the art is given in

/Gandin 98/.

Transparent model alloys have been used to experimentally investigate CET. /Gao 98/ observe so-

lidification of an ammonium chloride (NH4Cl) solution in a vertical test cell cooled from the top.

They find for this system that floating equiaxed grains are created by fragmentation of columnar

dendrites and CET is mainly controlled by sedimentation. Another way to experimentally investi-

gate CET is the metallographic evaluation of cast metal samples. /Ziv 89/ study the influence of

2.3 Movement of floating globulitic grains / twophase flow 37

cast with different heat transfer coefficients and find, that stronger cooling enhances the length of

the columnar region.

This observation is consistent with the results of the classical study by Flood and Hunt, /Flood

87a,b/, /Flood 88/ who numerically model the growth of equiaxed grains ahead of a columnar front.

They find that the position of CET is governed by the existence of nuclei in the bulk and the relative

growth speed of columnar and equiaxed grains. If equiaxed growth is quick relative to the colum-

nar, the large grains can block the growth of the columnar front. If equiaxed growth is slow, or if if

it is restricted to a narrow undercooled region ahead of the columnar front, the columnar growth

simply absorbs the existing small equiaxed grains and dominates the macrostructure. The growth

speed of the two types of grains is influenced by many casting parameters as initial superheat, com-

position, cooling etc.

An experimental study which confirms the theory is /Taha 00/, who measure a critical temperature

gradient GC and a critical growth rate RC at the CET in an Al-4.4wt%Cu casting to be about 6

K/mm and 0.6-0.9 mm/s respectively. /Dupouy 98/, who study CET in Al-4wt%Cu under micro-

gravity conditions, find some deviations and explain them mainly by a solute pile-up ahead the front

which is not considered by the Flood and Hunt model.

/Gandin 00a,b/ develop a one-dimensional solidification model and propose a new CET scenario

based on a constrained-to-unconstrained growth transition, leading to breakdown of the columnar

dendritic front. The find, that there are two stages of dendritic growth: The stage, where fast growth

is driven by superheat (constrained growth) and the stage, where growth slows down, when the

temperature gradients have almost disappeared (unconstrained growth). Comparing modelling re-

sults to experiments with Al-Si alloys, he finds that the position of the maximum velocity is in vi-

cinity to the position of CET.

Simulation models which are able to describe the formation of CET, are /Nastac 00/, who develops

a stochastic model for simulating the evolution of dendritic crystals during solidification, and

/Wang 94b/, who use their two-phase model to calculate CET in Al-Cu and Sn-Pb alloys.

Understanding the phenomenon CET does not give the whole answer to the question which part of

the solid is stationary and which one is mobile. The grain structure which is a result of CET, gives

information about where the corresponding grains may come from, but it does not tell much about

the mobility of the grains. Important for the flow characteristics in the two-phase area is the ques-

tion of dendrite coherency or impingement, the point where the single equiaxed crystals start stick-

ing together and forming a stationary dendritic network. /Chai 95/ measure the point of dendrite

coherency in equiaxed dendritic solidification of Al-Si alloys as a function of alloy composition.

They find that the solid fraction where coherency of the dendritic network occurs decreases from

0.39 to 0.09 when Si concentration rises from 0.5 to 11 wt%.

38 Chapter 2 Modelling of convection effects in casting and welding processes

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 necessity to describe the movement of a partially solidified surface has led to the development

of the Split Solid Model (SSM). Experiments have shown the formation of extremely deep external

shrinkage cavities in cast ingots of unalloyed steel (Fe-0.42wt%C). As this material has a long so-

lidification range, small amounts of solid form throughout the riser already in early stages of solidi-

fication. If this solid is assumed to be stationary / columnar, the lowering of the surface and forma-

tion of a deep shrinkage cavity is impossible.

To solve this problem a new concept was developed which introduces the existence of mobile solid

fractions without the cost of a second set of momentum equations. A Split Solid Function (SSF) was

developed which splits the existing solid into a stationary and a mobile part. The ratio between

these two parts is determined by the total amount of solid. For low solid fractions all solid is con-

sidered to be mobile. For high solid fractions all solid is stationary. In between there is a smooth

transition zone. If solid forms close to a wall or close to an area where the solid fraction is already

high, columnar growth is assumed, and all solid is considered to be stationary. Otherwise small

amounts of solid which form at the wall would be considered to be completely mobile and accu-

mulate at the bottom of the cast part. The formation of a solid shell would not be possible. The ex-

act conditions and behaviour of this growth can be adjusted in the Columnar Growth Model

(CGM).

The mobile solid is thermodynamically treated as solid, but hydrodynamically as liquid, i.e. the

mobile solid behaves like a liquid with solid density, solid heat capacity and solid concentrations.

This gives additional transport terms in all differential equations, but no additional transport equa-

tions. The only equations added by the model are the two split solid equations which are expres-

sions to calculate the fraction of mobile - and - solid using the SSF.

2.4 Solidification and fluid flow: Thermal and solutal effects 39

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.

Fluid flow in solidifying areas has a big influence on the solidification behaviour. The transport of

solute and heat changes local composition and enthalpy. Important effects are the lowering of the

solidus temperature in highly enriched areas, which extends the total solidification time, and the

formation of very different weld pool shapes depending on the flow pattern during the melting pro-

cess, section 7.1. Another well known effect is that dendrites which grow perpendicular to the flow

bend upstream towards the flow. This is due to the displacement of the solutal boundary layer

which changes the local liquidus temperature. In the context of this work effects of this type cannot

be described. The focus here is on volume averaged effects which made the development of a new

solidification algorithm necessary. These effects are horizontal and vertical movements through the

binary phase diagram due to liquid transport phenomena and solid transport effects.

Industrial problems which are linked to this subject have all been treated in the sections about mac-

rosegregations and mobile solid.

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

remelting at a) constant tem-

perature and b) constant con-

centration in the liquid-solid

area of an alloy with with

segregation coefficient ls < 1.

The solid fraction decreases

by a) locally rising mixture

concentration b) locally rising

a) b) temperature.

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.

tational cell in a calculation of a stationary

welding process with Fe-0.42wt%C. When the

cell starts to melt, solute is removed by convec-

tion and the mean concentration is reduced.

During the primary -melting the laser source is

switched off and the cell starts to cool down. At

the end of the peritectic resolidification the mean

concentration of the cell has been reduced so far

that some -phase remains in the solid. The new

solidification algorithm now correctly treats the

(equilibrium-) solid state transformation inside

the + -phase area.

The consideration of the movement of floating globulitic grains adds some severe problems to the

task of the solidification algorithm. Solid grains which enter a liquid volume replace the liquid by

solid with different properties like different density, concentration, heat capacity etc. In addition

they change the local fraction of solid. The task of a solidification algorithm must be to correctly

describe the influence of the newly arriving (or leaving) solid, and to find a new equilibrium which

yields the new solid fraction considering the effects of advection, solidification and remelting, as

well as the new temperatures, concentrations and other physical properties.

Considerable effort has been directed to the development of phase change algorithms for metallur-

gical systems. An overview over a number of techniques which are available for the solution of

these special moving boundary problems and a comprehensive review can be found in /Crank 84/.

For cases, where convection effects are of importance, the modelling methodologies may broadly

be classified into three groups, /Brent 88/:

1. Empirical: Effects of convection are included in some kind of rised effective liquid conductivity

2. Classical: Full calculation of convection in the liquid, using deformed grids or transformed co-

ordinate systems to account for the position of the phase front.

3. Enthalpy: Fixed grid techniques which use enthalpy and latent heat formulations. The position

of the phase boundary is described by the solidification status, i.e. the solid fraction of the cells.

42 Chapter 2 Modelling of convection effects in casting and welding processes

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

because most technical alloys are at least ternary. As an example a method of following a complete

solidification path in ternary alloys, including the primary phase, binary eutectic, ternary eutectic

and peritectic reaction can be found in /Krane 97a,b/. The current work heads into a different direc-

tion. It adds to the modelling of more complex binary phase diagrams and is a step on the way to

find a generalised solidification algorithm for the complete description of extremely complex binary

phase diagrams like AlCu, AlMg, FeMn.

Modelling the transport of heat, concentration and latent heat due to the movement of floating

globulitic grains adds still more difficulties to the problem of solidification modelling. /Beckermann

95/ gives a detailed review on this subject.

The new solidification algorithm which is one of the core parts of the new model, can be classified

as a fixed grid enthalpy-porosity technique. Based on the concepts by /Prakash 89/ and /Schneider

95a/ a stable algorithm has been developed which can be applied to many practically important al-

loy systems like Fe-C, Al-Si, Al-Cu, Al-Mg, Pb-Sn, Ag-Cu etc.

The algorithm has the following important properties:

It always gives exact values for one liquid and two solid fractions and the corresponding con-

centrations as predicted by the equilibrium phase diagram. This is a property of the basic idea.

In the present state it can be used to simulate arbitrary (C , T ) paths through typical piecewise

linear eutectic and peritectic phase diagrams including primary solidification, eutectic solidifi-

cation and peritectic transformation as well as (equilibrium-) solid state transformations.

It is unconditionally stable and suitable for strong convective solute transport which occurs

mainly in welding problems and which results in horizontal movements through the phase dia-

gram. One of the main achievements of the new algorithm is the ability to cross phase lines and

even jump over complete phase areas without neglecting any of the consequences of the physi-

cal processes happening in between.

It is able to describe the advection of mobile solid in the framework of the volume averaging

model. The advection of the solid is not described explicitly, but in terms of advection of the

properties of the solid. The solidification algorithm finds the new equilibrium which results in a

locally rised solid fraction.

It can be extended for arbitrary types of phase diagrams, including intermetallic phases. To re-

alize this, the algorithm has to be written in a more generalised form and new solidification

modules have to be created for new solidification types.

44 Chapter 2 Modelling of convection effects in casting and welding processes

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 ...).

The previous sections describe the state of the art in modelling convection effects. They show a

high standard of sophisticated models that are able to describe various flow related physical phe-

nomena. The main aim of the present work is, in addition to introducing a number of innovative

models which describe single effects in a new way, to couple these models and to form a complex

framework that is able to describe phenomena which cannot be described by simple models.

The main task which has been governing the development was the wish to simulate the influence of

macroscopic shrinkage cavities on the macrosegregation pattern in castings. A second task was the

transient simulation of pool shapes and solute redistribution in welding processes.

The complexity of the processes in question has already been described in the introduction. There-

fore we directly proceed to the investigation of the current state of the art of highly integrated

simulation models, and of the contributions of this work.

As the number of approaches in this area is equally almost infinite, this overview is going to focus

on work directly related to the process in question, the modelling of macrosegregations governed by

the movement of the free surface.

To the authors knowledge there is no published work on the influence of deep external shrinkage

cavities on the final macrosegregation pattern of cast parts.

The MAGMAsoft module MAGMAsteel, /MAGMAsoft 02/, is able to simulate convection and

macrosegregation phenomena in steel casting together with the formation of shrinkage cavities. The

fluid flow solver has been coupled with a feeding algorithm, cf. section 2.2.2.1. The new surface

shape is taken as a boundary condition for the convection solver, and convection in turn influences

the temperature field that is an input for the feeding algorithm. To avoid problems with deep

shrinkage cavities, in most published applications so far, e.g. /Schneider 98/, insulated risers have

been used which as long as possible keep the feeding reservoir completely liquid and thus the free

surface even. Only at the very end of solidification flat shrinkage cavities form. Work is in progress

to integrate a pressure based porosity and shrinkage cavity formation model into the module,

/Schneider 03/. /Gu 99/ model the solidification of a large steel ingot and consider the influence of

the formation of a flat external shrinkage cavity on the formation of macrosegregations. The devel-

opers of /FLOW-3D 02/ are doing research in the simulation of shrinkage cavity formation and cal-

2.5 Complex integrated models for casting and welding 45

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.

The complete mathematical model which is presented in this thesis and which has been imple-

mented as the software package SoliCon integrates the following submodels:

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.

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.

The number of approaches again being large, this section will only deal with models that are com-

parable to the authors, and that are dealing with conduction mode welding models with special

weight on the prediction of weld pool phenomena.

46 Chapter 2 Modelling of convection effects in casting and welding processes

Numerical models which are designed to predict the pool shape in welding processes need to be

very complex. A lot of important interacting influences have to be considered. This is why many of

the existing models are highly sophisticated. State of the art welding models generally contain

Solution of differential equations for temperature, velocities, pressure

Description of Marangoni convection with constant or temperature dependent Marangoni coef-

ficients

Welding heat source with constant, Gaussian or parabolic power distribution for the simulation

of laser and GTA welding

Electromagnetic forces for the simulation of GTA welding

Heat losses at the surface due to radiation, air convection and vaporization (not always)

The existing models for conduction mode welding can be characterised using the following criteria:

Laser / Gas Tungsten Arc (GTA)

3D / 2D axisymmetric

FEM (Finite Element Method) / FDM (Finite Difference Method) / FVM (Finite Volume

Method) / CVM (Control Volume Method) / DEA (Discrete Element Analysis)

Steady state / transient

spot welding / line welding

Tables 2.1 (GTA welding) and 2.2 (laser welding) give an overview over important welding mod-

els of the last two decades which take into account Marangoni convection. In addition to the classi-

fication criteria from the list a number of special features of the models and the materials used in

the publications are given. If the welding model has been implemented into an existing software

package, the name is given in the middle column.

The welding model presented in this work can be classified as 2D axisymmetric CVM model for the

transient simulation of conduction mode laser and GTA spot welding processes. It has all the basic

features which are state of the art in modern welding simulations, but adds some important submod-

els which allow to make contributions to the area of concentration modelling in welding processes.

The model is able to describe solute redistribution by solving the liquid concentration equation for

the binary alloy element. It has a sophisticated solidification algorithm and considers the flow inside

a dendritic mushy zone including isotropic and anisotropic permeability models.

2.5 Complex integrated models for casting and welding 47

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

/Chan 84/ Laser, 2D, FDM, transient, spot and stationary band source, dimensionless,

line, SOLA solute redistribution in laser surface

alloying

/Lei 95/ Laser, CVM, 2D-axisymmetric, tran- Pulsed Nd:Yag Laser, AISI 304 stain-

sient, series of spots less steel

/Pericleus 95/ Laser, CVM, 2D-axisymmetric, tran- Mushy zone, post solidification stress

sient, spot history, steel

/Pirch 95, 98/ Laser, 2D/3D, FEM, steady state, line Laser surface alloying, solute redistri-

bution, AlCu

/Pitscheneder 96/ Laser, 2D-axisymmetric, CVM, tran- Top hat laser profile, 3 high speed

sient, spot steels

/Mundra 96/ Laser, 3D, CVM, stationary, line steel with increased liquid heat con-

ductivity

/Takeshita 01/ Laser, 3D, FDM, transient, line Laser-surface-melting Al-32.7wt%Cu

/Taylor 00, 02/ Laser, 3D, FVM with unstructured Solid mechanics, application to girth

meshes, transient, spot and line, welding

PHYSICA

Somehow as an appendix to this chapter which describes the modelling of convection effects, this

section is dedicated to describe the application of the complex welding model to convection effects

which are specific to welding processes. Even if some parts of the welding model are new and in-

novative, the weight was not on the development of a new model, but on the scientific application.

In this work a considerable number of basic investigations have led to important theoretical predic-

tions. For reasons of completeness the physical background is described in this place in close

neighbourhood to the general convection effects.

2.6 Application: Convection effects in welding processes 49

Marangoni convection is driven by local variations of the surface tension . Fig. 2.17 shows the

basic mechanisms that lead to Marangoni flow. Surface tension is due to intramolecular forces, Fig.

2.17 a). Molecules that are completely embedded in the liquid feel no resulting force because the

forces are equal for all directions. Molecules in the surface layers have one side without neighbours.

The missing force from this side causes a resulting force on the molecule that is directed into the

liquid (vertical to the liquid/air surface). This leads to the well known fact that liquids try to mini-

mize their surface because all surface molecules try to enter the bulk liquid.

The force which leads to Marangoni convection emerges in a similar way. It is initiated parallel to

the liquid/air surface by surface molecules with differing intramolecular forces. As a simplified ex-

ample Fig. 2.17 b) shows a top view onto a liquid/air surface in an area where two liquids with dif-

fering intramolecular forces, respectively surface tensions, meet. At this boundary between the liq-

uids the forces are not in equilibrium. If the two molecules A and B were considered to be only one

molecule, summing of all forces would lead to a resulting force and thus a fluid flow perpendicular

to the liquid/liquid boundary and directed towards liquid B. This is a very simple case, which, e.g.,

explains the behaviour of a thin oil film on water, if a detergent is added. The detergent locally low-

ers the surface tension and a very quick flow from these areas towards the detergent free areas re-

sults. It looks like the areas with detergent spread, but in reality the areas without detergent contract.

All this happens in a split second.

Similar mechanisms apply if the surface tension does not change abruptly but if surface tension

gradients occur. The gradients always result in flows towards regions with higher surface tension.

Another way to put it, is that the flows always tend to minimize the surface energy and therefore try

to minimize the areas of high surface tension (= high surface energy). As surface tension depends

on temperature as well as on the activities of surface active elements, two different flow types oc-

cur. Thermal Marangoni convection is due to temperature gradients along the surface, solutal Ma-

rangoni convection due to concentration gradients of the surface active elements. In welding proc-

esses, due to the high temperature gradients of about 1000 K / cm the thermal effect is much bigger

than the solutal one.

While Marangoni convection is limited to the very first (maybe monoatomic) surface layer, the bulk

liquid does not remain unaffected. The high surface velocity is rather coupled into the bulk by

means of viscosity, which causes the formation of convection rolls, Fig. 2.17 c). For more details

about Marangoni convection, the governing equation and the temperature and concentration de-

pendence of surface tension cf. section 3.3.2.9.

50 Chapter 2 Modelling of convection effects in casting and welding processes

Weld pool shapes, i.e. the shape of the area which has been molten at any time of the welding proc-

ess, are determined by the flow patterns during remelting and solidification. In conduction mode

laser weld pools thermal Marangoni convection is the governing effect. Surface velocities of up to 1

m/s have been observed /Kreutz 99/, /Ogawa 00/. Effects like thermosolutal convection (order of

magnitude: 1 mm/s or 1 cm/s) and solutal Marangoni convection (1 mm/s) can almost be neglected.

In GTA weld pools electromagnetic forces of about the same order of magnitude as the Marangoni

forces lead to complex interactions and flow patterns, section 7.1.4.

Already small amounts of surface active elements like sulphur and oxygen can change the behav-

iour of the Marangoni flow and thus influence the final shape of the resolidified area considerably.

Fig. 2.18 shows examples of how the addition of shielding gas containing surface active elements

can influence the shape of the molten area.

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

of shielding gas con-

taining surface active

elements on the pool

shape in welding pro-

cesses. a) GTA welds

with SO2 addition to

a) pure argon argon + 750 ppm SO2 shielding gas /Heiple

85/. b) GTA welds of

technically pure iron

with O2 addition to

shielding gas /Xiao

97/.

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/.

The local composition of the alloy in the weld pool changes during the welding process. At least

two principal mechanisms can be responsible for that:

Species losses due to evaporation at the surface. The high surface temperatures raise the vapour

pressure of the matrix material and the different alloying elements and lead to species and en-

ergy losses close to the surface. This effect is well known (burn-out, scale loss, german: Ab-

brand). The arising concentration inhomogeneities are attenuated by strong convective trans-

port.

Solute redistribution due to convection effects; the principle effects may be the same as in cast-

ing processes, which have been described in section 2.1. So far there is not much experimental

evidence of this effect, because it is superposed by the species loss due to evaporation.

Many experimental and theoretical papers have been published concerning the description and pre-

diction of pool shapes in laser and GTA spot welding. Especially the formation of flat and V-shape

pools have been largely predicted and experimentally verified.

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

The new welding model has been applied in the investigation of mainly two areas:

The influence of convection on the weld pool shape

Intensive parameter studies have been performed to study the influence of welding parameters

like laser radius, welding power, welding duration, weld material composition, viscosity, and

Marangoni coefficient on the shape of the weld pool.

A new terminology has been developed which allows the description and visualization of time

dependent pool shapes. It extends the classical depth-width-ratio and introduces a second, lateral

depth and the offset of this depth from the middle of the weld pool.

This terminology has been used to evaluate the parameter studies of calculations with tempera-

ture dependent Marangoni coefficient. In this context a number of important transient effects

has been observed and described, section 7.1.3. The results can be used by researchers to design

experiments on their search for specific effects.

This work tries to contribute to this area by a number of numerical and experimental investiga-

tions concerning the formation of macrosegregations on the scale of the weld pool. The most

important investigations are:

Simulation of the formation of macrosegregations on the scale of the weld pool, section

7.2.2.

Influence of the melt viscosity on the formation of macrosegregations, section 7.2.3.

Observation of the mushy zone in welding of Al-7wt%Si-0.3wt%Mg, section 7.2.4.

3 Theory

In this chapter, the framework of mathematical and physical models for use in the casting simula-

tion code SoliCon is developed. The models used in the welding simulation code EL2D are basi-

cally those described in /Schneider 93, 95b/. They were extended by a laser and GTA welding

model (Marangoni + Lorentz convection), cylindrical co-ordinates and by early versions of the new

solidification algorithm and the algorithm for the solidification of moving free surfaces.

Except of the momentum equation, which has been written in 2D formulation, all equations have

been developed in vector notation and can therefore be used for full 3D treatment. The extension of

the momentum equation to 3D is some work, but should be easy enough to do.

Section 3.1 gives a general overview of the problem, which quantities are to be calculated, which

equations are used and which assumptions reduce the complexity of the model enough to make it

solvable. Section 3.2 describes further assumptions and definitions used during the derivation of the

equations which are then developed in section 3.3.

One central idea of the model is the introduction of mobile solid fractions without introducing addi-

tional solid momentum equations. This is achieved by treating the mobile solid fraction thermody-

namically as solid, but hydrodynamically as liquid with solid density. This model will be called

Split Solid Model (SSM).

The mathematical model considers the nine phases

liquid: q=l

mobile solid - phase: q = move

mobile solid - phase: q = move

stationary solid - phase: q = dead

stationary solid - phase: q = dead

shrinkage porosities: q = pore shrink

hydrogen porosities: q = pore hydro

obstacles: q = obst

void: q = void

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 hydrogen and

3.1 Mathematical model 59

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:

Pressure: pq

Temperature: Tq

Concentration of component i in phase q: C qi

Phase fraction of component q: q

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 binary alloy element (C in case of Fe-C and Si in case of Al-

Si). The high number of variables can be reduced by a number of assumptions:

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)

Tl = T move = T move = T dead = T dead = Tobst = T pore shrink = T pore hydro = Tvoid = T (3.3)

Remaining temperature variables: T (1)

Thermodynamically the mobile and stationary fractions of - and -phase are considered to be

one single phase. They have identical concentrations:

60 Chapter 3 Theory

C iporeshrink = C iporehydro = Cobst

i

= Cvoid

i

=0 (3.4)

Remaining concentration variables: Cli, Ci, Ci (3nelements)

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

Continuity equation f(ul, vl, q) 1 equation

Energy equation f(T, ul, vl, q) 1 equation

Liquid concentration equation Cl f(Cl , ul, vl, q)

i i

nelements equations

Solid concentration equations C , C f(C /C , ul, vl, q)

i i i i

2 nelements equations

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:

l = 1 move move dead dead obst pore shrink pore hydro void (equation 1)

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

i i

i i

T (C ) = T

i

(equation 8)

62 Chapter 3 Theory

+ = alloy (equation 9)

Now there is a complete system of equations to solve the problem. The next section will give a de-

tailed description of these equations.

In addition to the assumptions made in section 3.1:

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

T

hq (T0 q ) + c P q dT = h0 q + c P qT T h0 q + c P q (T T0 q )

T

hq = = (3.6)

0q

T0 q

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

mat = q (3.10)

q void

q ( q q ) q ( q q )

eff mat = = ( := 0) (3.11)

q

void

mat

q void

*q = (h0 l c P l T0 l ) (h0 q c P qT0 q ) = const . (3.12)

The current software can be used for all typical eutectic and peritectic equilibrium phase diagrams

like the two examples shown in Fig. 3.1. The phase diagrams do not have to be complete, but can be

clippings of more complex phase diagrams.

For the applications of this work the complete eutectic phase diagram for Al-Si and a clipping of the

peritectic area of the Fe-C diagram (0 1wt% C, 1400 1850 K) were used. The complete data are

listed in App. 9.1.3.

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.

The mass conservation equation for a single phase in multi-phase situations is (units:[kg/(s m3)]):

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

v

q q = q ( q q )

t

+ [( l l + move move + ) ]= 0

v

move move v l (3.15)

v

q q =

t

( move eff move + dead eff dead ) + ( move eff move vvl ) = 0 (3.16)

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

For each phase there is a separate complete momentum conservation equation:

( vv ) + v ( vv vv ) = v p + v t + gv + Mv + S (3.17)

t q q q q q q q q q q p q p q external q

(1) (2) (3) (4) (5) (6) (7)

The meaning of the terms is:

1. Transient term, time dependence of the momentum components

2. Momentum transport due to convection (fluid flow)

3. Acceleration due to pressure gradients

4. Dissipation term, momentum loss due to friction inside the single phase

5. Buoyancy term

6. Momentum transfer due to interaction with other phases

7. Momentum sources or sinks due to external influences

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

seepage velocity

u q

( q vq ) =

v

q q q

v

q t q

t q

t

(

l l + move move + move move ) ul

=

move eff move

ul

v

vl t

l

66 Chapter 3 Theory

This term contains a dyadic product that describes the coupling of the different velocity compo-

nents. In the same way as the energy equation allows energy to advect from all directions, momen-

tum can be advected from all directions. Thus the x-momentum equation contains convection terms

including as well as terms including .

u v

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 ]

It is assumed that only mobile phases add to the acceleration due to the pressure gradient.

v v v

( q p) = ( l + move + move )p = move p

q

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.

This term contains a dyadic product too to couple the velocity components. Following /Ni 91/ in the

momentum equation for the liquid phase t can be written as

v v

t

= { ( vv ) + [ ( vv )] }

l l l l l

t

3.3 Equations 67

( q u q ) ( q u q ) ( q v q )

2 +

x

q

v t

q

v

{

v

[ v

= q ( q v q ) + ( q v q )

v v

]}

t v

= q

( u )

x

( v )

y

( q q)

v

q q q q + q q

2

y x y

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

v v

v t [(

l l + move move + move move ul ) ]

q

q = v

+ [(

v

+ move move v l ) ]

l l move move

+

l

x

l

x

+ move

x

+ move

x

v

( )

+ l l + move move + move move ul

l

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

v v

[

eff move move u l ] move v v move v v

] + x eff move x vl + ( moveul ) + y eff move y vl + ( movevl )

v v

[eff move move v l

68 Chapter 3 Theory

The general buoyancy term

q [ q q ]g

v

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:

Solutal convection, induced by differences in the local chemical composition. The density dif-

ferences due to different concentrations of the binary alloy element are described by the solutal

q

expansion coefficients for each single phase q. The solutal term assumes that the density

Cq

of the single phases is modified linearly with the concentration of the binary alloy element.

Sedimentation effects due to different local phase fraction distributions.

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

3.3 Equations 69

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

v

Mqp

qp

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

M obst move

M obst dead

M obst dead

M obst obst M obst pore sh M obst pore hy M obst void

Mvv pore sh move

Mvv pore sh dead

Mvv pore sh dead

Mvv pore sh obst Mvv pore sh pore sh Mvv pore sh pore hy Mvv pore sh void

Mvpore hy l Mvpore hy move

Mvpore hy move

Mvpore hy dead

Mvpore hy dead

Mvpore hy obst Mvpore hy pore sh Mvpore hy pore hy Mvpore hy void

M void l M void move M void move M void dead M void dead M void obst M void pore sh M void pore hy M void void

1. Interactions of the phase with itself are already considered in the dissipative term, therefore the

diagonal elements disappear (9 terms).

2. All interactions with the void phase are neglected (16 terms).

3. All interactions with the pores are neglected (26 terms).

4. All interactions with the blocked cell portions (obstacles) are neglected (10 terms).

70 Chapter 3 Theory

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

v ( l l + move move + move move )( dead + dead )2

M Darcy =

vvl

K 0 ( l + move + move )

2

This term contains mathematical expressions, which cause (volume-) momentum gain or loss by

external sources. Introduction of the model assumptions yields

( q S external q ) = ( l Sexternal l +

q

move S external move + move S external move )

The complete momentum equation for the case of isotropic permeability reads

v

u

[

l move eff move l l

v

v

= move

vu]

p

move eff move v

t

l

+

v move eff [

move

v

l l

vv]

v v

+ v

[

eff move move l

v +

u]

move v v

v move v

u

v

v v

l + ( move l ) eff move l + ( move l )

eff [

move move l v]

eff move

x +

x

y y

( l l + move move + move move )( 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)

As Marangoni convection is driven by surface tension forces inside a thin, maybe monoatomic sur-

face layer, the corresponding forces are not directly written as part of the momentum equations, but

they are expressed as stress terms and added as a boundary condition. The governing equation de-

scribing both thermal and solutal Marangoni stresses is

u T a

Mara = l

=

+

i

(3.23)

z T r i

a i

r

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

Gas Tungsten Arc (GTA) welding is a welding technique which uses high currents to melt the weld

material. The high currents come with strong electromagnetic (EM) fields and therefore influence

the flow field of the molten metal in the weld pool.

The model used here to describe EM forces for GTA welding follows /Kou 85/. It is formulated in

cylindrical coordinates and has been implemented into the old software (EL2D) used for the main

part of the welding calculations.

The electromagnetic body forces are due to the interaction between welding current and the selfin-

duced magnetic field and can be expressed as

v v v

FEM = J B (3.25)

The model assumes that the electric and magnetic fields are not significantly changed by the flow of

the molten metal. In this case the vertical current distribution Jy on the weld pool surface can be

described as

2

3I 3r

Jz = 2

exp

2

at z =0 (3.26)

b b

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.

The energy conservation equation for a single phase q is (units: [W/m3]):

v v v

( q hq ) + ( q q v q hq ) = q q

v

Hq p q S external q (3.29)

t q + +

p l

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

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

+ [(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 exchange with the environment is described by boundary source terms. The heat flux

through the surface can be written as

qsurface = qheat source qradiation qconvection qevaporation units: [W/m2] (3.33)

The terms have been modelled as follows.

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

~ v

t

( v

) ( v

)v

(

l l Cli + l l vl Cli = l l Dli Cli + l S external l )

(3.42a)

~ v

+ (

v

) (

v v

) (

q q Cqi q q vq Cqi + q q Dqi C qi + q S external q )

q t

l

(

v

) (

v v

) ~v(

l l Cli + l l vl C li = l l Dli Cli + l S external l )

t

[ ] v v ~i v i ( ) (3.42b)

( )

t ( q move + q dead ) q C q q move q v l C q + q dead q Dq dead C q

i v i

+

+ q move S external q move + q dead S external q dead

q = ,

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

open = l + move + move + pore shrink + pore hydro + void

(3.45)

closed = dead + dead + obst

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

v

t

( open eff open ) + ( open

v

eff open vl ) =

t

( closed eff closed )

If the continuity equation is multiplied by Fmove:

v

Fmove

t

( open eff open ) + Fmove ( open eff openvvl ) = Fmove

t

( closed eff closed )

and the VOF equation by open eff open :

Fmove + vv v F = 0

t open eff open l move

v

( 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.

For solid fractions 0 mv dd 12 mv dd the fraction of mobile solid decreases slowly down to zero

For high solid fractions all solid is stationary (SSF = 0)

SSF must be continuous and continuously differentiable

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]

In the cubic spline function the sinus segment is replaced by a cubic polynom. The four coefficients

have been determined by the conditions that the function values and the derivatives at the connec-

tion points must be equal to the values and derivatives of the constant parts. The function is defined

as:

1 for + [0 , ]

SSFspline ( + ) = a x 3 + b x 2 + c x + d for + ] , + [ (3.50)

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

+ + +

The third function, the Fermi-Dirac distribution function, was originally developed to describe the

temperature dependent distribution of electrons on the energy levels in solids. It is not defined

piecewise but as one term:

SSFFD ( + ) =

1

+ [ 0 , 1] (3.51)

for

( + 0 mv dd ) +

10

exp 1

mv dd

82 Chapter 3 Theory

tion types which can be used as Split

Solid Functions a) SSF b); derivatives

of SSF; c) fraction of mobile solid as a

function of the total solid fraction:

move = SSF ( +)

+

a)

b)

c)

3.3 Equations 83

Fig. 3.2 shows the different function types and their derivatives for the case 0 mv dd = 0.3 and mv dd

= 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.

The Columnar Growth Model (CGM) describes the fact that solid which forms at the wall grows as

columnar, stationary solid, even if the amount is small. In a pure split solid model, this solid would

be considered as completely mobile and accumulate on the ground. Therefore no solid shell could

form at a vertical wall.

To avoid this the Split Solid Model is combined with a Columnar Growth Model, which decides, in

which areas the Split Solid Function is applied and in which all solid is considered to be stationary,

independent from the total amount of solid.

As the model is a volume averaging model it is not possible to distinguish between columnar solid

which grows at the wall and equiaxed grains which form in the liquid close to the wall, a criterion

has to be developed which decides if the local area is an area of columnar or equiaxed growth.

It is not objective of this work to develop a sophisticated model of all processes, but to develop a

simple model which can describe some of the phenomena and to provide the framework for the im-

plementation of more sophisticated models in the future.

So far a very simple model has been applied which checks a circle with fixed radius RCGM around

each point. If inside the circle there is any point which is either wall or obstacle or has a solid frac-

tion bigger than CGM, the split solid model is not applied but all solid is considered to be stationary.

Fig. 3.3 demonstrates the ideas behind these assumptions. Fig. 3.3 a) shows an area close to a wall.

Solid which forms in a small distance, RCGM , is considered to stick to the wall and to help forming

the solid shell. Solid which forms in a greater distance becomes a floating grain. Fig. 3.3 b) consid-

ers the case that new solid forms not close to a wall but close to an area where there are already

high fractions of solid present. If inside a radius RCGM around the position where the new solid

84 Chapter 3 Theory

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.

A new model has been developed which is able to quantitatively predict the formation of hydrogen

and shrinkage porosities as well as internal shrinkage cavities. Of course the prediction hydrogen

porosities only makes sense for aluminium alloys.

There are two different mechanisms responsible for the formation of hydrogen porosities on the one

hand and shrinkage induced porosities and cavities on the other hand. Hydrogen porosities form by

rejection of hydrogen bubbles during solidification and cooling due to the reduced hydrogen solu-

bility in the solid phase. If hydrogen bubbles form they push the local liquid and mobile solid

3.3 Equations 85

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.

In a first step the assumption is used, that the governing mechanism of hydrogen porosity formation

in aluminium alloys is the severely reduced solvability of hydrogen in the solid phases compared to

the liquid phase. We assume that the amount of hydrogen which can be dissolved in the liquid

phase, is limited to the initial amount. In the solid phases a fixed amount VH max which is smaller

than the maximum amount in the liquid can remain dissolved. In the case of aluminium alloys we

use the rule applied by the foundry industry that a maximum of VH max = 0.1 cm3 / 100 g alloy is

allowed to prevent the formation of hydrogen porosities. If solidification occurs, the excess hydro-

gen is rejected and forms bubbles.

If the formation rate of new solid is t , the volume rejection rate of hydrogen in the volume

s

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

ume fraction of hydrogen porosities pore

hydro. How much volume a certain mass

of rejected hydrogen occupies, depends

on the local pressure of the fluid which

compresses the hydrogen until the pres-

sure inside the bubbles is in equilibrium

with the outside pressure. Even for high

pressures hydrogen porosities should

appear, but small ones. If local pressure

is reduced inside a dendritic network,

rejected hydrogen should form large

amounts of low pressure hydrogen po-

rosities which look like shrinkage poro-

sities.

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).

and velocities in a rectangular elbow of a tube.

The change of direction is achieved by means of

a pressure hill in the corner. The horizontal

velocity component is decreased because it has

to run uphill, the vertical component is in-

creased by running downhill. a positive pres-

sure gradient slows down, a negative one accel-

erates.

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.

function of the solidification status of the mushy

zone. Solid curve: Real pressure inside the hot

spot; dashed curve: pressure that would be nec-

essary to fully compensate the volume losses due

to shrinkage.

4 Algorithms

4.1 General properties of the code

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

and phase fractions and calculate new volume averaged phase fractions to be expected. This

step contains a pre-evaluation of the energy and concentration equations. Calculate new volume

fraction of hydrogen porosities.

3. Iteration 1, SIMPLER algorithm: (simpler) Take new volume averaged phase fractions, old

VOF and calculate new pressures and volume fractions of shrinkage porosities, velocities, tem-

peratures and concentrations.

4. Iteration 1, free surface algorithm: (update_free_surface) Take new volume averaged

phase fractions and velocities and calculate new VOF.

4.3 Discretisation of differential equations 95

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.

The rectangular calculation grid (normal grid) is subdivided into rectangular cells. Most of the

physical quantities, especially phase fractions, temperature, concentrations, densities and pressure

are defined in the middle of these cells and meant to prevail over the whole cell (control volume).

This does not hold for the localised phase fractions used to define the free surface configuration and

obstacles. These phase fractions are defined to be 1 in some parts of the cell and 0 in others. Due to

numerical reasons the velocities are calculated on staggered grids that are shifted one half of a

cell width in the direction of the corresponding velocity component (to the left for u, to the bottom

for v). This results in the velocity values being located on the cell faces of the normal grid. Fig. 4.2

shows the definition of the different grids. Each grid is surrounded by a line of virtual boundary

cells with zero volume. There are no equations solved for these cells, but they are used to set

boundary conditions.

The notation of the discretisation follows the one used by /Patankar 80/. All equations are cast into

the shape of the discretised general differential equation for the variable :

a P P = aW W + a E E + a S S + a N N + b

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.

For all equations the convection-diffusion term is discretised using a power law scheme. The influ-

ence of diffusion is determined by the function

4.3 Discretisation of differential equations 97

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.

The discretisation of the differential equations requires different types of discretised continuity

equations:

Implicit continuity equation on the normal grid for implicit pressure equation.

Explicit continuity equation on the normal grid for liquid concentration equation and the same

using effective quantities for temperature equation.

Explicit continuity equation on the x-/ y-staggered grid for momentum equations.

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

Integration using an implicit discretisation scheme yields

t + t

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

An explicit discretisation scheme yields

t + t

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

A special case is the continuity equation of the x- and y-staggered grid. Explicit integration over the

x-staggered control volume ( x ) y z yields

t + t

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

As the discretisation is carried out on different staggered grids for each momentum component the

components of the momentum vector equation can be considered separately. The procedure will be

shown for the x-momentum equation only. The result for the y-momentum equation can be derived

by easy transformation rules.

( move eff move ul ) + v [ move eff move vvl ul move eff move v ul ] = move p

t x

+ eff move ul + eff move + eff move u l + eff move

x x x y y x

(

move eff move dead + dead )2

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

v v v v

[ move eff move vvl ul move eff move ul ] = Ju

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 ]

t + t

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

( move eff move )w [(ul )w (ul )w ]( x ) y z [(qdiscretised )w (ul )w ]

0 0 0

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

x P x W ( x )

) (

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 )

t + t

( moveul )

(

d x d y d z t

x yz

)

x

eff move

x

dt

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

0

= t

( x )

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

)

=

( x ) y z t

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

( (

( eff move )W0 (ul )W0 move w

x ( x ) ( x ) ( x )

u ) ( move ul )w

0 0

u ) ( moveul )ww

0 0

( (

( eff move )W0 move l w ( x ) y z

x ( x ) x ( x )

0 ( move )Nw ( move )w

0 0

( ) ( move )Sw

0 0

( x ) y z

+ ( eff move )0nw (ul )nw ( eff move )0sw (ul )0sw move w

( y )+ y ( y ) y

102 Chapter 4 Algorithms

v ) (

0 0

v ) ( move vl )sW

0 0

( v ) (

( 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

+ l C external l

( x ) y z

( 4.6 )

+ (

l P external l

S + S

move P external move

+ S

)

move P external move w l w

(u )

a w (u l )w = a e0 (u l )e + a ww (u l )ww + a Nw (u l ) Nw + a Sw (u l )Sw + b ( move )w [( p )P ( p )W ] y z

0 0 0 0 0 0 0

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

a w0 =

t

0

= (Du )W0 A((Pu )W0 ) + [(Feff move )W0 ,0]

a Nw

0

= (Du )0nw A((Pu )0nw ) + [ (Feff move )0nw ,0] a Sw

0

= (Du )0sw A((Pu )0sw ) + [ (Feff move )0w ,0]s

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 )+ /

(Pu )P / W / nw / sw =

0

(Du )P / W / nw / w

0

s

0

( 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

0 0 0 0

( (

+ ( eff move )0nw move l n ( eff move )0sw move l s

( x ) y ( x ) 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

u

( l )w

0

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 by , by .

u v v u

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

Temperatures of the old time step are used to calculate the temperature gradient. This is a meas-

ure to prevent oscillations of the implicit SIMPLER algorithm. As in general the time step width

is extremely small (1/1000 s 1/20000 s), this seems to be reasonable.

104 Chapter 4 Algorithms

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

The equation to be discretised is:

[( move eff move c P eff move + dead eff dead c P eff dead )T ] v v

+ [ move eff move c P eff move v l T mat eff mat T ]

v

t

* ( q q )

v

[* ( q q vl )]+ q ( q S external q )

v

= q +

q l t q = move q

move

4.3 Discretisation of differential equations 105

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

t + t t + t

v v (J T )x ( J T ) y

dx dy dz t ( J T )dt = xydx

dy dz + dt

x y

x y z z t

0

( J )0

[ ]

(J T ) y

T x dx dy dz = ( J )0 ( J )0 + ( J )0 ( J )0

= t

x

x y z

+

y

T e T w T n T s t

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]

t + t

[(

move eff move c P eff move + dead eff dead c P eff dead )T ] dt =

dx dy dz

x y z

t t

(

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

W

(Feff* move )e / w / n / s = (Feff move )e / w / n / s (c P eff move )P

0 0 0

Einheit :

K

106 Chapter 4 Algorithms

dead eff dead )0P ( dead eff dead )P ](c P eff move )0P TP0

[(

+ x yz

+ ( dead eff dead )P (c P eff dead )P TP ( dead eff dead )P (c P eff dead )P TP

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

(Feff* move )e / w / n / s

0

(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

q

[(

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

a P TP = a E0 TE0 + aW0 TW0 + a0N TN0 + a S0 TS0 + b

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

aP0 = (cP eff move )0P

t

x y z

aP0 * = [(cP eff move )P ( move eff move )P + (cP eff dead )P ( dead eff dead )P ]

t

[

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

( C i ) + v ( vv C i D~ i v C i ) = S i + S 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 + t

(

l l Cl

i

) dt = [( ) (C i ) ( ) (C i ) ] x y z

dx dy dz t

0 0

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

t + t

dx dy dz (S l + l S external l )dt

i i

x y z t

v v

= 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

0 0 0 0 0 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

a P (Cli )P = a E0 (Cli )0E + aW0 (Cli )W0 + a N0 (Cli )0N + a S0 (Cli )0S + b

with

a P = (a P0 S P x y z )

x yz

a P0 = ( l l ) P

t

l

0

l

0 0

aW0 = (DCi l )w A (PCi )w + [(Fl )w , 0]

l

0

l

0 0

l

0

l

0 0

a S0 = (DCi l )s A (PCi )s + [(Fl )s , 0]

l

0

l

0 0

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

+ [

q =

e

0

E l w W

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

l

0 0

aW0 = (DCi l )w A (PCi )w + [(Fl )w , 0]

l

0

l

0 0

l

0

l

0 0

aS0 = (DCi l )s A (PCi )s

l

0

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

The SIMPLER (Semi Implicit Method for Pressure Linked Equations Revised) algorithm by

/Patankar 80/ can be used to find the solution of a system of linked nonlinear partial differential

equations. Inits original version the differential equations to be solved are used in implicit discreti-

sation. The non-implicit part which makes the method semi-implicit is the use of an iterative proce-

dure to solve for nonlinearities.

In the model presented here the ideas of the SIMPLER algorithm are at least in parts applied to ex-

plicit differential equations. This causes some difficulties. A completely explicit version of the algo-

4.4 Semi-explicit version of the SIMPLER algorithm 111

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.

The x-momentum equation in semi-explicit discretisation can be written as

(u l )e / w = (u l )e w + (d )e

/ / w [( p )P / W ( p )E / P]

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 /

To get the implicit pressure equation we insert the semi-implicit velocity equations

(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 ]

112 Chapter 4 Algorithms

( ) [(u ) + (d ) [( p ) ( p ) ]] y

[( q q )P ( q q )0P ] x y z + ( move eff move )w[(v l) w+ (d ) w[( p ) W ( p ) P]] zt = 0 (4.9)

move eff move n l n N

+ x

q n P

( move eff move )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

aN / S = (d )n / s ( move eff move )n / s x z

b = [( )

q

q q P

0

( q q )P ]

t

+

+ [( move eff move )s (vl )s ( move eff move )n (vl )n ] x z

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.

The discretised pressure equation is overrelaxed using a method proposed by /Patankar 80/.

4.5 Stable phase change algorithm 113

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 aim to have a program which is especially suitable for the calculation of convective solute

transport in casting and welding processes makes great demands on the phase change algorithm (in

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 fraction values

, , l with + + = alloy

l (4.11)

which guarantees the simultaneous conservation of mass, heat and solute. alloy is the total volume

fraction covered by alloy material and therefore taking part in the solidification process. The total

volume balance for each cell is:

alloy + pore + obst + void = 1

One key problem of solidification algorithms is how to treat the onset and end of solidification. For

cases that 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. 2.16 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.

The problem to be solved by the solidification algorithm can be described as follows: For each new

time step the discretised differential equations for temperature and solute concentration start from a

point (C 0, T 0 ) in the phase diagram and yield a point (C , T ). The position of this new point

(C , T ) depends on

114 Chapter 4 Algorithms

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 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.

The initial step of the solution strategy is to find out the region of the phase diagram where the final

solution may lie and the direction in which the system is likely to develop. This information is

given by two positions of the alloy system inside the phase diagram:

The position (C 0, T 0 ) at the end of the last time step.

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 correct solidification type has been found, the set of new phase

fractions is returned to the main algorithm and the differential equations are now solved using the

correct new phase fraction values.

Fig. 4.3 shows two flow charts of the new solidification algorithm, the main subroutine calcu-

late_phase_fractions which is responsible for preparations and application of correction

rules and the core routine solidification_type_new which checks all possible solidification

types for the path between (T 0, C 0 ) and (T *, C * ) using the solidification modules.

After this overview of the solution strategy the following sections give a detailed description of the

most important parts of the solidification algorithm: The solidification modules, the guessing algo-

rithm, the checking rules and the correction rules.

116 Chapter 4 Algorithms

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

module for primary -solidification

based on /Prakash 89/. The other solidifi-

cation modules work similarly.

To model simple eutectic and peritectic phase diagrams three modules are necessary:

primary_solidification(delta/gamma) yields one new solid fraction as well as the new

temperature and liquid concentration and gives reasonable results inside the + liquid and +

liquid area.

solid_state_trans yields one new solid fraction as well as the new temperature and one of

the solid concentrations. It gives reasonable results inside the + area.

eutec_peritec yields two new solid fractions (temperature and liquid concentration are

fixed anyway) and gives reasonable results on the eutectic / peritectic plateau.

All modules use the terms of the discretised temperature and liquid concentration equations, write

them as functions of the new phase fractions and solve the resulting equations using one- and two-

dimensional Newton-Raphson iteration algorithms. The next sections give a brief overview of the

procedures.

The concept which is used in all three solidification modules is basically described in /Prakash 89/.

Fig. 4.4 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

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)

0 0 0 0 0 0

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 0*

a P + a TP 0* ( , ) (S PT + S PT ( , )) x y z )TP

= nb

0 0 T0 T0

a nbTnb + a P + a P ( , )

nb

0 0

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

4.5 Stable phase change algorithm 119

0

l

0 l

0

, ) a nb (Cli )P + (SCC + SCC ( , )) x y z

0

l l

(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

( ) = ( )

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

The solid state transformation inside the + phase area has been modelled in a very simple way.

Same as for solidification we assume equilibrium. Even if this is a reasonable assumption for some

solidification systems it is quite far from reality for solid state transformations on the time scales

considered. The decision to use this kind of model was based on the desire to have a closed system.

The current model does not allow the use of one non-equilibrium phase area in the framework of

equilibrium phase diagrams.

The mechanism of equilibrium solid state transformation is as follows. The system at a certain tem-

perature and a certain mixture concentration

C + C

C= (4.18)

+

(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 ) +

f ( ) = [( )P ( )] ( alloy

T C

T

, )= 0 (4.20)

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

each single cell with indices i, j.

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

)= T 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

T * ( , ) = T ( , ) Teut / peri = 0

(4.22a,b)

C * (

l , ) = C ( l , ) C l eut / peri =0

122 Chapter 4 Algorithms

T * ( 0 , 0 ) T * ( 0 , 0 )

T * ( 0 , 0 ) + ( 1

0 ) + (

1

0 ) = 0

(4.23a)

C * ( 0 , 0 ) C * ( 0 , 0 )

C * ( 0 0 ) +

,

l

( 1

0 ) +

l

( 1

0 ) = 0

l

T *0 T *0

( 1 0

) = T *0

(4.23b)

*0

l

C C*0

l

1 0

( ) C *0

l

T *0

1 0 T *0

+ C *0 T *0

T *0

C *0 T

* 0 *0

*0

C

*0

l T C

=

l l

T T

*0 l *0

(4.24)

C *0

l

T *0

1 0

+ T *0

C

*0

C

*0

l

*0

l C

=

T *0 l

T *0

The equations are implemented as subroutine eutec_peritec and yield , , T = T eut / peri ,C =

l

l eut / peri

As each call of a solidification module entails a time consuming iterative process, the number of

calls should be kept as small as possible. Therefore a guessing strategy has been developed which

determines the correct solidification module to apply in the current situation as quick as possible.

As said before, the decision is based on two positions of the alloy system inside the phase diagram,

(C , T ) at the end of the last time step and the preliminary position (C , T ) determined by

0 0 * *

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

algorithm. Example for the case of a peritectic

system cooling down in mid of primary -

solidification at the moment just before the

primary solidification is finished and peritectic

solidification starts.

(C , T ): State of the system after the last

0 0

* *

cured. The numbers denote possible final

states (C, T ) of the system if different solidifi-

cation types occur (1) primary -

solidification; (2) primary -solidification is

finished and peritectic solidification starts; (3)

primary -solidification and peritectic solidi-

fication are finished and primary -

solidification starts.

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

If, after finishing the corresponding sequence, no solidification module has been applied with suc-

cess, there are two possibilities: If the target phase area PA is one of the single phase areas 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 *

transformation / , e/p = eutectic / peritectic solidification; ? : Check for success. If success, exit

sequence, if no success, continue sequence; EXIT: Sequence is finished, finish guessing algorithm.

PA : l

0

PA :

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

PA : eutectic / peritectic plateau

0

NO SUCCESS:

PA : eut / peri

*

e/p ? EXIT PA : l, ,

*

e/p ? EXIT Do nothing, assume that solidification /

e/p ? EXIT remelting process is finished.

+l e/p ? p ? EXIT +l,l+,+

l + e/p ? p ? EXIT use old solidification type (security

measure), apply corresponding solidifi-

+ e/p ? _ ? EXIT cation module:

l e/p ? +l p

if(C < C

l l eut/peri ) p ? EXIT l+ p

if(C C

l l eut/peri ) p ? EXIT + _

Apply corrections

There is one set of rules to detect and one to correct non-physical phase fraction values yielded by

the solidification modules. To understand why they are reasonable we have to investigate the be-

haviour of the solidification modules inside and outside their corresponding phase area.

Each solidification module yields a consistent, reasonable set of phase fractions, temperature and

concentrations only for one phase area. This means, application of the module in a correct situation

results in a (C, T ) point inside the corresponding phase area, and physically reasonable phase frac-

tion values that are consistent with the phase fractions allowed in the corresponding phase area.

Whereas application of the module in an improper situation results in some inconsistency as a

(C, T ) point outside the corresponding phase area or non-physical phase fractions.

To illustrate the theory the characteristics of the solidification modules pri-

mary_solidification(delta) for two-phase-areas and eutec_peritec for the three-

phase-area will be demonstrated.

Fig. 4.6 shows the situation in the case of primary delta solidification in a peritectic system. For all

time steps of a solidification process which start end end inside the + liquid phase area the solidi-

fication module primary_solidification(delta) always yields a (C , T ) point inside the

+ liquid area and solid fraction 0 . This is guaranteed by the lever rule.

alloy

4.5 Stable phase change algorithm 127

solidification module for primary -

solidification. If in one time step the system

leaves the + liquid phase area, the applica-

tion of the solidification module to calculate

the new phase fractions, temperature and liq-

uid concentration yields inconsistencies like

< 0, > or C outside the phase area.

alloy l

point and phase fraction inside the + liquid

area, the solution is correct and the module for

primary -solidification applicable. If not,

other solidification modules have to be tried.

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

Similar behaviour is observed for all liquid-solid phase areas in the eutectic and peritectic systems

considered.

During the three-phase-solidification process on the peritectic plateau the situation has to be consid-

ered in a different way. As peritectic solidification happens at constant temperature T a T/C phase

peri

128 Chapter 4 Algorithms

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

Even if the eutectic reaction works different, similar events signalise that the solidification module

has left its definition area. In contrast to the peritectic reaction where the amount of -solid

4.5 Stable phase change algorithm 129

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

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

l l

130 Chapter 4 Algorithms

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

peri 0

eut l = alloy 0 + alloy

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

eut 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.

Now we have all information necessary to formulate the rules used to check the results of the dif-

ferent solidification modules. This is done in a logical function l_judge_phase_fractions.

Table 4.2 shows the rules defining the criteria for successful application of the solidification mod-

ules. For eutectic and peritectic reactions the phase fraction conditions are sufficient and no addi-

tional concentration conditions needed. They only give hints in which concentration interval e.g.

< 0 can occur.

The correction rules result directly from the checking rules. For the case of the system leaving a

two-phase area and entering a single-phase area the phase fractions are just cut down to the physi-

cally possible range 0 . q alloy

4.5 Stable phase change algorithm 131

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

concentration and lever rule:

< 0

= 0

= ( )/( )

alloy C C eut peri C eut peri C eut peri

> =

=

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

Special care had to be taken with the interpretation of blocked out areas. NASA-VOF2D offers a

partial treatment which allows to define a part of a cell as filled with solid obstacle. This obstacle is

localised in the cell and defines the fractions of cell faces which are closed to the flow. The obstacle

configuration is defined from the beginning and is not time dependent. The flow area is only cov-

ered with liquid.

The new algorithm considers solidification processes at the moving free surface as well as obsta-

cles. Therefore it has to take into account time dependent configurations of solid and obstacles. In

addition the forming solid is not a localised phase fraction, but a mixed one which has volume aver-

aged and localised properties at the same time . The volume which contains the cast part is a lo-

10

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.

The discretisation of the VOF equation

v

t

( open eff open Fmove ) + ( open eff open vvl Fmove ) = Fmove

t

( closed eff closed )

(eff move )P 0

( closed ) (eff closed ) ( closed ) (eff closed ) 0 0

(Fmove )P 1

0 P P P P

(eff move )P

( eff move )P

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 density of the void is taken to be equal to the effective density of the mobile phases. Other-

wise the volume of the fluid advected from the neighbour cells would depend on the filling state

at the cell face. Defining the densities to be equal yields eff open = eff move, and terms like

( eff move )s ( eff move )P describe normal shrinkage phenomena.

The new value (Fmove )P is defined as the relative fluid level which would result under action of

gravity if all the solid would be piled up in a vertical column of the full cell height. This defini-

tion, which sounds a bit arbitrary, is just another way to put, that (Fmove )P is the relative fraction

of the area open to flow which is filled by mobile phases. The fraction ( move )P therefore turns

out to be ( open )P (Fmove )P .

In this model the fraction of the closed areas consisting of solid and obstacle is defined as

( dead )P = ( closed )P (Fclosed )P where (Fclosed )P = 1 . This makes disappear the factor Fmove from

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

in the free surface boundary equation,

eq. 4.32. The fluid is assumed to be

below the free surface in this example.

Source: /Torrey 85/.

The surface reconstruction algorithm determines the new position of the surface line inside each

cell, after the advection algorithm has calculated the new filling states. The currently used surface

reconstruction algorithm is similar to the one described in /Nichols 80/ and will not be described in

detail. The principle is the following: The filling states of the current cell and its neighbour cells are

used to calculate an average slope of the surface line. Once the slope is known, the position of the

line inside the cell can be determined as the one where it divides the cell in a way, that the two vol-

umes exactly represent the filled volume and the void volume.

The important difference between the original surface reconstruction algorithm and the new one is

the fact that the new algorithm takes the volume of liquid, solid and porosities, i.e. the volume of

the cast part, as filled volume. This is necessary to describe liquid and solid surfaces using one and

the same algorithm. Of course this leads to errors in the description of the real surface position, but

these can be kept small by using a high grid resolution.

Following the well known procedure from /Torrey 85/ the pressure p ,j inside surface cells is ob-

i

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

new free surface boundary conditions. Cell

(1) is not a free surface cell because it does

not contain mobile phases. Cell (2) is not a

free surface cell either even if it has a neigh-

bouring cell which does not contain mobile

phases. The rule is: To become a free sur-

face cell, the cell needs direct contact to

void area.

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

The velocity boundary condition follows the technique described by /Torrey 85/. They apply the

condition of vanishing stresses for both tangential and normal flow. Plans to implement a Maran-

goni boundary condition at the free surface are in progress.

So far only adiabatic boundary conditions have been applied at the free surface for temperature and

concentration equations. Plans to implement radiation, air convection, vaporization and laser heat

input for the energy equation and species losses due to vaporization (Abbrand) for the concentra-

tion equation are in progress.

An algorithm of crucial importance for the performance of the highly coupled models has been

adopted from /Torrey 85/. Even when the coupled solution of pressure and velocities has converged

there is always a mass and volume conservation violation remaining. The amount of this volume

source or sink depends on the convergence criterion applied.

Each volume source or sink disturbs the free surface algorithm inside the fluid domain. For com-

pletely filled cells (F = 1) without any empty neighbours a completely conservative velocity field

should lead to no change in the filling function. A volume source or sink would cause deviations.

To prevent small numerical errors from piling up, after each time step all cells which have filling

functions of Fmove > 1 - VOF , where VOF is a small number (e.g. 110 ), are set back to 1. Of

-6

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.

In order to understand how the modified VOF equation works, it is useful to first review some of

the definitions. The quantity which requires the most detailed interpretation is the new variable for

the filling state, (Fmove )P . It has been defined in analogy to the definition of the filling state F in the

original VOF algorithm with partial cell treatment, where the fraction of fluid is linked to the filling

state by the equation l = F (1 obst ) = F open . This can be interpreted as F being the height fraction

of the area open to the flow filled with liquid under influence of gravity for the case that the obsta-

cle forms a vertical column with height fraction 1. This is a logical definition for the case of the

blocked out regions being part of a vertical wall. It becomes difficult to understand in the case of

horizontal or diagonal walls. These cases show that the definition is just a means of quantification

and does not mirror the real physical situation in the cell.

In the new model the liquid phase l is replaced by the mixed mobile phases move and the blocked

out regions may contain stationary solid. The nature of this solid as a volume averaged quantity

allows to interpret the filling heigth in a new way. If the solid is not considered to be piled up as a

massive vertical column but to be homogeneously distributed throughout the cell as a porous me-

dium, the quantity (Fmove )P naturally becomes the height fraction of the fluid settling inside the

solid network.

All trials to find a physical interpretation come to an end if solidification is considered and the

blocked out fraction of the cell changes. In this case the newly formed solid is assumed to be added

homogeneously to the vertical solid column, Fig. 4.11. This does not make much physical sense,

because why should solid form at places which are not in contact with the liquid? The only way to

deal with the problem is to accept the definition as a means of quantification which helps to de-

scribe the simultaneous effects of solidification and free surface flow. The advantage of the model

is the complete mass conservation and it can be used for general situations and solid / surface con-

figurations, even if it is dos not mirror the exact physical mechanisms.

4.6 Algorithm for solidification of moving free surfaces 139

states (Fmove )0P and (Fmove )P in the case where

solidification is present (cell height = 1). The

solid is always considered to be vertically piled

up in a column. The filling states are always

defined by the level of the liquid which fills the

space open to flow under influence of gravity.

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

(eff move )P which accounts for the shrinkage due to the temperature

dependent density of the liquid phase and phase transitions from liquid into mobile solid phases.

In general (for solidification processes) the new density is higher than the old one. Therefore the

factor becomes smaller than one and the volume of mobile phases decreases.

The second term describes the influence of the newly formed stationary solid. It becomes easier

to understand if we introduce the assumption ( eff close )P = const . , which is the case for constant

[ 0

( closed )P ( closed )P ] (( )

)

eff closed P

and

eff move P

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

2. The newly formed solid is added as a (virtual) layer to the vertical solid block at the left wall,

Fig. 4.12 b). By that it reduces the area open to flow into which the remaining liquid can settle.

As the density of the solid is in general higher than the density of the liquid, the liquid loses

142 Chapter 4 Algorithms

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.

There is not much to say about the implementation of the Split Solid Model as most features like the

solid splitting and the velocity assumptions have been integrated in the new set of differential equa-

tions. There are only a few remarks remaining:

So far the Columnar Growth Model has been implemented in a simplified form. For the deci-

sion whether to apply the Split Solid Model in a cell or not, the current model does not check a

radius around the cell centre, but the neighbour cells. This makes the current algorithm more

grid dependent than the full algorithm would be.

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.

The new porosity formation algorithm is capable of predicting the formation of hydrogen and

shrinkage porosities as well as internal shrinkage cavities.

At the beginning of each new time step the energy and liquid concentration equations are used to

calculate the new phase fractions of the alloy. This procedure yields the following results which can

be used for the prediction of hydrogen porosities:

The rate s / t of newly formed solid

The temperature Tnew to be expected at the end of the time step

The pressure is taken from the last time step.

The amount of mobile phases present is a limiting factor for the formation of hydrogen porosities.

Time integration of the hydrogen porosity formation rate pore hydro t yields

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

The shrinkage porosity calculation has been implemented into the iterative solver. At the beginning

of each time step the critical pressure for shrinkage pore formation

pcrit pore shrink = pvapour psurface tension

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.

In a volume averaging model it is not trivial to define a criterion to detemine whether microporosi-

ties or macroscopic shrinkage cavities form. As the shrinkage defects in general form in cells where

there is already solid present, it must be decided whether the solid has a planar front or forms a den-

dritic network. Observation of the immediate neighbourhood can help. There are different possibili-

ties:

4.8 Porosity formation algorithm 147

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.

An important matter is the accurate definition of the phase fractions on cell faces. The control vol-

ume method with staggered velocity grids needs to calculate many complex quantities on the cell

faces between two cells of the grid. There are two ways to determine these quantities:

To calculate the complex quantities for the cells of the main grid and to interpolate them there-

after.

To calculate interpolated values of all basic quantities like densities, phase fractions, conduc-

tivities etc., to store them in lookup-tables and then to calculate more complex quantities from

these staggered values.

The present model uses the latter procedure. It has mainly two advantages. On the one hand the

amount of calculations necessary is greatly reduced, on the other hand the danger of inconsistent

interpolations is reduced.

The interpolation of the phase fractions is especially difficult because two different types of phase

fractions are used:

localised phase fractions ( obst , void )

volume averaged phase fractions ( l , move , move , dead , dead , pore shrink , pore hydro )

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

cell faces for localised phase fractions.

The area fractions are calculated from

those localised phases which touch the

particular cell face. The area of vofm is

the cast part containing the averaged

phase fractions including alloy and po-

rosities.

vofm + void + obst = 1

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.

alloy + pore shrink + pore hydro = vofm

It is used for all thermodynamic calculations like phase transformations which only take into ac-

count the alloy phases.

In the last section we have seen how localised and volume averaged phase fractions on cell faces of

each single cell can be calculated. The next difficulties arise when the values of a phase fraction on

a cell face is different for the two neigbouring cells. This can happen in case of vofm which is a re-

sult of the VOF advection and surface reconstruction algorithm. The free surface is not a closed line

150 Chapter 5 Implementation Details

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.

The volume fractions and area fractions of the volume averaged quantities are now projected onto

the volumes and cell face areas of the localised volume fraction vofm. As for the volume fractions

the relation

vofm = l + move + move + dead + dead + pore hydro + pore shrink

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 *

and with p = l , move, move, dead, dead, pore hydro, pore shrink

The treatment of the effective densities of the mobile phases which are used in the convective terms

of the transport equations follows a concept which is very often used for the treatment of the con-

vection-diffusion term. As the staggered velocities are defined on the cell faces of the normal grid,

but the densities in the cell centres, the density values used for the convective terms must be inter-

polated to the cell faces. Instead of taking the average density of the two adjacent cells of the cell

face, an upwind model is used, i.e. always the density of the upstream cell is used. This is reason-

able, because all mass flowing through the cell face should have the density of the cell where the

flow comes from.

To ensure numerical stability of the calculations an adaptive time step control has been imple-

mented. The algorithm checks the stability criterion, that the flow must not cross more than one half

of a cell per time step. Therefore the time step width is reduced and the time step redone if the ve-

locities in at least one cell become too high.

The velocity used for the evaluation of the criterion is the superficial velocity. Especially in cases

with shrinkage flow the pore velocities sometimes become very high and would stall the calculation

if they were used for the criterion. Therefore the adapted time step width for an equidistant 1D-grid

becomes

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).

during a welding process with constant posi-

tive MC; the reduction of the time step width

is a result of the adaptive time step control;

when the velocities become so high that they

cross more than one half of a cell per time

step, the time step width is reduced; when the

HS is switched off, the velocities decrease

quickly and the time step width is reraised

until it reaches the maximum time step width

allowed by the user.

Convergence criteria work on two levels: The solver level and the SIMPLER level.

On the solver level convergence of the implicit discretised differential equations (u, v, p, pc, t, Cl,

C , C for EL2D and p only for SoliCon, which mainly uses explicit differential equations) is

checked by a normalised residuum criterion. Convergence of the equation for the variable is

reached if

154 Chapter 5 Implementation Details

nb a nb nb + b a P P

cells

<R (5.2)

a P P

cells

The SIMPLER loop is considered to be converged if the maximum change of the variable f of all

cells from SIMPLER iterations n1 to n becomes smaller than the maximum allowed change

SIMPLER :

max P n 1 < SIMPLER

cells P n

(5.3)

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

ware Bild2D. Normal velocity plots only show velocities

which have been interpolated to the centres of the real cells.

This module can show the staggered velocities and their posi-

tions on the cell faces. This is important for debugging pur-

poses, especially for the development of the free surface algo-

rithm, where it is crucial to know which velocities are defined

at exactly which position.

In addition, a shrinkage problem needs much more time for the pressure / velocity coupling than a

welding problem.

An image processing software (Bild2D) has been developed which has been used to create most of

the simulation pictures in this work. A reasonable question is, why creating a new software while

very sophisticated programs are available on the market? The main reasons for the development of

the software are:

Many image processing softwares can make one picture at a time. Bild2D can make pictures of

a series of time steps, and each of these pictures can themselves be composed from subimages

showing different physical quantities.

The program can be adapted to specific needs like representation of blocked off regions, free

surface lines, etc. and these features can be combined to find specific solutions. An important

example is the module which allows to draw the single components of the staggered velocity

fields at the right positions into a representation of the grid, Fig. 5.6.

Some programs can do all these things, but most of the times it takes a long time to find out how

and very often requires a high effort from the user.

5.5 Image processing software 157

Bild2D can make scalar plots with or without contour lines and with or without interpolation,

vector plots and streamlines, plot free surface lines, blocked out regions, trace marker particles

etc. A special module creates an RGB representation of three phase fractions in one picture. By

this module areas with eutectic or peritectic transformations can be visualized.

A powerful set of modules is available which allow to create user defined colour scales with

different properties. Modules are available which create split scales for positive and negative

values, or which make weighted scales adjusting the areas plotted in each colour to be approxi-

mately equal. Of course the scales can be exported and imported.

A toolbox contains tools for extracting the time development of values in certain cells (e.g. to

create cooling curves), plotting the mesh (cf. Fig. 7.3), plotting colour scales and drawing cool-

ing curves written by SoliCon / EL2D.

The area where nonzero velocities are present can be magnified automatically to fit the picture

size. This is important in cases where the molten area is comparably small, e.g. during welding

processes.

Bild2D can be controlled via control files and prepared scales, i.e. even complex pictures can be

made quickly.

Bild2D is easy and quick to handle, comfortable and (almost) self-explaining (End of the com-

mercial part...).

6 Applications: Simulations and experiments, Casting processes

6.1 Test problems

This chapter contains some benchmark problems, i.e. comparison of simulation results of the new

model with analytical solutions or well defined experimental results. As there are no benchmarks

for the complex problems which are subject of this work, the problems concentrate on testing the

principal function of single models.

The energy equation was tested against an analytical solution from /Baehr 98/. The test case is the

heating of a one-dimensional long (half-infinite) bar at a constant starting temperature T0. At t = 0

the end of the bar is put to a fix temperature TS. The development of the temperature field T(x,t) can

be calculated by analytically solving the one-dimensional partial differential equation

T 2 T

=a with a :=

t x2 cP

The analytical solution meeting the initial and boundary conditions is:

x

T = T0 + (TS T0 ) erfc (/Baehr 98/, p. 158)

2 at

2 w2

erfc( ) := 1 erf ( ) = 1

0 e dw

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 fluid flow calculation with free surfaces was tested against an analytical solution from /Torrey

85/. The test case is the sloshing motion of a tank of water. The calculation domain is an equidistant

20 by 13 grid of L = 0.1 m by H = 0.065 m. Initial condition is the fluid at rest with the surface

height function

h( x ) = h0 + h0 cos x

L

where

h0 = 0.05 m average fluid height

h0 = 0.005 m initial perturbation amplitude

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

l = 1000 kg / m3

eff move = 1.002 10-3 Pa s

6.1 Test problems 161

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.

In order to check the phase change algorithm the solidification of a test geometry, an ingot of 50

mm x 100 mm, Fig. 6.3, was simulated. The whole ingot was set to an initial temperature and T0

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.

and initial settings of ingot used to

test the solidification algorithm.

To test the algorithm for the eutectic phase diagram, two alloys of the system Al-Si, Fig. 9.1, with

silicon concentrations of 7 wt % (hypoeutectic) and 17 wt % (hypereutectic) were solidified. Fig.

6.4 shows the evolution of temperatures, phase fractions and concentrations. The curves exactly

reproduce the evolution to be expected from the analytical solution.

To test the algorithm for the peritectic phase diagram, two alloys of the system Fe-C, Fig. 9.2, with

carbon concentrations of 0.12 wt % (left part of peritectic plateau, < = 0.17 wt %) and 0.25

C C peri

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

Fig. 6.4: Test of phase change algorithm for two alloys from the eutectic Al-Si phase diagram; time

evolution of temperature, phase fractions and concentrations of the three phases liquid (l), alumin-

ium rich solid, (Al), and silicon rich solid, (Si), during solidification.

164 Chapter 6 Applications: Simulations and experiments, Casting processes

Fig. 6.5: Test of phase change algorithm for two alloys from the peritectic Fe-C phase diagram;

time evolution of temperature, phase fractions and concentrations of the three phases liquid (l), -

solid and -solid during solidification.

6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 165

To test the algorithm for a technically pure substance, an alloy of the system Al-Si, Fig. 9.1, with

silicon concentration of 0.01 wt %, i.e. 99.99 Al was solidified. Fig. 6.6 shows the evolution of

temperatures and phase fractions. The concentrations do not change much in this case. The curves

exactly reproduce the evolution to be expected from the analytical solution.

Fig. 6.6: Test of phase change algorithm for technically pure substance Al99.99 calculated using an

Al-Si phase diagram and an alloy with 0.01 wt % Si; time evolution of temperature, phase fractions

and concentrations of the two phases liquid (l) and (Al) during solidification.

Correct riser dimensioning in casting is a matter of great economical impact. Risers are responsible

for the soundness of the casting, but they should be kept as small as possible, because all material of

the riser which is not used for feeding is waste material. So far the main criterion for good riser de-

sign has been to make them as small as possible, but large enough that the external shrinkage cavity

which occurs due to feeding does not enter the proper cast part. Experimental and numerical work

by /Bhmer 97/, who have cast and analysed a steel ingot with riser, has demonstrated that this cri-

terion is not sufficient. Spark spectrum analysis of the final macrosegregation pattern shows se-

verely carbon enriched areas directly beneath the shrinkage cavity, which penetrate the proper cast

part. The enriched areas contain up to 130 % more carbon than the initial mixture. The result is a

cast iron area inside a steel part. Apart from the problem that this unwanted modification of local

mechanical properties can lead to failure of the component, the cast part can already be damaged

during the production process when the riser is cut off.

In this section the experiments and simulation results by /Bhmer 97/ will be briefly reviewed and

then the investigation continued by new simulations using the new integrated model which allows

166 Chapter 6 Applications: Simulations and experiments, Casting processes

the description of the influence of the macroscopic shrinkage cavity on the final macrosegregation

pattern.

All details of this study can be found in /Bhmer 95/.

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.

Fig. 6.8: Results from /Bhmer 97/; comparison of simulated carbon concentration deviation of

GS60 unalloyed steel using a) isotropic, b) anisotropic permeability model with c) experimental

results from spark spectrum analysis; (x) strong macrosegregations at the last point to solidify.

168 Chapter 6 Applications: Simulations and experiments, Casting processes

The present work continues the studies by /Bhmer 97/. In order to integrate the complex models

some features have been temporarily removed. The new model does not consider

multicomponent steel, but an Fe-0.42wt% binary alloy

backdiffusion or Scheil microsegregation model, but uses the lever-rule and an equilibrium

phase diagram

This is a reason, why so far the results are only qualitatively comparable. The following models

have been added:

solidification of the moving free surface,

SSM, which allows the formation of a deep shrinkage cavity and the modelling of sedimentation

effects,

CGM which, together with the SSM, governs the transition between columnar and equiaxed

growth

piecewise linear phase diagram instead of linear one.

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.

For most of the calculations the grid configuration, boundary conditions and model parameters

shown in Fig. 6.9 were used. The cast material was a binary Fe-0.42wtC alloy. The carbon content

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. If shrinkage flow is considered, temperature

dependent liquid density is used and constant and higher solid density. It is important that the solid

density is higher as the highest possible liquid density. Otherwise solidification would lead to ex-

pansion instead of shrinkage. The choice of the right solid density is difficult, because a 2D calcu-

lation is compared to a 3D problem. Even if the experimental set-up tries to create a quasi-2D situa-

tion by using isolating front plates (dark grey areas in Fig. 6.7 a) ), the experimentally observed

shrinkage cavity is not really 2D. In addition the constant solid density cannot correctly describe the

effects of solid state shrinkage. Trials with different constant solid densities between 7344 and 8344

kg/m3 were made. It turned out that for low densities the formation of the shrinkage cavity was in-

hibited by numerical problems due to the coarse grid (the feeding path was blocked even if there

6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 169

tions and parameters of SSM and CGM for steel

ingot with macroscopic shrinkage cavity.

SSM: 0 mv dd = 0.3

mv dd = 0.2

CGM: RCGM = 0.01 m

CGM = 0.6

for the split solid

model: This calcula-

tion crashed after

880 seconds due to a

pressure failure,

when the free sur-

face model tried to

empty a cell which

already contained

small amounts of

stationary solid; a)

solid fraction; light-

est grey: 0 < s 0.1;

b) velocities; c) pres-

sure; arrow: Position

a) b) vmax = 0.0018 m/s c) of pressure failure.

170 Chapter 6 Applications: Simulations and experiments, Casting processes

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.

A calculation with solidification of free surface, shrinkage flow and thermosolutal convection, but

without SSM shows the importance of this model for the simulation of a deep shrinkage cavity. The

calculation crashed at t = 880 s. The problems arise when the free surface model is about to empty

a cell or cell row which already contains some stationary solid. Without SSM all solid which forms

is assumed to be stationary and therefore stays in the place where it has formed. As Fe-0.42wt%C is

an alloy with a long solidification range, small amounts of solid form all over the cast part, and also

close to the free surface, already in early stages of the solidification process, Fig. 6.10 a). Shrinkage

flow which is in this case superposed by thermosolutal convection, 6.10 b) lowers the liquid level

and causes the free surface algorithm to successively empty the top layer of cells. When the free

surface has removed all liquid from the topmost cell row it starts to empty the next row, even if the

row above is not empty but still contains solid. The free surface is deadlocked in the topmost cell

layer while the free surface algorithm tries to create a new surface in the next row. This leads to

contradictions which cause the pressure iteration to fail, 6.10 c).

The SSM alone, which keeps the solidifying free surface mobile while the solid fraction is low, is

not sufficient to model the formation of deep shrinkage cavities. Fig. 6.11 shows a calculation with

SSM and thermosolutal convection, but without CGM. All solid forming close to the walls is con-

sidered to be mobile and sinks down with the fluid. It is clearly visible that the flow along the walls

speeds up when the first solid phase, which has a higher density than the liquid phase, occurs. A

sedimentation zone forms at the bottom. In the bottom corners the total solid fraction exceeds the

limit for the transition to stationary solid first. In this region the first stronger macrosegregations

form because the now stationary mushy zone is washed out by flow going through. The free surface

always remains flat, because it never contains stationary solid. With proceeding solidification the

liquid level falls homogeneously. To form a deep shrinkage cavity some of the solid must stick to

the walls. To solve these problems the Columnar Growth Model (CGM) has been developed. It will

be used in the next section.

In the present case without CGM the SSM gives unrealistic results when solidification continues.

Solid fraction fluctuations which locally exceed the limit lead to the formation of stationary islands

which are not connected to any wall, Fig. 6.11, t = 700 s. These islands constitute flow obstacles

and cause numerical stability problems.

6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 171

vmax = 0.018 m/s vmax = 0.020 m/s vmax = 0.022 m/s vmax = 0.025 m/s vmax = 0.027 m/s

vmax = 0.026 m/s vmax = 0.025 m/s vmax = 0.024 m/s vmax = 0.053 m/s

Fig. 6.11: Simulation of steel ingot with SSM, but without CGM; velocity vectors and development

of the solid fraction.

A simulation with SSM, CGM and shrinkage flow, but without thermosolutal convection, demon-

strates the importance of thermosolutal convection for the formation of the deep shrinkage cavity.

Fig. 6.12, left half of the pictures, shows the results of the calculation which starts to loose mass due

to non-convergence at about t = 1200 s because the surface is deadlocked after the feeding path has

172 Chapter 6 Applications: Simulations and experiments, Casting processes

vmax = 0.00001 m/s 0.0152 m/s vmax = 0.00002 m/s 0.0101 m/s vmax = 0.00139 m/s 0.0120 m/s

solid fraction []

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

perature distributions after

1200 s for case with shrinkage

flow only (left) and shrinkage

flow with thermosolutal con-

vection (right)

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 1, continued on following pages)

176 Chapter 6 Applications: Simulations and experiments, Casting processes

Fig. 6.14 (part 2, continued on following page)

6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 177

deviation [%]

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 [%]

Fig. 6.15: Shrinkage cavity formation of Fe-0.42wt%C ingot; comparison of final macrosegregation

patterns of a) the new results of the simulation with SSM, CGM, shrinkage and thermosolutal con-

vection with b) the experimental results by /Bhmer 97/.

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.

Of course the grid resolution of 10x23 cells is too small to give reliable results. This is why com-

parative calculations were made with grid resolutions 20x46 and 40x92 cells. It turned out that the

numerical problems observed during the calculations with grid resolution 10x23 cells grew worse.

The 20x46 calculation finally crashed after after 1630 s, the 40x96 after 550 s. The exact reasons

have not yet been determined. However, some observations can still be made. Fig. 6.16 compares

the simulations results after 540 s for all three grid resolutions, Fig 6.17 the results after 1600 s for

grid resolutions 10x23 and 20x46 cells.

The high resolution calculations give a much better representation of the surface shape, but only as

far as the surface resolution is concerned. The general shape is similar to the case with the coarse

grid 10x23. The solid fractions after 540 s are qualitatively similar, but the coarse calculation has

built more stationary solid, while the calculations with higher resolution show more mobile solid.

This can be explained by the higher absolute velocities which remove the forming solid from the

front, before it can become stationary. After 1600 s the most obvious difference in the high resolu-

tion calculation is the appearance of many segregation channels, solutally remolten by enriched

melt drawn into the mushy zone by shrinkage. They are so small that they cannot be resolved by the

10x23 calculation. Even in the 20x46 cells calculation they are only one cell wide.

The velocity fields after 540 s compare qualitatively well. Even if the resolution is very different,

the main flow patterns which are driven by thermal (downward flows along the walls and mushy

zone) and solutal gradients (upward flow at the internal edge), are found in all three resolutions. The

absolute velocities rise with higher resolution. After 1600 s the velocities differ considerably, which

is normal, considering the higly complex and quickly changing flow patterns due to the sedimenta-

tion of the mobile solid.

The macrosegregation patterns after 540 s differ in details. The level inside the liquid phase is

sligtly higher for higher resolution. All calculations show plumes of enriched melt washed out of

the mushy zone by the downward flow. After 1600 s the positive segregation channels which have

already been described, are the main difference. Fig. 6.18 shows the segregation channels together

with the colour coded velocity field. Now the shrinkage flow which is responsible for the channel

formation becomes visible. The velocities are in the order of magnitude of about 510-6 m/s, corre-

sponding to 1 mm in 200 s. It becomes obvious, that the small enriched area in the riser is due to the

same effect.

180 Chapter 6 Applications: Simulations and experiments, Casting processes

deviation [%]

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

The results of this section are preliminary. Some important models have just been implemented and

are not yet completely debugged. However, the results show that the complex interaction of the

models works. The following errors and problems have been observed during the production runs:

The program tended to crash when one horizontal cell row was about to be emptied. This is a

consequence of the complexity of the models working together inside the free surface cells.

6.2 Formation of deep shrinkage cavity in riser of Fe-0.42wt%C ingot 181

solid fraction [] carbon concentration Fig. 6.17: Shrinkage cavity formation of Fe-

deviation [%] 0.42wt%C ingot; comparison of solid fractions

(left half), velocity fields and macrosegregations

(right half) after 1600 s for grid resolutions a)

10x23 cells, b) 20x46 cells.

deviation [%] mation of Fe-0.42wt%C ingot;

macrosegregation pattern and

colour coded velocity field after

1600 s for grid resolution 20x46

cells. The dark arrows inside the

mushy zone and the segregation

channels have velocities in the

order of magnitude of 510-6 m/s.

182 Chapter 6 Applications: Simulations and experiments, Casting processes

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.

liquid density (const) l = 2400 kg/m3 moving free surfaces

3 shrinkage induced flow

solid density (const) s = 2600 kg/m

mushy zone flow

convective solute transport

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.

To test the porosity formation model which is able to simultaneously but separately predict the for-

mation of hydrogen and shrinkage porosities, a 2D numerical setup was designed which provokes

the formation of both hydrogen and shrinkage porosities. Fig. 6.21 shows the configuration, the

boundary conditions and some important parameters used for all calculations. A tall ingot of Al-

7wt%Si (physical data from section 9.1.3.1) is cooled uniformly from the bottom to initiate upward

directional solidification and a shrinkage flow towards the bottom. At the same time the ingot is

cooled from the sides by two chills in order to slowly close the feeding path by creating a mushy

zone with increasing solid fraction and to interrupt the shrinkage flow. Expected results are a more

or less uniform distribution of hydrogen porosities all over the cast part (if the initial hydrogen

contents exceeds the maximum contents in the solid), and an area of shrinkage porosities beneath

the point where the solidification from the sides closes the feeding path. The calculations were done

without SSM. Due to numerical problems of the first implementation of the algorithm, the results of

the calculation with shrinkage porosity formation only have been made with zero viscosity.

A test calculation was made to check and demonstrate some basic properties of the shrinkage po-

rosity formation algorithm. As viscosity was set to zero porosity formation started just after the

feeding path had completely closed. Several interesting observations were made:

The porosity formation starts beneath the point where feeding is blocked and then develops

downwards.

186 Chapter 6 Applications: Simulations and experiments, Casting processes

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.

tions and some important parameters used for all

for porosity calculations.

tents in the liquid: VH = 0.3 cm3 / 100 g

Maximum hydrogen

contents in the solid: VH max = 0.1 cm3 / 100 g

Critical pressure for

shrinkage pore for-

mation: pcrit pore shrink = 0.1 p0

= 1.013104 Pa

6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys 187

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.

To validate the model an experimental set-up was designed which is shown in Fig. 6.24. Due to

practical reasons (last minute experiment...) the geometry differs from the numerical set-up. The

side chills have been fixed 95 mm from the bottom instead of 50 mm. To yield 2D conditions in the

middle cross section, the third dimension of the cast part was chosen to be 300 mm. The industrial

alloy Al-7wt%Si-0.3wt%Mg with low hydrogen contents of less than 0.1 cm3 / 100g at a casting

temperature of T0 = 700 C [973 K] was cast into the sand mould provided with cast iron chills.

shrinkage porosities.

6.4 Simultaneous prediction of hydrogen and shrinkage porosities in Al-Si alloys 189

cation of shrinkage porosity

formation in absence of hydro-

gen; a) complete ingot showing

the positions of the three cast

iron chills; b) horizontal clip-

b)

ping, showing the main porosity

area (height: 8 mm); c) large

magnification of the main po-

rosity area; the horizontal size of

the ingot is 48 mm.

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.

Two test calculations were made to check and demonstrate some basic properties of the hydrogen

porosity formation algorithm.

The first calculation, without shrinkage flow, shows surprising but understandable effects: When

solidification starts, an expansion flow is initiated which drives the liquid away from the solid, Fig.

6.26 a). This is physically correct and a consequence of the formation mechanism of hydrogen

porosities. The hydrogen which is rejected from the solidifying alloy becomes gaseous and requires

190 Chapter 6 Applications: Simulations and experiments, Casting processes

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.

The coupled calculations with both models switched on gave fascinating results. Three calculations

were made with different permeability conditions of the mushy zone:

1. zero viscosity, i.e. infinite permeability of the mushy zone.

2. small viscosity = 1.12510-3 kg m-1 s-1 (1/10 of realistic viscosity for Al-Si) and isotropic per-

meability with permeability constant K0 = 810-10 m2.

3. realistic viscosity = 1.12510-2 kg m-1 s-1 of Al-Si and and isotropic permeability with small

permeability constant K0 = 810-10 m2.

Fig. 6.27 shows the porosity distributions for case 1 after complete solidification of the bottom part

of the casting14 separately for hydrogen and shrinkage porosities, and both combined.

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

area is 6.9 mm2 instead of 22,8 mm2. This is due to the fact that the hydrogen pores are additional

volume compensating for the volume losses caused by shrinkage.

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.

The set-up used for the calculations, Fig. 7.1, is a stationary welding process on an infinite 6 mm

steel plate. A cylindrical section of this plate with a radius of 9 mm is simulated. The plate is melted

by a stationary heat source with vertical incidence and Gauss-shaped intensity profile. The calcula-

tion domain is an equidistant fixed grid with mesh resolutions between 60x40 and 120x80 square

cells (constant axial and radial grid spacing between 0.15 mm and 0.075 mm) with the cylinder axis

on the right boundary. If not otherwise mentioned the material parameters are those of the binary

Fe-C system, section 9.1.3.2. Table 7.1 shows additional parameters and physical constants used for

all welding calculations in this chapter if not otherwise mentioned. To investigate the influence of

constant surface parameters instead of temperature dependent ones a sensitivity analysis for the

value of emissivity b has been performed. Calculations with b = 0.2 and b = 0.7 did not show im-

portant changes in temperature distribution or pool shape.

Initially the domain is set to T = 300 K. To simulate infinite dimension at the left boundary, the

temperature gradients are continued, multiplied by a relaxation factor 0.8. This is an empirically

determined factor which ensures that the curvature of the isotherms does not change at the

7.1 Systematical investigation of pool shape formation in welding 197

welding calculations; boundary and

initial conditions.

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.

The task to study the influence of welding parameters on the resulting weld pool shape is difficult

due to the multidimensional parameter space and the long calculation times for the simulations. In

this first study the parameter space is limited to the case of constant MC. The following strategy

was chosen to tackle the problem.

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

temperature plots in this section. The

grey area shows the mushy zone for

the initial composition. The liquidus

and solidus temperatures are always

given in the caption of the corre-

sponding figure. In this case, Fe-

0.1wt%C, Tliq = 1801 K, Tsol = 1766

K; the other isolines are identical for

all compositions.

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.

Several numerical parameters were checked to determine their influence on the final pool shape and

to find suitable settings for the studies of the welding parameters.

The correct grid spacing for the simulation of Marangoni convection is subject to ongoing scientific

discussions. As Marangoni flow is initiated in a thin surface layer, it can be expected that a correct

7.1 Systematical investigation of pool shape formation in welding 199

a)

b)

c)

Fig. 7.3: Influence of grid resolution on the pool shape during laser welding of Fe-0.1wt%C with P

= 2000 W. Temperature distributions after 1.5 s for different constant MC a) +10-4 N m-1K-1, b) 0,

c) 10-4 N m-1K-1. Scale cf. Fig. 7.2; Tliq = 1801 K, Tsol = 1766 K; d) Clippings of the corresponding

grids.

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.

To cope with the high nonlinearity of the problem, the SIMPLER algorithm, section 4.4, repeats the

solving of all differential equations until the results of the single equations do not influence each

other any more. Convergence is reached when the maximum change of any temperature, velocity or

concentration value since the last SIMPLER iteration becomes smaller than a limit. If the maximum

number of SIMPLER iterations is limited, the coupling between the differential equations is eased,

SIMPLER convergence is not always reached. Of course the single differential equations still con-

verge, the mass, energy and momentum balances are still fulfilled. As the time steps in this problem

are very small, there is some evidence, that the influence of a limited number of SIMPLER itera-

tions may not have too much impact and can be used to reduce the calculation times. Comparative

calculations of a laser welding process of Fe-0.1wt%C with P = 2000 W, rlaser = 4 mm, positive MC

and SIMPLER iteration limits of 200 and 10 were made. The pool shapes after 1.5 s and the solidi-

fication times are practically identical, and the maximum velocities differ less than 1 %. The im-

portant difference: The overall duration of the calculation reduces from 13 days for 200 iterations to

21 hours for 10 iterations.

The adaptive time step control, section 5.2, guarantees that the time step width is small enough that

the maximum velocity does not cross more than one half of a cell during one time step. This is a

stability criterion and should prevent unphysical results. Due to the high Marangoni velocities the

time step width which is initially set to t = 0.001 s, is reduced and becomes t = 0.00005 s. Test

calculations were made, if the adaptive time step control is really necessary and how the choice of

an equidistant time step width of t = 0.001 s influences the pool shape. Fig. 7.4 shows an example

of a laser weld pool with positive MC after t = 1.4 s, once calculated using equidistant time steps,

and once using an adaptive time step control. The second pool is deeper. Using too large time steps

seems to cut down the convective heat transport. The qualitative pool shape has not been changed.

7.1 Systematical investigation of pool shape formation in welding 201

time step control; temperature

distributions after 1.5 s for

welding of Fe-0.1wt%C; scale cf.

Fig. 7.2; Tliq = 1801 K, Tsol =

1766 K; the pool calculated using

the adaptive time step control is

deeper than the one calculated

using equidistant time steps with

t = 0.001 s.

equidistant adaptive

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.

A series of welding simulations of Fe-0.1wt%C with different constant MC values 0, 110-5,

210-5, 510-5, 110-4, 1.510-4, 210-4 and 410-4 N m-1K-1 were made. Fig. 7.6 shows the

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

composition on the pool shape

during laser welding with P =

2000 W. Temperature distribu-

tions after 1.5 s for different con-

stant MC a) +10-4 N m-1K-1, b) 0,

c) a) 10-4 N m-1K-1; grey area:

Solid/liquid interval of initial

composition, determined from

the binary Fe-C equilibrium

phase diagram. Isotherms inside

a) Fe-0.1wt%C Fe-0.8wt%C and outside the weld pool have

the same values as in Fig. 7.2. d)

Solid/liquid intervals of the used

alloys.

wt % C Tliquidus [K] Tsolidus [K]

0.1 1801.04 1766.17

0.2 1792.93 1760.67

b) Fe-0.1wt%C Fe-0.8wt%C

0.3 1784.82 1742.37

0.4 1776.71 1724.06

0.5 1768.60 1705.75

c) Fe-0.1wt%C Fe-0.7wt%C 0.6 1761.99 1687.44

0.7 1756.03 1669.14

d) 0.8 1750.07 1650.83

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

Fig. 7.6: 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. 7.2; Tliq = 1801 K, Tsol = 1766 K.

For positive MC the surface flow drives hot melt towards the centre of the pool, where it is redi-

rected downward and digs a deep pool. Therefore with increasing positive MC the pool depth in-

creases as a function of the increasing surface velocity while the pool radius remains almost con-

stant and equal to the radius of the calculation without Marangoni convection. This radius as well as

the characteristic edge of the deep V-shape pools seem to be a result of heat conduction only

(Thermosolutal convection does not play an important role).

The increase of the total molten volume after 1.5 s for any kind of Marangoni convection compared

to the case with MC = 0 can, in the case of negative MC, be explained by the more homogenous

heat distribution along the surface due to convection which decreases the maximum surface

7.1 Systematical investigation of pool shape formation in welding 205

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.

A series of welding simulations of Fe-0.1wt%C with MC = 110-4 N m-1K-1 and 0 and different

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

experimental observations. An explanation for the 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 have been tried in order to understand the effect and to find the best descrip-

tion:

206 Chapter 7 Applications: Simulations and experiments, Welding processes

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.

Fig. 7.9: Influence of laser radius on pool shape

for welding of Fe-0.1wt%C with P = 2000 W

and positive MC +110-4 N m-1K-1; scale cf. Fig.

7.2; Tliq = 1801 K, Tsol = 1766 K; the pool shapes

marked with (V) are not realistic, because for

them the formation of a vapour capillary must be

expected.

rlaser = 5 mm rlaser = 7 mm

7.1 Systematical investigation of pool shape formation in welding 207

Fig. 7.10: Influence of laser radius on pool shape

for welding of Fe-0.1wt%C with P = 2000 W

and no Marangoni convection; scale cf. Fig. 7.2;

Tliq = 1801 K, Tsol = 1766 K; the pool shapes

rlaser = 5 mm rlaser = 7 mm marked with (V) are not realistic, because for

them the formation of a vapour capillary must be

expected.

Fig. 7.11: Influence of laser radius on pool shape

for welding of Fe-0.1wt%C with P = 2000 W

and negative MC 110-4 N m-1K-1; scale cf. Fig.

7.2; Tliq = 1801 K, Tsol = 1766 K; the pool shapes

rlaser = 5 mm rlaser = 7 mm marked with (V) are not realistic, because for

them the formation of a vapour capillary must be

expected.

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, Weld