Entdecken Sie eBooks

Kategorien

Entdecken Sie Hörbücher

Kategorien

Entdecken Sie Zeitschriften

Kategorien

Entdecken Sie Dokumente

Kategorien

0 Bewertungen0% fanden dieses Dokument nützlich (0 Abstimmungen)

122 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)

122 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

multicom