physical metallurgy of direct chill casting of aluminum alloys 9781420062816, 1420062816

Pulling together information previously scattered throughout numerous research articles into one detailed resource, Phys

316 104 6MB

English Pages 290 Year 2008

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

physical metallurgy of direct chill casting of aluminum alloys
 9781420062816, 1420062816

  • Commentary
  • Concluding Remarks missing
  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Copyright 2008 by Taylor and Francis Group, LLC

Advances in Metallic Alloys A series edited by J.N. Fridlyander, All-Russian Institute of Aviation Materials, Moscow, Russia and D.G. Eskin, Netherlands Institute for Metals Research, Delft, The Netherlands Volume 1 Liquid Metal Processing: Applications to Aluminum Alloy Production I.G. Brodova, P.S. Popel and G.I. Eskin Volume 2 Iron in Aluminum Alloys: Impurity and Alloying Elment N.W. Belov, A.A. Aksenov and D.G. Eskin Volume 3 Magnesium Alloy Containing Rare Earth Metals: Structure and Properties L.L. Rokhlin Volume 4 Phase Transformations of Elements Under High Pressure E.Yu Tonkov and E.G. Ponyatovsky Volume 5 Titanium Alloys: Russian Aircraft and Aerospace Applications Valentin N. Moiseyev Volume 6 Physical Metallurgy of Direct Chill Casting of Aluminum Alloys Dmitry G. Eskin

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys DMITRY G. ESKIN

Boca Raton London New York

CRC Press is an imprint of the Taylor & Francis Group, an informa business

Copyright 2008 by Taylor and Francis Group, LLC

CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2008 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group, an Informa business No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-13: 978-1-4200-6281-6 (Hardcover) This book contains information obtained from authentic and highly regarded sources Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The Authors and Publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www. copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com

Copyright 2008 by Taylor and Francis Group, LLC

To my mama and papa

Copyright 2008 by Taylor and Francis Group, LLC

Contents List of Symbols and Abbreviations ............................................................xi Preface .........................................................................................................xv Author ........................................................................................................xix 1

Direct Chill Casting: Development of the Technology ................. 1 References ..................................................................................................... 17

2

Solidification of Aluminum Alloys ............................................... 19 2.1 Effect of Cooling Rate and Melt Temperature on Solidiication of Aluminum Alloys ............................................ 19 2.2 Microsegregation in Aluminum Alloys .......................................... 28 2.3 Solidiication Reactions and Phase Composition...........................43 2.3.1 Commercial Aluminum ........................................................43 2.3.2 Wrought Alloys with Manganese (3XXX Series)................ 46 2.3.3 Al–Mg–Si Wrought Alloys (6XXX Series)............................ 47 2.3.4 Al–Mg–Si–Cu Wrought Alloys (6XXX and 2XXX Series) ........................................................ 50 2.3.5 Al–Mg–Mn Wrought Alloys (5XXX Series)......................... 51 2.3.6 Al–Cu–Mn (Mg, Si, Ni) Wrought Alloys (2XXX Series) ...........................................................................54 2.3.7 Al–Mg–Zn–(Cu) Wrought Alloys (7XXX Series)............... 55 2.3.8 Wrought Alloys Containing Lithium .................................. 56 2.3.8.1 Al–Cu–Li Commercial Alloys ............................... 56 2.3.8.2 Al–Li–Mg Commercial Alloys .............................. 57 2.3.8.3 Al–Li–Cu–Mg Commercial Alloys ...................... 57 2.4 Effect of Alloy Composition on Structure Formation: Grain Reinement ................................................................................ 59 2.4.1 Mechanisms of Grain Reinement........................................ 59 2.4.2 Al–Sc–Zr Phase Diagram ...................................................... 68 2.4.3 Al–Ti–B Phase Diagram ........................................................ 70 2.4.4 Al–Ti–C Phase Diagram ....................................................... 71 References ..................................................................................................... 75

3

Solidification Patterns and Structure Formation during Direct Chill Casting ............................................................ 79 3.1 Shape and Dimensions of the Billet Sump...................................... 79 3.2 Solidiication Rate and Cooling Rate during DC Casting ............................................................................. 91 3.3 Effects of Process Parameters on the Formation of Grain Structure ............................................................................. 101 vii

Copyright 2008 by Taylor and Francis Group, LLC

Contents

viii 3.4

Effect of Process Parameters on the Amount of Nonequilibrium Eutectics ............................................................... 111 3.5 Effect of Process Parameters and Alloy Composition on the Occurrence of Some Casting Defects ................................ 115 References ................................................................................................... 122 4

Macrosegregation ........................................................................... 125 4.1 Mechanisms of Macrosegregation ................................................. 125 4.1.1 Historic Overview ................................................................ 125 4.1.2 Permeability........................................................................... 132 4.1.3 Convection-Driven Macrosegregation .............................. 136 4.1.4 Shrinkage-Driven Macrosegregation ................................ 141 4.1.5 Floating Grains and Macrosegregation ............................. 145 4.1.6 Deformation-Driven Macrosegregation ............................ 151 4.2 Effects of Process Parameters on Macrosegregation during DC Casting ........................................................................... 152 4.2.1 Effect of Casting Speed on Macrosegregation during DC Casting ............................................................... 152 4.2.2 Effect of Melt Temperature on Macrosegregation during DC Casting ............................................................... 159 4.2.3 Dimensions of the Billet: Scaling Rule of Macrosegregation ................................................................. 159 4.2.4 Effect of Forced Convection on Macrosegregation during DC Casting ............................................................... 169 4.3 Effect of Composition on Macrosegregation: Macrosegregation in Commercial Aluminum Alloys ................ 170 References ................................................................................................... 179

5

Hot Tearing ...................................................................................... 183 5.1 Thermal Contraction during Solidiication .................................. 185 5.2 Mechanical Properties of Semi-Solid Alloys ................................200 5.2.1 Testing Techniques ...............................................................200 5.2.2 Strength Properties of Aluminum Alloys in the Semi-Solid State ......................................................... 205 5.2.3 Ductility of Aluminum Alloys in the Semi-Solid State ......................................................... 208 5.3 Mechanisms and Criteria for Hot Tearing .................................... 216 5.3.1 Historic Overview of Hot Tearing Research and Suggested Mechanisms................................................ 216 5.3.2 Hot Tearing Criteria .............................................................225 5.3.2.1 Stress-Based Criteria .............................................225 5.3.2.2 Strain-Based Criteria ............................................. 229 5.3.2.3 Strain Rate-Based Criteria .................................... 231 5.3.2.4 Criteria Based on Other Principles......................234

Copyright 2008 by Taylor and Francis Group, LLC

Contents

ix 5.3.3

Application of Hot Tearing Criteria to DC Casting of Aluminum Alloys .................................. 236 5.3.4 Quest for a New Hot Tearing Criterion ............................. 240 5.4. Effects of Process Parameters on Hot Tearing during DC Casting ........................................................................... 245 5.4.1 Thermo-Mechanical Behavior of a DC-Cast Billet (Ingot) ........................................................................... 246 5.4.2 Effects of Composition and Casting Speed on Hot Tearing during DC Casting ................................... 251 5.4.3 Effect of Melt Temperature on Hot Tearing during DC Casting ............................................................... 259 5.4.4 Structural Features Associated with Hot Tearing in DC Casting ......................................................... 262 References ................................................................................................... 269 Concluding Remarks ...............................................................................275

Copyright 2008 by Taylor and Francis Group, LLC

List of Symbols and Abbreviations 1D 2D 3D A Bi C0 CE CL; CL Climit CS; CS L CSS D Dgr DS E Eq, eq Fz G G H K K L Lm N Ni NEq, neq P Pe Q Q Qw R S Se Sv T Tcoh Tm

one-dimensional two-dimensional three-dimensional a constant Biot number nominal alloy composition eutectic composition solute concentration in the liquid at the solid/liquid interface limit solubility of a solute in aluminum solute concentration in the solid at the solid/liquid interface solute concentration in the supersaturated liquid at the solid/liquid interface billet diameter grain size diffusion coeficient of a solute in the solid phase Young’s modulus equilibrium separation force shear modulus thermal gradient, temperature gradient depth of the sump partition coeficient permeability liquidus isotherm vertical dimension of the mushy zone amount (density) of particles in the melt principal quantum number of the i electron shell (Ai for aluminum) nonequilibrium undercooling parameter Peclet number low rate growth restriction factor water low rate billet radius; regression coeficient solidus isotherm interfacial area concentration of the grain envelope speciic surface area of the solid phase temperature coherency temperature melting temperature xi

Copyright 2008 by Taylor and Francis Group, LLC

List of Symbols and Abbreviations

xii Tsurf Tth · T Vc Vcast Vgrowth VL Vs Vshr V acr b c c d f fe fe fl fr fs g h kD kKC m m n na ni pr Ps q q/A t tf tR tV v ε ∆Hf ∆Hv ∆lexp ∆P ∆Pcr ∆Pmech

surface temperature temperature of thermal contraction onset (rigidity) cooling rate cooling rate casting speed interfacial growth velocity volume of liquid solidiication rate (linear velocity of the solidiication front) shrinkage-low velocity volume element critical defect size liquid ilm thickness; coeficient speciic heat tortuosity constant of dendrite network (secondary) dendrite arm spacing fraction of nucleating particles feeding rate volume fraction of grain envelope liquid fraction shrinkage (contraction) rate solid fraction gravity acceleration melt level permeability coeficient Kozeny–Carman constant microstructure parameter liquidus slope coarsening exponent, material parameter, coeficient number of atoms in a stable nucleus number of electrons on the i level (Ai for aluminum) reserve of plasticity effective feeding pressure heat transfer coeficient heat lux through surface A running (solidiication) time local solidiication time time available for stress relief vulnerable time period average (supericial) low velocity thermal strain rate enthalpy (latent heat) of fusion enthalpy (latent heat) of evaporation pre-shrinkage expansion pressure drop critical pressure drop pressure drop due to the deformation-induced luid low

Copyright 2008 by Taylor and Francis Group, LLC

List of Symbols and Abbreviations ∆Psh ∆sf ∆T0 ∆T br ∆Tc ∆Tn Γ Ξ Ω α αn α′ β βd βl βT ε εapp εfree εint εp εth φ γ λ λ1 λ2 µ ν θ ρ σfr σ ls ψ ψ( fs) ζcr ∆εres CET DAS DC casting El EMC EPMA

pressure drop due to the solidiication shrinkage volumetric entropy of fusion solidiication range brittle temperature range constitutional undercooling undercooling required for nucleation Gibbs–Thompson coeficient solutal supersaturation grain-reining criterion angle between the tangent to the coherency isotherm and the horizon; coeficient angle between the billet axis and the normal to the solidiication front Fourier number volumetric shrinkage; coeficient intradendritic permeability extradendritic permeability volumetric thermal expansion coeficient total solidiication shrinkage; strain; deformation actual (apparent) strain free thermal contraction strain internal strain elongation to failure linear thermal contraction porosity surface tension; effective fracture surface energy thermal conductivity primary dendrite arm spacing secondary dendrite arm spacing dynamic viscosity of the liquid Poisson’s ratio dihedral angle density fracture stress solid/liquid interfacial energy shape factor function discriminating the mechanical behavior of a semi-solid sample critical feeding rate reserve of technological strain (plasticity) columnar-to-equiaxed transition dendrite arm spacing direct chill casting tensile elongation electromagnetic casting electron probe microanalysis

Copyright 2008 by Taylor and Francis Group, LLC

xiii

List of Symbols and Abbreviations

xiv FEM GR HCS LTEC NES NGR PFT SC SEM SPV SRG TCC UTS

inite element method grain reined hot cracking susceptibility linear thermal expansion coeficient nonequilibrium solidus not grain reined pseudo front tracking model similarity criterion scanning electron microscope maximum volumetric low rate per unit volume (feeding term) rate of volumetric solidiication shrinkage thermal contraction coeficient ultimate tensile strength

Copyright 2008 by Taylor and Francis Group, LLC

Preface Direct chill (DC) casting is not a new technology; it has been known for more than 70 years. Despite important developments in this casting technique over the years, the core features have remained the same, and the prototypes invented in the 1930s are still evident in modern DC casting machines. For more than 60 years, this casting technology has been the primary means for producing extrusion billets and rolling ingots from a wide range of aluminum wrought alloys. There is no other technology on the horizon that could replace DC casting. This book is not about the technology of casting, though some principles and milestones are discussed in Chapter 1. The principles of solidiication and structure formation are very important for understanding the quality and performance of billets and ingots produced by DC casting. This book is not about the fundamentals of solidiication, but Chapter 2 gives all necessary information for understanding the formation of structure and defects during casting of aluminum alloys. Today most efforts in the ield of solidiication and practical casting are concentrated on computer modeling and process simulation. Physicists and mathematicians are replacing materials scientists at the forefront of research. This book is not about modeling and simulation, but the results obtained using these methods are widely used throughout the book as tools for interpreting experimental results and for studying the physical mechanisms involved. At this point, the reader might ask: What is this book about? The answer is in the book title: the book is about physical metallurgy of DC casting. This means that the formations of structure, properties, and defects in the as-cast material are considered in close correlation with the physical phenomena that are involved in the solidiication and with the process parameters. There is a serious lack of such information in the Western literature. The formation of structure in relation to the peculiarities of heat and mass low during DC casting is the main subject of Chapter 3, while the formation of major defects—macrosegregation and hot cracking—is the topic of Chapters 4 and 5. Throughout the book, the formation of structure and defects is considered in relation to the main process parameters: casting speed, melt temperature, water-low rate, and grain reinement. In this book, I have tried to present a logical system of structure and defect formation based on the speciic features of the DC casting process. The basis of this system is, on the one hand, melt low and heat low (cooling) in different parts of the solidifying domain and, on the other hand, kinetics of solidiication and the solidiication path of the alloy. The complexity of the mechanisms involved in the structure and defect formation is the main problem that frequently hinders clear understanding of the experimentally observed patterns. In this book, I try to single out these mechanisms and xv Copyright 2008 by Taylor and Francis Group, LLC

xvi

Preface

demonstrate that the seemingly controversial results reported in the literature are, in fact, caused by different ratios of the same mechanisms. This book is inspired by and based on results that have been obtained in the last 8 years from the extensive scientiic program on solidiication and casting of the Netherlands Institute for Metals Research. The scientiic foundation for understanding DC casting was established long before that, by extraordinary scientists and visionaries such as W. Roth, W. Patterson, V. Kondic, S.M. Voronov, V.I. Dobatkin, and V.A. Livanov. Among these names, Roth, Dobatkin, and Livanov should be distinguished as three individuals who successfully combined the practical development of DC casting technology with fundamental studies. Dobatkin and Livanov have also written monographs (V.I. Dobatkin, Direct Chill Casting and Casting Properties of Alloy, 1948; Ingots of Aluminum Alloys, 1960; V.I. Livanov et al., Direct Chill Casting of Aluminum Alloys, 1977) that were published in the former Soviet Union and remain to date the only scientiic books on DC casting. I frequently refer to these books because they are still very valuable in many respects. Two other important books provide a background for scientiic understanding of structure and defect formation during solidiication: Solidiication Processing by M.C. Flemings (1974) and Hot Tearing of Non-Ferrous Metals and Alloys by I.I. Novikov (1966). The current book is based on the important achievements of previous generations of scientists. We realize that we are building on an already existing foundation, and that we can only move forward if we have taken into account what has been achieved before us. Chapters 1, 4, and 5 include historic overviews of technologies, theories, and hypotheses that demonstrate the brilliant heritage we can utilize in our work. The original data that are presented in this book result from a team effort. I would like to acknowledge the contribution of post-doctoral researchers Dr. Q. Du and Dr. R. Nadella and Ph.D. students Suyitno, J. Zuidema, A. Stangeland, V. Savran, A. Turchin, and D. Ruvalcaba, who actively participated in the solidiication research. Our joint papers are extensively cited in the book. Technical assistance in setting up and performing experiments was provided by J.J. Jansen and J.J.H. van Etten, whose help and expertise are gratefully acknowledged. I am a metals scientist who graduated from the distinguished Department of Metals Science of the Moscow Institute of Steel and Alloys. I am greatly indebted to my teachers, Professor V.S. Zolotorevsky and Professor I.I. Novikov, for the knowledge they imparted to me and for the passion and critical approach to the research they fostered in me. DC casting has been a familiar subject for me since childhood. My father, Professor G.I. Eskin, is a well-known scientist in solidiication processing who worked with the renowned Professor V.I. Dobatkin for many years. It was, however, not until 1999 when I became involved in solidiication science. Two professors gave me this opportunity: Professor S. Radelaar, then the scientiic director and general manager of the Netherlands Institute for Metals Research, who hired me as a senior researcher and always supported me, and Professor L. Katgerman,

Copyright 2008 by Taylor and Francis Group, LLC

Preface

xvii

who heads the light-metals processing research at Delft University of Technology. I would like to express my profound gratitude to Professor Katgerman, who gave me many opportunities, supported me in my work, and shared with me his vast expertise. Many of the results presented in this book were obtained within the framework of the research program of the Netherlands Institute for Metals Research (www.nimr.nl), and I would like to thank Dr. S. Hoekstra for the opportunity to carry on our solidiication research. Constant help and support from Corus Aluminium are also gratefully acknowledged. Over the years I have met many individuals involved in solidiication research all over the world who have impressed me with their knowledge, attitude, and experience, including Professors A. Mo, L. Arnberg, M. Rappaz, C. Beckermann, and Drs. R. Mathiesen, Ø. Nielsen, R. Kieft, W. Boender, A.L. Dons, G.-U. Grün, J. Grandield, and J.-M. Drezet. And inally, I would like to thank my father and mother for their love and encouragement. Special thanks go to my wife, Natasha, who courageously and critically read the drafts of this book. I hope that this book will ill the gap in the literature on solidiication processing and provide new insight and perspective for DC casting research. The book is written for a wide audience that includes scientists, engineers, postgraduate and graduate students, and hopefully can be easily understood by readers without a special background in the subject. Dmitry G. Eskin

Copyright 2008 by Taylor and Francis Group, LLC

Author Dmitry G. Eskin, Ph.D., received his M.Sc. and Ph.D. in physical metallurgy in 1985 and 1988, respectively, from the Moscow Institute of Steel and Alloys (Technical University). He worked at the Baikov Institute of Metallurgy (Russian Academy of Sciences) until 1999, researching precipitation hardening, alloy development, and phase composition of aluminum alloys. Since 1999, he has worked at the Netherlands Institute for Metals Research in Delft, the Netherlands. As a senior scientist and fellow, Dr. Eskin is involved in an extensive research program on solidiication phenomena in light alloys, including industrial applications. The focus of this research is the interconnection among solidiication parameters, as-cast structure, and casting defects. A well-known specialist in the ield of aluminum alloys, Dr. Eskin has published more than 90 scientiic papers and co-authored three monographs.

xix Copyright 2008 by Taylor and Francis Group, LLC

1 Direct Chill Casting: Development of the Technology Direct chill casting of aluminum alloys, commonly known as DC casting, is an example of a technology that appeared just in time to serve the needs of the industry. Before discussing the history of DC casting, we will briely consider the history of aluminum. Aluminum is a “young” metal whose existence was established only 200 years ago. However, Pliny the Elder mentions a strange, light, and silvery metal in his Historia Naturalis, which might indicate that aluminum may have been discovered accidentally and then forgotten almost 2000 years ago: One day a goldsmith in Rome was allowed to show the Emperor Tiberius a dinner plate of a new metal. The plate was very light, and almost as bright as silver. The goldsmith told the Emperor that he had made the metal from plain clay. He also assured the Emperor that only he, himself, and the Gods knew how to produce this metal from clay. The Emperor became very interested, and as a inancial expert he was also a little concerned. The Emperor felt immediately, however, that all his treasures of gold and silver would decline in value if people started to produce this bright metal of clay. Therefore, instead of giving the goldsmith the regard expected, he ordered him to be beheaded.

Modern scientists and inventors were more fortunate and, as a result of their efforts, we now have a variety of aluminum alloys that are used in a wide range of applications. In 1808 the Englishman H. Davy established the existence of a new metal that he called alumium. This name later was changed to aluminum (U.S.A.) and aluminium (U.K.). It was not until 1825 that minute amounts of pure aluminum were extracted by the Dane N.C. Oersted. Between 1827 and 1845, the German F. Wöhler developed the irst process to produce aluminum powder by reacting potassium with anhydrous aluminum chloride. He also determined some physical properties of aluminum, including density, which appeared to be the most remarkable characteristic of the new metal. Jules Verne wrote in his “From the Earth to the Moon” in 1865 about the material of his ictitious space capsule: This valuable metal possesses the whiteness of silver, the indestructibility of gold, the tenacity of iron, the fusibility of copper, the lightness of glass. It is easily wrought, is very widely distributed, forming the base 1 Copyright 2008 by Taylor and Francis Group, LLC

2

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys of most of the rocks, is three times lighter than iron, and seems to have been created for the express purpose of furnishing us with the material for our projectile.

The era of light metals would have begun if not for one vital problem—the price. Despite some improvement to Wöhler’s process in 1854 by the Frenchman H.E.S.-C. Deville, aluminum remained very expensive, infact it cost more than platinum and gold. An ingot of aluminum was presented at the Paris World Exhibition in 1855 as a new precious metal. During the next 30 years aluminum remained an exotic, expensive material used in jewelry, royal cutlery, plates, and parade decorations. “Silver from clay” was an unoficial name of the metal. Napoleon III allowed only the most honored guests to eat from aluminum plates; the others were forced to settle for silver and gold ones. In 1889 the British Royal Society presented a set of scales made from gold and aluminum to the Russian scientist D.I. Mendeleev to honor his achievements in chemistry. In 1886, two 22-year-old scientists, Paul-Louis Toussaint Héroult of France and Charles Martin Hall of the United States, independent of each other and in different countries, invented a commercial process of producing pure aluminum by electrolysis of alumina dissolved in molten cryolite. Their process, the Hall–Héroult process, is still used today. In 1888, together with A.E. Hunt, C.M. Hall founded the Pittsburgh Reduction Company, now known as the Aluminum Company of America (ALCOA). By 1914, the cost of a kilogram of aluminum was down from $40 in 1860 to 40¢, and aluminum was no longer considered a precious metal. The annual production went from 15 tons in 1885 to 65,000 tons in 1913. Simultaneous or almost simultaneous discoveries and inventions occur frequently in the history of aluminum. This is an unmistakable sign of the need for such discoveries and inventions and the research that is under way throughout the industrial world. Remarkably, Jules Verne foresaw the most promising application for aluminum—airspace vehicles. The irst aircraft that took to the air on December 17, 1903 at Kitty Hawk was designed and built by the Wright Brothers. Their biplane was made from wood and fabric and powered by a 12-horsepower, 4-cylinder engine. And yet, even in this irst successful attempt to ly with a heavier-than-air machine, aluminum had a vital role as a one-piece cast crankcase of the engine. One more discovery was needed to fully implement the potential of aluminum in construction. Although castings from aluminum alloys were occasionally used in manufacturing some details, the strength of deformed aluminum was not suficient for its use as a structural material. The breakthrough came in 1908–1909 when A. Wilm of Germany found that the strength of an Al–Cu alloy would increase after several days of so-called “natural aging” after quenching. He patented the alloy and the technology of its heat treatment as “duralumin.” Now aluminum alloys could acquire the mechanical properties that would make them suitable for use in structures. The true era of aluminum had begun. It is interesting to note that the nature of hardening zones that cause room-temperature aging was discovered in

Copyright 2008 by Taylor and Francis Group, LLC

Direct Chill Casting: Development of the Technology

3

1938 simultaneously and independently by A. Guinier in France and G.D. Preston in Great Britain. Not surprisingly, the demand for wrought aluminum alloys came from the aircraft industry, fueled by World War I and later by increased requirements for speedy delivery of mail. The irst all-metal (mainly duralumin) airplanes were built by the German designer H. Junkers in 1915–1916. Junkers started from steel (J1) and then shifted to duralumin in the J7 and J9. The fuselage was made from plane and corrugated sheets. In 1919 Junkers together with O. Reuter designed the irst all-metal passenger plane, the F13 (J13). This design served as a benchmark for several aircrafts made in different countries [1]. The Soviet Union was quick to follow. After World War I, Germany was not allowed to manufacture military aircraft, so the U.S.S.R. provided concessions for Junkers to build his aircraft in Moscow from 1923 to 1927. The designer A.N. Tupolev made his irst aluminum plane, the ANT-2, in 1924 using Russian-made duralumin. Just a year later the United States followed with the passenger Stout Air Sedan and Ford 3-AT. Tupolev made a very large, all-metal heavy bomber, the ANT-4 (TB1), in 1925. This plane was 18 m long, with a 28.7-m wing span, and was powered by two engines. In the late 1920s, Ford produced several passenger planes, including the famous 4-AT “Tin Goose.” Despite all these efforts, only 5% of all aircraft in production by 1930 were of all-metal construction. In the 1930s the corrosion resistance of duralumin had been improved by cladding with pure aluminum, and a wide variety of airplanes were manufactured from aluminum alloys. The U.S.S.R. continued to design and build record-size heavy bombers and passenger airliners including the ANT25 in 1933 that lew 11,500 km nonstop in record time, in 1937 from the U.S.S.R. to the United States; the 8-engine ANT20 “Maxim Gorky” made in 1934 that was the largest aircraft of its time (33 m long with a 63-m wing span, and a passenger capacity of 48); and the ANT37 (DB2), a long-range bomber put in service in 1936. Junkers in Germany designed and produced midsize and practical passenger planes such as the Ju60 in 1932 and the Ju160 in 1934. In the United States Boeing started to produce passenger planes, the Boeing 200 in 1930 and the Boeing 247 in 1933, and was joined by Douglas with the DC1–DC3 in 1933–1935. The DC3 was the irst aircraft that actually enabled airlines to make money from passenger rather than mail transport. In addition to the “main players”—Germany, the United States, and U.S.S.R.—other countries made their efforts as well. France manufactured the Devoitine D332 in 1933, and the United Kingdom produced the Ensign-1 in 1934 and the de Havilland DH95 in 1938. Some of the benchmark planes are shown in Figure 1.1 [1]. Throughout the industrialized world all-metal, mainly aluminum aircraft were dominant by the late 1930s. It became obvious that mass fabrication of wrought aluminum products was necessary to sustain the development of military and civil airplanes. Let us now consider the manufacturing technology for large-scale wrought aluminum ingots and billets required for rolling, extrusion, or forging. From 1924 to 1939 the average weight of an aluminum lat ingot cast in a permanent mold increased from 20 to 500 kg [2]. By the mid-1930s most

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

4

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

FIGURE 1.1 Examples of all-metal airplanes designed and built in the 1920s and 1930s: (a) the Junkers F13 (1919, Germany); (b) the ANT-4 (1925, U.S.S.R.); (c) the Ford 4-AT (1926, U.S.A.); (d) the Junkers Ju60 (1932, Germany); (e) the Boeing 247 (1933, U.S.A.); (f) the ANT-20 (1934, U.S.S.R.); (g) the Douglas DC-3 (1935, U.S.A.); and (h) the ANT-37 (DB2) (1936, U.S.S.R.) [1].

bulk ingots and billets were made by permanent-mold casting, while the increased volume and cross-section resulted in exponential deterioration of the structure and properties of the cast metal. Three main problems were intrinsically present in permanent-mold casting: turbulence during pouring; air-gap formation during cooling; and variable low cooling rate across the ingot. Several innovations were proposed to deal with these drawbacks. In the Züblin method the mold was illed gradually from the bottom to the top by moving the feeding nozzle upward and corresponding closure of the sidewall [3, 4]. This technique helped avoid large temperature gradients

Copyright 2008 by Taylor and Francis Group, LLC

Direct Chill Casting: Development of the Technology

5

and decreased chemical and structure inhomogeneity of the ingot. Another widely used technique was the so-called tipping mold, the main purpose of which was to avoid turbulence while illing the large volume. The idea was the same as the process of illing a glass with beer in order to avoid too much foaming: one starts to ill the tilted glass by pouring the beer along the glass wall and gradually straightens the glass until it is completely full without foam. Another casting technology that was a transition stage to the direct chill casting was casting by dipping the illed mold gradually into the water [5]. With this technique, solidiication was almost directional from the bottom to the top and the solidiication front was lat. As a result, the thermal gradients were low, and large ingots of 800 × 1300 mm2 could be cast. Fig ure 1.2 illustrates schematically these methods of casting. However, the problems of airgap and low cooling rates, and thus the problems of uniform structure and properties, remained unsolved and required revolutionary change in technology. It was absolutely clear that there was no further room for improvement in the permanent-mold casting of large ingots and billets. In response to these demands, German and American engineers devised a solution that addressed two problems successfully and simultaneously: control of solidiication and production of large ingots and billets. The solution was in the modiication of a so-called continuous casting process that had been proposed and sluggishly developed for almost 80 years but never used in mass production.* Henry Bessemer, the inventor of modern steelmaking, suggested in 1856 the irst method of steel strip casting by pouring the melt in the opening between two rotating wheels [6]. In 1886 B. Atha of the United States suggested casting molten steel into a high, water-cooled, bottomless mold and extracting the resulting billet with withdrawal rolls [6]. This method was used for semi-commercial production of 100 × 100 mm steel bars in the early twentieth century. However, these inventions had limited, if any, application, and the continuous casting of steel was not in high demand until the 1950s. A similar technique was developed by Siegfried Junghans in the early 1930s [7]. His machine was initially used at Wieland-Werke in Germany for casting brass [4]. The mold consisted of a copper tube open at both sides and surrounded by a water jacket. The melt was fed into the mold from the top and the solid part was withdrawn by rolls from the bottom. The melt feeding was adjusted to the withdrawal speed by a special system in such a way that a constant melt level was maintained in the mold. The mold was lubricated and given an up-and-down oscillating movement to prevent the sticking of the solid metal to the mold walls. Flying saws were positioned in the pit below the installation for continuous cutting of the billet into required lengths. This successful scheme was widely used for casting copper and aluminum alloys in Germany, the United States, and the U.S.S.R. Figure 1.3 depicts the Junghans method of continuous casting. Junghans later added * The vertical DC casting of aluminum alloys is the main subject of this book. Other types of continuous casting are mentioned here only as predecessors or analogs of vertical DC casting.

Copyright 2008 by Taylor and Francis Group, LLC

6

Crucible

Mold in the starting position

Ladle with melt Dipping mechanism

Water

Built-up wall

Mold in an intermediate position during casting

(a)

(b)

(c)

FIGURE 1.2 Some typical methods used for casting large ingots and billets in permanent molds: (a) the Züblin mold with a feeding trough and a built-up wall [4]; (b) a tipping mold in two positions [4]; and (c) a mold with a melt before being dipped in a water tank [5].

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Melt

Direct Chill Casting: Development of the Technology

7

Holding furnace

Melt

Feeding channel

Mold

Water Solid billet

Withdrawal rolls

FIGURE 1.3 A diagram of Junghans method of continuous casting [8].

water spraying directly on the billet and made several innovations to the proper melt feeding/distribution system. Dobatkin [8] mentions that the maximum casting speed achieved for aluminum alloys with the Junghans machine was 300 mm/min for relatively small rods. Compared with permanent-mold casting, the Junghans method offered the following advantages [9]: • A truly continuous process with the possibility of advanced automation that increased productivity with less manpower • Reproducible casting regimes that allowed the reproducible quality of billets • Improved feeding of the central portions of billets with a correspondingly increased soundness of billets • More uniform structure across the billet • Better removal of gases during casting through the liquid portion of the billet • Less scraped material

Copyright 2008 by Taylor and Francis Group, LLC

8

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

At the same time, this method did not solve all problems. The following drawbacks remained, most of which were related to the fact that the heat was still extracted through the walls of the mold [9]: • Air gap formation, deep sump, and high thermal gradients • Macrosegregation and inhomogeneous structure in large billets 300–500 mm in diameter • Low casting speeds • Need for a long mold with high surface quality requirements To eliminate these shortcomings, it was necessary to develop a technology where the heat would be extracted mainly through the solid part of the casting. As a result, the sump of the casting would be shallower and the solidiication proile would be latter. The macrosegregation, structure inhomogeneity, and radial stresses would be much less pronounced. These needs were met by a new casting technology developed, again almost simultaneously and independently, in Germany and the United States. The technology was given the German name “Wasserguß” or “water-casting,” and was later called “direct chill casting.” Berthold Zunkel in 1935 [10], Walter Roth in 1936 [11], and William T. Ennor in 1938 [12] iled and subsequently received patents for the casting technology with several common features. Melt was poured from the top in an opened, relatively short, water-cooled mold that at the beginning is closed from the bottom by a dummy block connected with a hydraulic or mechanical lowering system. After the melt level in the mold reaches a certain level, the ram is lowered and the solid part of the billet or the ingot is extracted downward. The melt low rate and the casting speed are adjusted in such a way that the melt level in the mold remains constant. As soon as the solid shell emerges from the bottom part of the mold, water is applied to the surface in the form of spray or water ilm. Cooling of the solid billet (or ingot) is further intensiied by lowering it into a pit illed with water. The process is semi-continuous. As soon as the ram reaches its lowest position in the pit, the casting is stopped and the billet (or ingot) is removed from the pit. Figure 1.4 shows schematic drawings of these three inventions, clearly demonstrating that their similarities are more important than any differences. The new casting technology provided lexibility in casting applications. Trials started immediately with round billets, rectangular ingots (or rolling slabs), and hollow billets (pipes). In addition, multiple casting was possible with several molds positioned on a single casting table. DC-casting inventions were commercialized in Germany at Vereinigte Leichtmetall-Werke beginning in 1936 [4] and in the United States at ALCOA from 1934 [13]. The water cooling or direct chill in these irst patents was done either by dipping the billet directly into the water bath or by spraying water onto the surface using separate sprinklers located along the billet length. Already in his 1937 patent application in Great Britain (GB Patent 492216) Roth suggested using openings in the lower part of the water-cooled mold for spraying the coolant onto the billet surface. Roth, Patterson, and Kondic

Copyright 2008 by Taylor and Francis Group, LLC

Direct Chill Casting: Development of the Technology

9

were also among the irst to publish scientiic papers based on extensive research of this new technology [14–18]. In 1939, the U.S.S.R. began extensive work on the development and use of DC casting of aluminum alloys, building mainly on the German inventions as shown in Figure 1.4d [8]. This rapid advance was facilitated by close economical links between the U.S.S.R. and Germany with many Russian and German engineers exchanging ideas and technological knowledge. Well-known names in Russian DC casting include S.M. Voronov, V.A. Livanov, R.I. Barbanel, and V.I. Dobatkin. Extensive reports on the Russian experience in DC casting and physical metallurgy of the process were published immediately after World War II [8, 9, 19]. Roth, Livanov, and Dobatkin recognized the important role of heat transfer, thermal contraction, and the dimensions of the semi-solid region in the billet and made the irst attempts to explain the formation of hot and cold cracks, macrosegregation, and the homogeneity of structure and properties distribution. The early reports on the application and mastering of the new technology highlight the great dificulties that engineers had to overcome on the casting loor. In addition to mechanical dificulties, one of the main problems reported was splitting and fracture of ingots, especially from high-strength alloys such as duralumin and emerging Zn-containing aluminum alloys [13]. Today we still face these problems, known as cold cracks and hot tearing. It was soon acknowledged that the main drawback of DC casting that caused cracking problems was the high thermal gradient between the surface and the interior of the billet (ingot) that resulted from direct cooling with water. Kondic put it this way: “The problem of continuous casting is essentially a matter of heat exchange between the metal cast and the cooling medium” [17]. Another problem was bleed-out, when the solid shell was fractured below the mold and the melt went through it. On the other hand, high cooling (and solidiication) rates achieved during DC casting, especially in short molds, were advantageous for structure formation, producing ine and homogeneous structure across the billet (ingot). Livanov wrote in 1945: “The essence of direct chill casting is the possibility to sharply increase the cooling rate” [19]. Kastner suggested two ways of decreasing the danger of cracking and bleed-outs: either increase the distance between the point where direct cooling was applied and the bottom of the mold or decrease the casting speed in such a way that the billet was solidiied completely over the whole crosssection before it reached the region of direct cooling [4]. Later, it was shown that sound billets and ingots could be obtained under a variety of conditions, provided the casting processes were well understood. Direct chill casting had a unique feature that distinguished it from other casting techniques. The solidiication occurred in a narrow layer of the casting inside and below the mold. During the steady-state stage of casting, the shape and the dimensions of this region remained constant and reproducible from one heat to another. By regulating the melt distribution during feeding the mold, cooling conditions inside the mold, direct cooling below the mold, and the casting speed, the shape and dimensions of the solidiication region

Copyright 2008 by Taylor and Francis Group, LLC

10

Melt

Melt Melt

Mold

Water-cooled mold

Water

Water

Billet

Billet

Billet

Water

Water

Withdrawal ram (a)

Withdrawal ram (b)

FIGURE 1.4 DC casting methods patented by Zunkel [10] (a), Roth [11] (b), and Ennor [12] (c), and a scheme of a working DC casting machine used in the 1940s [8] (d).

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Ladle with melt

Distribution box Melt Melt

Mold

Water

Mold

Water

Billet Billet

20°

Direct Chill Casting: Development of the Technology

Melt level control

Withdrawal ram Withdrawal ram

(c)

Copyright 2008 by Taylor and Francis Group, LLC

11

FIGURE 1.4 (Continued)

(d)

12

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

could be controlled. Because shape and dimensions determined the thermal gradient and were responsible for cracking, macrosegregation, and structure homogeneity, the occurrence of these defects could also be controlled. Determination of a correct casting recipe for each alloy and size was essential. The main rules were formulated as follows [8, 19]. The melt level in the mold should be minimum; the water should be applied onto the billet surface as close to the bottom part of the mold as possible and at the minimum angle to the billet axis; and the melt should be evenly fed into the sump using special distribution boxes or bags. Maintaining these rules, Dobatkin reported that the optimal casting speed for 400-mm billets from high-strength aluminum alloys ranged between 40 and 85 mm/min [8]. The interrelation between the casting speed and the shape and dimensions of the sump and the transition region (between liquidus and solidus isotherms) was quickly recognized and main relationships were established between the physical properties of the alloy, the dimensions of the billet, and the process parameters [8, 15, 19]. Roth showed that the depth of the billet sump (H) is directly proportional to the casting speed (Vcast) and to the squared radius (R) [15, 16]: H = AVcastR2 (1.1) This dependence was conirmed by Livanov [19] and Dobatkin [8]. Livanov found out that the solidiication rate (the velocity of the solidiication front) depends on the casting speed as Vs = Vcastcos αn,

(1.2)

where αn is the angle between the billet axis and the normal to the solidiication front [19]. As a result, the solidiication rate is a function of the shape of the solidiication front and is usually maximum in the center and at the periphery of the billet. It was also shown that the ratio between the sump depth and the billet diameter can be maintained constant if VcastR = const [8]. However, for each billet size and alloy there exists a maximum solidiication rate that cannot be exceeded by further increasing the casting speed [8]. This maximum solidiication rate is lower for larger billet diameters. These fundamental relationships provided the basis for scaling up the technology to very large castings. Today round billets more than 1000 mm in diameter and 4700 mm long and ingots over 2400 × 1000 mm2 in cross-section and 4400 mm long can be successfully cast from high-strength aluminum alloys [20]. Extensive research undertaken in close connection with industrial DC casting produced outstanding results. By the end of World War II all highstrength aluminum alloys in the United States and the U.S.S.R. were cast using this new technique. The comparison of DC casting and Junghans’ method showed the following advantages of the former [9]: • Considerably reduced centerline segregation • Increased density of the central portion of a billet

Copyright 2008 by Taylor and Francis Group, LLC

Direct Chill Casting: Development of the Technology

13

• Finer and more homogeneous structure with correspondingly improved mechanical properties • Better surface quality • Lower operation costs The irst three features were attributed to the shallower sump and latter solidiication front. It was shown that, for the same casting speed and billet diameter, the sump depth would be four times greater for the Junghans’ method than for DC casting [9]. Therefore, using the scaling rules, one can say that the billet with the same sump depth can be cast four times faster with DC casting than with Junghans’ technique. The DC-cast billet would also have four times higher solidiication rate and, hence, better structure and properties, as demonstrated by Dobatkin [8]; see Figure 1.5. The economical beneits of the new casting process were summarized as a decrease of 1.5 times in workplace occupied; scrap reduction from 25 to 10%; and a 50% reduction in workforce [9]. The working conditions became so much easier that Voronov [9] noted that women could be employed in the cast house, which was very important in wartime. Despite the great and obvious success of DC casting as a major technique for producing billets and ingots for further deformation, there remained some intrinsic drawbacks that inspired inventive people to suggest numerous improvements to the technology. Some of these inventions were shortlived and some were minor, but others were milestone achievements.

Average dendrite arm spacing, µm

250

× 200 238

200

× 200 183

150

× 200

100

100 × 200

52

50

× 200 20

0

× 200 7

0.2

0.5

3.1 8.0 1.7 Solidification rate, m/h

FIGURE 1.5 Dependence of dendrite arm spacing on the solidiication rate in DC casting [8].

Copyright 2008 by Taylor and Francis Group, LLC

35

6

55

14

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

The necessity to maintain low melt level in the mold became a more serious problem as the cross-sections of billet and ingots became larger and larger. But there was no other option—the melt level had to be low in order to avoid large thermal stresses, a wide transition region, and a large zone of air-gap within the mold. Due to thermal contraction of the just solidiied shell, this shell withdraws from the mold surface, reheats, and may remelt. Because the metallostatic head is large, the melt can penetrate through the remelted crust and cause periodic segregation bands at the billet surface or bleed-out, resulting in deterioration of the quality of the casting or abandonent of the casting altogether. The shape of the sump and the distribution of cooling rates for the cases of high and low melt levels are demonstrated in Figure 1.6a and b [21]. Note that the region of very low cooling rates is present inside the mold when casting with a high melt level and is absent when casting with a low melt level. In 1958, Gunther E. Moritz of Reynolds Metals proposed a solution for the problem of the low melt level by simply isolating the upper part of the mold with a heat-insulating material [22]. The effective mold length (similar to the former melt level) could be controlled by the sizes of this insert. As a result, the effective mold length or the effective melt level decreased while the real melt level could be maintained as high as suitable for the casting control. The level of the melt above the insert was no longer important. What mattered was the distance between the lower edge of the refractory insert and the lower edge of the water-cooled mold. Figure 1.7a shows the schematic of this invention. As the next step, Moritz suggested making the heat insulating insert composite with the internal graphite ring for better surface quality [23]. This was the beginning of the “hot top” era in DC casting. Later, the refractory part was extended above the mold and became a part of the top melt container connected to the feeding system patented by Furness and Harvey (Great Britain, British Aluminium) in 1973 [24], hence the name “hot top”; see Figure 1.7b. The control of heat transfer in the mold was further improved by constant feeding of gas/lubricant mixture into the mold. This process was invented by Ryota Mitamura and Tadanao Itoh at Showa Denko in Japan in 1977 [25] and later reined by Wagstaff Engineering as the “air-slip” mold where the gas/ lubricant mixture was supplied to the inner surface of the mold through a ineporous graphite ring [26]. The main improvement was in the surface quality. The introduction of hot-top mold was an important and popular development that spread through casting houses around the world in the 1970s. Despite all these efforts air-gap formation could not be avoided. An area of coarser structure was formed inside the casting at some distance from the surface because of the lower cooling rate at which the billet was solidiied while traveling from the point of the air-gap formation to the point of water impingement below the mold. An elaborate attempt to overcome this problem was the invention of a so-called “dual-jet” mold by Wagstaff Engineering [27]. In this mold design two water jets exited the mold and hit the billet surface at two different angles, 22° and 45°, as shown in Figure 1.7c. The region of slower cooling was thus considerably reduced. Wagstaff Engineering also produced a combination of air-slip and dual-jet molds.

Copyright 2008 by Taylor and Francis Group, LLC

Melt level

Melt level 60

100

120

50

35 30

Melt level

115 100

Inductor 0

15 20

60

0

360 1180

280

160 130 505° 130

50

645°

180 ° 05 160

615°

150

160

150

100



61

5

130

160

60

100

100

150

200

250

60 740 340

645° 240

50

645° 60

200

280 350

600 380 1000°/min

Distance from the bottom of the inductor, mm

Distance from the bottom of the mold, mm

50

200 160

100 130

50

160

100

60



130

200

100

150

200

250

300

350

300

400

350

450

Direct Chill Casting: Development of the Technology

Mold

150

500 400

(a)

(b)

550

(c)

Copyright 2008 by Taylor and Francis Group, LLC

15

FIGURE 1.6 Shape of the sump, transition region (gray zone) and isolines of cooling rates (K/min) upon casting in the normal mold with low (a) and high (b) melt levels and upon casting in the electromagnetic mold (c) [21].

Melt

Actual melt level

Water

Melt

Water

Sump depth

Mold

Billet

(a) Water chamber

Graphite insert

Water

Water Billet Solid shell

(b)

Mold Distribution system Melt

Cooling water

Electromagnetic inductor

Water chamber

Billet Billet Water jets

(c)

(d)

FIGURE 1.7 Development of DC casting molds in the 1960s–1990s: (a) refractory insert, 1958 [22]; (b) hot top, 1973 [24]; (c) dual-jet mold, 1996 [27]; and (d) electromagnetic casting, 1969 [28].

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Effective melt level

16

Refractory hot top

Refractory insert

Direct Chill Casting: Development of the Technology

17

The revolutionary invention was electromagnetic casting (EMC) or moldless casting. The idea of DC casting included the mold with primary cooling of the melt where the solid shell was formed and the direct cooling of this shell with water outside the mold. The existence of the mold created some dificulties: premature solidiication with potential hanging of the billet or ingot, freezing of liquid meniscus and formation of a banded surface with cold shuts, air-gap formation and corresponding change of solidiication conditions, requirements for lubrication, etc. Engineer Zinovy N. Getselev, who worked at Kuibyshev Aluminum Works in the U.S.S.R., came up with an astonishingly simple and controversial idea: to abandon the mold altogether. The molten metal is shaped and held in the required position by an alternating electromagnetic ield that induced electromotive forces and eddy currents in the melt. The interaction of these eddy currents with the external magnetic ield creates electrodynamic forces that compress the melt. The cooling water is applied onto the surface of this molded melt and the solidiied part is withdrawn downward as in the normal DC casting [28]. A successful start-up is, however, dificult and requires very accurate maintenance of several parameters such as the pouring rate, water low rate, initial casting speed, and strength of the magnetic ield. Figure 1.7d depicts the principle of the proposed method. The most important features of this method of casting are the absence of the air gap between zones of primary and secondary cooling typical of traditional DC casting and stirring of the melt in the sump by electromagnetic forces. This technology and its variations were patented in most industrialized countries between 1969 and 1977. The advantages of EMC include very good surface that does not require scalping, less macrosegregation due to the melt stirring during casting; and more uniform and generally iner structure [21]. The sump during electromagnetic casting is shallower and the thermal gradients are less, as shown in Figure 1.6c. The disadvantage of this method that limits its wider application is the relative complexity of the casting machine. The history of DC casting shows that engineering ingenuity has always matched the demands of the industry and met the challenge for new products and new materials. DC casting is a relatively new technology that is rapidly developing. In many cases this advancement is limited by the lack of understanding of mechanisms of structure and defect formation. It is essential that physical metallurgy catch up with the technology and provide the scientiic basis for even better technology.

References 1. http://www.airwar.ru/main.html. 2. V.A. Livanov. In: Belov A.F., Agarkov G.D. (ed.), Aluminum Alloys, Collection of Papers, Moscow: Oborongiz, 1955, pp. 128–168.

Copyright 2008 by Taylor and Francis Group, LLC

18

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

3. J. Züblin. Apparatus for Casting Metals, 1929, US Patent 1734786. 4. H. Kastner. Metal Industry, 1947, vol. 71, August, pp. 83–85; 106–108; 131–132. 5. V.I. Dobatkin, E.G. Safonova. In Aluminum Alloys, Collection of Papers, Moscow: Oborongiz, 1955, pp. 188–215. 6. M. Wolf. Rev. Metall.–CIT, 2001, no. 1, pp. 63–73. 7. S. Junghans. Apparatus for Continuous Casting of Metal Rods, 1938, US Patent 2135184, iled 1936; German Patent 750301, iled 1933. 8. V.I. Dobatkin. Continuous Casting and Casting Properties of Alloys, Moscow: Oborongiz, 1948. 9. S.M. Voronov. Continuous Casting of Round Billets from Light Alloys, Selected Works on Light Alloys (1946), Moscow: Oborongiz, 1957, pp. 336–362. 10. B. Zunkel. Gießvorrichtung zum ununterbrochenen Geißen von Blöcken und ähnlichen Werkstücken aus Lichtmetall oder Leichmetallegierungen, 1939, DR Patent 678534, iled 1935. 11. W. Roth. Verfahren zum Gießen von Metallblöcken mit Ausnahme solcher aus Leichtmetallen, 1960, BRD Patent 974203, iled 1936. 12. W.T. Ennor. Method of Casting, 1942, US Patent 2301027, iled 1938. 13. M.B.W. Graham, B.H. Pruitt. R&D for Industry. A Century of Technical Innovation at Alcoa, Cambridge: Cambridge University Press, 1990, pp. 251–262. 14. P. Brenner, W. Roth. Metallwirtschaft, Metallwissenschaft, Metalltechnik, 1942, vol. 21, pp. 695–699. 15. W. Roth. Aluminium, 1943, vol. 25, pp. 283–291. 16. W. Roth. Z. Metallkde., 1949, vol. 40, pp. 445–456. 17. V. Kondic. Metal Industry, 1944, vol. 65, pp. 56–58. 18. W. Patterson. Aluminium, 1943, vol. 25, pp. 75–78. 19. V.A. Livanov. In Proceedings of the First Technological Conference of Metallurgical Plants of Peoples’ Commissariat of Aviation Industry, Moscow: Oborongiz, 1945, pp. 5–58. 20. Light Metal Age, 2003, no. 1, pp. 47–49; 2005, no. 4, pp. 32–33. 21. V.A. Livanov, R.M. Gabidullin, V.S. Shipilov. Continuous Casting of Aluminum Alloys, Moscow: Metallurgiya, 1977. 22. G.E. Moritz. Metal Casting System, 1961, US Patent 2983972, iled 1958. 23. G.E. Moritz. Continuous Casting System, 1965, US Patent 3212142, iled 1962. 24. A.G. Furness, J.D. Harvey. Mould Assembly and Method for Continuous or Semi-continuous Casting, 1975, GB Patent 1389784, iled 1973. 25. R. Mitamura, T. Itoh. Process for Direct Chill Casting of Metals, 1979, US Patent 4157728, iled 1977. 26. F.E. Wagstaff, W.G. Wagstaff, R.J. Collins. Direct Chill Metal Casting Apparatus and Technique, 1983, UK Patent Application 2129344A; 1986, US Patent 4598763, iled 1985. 27. R.B. Wagstaff, D.A. Salee. Direct Cooled Metal Casting Process and Apparatus, 1996, US Patent 5582230, iled 1994. 28. Z.N. Getselev, G.A. Balakhontsev, V.K. Zinoviev, et al. Method of Continuous and Semicontinuous Casting of Metals and a Plant for the Same, 1969, US Patent 3467166, iled 1967. Z.N. Getselev. Continuous Casting Apparatus with Electromagnetic Screen, 1971, US Patent 3605865, iled 1969.

Copyright 2008 by Taylor and Francis Group, LLC

2 Solidification of Aluminum Alloys

2.1

Effect of Cooling Rate and Melt Temperature on Solidiication of Aluminum Alloys

There are two vital factors to consider regarding the formation of structure and the quality of any casting—cooling rate and melt temperature. The cooling rate relects the heat extraction rate and is measured in K/s. Cooling rate is closely connected to the solidiication rate, which can be deined either as the velocity of the solidiication front or as the liquid–solid phase transformation rate, measured in units of m/s or s−1, respectively. The solidiication rate and cooling rate affect the structure formed during solidiication in a manner that was demonstrated in Figure 1.5. Generally, the structure is reined by increasing the heat extraction and corresponding increase in the solidiication rate. Let us look at these relationships in some detail. In fundamental and theoretical studies casting structure is often represented as unidirectional and columnar, with the grains or their branches represented by cylinders with rounded tips. In reality the structure of commercial alloy castings consists of dendritic equiaxed grains. The structure is therefore conventionally considered on two levels: grain size and dendrite arm spacing, as illustrated in Figure 2.1a. The grain usually has a rather complex, branched shape, hence the name— dendrite. Its branches are called dendrite arms. Dendrite grains are formed in alloys solidiied in the presence of thermal gradient at the solid–liquid interface and their shape is a result of chemical inhomogeneity and corresponding instability at this interface, as shown in Figure 2.2. The grain size is obviously a function of several parameters, the most important of which are the nucleation rate and the growth rate. The nucleation rate depends on the amount of energy required for the creation of a new phase structure and new surface area. Therefore, the nucleation rate can be affected by the melt undercooling below the liquidus (which gives a direct thermodynamic stimulus for nucleation, decreasing the critical size of the solidiication nucleus) and/or by the presence of solidiication sites for heterogeneous nucleation (that provide the suitable surface for nucleation, decreasing the amount of required energy). The latter mechanism can be facilitated by special additions called grain reiners. The growth rate is

19 Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

20

1 mm (a)

1 mm (b) FIGURE 2.1 The structure of an Al–1.8% Cu alloy solidiied at a cooling rate of 0.4 K/s (a) and 13 K/s (b). Grain size D and dendrite arm spacing d are shown in (a).

a function of the crystallography (growth kinetics is different at different crystal planes), heat extraction, and mass transfer. The liquid undercooling is higher at higher cooling rates. As a result, grains generally nucleate and grow faster at higher cooling rates, with corresponding grain reinement; see Figure 2.1b.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

21

200 µm FIGURE 2.2 Instabilities (branches shown by arrows) at the dendrite tip interface caused by an abrupt change of thermal gradient as a result of quenching in the semi-solid state (an Al–3% Si alloy).

The dependence of the grain size (Dgr) on the important kinetic parameters of solidiication can be written as [1] Dgr = A∆T0.25 V–0.25 G –0.5, s 0

(2.1)

where ∆T0 is the alloy solidiication range, Vs is the velocity of the solid–liquid interface (solidiication rate), G is the interfacial temperature gradient, and A is the alloy-dependent coeficient that takes into account the interface surface energy, solute diffusion, distribution coeficient, and latent heat of fusion. For the same alloy, the most decisive factor for the grain size is the temperature gradient. Under identical cooling conditions, the equilibrium solidiication range is one of the important parameters that govern the grain size [2]. This explains the experimentally observed fact that the increased amount of solute in alloys results in grain reinement [2, 3], as demonstrated in Figure 2.3a for binary Al–Cu alloys. The effect of the alloy composition on grain reinement is considered in more detail in Section 2.4. The internal ineness of the grain structure is characterized by the size and spacing of dendrite branches. The analysis of dendrite branching usually involves the phenomenon of so-called constitutional undercooling [1, 4]. During solidiication of a hypoeutectic alloy (with the partition coeficient K = CS/CL < 1), the solute concentration in the liquid very close to the solid– liquid interface is higher than the bulk liquid concentration because of the

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

22

4.3% Cu 3.24% Cu 2.12% Cu 0.98% Cu

Grain size, mm

4 3 2 1

Dendrite arm spacing, µm

5

100

10

0 0 (a)

4.3% Cu 3.24% Cu 2.12% Cu 0.98% Cu

4

8

Cooling rate, K/s

0.1

12 (b)

1.0

10.0

Cooling rate, K/s

FIGURE 2.3 Effect of cooling rate and composition on the grain size (a) and dendrite arm spacing (b) in binary Al–Cu alloys [3]. (Reproduced with kind permission of Elsevier.)

solute rejection from the solute-lean solid phase. This situation is shown in Figure 2.4. As a result of this liquid enrichment, the equilibrium liquidus at the interface may lower signiicantly. Local equilibrium that usually exists at the interface under normal solidiication conditions dictates that the actual temperature of the melt at the interface should be equal to the liquidus temperature corresponding to the local composition. The existing melt temperature gradient creates the situation that the melt temperature in the vicinity of the interface is lower than the actual liquidus, forming the region of constitutional (in contrast to thermal) undercooling. In this case, any protrusion at the solid–liquid interface happens to be in the constitutionally undercooled region and has a possibility to become stable and grow. This effect is more pronounced in the case of columnar growth when there is no thermal undercooling (marked b in Figure 2.4). For equiaxed grains, the thermal undercooling is always present so it seems that contribution of the constitutional undercooling is not so important (marked c in Figure 2.4). In real conditions, however, the thermal undercooling in liquid regions between equiaxed grains is to a great extent compensated for by the latent heat evolution and the very low ratio between the solute and heat diffusion, which makes the impact of constitutional undercooling much more pronounced. It is important to note that the region of the maximum constitutional undercooling occurs at some distance from the solid–liquid interface, which has profound implications for the columnar–equiaxed transition and grain reinement. Dendrite branches are initially formed as periodic disturbances of the lat interface. The average distance between the interface protrusions relects the minimum segregation distance when the diffusion ields of the neighboring branches do not overlap yet [1, 5]. As soon as the diffusion ields overlap, the branches stop growing and the process of their ripening (coarsening) starts,

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

C iS

23

Liquid

TC0

Tx

Solid

S

Cx

(a)

L

C0

C

Cx

Heat flow Heat flow

Heat flow

CxL

Cis

C is

Tmelt Tc0

Solid

Liquid C0

C0

Cxs

Cxs

Tc0 Tinterface = Tx

Tinterface = Tx

(b)

CxL

Liquid

Region of constitutional undercooling

(c)

Solid

Region of constitutional undercooling

Tmelt

FIGURE 2.4 Concentration and temperature ields around a dendritic grain showing the concept of constitutional undercooling [1]: (a) section of a hypoeutectic region of a binary eutectic phase diagram; (b) columnar grain growth; and (c) equiaxed grain growth. C0, CSx, and CLx are the compositions of the alloy, solid, and liquid phases at a particular melt temperature Tx, respectively. CSi is the initial solid concentration at the liquidus temperature of the alloy C0. Tinteface is the equilibrium liquidus temperature corresponding to the solute concentration in the melt. Tmelt is the melt temperature far from the solid–liquid interface (or solidiication front) corresponding to the nominal alloy composition C0. Regions of constitutional undercooling are shown in (b) and (c).

when iner branches disappear and coarser ones dominate [1, 4]. Eventually, the spacing between branches becomes wider. This process is considered in more detail in the next section. The distance between these branches is called the secondary dendrite arm spacing because these branches appear on the primary trunks of the dendrite.

Copyright 2008 by Taylor and Francis Group, LLC

24

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

It has been shown that the secondary dendrite arm spacing (d) is a unique function of the local solidiication time (tf) [4]: d = Btnf .

(2.2)

As tf = ∆T0/GVs and GVs is the cooling rate (Vc) in the solidiication range, we can write the same expression in a well-known form, d = CV–n c ,

(2.3)

where n, a so-called coarsening exponent, varies between 0.4 and 0.2 for various aluminum alloys. Figure 2.3b illustrates this relationship for binary Al–Cu alloys. This unique correspondence between the secondary dendrite arm spacing and the cooling rate (which is an important technological parameter) has paramount importance and allows us in many cases to estimate the cooling rate based on metallographic examination of the as-cast structure. However, it is virtually impossible to single out secondary dendrite arms on cross-sections of real as-cast samples, like those shown in Figure 2.1. Instead, the average thickness of all dendrite branches is measured. Numerous experimental studies conirm that Equation 2.3 is also valid for the average dendrite arms’ thickness or spacing. The relationships for the grain size and the dendrite arms spacing can look similar (as Equation 2.3) providing that the solidiication conditions are all the same, except for the cooling rate. Our results for Al–Cu alloys show that the exponent n is close to 0.3 for the grain size and to 0.4 for the dendrite arms spacing, and the coeficient C decreases with the increasing copper concentration [3]. It is important to note here that the cooling rate Vc and the solidiication rate Vs are, in general, proportional to each other but not always directly. In the case of DC casting, the solidiication rate is equal to the velocity of the solidiication front and depends on the casting speed as shown by Equation 1.2. The cooling rate for a structure found in a particular section of the billet is a function of its solidiication history, as will be discussed in more detail in Chapter 3. In addition to the cooling rate, the superheating of a melt over the liquidus temperature is an important process parameter when preparing and casting alloys. The melts of commercial aluminum alloys are seldom heated over 800°C because a higher superheating can result in structure coarsening, development of columnar grain structure, and increased saturation of the melt with gas. In the 1950s some scientists and engineers involved in the production of foundry aluminum alloys observed interesting effects from melt superheating. They found out that after a stage of structure coarsening upon increasing the melt temperature, the structure tends to reine. Later, there were many studies and attempts to explain this phenomenon. First, some anomalies in the temperature dependences of viscosity and density were observed upon heating and cooling of the melt [6, 7]. Then, with the development of new techniques of X-ray and neutron diffraction, the short-range

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

25

order in aluminum melts was revealed and interpreted [6, 8]. Some scientists explain the phenomenon of structure reinement at high superheating from the viewpoint of changes in the melt structure. There are hypotheses about the melt structure, phase transformations in the liquid phase, and even about quasicrystalline or polycrystalline constitution of melts [6–8]. Other specialists explain these anomalies in terms of impurities in aluminum and insuficiently correct experiments [9]. There is also a viewpoint that the spontaneous solidiication of homogeneous melt can occur, showing the direct dependence of the grain size on melt undercooling [10, 11]. Most commercial aluminum alloys belong to the eutectic type with respect to main alloying components. The solidiication of such aluminum alloys (except for purely eutectic alloys) starts with the formation of primary phases. In this case, the primary phase is considered to be a phase of the main alloying system that is the irst to crystallize. In hypoeutectic aluminum alloys, this phase is the aluminum-based solid solution. In most cases the so-called primary solidiication occurs heterogeneously. The main primary phase can nucleate onto intermetallic particles formed by specially added grain reiners (Ti, Zr, Sc, Mn, etc.), on active or activated impurities (alumina particles with remains of the solid phase in surface defects), and on clusters of the melt with the short-range order similar to the crystal structure of the main primary phase. We should not overlook the possibility that physico-chemical properties of some inert impurities can change with temperature, and these impurities can become active with respect to the melt, which is what happens with alumina that undergoes polymorphic transformation at high temperatures and becomes wettable [12]. Figure 2.5 shows the experimental results on the structure change with an increasing pouring temperature. In pure aluminum, the equiaxed grain coarsens in the temperature range up to 950°C, then its size stabilizes and somewhat decreases upon heating above 1000–1050°C; see Figure 2.5a. The columnar zone narrows while casting from temperatures above 950°C. Apparently, the peculiar temperatures of these effects depend on the purity of the aluminum and the cooling rate. This behavior can be explained by the effect of alumina ine particles that are inert with respect to liquid aluminum at temperatures below 950°C and then become wet from aluminum and act as solidiication nuclei at higher temperatures [12]. These processes affect the re-coarsening of the structure in a 2024 alloy cast from very high temperatures; see Figure 2.5b. The 2024-type alloy is the most common of the widely used wrought aluminum alloys. Its structure is predominantly equiaxed due to the small additions of Ti and Zr; see Figure 2.5c. The grains of aluminum solid solution grow markedly (from 63 to 622 µm) as the pouring temperature increases from 700 to 980°C, then the size of equiaxed grains essentially decreases (to 332 µm at 1030°C) and coarsens again on further heating; see Figure 2.5b,c. An examination of the microstructure shows that the growth of the grain size is accompanied by the development of dendritic branching with relevant reinement of the average section of dendritic arms. The grain reinement achieved during high-temperature pouring results in the decreased dendritic branching.

Copyright 2008 by Taylor and Francis Group, LLC

14

1.2

12

1.0

10

1.8

8

6 0.6 600 700 800 900 1000 11001200 Temperature of pouring, °C (a)

Size of grains and dendritic cells, µm

: Size of equiaxed grains, mm

1.4

: width fo columnar zone, mm

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

26

1000

8 6 4 2

100

Size of grains

8 6 4

Size of dendritic cells

2

10

650

(b)

750 850 950 1050 1150 Temperature of pouring, °C

200 µm (c) FIGURE 2.5 Effect of melt superheating on the structure of pure aluminum (a) and a 2024 alloy (b, c) [14]. Cooling rate during solidiication 5.5 K/s. (Reproduced with kind permission of Carl Hanser Verlag.)

We can say that there is a characteristic melt temperature typical of each alloy starting from which the grain structure or primary intermetallics start to reine. This temperature depends on the cooling rate: the higher cooling rate, the lower the temperature [13]. The occurrence of an additional high-temperature heterogeneity factor can delay or even suppress the reinement [14]. The observed phenomena can be explained, in our opinion, based on the following ideas [13, 14]: • The melt is a heterogeneous liquid in the particular temperature range above the liquidus: in other words, the melt contains various ready-to-solidify sites that become nuclei at small undercooling (these are the remains of solid phase in defects of nonmetallic particles and remains of intermetallics that need some time and thermodynamic stimulus to dissolve).

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

27

• Upon increasing the melt temperatures these structural and compositional heterogeneities dissolve: this process requires melt superheating about 200–250°C. • At higher temperatures the melt can be considered as homogeneous liquid. • The formation of new solidiication nuclei upon undercooling occurs in a way similar to the decomposition of supersaturated solution. These new nuclei are represented mainly by aluminides of transition metals in hypoeutectic aluminum alloys. The inal structure depends on the composition (undercooling and corresponding supersaturation) and sequence of phase precipitation (solidiication). All processes determining the structure formation upon solidiication from different melt temperatures are summarized in Figure 2.6. Generally, the primary structure constituents coarsen due to the gradual decrease in the amount of ready nuclei. Starting from a certain temperature, the liquid becomes homogeneous and tends to undercool before solidiication. The higher the melt temperature and the cooling rate, the greater the undercooling and the supersaturation of the liquid solution. The greater the undercooling, the larger the amount of new nuclei and the i ner the grain or primary intermetallics. The increased amount of solidiication nuclei, the narrowed two-phase zone, the increased temperature gradient in the melt, and the decreased

Overheating of the melt Increased temperature gradient in the melt Narrowed solidification zone

Deactivation of impurities

Grain growth Columnar structure

Dissolution of intermetallic remains

Increased thermal undercooling

Increased amount of active solidification sites

Increased diffusion coefficient

Increased content of alloying elements in the melt

Decreased constitutional undercooling Arrested dendritic branching

Increased supersaturation of the liquid solution (melt)

Higher fineness and precipitation density of products of decomposition of the supersaturated liquid solution Formation of fine equiaxed grains with slight dendritic branching

Refinement of primary structure constituents (grains)

FIGURE 2.6 Processes occurring during melt superheating and relevant structure changes [13, 14]. (Reproduced with kind permission of Carl Hanser Verlag.)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

28

constitutional undercooling result in the formation of equiaxed grain structure with only slight dendritic branching. This occurs when the homogeneous state of the liquid can be achieved. Active inhomogeneities, such as wet alumina particles, prevent large undercooling and promote nucleation of few grains at high temperatures, effectively coarsening the structure; see Figure 2.5b. The commercial application of melt superheating as a means of structure reinement in conventional ingot and shape casting is impractical because of the temperature limitation of furnaces and increased gas absorption and porosity. In conventional casting practice only the coarsening stage is usually observed.

2.2

Microsegregation in Aluminum Alloys

The occurrence of additional solidiication reactions and the formation or retention of additional phases during nonequilibrium solidiication upon direct chill casting are caused, in most cases, by the phenomenon called microsegregation.* Microsegregation means the inhomogeneity of the chemical composition on the scale of a single grain (dendrite). The origins of microsegregation are well known. All phase transformations during solidiication occur by diffusion, and therefore require some time to be accomplished. The degree of this accomplishment is controlled by the following three diffusion processes: • Diffusion of alloying elements in the bulk liquid toward and from the solid–liquid interface in order to create the equilibrium difference of concentrations according to the difference in concentrations in the liquid and solid phases relected by the liquidus and solidus lines in the equilibrium phase diagram (see Figure 2.4). • Diffusion in the liquid close to the solid–liquid interface in order to dissipate the solute (partition coeficient K < 1) or solvent (K > 1) pile-up at the solidiication front (Figure 2.4). • Diffusion in the solid phase in order to equilibrate the solute concentration within the solid grain. If all three diffusion processes proceed fully during solidiication, then the solidiication occurs according to the equilibrium phase diagram, with the average compositions of solid and liquid changing according to the solidus and liquids lines, respectively. In most cases, even at relatively low cooling rates, these diffusion processes are to some extent incomplete, which results * Incomplete peritectic reactions and corresponding retention of high-temperature phases in the as-cast structure are usually not considered as microsegregation, but are also a result of hindered diffusion.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

29

in the incomplete peritectic reactions, change of the average compositions of the liquid and solid phases, appearance of excess phases, and inhomogeneous distribution of alloying elements in the volume of the grain. One of the main approaches to treat microsegregation was suggested by Gulliver in 1913–1922 and developed by Scheil in 1942. These scientists assumed that the diffusion in the liquid phase was complete whereas the diffusion in the solid phase was fully suppressed, and the conditions of local equilibrium were preserved at the solid–liquid interface. For the cooling-rate range of commercial shape and direct chill casting, these assumptions are frequently believed to be valid. For the sake of simplicity, we consider in this section mainly cases of binary alloys, though the same mechanisms act in more complex and commercial alloys. The Gulliver–Scheil approximation can be written in the following form for the one-dimensional case: CL = C0(1 – fs)(K–1),

(2.4)

where CL is the composition of the liquid phase, C0 is the nominal alloy composition, fs is the fraction of solid, and K is the partition coeficient (K = CS/CL ). It is important to note that the Gulliver–Scheil model presents the limited case when no diffusion occurs in the solid phase, and it does not take into account either morphology of real structure or phase transformations occurring during solidiication. In principle, the solidiication occurs within this approximation as follows (Figure 2.7): • The liquid phase changes its composition during primary solidiication according to the liquidus line, from C0L to CFL and eventually to CE (which is the eutectic concentration). • The composition of the solid phase at the interface with the liquid is according to the equilibrium solidus at each temperature in the primary solidiication range and is locally preserved at lower temperatures. • Therefore, the solute distribution in the solid phase changes from C0S in the center of the grain (dendrite) to CFseq and, eventually, to Climit (which is the limit solubility of the solute in the matrix) at the periphery of the grain (dendrite). • The average composition of the solid phase CSav changes with the decreasing temperature from C0S to CFsneq. As a result of this nonequilibrium solidiication, the solid phase at the moment when the alloy reaches the equilibrium solidus contains less solute than the nominal alloy composition C0. Therefore, there is still liquid existing in the system to accommodate the remaining solute. This liquid continues to solidify according to the equilibrium liquidus. It may happen that the difference between the alloy composition and the actual average

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

30

660

C0

Temperature, °C

C0s

TL; C0L

620

580

CSav

CFSeq

TS; CFL

TSneq TE

CE

CFSneq 540 0

10

20

30

Cu, wt% FIGURE 2.7 A scheme of nonequilibrium solidiication as applicable to binary Al–Cu alloys. C0 is the Al–4% Cu alloy; TL is the equilibrium liquidus temperature; TS is the equilibrium solidus temperature; TE is the equilibrium eutectic temperature; C0S is the initial concentration of solid (C0S = KC0L); CFL and CFSeq are the equilibrium liquid and solid concentrations at the end of the equilibrium solidiication and the liquid and solid concentrations at the interface at TS; CFSneq is the average concentration of the solid phase at the eutectic temperature TE; CSav is the line of the average solid concentration; and TSneq is the experimental nonequilibrium solidus [17].

composition of the solid phase will be so large that the solidiication process will continue until the eutectic temperature is reached. At this moment, the Gulliver–Scheil equation just shows the amount of the remaining liquid phase of the eutectic composition (Figure 2.8). This liquid will solidify as the eutectic, forming the “nonequilibrium eutectic” or “nonequilibrium phase” (Al2Cu in the case of Al–Cu alloys), which will be discussed in the next section of this chapter. The reality of solidiication is obviously much more complicated than the Gulliver–Scheil approximation. Measurable diffusion of solute occurs in the solid phase during and after solidiication, diffusion in the liquid phase may be hindered at high cooling rates, dendritic grains evolve during solidiication with dissolution of some i ner arms, thermo-solutal convection may cause local changes in the composition with correspondingly

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys 100

31

Eutectic reaction

90 80

Fraction solid, %

70 60 50 40 30 20 10 0 540

560

580

600

620

640

660

Temperature, °C FIGURE 2.8 Variation of the volume fraction of solid during nonequilibrium solidiication of an Al–4% Cu alloy. About 8 vol.% of liquid remaining at the eutectic temperature will form the eutectic.

accelerated or slowed-down local solidiication, and the temperatures of liquidus and solidus (eutectic reaction) may be lower than the equilibrium ones as a result of undercooling during solidiication. All these phenomena occurring simultaneously and in different proportions can dramatically change the distribution of the solute in the solid grain and the ratio of phase fractions. One of the most important external parameters that affects microsegregation is the cooling rate during solidiication. Experimental results on the inluence of the cooling rate on the extent of microsegregation are seemingly controversial, as illustrated in Table 2.1 by the variation in the amount of nonequilibrium eutectics. The most frequently cited explanation of dependence is the one published by Sarrial and Abbaschian [15], shown in Figure 2.9a, which agrees in the range of moderate cooling rates with much earlier results by Michael and Bever [16]. The ascending shoulder of the dependence is conventionally explained by an increasing degree of microsegregation due to the limited diffusion in the solid; and the descending shoulder is explained by the limited diffusion in the liquid at very high cooling rates. It is interesting to note, however, that in this dependence the range of cooling rates below 1 K/s has not been well researched. Novikov and Zolotorevsky [17] performed thorough experiments by cooling the samples at different rates

Copyright 2008 by Taylor and Francis Group, LLC

32

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

TABLE 2.1 Dependence of the Amount of Nonequilibrium Eutectics on the Cooling Rate Reported in Different Sources (as compiled in [3]) Alloy Composition, wt%

Cooling Rate Range, K/s

Variation in the Amount of Nonequilibrium Eutectics with Cooling Rate

2Cu; 3Cu; 4Cu; 4.8Cu 1Cu; 3Cu; 4.5Cu

0.01; 0.8; 5; 50 1–38

Increase Increase

2.8Cu; 4.9Cu; 1Si 2Cu; 5Cu; 6Mg; 1.4Mn; alloy 2024 5Cu 7Mg; 11Mg 5Mg (alloy 5182) 7.5Si–0.45Mg (alloy 356) 0.94Mg–4.11Cu 0.87Mg–5.07Cu Al–Cu–Mg (alloy 2024); Al–Mn–Mg (alloy 3004) Steel (1.25C. 1.06Si, 6.6Mn, 1.06Al, etc.)

0.1–37000 0.001–4 1; 3 0.5–10000 0.5–2 0.25–1.5 1.3–21.3 0.9–18.7 0.05–8.5

Growth rate during directional solidiication 7–450 µm/s

Increase to 190 K/s, then decrease Initial increase to 0.002–1 K/s followed by decrease Increase Increase Decrease** Increase** Initial increase followed by decrease Increase Decrease

Initial increase followed by slight decrease

Comments* RT End-chill casting DS Q Calculated RT RT RT Q Q DS

Q

* RT = cooling was not interrupted until the room temperature is reached, DS = directional solidiication, and Q = alloy was quenched after the end of solidiication. ** Estimated from the fraction of solid at which the eutectic reaction starts.

and quenching them at the eutectic temperature to reduce the effect of structure homogenization. They showed for aluminum- and copper-based binary and commercial alloys that the dependence of the nonequilibrium eutectics on the cooling rate is quite special in the range of low cooling rates, the one that had not been thoroughly examined by other researchers. An example is given in Figure 2.9b for an Al–5% Cu alloy, showing the distinct peak at a low cooling rate. This dependence cannot be easily explained without taking into account the complexity of the microsegregation phenomenon. Recent results obtained on Al–Cu alloys conirm the dependence reported by Novikov and Zolotorevsky, as shown in Figure 2.10 [3, 18]. Several mechanisms mentioned previously can contribute to this phenomenon. An attempt to reproduce the observed behavior using one of the known models of microsegregation failed. The Brody–Flemings, Alstruc, and Pseudo Front Tracking (PFT) models predict the increase of the nonequilibrium eutectic fraction with the increasing cooling rate.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

33

Relative eutectic fraction

1

0.8

0.6

0.4

0.2 0

−3

−2

(a)

−1

0 1 2 3 4 Log of the cooling rate, K/s

5

6

Volume fraction of eutectics, %

8 7 6

0.3 (b)

0.7

1.0

1.3

1.7

3.3

Cooling rate, K/s

FIGURE 2.9 Effect of cooling rate on the degree of microsegregation in an Al–5% Cu alloy expressed by the amount of nonequilibrium eutectics: (a) after [15]; (b) after [17].

The Brody–Flemings model correlates the composition of the liquid at the solid–liquid interface to the solid fraction and the so-called Fourier number that is deined as α′ = 4DStf/d2,

(2.5)

where DS is the diffusion coeficient of the solute in the solid phase, tf is the local solidiication time, and d is the dendrite arm spacing. It is assumed that the contribution of the solid diffusion (so-called back diffusion) to the microsegregation will be appreciable if the dimensionless parameter α′K ≥ 0.1. For the cooling rates ranging from 0.1 to 10 K/s, the calculated solid-state Fourier numbers are at the order of 10 –2 and decrease very slowly as the cooling rate increases [3]. It is not surprising that the Brody–Flemings model cannot produce the trend observed experimentally. The Alstruc model is a 1D model of solidiication and homogenization that takes into account phase transformations during solidiication and considers

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

34

Non-equilibrium eutectic fraction, %

4

3

2

1

0

Non-equilibrium eutectic fraction, %

10

1

0 (a)

Cooling rate, K/s 3.5 3 2.5 2 1.5 1 0.5 0 0.1

(b)

1

10

100

Cooling rate, K/s

FIGURE 2.10 Effect of cooling rate on the amount of nonequilibrium eutectic in Al–Cu alloys: (a) Al–1.8% Cu and (b) Al–2.5% Cu, grain rei ned. Samples were quenched at the temperature of nonequilibrium solidus (eutectic).

the structure as a cylindrical dendrite arm, as described in detail elsewhere [19]. This model also cannot reproduce the experimentally observed trend. The Pseudo Front Tracking model solves the diffusion equations both in the solid and the liquid numerically and allows one to use a temperaturedependent diffusion coeficient and a non-linear phase diagram as the input parameters [20]. In a 1D model of the PFT model, calculations performed with the domain size equal to the half of the dendrite arm spacing show that, for

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

35

cooling rates of 0.4 and 13 K/s, the calculated eutectic fraction still increases with the cooling rate (2.72 and 3.22 wt%, respectively). However, if the calculation domain size is artiicially decreased (13 µm instead of experimentally measured 20 µm), then the calculated eutectic fraction (2.68 wt%) is indeed less than the calculated result for a cooling rate of 0.4 K/s (2.72 wt%). This artiicial calculation supports the explanation of Novikov and Zolotorevsky [17] as shown in Figure 2.11a: more solute remained in the primary phase at a higher cooling rate as a result of eficient diffusion in the solid contributed by a higher solid concentration gradient. That means that if microstructure is ine enough and the corresponding solute gradients are high, the solid-state (back) diffusion is so eficient that it decreases the degree of microsegregation. Upon slow cooling the degree of microsegregation is also low because of a longer solidiication time and correspondingly more time that is available for diffusion. In both cases, the fraction of nonequilibrium eutectics can be lower than at some intermediate cooling rates. Although this artiicial calculation reveals the important effect of the iner microstructure on the inal eutectic fraction, it cannot be directly used to explain the observed trend due to the deliberately decreased calculation domain size. Yet another indication of the role of dendrite ineness on the degree of microsegregation is a simple exercise in normalizing the experimentally measured fraction of nonequilibrium eutectic to the experimentally determined dendrite arm spacing, as shown in Figure 2.11b. The dependences appear to be linear and the normalized value of the eutectic fraction increases with the increasing cooling rate. These results show that the tested 1D models cannot adequately reproduce experimentally observed dependence. It should be noted that none of the existing models of microsegregation includes all the known effects that may occur during solidiication, namely, limited diffusion in the solid phase, back diffusion in the solid during solidiication, temperature dependences of diffusion coeficients in the solid and liquid, structure reinement with increasing cooling rate, dendrite coarsening with dissolution of iner branches, undercooling of the eutectic reaction with the shift of the eutectic point, thermodynamics of alloy solidiication, and solute accumulation in the liquid with mixing dependent on the diffusion rate. In addition, these models do not include the geometry of the dendritic structure. The scale of the structure is usually taken into account only as the secondary dendrite arm spacing or the grain size. Three factors may be quite important in the understanding of the experimentally observed phenomenon. First, the coarsening of the dendritic arms, which means that ine branches formed at higher temperatures and containing less solute dissolve and coarser branches thicken by “attaching” the solid phase containing more solute. As a result, the surrounding liquid is diluted of the solute, and the amount of nonequilibrium eutectic eventually decreases. This process should be active in the medium range of cooling rates (at very low cooling rates the branches are too thick to be dissolved) and the effect has to increase with the increasing cooling rate. Second, the undercooling of the eutectic reaction with the shift of the eutectic point to a higher solute

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

36

Concentration profiles in 1D PFT calculations 0.04 Slow cooling (0.4 K/s) Fast cooling (13 K/s)

0.035

Cu atomic fraction

0.03 0.025 0.02 0.015 0.01 0.005 0 0

0.1

0.23 0.46 0.69 0.77 0.84 0.92 0.95

(a)

1

Solid fraction

0.16

NEqE/d, %/µm

0.12

0.08

0.04

0.00 0 (b)

4

8

12

16

Cooling rate, K/s

FIGURE 2.11 (a) Accumulation of copper in the primary (Al) phase during solidiication of an Al–1.8% Cu alloy calculated using a 1D PFT model and (b) dependence of the nonequilibrium eutectic fraction (NEqE) normalized to the dendrite arm spacing (d) on the cooling rate during solidiication of Al– Cu alloys (experimental data, see symbols in Figure 2.3) [3]. (Reproduced with kind permission of Elsevier.)

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

37

concentration and the increased concentration of the solute in the solid phase at the solid–liquid interface under conditions of the hindered diffusion in the solid should result in the decreased amount of nonequilibrium eutectics. Third, the structure reinement with increasing cooling rate may result in more active back diffusion with the corresponding decrease in the amount of nonequilibrium eutectics, as has been shown by the 1D PFT model. A recently developed 2D PFT microsegregation model can predict the grain morphology of the primary phase and its effect on secondary phase formation [20]. Rather complex and time-consuming 2D PFT calculations that use the measured grain sizes as input parameters of calculation domain size to reproduce the dendrite arm spacing observed in the measurements are, in our opinion, required for taking into account the effect of grain morphology on the formation of eutectic fraction. Before discussing the outcome of the 2D PFT simulation results, let us consider separately the possible contribution of dendrite coarsening to the eutectic volume fraction. A model suggested by Voller and Beckermann [21] is able to predict the inal eutectic fraction as a function of the coarsening exponent. This model assumes that the dendrite arm spacing is not constant during solidiication, but is a function of solidiication time in the following form: d(t) = d0(t/tf)n,

(2.6)

in which d(t) is the dendrite arm spacing during solidiication, d0 is the inal dendrite arm spacing, t is the running solidiication time, tf is the total local solidiication time, and n is the coarsening exponent. The details of the calculation can be found elsewhere [18]. The results are shown in Figure 2.12.

0.05

Fast cooling (16 K/s) Slow cooling (0.8 K/s)

Eutectic fraction

0.04 0.03 0.02 0.01 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Coarsening exponent, n FIGURE 2.12 The mass eutectic fraction in an Al–2.5% Cu alloy as a function of the coarsening exponent as predicted by the approximate microsegregation model [21] for two given cooling rates [18]. (Reproduced with kind permission of Elsevier/Pergamon.)

Copyright 2008 by Taylor and Francis Group, LLC

38

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

It is interesting to note that when coarsening is absent, i.e., n = 0, the calculated eutectic fraction is almost double the one calculated with n = 0.33, which is regarded as a default value for aluminum alloys [1]. According to this model, coarsening indeed makes a difference in the inal eutectic fraction formed. For the same coarsening exponent, the eutectic fraction predicted for a faster cooling is always higher than the one obtained for a slower cooling. This result seemingly contradicts experimental observations in Figure 2.10. But should the coarsening exponent be held as the same value for fast and cooling solidiication conditions? In other words, does the coarsening exponent depend on cooling rate? Diepers et al. [22] showed numerically that the change of coarsening conditions from diffusion to convection controlled increased the coarsening exponent from 0.33 to 0.5. Recently, it has been demonstrated experimentally that the coarsening exponent indeed increases for a Al–4.5% Cu alloy from 0.3 to 0.5 in the presence of forced low [23]. In other in situ experiments on an Al–30 wt% Cu alloy, the analysis of dendrite coarsening shows that the coarsening behavior during solidiication is affected by natural luctuations in the solidiication front velocity that occur due to accumulation and settlement of solute-rich liquid [24]. The coarsening exponent is found to be equal to 0.4 during front deceleration and to vary between 0.28 and 0.5 upon acceleration of the solidiication front. It is argued that solute accumulation and drainage within the mushy zone may have a signiicant effect on the kinetics of solidiication, being the function of the dendrite geometry. A greater coarsening is associated with a higher solute accumulation caused by a lower mush permeability. Given the complex circumstances in which coarsening may occur during solidiication and different coarsening mechanisms (local melting, ripening, coalescence, and fragmentation), it seems there is no reason to restrict the coarsening exponent to a constant value. If in the fast cooling case the coarsening exponent is slightly higher than the one at a slower cooling rate, then the experimentally observed tendency could be explained, as illustrated in Figure 2.12. But certainly this explanation is not complete without a physically based reason to adjust the coarsening exponent. Since the 2D PFT model can release the grain morphology restriction made in the 1D PFT model and track the grain morphology evolution and dendrite arm coarsening, the corresponding simulation results can shed the light on the development of the structure and local composition during solidiication. Two cooling rates were taken according to Figure 2.10b, i.e., 0.8 and 16 K/s. Indeed, the application of the 2D PFT model indicates the trend in the cooling-rate dependence of the eutectic fraction that agrees well with that observed in the experiments; see Figure 2.13a. Although the amount of nonequilibrium eutectic is overestimated, the difference can be due to two factors: (1) the model does not take into account the undercooling of the eutectic reaction and (2) the real structure develops in the 3D space rather than in two dimensions. What is even more important is that the computational results allow us to follow the evolution of the structure, which can be instrumental in understanding the observed phenomena.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

39

Volume fraction of eutectic (%)

4.5 4 3.5 3 2.5 2 1.5

2D

E

1D

E

2D

1D

1 0.5 0 Slow cooling (0.8 K/s)

Fast cooling (16 K/s)

7.2E+04

120

6.3E+04

105

5.4E+04

90

4.5E+04

75

3.6E+04

60

2.7E+04

0.8 K/s

45

16 K/s

1.8E+04

30

9.1E+03

15

1.0E+02 0.00

0.20

(b)

0.40

0.60

0.80

DAS (µm)

Sv (1/µm)

(a)

0 1.00

Solid fraction

FIGURE 2.13 (a) Volume fraction of nonequilibrium eutectic predicted by PFT modeling (1D, 2D) and observed experimentally (E) in an Al–2.5% Cu alloy and (b) development of interfacial concentration (Sv) and dendrite arm spacing (DAS) during solidiication of an Al–2.5% Cu alloy as calculated by the 2D PFT model for two cooling rates [18]. (Reproduced with kind permission of Elsevier/Pergamon.)

Fast cooling leads to more dendritic grain morphology compared to slow cooling. The direct consequence of such a structure reinement in the fast cooling case is the higher interfacial concentration value (solid–liquid interface area per unit volume, or solid–liquid interface length per unit area); see Figure 2.13b. An immediate result of this could be the reduced extent of microsegregation in the faster cooled structure by the means of back diffusion. The comparison of Fourier numbers calculated by setting characteristic solidiication length to the dendrite arm spacing obtained from the completely solidiied microstructure for both cases (0.009 and 0.0165 for the

Copyright 2008 by Taylor and Francis Group, LLC

40

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

fast and slow cooling, respectively) tells us, however, that the inal structure reinement is less effective in inluencing the microsegregation than the decreased solidiication time. Therefore, the structure reinement due to the increased cooling rate cannot be the main contributor to the observed trend in the eutectic fraction variation. Another phenomenon that occurs at later stages of solidiication, approximately at volume fraction solid above 70%, is the coalescence of dendrite arms. At the beginning of solidiication when the solid fraction is low, dendrite arm spacings are quite large, relecting the fact that side branching has not yet started. The dendrite arm spacing decreases as the solidiication progresses, which indicates that dendrite multiplication is dominant over the ripening. With the further increase of the solid fraction, dendrites begin to impinge onto one another, and ripening and coalescence take over the side branching. As a result, the dendrite arm spacing starts to increase; see Figure 2.13b. This process works more eficiently for the iner structure, i.e., at a higher cooling rate, as demonstrated by the variation of the interfacial concentration. This can be readily observed on simulated microstructures shown in Figure 2.14. In the fast cooling case, tertiary arms are formed and they experience quick and severe remelting and re-solidiication (ripening). The surviving tertiary arms inally coalesce. In the slow cooling case, only secondary arms are induced, and all of them survive. The inal volume fraction of eutectic is, of course, a measure of the amount of liquid that is retained, due to the nonequilibrium regime of solidiication, at the temperature of the eutectic reaction. It is interesting that coarsening starts to affect the volume fraction of liquid at rather high temperatures, at the liquid fraction of about 0.46 [18]. Compared to the slow cooling rate, the severe coarsening in the fast cooling case implies that fast cooling must be accompanied by a higher coarsening exponent, as dei ned in Equation 2.6. This implication is supported by tracing how the dendrite arm spacing evolves in later stages of solidiication in Figure 2.13b, where ripening occurs and Equation 2.6 can be applied. As solid fraction increases, the difference between dendrite arm spacings in the two given cases decreases, which means that in the fast cooling case dendrite branches coarsen faster than in the slow cooling case. Therefore, the difference in coarsening kinetics at different cooling rates can produce the results that are observed experimentally, as suggested by Voller’s model. Another important factor that can be estimated using the 2D PFT modeling is the enrichment in solute of tertiary arms as compared to secondary arms. The enrichment is logical because the n-order arms are generated at lower temperatures. What is more interesting is that the solute concentration on the tertiary arms keeps increasing during solidiication because of the back diffusion while that in the secondary arms remains virtually unchanged due to a longer diffusion distance [18]. This enrichment in the core of tertiary arms results in general enrichment of the solid phase in the solute and, as a result, in a lesser nonequilibrium eutectic fraction.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

41

FIGURE 2.14 Structure development during solidiication of an Al–2.5% Cu alloy as modeled by the 2D PFT model [18] (a, b, d, e) and observed experimentally (c, f). a–c, 16 K/s; d–f, 0.8 K/s. a, d, 50% solid phase; b, e, 94% solid phase; c, f, 100% solid phase. The domain size is 330 × 330 µm in (a–c) and 524 × 524 µm in (d–f). Black ield in (a, b, d, e) is the solid phase; black ield in (c, f) is the nonequilibrium eutectic. (Reproduced with kind permission of Elsevier/Pergamon.)

Copyright 2008 by Taylor and Francis Group, LLC

42

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

The correct trend in the dependence of the eutectic fraction in the range of moderate cooling rates predicted by 2D PFT calculations is indicative of the important role played by the structure morphology. It is clear that 3D modeling could provide even better correspondence with the reality, though such a simulation would be very computer demanding. In situ observations of the structure evolution during solidiication provide valuable insight into coarsening kinetics. Recently, such experimental results started to emerge thanks to the new technique of X-ray microscopy and tomography. Figure 2.15a demonstrates signiicant coarsening of secondary dendrite arms accompanied by dissolution and fragmentation during primary solidiication in Al–Cu alloys [25]. A similar process of coarsening, but in 3D, is shown in Figure 2.15b for an Al–Si–Cu alloy [26]. It will be quite exciting to further employ this technique for studying the coarsening kinetics, since the conventional solidify-and-quench technique can only produce the inal microstructure, and cannot show the solidiication process.

1200

µm

1000

800

600

(a)

0 100 200 300 400 500 600 700 µm

50 µm (b) FIGURE 2.15 In situ observation of structure evolution during solidiication. (a) 2D columnar growth in an Al–30% Cu alloy [25]. The sequence is courtesy of Dr. R. Mathiesen (SINTEF, Norway) and Prof. L. Arnberg (NTNU, Norway). (b) 3D images of coarsening by dendrite-arm coalescence in an Al–Si–Cu alloy [26]. The images are courtesy of Prof. M. Suéry (INPG, France). Growth occurs from the bottom to the top; three sequential stages are shown. Note the coarsening of dendrite arms in frames.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

43

There is one more phenomenon that occurs in real solidiication of multicomponent alloys: local change of composition and solidiication path [27, 28]. The local liquid enrichment as compared to the local equilibrium conditions results in remelting of the solid phase, and the local liquid depletion causes accelerated freezing. The consequence of this process is the formation of less primary phase in the former case and more primary phase in the latter case. Hence, the local depletion of the liquid of the solute may result in the formation of less eutectic phase.

2.3

Solidification Reactions and Phase Composition

The structure and, ultimately, the mechanical and technological properties of the as-cast material depend to a considerable extent on the solidiication reactions and the selection of phases that occur during casting. Most commercial alloys contain several alloying elements and impurities, which makes their phase composition rather complex. In addition, solidiication during real casting does not follow the path of equilibrium solidiication that can be derived from the phase diagram. Additional reactions proceed and excess phases are formed under conditions of nonequilibrium solidiication. In this section we briely consider the solidiication and phase composition of main groups of commercial wrought alloys. For more detailed treatises of this subject the reader is referred to more complete accounts such as Refs. [29–32]. 2.3.1

Commercial Aluminum

Low alloyed aluminum, often called commercial aluminum or 1XXX-series alloys, belongs to the Al–Fe–Si phase diagram. This phase diagram is very complex, as shown in Figure 2.16a. The ternary phases in the solid state exist mainly outside the ields of their primary crystallization; therefore, numerous peritectic reactions should be completed for the equilibrium to be achieved. As a result, real alloys produced at commercial cooling rates can have the Al3Fe, Al6Fe, α(Al5FeSi), β(Al8Fe2Si) and δ(Al4FeSi2) phases co-existing in their structure [29]. Identiication of the phases based only on their morphology can often lead to a mistake because the same phase can have different morphologies depending on its origin: primary crystals or products of peritectic and eutectic reactions. In addition, silicon and other stable, metastable, and nonequilibrium binary and ternary phases can precipitate during the decomposition of supersaturated solid solutions in the process of homogenization or upon cooling of ingots or billets. Some of the phases are also known to undergo transformations during heat treatment. The Al–Fe–Si system is among the most important for the analysis of as-cast structure of aluminum alloys, and there are quite a few works suggesting its nonequilibrium variants. For example, the phase-ield distribution in

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

44

(Al)+Al3 (Al)+Al3+Al8

8 Si,%

576

6

590

600

630

4

Al3Fe, %

1.5 1.0

E (Si)

10 e2 12

(Al)+Al3+Al8+Al5+(Si)

(Al)+Al8+ Al5

0.5

600

610

2

Al4

640

P2 620

650

0

Al5

Al8 630 (Al)

1

680

650

P1 640

Fe,%

690 670

680

2 e1

(a)

Al3

710

3

720

710

740

(Al)+Al3+Al8+Al5

2.0

(Al)+Al5

14 (b)

Al3 - Al3Fe; Al5 - Al5FeSi;Al8 - Al8Fe2Si; Al4 - Al4FeSi2

0

0.5

1.0 Si,%

(Al)+Al8+ Al5+(Si)

(Al)+Al5+(Si)

1.5

2.0

Al3 - Al3Fe; Al8 - Al8Fe2Si;Al5 - Al5FeSi

(Al)+Al3+ Al8+(Si)

Al3Fe,%

4 (Al)+Al3+ Al8

3

(Al)+Al8+ (Si)

2

(Al)+Al3

(Al)+Al8+ Al5+(Si)

1

(Al)+Al5+(Si)

(c)

(Al)+(Si)

2 3 4 Si,% Al3 - Al3Fe; Al8 - Al8Fe2Si; Al5 - Al5FeSi 0

1

FIGURE 2.16 Projection of the solidiication surface under conditions of nonequilibrium solidiication (a) and distribution of phase ields in the solid state after casting at a cooling rate 10–1 K/s (b) and 10 K/s (c) in Al–Fe–Si alloys [32]. (Reproduced with kind permission of Elsevier.)

the as-cast state given by Philips [33] shows 4- and 5-phase regions (Figure 2.16b), which is the most evident feature of the nonequilibrium structure. A shift in the primary solidiication ields of the Al3Fe, Al8Fe2Si, and Al5FeSi phases depending on the cooling rate during solidiication (Vc) is reported by Langsrud [34], who shows that these ields drift toward a lower Si concentration with the increasing Vc. As a consequence, the Al3Fe formation is less probable at high cooling rates, even in alloys containing 2–3% Fe at 2–3% Si. At low cooling rates (Vc1 = 10 –2 – 10 –1 K/c), the onset of solidiication can be analyzed with suficient accuracy by the equilibrium phase diagram; see Figure 2.16a. Binary eutectic reactions with participation of Fe- and Si-containing phases take place after the primary crystals are formed. However, due to the

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

45

inhibition of peritectic transformations, some alloys can simultaneously form all three binary eutectics with participation of the Fe-containing phases. The nonequilibrium solidus of most alloys is equal to 576°C and corresponds to the ternary eutectics L → (Al) + (Si) + Al5FeSi (point E in Figure 2.16a), and only the alloys whose compositions are within a narrow region close to the binary systems complete solidiication by the binary eutectic reactions. The inhibition of the peritectic transformations L + Al8Fe2Si → (Al) + Al5FeSi (P2 in Figure 2.16a) and L + Al3Fe → (Al) + Al8Fe2Si (P1 in Figure 2.16a) leads to the following result. With the Si concentration increasing, the sequence of phase regions (not taking (Al) into account) in slowly solidiied alloys containing more than 0.5% iron will be Al3Fe, Al3Fe + Al8Fe2Si, Al3Fe + Al8Fe2Si + Al5FeSi, Al3Fe + Al8Fe2Si + Al5FeSi + (Si), Al8Fe2Si + Al5FeSi + (Si); Al5FeSi + (Si). This is in very good agreement with the distributions of the phase regions proposed by Philips and shown in Figure 2.16b. At higher cooling rates, more relevant to DC casting (Vc2 = 100–102 K/c), a noticeable swing of the liquidus surface (Figure 2.16c) shifts the boundaries of intermediate reactions and phase regions in the as-cast state as compared with slower solidiication. Apart from these changes, the eutectic reactions L → (Al) + Al3Fe, L → (Al) + Al5FeSi and L → (Al) + (Si) + Al5FeSi are hindered at certain concentrations of Fe and Si. As a consequence, the Al8Fe2Si and (Si) phases are present and the Al3Fe and Al5FeSi phases are absent in the as-cast structure. Literature data indicate that, besides the stable phases, various metastable phases are formed in commercially pure aluminum and low-alloyed materials containing up to 0.5% Fe and 0.5% Si, at cooling rates typical of industrial casting; see, e.g., [35–37]. In the chemical composition, these metastable phases are close to the stable phases but differ in crystal structure (Table 2.2).

TABLE 2.2 Metastable Phases Occurring in Cast Al–Fe–Si Alloys [31] Lattice Parameters Structure

a, nm

b, nm

c, nm

β

1.256







1.23 1.26–1.3

— —

2.62 3.70

— —

β′, β* αv(Al14.8–12.4Fe3Si1–2.1) αγ(29% Fe, 5.5% Si) q2 19–27% Fe, 0–3% Si

Primitive cubic or bcc Hexagonal Tetragonal or orthorhombic Monoclinic Monoclinic Monoclinic Monoclinic Rhombohedral

29.2% Fe; 11.3% Si 19% Fe; 1% Si

Hexagonal Hexagonal

0.89 0.847–0.869 2.795 1.250 0.89 2.082 0.836 1.776

0.49 0.635 3.062 1.230 — — — —

4.16 0.610–0.632 2.073 1.970 — — 1.458 1.088

92° 93.4° 97.74° 111° 111.8° 95.2° — —

Phase α(Al12.7–12.9Fe3Si1–1.5) (27–35% Fe, 4–8% Si) α′(Al12–12.6Fe3Si1.6–2) α″, q1

Copyright 2008 by Taylor and Francis Group, LLC

46

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

The complete account of metastable phases occurring in Al–Fe–Si alloys can be found elsewhere [31]. Metastable phases can also precipitate from the aluminum solid solution during homogenization of ingots. Among these phases, metastable modiications of the equilibrium Al8Fe2Si (α, α′, α″) phase are well documented (Table 2.2). If the homogenization temperature is suficiently high, these metastable phases are, as a rule, transformed into respective equilibrium phases. One should bear in mind that, due to the low solubility of iron in (Al) (under typical casting conditions), the maximum amount of Fecontaining phases of secondary origin cannot exceed 0.1–0.2 vol.%. Free (Si) is rare as the Fe:Si ratio is usually maintained above unity in order to prevent hot tearing during casting, by avoiding the low-temperature eutectic reaction L → (Al) + Al5FeSi + (Si) (E in Figure 2.16a). The phase selection in as-cast commercial aluminum is a function of the Fe:Si ratio and the cooling rate. Table 2.3 shows experimentally observed temperatures during solidiication of a 1050 alloy containing 0.37% Fe and 0.05% Si [5]. The solidus temperature decreases with increasing the cooling rate reaching 630°C (close to point P1 in Figure 2.16a) at which the invariant peritectic reaction speciied by Bäckerud et al. in Table 2.3 should occur under equilibrium conditions. The formation of the metastable Al6Fe phase is possible in a 1050 alloy cast at cooling rates above 1 K/s [5]. 2.3.2

Wrought Alloys with Manganese (3XXX Series)

Analysis of Mn-containing alloys is quite complex and depends on the amount of alloying elements and impurities. For example, an 8006 alloy with a low content of Si impurity may contain Al6(FeMn) and Al3Fe phases formed upon secondary reactions during casting. Primary crystals of these phases are likely to be formed only at low cooling rates and high concentrations of iron. Silicon-containing 3009 and 3003 alloys contain the ternary Al15Mn3Si2 compound at a low Fe concentration in these alloys, by 60°C, Silicon dramatically, decreases the solidus of Al–Mn alloys. Typically, however, commercial alloys contain Mn, Fe, and Si together. As a result, in addition to binary phases, the

TABLE 2.3 Solidiication Reactions under Nonequilibrium Conditions in Commercial Aluminum Containing 0.37% Fe and 0.05% Si [5] Temperatures (°C) at a Cooling Rate Reaction L ⇒ (Al) L ⇒ (Al) + Al3Fe L+Al3Fe ⇒ (Al) + Al8Fe2Si Solidus

Copyright 2008 by Taylor and Francis Group, LLC

0.4 K/s

1.2 K/s

18 K/s

659 650 —

659 649 642–638

659 647 630

642

638

630

Solidification of Aluminum Alloys

47

TABLE 2.4 Solidiication Reactions under Nonequilibrium Conditions in a 3003 Alloy Containing 1.19% Mn, 0.55% Fe, and 0.18% Si [5] Temperatures (°C) at a Cooling Rate Reaction

0.5 K/s

17 K/s

L ⇒ (Al) L ⇒ (Al) + Al6(FeMn) L + Al6(FeMn) ⇒ (Al) + Al15(FeMn)3Si2 and/or L ⇒ (Al) + Al15(FeMn)3Si2 Solidus

655 653 641–634

655 646–615 589

634

589

complex Al15(FeMn)3Si2 phase is formed as a result on bi- and monovariant eutectic and invariant peritectic transformations during solidiication. Nonequilibrium solidiication further complicates the phase composition, which is due to the inhibition of invariant and monovariant peritectic reactions. Bäckerud et al. [5] experimentally observed the solidiication reactions given in Table 2.4 during nonequilibrium solidiication of a 3003 alloy. Note the considerable decrease in the solidus temperature with increasing the cooling rate. The as-cast structure contains Al6(FeMn) and Al15(FeMn)3Si2 phases. The amount of the latter phase increases with the silicon concentration. If the concentration of silicon is at the upper level (0.6%) and that of iron is at the lower level, the formation of free (Si) is possible with corresponding decrease of the solidus to 573°C (temperature of the monovariant eutectic reaction L → (Al) + (Si) + Al15(FeMn)3Si2). 2.3.3

Al–Mg–Si Wrought Alloys (6XXX Series)

Alloys of the 6XXX series would be easy to analyze if they did not contain other elements, apart from magnesium and silicon, capable of affecting the phase composition. However, this is not always the case, as these alloys usually contain Mn, Fe, and Si, and sometimes Cu. Nevertheless, the Al–Mg–Si phase diagram gives important information and it is appropriate to start the analysis of commercial 6XXX alloys with this diagram. Figure 2.17a shows the equilibrium phase composition (solidus surface) in the as-cast state. In the compositional range of 6XXX series alloys all phases that form during solidiication are of eutectic origin and are, generally, a result of nonequilibrium solidiication. If the ternary eutectics (Al) + (Si) + Mg2Si is present, the solidus is as low as 555°C. The presence of iron in 6XXX alloys results in the formation of different Fe-containing phases, i.e., α(Al8Fe2Si), β(Al5FeSi), Al3Fe, and π(Al8FeMg3Si6). In a 6003 alloy (0.8–1.5% Mg, 0.35–1% Si, up to 0.6% Fe, 0.8% Mn) the selection of phases depends on the ratio between alloying elements; see Figure 2.17b. At a high Mg:Si ratio, all iron should be bound to the Al3Fe

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

48

700

1.5

555 545

(Al)+Mg2Si+(Si) [555°C]

500

(Al)

+F+

α

0.88 0.8 533 518

(Al)+ α

(Al)+F+Mg2Si 0.74 500

0.23

1.0

1.5

2.0

3.0

2.5

3.5

4.0

(Al)+α+Mg2Si (Al)+α+β+Mg2Si (Al)+β+Mg2Si

0.68 0.8

0.12 0.34 Al−0.5%Si−0.2%Fe

Si, %

(Al)+F+α+Mg2Si

(Al)+β+π+Mg2Si

0.5

(Al)+β+π

(Al)+β+(Si)

(Al)+β+π+(Si)

610 620

80 0 5 0 59

0

570

300 0.0

(a)

57

60

0.0

(Al)+(Si)

560

L+(Al)+F+Mg2Si

(Al)+β+α

435

0.5

643

1.9 631

L+(Al)

(Al)+F

(Al)+β

1.0

L+(Al)+Mg2Si

L+(Al)+F

612 606

T°C

ina sib

2.0

Qu a

Mg, %

2.5

58

ry se 590 ctio n[ 0 59 59 0 5°C 56 5 ] 0 70

(Al 58 )+ M g2 S 0 i

3.0

L

657 654 635

3.5

0.89

0.7 0.88 1

2

Mg,%

(b) 700 6162

L L+(Al)

600 (Al)

L+(Al)+(Si)

500 T°C

(Al)+Mg2Si (Al)+Mg2Si+(Si) 400 (Al)+Mg2Si

L+(Al)+Mg2Si L+(Al)

300

L+(Al)+(Si) 1.24 1.12 555 (Al)+Mg2Si+(Si) 1.1

(c)

200 Al−0.9% Mg

1

1.2

1.3

2

3

Si, %

FIGURE 2.17 Solidus surface of the Al–Mg–Si system (a) and isopleths at 0.5% Si and 0.2% Fe (b) and 0.9% Mg (c) [32]. F denotes Al3Fe. (Reproduced with kind permission of Elsevier.)

phase, at the inverse ratio of these elements phases β(Al5FeSi) and π become dominant in the equilibrium state. The compositional range where iron is completely bound in the α(Al8Fe2Si) phase, which has the most favorable morphology among all Fe-containing phases, is quite narrow. The concentration of alloying elements can considerably affect the extent of the solidiication range. Figure 2.17c shows that even a small increase in Si concentration can strongly lower the solidus temperature (TS). For example, in a 6162 alloy containing 0.9% Mg, the change of TS within the grade compositional limits (0.4–0.8% for Si) can be as large as 25°C. The effect of magnesium is less strong. The effect of iron on TS depends on the formation of α(Al8Fe2Si) and β(Al5FeSi). These phases bound silicon and, therefore decrease

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

49

TABLE 2.5 Solidiication Reactions under Nonequilibrium Conditions in a 6063 Alloy (0.43% Mg, 0.39% Si, and 0.2% Fe) [5] Temperatures (°C) at a Cooling Rate Reaction

0.5 K/s

15 K/s

L ⇒ (Al) L ⇒ (Al) + α(AlFeSi)* L + α(AlFeSi)* ⇒ (Al) + Al5FeSi L + α(AlFeSi)* ⇒ (Al) + Al5FeSi + Mg2Si Solidus

655–653 618–615 613 576** 576**

654 617 610 576** 576**

* The crystal structure of α(AlFeSi) is cubic, hence this is a metastable phase (see Table 2.2). ** Estimated value from [29].

its free amount, resulting in a higher solidus temperature. According to Belov et al. [32], up to 0.1% Si can be bound in the Fe-containing phases in 6XXX alloys containing 0.2% Fe. Nonequilibrium solidiication causes deviation from the equilibrium phase composition. For example, in an alloy containing 0.5% Mg, 0.5% Si, and 0.2% Fe, the Al3Fe phase is formed during equilibrium solidiication as follows from the isopleth shown in Figure 2.17b. However, as the formation of this phase requires larger undercooling as compared to the α(Al8Fe2Si) phase, the latter usually forms under real casting condition (at this alloy composition). High-temperature peritectic reactions are hindered during solidiication at real cooling rates and, as a result, the phases that should disappear upon these reactions are retained in the structure. Experimental studies of nonequilibrium solidiication of a 6063 alloy were performed by Bäckerud et al. [5]. The results given in Table 2.5 show the simultaneous presence of α(AlFeSi) and β(Al5FeSi) phases in the as-cast structure. This agrees well with the casting practice. The amount of Mg2Si formed at the end of solidiication is small. Frequently, particles of Mg2Si form conglomerates with iron-containing particles that testiies for the occurrence of the last, peritectic reaction in Table 2.5. On increasing the concentration of Mg and Fe in 6XXX series alloys, the probability of Al3Fe formation increases, especially at moderate cooling rates. Hsu et al. [38] examined the phase composition of an Al–0.8% Mg–0.6% Si–0.3% Fe alloy (6063 type, high-alloyed) after nonequilibrium solidiication at ~0.1 K/s and revealed the eutectic formation of the Al3Fe phase at 625°C and of the metastable α(AlFeSi) phase at 593°C. On further increasing concentration of silicon (and low iron), the formation of Al3Fe is unlikely even upon slow cooling, and the probability of the quaternary π phase formation increases. For example, simultaneous presence of Mg2Si, (Si), β, and π crystals within one conglomerate is observed in an Al–0.86% Mg–1.61 Si–0.072% Fe alloy cast at 0.03 K/s [39].

Copyright 2008 by Taylor and Francis Group, LLC

50 2.3.4

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys Al–Mg–Si–Cu Wrought Alloys (6XXX and 2XXX Series)

Several widely used commercial alloys of the 6XXX series contain copper that is added to improve strength. In most cases, the solidiication of 6XXX series alloys containing copper can be described using the Al–Mg–Si phase diagram (Figure 2.17). With taking into account typical additions of transition metals, the as-cast structure exhibits the aluminum solid solution with Mn, Ti, and Cr-containing particles formed during high-temperature eutectic (Mn) and peritectic (Ti, Cr) reactions. Only at the concentrations of alloying elements close to the upper limits do the alloys fall at different phase ields (containing Si and Mg2Si) at the end of equilibrium solidiication as shown in the isothermal section at 500°C (close to the solidus) for alloys containing 1% Si (Figure 2.18). Impurity of Fe reacts with Si and Mn to form (AlFeMnSi) particles of eutectic origin. Under real casting conditions nonequilibrium eutectics containing Mg2Si and Si are formed, decreasing the solidus of the alloy. The main implication of copper addition to Al–Mg–Si alloys is the formation of the quaternary Al5Cu2Mg8Si6 (Q) phase through peritectic reactions with Mg2Si and Si occurring at 529 and 512–514°C or by a multi-phase eutectic reaction at 505–507°C. This phase

1.34 1.72

4.05 4

Al2Cu+Q Q+(Si)

(Si)+Al2Cu

Cu, %

0.28 0.58 6

Mg2Si+Al2Cu [1.35;4.2]

[0.56;4.05] [1.72;3.74]

Q+

Mg

2 Si

[0.29;3.9]

2 [0.6;1.52]

[0.5;1.04]

Al − 1% Si−0.5

[1.1;1.03]

1 1.1

(Si)+Mg2Si

2

3 Mg, %

FIGURE 2.18 Isothermal section of the Al–Cu–Mg–Si phase diagram at 500°C and 1% Si [32]. (Reproduced with kind permission of Elsevier.)

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

51

is very stable and can be found in Al–Mg–Si alloys with an excess of Si even at additions of copper as small as 0.25% [40]. According to Mondolfo this phase forms upon solidiication and is in equilibrium with aluminum under the following conditions: Mg:Si < 1.73, [Mg] > 2 [Cu], and [Cu] > 1% [29]. Depending on the ratio of copper, magnesium, and silicon, the Q phase can coexist with Al2Cu (θ) (that appears in Al–Cu–Mg–Si alloys with the suficient amount of copper, > 4%), Mg2Si and Si. The phase composition of cast alloys signiicantly differs, as a rule, from the equilibrium selection of phases, which is due to incomplete peritectic reactions and hindered diffusion of copper, magnesium, and silicon in (Al) during solidiication. Almost all phases from the aluminum corner of the Al–Cu– Mg–Si system—(Si), Mg2Si, Al2Cu, Al2CuMg and Q—can be found simultaneously in as-cast ingots and billets of wrought alloys of the 2XXX series. 2.3.5

Al–Mg–Mn Wrought Alloys (5XXX Series)

Let us irst look at the general effect of magnesium and manganese on the solidus and liquidus of commercial alloys with minor silicon content (< 0.5%). Plots in Figure 2.19a and b are obtained by interpolation of data from [41]. Obviously, magnesium has much stronger inluence on both characteristic temperatures than manganese. These charts can serve as a directory on the correct choice of casting and annealing temperatures. The polythermal section of the ternary Al–Mg–Mn phase diagram shown in Figure 2.19c gives us the irst approximation of phase transformation history for alloys like 5083, 5182, and 5456 containing 4–7% Mg and 0.3–0.8% Mn. Solidiication starts at 635–640°C with the formation of (Al) grains. After that, providing that the concentration of Mn is suficient, the (Al) + Al6Mn eutectics is formed in the range of temperatures from 627 to 617°C [5]. Other phases that may be present in the solid state are formed by precipitation from the aluminum solid solution. These reactions usually seldom occur during cooling after the end of solidiication because, due to a relatively low diffusion coeficient, manganese usually remains in a supersaturated solid solution. During annealing of cast material, Mn-containing phases precipitate from the solid solution supersaturated during solidiication and form dispersoids. The inal equilibrium phase composition of alloys containing more than 4% Mg and 0.1% Mn is (Al) + Al8Mg5 + Al10(MgMn)3. If the concentration of Mg is less than 2–3%, the equilibrium phase composition at room temperature would be (Al) + Al6Mn + (traces) Al10(MgMn)3. The presence of silicon, as an impurity or an alloying element, in Al–Mg–Mn alloys results in the formation of Mg2Si in addition to other phases that we have already considered. After formation of primary aluminum grains, the binary eutectics (Al) + Al6Mn is solidiied. During further cooling, Al15Mn3Si2 is formed. This phase then reacts with liquid to produce Al10(MgMn)3 and Mg2Si phases, and these phases remain in equilibrium with aluminum down to the room temperature. The Al8Mg5 phase is formed by precipitation from the aluminum solid solution upon cooling in the solid state.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys 1.2

530 510

550

0.6

570

1.0 Mn, %

0.8

590

645

625

Mn, %

1.0

635

1.5

610

52

0.5

0.4 0.2

0.0 2 (a)

4 6 Mg, %

8

10

0

2

4 6 Mg, %

(b)

T°C 700

615 649

627

L

8

10

L+Al6

600 L+(Al)

L+(Al)+Al6

500 (Al)+Al6

(Al) 400 300

0.37 264

205

200

(Al)+Al10+Al6

0.1

(Al)+Al10+Al8

100 Al−5% Mg (Al)+Al8

(c)

1

2.75

2 2.73 Mn,%

3

Al10: Al10(MgMn)3 Al8: Al8Mg5 Al6: Al6Mn

FIGURE 2.19 Isotherms of liquidus (a) and solidus (b) for commercial Al–Mg–Mn alloys containing silicon and iron on the impurity level; and (c) isopleth of the Al–Mn–Mn phase diagram at 5% Mg. Compositional range of some 5XXX series alloys is marked by dashed lines [32]. (Reproduced with kind permission of Elsevier.)

Most commercial alloys contain iron as an impurity. As a result, the (Al) + Al6(FeMn) eutectics is formed in the temperature range from 600 to 570°C, and the solidiication ends at approximately 570°C with the formation of the (Al) + Al6(FeMn) + Al3Fe eutectics. Under equilibrium, other phases that are frequently observed in Al–Mg–Mn–Fe alloys, e.g., Al8Mg5 and Al10(MgMn)3, are formed only by precipitation from the aluminum solid solution upon cooling in the solid state. Under real, nonequilibrium conditions, these phases can form during solidiication as a result of eutectic reactions. In the case of simultaneous presence of iron and silicon in a 5XXX series alloy containing 5% Mg (e.g., the 5182 alloy), the solidiication will end at 576–578°C with the formation of Mg2Si by a eutectic reaction.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

53

The solidiication paths that we considered above describe the phase equilibrium and can hardly be accomplished under real casting conditions when cooling rates are high and the diffusion processes, especially in the solid phase, cannot be completed to such an extent that the compositions of the phases change with temperature in accordance with the equilibrium phase diagram. Local deviations from equilibrium result in microsegregation and eventually in the shift of local equilibrium to the concentrations where new phases are formed. In addition, some high-temperature peritectic reactions remain uncompleted, and high-temperature phases—that have to disappear as a result of these reactions—are retained at lower temperature and can be found in the solid sample. The consequences of nonequilibrium solidiication for the phase transformations and phase selection are demonstrated below for a 5182 alloy that has been tested upon solidiication at different cooling rates, from 0.3 to 11 K/s, these cooling rates being characteristic of direct-chill and die casting [5]. The equilibrium phase diagram gives a solidus temperature of 576–578°C. However, thermal and phase analyses performed during and after solidiication clearly demonstrate that solidiication continues at lower temperatures. Table 2.6 gives solidiication reactions and corresponding temperature ranges at different cooling rates for this alloy [5]. The solidiication starts with the formation of aluminum grains. The solidiication temperature slightly decreases with increasing cooling rate due to increased undercooling. Then two concurrent reactions may occur at 620°C. After that the eutectics containing Mg2Si and Al3(FeMn) is formed at about 580°C. The equilibrium solidiication ends at this point. However, experimental results show that, after reaching the equilibrium solidus, the alloy still contains a liquid phase that undergoes a complex eutectic reaction and completely vanishes only at 470°C, thereby extending the solidiication range by almost 100°C. Thompson et al. studied the nonequilibrium solidiication of a 5182 alloy cast at different cooling rates [42]. They observed that the temperature of (Al) + Al6(FeMn) eutectics decreases from 588 to 575°C and the solidus decreases from 510 to 461°C on increasing the cooling rate from 0.5 to 2 K/s. The resultant phase

TABLE 2.6 Solidiication Reactions under Nonequilibrium Conditions in a 5182 Alloy (4.74% Mg, 0.34% Mn, 0.28% Fe, 0.1% Si) [5] Temperatures (°C) at a Cooling Rate Reaction

0.3 K/s

11 K/s

L ⇒ (Al) L ⇒ (Al) + Al6(FeMn) and/or L ⇒ (Al) + Al3(FeMn) L ⇒ (Al) + Al3Fe + Mg2Si L ⇒ (Al) + Al3Fe + Mg2Si + Al8Mg5 Solidus

632 621–617 586 557–470 470

632–623 620 583 543–470 470

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

54

composition of a 5182 alloy is (Al), Al3Fe, Mg2Si, Al8Mg5, and Al6(FeMn), the last phase frequently replaced by Al4(FeMn) due to the incomplete high-temperature peritectic reaction and enrichment of liquid in Mg and Mn [43]. 2.3.6

Al–Cu–Mn (Mg, Si, Ni) Wrought Alloys (2XXX Series)

The combined presence of Mn, Fe, and Si in these alloys mainly results in the formation of the Al15(FeMn)3Si2 phase through different eutectic reactions. The number of phases in the as-cast (nonequilibrium) state can be larger than the number under equilibrium conditions, but the sequence of solidiication reactions is in general agreement with the corresponding phase diagrams. Table 2.7 gives as an example the solidiication reactions identiied during nonequilibrium solidiication of a 2024 alloy [5, 30]. This alloy belongs to the Al–Cu–Mg–Mn system but the presence of Fe and Si impurities complicates the situation and causes the formation of phases containing these elements (Table 2.7). The as-cast structure of a 2024 alloy exhibits particles of the Al15(MnCuFe)3Si2 phase that is formed after primary (Al) grains. As iron is completely bound in this phase, the early solidiication reactions can be analyzed using the Al–Cu–Mn–Si phase diagram. Therefore, the binary eutectic reaction L → (Al) + Al15Mn3Si2 transforms to the ternary one L → (Al) + Al20Cu2Mn3 + Al15Mn3Si2. Then, Al20Cu2Mn3 must disappear through the peritectic reaction L + Al20Cu2Mn3 → (Al) + Al15Mn3Si2 + Al2Cu. The next reactions can be analyzed using the Al–Cu–Mg–Si phase diagram because manganese and iron have already been consumed by the earlier formed phases. A stringent analysis of 2618-type alloys requires the ive-component Al– Cu–Fe–Mg–Ni phase diagram because these alloys contain enough copper and magnesium to produce the Al2CuMg phase. As this phase diagram is yet to be constructed, only a simpliied analysis of the phase composition of this type of alloys is possible using constitutive phase diagrams and considerable experimental data on the structure of 2618-type alloys [32]. According to TABLE 2.7 Solidiication Reactions under Nonequilibrium Conditions in a 2024 Alloy (4.44% Cu, 1.56% Mg, 0.55% Mn, 0.21% Si and 0.23% Fe) [5] Temperatures (°C) at a Cooling Rate Reaction

0.8 K/s

13 K/s

L ⇒ (Al) L ⇒ (Al) + Al15(MnCuFe)3Si2 L ⇒ (Al) + Al15(MnCuFe) 3Si2 + Al20Cu2Mn3 L + Al20Cu2Mn3 ⇒ (Al) + Al15(MnCuFe)3Si2 + Al2Cu L ⇒ (Al) + Al2Cu + Mg2Si L ⇒ (Al) + Al2Cu + Al2CuMg + Mg2Si Solidus

637–633 633–613 551–538

637–627 613 544

486

480

486

480

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

55

these data, the microstructure of these alloys in the as-cast state contains particles of the Al9FeNi and Al2CuMg phases. The former probably forms through the binary eutectic reaction L → (Al) + Al9FeNi which, due to the presence of copper and magnesium in an alloy, occurs over a wide temperature range, from approximately 640–645°C down to 505–515°C. The Al9FeNi phase is then preserved in the inal structure of deformed semiinished items. A subsequent treatment can only change the morphology of particles that, as a rule, is rather compact. The appearance of the Al2CuMg phase in the as-cast structure is a consequence of nonequilibrium solidiication. During homogenizing annealing, it completely dissolves in solid (Al). The nonequilibrium solidus of 2618-type alloys corresponds to the temperature of the quasi-ternary L → (Al) + Al9FeNi + Al2CuMg eutectics and is about 515°C. At a maximum copper concentration (within the alloy nominal composition), the Al2Cu phase can form, in this case the solidiication completes at a lower temperature, ~505°C according to the Al–Cu–Mg phase diagram. The formation of Al9FeNi primary particles is unlikely in the entire compositional range of a 2618 alloy. 2.3.7

Al–Mg–Zn–(Cu) Wrought Alloys (7XXX Series)

Copper-less 7XXX-series alloys, e.g., 7004 and 7005, contain, as a rule, less than 6–7% (Zn + Mg) because higher concentrations facilitate stress corrosion. In the as-cast condition, the structure of these alloys consists mainly of (Al) grains. The presence of iron and silicon impurities causes the formation of Al3Fe and Mg2Si phases as has been shown experimentally by Bäckerud et al. for a 7005 alloy [5] (Table 2.8). However, the Al8Fe2Si phase should form in this type of alloy at Si:Fe > 3 [29]. The probability of Al8Fe2Si formation increases at high cooling rates [32]. Slow cooling favors the formation of Al3Fe because magnesium in 7XXX alloys is in excess of Mg2Si. Additions of Mn in some commercial 7XXX alloys promote the formation of Al15(FeMn)3Si2 particles that can hardly be dissolved during heat treatment and are retained in the structure of the inal product. Small additions of Cr and Zr do not

TABLE 2.8 Solidiication Reactions under Nonequilibrium Conditions in a 7005 Alloy Containing 5.14% Zn, 1.12% Mg, 0.18% Mn, 0.19% Fe, and 0.08% Si [5] Temperatures (°C)* at a Cooling Rate Reaction

0.3 K/s

15 K/s

L ⇒ (Al) L ⇒ (Al) + Al3Fe L ⇒ (Al) + Mg2Si L ⇒ (Al) + Al2Mg3Zn3 Solidus

641 632–596 596 470 470

638 610 560 470 470

* Start of reaction.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

56

TABLE 2.9 Solidiication Reactions under Nonequilibrium Conditions in a 7075 Alloy (5.72% Zn, 2.49% Mg, 1.36% Cu, 0.19% Cr, 0.28% Fe, and 0.11% Si) [5] Temperatures (°C)* at a Cooling Rate Reaction

0.3 K/s

2.3K/s

L ⇒ (Al) L ⇒ (Al) + Al3Fe L ⇒ (Al) + Mg2Si L ⇒ (Al) + Al2Cu**+MgZn2+ Al2Mg3Zn3 Solidus

630–623 618–615 568–563 469 469

628 — 558–550 466 466

* Start of reaction. ** Al2CuMg

affect the as-cast phase composition much because these elements enter the aluminum solid solution during solidiication. High-strength wrought aluminum alloys have a complex chemical composition, i.e., they contain at least ive or six components. With certain assumptions, these alloys can be assigned to the Al–Cu–Mg–Zn system. According to Bäckerud et al. the nonequilibrium solidiication of a 7075 alloy ends at 466–469°C with the invariant eutectic reaction L → (Al) + Al2Cu* + MgZn2 + Al2Mg3Zn3 [5]. In our opinion, the reaction with the participation of the Al2CuMg (S) phase is more likely (see Table 2.9). Inclusions of the η (MgZn2) and T (Al2Mg3Zn3) phases formed during nonequilibrium solidiication are completely dissolved in (Al) during homogenization in the temperature range 435°C–445°C. 2.3.8

Wrought Alloys Containing Lithium

Commercial Al–Li alloys can be tentatively divided into three groups: Al–Li–Cu (2090, 2020, VAD23rus); Al–Li–Mg (1420rus), and Al–Li–Cu–Mg (2091, 8090, Weldalite049, CP276, 1441rus, 1464rus). Most Al–Li alloys contain Zr; some contain other small additions such as Ag (Weldalite049), Sc (1421rus, 1424rus, 1464rus), Cd (VAD23rus). The implications of these additions for the phase composition are discussed in this section. 2.3.8.1

Al–Cu–Li Commercial Alloys

Commercial alloys that belong to this system contain 1–2.5% Li, 2.5–5.5% Cu and small additions of Zr (2090) or Mn and Cd (VAD23rus). The solidiication of a VAD23-type alloy (1.15% Li, 5.15% Cu) starts with the formation primary (Al) grains, then a small amount of binary (Al) + Al2Cu eutectics precipitates. The remaining liquid reacts with Al2Cu according to the transition reaction L + Al2Cu → (Al) + Al7.5Cu4Li (TB). During that reaction (under equilibrium conditions) Al2Cu disappears and T1 (Al2CuLi) and TB

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

57

phases are formed. The solidiication is likely to end with the formation of (Al) + TB binary eutectics. During cooling in the solid state, TB and T1 phases precipitate from the aluminum solid solution as a result of decreasing solubility of Cu and Li in (Al). The resultant phase composition of a VAD23-type alloy is (Al) + T1 + TB, the ternary phases forming different structure constituents. In the case of a 2090-type alloy with a higher concentration of Li (2.25%) and a lower concentration of copper (2.7%), after primary formation of (Al) grains the (Al) + AlLi binary eutectics solidiies. The alloy then undergoes transition reactions with the formation of T1 and T2 (Al6CuLi3) phases. Hence, the inal equilibrium phase composition at room temperature of a 2090-type alloy is (Al) + T1 + T2. Under nonequilibrium solidiication conditions, the amount of the binary eutectics in both alloys increases. Transition (peritectic) reactions may not complete, hence some particles of Al2Cu and AlLi, respectively, remain in the fully solidiied structure. The formation of the ternary eutectics (Al) + T1 + TB becomes possible, and the solidus temperature may decrease to 518– 528°C. The structure of all eutectics in wrought Al–Cu–Li alloys is divorced and appears as individual particles at grain and dendritic cell boundaries. Nonequilibrium phases Al2Cu and AlLi will dissolve during homogenization annealing of commercial alloys. 2.3.8.2

Al–Li–Mg Commercial Alloys

Commercial Al–Li–Mg alloys (without copper) were developed and used in Russia. All alloys of this group (142Xrus) contain small additions of transition metals such as Zr, Mn, and Sc. The equilibrium phase composition and the solidiication path of 142Xtype alloys can be evaluated from ternary and quaternary phase diagrams, e.g., an isopleth shown in Figure 2.20a. The equilibrium structure (at 4–6% Mg, 1.5–3% Li) consists of (Al) grains with secondary particles of AlLi and Al2LiMg phases formed during cooling in the solid state. Additional alloying with Zr and Sc results in more complex solidiication behavior with primary solidiication of Al3Zr and Al3Sc phases. There are indications that the Al2LiMg phase may form peritectically at higher Li concentrations [44]. The (nonequilibrium) solidus of Al–Li–Mg–Zr alloys is 530°C. At cooling rates realized during real casting, scandium and zirconium enter the aluminum solid solution during solidiication and then precipitate upon high-temperature annealing to form coherent particles of stable Al3Sc and metastable Al3Zr phases. 2.3.8.3

Al–Li–Cu–Mg Commercial Alloys

The majority of commercial Al–Li alloys contain copper and magnesium within the compositional ranges 0.3%–6% Mg, 1–5.5% Cu, and 1–3% Li. All three elements contribute to the formation of phases and structure during solidiication, deformation, and aging. The as-cast and annealed alloys may contain the following phases: (Al), T1 (Al2CuLi), T2 (Al6CuLi3), Al2Cu,

Copyright 2008 by Taylor and Francis Group, LLC

58

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys T°C

700 L+Al3Zr 650 (Al)+Al3Zr+Al2LiMg+AlLi

L+(Al)+Al3Zr 600 L+(Al)+Al3Zr+Al2LiMg 550 (Al)+Al2Zr 500 1.0

(Al)+Al3Zr +Al3LiMg 2.0

4.0

3.0

5.0 Li, %

Al − 4.5% Mg − 0.2% Zr (a) 2090, CP276

2091, 8091

T°C

L+(Al)

L+(Al)+T2

L

600 (Al)+T2 484°C 400

L+(Al)+T2+S

(Al)+T2+S (Al)+T1+T2+S

200 (Al)+T2+S+Al2LiMg

(Al)+T1+T2

2 Al − 3% Cu − 2.5% Li

4

6

8

10 Mg, %

(b) FIGURE 2.20 Isopleths of Al–Li–Mg–Zr (a) and Al–Cu–Li–Mg (b) phase diagrams. (Reproduced with kind permission of Elsevier.)

and S (Al2CuMg). Figure 2.20b demonstrates the polythermal section of the quaternary phase diagram at 3% Cu and 2.5% Li, which is relevant for the analysis of such commercial alloys as CP276, 2091, 8091 (up to 3% Mg). One can easily see that the solidiication for these alloys starts with the formation of the aluminum solid solution, then T2 and S phases are formed through eutectic reactions. The T1 phase is formed by precipitation from the solid

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

59

solution. The lowest possible eutectic reaction in this system is L → (Al) + S + T2 + Al2LiMg at 484°C. The effects of alloying elements on the solidus temperature of Weldalitetype alloys (4–6.3% Cu, 0–2% Li, 0–0.8% Mg) were studied by Montoya et al. [45]. The variation in copper concentration above 4% has virtually no effect on the solidus temperature of the base alloy Al–1.3% Li–0.4% Mg, the solidus being at 512–513°C. Increasing the concentration of magnesium in the Al–(5–6)% Cu–1.3% Li alloys results in the continuously decreasing solidus temperature, from 521 to 507°C. A minimum solidus temperature of 507– 513°C suggests the occurrence of a eutectic reaction at this temperature. The temperature is, however, about 10°C lower than that of the eutectic reaction L → (Al) + T1 + TB of the Al–Li–Cu system. Mukhopadhyay et al. suggest that the eutectic reaction L → (Al) + Al2Cu + Al2CuMg (S) that takes place at 507°C is responsible for the lowest melting temperature in Weldalite-type alloys [46]. They also observed the melting of the ternary eutectics (Al) + T1 + TB at 521°C. The phase composition of as-cast Weldalite-type alloys is (Al) + T1 + TB + Al2Cu + Al2CuMg (S), with the last phase being present only in small quantities [46]. It should be noted that the possibility of the invariant eutectic reaction L → (Al) + T1 + S that occurs at 505 ± 10°C cannot be excluded under nonequilibrium solidiication conditions [40]. The existence of low-melting eutectics in Weldalite-type alloys should be taken into account while choosing the correct regime of solution treatment.

2.4 2.4.1

Effect of Alloy Composition on Structure Formation: Grain Refinement* Mechanisms of Grain Refinement

It is well known that that the inal structure formed in a casting depends on both the alloy composition and the casting conditions. Figure 2.3 illustrates this fact for binary Al–Cu alloys and Figure 2.4 explains the principle of constitutional undercooling, the phenomenon driven by the alloy composition and phase diagram constitution. In this section we look at the interrelation between the alloy composition and the as-cast structure in more detail. The ranking of solute effects on the grain size in hypoeutectic aluminum alloys can be done simply by comparison of their equilibrium solidiication ranges [2]. Two parameters have been suggested for quantifying the effect of alloy composition on grain size, both parameters being proportional to the solidiication range. These are the undercooling parameter [47]: P = mC0(K – 1)/K = ∆T0

(2.7)

* Fruitful discussions with Dr. R. Mathiesen of SINTEF (Norway) and Dr. M. Easton of CAST CRC (Australia) on the issues of solute effects in grain rei nement are greatly appreciated.

Copyright 2008 by Taylor and Francis Group, LLC

60

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

and the growth restriction factor [48]: Q = mC0(K – 1) = K∆T0,

(2.8)

where m is the liquidus slope at the alloy composition C0, K is the partition coeficient, and ∆T0 is the equilibrium solidiication range, i.e., the difference between equilibrium liquidus and solidus temperatures. The undercooling parameter represents the potentially possible undercooling and is equal to the solidiication range of an alloy with composition C0, while the growth restriction factor relects the solute rejection at the solid–liquid interface. The constitutional undercooling at the solid–liquid interface is proportional to the growth restriction factor: ∆Tc = ΞQ,

(2.9)

L L L where Ξ = (Css – C0)/(Css (1 – K)) is the solutal supersaturation [1]. Css represents the liquid constitution at the interface, which is different from the equilibrium C0, when undercooling is present (see Figure 2.22). Generally the constitutional undercooling at the solid–liquid interface will drive the growth according to the general formulation for the maximum growth velocity [1, 49]:

Vgrowth = Ds∆T2c /QΓ,

(2.10)

where Ds is the diffusion coeficient of the solute in the melt, ∆Tc is the undercooling, Q is the growth-restriction factor, and Γ is the Gibbs–Thompson coeficient relecting the effect of the dendrite tip curvature on the tip growth velocity. (Γ = σ ls/∆sf, where σ ls is the solid–liquid interfacial energy and ∆sf is the volumetric entropy of fusion, which corresponds to the latent heat of solidiication.) We can see, at least qualitatively, that slow solute diffusion (Ds), small undercooling (∆Tc), large solidiication range and partitioning (Q), and large interfacial energy (σ ls) would result in limited growth of the solid phase into the liquid. It might seem awkward that the growth-restriction factor is proportional to solute-caused* undercooling (compare Equations 2.8 and 2.9) and, at the same time, determines the hindering of solid-phase growth in the presence of solute in the melt. This apparent dilemma can be resolved if we accept the following. 1. Growth of the interface requires the local equilibrium between solid and liquid concentrations that is easily shifted under real solidiication conditions from the global equilibrium according to the equilibrium phase diagram for the alloy composition C0. In the case of hypoeutectic aluminum alloys, the liquid becomes supersaturated with the solute. * Here and below we mainly consider hypoeutectic alloys where the constitutional undercooling is a result of solute enrichment. In the case of peritectic systems and hypereutectic alloys, the constitutional undercooling is caused by the solvent (aluminum).

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

61

1

2

3

4

FIGURE 2.21 A sequence of X-ray micrographs (from 1 to 4) taken in the European Synchrotron Radiation Facility from an Al–30 wt% Cu alloy solidifying in a Bridgeman-type furnace with the cooling at the bottom. Gray shade in the liquid area changes from black to light gray on decreasing the concentration of copper from 33 to 30 wt% Cu. An arrow points at the columnar dendrite tip, the growth of which depends on the Cu concentration ahead. (The sequence is courtesy of Dr. R. Mathiesen (SINTEF, Norway) and Prof. L. Arnberg (NTNU, Norway).)

2. The so-called growth restriction can be caused not only by the solute rejected from the advancing front (“parent” interface) but also by the solute that is brought to the front from elsewhere, e.g., by convection. The solute supersaturation at the interface should be dissipated either by diffusion (convection) or by remelting of the solid phase. Figure 2.21 [25] brilliantly illustrates how the uneven accumulation of Cu in the melt at the columnar front (frame 2) hinders the growth of a grain marked by an arrow. As soon as this accumulation is dissipated by diffusion and convection (frame 3), the grain resumes its growth (frame 4). The reasons for the solute accumulation or the solute inlux to the solid/ liquid interface can be many. Most frequently, the growth restriction is considered in relation to grain reinement by inoculants and to the columnarto-equiaxed transition (CET) [50–52]. In the course of grain rei nement, the solute rejected by the growing grain slows down its growth and, at the same

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

62 T CiS

TC0

Liquid Tx

Tss

Solid

(a)

CxS

C0

CxL

CssL

C CxL

CssL CxL C0

Cxs

C0

s

Cx Cis

Cis

Tmelt

Tmelt

TC0

TC0 Tinterface=Tx Tss Solid

(c)

Tinterface =Tx

Region of constitutional undercooling Region of constitutional superheating

Liquid

Solid

Region of constitutional undercooling

Liquid

(b)

FIGURE 2.22 Diagram of solute effects on the advance of the solidiication front: (a) a scheme of a binary diagram with characteristic temperatures and compositions; (b) change in the constitutional undercooling at the “parent” interface in the case of nucleation ahead of this interface (solid lines and arrows = no or weak nucleation ahead; dashed lines and arrows = intensive nucleation ahead [53]); and (c) the remelting of the “parent” interface due to the excess solute accumulation.

time, may create the constitutional undercooling suficient for the nucleation of a new grain at the substrate that pre-exists in the melt [52]. In the case of CET, the “added” solute is the solute rejected by the columnar grain tip, but it also can come from the neighboring columnar grain or the solid phase nucleated ahead of the columnar front. The constitutional undercooling due to the solute rejection at the solid–liquid interface creates favorable conditions for heterogeneous nucleation at some distance ahead of the solidiication front (see Figures 2.4 and 2.22b). The nucleated grains start to grow and reject solute as well as generate the latent heat. As a result, the growth of the “parent” interface stops. The greater the solidiication rate (solid fraction evolution per unit time), the more effective this growth restriction.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

63

Martorano et al. showed that the rejection of solute by newly formed grains with a high grain density will, in fact, effectively decrease the constitutional undercooling initially created at the “parent” interface and stop its advance, as shown in Fig ure 2.22b [53]. The solute also can be brought to the interface by thermo-solutal and shrinkagedriven lows during solidiication (see (2) above), as well as be caused by nonequilibrium solidiication conditions (see (1) above). In both cases the volume around the interface cannot be considered a closed system. The volume becomes open. The extent of this solute accumulation depends on the effective diffusivity of the solute in the melt (see Equation 2.10). In order to maintain the local equilibrium at the interface and obtain the momentum for further growth (solidiication) the interface should irst remelt and absorb the excess of the solute. This situation is shown in Figure 2.22c and is often a case upon fragmentation of dendrite branches [54, 55]. It is worth mentioning here that fragmentation or “grain multiplication” is considered to be one of the mechanisms of grain reinement in castings. An example of fragmentation observed in situ during solidiication of an Al–20% Cu alloy is demonstrated in Figure 2.23 [56]. It is quite obvious that the accumulation of the solute copper at the root of the dendrite branch causes its remelting and, eventually, detachment from the parent branch, as shown in Figure 2.24 [55]. In this particular case the accumulation of FIGURE 2.23 the solute is a result of solute transport Fragmentation during columnar growth from the solidiication front (far from the of an Al–20% Cu alloy observed in situ by fragmentation site), from the tip of the X-ray microscopy [56] (images are courtesy of R. Mathiesen of SINTEF, Norway.) fragment-to-be, and due to the local solidiication in the place of detachment [55]. It is also suggested that the solute accumulated at the interface hinders the growth only until some values of Q, when the solid–liquid interface is relatively smooth (which is the case for globular and little-branched grains)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Cu wt%

64

26 24 22 20

150

150

100

100

50

50

120

80

60 40

40 20 0

0 0

20

0

20

0

0

20

40

0

0

20

40

FIGURE 2.24 Fragmentation of a tertiary branch (gray in igure) during columnar solidiication of an Al–20% Cu alloy, similar to that shown in Figure 2.23. Images are quantiied in terms of copper concentration in the liquid as shown in the insert scale [55]. The accumulation of copper at the root of the fragment-to-be is demonstrated. (The frame sequence is courtesy of Dr. R. Mathiesen (SINTEF, Norway).)

[52, 57]. In the case of high Q, and hence high constitutional undercooling, grains start to grow faster and in a very branched manner. As a consequence, the solute is rejected from the sharp and elongated dendrite-branch tips not only to the front but also sideways. This solute is accumulated in the mush between the dendrite branches and is, upon decreasing the temperature, absorbed by the solid phase. In such a way, the effective growth restriction is decreased and the dendrite grains have the opportunity to grow faster, even if the average concentration of solute in the melt is high. The concept of growth restriction is rather popular nowadays for the explanation of the effect of different solute levels on the eficiency of grain reinement in aluminum alloys [50–52, 58]. There are, however, some problems that are still exist with this concept. First of all, the idea that the growth restriction factor Q of an alloy can be estimated by summation of growth restriction factors of its individual alloying elements can be argued for two main reasons: (1) the interaction between different alloying elements can dramatically change not only the slope of the liquidus but also the amount of solute in the melt at different stages of solidiication [58] and (2) the effects of hypoeutectic, e.g., Cu, and “hypoperitectic,” e.g., Ti, elements on the composition of liquid in equilibrium with solid (Al) during solidiication are different; the former add solute into the melt while the latter add solvent. At the same time, this “summation” approach works

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

65

in practice quite well, and is explained by the fact that, though different elements affect the solute distribution ahead of the interface differently, their individual effects on the constitutional undercooling are similar [59]. The second problem is that the growth restriction factor is usually used in order to explain why Ti is so powerful as a grain reiner. Indeed, the growth restriction factor of Ti is Q ≈ 35 for the region of (Al) primary solidiication and at C0 = 0.15% Ti. This value is well above the typical values of the usual solutes in aluminum alloys [51]. However, there is no explanation as to why the most powerful grain reiner in aluminum alloys—scandium—has a relatively low growth restriction factor, Q = 2.5 for C0 = 0.6% Sc, and yet remains very eficient [60]. It is noteworthy that the undercooling factors P of both elements are quite close: 5 for Sc versus 4.4 for Ti. It is also very important to realize that the typical concentrations of Ti that are added to commercial aluminum alloys in the form of Al–Ti–B master alloys are in the range 0.001%–0.005% (C0 in Equation 2.8), and hence the growth restriction caused by titanium effectively decreases. It is, however, possible that the local enrichment of aluminum melt in Ti due to nonequilibrium segregation in the vicinity of grain-reining particles of TiB2 compensates for the otherwise very low bulk Ti concentration. In any case, the growth-restriction theory helps rank commercial aluminum alloys with regard to their susceptibility to grain reining. The grain reining criterion (grain size) can be determined as b∆Tn 1 + _____ ____ Ω = _____ , 3 Q √ Nf

(2.11)

where N is the amount (density) of particles in the melt, f is the fraction of these particles that can act as nucleants, ∆Tn is the undercooling required for nucleation, and Q is the growth restriction factor [61]. The irst term is related to the availability of solidiication sites, while the second term shows the possibility for nucleation (see Figure 2.28). Another theory of grain reinement links the formation of solidiication nuclei to compositional and structural luctuations in liquid. Atoms of the matrix metal and the addition are able to form stable conigurations. Samsonov and Lamikhov [62] suggested that the incompleteness of d electron shells in transition metals was the main factor determining their reining ability. Then the grain-reining criterion looked as Ω = 1/Ndnd,

(2.12)

where Nd is the principal quantum number of the d electron shell and nd is the number of electrons on the d level. The greater the Ω, the larger the grain reining eficiency. The following values can be obtained for Sc, Ti, Mn, Fe, and Zr: 0.333, 0.167, 0.067, 0.055, and 0.125. The combined eficiency of several transition elements can be estimated from the electron exchange between s and d electron shells. This theory explains the high reining ability of Sc and Ti from the statistically stable coniguration of atomic clusters that can act as solidiication nuclei.

Copyright 2008 by Taylor and Francis Group, LLC

66

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Further development of this theory included the interaction of valence electrons of both transition metal and aluminum [63]. Collective behavior of these electrons on s – p electron shells determines the stability of the nucleus. The following criterion has been suggested [63]: 1 Ω = _________________________ , Nsns + Npnp + Asas + Apap

(2.13)

where Ns, Np are the principal quantum numbers of and ns, np are the number of electrons on the s- and p-levels of the addition atom, whereas As, Ap, as, and ap are the same for the aluminum atom. Figure 2.25a illustrates the application of this criterion to two groups of additions to aluminum alloys. The irst group consists of typical grain reiners that act through the nucleation mechanism, facilitating the nucleation of solid aluminum by providing a better structural match between the substrate and aluminum. The second group represents so-called surface-active additions that can reine grain by changing surface properties of substrates, e.g., facilitating nucleation by lowering the interfacial surface energy. The results show the obvious advantage of Ti and Sc for the irst group and Ca for the second group of additions. The potency for grain reinement is also a function of the size of an atomic cluster that can become a nucleus. The number of atoms in such a stable nucleus is a function of the ratio between enthalpy (latent heat) of evaporation ∆Hv and enthalpy (latent heat) of fusion ∆Hf of an Al–Addition compound: ∆H na = J ____v , ∆Hf

(

)

,

(2.14)

where J is the coeficient equal to 1/4 for hexagonal closed packed and fcc lattices and to 9/16 for bcc lattices [63]. Using the number n, it is possible to determine the size of the cluster, as shown in Figure 2.25b. The larger the size, the lower the undercooling required for stable nucleation, and the smaller the grain size. Note, however, that the data shown in Figure 2.25b relect only the stability of the nucleus and do not consider its further growth, and therefore do not fully agree with the casting practice. Let us now look at some of the mechanisms of grain reinement in aluminum alloys. Generally, the idea behind grain reinement is the sharp increase in the number of solidiication sites for heterogeneous nucleation of the primary aluminum phase. This is usually done by special additions to an alloy, which are called grain reiners. However, it is possible to achieve grain reining by multiplication of solidiication sites either by fragmentation of existing solid grains or by activation of usually inert submicron solid particles that are always present in melts (e.g., oxides and carbides). The grain reinement by special additions is by far the most common method of producing castings with small, equiaxed, and uniformly distributed grains. Particles that act as substrates for primary (Al) are usually represented by high-temperature phases with crystal lattices that have a small

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

67 Mn

12 Cr

Zr Grain size, mm

B Cu 8 Fe

Mg

Ni

V 4

Si Ti

Zn

Ca Sc

0 0.03

0.04

0.05 0.06 Interaction parameter Ω

(a)

0.07

0.08

2.5 Y

Hf

Zr

2

Cluster size, nm

Nb 1.5

Ti

Mn Ta Mo

W

Fe

V Sc

Ni Co

1

0.5

Cr

B

0 (b)

FIGURE 2.25 (a) Variation of grain size with respect to the interaction criterion (Equation 2.13) for nucleation-type (solid line) and surface-active-type (dashed line) grain rei ners and (b) the size of a stable cluster [63].

Copyright 2008 by Taylor and Francis Group, LLC

68

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

mismatch with the crystal lattice of (Al), at least along some crystal planes. The demand for their formation in the alloy prior to the (Al) phase is obvious: they should act as a substrate and be stable in the melt from which (Al) is formed. In addition, (Al) will “choose” these phases as substrates only if the heterogeneous nucleation on them will reduce the energetic barrier for the “critical” nuclei [1]. Hence, these solidiication sites should present at least one of the surfaces with low interfacial (nucleation) energy, which is possible if the crystal planes of the substrate and (Al) have as little a mismatch as possible. The similarity of chemical composition can also be helpful in nucleation of a primary phase, but it usually acts in the case of fragmentation or, sometimes, activation of solid impurities, but not when special grain reiners are added. Among the most potent grain reiners are Ti (Al3Ti) with the peritectictype phase diagram with Al and Sc (Al3Sc) with the eutectic-type phase diagram with Al. The Al3Ti phase has a tetragonal crystal lattice and has a small mismatch with (111)Al at (112)Al3Ti plane [64]. The Al3Sc phase is cubic with a = 0.4104 nm, which is very close to the lattice parameter of (Al), a = 0.404 nm [60]. But, of course, the grain-reining effect in binary Al–Ti or Al–Sc alloys can be obtained only at concentrations where the corresponding primary phases are formed, >0.15% Ti and >0.6% Sc, respectively. In commercial practice, much less Ti or Sc is added, but then in combination with other elements such as B, C, and Zr. The question of how eficient grain reinement can be achieved at low concentrations of active elements is a problem in modern solidiication theory [52, 65]. As a irst approach to solving this problem, we need to look at what is known about the sequence and nature of solidiication in complex systems that are used as grain reiners, i.e., Al–Sc–Zr, Al–Ti–B, and Al–Ti–C [32]. 2.4.2

Al–Sc–Zr Phase Diagram

In the aluminum corner of this system only binary Al3Sc and Al3Zr phases are in the equilibrium with (Al). The Al3Sc phase is formed at 1320°C during a peritectic reaction, has a cubic ordered structure of L12 type with a = 0.4104 nm, and can dissolve Zr up to the composition Al3Sc0.6Zr0.4 (35% Zr) [60]. The Al3Zr phase melts congruently at 1577°C, has a tetragonal structure of D023 type with a = 0.4006–0.4014 nm and c = 1.727–1.732 nm, and dissolves Sc up to the composition Al3Zr0.8Sc0.2 (5% Sc) [60, 66]. These phases participate in the invariant reaction L + Al3Zr ⇒ (Al) + Al3Sc at 659°C. Mutual equilibrium solubility of Zr and Sc in solid (Al) is 0.06% Zr, 0.03% Sc and 0.09% Zr, 0.06% Sc at 550 and 600°C, respectively [60]. The typical concentration of scandium and zirconium in commercial aluminum alloys is lower than 0.3% Sc and 0.15% Zr or 1) solidiies as the primary phase. The eficiency of Sc as a grain reiner is much improved in the presence of Zr. In this case usually 0.15%

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

69

Al3Sc

Sc, wt.%

0.8

0.6 Al3Zr

659 0.4

900 850

0.2

800 750

(Al)

700 661

Al

0.4 0.6 Zr, wt. %

0.2

(a) 0.6

(Al)+Al3Sc

0.4

Sc, wt.%

0.8

(Al)+Al3Sc+(Al)+Al3Zr

0.2 (Al)+Al3Zr

(Al)

0.2

Al

0.4 0.6 Zr, wt. %

(b)

0.8

~850

T°C L L+Al3Zr 670 L+(Al)+Al3Zr

L+(Al)

650 L+(Al)+Al3Sc

(Al)+Al3Zr

(Al)+Al3Sc (Al)+Al3Zr+Al3Sc 630 Al−0.4% Sc

(c)

Zr, %

0.2% Sc 0.4% Zr Sc, %

Al−0.8% Zr

FIGURE 2.26 Al–Sc–Zr phase diagram: (a) projection of liquidus [68]; (b) isothermal sections at 450°C (dashed line) and 600°C (solid line) [32, 60]; and (c) isolpleth [60]. (Reproduced with kind permission of Elsevier.)

Copyright 2008 by Taylor and Francis Group, LLC

70

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

FIGURE 2.27 Precipitation of Al3Sc onto primary Al3Zr phase during Solidiication, SEM, secondary electrons, arrows show the Sc-rich layer [69].

Sc and 0.15% Zr are added to an alloy. The reason why is under discussion. Some authors suggest the formation of a ternary Al3(ScZr) phase with the crystal structure similar to that of stable Al3Sc and metastable Al3Zr. However, there is no evidence in favor of the formation of a new phase in the Al–Sc–Zr system. Metallographic examination of primary particles in Al–Sc–Zr alloys show that the Al3Sc phase (as a result of the peritectic reaction) forms a rim on primary Al3Zr particles (Figure 2.27). This surface layer possesses very good reining ability of Al3Sc and “activates” Al3Zr, allowing strong grain reinement at relatively low Sc concentrations [69]. 2.4.3

Al–Ti–B Phase Diagram

The Al–Ti–B phase diagram has been a subject of investigation since the early 1970s. Under equilibrium solidiication conditions, the following phases can be found together with aluminum in the solid state: Al3Ti, AlB2, and TiB2. The Al3Ti phase has a tetragonal crystal structure (space group I4/mmm) with a = 0.385 nm and c = 0.86 nm [29]. The open question is the existence of continuous solid solution between AlB2 and TiB2. Both compounds have a hexagonal crystal structure (space group P6/mmm) with close lattice parameters: a = 0.3003 nm and c = 0.3251 nm for AlB2 and a = 0.3032 nm and c = 0.3231 nm for TiB2 [70]. Thermodynamical calculations show the

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

71

TABLE 2.10 Invariant and Monovariant Reactions in the Aluminum-Rich Al–Ti–B Alloys [71] T, °C

Point in Figure 2.28

665 ~665 659 ~659 > 665 > 659 — 659 < T < 665

p1–P1 P1 e1 P2 S–P1 F–P2 P2–e1 P1–P2

Reaction L + Al3Ti ⇒ (Al) L + Al3Ti ⇒ (Al) + TiB2 L ⇒ (Al) + AlB2 L + TiB2 ⇒ (Al) + AlB2 L ⇒ TiB2 + Al3Ti L ⇒ TiB2 + AlB2 L ⇒ (Al) + AlB2 L + TiB2 + (Al) either peritectic or eutectic

2

3

B

F

Ti

AIB2

4

Liquid (Al)

6 e1

P2

1 5

2

S

h

e

J D

P1

AI

p1

p

Al3Ti

FIGURE 2.28 Scheme of the Al–Ti–B phase diagram [32, 71]. (Reproduced with kind permission of Elsevier.)

possible formation of (AlTi)B2 solid solution. However, numerous experimental results attest to the formation of separate, well-distinguished borides. The liquid–solid equilibria in Al-rich Al–Ti–B alloys have been studied in detail [71]. The invariant and monovariant solidiication reactions are shown in Table 2.10, and the schematic projection of the solidiication surface is given in Figure 2.28. 2.4.4

Al–Ti–C Phase Diagram

Another system that is important for grain reining is Al–Ti–C. In the aluminum corner of this system three phases are in equilibrium with (Al): Al3Ti, Al4C3, and TiCx (0.48 < x < 0.98). The Al4C3 carbide (25.3% C) has a rhombohedral structure with a = 0.855 nm and β = 22o28′ [29] or a hexagonal structure with a = 0.33328 nm and c = 2.5026 nm [66]. TiCx has a cubic structure (space group Fm3m) with a = 0.43176 [66].

Copyright 2008 by Taylor and Francis Group, LLC

72

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys C Al4C3

TiC0.98

L+TiCx1+Al4C3 TiCx1

L+Al4C3

L+

1

2 L

Al

TiC0.48

Ti C

x

TiCx2

L+TiCx2+Al3Ti

L+Al3Ti

Al3Ti

Ti

FIGURE 2.29 Scheme of the Al–Ti–C phase diagram [32, 72]. (Reproduced with kind permission of Elsevier.)

The invariant equilibrium L + TiC ⇒ Al3Ti + Al4C3 occurs in the aluminum corner of the phase diagram at 693°C (0.53 at.% Ti, 7 × 10–6 at.% C) [72]; other temperatures of 700 or 812°C also have been reported for the same equilibrium. In the structure of solidiied samples, the association of Al3Ti with TiC and TiC with Al4C3 is frequently observed, suggesting complex character of solidiication reactions in this system [73]. However, the full sequence of solidiication is not clear yet. A schematic isothermal cross-section is shown in Figure 2.29. Line 1 connects Al with the stoichiometric composition of TiC, whereas line 2 relects the equilibrium with the nonstoichiometric titanium carbide. The latter line does not cross phase ields with Al4C3 present and, therefore, the Al–TiCx (off-stoichiometric) system can be considered as quasi-binary [72]. As the reader can see, the mechanism of grain reining in Al–Sc–Zr alloys is more or less clear—a peritectic reaction stimulates the formation of Al3Sc layer onto primary Al3Zr particles with subsequent eficient nucleation of (Al). The situation is somewhat more complicated in Ti-containing alloys. It is obvious that Al3Ti is a very powerful grain reiner for (Al), when present at “hyperperitectic” concentrations, i.e., when Al3Ti is formed as a primary phase. In practice, the grain reining additions of Ti in the form of Al–Ti–B and Al–Ti–C master alloys are seldom above 0.05% Ti and, in many cases, much less. Yet the result they produce is remarkable. One set of theories is based on the assumption that there is a peritectic reaction that can produce (Al) from Al3Ti at very low concentrations of Ti, much less than in the binary Al–Ti system. An analysis of solidiication reactions in the Al–Ti–B system yields the following. Alloys in compositional ranges 1 and 2 (Figure 2.28) start solidiication with the formation of TiB2, then Al3Ti and TiB2 are formed through the eutectic reaction (S–P1). Small TiB2 particles are observed inside bulk Al3Ti crystals [64, 71, 74]. The equilibrium crystallization inishes with the invariant reaction P1 when the aluminum solid

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

73

solution is formed. Under nonequilibrium conditions, the peritectic reaction P1 is uninished and the remained liquid solidiies as divorced eutectics along line P1–P2. The described solidiication path is typical of titanium and boron concentrations in Al–Ti–B master alloys, i.e., 3–5% Ti, 1–3% B. In compositional ranges 3 and 4, the primary phase is TiB2. The solidiication continues along line F–P2 (formation of TiB2 + AlB2 eutectics) to point P2, where the invariant peritectic reaction occurs with the formation of AlB2 and (Al). Under real casting conditions, some liquid may remain at temperatures below the temperature of the peritectic reaction. This liquid solidiies according to line P2–e1, forming the (Al) + AlB2 eutectics. Alloys with intermediate compositions 5 and 6 also start to solidify with the formation of TiB2 as the primary phase. Then the aluminum solid solution is formed as a result of the peritectic reaction L + TiB2 ⇒ (Al) (P1–P2), and this reaction may transform to the eutectic reaction L ⇒ (Al) + TiB2 until all liquid vanishes at point P2. In compositional range 6, the peritectic reaction L + TiB2 ⇒ (Al) + AlB2 is likely to occur. As we can see, the analysis of the phase diagram does not give any evidence as to the existence of a peritectic reaction with participation of Al3Ti at low Ti concentrations. In a real casting situation the total amount of Ti is less than 0.05% and there is an excess of Ti with respect to TiB2. In this case, rims of Al3Ti onto TiB2 and rims of TiB2 onto AlB2 are frequently observed, suggesting complex incomplete solidiication reactions and possible mechanisms of grain reinement [64, 70]. The peritectic reaction may not be necessary if it were possible to have Al3Ti present in some form at these small Ti concentrations. There is a possibility that Al3Ti survives in the surface defects of boride particles and then acts as a nucleant for (Al) [75]. A more feasible mechanism is, however, as follows. During dissolution of Al–Ti–B master alloys in aluminum-alloy melt, the borides remain virtually unaffected while Al3Ti particles dissolve, creating the excess of solute Ti in the melt. This Ti can then segregate to the surface of boride particles and form there a very thin layer of Al3Ti phase. This layer is metastable and can be preserved for some time at temperatures above the melting point of the main alloy, creating the substrate for (Al) nucleation [74]. The sequence of habitus crystal planes is suggested as (0001)TiB2 || (112)Al3Ti || (111)Al [64]. Therefore, the aluminum solid solution nucleates on Al3Ti that is dispersed in the melt by primary solidiied borides (or pre-existing borides from the master alloy), and much less titanium is required for the same reining effect. Hence, there is a similarity in the ultimate mechanism of activation in Al–Ti–B and Al–Sc–Zr alloys: less eficient solidiication sites are covered with the layer of more potent nucleant, effectively decreasing the required amount of the latter. Recently, very interesting results have been produced by in situ X-ray diffraction measurements using a synchrotron radiation [76]. It has been demonstrated that in Ti-rich liquid aluminum inoculated by TiB2 the (metastable) Al3Ti phase is indeed formed prior to the solidiication of primary aluminum and then vanishes (is consumed) in the process of aluminum

Copyright 2008 by Taylor and Francis Group, LLC

74

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

100

Grain size, µm

(AI) AI3Ti ×10 50

0 930

940 Temperature, K

950

FIGURE 2.30 Growth of an aluminum grains (Al) and metastable Al3Ti phase in an Al–0.1% Ti–0.1% TiB2 alloy as a function of temperature during continuous cooling at a rate of 1 K/min [76]. Note that the Al3Ti phase appears at about 10 K above the onset of (Al) nucleation and is consumed during (Al) growth. (Reproduced with kind permission of Elsevier/Pergamon.)

nucleation, as shown in Figure 2.30 [76]. The nucleation process of primary aluminum is limited to the upper part of the solidiication range and is almost complete when about 20% of the solid phase is formed. Further nucleation is hindered by the latent heat released by previously formed grains. It also has been found that the presence of solute titanium in the melt indeed enhances the nucleation of new grains, not by a growth-restriction mechanism but rather by improved wetting of TiB2 particles by the melt. Al–Ti–C master alloys contain about 3% Ti and 0.15% C with the phase composition (Al) + TiC + Al3Ti. Titanium carbide can act as a solidiication sites for (Al) by itself as having a suficiently good structural match with aluminum. It was suggested by Cibula in the 1950s that even upon adding Ti to aluminum alloys, the actual nuclei for (Al) are represented by carbides formed by the reaction of Ti and Al with carbon impurities in the melt [77]. Later it was conirmed that TiC indeed acts as a powerful grain reiner in aluminum alloys [78]. However, in this system there is a possible peritectic reaction that produces Al3Ti from TiC, which may facilitate the formation of an Al3Ti envelope onto TiC particles and increase their nucleation potential by the same mechanism as in the case of the systems discussed earlier. The research interest has shifted lately from the search for the nucleation sites to the understanding why the nucleated grains do not grow big. The growth restriction caused by highly segregated Ti is the basis of this school of thought [50–52]. In this case, the “center of gravity” is shifted from the number of nucleation centers as a condition of eficient grain reinement to

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

75

the amount of growth restriction that is necessary to suppress their growth. Excess titanium that is brought to the melt during dissolution of master alloy then hinders the growth of (Al) grains that are nucleated on either borides or carbides. As we mentioned in the beginning of this section, the growthrestriction theory is not lawless, but then there are drawbacks to each hypothesis of grain reining with complex master alloys. It is quite possible that the initial formation of aluminum grains is controlled by the constitutional and thermal undercooling that drives the heterogeneous nucleation onto active envelopes of high-temperature particles. The accumulation of solute (solvent) that assists the nucleation simultaneously prevents the growth of the parent particles. The latent heat released during solidiication acts in the same direction. The greater the nucleation density of the primary grains, the sooner their thermo–solutal ields start to interact and the less their dendritic development. As a result, the ine and not well-developed dendritic equiaxed structure is formed. Sometimes the branching is so negligible that the grains are called globular or nondendritic. Another way to produce very ine grains is the activation of impurities that are always present as “plankton” in aluminum melts. These impurities could be oxides and carbides that are usually not wettable by liquid aluminum and, therefore, are excluded from the process of solidiication. If, however, they could be wet with liquid aluminum and form substrates for either aluminides or aluminum, then they would become a means for very eficient reinement. It has been suggested that cavitation melt treatment can facilitate the activation process and produce very ine, nondendritic structure in commercial-size billets and ingots of aluminum alloys [79]. The activation of nonmetallic impurities facilitates and reines the primary phases that solidify in alloys prior to (Al) [80]. The cavitation treatment can also disperse boride and carbide agglomerates that are frequently formed during introduction of master alloys into the melt and thereby involve more substrate particles into the solidiication process [81].

References 1. W. Kurz, D.J. Fischer. Fundamentals of Solidiication, Aedermannsdorf: Trans Tech Publications, 1992. 2. H. Xu, L.D. Xu, S.J. Zhang, Q. Han. Scr. Mater., 2006, vol. 54, pp. 2191–2196. 3. D.G. Eskin, Q. Du, D. Ruvalcaba, L. Katgerman, Mater. Sci. Eng. A, 2005, vol. 405, pp. 1–10. 4. M.C. Flemings, Solidiication Processing, New York: McGraw-Hill, 1974. 5. L. Bäckerud, E. Król, J. Tamminen. Solidiication Characteristics of Aluminium Alloys. Vol. 1: Wrought Alloys, Oslo: SkanAluminium/Univesitetforlaget, 1986. 6. V.Z. Kisun’ko, I.A. Novokhatskii, A.I. Pogorelov, Liteinoe Proizvod., 1986, no. 11, pp. 10–12.

Copyright 2008 by Taylor and Francis Group, LLC

76

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

7. I.G. Brodova, P.S. Popel, G.I. Eskin. Liquid Metal Processing. Applications to Aluminium Alloy Production, London: Taylor & Francis, 2002. 8. H. Fredriksson, L. Fredriksson. Mater. Sci. Eng. A, 2005, vol. 413–414, pp. 455–459. 9. P.P. Arsent’ev, D.I. Ryzhonkov, K.I. Polyakova, Yu. A. Anikin. Liteinoe Proizvod., 1987, no. 3, pp. 9–10. 10. G.G. Krushenko, V.I. Shpakov. Tekhnol. Legkikh Splavov, 1973, no. 4, pp. 59–62. 11. D.M. Herlach. Mater. Sci. Eng. A, 1997, vol. 226–228, pp. 348–356. 12. P. Šebo, J. Ivan, L. Táborský, A. Havalda. Kovové Materiály, 1973, vol. 11, no. 2, pp. 173–180. 13. D.G. Eskin. Tsvetn. Met., 1989, no. 5, pp. 97–99. 14. D.G. Eskin. Z. Metallkde., 1996, vol. 87, pp. 295–299. 15. J.A. Sarrial, C.J. Abbaschian. Metall. Trans. A, 1986, vol. 17A, pp. 2063–2073. 16. A.B. Michael, M.B. Bever. Trans. AIME, J. Met., 1954, vol. 200, no. 1, pp. 47–56. 17. I.I. Novikov, V.S. Zolotorevsky. Dendritnaya likvatsiya v splavakh (Dendritic Segregation in Alloys), Moscow: Nauka, 1966. 18. Q. Du, D.G. Eskin, A. Jacot, L. Katgerman. Acta Mater., 2007, vol. 55, pp. 1523–1532. 19. A.L. Dons. Ph.D. Thesis, NTNU, Trondheim, Norway, 2002. 20. Q. Du, A. Jacot. Acta Mater., 2005, vol. 53, pp. 3479–3493. 21. V.R. Voller, C. Beckermann. Metall. Mater. Trans. A, 1999, vol. 30A, pp. 2183–2189. 22. H.J. Diepers, C. Beckermann, I. Steinbach. Acta Mater., 1999, vol. 47, pp. 3663–3678. 23. A.N. Turchin, D.G Eskin, L. Katgerman. Metall. Mater. Trans. A, 2007, vol. 38A, pp. 1317–1329. 24. D. Ruvalcaba, R.H. Mathiesen, D.G. Eskin, L. Arnberg, L. Katgerman. In Proc. 7th Intern. Symp. Liquid Metal Processing and Casting, Nancy, France, September, 2007, pp. 181–185. 25. R.M. Mathiesen, L. Arnberg. Acta Mater., 2005, vol. 53, pp. 947–956. 26. N. Limodin, E. Boller, L. Salvo, M. Suéry, N. DiMichel. In: H. Jones (Ed.), Proc. 5th Decennial Intern. Conf. on Solidiication Processing, University of Shefield, Padstow: TJ Intern, Ltd., 2007, pp. 316–320. 27. S.A. Cefalu, M.J.M. Krane. Mater. Sci. Eng. A, 2003, vol. 359, pp. 91–99. 28. I. Vušanović, B. Šarler, M.J.M. Krane. Mater. Sci. Eng. A, 2005, vol. 413–414, pp. 217–222. 29. L.F. Mondolfo. Aluminum Alloys: Structure and Properties, London/Boston: Butterworths, 1976. 30. L. Bäckerud, G. Chai, J. Tamminen. Solidiication Characteristic of Aluminum Alloys. Vol. 2: Foundary Alloys, Des Plaines, IL: AFS/Skanaluminium, 1990. 31. N.A. Belov, A.A. Aksenov, D.G. Eskin. Iron in Aluminum Alloys: Impurity and Alloying Element, London, New York: Taylor & Francis, 2002. 32. N.A. Belov, D.G. Eskin, A.A. Aksenov. Multicomponent Phase Diagrams: Applications for Commercial Aluminum Alloys, Amsterdam: Elsevier, 2005. 33. H.W.L. Philips. Annotated Equilibrium Phase Diagrams of Some Aluminum Alloy Systems, Monograph 25, London: Institute of Metals, 1959. 34. Y. Langsrud. In Proc. Workshop Effect of Iron and Silicon in Aluminum and Its Alloys, Balatonfured, Hungary, May 1990, pp. 95–116. 35. A.L. Dons. Z. Metallkde. 1985, vol. 76, pp. 609–612. 36. P. Skjerpe. Metall. Trans. A, 1987, vol. 18A, pp. 189–200.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification of Aluminum Alloys

77

37. G. Ghosh. In: G. Petzow and G. Effenberg (Eds.), Weinheim: VCH, 1992, vol. 5, pp. 394–438. 38. C. Hsu, K.A.Q. O’Reilly, B. Cantor, R. Hamerton. Mater. Sci. Eng. A, 2001, vols. 304–305, pp. 119–124. 39. Y.L. Liu, S.B. Kang, H.W. Kim. Mater. Lett., 1999, vol. 41, pp. 267–272. 40. M.E. Drits, E.S. Kadaner, E.M. Padezhnova, L.L. Rokhlin, Z.A. Sviderskaya, N.I. Turkina. Phase Diagrams of Aluminum- and Magnesium-Based Systems, Moscow: Nauka, 1977. 41. J.R. Davis (Ed.). Aluminum and Aluminum Alloys, ASM Specialty Handbook, Materials Park: ASM International, 1993. 42. S. Thompson, S.L. Cockroft, M.A. Wells. Mater. Sci. Technol., 2004, vol. 20, pp. 497–504. 43. X.-Y. Yan, Y.A. Chang, F.-Y. Xie, S.-L. Chen, F. Zhang, S. Daniel. J. Alloy Comp., 2001, vol. 320, pp. 151–160. 44. I.N. Fridlyander, L.L Rokhlin, T.V. Dobatkina, N.R. Bochvar, E.V. Lysova, I.G. Korolkova. In Metallovedenie i tekhnologiya legkikh spalvov (Physical Metallurgy and Technology of Light Alloys), Moscow: VILS, 2001, pp. 39–51. 45. K.A. Montoya, F.H. Heubaum, K.S. Kumar, J.R. Pickens. Scr. Metall. Mater., 1991, vol. 25, pp. 1489–1494. 46. A.K. Mukhopadhyay, V.V. Rama Rao, P. Ghosal, N. Ramachandra Rao. Z. Metallkde., 2000, vol. 91, pp. 483–488. 47. L.A. Tarshic, J.L. Walker, J.W. Rutter. Metall. Trans., 1971, vol. 2, pp. 2589–2597. 48. I. Maxwell, A. Hellawell. Acta Metall., 1975, vol. 23, pp. 229–237. 49. M.H. Burden, J.D. Hunt. J. Cryst. Growth, 1974, vol. 22, pp. 109–116. 50. T.E. Quested, A.T. Dinsdale, A.L. Greer. Acta Mater., 2005, vol. 53, pp. 1323–1334. 51. M.A. Easton, D.H. St. John. Acta Mater., 2001, vol. 49, pp. 1867–1878. 52. M.A. Easton, D.H. St. John. Metall. Mater. Trans. A, 1999, vol. 30A, pp. 1613–1623. 53. M.A. Martorano, C. Beckermann, Ch.-A. Gandin. Metall. Mater. Trans. A, 2005, vol. 34A, pp. 1657–1674. 54. K. Jackson, J. Hunt, D. Uhlmann, T. Seward. Trans. Metall. Soc. AIME, 1966, vol. 236, pp. 149–158. 55. D. Ruvalcaba, R. Mathiesen, D.G. Eskin, L. Arnberg, L. Katgerman. Acta Mater., 2007, vol. 55, pp. 4287–4292. 56. R.H. Mathiesen, L. Arnberg, P. Bleuet, A. Somogyi. Metall. Mater. Trans. A, 2006, vol. 37A, pp. 2515–2524. 57. M. Johnsson. Z. Metallkde., 1994, vol. 85, pp. 781–789. 58. T.E. Quested, A.T. Dinsdale, A.L. Greer. Mater. Sci. Technol., 2006, vol. 22, pp. 1126–1134. 59. M.A. Easton. Private communication, 2007. 60. L.S. Toropova, D.G. Eskin, M.L. Kharakterova, T.V. Dobatkina. Advanced Aluminum Alloys Containing Scandium: Structure and Properties, Amsterdam: Gordon & Breach, 1998. 61. M.A. Easton, D.H. St. John. Metall. Mater. Trans. A, 2005, vol. 36A, pp. 1911–1920. 62. L.K. Lamikhov, G.V. Samsonov. Tsvetn. Met., 1964, no. 8, pp. 79–82. 63. V.I. Napalkov, G.V. Cherepok, S.V. Makhov, Yu. M. Chernovol. Continuous Casting of Aluminum Alloys, Moscow: Intermet, 2005. 64. P. Schumacher, A.L. Greer, J. Worth, P.V. Evans, M.A. Kearns, P. Fisher, A.H. Green. Mater. Sci. Technol., 1998, vol. 14, pp. 394–404.

Copyright 2008 by Taylor and Francis Group, LLC

78

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

65. T.E. Quested. Mater. Sci. Technol., 2004, vol. 20, pp. 1357–1369. 66. P. Villars, L.D. Calvert. Pearson’s Book on Crystallographic Data for Intermetallic Phases, Metals Park, OH: ASM, 1985. 67. V.G. Davydov, T.D. Rostova, V.V. Zakharov, Yu. A. Filatov, V.I. Yelagin. Mater. Sci. Eng. A, 2002, vol. A280, pp. 30–36. 68. N.A. Belov, A.N. Alabin, D.G. Eskin, V.V. Istomin-Kastrovskii. J. Mater. Sci., 2006, vol. 41, pp. 5890–5899. 69. D.G. Eskin. Ph.D. (Candidate of Science) Thesis, Moscow, Moscow Institute of Steel and Alloys, 1988. 70. J. Fjellstedt, A.E.W. Jarfors, L. Svendsen. J. Alloys Comp., 1999, vol. 283, pp. 192–197. 71. A. Abdel-Hamid, F. Durnad. Z. Metallkde., 1985, vol. 76, pp. 739–746. 72. N. Frage, N.N. Frumin, L. Levin, M. Polak, M.P. Dariel. Metall. Mater. Trans. A, 1998, vol. 29A, pp. 1341–1345. 73. L. Svendsen, A. Jarfors. Mater. Sci. Technol., 1993, vol. 9, pp. 948–957. 74. P.S. Mohanty, J.E. Gruzleski. Acta Metall. Mater., 1995, vol. 43, pp. 2001–2012. 75. D. Turnbull. J. Chem. Phys., 1950, vol. 18, pp. 198–203. 76. N. Iqbal, N.H. van Dijk, S.E. Offerman, M.P. Moret, L. Katgerman, G.J. Kearley. Acta Mater., 2005, vol. 53, pp. 2875–2880. 77. A. Cibula, G.W. Delamore, R.W. Smith. Metall. Trans., 1972, vol. 3, pp. 751–754. 78. A. Banerji, Q.L. Feng, W. Reif. Metall, 1987, vol. 41, pp. 1237–1242. 79. G.I. Eskin. Ultrasonic Treatment of Light Alloy Melts, Amsterdam: Gordon & Breach, 1998. 80. G.I. Eskin, D.G. Eskin. Z. Metallkde., 2004, vol. 95, pp. 682–690. 81. Y. Han, K. Li, J. Wang, D. Shu, B. Sun. Mater. Sci. Eng. A, 2005, vol. A405, pp. 306–312.

Copyright 2008 by Taylor and Francis Group, LLC

3 Solidification Patterns and Structure Formation during Direct Chill Casting

3.1

Shape and Dimensions of the Billet Sump

The billet (ingot) during casting can be subdivided into several stages with distinctly different characteristics. We mentioned in Chapter 1 (Figure 1.6) the sump and the transition region. Other terms can be found in the literature, e.g., liquid pool, slurry zone, and mushy zone. There is also the solid part of the billet (ingot). We have to deine these terms because there are discrepancies in the literature about what is actually what. Figure 3.1a shows a schematic cross-section of a billet with isotherms of liquidus, coherency,* [1] and solidus, and the deinition of the different zones of the billet. The real sump proile is shown in Figure 3.1b. The dimensions and geometry of these regions are determined by temperature distribution that is the product of heat transport in and out of the system. Heat input is the energy introduced into the system by the melt (temperature and speciic heat of the liquid) and generated during solidiication (latent heat of solidiication and the speciic heat of the solid). Heat extraction occurs by convection of the melt in the sump; by conduction of heat through liquid, semi-solid, and solid part of the billet; and by convection of cooling water inside the mold and at the billet surface. The heat transport is usually characterized by the following two dimensionless numbers. The thermal Peclet number is Pe = ρcVcR/λ,

(3.1)

where ρ is the density, c is the speciic heat, Vc is the casting speed, R is the billet radius, and λ is the thermal conductivity. The Peclet number shows the ratio between heat convection (advection) and heat conduction. The Biot number is Bi = qR/λ,

(3.2)

* Coherency is a conditional term denoting the change in the slurry behavior, e.g., the change in the viscosity or heat conductivity related to the beginning of interaction between solid grains. The coherency isotherm can be also called the “solidiication front,” marking the macroscopically continuous front of the growing solid phase (see Section 3.2). A detailed discussion of the coherency phenomenon can be found elsewhere [1].

79 Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

80

Reference point at the bottom of hot top 1 L 30%

2

S 3

Melt temperature 707 deg.C Water rate 200 l/min Water temperature 28.5 deg.C Alloy Al−3.7% Cu Casting speed 7 cm/min

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

(a)

(b)

FIGURE 3.1 (a) Deinition of billet portions during casting (L = liquidus; S = solidus, and 30% = coherency isotherm): 1 = molten pool; 2 = transition region; (1 + 2) = sump; (2–3) = slurry zone; and 3 = mushy zone; and (b) real sump proile revealed by adding liquid Zn during casting: the lighter part shows the solid part of the billet at the moment of Zn addition and the darker part shows the sump.

where q is the heat transfer coeficient. The Biot number relects the ratio between heat convection (advection) at the surface and heat conduction inside the sample. Typical values for direct chill casting of aluminum alloys are 1.8 < Pe < 4.5 and 1 < Bi < 25 [2]. The temperature distribution in the billet sump depends on the melt temperature, melt low, and cooling conditions in the mold and at the billet surface and, therefore, is the function of process parameters such as casting speed, water low rate, melt temperature, and the melt distribution system. On the other hand, the dimensions and geometry of these regions determine to a great extent the formation of structure and casting defects during solidiication. It became clear early on in DC casting practice that there was a direct link between the casting speed and the depth of the sump as well as between the casting speed and the solidiication rate. These relationships were briely mentioned in Chapter 1, Equations 1.1 and 1.2, and they remain valid despite later advances in the analysis of DC casting. Roth [3] and Dobatkin [4] showed that the sump depth (H) depends on the casting speed (Vcast) and billet radius as H = AVcastR2/4λs(Tm − Tsurf),

Copyright 2008 by Taylor and Francis Group, LLC

(3.3)

Solidification Patterns and Structure Formation during Direct Chill Casting

81

where Tm is the melting temperature, Tsurf is the temperature of the billet surface or the temperature of the cooling medium, R is the billet radius, λs is the thermal conductivity of solid, and A is determined as follows: A = ∆Hf ρs + 1/2cs ρs(Tm – Tsurf),

(3.4)

where ∆Hf is the latent heat of fusion, ρs is the density of solid, and cs is the speciic heat of solid. Similar relationships can be written for sheet ingots using the ingot thickness instead of billet radius in Equation 3.3. These formulations, though derived based on the assumption of the constant surface temperature, have proved to be valid in practice. Figure 3.2a shows that the experimentally measured sump depth depends linearly on the

250

20% Cu

150

1% Cu

100 50 0 0

50 100 150 200 Casting speed, mm/min

250

Parameter of liquid pool and sump, mm

Sump depth, mm

160 5% Cu

200

(1+2)′

140

(1+2)

120 100

2

80 60 1

40 20 0 100

140 180 Casting speed, mm/min

(a)

(b)

100 Distance between liquidus and solidus, mm

220

120 mm/min; 150 l/min 200 mm/min; 150 l/min

80 60 40 20 0 0

20 40 60 80 Distance from center, mm (c)

100

FIGURE 3.2 Dependence of the sump depth in the center of the billet on the casting speed: (a) experimental data for 200-mm billets from three binary Al–Cu alloys [4]; (b) experimental sump depth (1 + 2)’ and calculated molten pool depth (1), transition region (2), and sump depth (1 + 2) for a 200-mm Al–4.3% Cu billet cast at two casting speeds (numbers are as in Figure 3.1a) [5]; and (c) the variation of the transition region vertical dimension along the billet radius for two casting speeds [5]. (Parts (b) and (c) reproduced with kind permission of Elsevier.)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

82

casting speed for different alloy compositions [4]. Figure 3.2b conirms this dependence by numerical simulations [5]. The sump depth normalized to the billet radius is also shown to increase linearly with the increasing Peclet number, which means that the sump depth is inversely proportional to the alloy thermal conductivity [2]. The direct consequence of Equation 3.3 is the rule that the ratio between the sump depth and the billet radius is constant if the following condition is valid: VcastR = const.

(3.5)

The dimensions of the transition region, however, do not change uniformly along the billet cross-section. The effect of casting speed is mostly felt in the central part of the billet, as demonstrated in Figure 3.2c. It is important to note that Equation 3.3 is derived and valid only for the steady-state regime of casting. If the casting speed is ramped up or down the sump does not change its shape and depth instantly, but rather with some thermal inertia as has been shown experimentally [8] and by numerical simulations [9]. Similar phenomena occur during the start-up phase of casting [6, 7, 10]. Let us look at this phenomenon in more detail. The typical beginning of DC casting involves either ramping of the casting speed from some low value (or zero) to the run speed according to the recipe, or starting the casting directly with the run casting speed. During this transient stage, the cooling from the direct water jet is felt in the center of the billet with a delay. As a result, the sump depth increases to an extent greater than in steady-state casting, and only after some time does the sump become shallower, approaching the steady-state depth. The “overshot” of the sump depth is a function of the run casting speed, as shown in Figure 3.3. The evolution of the sump depth at the beginning of casting depends on the start-up regime and on the geometry of the starting block [7, 11].

Sump depth, mm

100 110 mm/min 80 90 mm/min

60 40

75 mm/min

20 0

100

200 300 Billet length, mm

400

FIGURE 3.3 Experimentally measured sump depth in the start-up phase of DC casting of a 228-mm billet from a 6061 alloy as a function of the billet length and the run casting speed [6, 7]. The starting casting speed was equal to the run speed.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

83

Suggestions have been made to raise the central part of the starting block or make the starting casting speed higher than the run speed to diminish the initial deepening of the sump [7]. Figure 3.4 illustrates how the starting regime affects the sump depth. Gentle ramping of the casting speed to the run speed (regimes 2 and 4) decreases the “overshot” [11]. It is interesting to note that the length of the mushy zone changes in a manner similar to the total sump depth [7, 11], although the response of the mushy zone to the casting speed is stronger than that of the sump depth. During DC casting, the casting speed may change either stepwise or in the form of ramping, especially when experiments are performed. Several casting regimes have been simulated using the numerical model of DC casting and different casting regimes, as shown in Figure 3.5a [9]. Figure 3.6 demonstrates velocity vectors with superimposed solid fraction contours upon casting with regimes 1 and 2 denoted in Figure 3.5a. All of these vector ields characterize four distinct low zones: the bulk liquid zone (solid fraction 0.0), the slurry region (between solid fractions 0.0 and 0.3), 160 3

Casting speed, mm/min

150 140

4

130 1

120 110

2

100 90 80 0

100

(a)

200

300

400

160

Length of the mushy zone, mm

180 Sump depth, mm

600

80

200

3

140

4

120

2

100

1

80 60 40 20

70 60

3

50

4

1 2

40 30 20 10 0

0 0 (b)

500

Billet length, mm

50

100

150 200 Time, s

250

0

300 (c)

50

100

150 200 Time, s

250

300

FIGURE 3.4 Effect of the start-up regime (a) on the sump depth (b) and the length of the mushy zone (c) during casting of a 200-mm billet from an Al–4.5% Cu alloy (computer simulations) [11]. (Reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

5

6

2

100

50

Ramping Depth (m) Steady Depth (m) Measured Depth Casting Speed

0

10

20

30

40

(b)

50 60 Time, s

70

80

90

100

Casting speed, mm/min

4

150

Sump Depth, m

1,3

2, 3

200 190 180 170 160 150 140 130 120 110 100 110

0 0

200

400

(a)

600 Time, s

800

1000 0.17

6

5

0.15

0.17 0.16 0.15 0.14 0.13 0.12 0.11 0.1 0.09 0.08 0.07

Regime 1 steady state

Sump Depth, m

Sump Depth, m

0.16

Regime 3

2

0.14 0.13 4

0.12 0.11 0.1 0.09 0.08

0

20

40

(c)

60

80 Time, s

100

120

140

0.07 100

160 (d)

119

138

157

175 200 187 168 Casting speed, mm/min

149

130

111

100

FIGURE 3.5 Calculated dependences of the sump depth in the center of the billet on the casting speed and ramping rate upon casting of a 200-mm Al–4% Cu billet: (a) casting regimes at a casting temperature of 675°C with 1 = steady state casting at 200 mm/min, 2 = ramping at 113.2 mm/min2, 3 = ramping at 113.2 mm/min2 then casting at 200 mm/min, 4 = ramping at 56.6 mm/min2, 5 = ramping at 34.9 mm/min 2, and 6 = ramping at 11.32 mm/min 2; (b) evolution of the sump depth in a transient regime 2 with experimentally measured values shown by triangles; (c) evolution of the sump depth in regimes 1 (steady state) and 3; and (d) evolution of the sump depth at different ramping rates [9]. Vertical lines in (b) and (d) show the position of the maximum casting speed. (Reproduced with kind permission of Elsevier.)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Casting speed, mm/min

1 200

0.17 0.16 0.15 0.14 0.13 0.12 0.11 0.1 0.09 0.08 0.07

84

250

Solidification Patterns and Structure Formation during Direct Chill Casting

(a)

85

(b)

FIGURE 3.6 Computer-simulated low patterns together with the contours of solid fraction at values of 0.0, 0.3, and 1.0, for regimes 1 (a) and 2 (b) after 53.5 s of casting as shown in Figure 3.5a,b. Only half of the 200-mm Al–4% Cu billet is shown: the left-hand side of the igures is the billet center and the right-hand side is the surface.

the mushy zone (between solid fractions 0.3 and 1.0), and the solid zone (solid fraction 1.0). Below the solidus contour is the solid zone, in which solid moves uniformly at the casting speed. The thermal inertia in the transient state is obvious from the comparison of the steady-state situation shown in Figure 3.6a with the transient state shown in Figure 3.6b. This thermal inertia is quantiied by calculating the evolution of the sump depth, as shown in Figure 3.5b. After 53 sec of casting when the peak casting speed is reached during ramping, the depth of the sump is about 4 cm lower than the steady-state depth, shown as a horizontal dashed line in Figure 3.5b. It is interesting to note that although casting speed begins to decrease after this peaking point, the depth of the sump keeps increasing up to 14.2 cm, still below the steady-state value. These calculation results are qualitatively similar to the experimental results reported by Commet et al. [8], though the quantitative differences in the sump depth response to the changing casting speed is, obviously, a function of the particular casting conditions. To answer the question as to how long it would take for this transient state to reach steady state, calculation 3 was performed. In this regime, the casting

Copyright 2008 by Taylor and Francis Group, LLC

86

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

speed is maintained constant and equal to 200 mm/min after the transient stage of ramping. All other calculation parameters are the same as in regime 2. The evolution of the sump depth is shown in Figure 3.5c. After 45 sec of simulated casting time (or 150 mm of the casting length) in calculation 3, the same sump depth as in steady-state calculation 1 is reached. Ramping rate is an obvious controlling parameter that determines the magnitude of the thermal inertia. Calculations 4, 5, and 6 (corresponding casting regimes shown in Figure 3.5a) were made to ind the critical ramping rate at which no thermal inertia exists. The sump depths as a function of casting speed in these calculations are shown in Figure 3.5d. Almost realtime response of the sump depth to the ramping casting speed is achieved in casting regime 6 at a ramping rate of 11.32 mm/min2, while regime 2 shows a very big lag. Two more casting parameters may affect the sump depth, i.e., cooling water low rate and melt temperature. Both parameters may potentially inluence the thermal conditions inside the mold. The water low promotes heat extraction from the melt whereas the melt temperature adds heat to the system. The dimensions of the sump and transition region as functions of cooling intensity characterized by the Biot number (Equation 3.2) have been assessed by numerical and physical modeling [12]. The sump depth varies, depending on the cooling intensity, as shown in Figure 3.7a. For Bi < 4, increased cooling results in a signiicantly decreased sump depth, whereas further intensiied cooling has only slight consequences for the depth of the sump and other characteristic dimensions. The temperature of the cooling liquid can also affect the sump depth and the thickness of the transition region, as demonstrated in Figure 3.7b. This inluence becomes considerable, however, only at coolant temperatures above 70°C [12]. The dependences shown in Figure 3.7 demonstrate that it is not possible to increase heat extraction beyond a certain limit that is a function of thermal properties of the alloy and the cooling medium. Few data are available in the literature on the effect of water low rate on sump geometry. Most of the research has been done on the interrelation between water low rate, water temperature, heat transfer coeficient, and the so-called Leidenfrost temperature that corresponds to the point at the billet surface where the evaporation of water takes the longest time and the heat lux density is minimum [13]. It is important to understand that actual water low rate in DC casting can be safely decreased only to a certain value because otherwise the solid shell will not be formed inside the mold and the melt will bleed out, which is dangerous. On the other hand, the increase of the water low rate above a certain value is useless because the water cannot extract more heat than its heat capacity and heat conductivity allows (see Figure 3.7a). Therefore, the actual range of water low variation is rather narrow. The Leidenfrost temperature can be shifted to higher surface temperatures by increasing water low rate [13]. The best conditions of heat extraction are created when so-called nucleate boiling occurs during direct water chill of the billet surface. In this case the heat transfer coeficient can be on the

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

87

Normalized sump parameter

1.2 1 0.8 0.6 1+2

0.4

1+2−3 0.2 1 0 0

5

10

(a)

15 Bi

25

20

30

Normalized sump parameter

0.6 1+2

0.5 0.4 0.3

2

0.2 0.1 0 0

50

(b)

100 Cooling water temperaure, °C

150

200

FIGURE 3.7 Effect of cooling intensity (a) and cooling media temperature (b) on the sump parameters normalized to (VcR2) in the center of a round 2024 billet [12]. (1) Molten pool depth, (2) transition region, (1 + 2) sump depth, (1 + 2 – 3) slurry zone. Numbers are as in Figure 3.1a.

order of 40 kW/m2 K with corresponding heat laxes up to 6000 kW/m2 [2]. The conditions of nucleate boiling are maintained during steady-state casting because the surface temperature stays below 300°C. Under these conditions the inluence of the water low rate is considered negligible because of very small effects of water low rate on the heat lux density and heat transfer coeficient [13]. However, during the start-up phase of DC casting when the water low rate is kept lower than during steady-state casting in order to decrease thermal gradients and thermal stresses, partial or even stable ilm boiling may occur, and the water low rate does have an effect [13].

Copyright 2008 by Taylor and Francis Group, LLC

88

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Typically, experimental studies are performed under laboratory conditions, using a hot slab with thermocouples at various positions at the surface and inside the slab, and a water source moving along the slab at a certain speed. However, Wells et al. [14] note that laboratory-based studies cannot directly reproduce heat transfer behavior in the industrial DC casting process. Prasso et al. have shown that the water low rate has only a small effect on the sump depth, i.e., the decrease of the water low rate by 20% causes the deepening of the sump by only 4% [15]. At the same time, the relative inluence of the water low rate on temperature distribution in the sump is more pronounced at the periphery of the billet [15]. Our results shown in Figure 3.8a,b demonstrate that the water low rate can inluence the parameters of the sump, but only at low casting speeds. The inluence of melt temperature on the geometry of the sump during DC casting has been studied using computer simulations, model experiments, and pilot-scale experiments but only within narrow margins of temperature variation. The golden rule of the cast shop is “melt hot, cast cold” because high melt temperatures are known to produce coarser structure (see Chapter 2, Figure 2.5) with greater porosity and to increase the probability of bleed-outs. Tarapore [16] reported experimental and computer-simulation results on DC casting of a 2024 alloy. The melt temperature varied from 660 to 715°C in the trough (the resultant melt temperature in the hot top changed from 642 to 696°C). The increased depth of the sump, higher temperature gradients in the liquid bath, and a thinner surface liquation layer corresponded to a higher melt temperature. Some results are available for computer simulation of DC casting under different process conditions, including the melt temperature. Reese [17] tested an analytical model for liquid aluminum low in the sump of a DCcast round billet at different superheats, from 30 to 70 K. The increase in melt superheat was shown to increase the depth of the sump and the melt low velocity along the mushy zone. At the same time, the thickness of the mushy zone and the upward melt low velocity (in the central part of the sump) remained virtually unaffected. A conclusion was made that the deep sump resulted from high melt superheat. Shestakov [18, 19] modeled the effect of melt superheat (∆T = 0–300 K) on the kinetics of progressive solidiication and the parameters of the transition region (between liquidus and solidus isotherms) of an aluminum alloy. The increase in melt temperature was shown to narrow the transition region in the billet and to accelerate the solidiication (local solidiication time is shortened). The optimum superheating for aluminum alloys was recommended as 150–200 K in contrast to the usual 60–80 K. However, this effect was only pronounced in alloys with low solidiication rate (when the solid phase was formed evenly throughout the solidiication range). We studied the effects of melt temperature on the temperature distribution in the liquid pool and on the sump geometry in DC casting of a binary Al–2.8% Cu alloy [20]. Both experiments and computer simulations were performed. The melt temperature in the furnace varied between 700 and

Copyright 2008 by Taylor and Francis Group, LLC

140

140

120 100 80

Exp 1+2

60 40 20

2

Parameter of liquid pool and sump, mm

(a)

180

220

120 100

2

80 60 40

1

(b)

160 120 80

1+2 2 1 3

40

725 750 Melt temperature, °C

0 140

260

Water flow, l/min

0 700

1+2

20

1

0 140

(c)

Parameter of liquid pool and sump, mm

Exp

Parameter of liquid pool and sump, mm

Parameter of liquid pool and sump, mm

160

775 (d)

180

220

260

Water flow, l/min

160

1+2

120

2

80 1

40

3 0 700

725 750 Melt temperature, °C

775

Copyright 2008 by Taylor and Francis Group, LLC

89

FIGURE 3.8 Effect of water low (a,b) and melt temperature (c,d) on the parameters of the liquid pool and sump at the casting speed 120 mm/min (a), 100 mm/min (c) and 200 mm/min (b,d) for the central portion of 200-mm billets from an Al–4.3% Cu alloy (a,b) and an Al–2.8% Cu alloy (c,d) [5, 20]. (1) Molten pool depth, (2) transition region, (1 + 2) sump depth, (3) mushy zone. The numbers are as in Figure 3.1a. These results were obtained from computer simulations of DC casting validated by experiments. “Exp” designates the experimentally measured sump depth. (Reproduced with kind permission of Elsevier (a,b) and Springer Science and Business Media (c,d).)

Solidification Patterns and Structure Formation during Direct Chill Casting

160

90

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

(a)

(b)

FIGURE 3.9 Computer-simulated low patterns and positions of liquidus, coherency and solidus isotherms in 200-mm Al–4% Cu billets cast at 200 mm/min with the melt temperature at the inlet to hot top 675°C (a) and 725°C (b) [9]. Only half of the billet is shown: the left-hand side of the igures is the billet center and the right-hand side is the surface. (Reproduced with kind permission of Elsevier.)

760°C, which corresponded to the range of 660°C and 740°C at the inlet to the hot top. The results are given in Figures 3.8c,d and 3.9. These data are taken from the computer-simulated patterns and are validated by temperature and sump depth measurements [20]. The dimensions and the position of the transition region in the billet (in relation to the hot top or to the mold) show the following trends. Increasing the melt temperature shifts the positions of liquidus and solidus isotherms in the center of the billet downward; however, the liquidus position is affected to a greater extent than the solidus position. This effect is understandable if one takes into account that the incoming hot melt introduces more heat into the system and extends the liquid part of the billet, whereas the cooling in the mold is eficient enough to form the solid shell within the limits of the mold. The transition region narrows in the center of the billet and at the periphery with increasing melt temperature. The dimensions of the mushy zone remain virtually independent of the melt temperature. In addition, the simulation of melt low patterns (Figure 3.9) shows that higher melt temperatures produce more severe melt low toward the surface and deeper penetration of the low into the mushy zone compared to lower melt temperatures.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

91

0.5 Normalized transition region

Bi=2.4 0.4

Bi=4.3 Bi=6.9 Bi=24.3

0.3 0.2 0.1 0 0

25

50 75 100 Solidification range, °C

125

150

FIGURE 3.10 Effect of the alloy solidiication range and the intensity of cooling on the thickness of the transition region in the center of the billet normalized to (VcR2) [12].

The inal factor that affects the shape and the dimensions of the sump is alloy composition. It is obvious that different alloys have different thermal properties and different solidiication ranges that determine the positions of the liquidus and solidus isotherms in the billet. The structure of the solidifying alloy, e.g., columnar or equiaxed grains, coarse or ine grains, may have a considerable effect on the position of the coherency isotherm and, therefore, the dimensions of the mushy zone. Figure 3.10 demonstrates that the greater the solidiication range, the wider the transition region in the billet. In the range of smaller solidiication ranges this dependence is almost linear. For alloys with a greater solidiication range, the transition region widens more quickly. The increased intensity of cooling, characterized by the Biot number, can narrow the transition region.

3.2

Solidification Rate and Cooling Rate during DC Casting

One of the most prominent features of the DC casting process is that the sump does not change shape and position during steady-stage DC casting. The stability of the sump shape can be expressed by Equation 1.2, which we repeat here: Vs = Vcastcos αn, where αn is the angle between the billet axis and the normal to the solidiication front. The steadiness of the sump during progression of the solidiication

Copyright 2008 by Taylor and Francis Group, LLC

92

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Solidification rate, mm/min

250 200 200 mm/min

150 100 120 mm/min

50 0 0

20 40 60 80 Distance from billet center, mm

(a)

100

(b)

FIGURE 3.11 (a) Scheme explaining the interrelation between the casting speed (Vcast), solidiication rate (Vs) and sump proile (αn) [53] and (b) the change of the solidiication rate along the billet radius for the two billets described in Figure 3.2b,c [5].

front with the velocity Vs can be compensated for by the withdrawal of the solid part of the billet with the casting speed Vcast only if this relationship holds. The relationship between the shape of the sump and the solidiication rate Vs is schematically illustrated in Figure 3.11a. The solidiication rates calculated based on the computer simulation of two billets cast at two casting speeds are shown in Figure 3.11b [5]. The maximum solidiication rates are, therefore, reached at the periphery and in the center of the billet, where the solidiication front is lat. The minimum solidiication rate occurs at about midradial position, where the solidiication front is the steepest [12]. The increase in casting speed affects mostly the solidiication rate at the periphery and in the center of the billet [5, 12]. It has been shown that the average solidiication rate of the entire solidiication front across the billet has limits beyond which it cannot be increased any more [4, 21]. This maximum average solidiication rate is a function of the billet radius and the thermophysical properties of the alloy [4]: Vmax = 4λ(Tm – Tsurf)/∆HρR, s

(3.6)

where λ is the thermal conductivity of the solid, ρ is the density of the solid, ∆H is the difference in the enthalpies of the liquid at the melting point Tm and the solid at a temperature (Tm − Tsurf)/2, Tsurf is the surface temperature,

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

93

Solidification rate, mm/min

100 R = 97.5 mm Vs max = 103 mm/min

80

60 R = 185 mm Vs max = 54 mm/min

40

20

0 0

50

100 150 Casting speed, mm/min

200

250

FIGURE 3.12 Scheme illustrating the concept of the maximum average solidiication rate for two billets of a 2024 alloy [4].

and R is the billet radius. This relationship, which is shown in Figure 3.12, is derived with the assumption that the sump has a conical shape. The concept of solidiication rate is based on the idea that the solidiication front is formed by solid grains that grow continuously from and in normal direction to the solidus; hence, the solidiication front is continuous. Only in this case, the solidiication rate can be taken as a parameter that determines the formation of structure, as we discussed in Chapter 2. The reality is, however, more complicated. Commercial alloys form mostly equiaxed grain structures. This means that the small equiaxed grains loat freely in the upper part of the transition region (the slurry zone). With the increasing fraction of solid and with decreasing temperature, these grains start to “feel” and touch one another, being still quite loose. Eventually, the grains start to tangle and bridge, forming a weak but macroscopically (and statistically) coherent network. At this temperature and in this portion of the billet, a macroscopically continuous solidiication front is formed. This network is, however, still microscopically quite discontinuous, with grain detachment and fragmentation occurring. Finally, at rather high fractions of solid, a rigid skeleton of the solid phase is formed. Then the solid phase is continuous on both macroscopic and microscopic levels. Therefore, the concept of solidiication rate can be applied to the formation of structure in the mushy zone, i.e., between the coherency and solidus isotherms, rather than to the general description of structure formation upon DC casting. The other parameter that is often used in practice and that can be seemingly easily measured is the cooling rate. We discussed in Chapter 2 the importance of the cooling rate for structure formation during solidiication. The typical set-up for measuring cooling rate during DC casting consists of

Copyright 2008 by Taylor and Francis Group, LLC

94

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

a rack of long thermocouples arranged in a regular manner along the billet diameter. These thermocouples are dipped into the liquid pool at a rate equal to the casting speed. Eventually, the tips of the thermocouples reach the slurry zone, then the mushy zone and, inally, are frozen into the solid part of the billet. Temperature readings from these thermocouples can be used to locate the isotherms of liquidus and solidus. If we know the time of passing the transition region, we can calculate the cooling rate. A scheme for such a set-up is shown in Figure 3.13, along with the actual rack used in our experiments. The cooling rate measured in this way should correspond well to the dimensions of the transition region, which in turn relects the cooling conditions of the billet. Previously, we assumed that the cooling of the billet occurs uniformly at the surface. In reality, there are several distinct cooling zones with different heat transfer conditions and different cooling rates, as illustrated in Figure 3.14a,b. Melt starts to cool and solidify inside the watercooled mold (primary cooling). The solid shell that is formed contracts as a result of the temperature-dependent density of the solid phase. This contraction is almost free as the interior of the billet is still liquid and does not offer much resistance. Consequently, the surface of the billet pulls away from the mold with the formation of so-called air gap. The heat transfer coeficient immediately drops several times, e.g., from 1–2.5 to 0.5 kW/m2K, and the billet undergoes reheating from the incoming hot melt and the latent heat of solidiication. In the worst scenario, this reheating can reach the point of

Rack with thermocouples

Hot top Inlet

Mold

Billet

Casting direction, Vc (a)

(b)

FIGURE 3.13 Scheme (a) of a rack for measuring temperatures during DC casting and a photo (b) of a similar set-up used for measuring temperatures in the sump and billet during DC casting.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

95

MENISCUS BASE MUSH

AIR-GAP

SECONDARY

SOLID

Temperature, °C

MOLD COOLING

REHEATING ZONE

COOLING

750 700 650 600 550 500 450 400 350 300 0.4

1 4

0.5

3

2

0.6 0.7 Billet length, m

0.8

0.9

(c)

(a)

4

0.25 cooling

3

1.0

2

Primary

1

2.5 1.5

L

1.0

6.0 11.7 19.0

16.0 14.0

2.7

505 °C

Downstream Impingement

1.7 615 °C

Billet length

645 °C

Air gap

0.25

Tcoh

S

2.2

(b)

(d)

FIGURE 3.14 Different cooling zones in DC casting [2] (a); experimental data on cooling rates (K/s) measured during DC casting of a 2024-alloy 330-mm billet, casting speed 75 mm/min [12] (b); cooling curves obtained upon computer simulation of DC casting of an Al–4% Cu 200-mm billet, casting speed 200 mm/min (c); and the corresponding sections of the same billet with liquidus (L), coherency (Tcoh), and solidus (S) isotherms, with billet surface on the right (d).

Copyright 2008 by Taylor and Francis Group, LLC

96

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys TABLE 3.1 Coeficients α and β for Different Surface Temperature Ranges in Aluminum DC Casting [22] Temperature Range,°C 150

α, 104 m2/W°C

β, 106°C

2.73 9.43 1.23

–1.27 –9.24 3.06

remelting with subsequent bleed-out of the melt through the remelted shell. The shell eventually goes out of the mold and at some point the cooling water impinges onto the billet surface (direct chill, secondary cooling), increasing the heat transfer by one or two orders of magnitude, e.g., to 48 kW/m2K. Finally, the surface of the billet is cooled by free-falling water with the heat transfer coeficient ranging from 30 to 40 kW/m2K. The heat transfer coeficients in the regions of water impingement and downstream cooling are the functions of the surface and water temperatures and the cooling-water rate as shown by Equations 3.7 and 3.8 for the case of aluminum DC casting [22]. Heat lux in the zone of impingement can be described as q/A = αT – β,

(3.7)

where q is the heat transferred through the surface A, and the coeficients α and β are shown in Table 3.1. Heat lux in the downstream cooling region can be itted by the following formula: q/A = (–1.67 × 105 + cT*)Qw1/3∆T + 100∆Tx3,

(3.8)

where c = −4.01 × 106 Q2w + 6.9 × 104Qw + 628; T* is the average of the water and the surface temperatures in K; ∆T is the difference between these temperatures; ∆Tx is the difference between the surface temperature and the saturation temperature of water (taken as 90°C); and Q w is the water low rate in m3/m s. As a result of such uneven cooling, different sections of the transition region solidify under different cooling conditions, as relected in structure formation. The consequences for structure formation will be considered in the next section of this chapter. Direct measurement [12, 23] and computer simulation of the coolingrate distribution [9] in cross-sections of lat ingots and round billets yield consistent results, as illustrated in Figure 3.14c,d. High cooling rates close to the billet surface gradually decrease toward the central portion of the billet (ingot). On approaching the central part, however, the cooling curve exhibits two distinct regions. Very slow cooling in the upper part of the transition region, appearing as an almost horizontal plateau on the cooling curve

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

97

(Figure 3.14c, curves 3 and 4), gives way to a much faster cooling in the mushy zone. The remarkable feature of this transition is that the cooling curve in the mushy zone is steeper in the center (curve 4) than at midradius (curve 3). Hence, there are two features to be addressed. The irst is the appearance and extent of the plateau at temperatures close to the liquidus, and the second is the “faster” cooling in the central part of the mushy zone versus intermediate radial (thickness) location. There are several reasons for these observations. First, the overall shape of the cooling curve can be explained by the extension of the slurry region. The temperature variation within the slurry region is rather small whereas its width increases toward the billet center (Figures 3.1a, 3.2c). The rate of solid-phase formation in aluminum alloys is, however, very high, which means that more than 50% of the solid phase can be formed within a few degrees below the liquidus (see Figure 2.8). The formation of the solid phase produces latent heat of solidiication and pumps additional thermal energy into the slurry zone, effectively slowing down the cooling. As a result, a nearly isothermal plateau appears in the irst portion of the cooling curve and lengthens on approaching the billet center. Second, the central part of a casting is formed in the last stage of solidiication. This means that the amount of latent heat that is released in a particular horizontal cross-section of the billet decreases toward the bottom of the sump, as most of the lower cross-section already has been solidiied and no longer releases latent heat. Therefore, the cooling in the central–bottom part of the mushy zone (second portion of the cooling curve) is more eficient than in the slurry zone. Four factors may further enhance the acceleration of cooling in the central part of the billet compared to an intermediate radial (thickness) position. (1) The lower part of the sump (central portion of a billet) is formed, in most cases, in the range of secondary and downstream cooling (Figure 3.14a,b), where the heat transfer is the highest. (2) In addition, the heat is extracted from this part of the billet mostly through the solid phase, which has a higher heat conductivity than the liquid. (3) Some inluence can be imposed by melt low in the liquid/slurry part of the billet (see, e.g., Figure 3.9). Hotter melt penetrates the slurry region about the midradius, possibly impeding the cooling eficiency, whereas in the central–bottom part of the slurry zone, the colder melt is driven upward, additionally cooling the melt in this section of the billet. (4) And, inally, the high solidiication-front velocity in the center of the billet (Figure 3.11) may narrow the mushy zone as a result of a higher upward velocity of the solidus isotherm compared to the liquidus velocity. We should, however be very careful in attributing the measured cooling curve to the actual cooling rate in the solidiication range. It is true that if we take the temperature versus casting length and assume the constant casting speed, we can recalculate the data in terms of temperature versus time. This “cooling rate” can be related to the structure formation only if the solid phase follows the tip of a thermocouple, in other words, travels from the liquidus to the solidus at the casting speed. The actual situation in the sump of the billet (ingot) is more complicated with thermo-solutal convection and gravity

Copyright 2008 by Taylor and Francis Group, LLC

98

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

involved and with the resultant complex low patterns that exist in the liquid and slurry zones of the billet (see Figures 3.6 and 3.9). This situation is relected in the real structure observed in different sections of the billet (ingot). As we know, the cooling rate during solidiication can be also estimated by examination of the as-cast structure. The measured dendrite arm spacing can be recalculated to the cooling rate using Equation 2.3 with the preliminary determined coeficients. In this case we assume that the structure relects its solidiication history. Let us look at the distributions of cooling rate estimated by different methods in DC-cast billets 200 mm in diameter from an Al–4.3% Cu alloy produced at 120 and 200 mm/min. The variation of the transition region size for these billets is shown in Figure 3.2c, and the estimated solidiication rate is shown in Figure 3.11b. Figure 3.15a,b shows the cooling rates recalculated from the distance between liquidus and solidus isotherms (dashed lines) and from the measured dendrite arm spacing (solid lines). The former rates relect the expected measurements from thermocouples, and the latter relect the solidiication history. Although the “thermocouple” results fall in the same range as the “structural” cooling rates, they fail to reproduce the effect of the casting speed on cooling rates in the central part of a billet. The larger distance between liquidus and solidus isotherms (Figure 3.2c) is fully compensated for by the increased casting speed, which is in direct contradiction with experimentally observed difference in structure (see next section of this chapter). Computer simulation shows that a low pattern that exists in the transition region may drag solid crystals nucleated at the periphery of the billet to its center. As a result, at least some of the grains found in the central part of the billet are solidiied during a longer time (and have coarser internal structure) than might be presumed based only on their inal position in the billet. Examples of streamlines showing simulated paths of “particles-tracers” in different parts of the billet during solidiication are shown in Figure 3.16 for the two given casting speeds. Cooling rates calculated using these, quantiied in time streamlines, are given in Figure 3.15b. Obviously, the low patterns existing in the slurry zone cause scatter in the solidiication times and, as a result, scatter in structure parameters related to the cooling rate. The cooling rate calculated using the streamlines shows that closer to the surface of the billet the cooling rates agree well with the “thermocouple” cooling rate, whereas closer to the center of the billet the cooling rates range from “thermocouple” to “structural,” relecting the mixed history of structure evolution in the central portion of the billet. The grains that travel from the periphery to the central portion of the billet do not have a ine-cell structure, as has been assumed by Chu and Jacoby [24]. The fact that they originate from the region with a higher cooling rate seems to be completely overruled by the longer total solidiication time. Some of these grains can actually go upward into the hotter region of the sump and partially remelt and considerably coarsen while being suspended in the slurry region. Generally, the thermal history of solid phase that loats within the transition region causes the formation of grains with different dendrite arm

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

99

25

Cooling rate, K/s

20

200 mm/min

15 120 mm/min 10

5

0 0

20

40 60 Distance from billet center, mm (a)

80

100

25 200 mm/min

Cooling rate, K/s

20

15 120 mm/min 10

5

0 0

20

40 60 Distance from billet center, mm (b)

80

100

FIGURE 3.15 Cooling rate along the billet radius (billet center on the left) estimated using (a) the transition region size (dashed lines) and the dendrite arm spacing (solid lines) and (b) streamlines shown in Figure 3.16. Billets 200 mm in diameter from an Al–4.3% Cu alloy produced by DC casting at casting speeds 120 and 200 mm/min, with water low rate 150 l/min [5]. (Reproduced with kind permission of Elsevier.)

spacing (cell size), depending on the overall solidiication time and the transition region where these grains spent more time. Coarse-cell structures can be expected in grains that grow in the upper part of the transition region and then settle to the bottom of the sump, whereas much i ner cell structures relect grains that grow in the lower part of the sump where the cooling rates are higher. This may result even in the formation of a duplex internal structure of grains with coarse core and i ne periphery, as shown in Figure 3.17a. In this case, one can imagine that the grain spent

Copyright 2008 by Taylor and Francis Group, LLC

100

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

(a)

(b)

FIGURE 3.16 Streamlines illustrating “tracer” paths during DC casting of the billet described in Figure 3.15 with casting speed (a) 120 mm/min and (b) 200 mm/min [5]. (Reproduced with kind permission of Elsevier.)

FIGURE 3.17 Typical duplex structure found in the center of a 200-mm billet from an Al–4.3% Cu alloy cast at a casting speed of 200 mm/min: (a) ine periphery of some coarse-cell grains, shown by arrows; (b) a mixture of coarse-cell and ine-cell grains.

considerable time in the slurry region and then settled down and inished solidiication at a much higher rate. Another phenomenon can also inluence the inal dendrite arms spacing in grains. We have discussed in detail the development of dendrite structures

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

101

in Chapter 2. It has been shown that the dendrite arm spacing actually decreases in the upper part of solidiication range due to intensive branching of grains (Figure 2.13b). Only at high fractions of solid, when the structure becomes coherent, do coarsening and coalescence of dendrite branches take over [25]. Considered from the perspective of the cooling curves in Fig ure 3.14c, one can argue that dendrite cells in the billet center are iner because they have had less time to coarsen in the lower part of the solidiication range where the cooling rate was higher. This logic holds, however, only if the grain in question has passed the transition region at a uniform speed without delays. Only in this case does the cooling curve relect the cooling rate. This ideal scenario, however, is apparently seldom realized in reality. We can conclude that the structure found in the central part of a solidiied billet has characteristics relecting its solidiication history. In particular, the dendrite arm spacing corresponds to the solidiication time required for a particular section of the grain to form. Therefore, attributing the cooling rates experimentally measured by a thermocouple moving with a billet to the structure found in the position of this thermocouple in the solid billet is not always correct. Flow patterns, transport of the solid phase within the slurry region, and the shape of the cooling curve should be taken into account. Ultimately, this explains the reinement of dendrite arm spacing and the occurrence of “coarse-cell” grains (Figure 3.17) that are frequently found in the central part billets and ingots. This will be discussed in more detail later in this chapter.

3.3

Effects of Process Parameters on the Formation of Grain Structure

The distribution of structure parameters in the cross-section of DC-cast billets and ingots is very important for the occurrence of defects and performance of as-cast metal during subsequent deformation. Despite that, relatively little is known about structure formation under different process conditions during DC casting. In this section we will examine how the main process parameters that we have considered in previous sections, i.e., casting speed, water low rate and melt temperature, inluence the grain size and dendrite arm spacing in billets. We already showed in Sections 2.1 and 2.2 of Chapter 2 and 3.2 of this chapter that the structure is mainly controlled by nucleation and growth conditions, which, in turn, are governed by composition and cooling conditions. In the previous section we discussed two parameters—solidiication rate and cooling rate—that could be the main factors in the structure formation, provided that the alloy composition is not changed. A simple comparison of solidiication rate (Figure 3.11b) and cooling rate (Figures 3.14 and 3.15) distributions suggests that the dendrite arm spacing, according to Equations 2.2 and 2.3, should change differently along

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

102

Center

Surface

the billet radius. A zone of i ne grains with ine dendrite branches is formed at the surface of the billet in the region of direct contact with the mold, and then a zone of coarser structure appears in the billet section when the transition region passes the air gap region. Further inside the billet, another zone of iner grains appears when solidiication occurs in the region of secondary cooling (direct chill). There is a general structure coarsening toward the center of the billet, though there are experimental observations showing the dendrite cell reinement in the very center, which is not surprising if we take into account our analysis of cooling rate variation in the previous section of this chapter. One should note that this reinement is usually observed when careful separation is made between ine-cell grains and coarse-cell grains, otherwise the average dendrite cell size (or dendrite arm spacing) always coarsens toward the center of a billet. Grain size in equiaxed structures typically increases toward the center of a billet. If the structure changes from ine grained at the periphery through a zone of columnar grains to equiaxed in the center (which is a common case for high-pure and not grain-reined alloys), then one can consider the columnar-to-equiaxed transition as the decrease in grain size in the central portion of a billet. A typical example of different structure zones in a DC-cast billet is given in Figure 3.18. Note the duplex structure in the center of the billet (see also Figure 3.17). Figure 3.19 demonstrates the distribution of grain size and dendrite arm spacing for different casting speeds and water low rates for a binary

Primary cooling

Air gap

Secondary cooling

Downstream cooling

(a)

(b)

(c)

(d)

FIGURE 3.18 Different structure zones relecting different cooling conditions: (a) macrostructure of a 200-mm billet from an Al–2.8% Cu alloy cast at 200 mm/min and casting temperature 740°C with the photos taken at equal distances along the billet radius; (b–d) microstructures of a 200-mm billet from an Al–4.3% Cu alloy cast at 120 mm/min and casting temperature of 715°C taken at 10 mm from the surface (b), 55 mm from the surface (c) and 100 mm from the surface (d).

Copyright 2008 by Taylor and Francis Group, LLC

0 16 0 12 , mm ion ect

80 s sros

40 c let Bil

30 28 26 24 22

(c)

0

20

0

40 let Bil

1 n, 80 ectio s s s cro

16 mm

30 28 26 24 22

0

20

19 Ca 18 sti 17 ng sp 16 ee 15 dc m 14 /m 13 in 12

200

180 0 25 40 2 30 2 20 W ate 2 210 0 rf low 20 90 rat 1 180 0 e, l/m 17 60 in 1 50 (b) 1

DAS, µm

Grain size, µm

(a)

DAS, µm

Grain size, µm

19 Ca 18 7 sti ng 1 6 sp 1 5 ee 1 dc 4 m/ 1 3 mi 1 n 12 0

220

25 0 12 , ion t c e

0

80 s-s 40 cros t e l Bil

0 18 mm

(d)

0 24 30 W 2 220 0 at er 21 0 0 flo w 2 90 ra 1 80 te , l/ 1 170 0 m in 16 50 0 1

0

16 0 12 mm n, ctio

80 e -s 40 ross c t le Bil

FIGURE 3.19 Effect of casting speed (a,c) and water low rate (b,d) on distribution of grain size (a,b) and dendrite arm spacing (c,d) in DC-cast, 200-mm billets of an Al–4.3% Cu alloy [5]. DAS-dendrite arm spacing.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting 103

240

240 220 200 180 160 20

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

104

Al–Cu alloy. These observations are in good agreement with previously reported data [12, 26, 27]. The dendrite arm spacing is more responsive to the increasing casting speed and water low rate than the grain size, which can be expected, taking into account the basics of solidiication theory discussed in Section 2.1. It is important to mention here that the structure in general and the grain size in particular are inluenced by the composition, as we already discussed in Section 2.4. Low-alloyed alloys frequently exhibit coarser grain structure with the tendency of columnar grain growth at the periphery of the billet, as shown in Figures 3.18 and 3.20 [28]. The general pattern of grain size and dendrite arm spacing distribution along the billet diameter corresponds well to the change of cooling condition as has been described earlier in this chapter; see, e.g., Figures 3.14 and 3.15. An interesting conclusion can be drawn comparing structure data in Figure 3.19 with the results of computer simulations of the same billets shown in Figures 3.2c and 3.16. The zone of iner structures (both grain size and dendrite arm spacing, Figure 3.19) correlates well with the region where the liquidus line shows delection (Figure 3.16) and the slurry region starts to widen toward the center of a billet (Figure 3.2c). Movement of solid and liquid phases is also most intense in this region (Figure 3.16). As a result, in addition to the faster cooling because of the water impingement onto the billet surface, the conditions for solid phase fragmentation (multiplication of solidiication sites) and rapid growth are created, and smaller grains with iner internal constitution are formed. General coarsening of the structure in the central portion of the billet can be associated with a longer total solidiication time and correspondingly slower total cooling rate in this region (Figures 3.14c and 3.15)

Grain size, µm

1800 1400 1000 600 200

5.0

0

20

4.0

0 16 Bill e

20

t cr 1 oss -s

3.0

80 ion ,m

2.0

ect

m

40

1.0

,%

Cu

20 0.0

FIGURE 3.20 Effect of copper concentration on distribution of grain sizes in the cross-section of 200-mm billets from Al–Cu alloys cast at 160 mm/min, water low rate 150 l/min, and melt temperature 715°C [28]. (Reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

300

1200

250

1000

Grain size, µm

Grain size, µm

Solidification Patterns and Structure Formation during Direct Chill Casting 105

200 150 100 50 0 10

800 600 400 200 0 10

30 50 70 90 Distance from billet surface, mm

40 35 30 25 20 15 10

30 50 70 90 Distance from billet surface, mm (c)

Dendrite arm spacing, µm

Dendrite arm spacing, µm

(a)

30 50 70 90 Distance from billet surface, mm (b)

40 35 30 25 20 15 10

30 50 70 90 Distance from billet surface, mm (d)

FIGURE 3.21 Variation of grain size and dendrite arm spacing along the diameter of 200-mm billets from 2024 (a,c) and 7075 (b,d) alloys: ■ = not grain reined, casting speed 80 mm/min; ○ = grain reined, 80 mm/min; △ = grain reined, 120 mm/min; and □ = additionally grain reined, 120 mm/min [29]. The billet surface is on the left and the billet center is on the right of the igures.

and with a wider region of relatively stagnant low where grains have ample opportunity to grow and coarsen (see Figure 3.6). There is experimental evidence, however, that the dendrite arm spacing can reine in the center of the billet because of the reasons already discussed in Section 3.2. Figure 3.21 shows the variation in grain size and dendrite arm spacing in 200-mm billets of two commercial alloys cast with and without grain reining [29]. In all cases, the grain structure coarsens from the surface to the center of the billet. The dendrite arm spacing, in contrast, exhibits some reinement in the central portion of the billet, mostly in grain-reined alloys. The structure of billets from a 7075 alloy is shown in Figure 3.22. Similar results have been reported elsewhere. Nagaumi [30] reported a slight reinement in the central portion of an Al–Mg slab that was correlated to the local increase of the solidiication and cooling rates. Chu and Jacoby [24] described the general increase of the average dendrite cell size from the periphery to the center of a 7XXX-series rolling ingot. However, the dendrite arm spacing in their measurements would actually decrease from 55 µm at the midthickness to 40 µm in the center (instead of increasing to 80 µm) if one measured only ine-cell grains in the duplex structure of the central portion of the ingot. We may conclude that the casting speed and the water low rate do not change the overall distribution of grain size and dendrite arm spacing across the billet section with the general tendency of structure coarsening toward

Copyright 2008 by Taylor and Francis Group, LLC

106

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

FIGURE 3.22 Microstructure of a 7075 billet: (a) not grain reined, close to the surface; (b) not grain rei ned, in the center; and (c) grain reined, in the center.

the center of the billet and some structure reinement with the increasing casting speed. There is a region of local structure reinement at some distance from the surface that is formed because this section of the billet is solidiied under conditions of accelerated cooling by water impingement onto the billet surface and intense low motion at the border of the mushy zone. There is also some dendrite-cell reinement in the central portion of the billet, which is more pronounced for grain-reined structure and when duplex grain structures are formed. The possible reason for this local dendrite-arm reining is in the local decrease of the solidiication time in the central part of the billet, as we discussed in Section 3.2. Let us now look at how the melt temperature may affect the grain structure of the billet. Figure 3.23 shows the distribution of grain size and dendrite arm spacing in cross-sections of billets cast at different melt temperatures and at two casting speeds. The grain size (Figure 3.23a,b) coarsens with increasing melt temperature, decreasing casting speed, and toward the center of the billet. The main cause of these phenomena is the decreasing cooling rate that affects the nucleation and growth of grains. In addition, the increased melt temperature deactivates available solidiication sites, decreasing the frequency of grain

Copyright 2008 by Taylor and Francis Group, LLC

450

450

Grain size, µm

Grain size, µm

500

400

350 300

400

350 300

250

250

200

200

0 40 m 75 m 40 80 tion, Me 7 c lt te 0 e 0 3 s mp 7 12 ss(b) 20 era 0 ro tur 7 16 llet c 10 e, 7 0 i 0 0 B °C 70 2

(a)

0 0 75

Me

40 m m n, tio

0

74

lt te

mp

0

era

73

20

7

tur

e,

°C

80 ec 0 12 ss-s o 0 r 0 16 llet c 71 0 Bi 0 20 70

28

DAS, µm

DAS, µm

28

26

26

24 24

22

0

0 40 m 75 m 40 80 tion, Me 7 lt te 0 0 -sec 2 (c) mp 73 1 ss 20 (d) era 0 ro tur 7 16 llet c 10 e, 0 i 0 0 B °C 7 70 2

0 0 40 m 5 7 m 40 80 tion, Me lt te 7 0 0 -sec 2 mp 73 1 ss 20 era 0 ro tur 7 16 llet c 10 e, 7 0 i 0 0 B °C 70 2

22

Copyright 2008 by Taylor and Francis Group, LLC

107

FIGURE 3.23 Effect of melt temperature (in the furnace) and casting speed on distribution of grain size (a,b) and dendrite arm spacing (c,d) in 200-mm billets of an Al–2.8% Cu alloy cast at a speed of 100 mm/min (a,c) and 200 mm/min (b,d) [20]. DAS-dendrite arm spacing.

Solidification Patterns and Structure Formation during Direct Chill Casting

500

108

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

nucleation. The appearance of a ine-structured zone at some distance from the billet surface relects the region formed under secondary (direct chill) cooling. A subsurface zone of coarser dendrites corresponds to the air gap. The general pattern of grain size distribution, therefore does not depend on the melt temperature. The melt temperature mostly inluences the dendrite internal structure in the center of the billet, causing its coarsening (Figure 3.23c,d). The increase in casting speed results in the general reinement of the dendrite internal structure across the billet. In the range of lower melt temperatures and at low casting speeds, one can see that dendrite cells reine in the central part of the billet (Figure 3.23c). This structure reinement is not observed at a higher casting speed (Figure 3.23d) and at higher melt temperatures (Figure 3.23c,d). The acceleration of the solidiication front in the center of the billet and a shallower sump with faster cooling of its bottom part through the mechanisms discussed earlier (Section 3.2) may be responsible for this phenomenon. The fact that this effect is not observed at a higher casting speed (Figure 3.23d) may be because the faster solidiication rate is completely overrun by the increasing casting speed and corresponding change in the shape of the solidiication front, the latter becoming less shallow with less lattening in the center of the billet. The peripheral regions of the billet are affected mainly by the increased casting speed, with iner structures (and higher cooling rates) occurring in the regions of primary cooling and secondary cooling. We already discussed that the structure found in the central part of a solidiied billet has characteristics relecting its solidiication history (see Figures 3.15–3.17). The dendrite arm spacing relects not only the total solidiication time required for a particular grain to form but also the particular cooling schedule. This is witnessed by the presence of so-called “coarse,” “coarse-cell” grains, or “daisies” in the central part of the billet (Figures 3.17 and 3.18). Therefore, attributing the cooling rates experimentally measured by a thermocouple moving with a billet to the structure found in the position of this thermocouple in the solid billet should be done cautiously. The solidiication rate also cannot be used directly to explain structure formation, otherwise the structure in the center of the billet would always be iner than, for example, at the midradius. Obviously, low patterns and transport of the solid phase within the slurry region along with the shape of the cooling curve should be taken into account and may explain the appearance of “coarse” grains. The origin of these grains is disputed. The literature discusses several hypotheses [24, 27, 31] that can be generally summarized as follows: (1) formation of these grains on the open surface of the melt, on a distribution bag, or on a hot top with subsequent separation and transporting by melt low or due to gravity and (2) fragmentation or separation of dendrites at the solidiication front as a result of local remelting or mechanical forces with detachment of fragments or dendrites and transport of them by melt low toward the center of the billet. Both mechanisms are possible. Note, however, coarse

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

109

grains that appear as a result of the irst mechanism are usually not only coarser internally but also much larger than the surrounding grains [31, 32]. The occurrence of these grains can be avoided by using proper technology. The second mechanism is more generic for solidiication processing and requires more attention. A low pattern existing in round billets during DC casting involves quite strong currents close to the liquidus isotherm in the direction from the periphery to the center of a billet (see Figures 3.6 and 3.16). The maximum of these currents is at approximately two thirds of billet radius, measuring from the center. These current transform further develop into a vortex in the central part of the billet, with part of the melt going upward and part going downward. As a result, dendrite fragments or even entire dendritic grains growing at the border between slurry and mushy regions in the billet section with strong melt currents can be detached easily. The further path of these solid-phase particles can be very different. Some of them will go along the low direction upward and vanish in the liquid part of the sump. Some will “roll” down along the coherency isotherm and end up in the center of the sump without much coarsening. Other grains will grow in the melt volume ahead of the solidiication front under very small undercooling and then drop to the central part of the billet, having large dendrite arm spacings. Some of these grains are captured earlier by the solidiication front, forced down by currents or sedimenting under gravity, and inish solidiication at a higher undercooling or cooling (solidiication) rate (that results in a iner dendrite arm spacing frequently observed at the grain periphery; see Figure 3.17a). Yet another grain trajectory that can result in grain-cell coarsening can be illustrated by streamlines shown in Figure 3.16. The corresponding cooling rates in Figure 3.15b demonstrate that at least some grains found in the central part of the billet could be nucleated at the liquidus (not separated from the mushy zone!) at the billet periphery and then transported to the center, growing as they travel. Therefore, the third mechanism of loating grain formation is a longer growth due to the grain trajectory within the slurry zone. All these variations of grains should be more correctly called “loating” grains. In addition, there is always an area of relatively slow melt low in the central part of the sump where loating grains may grow before settling down. As a result, the structure of the central part of the billet shows a scatter in cooling rates and, hence, in dendrite arm spacing. These grains are poor in solute concentration and are frequently cited as a cause of negative centerline segregation [12, 24, 31, 33]. This concept will be considered in more detail in Chapter 4. The process parameters affect the distribution of loating grains in the following manner. Figure 3.24 shows how the distribution of “coarse-cell” grains depends on casting speed, water low rate, and melt temperature. The increase in the casting speed obviously produces more loating grains though their distribution in the billet cross-section remains the same (see Figure 3.24a). This agrees well with the distribution of cooling rates estimated from streamlines, as shown in Figure 3.15b. The increase in the casting speed

Copyright 2008 by Taylor and Francis Group, LLC

20 Ca s

18 tin gs 16 pe ed ,c

14 mi n

m/

0 16 0 m 12 , m n 80 ctio e 40 ss-s 12 0 cro t le Bil

0.5 0.3 0.1

0 76 0 Me 75 40 lt t 7 0 em 73 20 pe (c) rat 7 710 ure 0 70 , °C

0.5 0.4 0.3 0.2 0.1 250

0 16 30 Wa 2 0 20 , mm 1 1 2 ter n flo 80 ectio 90 w 1 0 rat s-s 70 4 s 1 (b) e, cro 50 0 l/m let in 1 Bil

Coarse grains fraction

Coarse grains fraction Coarse grains fraction

(a)

0.5 0.4 0.3 0.2 0.1

Coarse grains fraction

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

110

0.5 0.3

0.1 0 76 0 40 m Me 75 0 m 0 , lt t 74 8 ion em 30 t 0 pe 7 720 12 s-sec r 0 a 0 s t (d) ure 71 16 t cro 0 , °C 70 le Bil

40 m 80 ion, m t 0 12 s-sec 0 s 16 t cro le Bil

FIGURE 3.24 Effect of process parameters on the distribution of loating grains along the diameter of a 200-mm billet: (a) Al–4.3% Cu alloy, water low 150 l/min, melt temperature 715°C; (b) Al–4.3% Cu, casting speed 200 mm/min, melt temperature 715°C; (c) Al–2.8% Cu, water low rate 150 l/min, casting speed 100 mm/min; and (d) Al–2.8% Cu, water low rate 150 l/min; casting speed 200 mm/min [5, 20]. (Parts (c) and (d) reproduced with kind permission of Springer Science and Business Media.)

results in a deeper sump with more severe lows at the boundary between the slurry and the mushy zone and a larger zone of relatively stagnant low in the central part of the billet [5]. As a result, the amount of loating grains that is generated and grows increases with the casting speed. Water low rate has virtually no effect either on the amount or the distribution of “coarsecell” grains (see Figure 3.24b). Melt temperature inluences the distribution and amount of loating grains depending on the casting speed. At a low casting speed, grains with coarse dendrite arms occupy most of the billet cross-section at low melt temperatures. With the increase in melt temperature their amount decreases and they are conined to the area around the billet centerline (Figure 3.24c). In this casting regime the area of the stagnant low in the central part of the sump (growth area) shrinks with increasing melt temperature, and a larger fraction of loating grains may go upward and remelt, as suggested by the low patterns in Figure 3.25. At high casting speeds, the amount of “coarse” grains increases with the increasing melt temperature but their distribution remains typical, i.e., they appear only in the central part of the billet (see Figure 3.24d). The reasons for this pattern are the same as in the case of casting speed: deepening of the sump, more severe lows along the coherency isotherm, and a larger stagnant region (see Figure 3.9 [9]).

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting 0.662

0.662

z

z

111

0.526

0.526 0.0

0.025

0.050 r (a)

0.075

0.100

0.0

0.025

0.050 r (b)

0.075

0.100

FIGURE 3.25 Melt low pattern in the liquid and slurry zones of a 200-mm billet (Al–2.8% Cu) cast at a casting speed 100 mm/min, at a melt temperature of 700°C (a) and 760°C (b) [20]. Only half of the billet is shown, the centerline on the left and the surface on the right. (Reproduced with kind permission of Springer Science and Business Media.)

3.4

Effect of Process Parameters on the Amount of Nonequilibrium Eutectics

The appearance of eutectics in most commercial wrought alloys is the consequence of nonequilibrium solidiication, as we discussed in Section 2.2 of Chapter 2. Under equilibrium conditions most wrought alloys should solidify as single-phase aluminum solid solution. That is why this eutectic is called nonequilibrium and does not usually attract much attention because during proper homogenization of billet and ingots it is dissolved anyway. The eutectic, by deinition, represents the last liquid that solidiies and, therefore, its occurrence in the as-cast structure is an indication of the solidiication conditions that have led to the appearance of this last liquid and of the availability of the liquid phase at late stages of solidiication. The presence of liquid at high volume fractions of solid in combination with the ability of this liquid to penetrate through the solid network is very important for the occurrence of such casting defects as pores and hot tears, as will be discussed in Chapter 5. The microstructure of different sections of a 200-mm billet from a binary Al–Cu alloy is shown in Figure 3.26. This alloy should be single-phase under equilibrium solidiication conditions but the structure unambiguously

Copyright 2008 by Taylor and Francis Group, LLC

112

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

FIGURE 3.26 Microstructures of a 200-mm billet from an Al–4.3% Cu alloy cast at 120 mm/min, casting temperature 715°C, water low rate 150 l/min taken at 10 mm (a), 55 mm (b) and 80 mm (c) from the billet surface. Arrows point to the areas of coupled eutectics.

exhibits the veins of excess phase at dendrite boundaries. The typical pattern is a mixture of eutectic areas, where one can distinguish the coupled eutectic structure, and individual phase particles that represent so-called divorced eutectic. In the latter case, the solid solution that should constitute the part of the eutectic mixture has joined the (Al) primary solid phase and only the excess phase from the eutectic remains visible. The quantiication of such a structure presents some dificulties. There are three ways of dealing with this problem. First, it is possible to accurately measure only the amount of the secondary phase and then, knowing the composition of the eutectics, to recalculate the measured fraction of the second phase to the total amount of the eutectics. Second, one can separately measure the fraction of the divorced eutectic particles and the fraction of eutectic areas, and then make a recalculation for the former only. And third, the general fraction of the mixture of divorced and coupled eutectics can be assessed. The last approach is used in this work. Taking into account that the variation of cooling rates in the crosssection of the billet is noticeable but not dramatic (see Figure 3.15), the effect of the cooling rate on the ratio between the divorced and coupled eutectic can be neglected, and the mixed fraction should be representative of the total amount of nonequilibrium eutectics.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

113

An important and quite persistent feature of nonequilibrium-eutectic distribution in the billet is the minimum in the central portion of the billet crosssection. This means that the amount of last available liquid is always lower in the center of the billet than at its periphery. The effect of the cooling rate on the eutectic fraction was discussed in detail in Section 2.2 of Chapter 2. There are controversial reports in the literature about this dependence, as illustrated in Table 2.1. The general consensus is that the increase in the cooling rate would increase the amount of eutectics, which seemingly explains the increase of the eutectic fraction toward the billet surface. On the other hand, in the range of low cooling rates the dependence can be opposite, as was shown in Section 2.2 of Chapter 2. In a commercial extrusion billet of 200 mm in diameter the cooling rates in the center and periphery of the billet are approximately 3–5 K/s and 15–20 K/s, respectively (see Figure 3.15). It is unlikely that this difference in cooling rates between the center and the periphery of the billet can cause the experimentally observed change in the eutectic amount. Several other phenomena can inluence the volume fraction of eutectics. We can suggest the following line of logic for the case of DC casting. First, in this cooling range, the eutectic fraction does not show signiicant response to the cooling rate (see Figure 2.10). Second, the solidiication during DC casting occurs under convection conditions that are rather active in the transition region of the billet (see Figure 3.6). According to Diepers et al. [34], convection can signiicantly increase the coarsening exponent, affecting the coarsening kinetics, as shown in Figure 2.12. The width of the transition region is maximum in the center of the billet, and the area of stagnant low where grains can ripen is always present there. As a result, the coarsening and back diffusion may occur in the center of the billet much more eficiently than at the periphery, decreasing the amount of eutectics. On the other hand, structure coarsening may facilitate liquid penetration into the mushy zone by increasing the permeability of the solid network [35]. In this case, depending on the direction of this low, the amount of eutectic formed by the solute-rich liquid increases or decreases in certain sections of the billet. And inally, the negative macrosegregation in the center of the billet can further contribute to the decreased amount of eutectics. As we will show in Chapter 4, the actual macrosegregation in an Al–4.25% Cu billet can be responsible for as much as 0.3 vol.% of variation in eutectic volume fraction along the billet diameter. There are also some kinetic effects of negative macrosegregation that may cause locally accelerated solidiication and, as a result, less eutectics [36–38]. The contribution of these effects is, however, not likely to be appreciable. At the present level of knowledge it is dificult to rank the mechanisms that may be responsible for the formation of nonequilibrium eutectics and single out the most signiicant one. Obviously, the scale and shape of the casting (lat ingot or round billet) and structure formed during solidiication can fundamentally affect the amount and distribution of nonequilibrium eutectic. Here we consider the effect of process parameters on the formation of eutectic in an average-size extrusion billet, as shown in Figure 3.27.

Copyright 2008 by Taylor and Francis Group, LLC

Eutectics, %

4.4

4.0 150 W 170 at er 190 flo w 210 ra te 230 , l/ (b) 50 m in 2

0 40 m 80 tion, m c 120 oss-se 160 illet cr B

Bi

5.0 4.0 3.0 2.0 1.0 0 20

60 tc 1 20 ro ss 1 80 -s ec 40 tio n, (e) 0 1.0 m m

3.0 2.8 2.6 2.4

Bi 140 lle t c 100 ro ss -s 60 (c) ec tio 20 00 n, 7 m m

Eutectics, %

Eutectics, %

lle

760 740 , °C 0 2 7 ture era p Tem

3.0 2.8 2.6 2.4

Bi

lle 140 tc ro 100 ss 0 -s (d) ec 6 tio 0 n, 2 700 m m

2.0

3.0 % Cu,

4.0

5.0

760 740 C ,° 720 rature e p Tem

FIGURE 3.27 Effect of process parameters on the amount and distribution of nonequilibrium eutectic in binary Al–Cu alloys: (a) Al–4.3% Cu, water low rate 150 l/min, melt temperature 715°C; (b) Al–4.3% Cu, casting speed 200 mm/min, melt temperature 715°C; (c) Al–2.8% Cu, water low rate 150 l/min, casting speed 100 mm/min; (d) Al–2.8% Cu, water low rate 150 l/min, casting speed 200 mm/min; and (e) water low rate 150 l/min, casting speed 200 mm/min; melt temperature 715°C [5, 20, 28]. (Parts (d) and (e) reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

20

0 40 m m , 0 8 tion 120 s-sec 160 et cros Bill

4.8

Eutectics, %

Eutectics, %

4.4

4.0 2 Ca 1 sti 14 ng sp 16 ee d, 8 cm 1 (a) /m in

114

4.8

Solidification Patterns and Structure Formation during Direct Chill Casting

115

At usual and high casting temperatures, the casting speed and water low rate slightly affect the amount of eutectic in the central portion of the billet, as shown in Figure 3.27a–d. The eutectic fraction increases with the casting speed, with the water low rate having an opposite effect. Interestingly enough, at very low casting temperatures, the amount of nonequilibrium eutectic decreases with increasing the casting speed, as illustrated in Figure 3.27c,d. The interrelation between cooling rate, solidiication rate, sump dimensions, and the ineness of structure can be responsible for such behavior. The casting temperature has a strong effect on the volume fraction and distribution of nonequilibrium eutectic, especially at high casting speeds, as illustrated in Figure 3.27d. More eutectic is concentrated in the center of the billet upon casting from high melt temperatures. In this case, the coarseness of the structure in combination with more active melt lows (see Figures 3.9 and 3.22) may facilitate a deeper penetration of the solute-rich melt into the mushy zone in the center of the billet, effectively increasing the amount of eutectic [9, 20]. More eutectic is formed in more alloyed billets. This obvious fact is demonstrated in Figure 3.27e. It is interesting, though, that the degree of eutectic decline in the center of the billet seems to be more pronounced in less alloyed billets. The experimental and computer-simulation results shown in this chapter constitute a background for the discussion in Chapters 4 and 5. The solidiication and structure formation during DC casting are intricate phenomena that are not yet fully understood and described. It is, however, obvious that the process parameters inluence the structure formation through their effect on the geometry and dimensions of the billet sump and on the low patterns that exist in the liquid and slurry part of the sump. A complex interplay between cooling rate, solidiication rate, and the dimensions of the sump sometimes results in seemingly controversial effects of process conditions on structure parameters. All in all, the material and discussion given in this chapter do not present the inal solution to the problems but rather provide broad information and offer some possible mechanisms involved in the structure formation during DC casting. This chapter would not be complete without a discussion of the effect of process parameters on the occurrence of some casting defects that can either result in the rejection of the casting or in the increased amount of removed (scalped) material.

3.5

Effect of Process Parameters and Alloy Composition on the Occurrence of Some Casting Defects

Many things may go wrong during DC casting if the casting recipe does not address the requirements of the alloy and particular mold design. Modern molds are designed in such a way that they help produce better surface

Copyright 2008 by Taylor and Francis Group, LLC

116

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

quality and more uniform temperature distribution across the casting, but at the same time these molds have a narrower allowed variation range of casting parameters and demand iner tuning of casting recipes. One of the problems that became a hot topic of discussion in the 1980s and 1990s was the formation of a so-called “ir-tree” zone in ingots and billets of diluted aluminum alloys. For example, the macrostructure of billets and ingots of 1XXX- and 5XXX-series alloys is inhomogeneous and demonstrates regions with different etching ability, e.g., in hot NaOH (see Figure 3.28a). After high-temperature annealing, the regions with different etching ability are no longer detected. Westengen [39] attributes the formation of this zone to the phase selection of iron-containing phases formed at different cooling rates in different zones of the ingot. This point of view was supported by a number of researchers; see [40–42]. It should be noted that the segregation of iron in the subsurface region is virtually unavoidable under real casting conditions (for further discussion on macrosegregation, see Chapter 4). The main reason for the formation of the ir-tree zone is the competition in nucleation and growth between different metastable and stable modiication of Fe-containing phases [41] under variable cooling conditions in the subsurface layer of DC casting [43] (see Figure 3.14a,b). It is important to note that the ir-tree structure is caused not by the formation of metastable phases proper but rather by formation of a speciic mixture of these phases due to uneven cooling conditions and nucleation-and-growth pattern. High cooling rate at the surface of the casting promotes the formation of metastable phases such as body-centered tetragonal or monoclinic AlmFe (Al9Fe2) and tetragonal α″(AlFeSi) that require about 20 K/s for their formation [44] (see also Table 2.2). On decreasing the cooling rate, in the area of air-gap formation and generally deeper under the surface, the cooling conditions (2–5 K/s) are more favorable for the formation of a mixture of monoclinic AlxFe (Al5Fe), orthorhombic Al6Fe, and cubic α(AlFeSi) phases [44]. Phases like Al6Fe nucleate at a lower cooling rate in the environment where AlmFe phase dominates but, having a higher growth rate, overgrow it, forming a speciic pattern, as shown in Figure 3.28b [41]. Different phases promote different etching ability of the alloy, which ultimately results in the appearance of the ir-tree zone. Brusethaug et al. [41] note that the ir-tree structure can be observed not only in subsurface layers but also in the central part of billets and small ingots, which is in line with the enhanced cooling and increased solidiication rate in this region, as we discussed in Section 3.2. The phase composition of aluminum ingots, containing 0.25% Fe and 0.13% Si, in the as-cast condition and after homogenization is given in Table 3.2 after Westengen [39] and Skjerpe [40]. The equilibrium phase composition of this alloy must be (Al) + A13Fe + α(Al8Fe2Si) [44]. The main change that occurs during high-temperature annealing is the transformation of binary and ternary metastable phases at the surface to more stable phases. Such an annealing may help reduce the appearance of the ir-tree zone but is not always eficient (note the persistence of the metastable α″(AlFeSi) phase).

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

117

(a)

Al3Fe, α′′

Al6Fe, Al3Fe, α(AlFeSi)cubic

(b) FIGURE 3.28 Fir-tree structure in a horizontal cross-section of an aluminum ingot (a) and a scheme illustrating the formation of the ir-tree zone (b) [41]. (Courtesy of Dr. S. Brusethaug, Hydro Aluminium, Norway.)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

118

TABLE 3.2 Phase Composition of Direct-Chill Cast Ingots in As-Cast and Annealed Conditions (600 × 1350 mm Ingots, Casting Speed 100 mm/min) [39, 40, 44] Phase

Al8Fe2Si

As–Cast Surface – Inner – region Annealed (590°C, 5 h) Surface + Inner + region

α(AlFeSi)

α″

Al3Fe

Al6Fe

AlmFe

AlxFe

β′(AlFeSi)

Si

+ –

+++ –

– +

– +++

+++ –

– +++

– +

– +

+ +

+++ +

+++ +++

– –

+ –

– +

– –

– –

FIGURE 3.29 Feathery grains formed in the presence of forced low (from left to right) in an Al–4% Cu alloy. (Courtesy of A.N. Turchin.)

Other ways of preventing formation of the ir-tree structure include the control of alloy composition (keep the Fe:Si ratio at about 3 [41], presence of small amounts of Cr and Mn [43]), and optimization of cooling conditions at the surface (low-head casting, reducing the air-gap zone [43]). Yet another undesirable structure pattern that is observed in DC-cast billets is formed by so-called feathery grains that appear as elongated crystals with both straight and wavy boundaries and are frequently assembled in fanshaped agglomerates. The appearance of such grains is favored by high melt temperature [31], high temperature gradient [45, 46], forced melt low [47, 48], and concentration of certain alloying elements [49–51]. Figure 3.29 shows a typical image of feathery grains, inclined toward the direction of melt low. In practice, feathery grains may appear during DC casting if the cooling rate is higher than 10 K/s (or the overheated melt comes in contact with the water-cooled mold promoting high thermal gradients), the alloy is relatively

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting

119

pure (e.g., 1XXX series) or the alloy contains Cr, Zr, or Ti as growth-restricting elements or Mg, Cu, Zn, or Si as the elements that increase the solidiication range and strongly segregate. A thorough study of crystallographic orientations and morphology of feathery grains shows that these grains, unlike normal columnar grains that grow in direction, are made of dendrites with trunks split by a {111} plane [46]. The straight boundaries in the structure shown in Figure 3.29 represent the coherent {111} planes, whereas the wavy boundaries represent incoherent twin planes between neighboring grains. It is suggested that the formation of feathery grains is a result of two mechanisms [46]. First, the anisotropic properties of the alloy, i.e., surface tension and kinetics of atom attachment, change at high growth rates facilitated by high thermal gradients and certain alloy compositions. As a result, the growth direction switches from to . Second, the twinning mechanism of growth is initiated by stacking faults and is promoted by convection and high thermal gradient. In the absence of convection or high thermal gradient normal, untwinned dendrites are expected. Figure 3.30 demonstrates different dendrite morphologies that can be obtained under forced convection in an aluminum alloy solidiied in different thermal conditions. The occurrence of feathery grains in the billet or ingot produces strong anisotropy of structure and properties of deformed products, which includes 20 equiaxed feathery columnar

18

Thermal gradient, K/mm

16 14 12 10 8 6 4 2 0 0.0

0.2

0.4

0.6 0.8 1.0 1.2 1.4 Solidification rate, mm/s

1.6

1.8

2.0

FIGURE 3.30 Effect of solidiication rate and thermal gradient on the morphology of dendrites formed in an Al–4% Cu alloy under forced low conditions [48]. (Reproduced with kind permission of Elsevier/Pergamon.)

Copyright 2008 by Taylor and Francis Group, LLC

120

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

inhomogeneous recrystallization, formation of intermetallic segregates, and low ductility in certain directions [50]. A proper grain-reining practice and a lower temperature gradient in the liquid (less melt overheating and slower casting) are among common practices for elimination of feathery grains [51]. Apart from the sensitivity of the as-casting structure to the casting speed and melt temperature as a natural consequence of the liquidus and solidus temperatures of the particular alloy and the corresponding extent of the characteristic regions on the billet (see Figure 3.1), there are other casting defects that originate and are controlled by the conditions of cooling, i.e., by melt temperature, casting speed, and water rate. The irst problem that may occur at the beginning of the casting or during casting if some casting parameters are changed on the ly is the melt lashing into the gap between the starting block and the mold or the bleed-out of melt through the weak solid shell and then into opening below the mold. This type of defect is shown in Figure 3.31a. Another manifestation of the same problem is the separation of the billet butt due to insuficient degree of its solidiication. All these defects are caused by the fact that the solid shell is supposed to be strong enough to contain the semi-liquid and liquid core is ruptured either due to insuficient strength or as a result of remelting in the air-gap region. Butt cracking may also be a result of solid shell weakness. Most likely the

FIGURE 3.31 Flashing, bleed-out, hang-up, butt separation, and cracking (a) and cold folding and liquation surface defects (b) on a DC-cast aluminum-alloy billet.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting 121 casting conditions are too hot and the cooling ability of the mold is insuficient. The usual remedy is to reduce the melt temperature and to ensure that the primary cooling and secondary cooling occur homogeneously along the perimeter of the casting. Sometimes the casting can be stopped for a short period to allow the solidiication and healing of the bleed-out section and then resumed. If the problem persists, the casting should be aborted. It is important to note that the position and strength of the solid shell are functions of the material structure. For example, the grain reinement that is generally a good answer to the requirement of better properties and quality of the as-cast material can promote bleed-out. The fact of the matter is that grain-reined alloy has a wider slurry zone and a narrower mushy zone (we will consider this in more detail in Chapter 5). The consequence of this can be the slip of the fully solid shell beneath the mold opening during casting if the melt temperature and the casting speed are not changed accordingly. The opposite of the bleed-out is the hang-up of the casting in the mold, also shown in Figure 3.31a. In this case, the thick section of the billet is solidiied inside the mold and gets stuck. This defect is especially harmful in hot-top molds when the solidiication spreads to the ceramic part of the mold assembly. In most cases the casting should be stopped immediately, and frequently the mold is damaged. Another danger of this defect is that the partial hangup can lead to the partial separation or horizontal cracking of the billet with subsequent massive bleed-out. The increase in the melt temperature along with the proper lubrication of the mold and the correct choice of casting speed help remedy this type of defects. The surface of the billet frequently bears the characteristic marks illustrated in Figure 3.31b. The reduction of this type of defect decreases the amount of scalping needed before further processing of billets and ingots. Cold folding is caused by the freezing of the melt meniscus inside the mold, e.g., below the ceramic ring of the hot top (see Figure 3.14a). The solid meniscus then moves down and the melt lashes into the opening, illing the cavity [2]. The process repeats with a periodicity that is a function of casting speed. Cold folds can penetrate fairly deeply inside the billet, forming special bands of ine structures with oxide inclusions and pores, as shown in Figure 3.32. The cold folding is caused by cold casting conditions and too high heat transfer through the mold. The increased melt temperature and casting speed and the decreased water low rate may help reduce this type of surface defect. Electromagnetic casting (EMC) and air-assisted (e.g., air slip) casting are much less prone to cold folding. The formation of the air gap (see Figure 3.14) and corresponding reheating of the solid shell up to the point of its remelting is the main reason for the appearance of liquation marks on the billet surface, as shown in Figure 3.31b. When the solid shell due to the thermal contraction pulls away from the mold surface, the solid shell starts to heat up from the latent solidiication heat and the heat coming from the melt. The appearance of the liquation relects the temperature of the thermal contraction onset [4, 12] (see Chapter 5 for details). The metallostatic pressure of the liquid metal head in the mold and

Copyright 2008 by Taylor and Francis Group, LLC

122

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

FIGURE 3.32 Cold folding in the cross-section of a 200-mm aluminum-alloy billet.

hot top together with the low patterns oriented toward the billet surface (see, e.g., Figure 3.25) and the shrinkage-induced low in the mushy zone drive the liquid toward the surface. Remelting of the shell opens wide channels for such a low. As a result, liquid can penetrate through the shell and reach the surface. In addition, this liquid is enriched in the solute elements up to the point of the eutectic concentration, which considerably lowers its solidus temperature and delays its solidiication. This liquid brings the solute to the surface and creates a region of highly positive surface segregation. The periodicity of shell pulling-off, remelting, liquation, and resolidiication results in the periodic subsurface structure with the bands of enriched (liquation marks) and depleted (between liquation marks) zones. Alloys with a larger solidiication range are more susceptible to this defect. Ohm and Engler [52] report the role of metallostatic pressure as a main reason for the liquation. Dobatkin [4] suggests the important role of the oxide surface layer. A stronger oxide skin, a better surface quality, and suficient lubrication of the mold can help reduce the liquation marks. The general remedy for the liquation is the proper thermal balance in the mold, e.g., cooler casting conditions and the reduction of air gap.

References 1. L. Arnberg, L. Bäckerud. Solidiication Characteristics of Aluminum Alloys. Vol. 3: Dendrite Coherency, Des Plaines, IL: AFS, 1996. 2. J.F. Grandield, P.T. McGlade. Materials Forum, 1996, vol. 20, pp. 29–51.

Copyright 2008 by Taylor and Francis Group, LLC

Solidification Patterns and Structure Formation during Direct Chill Casting 123 3. W. Roth. Aluminium, 1943, vol. 25, pp. 283–291. 4. V.I. Dobatkin. Continuous Casting and Casting Properties of Alloys, Moscow: Oborongiz, 1948. 5. D.G. Eskin, J. Zuidema, V.I. Savran, L. Katgerman. Mater. Sci. Eng. A, 2004, vol. 384, pp. 232–244. 6. J.F. Grandield, K. Goodall. P. Misic, Z. Zhang. In: R. Haglen (Ed.), Light Metals 1997, Warrendale: TMS, 1997, pp. 1081–1090. 7. J.F. Grandield, L. Wang. In: A.T. Tabereaux (Ed.), Light Metals 2004, Warrendale: TMS, 2004, pp. 685–690. 8. B. Commet, P. Delaire, J. Rabenberg, J. Storm. In: P.N. Crepeau (Ed.), Light Metals 2003, Warrendale, PA: TMS, 2003, pp. 711–717. 9. Q. Du, D.G. Eskin, L. Katgerman. Mater. Sci. Eng. A, 2005, vol. 413–414, pp. 144–150. 10. W. Schneider, E.K. Jensen. In: C.M. Bickert (Ed.), Light Metals 1990, Warrendale: TMS, 1990, pp. 931–936. 11. Suyitno, W.H. Kool, L. Katgerman. Metall. Mater. Trans. A, 2004, vol. 35A, pp. 2917–2926. 12. V.A. Livanov, R.M. Gabidullin, V.S. Shepilov. Continuous Casting of Aluminum Alloys, Moscow: Metallurgiya, 1977. 13. W. Schneider. Adv. Eng. Mater., 2001, vol. 3, pp. 635–646. 14. M.A. Wells, D. Li, S.L. Cockroft. Metall. Mater. Trans. B, 2001, vol. 32B, pp. 929–939. 15. D.C. Prasso, J.W. Evans, I.J. Wilson. Metall. Mater. Trans. B, 1995, vol. 26B, pp. 1243–1251. 16. E.D. Tarapore. In: P.G. Campbell (Ed.), Light Metals 1989, Warrendale, PA: TMS, 1989, pp. 875–880. 17. J.M. Reese. Metall. Mater. Trans. B, 1997, vol. 28B, pp. 491–499. 18. A.D. Shestakov. Izv. Ross. Akad. Nauk. Metally, 1996, no. 6, pp. 130–138. 19. V.I. Dobatkin, A.D. Shestakov. Izv. Ross. Akad Nauk. Metally, 1992, no. 4, pp. 55–60. 20. D.G. Eskin, V.I. Savran, L. Katgerman. Metall. Mater. Trans. A, 2005, vol. 36A, pp. 1965–1976. 21. V.A. Livanov. Metallurgical basics of continuous casting. In Transactions of the 1st Technological Conference of Metallurgical Plants of Peoples’ Commissariat of Aviation Industry, Moscow: Oborongiz, 1945, pp. 5–58. 22. J. Zuidema, Jr., L. Katgerman, I.J. Opstelten, J.M. Rabenberg. In: J.L. Anjier (Ed.), Light Metals 2001, Warrendale, PA: TMS, 2001, pp. 873–878. 23. J.-M. Drezet. Direct-Chill and Electromagnetic Casting of Aluminium Alloys: Thermomechanical Effects and Solidiication Aspects, Dr. Sci. Techn. Thesis, Lausanne, Switzerland: EPFL, 1996. 24. M.G. Chu, J.E. Jacoby. In: C.M. Bickert (Ed.), Light Metals 1990, Warrendale, PA: TMS, 1990, pp. 925–830. 25. Q. Du, D.G. Eskin, A. Jacot, L. Katgerman. Acta Mater., 2007, vol. 55, pp. 1523–1532. 26. A. Håkonsen, D. Mortensen, S. Benum, H.E. Vatne. In: C.E. Eckert (Ed.), Light Metals 1999, Warrendale, PA: TMS, 1999, pp. 821–827. 27. E.F. Emley. Int. Mater. Rev., 1976, no. 6, rev. no. 206, pp. 75–115. 28. Suyitno, D.G. Eskin, V.I. Savran, L. Katgerman. Metall. Mater. Trans. A, 2004, vol. 35A, pp. 3551–3561. 29. R. Nadella, D.G. Eskin, L. Katgerman. Metall. Mater. Trans. A, 2008, vol. 39A (in press).

Copyright 2008 by Taylor and Francis Group, LLC

124

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

30. H. Nagaumi. Sci. Techn. Adv. Mater., 2001, vol. 2, pp. 49–57. 31. V.I. Dobatkin. Ingots of Aluminum Alloys, Sverdlovsk: Metallurgizdat, 1960. 32. V.I. Napalkov, G.V. Cherepok, S.V. Makhov, Yu. M. Chernovol. Continuous Casting of Aluminum Alloys, Moscow: Intermet Engineering, 2005. 33. H. Yu, D.A. Granger. In: E.A. Starke, T.H. Sanders (Eds.), Aluminum Alloys: Their Physical and Mechanical Properties, Proc. ICAA 1, vol. 1, Warley (U.K.): EMAS, 1986, pp. 17–29. 34. H.J. Diepers, C. Beckermann, I. Steinbach, Acta Mater., 1999, vol. 47, pp. 3663–3678. 35. S. Chakraborty, P. Dutta. Mater. Sci. Technol., 2001, vol. 17, pp. 1531–1538. 36. S.A. Cefalu, M.J.M. Krane. Mater. Sci. Eng. A, 2003, vol. 359, pp. 91–99. 37. I. Vušanović, B. Šarler, M.J.M. Krane. Mater. Sci. Eng. A, 2005, vol. 413–414, pp. 217–222. 38. I. Vušanović, M.J.M. Krane. Personal communication, 2006. 39. H. Westengen. Z. Metallkde., 1982, vol. 73, pp. 360–368. 40. P. Skjerpe. Metall. Trans. A, 1987, vol. 18A, pp. 189–200. 41. S. Brusethaug, D. Porter, O. Vorren. In 8th Intern. Light Metals Congress, Leoben–Vienna 1987, Düsseldorf: Aluminium Verlag, 1987, pp. 472–476. 42. C.A. Aliravci, J.E. Gruzleski, M.Ö. Pekgüleryüz. In: B. Welch (Ed.), Light Metals 1998, Warrendale, PA: TMS, 1998, pp. 1381–1389. 43. D.A. Granger. In: B. Welch (Ed.), Light Metals 1998, Warrendale, PA: TMS, 1998, pp. 941–952. 44. N.A. Belov, A.A. Aksenov, D.G. Eskin. Iron in Aluminum Alloys: Impurity and Alloying Element, London: Taylor & Francis, 2002. 45. H. Anada, S. Tada, K. Koshimoto, S. Hori. J. Jpn. Inst. Light Met., 1991, vol. 41, pp. 497–503. 46. S. Henry, T. Minghetti, M. Rappaz. Acta Mater., 1998, vol. 46, pp. 6431–6443. 47. S. Henry, G.-U. Gruen, M. Rappaz. Metall. Mater. Trans. A, 2004, vol. 35A, pp. 2495–2501. 48. A.N. Turchin, M. Zuijderwijk, J. Pool, D.G. Eskin, L. Katgerman. Acta Mater., 2007, vol. 55, pp. 3795–3801. 49. L.-O. Gullman, L. Johansson. TMS paper no. A72-43, AIME, 1972, 38 pp. 50. D.A. Granger, J. Liu. JOM, 1983, no. 6, pp. 54–59. 51. L. Bäckerud, E. Król, J. Tamminen. Solidiication Characteristics of Aluminium Alloys. Vol. 1: Wrought Alloys, Oslo: SkanAluminium, 1986. 52. L. Ohm, S. Engler. Metall, 1989, vol. 43, pp. 520–524.

Copyright 2008 by Taylor and Francis Group, LLC

4 Macrosegregation

4.1 4.1.1

Mechanisms of Macrosegregation Historic Overview

The fact that large-scale castings and ingots are not homogeneous in chemical composition has been known for centuries. The Italian metallurgist and foundryman V. Biringuccio described segregation in bronze gun barrels in his book De la Pirotechnia as early as 1540 [1]. In 1574 the Austro-Hungarian chemist L. Ercker published his observations of liquation in precious alloys [2]. As Pell-Wallpole notes in a brilliant review [3], most observations and studies of macrosegregation during the nineteenth century were about precious metals, including works by W.C. Roberts-Austin (1875) and E. Matthey (1890) in Great Britain. In 1866–1867 Russian metallurgists A.S. Lavrov and N.V. Kalakutsky observed macrosegregation in steel ingots and noted that its degree depended on the size of the ingot [4]. Lavrov noted that macrosegregation was caused by the precipitation of carbon during steel solidiication and accumulation of low-melting components in the center of the ingot. It was not until the beginning of the twentieth century, however, that macrosegregation attracted real scientiic interest, irst as related to steel and bronze ingots and later as related to aluminum billets and ingots. Pioneering works include those of T. Turner, M.T. Murray, E.A. Smith, O. Bauer, H. Arndt, R.C. Reader, R. Kühnel, and F.W. Rowe on copper alloys and those of G. Masing, W. Claus, S.M. Voronov, and W. Roth on aluminum alloys (citation information can be found in Ref. [3]). This chemical inhomogeneity is called macrosegregation because it occurs on the scale of the grain. The main reason for it is the segregation (partitioning) of alloying elements during solidiication. This partitioning is considered in greater detail in Chapter 2. Somehow the large difference (sometimes tens of percent) in the compositions of solid and liquid phases occurring during solidiication is translated into a relatively small difference (about or less than 1%) of chemical composition across the whole section of a casting or a billet. What makes macrosegregation a really serious problem is that it cannot be eliminated during downstream processing, unlike microsegregation, which can be removed relatively easily by high-temperature anneal, the technique known as homogenization. Diffusion lengths are measured in micrometers

125 Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

126

Deviation of Cu concentration from the average

for microsegregation and in centimeters for macrosegregation. This difference in scales makes real difference in the severity of the problem. The simplest way to explain macrosegregation is to imagine that the advancing solidiication front pushes the liquid enriched in the solute (in the case of partitioning, coeficient K < 1) toward the hotter part of the casting, e.g., the center. As a result, after solidiication, the center of the casting will contain more solute than the periphery of the casting. In reality the soluterich liquid should be transported toward the hot spot of the casting (in the direction of solidiication). The driving force behind such transport is either convection or shrinkage-driven low. This kind of macrosegregation, called “normal” segregation, is a direct consequence of microsegregation and can easily be predicted and estimated using the phase diagram data. The frequently observed macrosegregation pattern in ingots and billets is just the opposite—the periphery of the casting is solute enriched while the center remains solute lean. This type of “inverse” macrosegregation is quite typical of billets and ingots from nonferrous alloys, including DC-cast aluminum alloys. The ultimate form of “inverse” segregation is liquation or exudation at the casting surface, when the solute-rich liquid penetrates through the outer shell of a casting and solidiies at the surface as liquates or eutectic exudates (see Figure 3.31b). Obviously, in this case the solute-rich liquid is transported in the direction opposite to the solidiication-front movement. The inverse segregation is quite typical of DC-cast aluminum billets when it is manifested as negative centerline segregation and positive surface segregation, as illustrated in Figure 4.1. During the irst half of the twentieth century several theories were suggested to explain inverse segregation [3]. These theories attempted to explain numerous observations, some of which are summarized below. It was found 0.04

Center

0.02 0 −0.02 −0.04 −0.06 −0.08 −0.1 0

40

80 120 Cross-section of billet, mm

160

200

FIGURE 4.1 A typical pattern of macrosegregation in a 200-mm billet of an Al–4.2% Cu alloy cast at 120 mm/min. Note the negative segregation in the center, positive at the surface, and a heavily depleted subsurface layer.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

127

experimentally that inverse segregation occurred in alloys with a considerable freezing range and the extent of the segregation increased with the freezing range, e.g., as reported in the works of Claus and Goederitz in 1928 and Voronov in 1927 and 1929. The presence of hydrogen was shown to promote exudations, while melt overheating increased the degree of segregation. As early as 1926, Masing and Dahl concluded that hydrogen in aluminum alloys would adversely affect macrosegregation if it were trapped in the mushy zone. Therefore, this inluence was typical only of moderate cooling rates, when hydrogen was neither quenched in solid aluminum nor escaped the solidifying metal. Early accounts, such as J.T. Smith’s in 1875, noted that the cooling rate was a determining factor in segregation. Bauer and Arndt (1921) emphasized a steep temperature gradient in the ingot as an essential condition for macrosegregation. Voronov (1929) showed that any change in casting conditions that increased the cooling rate, i.e., reduced casting temperature, colder mold, lower pouring rate, or increased mold conductivity, would increase the degree of inverse segregation in duralumin ingots, as demonstrated in Table 4.1 [5]. The relationship of macrosegregation to alloy properties and grain structure was also examined [3]. It was shown that segregation developed during solidiication, not in the liquid state. The transition from normal to inverse segregation was experimentally observed by increasing the thickness of the solidiied shell of an ingot, as reported on Fraenkel and Gödecke in 1929. Inverse segregation was often less in iner equiaxed structures than in columnar or coarse dendritic structures because of the different mechanisms of

TABLE 4.1 Effect of Casting Parameters on Macrosegregation in a Permanent-Mold Ingot from Duralumin (Al–Cu–Mg alloy) [5]

Parameter Casting temperature, °C

Mold temperature

Thermal conductivity

Ingot thickness, mm

Values 750 730 700 200°C Ambient Water cooled Sand mold Chill with the wall thickness equal to the ingot thickness Chill three times thicker than the ingot 20 90

Copyright 2008 by Taylor and Francis Group, LLC

Difference in Cu Concentration between the Center and the Surface of an Ingot, % 0.28 0.37 0.57 1.3 3.3 4.7 0.08 0.33 0.71 0.24 0.54

128

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

feeding the solidiication contraction, i.e., liquid feeding in columnar structures and mass feeding in equiaxed structures [3]. As early as 1925, Masing et al. correlated inverse segregation to volume contraction during solidiication of metallic alloys [3]. This theory was further developed by Phelps (1926) and Verö (1936) and formed a basis for modern views on macrosegregation. Tables 4.2 and 4.3 summarize some early theories that have been offered to explain the phenomenon of inverse segregation. Some of these theories were short-lived (Table 4.2), and others were developed into the contemporary theory of macrosegregation [6–8] (Table 4.3). The current understanding of macrosegregation mechanisms can be formulated rather simply: relative movement of liquid and solid phases during solidiication. On the phenomenological level, we can single out several types of such relative movement that occurs in the sump of a billet during DC casting: • Thermo-solutal convection in the liquid sump caused by temperature and concentration gradients, and the penetration of this convective low into the slurry and mushy zones of a billet (see Figures 3.6, 3.9, and 3.25); • Transport of solid grains within the slurry zone by gravity and buoyancy forces, convective or forced lows (see Figure 3.16; the origins of such grains are discussed in Section 3.3); • Melt low in the mushy zone that feeds solidiication shrinkage and thermal contraction during solidiication; • Melt low in the mushy zone caused by metallostatic pressure; • Melt low in the mushy zone caused by deformation (thermal contraction) of the solid network; • Forced melt low caused by pouring, gas evolution, stirring, vibration, cavitation, rotation, etc., which penetrates into the slurry and mushy zones of a billet or changes the direction of convective lows. We know that commercial alloys usually solidify as dendrites, forming overall equiaxed structure in a billet (see Section 3.3). In the slurry zone (above the coherency isotherm in the transition region) the equiaxed grains are free to move and can travel short or long distances, depending on their size and direction of melt low. In the mushy zone, however, these dendrites form a continuous solid network and have a ixed position in the billet. They move only in the direction of billet withdrawal and with the casting speed. Liquid low within the mushy zone is limited to distances on an order of magnitude of several grain sizes. Before we discuss the mechanisms of macrosegregation we need to familiarize ourselves with the phenomenon of permeability, which relects the ability of liquid to penetrate through the ilter, e.g., formed by a continuous solid-phase network.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

129

TABLE 4.2 Summary of Unsubstantiated Theories of (Inverse) Macrosegregation [3] Theory

Author(s), year

Mobile equilibrium

S.W. Smith, 1917–1926

Thermosolutal segregation

Benedicks, 1925

Undercooling

Hanson, 1917; Johnson, 1918; Masing, 1922

During initial undercooling, the solid phase is enriched in the solute; next layers will be solute lean.

Crystal migration

Voronov, 1927; Watson, 1933

Contraction pressure

Kühnel, 1922; Price and Philips, 1927; Voronov, 1929

Solute-lean primary crystals are detached from the ingot’s shell and pushed by the solidiication front toward the center of the ingot under conditions of small thermal gradient. Solid shell formed at the top of a casting exerts pressure onto the liquid and forces it through the side shell to the surface.

Gas evolution

Genders, 1927

Copyright 2008 by Taylor and Francis Group, LLC

Essence

Comments

System tends to lower its melting point in the region of solidiication by segregation of lowmelting components to this region. Temperature and solute gradients in the liquid drive the segregation.

Based on an incorrect assumption that segregation occurs in the liquid state.

Thermal contraction of an ingot shell may result in its rupture (hot cracking!). The gaps between grains are illed with the solute-rich liquid. Dissolved gas concentrates in the residual, low-melting liquid and then evolves, pushing the solute-rich liquid along the grain boundaries toward the ingot surface.

Based on an incorrect assumption that segregation occurs in the liquid state. Cannot explain the segregation over the cross-section in relatively large ingots and upon slow solidiication. Contradicts the experimental observations on inverse segregation in ingots where no primary phase movement has taken place. Contradicts calculations of pressure and experimental observations of inverse segregation in ingots with the open (liquid) surface. Explains the exudations and surface segregation.

Fits many experimental observations. Requires the continuous solid network to be formed. Cannot explain inverse segregation in gas-free alloys and upon fast cooling.

130

TABLE 4.3 Summary of Theories of (Inverse) Macrosegregation That Form the Basis of the Modern School of Thought [3] Interdendritic feeding

Author(s), Year

Essence

Comments

Gulliver, 1920 Bauer and Arndt, 1921

Interdendritic feeding is assisted by capillary forces Interdendritic movement of solute-rich liquid from the interior toward the surface of an ingot. Segregation depends on the ratio between the solid– liquid interfacial diffusion rate and the growth velocity of primary phase. Normal segregation occurs in equiaxed structures and the inverse in columnar structures. Feeding of solidiication shrinkage occurs also through intradendritic channels. Volume solidiication contraction is a driving force for the interdendritic feeding. Solidiication expansion results in normal segregation, while solidiication shrinkage occurs in inverse segregation. Segregation in DC cast billets occurs in layers: the surface receives solute-rich liquid and does not part with it; the center expels liquid but, due to restricted feeding, does not receive the fresh one. Inverse segregation is driven by interdendritic melt low caused by solidiication shrinkage and is not limited to columnar structures. Emphasized the role of partitioning in the extent of segregation, rather than the solidiication range. Showed that segregation occurs in the transition region of an ingot.

Additional mechanism of liquid transport Does not explain inverse segregation in equiaxed structures

Dobatkin, 1948 [7] Volume contraction

Phelps, 1926

Gradual segregation

Brenner and Roth, 1940 [6]

Shrinkage-driven low

Verö, 1936 [8]

Copyright 2008 by Taylor and Francis Group, LLC

Very important contribution to modern theory

A basis of the modern theory of macrosegregation during DC casting Important contribution to modern theory

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Theory

Dobatkin, 1948 [7]

Shrinkage-driven low and temperature gradient

Pell-Walpole, 1949

Relation to the geometry of the billet sump

Brenner and Roth, 1940 [6]

Dobatkin, 1948 [7]

Related the occurrence of macrosegregation to the formation of a coherent solid skeleton of primary phase (mushy zone). In the slurry zone only the gravity segregation of solid (loating) grains is possible. Extended the mechanism of Brenner and Roth to equiaxed structures and linked the development of macrosegregation to the temperature gradient across the billet. Correlated the occurrence of exudations to the formation of air gap. The extent of macrosegregation in vertical DC casting is proportional to the vertical length of the transition region, and increases with the casting speed and the size of the ingot. The extent of macrosegregation in vertical DC casting is proportional to the horizontal thickness of the transition region and to the inclination of the solidiication front to the horizon.

Important contribution to modern theory

Important link to process parameters

Macrosegregation

Importance of coherency

Approach to the modern theory of macrosegregation in DC casting

131

Copyright 2008 by Taylor and Francis Group, LLC

132 4.1.2

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys Permeability

In order to understand segregation formation, porosity, and hot tearing during DC casting, we need information on the permeability of the transition region and, in particular, of the mushy zone. This parameter is very important during the later stages of the solidiication process, when the channels where liquid lows are very narrow. Here the interdendritic low caused by solidiication shrinkage might not be enough to feed the solid phase, which may result in discontinuity of the solid material in the form of microporosity and hot tearing. In macrosegregation, the permeability in the entire transition region is crucial because it determines the depth of liquid penetration into the mushy zone and the length of liquid transport within the mushy zone. Figure 4.2 shows a 3D image of the mushy zone at approximately 70 vol.% of the solid phase [9]. Note that the solid phase (darker gray portions of the image) and liquid channels (light gray areas) are continuous. The permeability of a porous structure (in the semisolid material such a porous structure is represented by the solid-phase network) can be determined

FIGURE 4.2 3D image of a semi-solid sample of an Al–13% Cu alloy at 550°C. The volume fraction of solid is 71 to 74%. Light areas correspond to the liquid phase (removed from the image for clarity) and darker gray areas correspond to the solid phase. Image is obtained by 3D tomography at the European Synchrotron Radiation Facility. (Courtesy of D. Bernard (ICMCB-CNRS).)

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

133

from the low rate of a penetrating liquid at the exit of sample Q and the pressure drop per unit length, ∆P′/L. We use Darcy’s law [10]: kD A∆P′ Q = _______ L

(4.1)

where Q is the volume low rate, A is the cross-sectional area, and kD is the permeability coeficient. The permeability coeficient depends on the liquid properties. To make the permeability dependent only on the geometry of the solid phase, a speciic permeability, K, is introduced as: K = µk D ,

(4.2)

where µ is the dynamic viscosity of the liquid. This permeability depends on the packing of dendrites, hence on the volume fraction and morphology of the solid phase. The same law can be written in terms of the average (supericial) low velocity v as follows [11]: K ___ ∆P + ρ g , v = – ___ 1 ) µf 1 ( L

(4.3)

where f l is the volume fraction of liquid, ρl is the density of liquid, and g is the gravity acceleration. The value of permeability is quite dificult to measure for a particular type of alloy and structure. Available experimental methods are limited to alloys with high volume fraction of liquid tested upon remelting [12–14]. Recent attempts were made to perform measurements upon solidiication of casting aluminum alloys that naturally contain considerable amounts of liquid phase (eutectics) even at temperatures close to the solidus [15]. In fact, up to now these experimental techniques have been used mainly to validate different analytical or numerical models. The Kozeny–Carman relationship is one of the analytical models that describe the permeability of a coherent solid network using structure parameters. This relationship couples the porosity and tortuosity of the structure to the permeability as follows [16, 17]: φ3 K = B ___2 , Sv

(4.4)

where φ is the porosity, Sv is the speciic surface area of the solid phase, and B is a geometrical factor. For solidiication, this relationship is frequently written in the following form: (1 – fs)3 , K = _______ kKC S2v fs2

(4.5)

where fs is the solid fraction and kKC is the Kozeny–Carman constant taken equal to 5 for equiaxed structures [14].

Copyright 2008 by Taylor and Francis Group, LLC

134

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

It should be noted that the Kozeny–Carman relationship can be written in different ways depending on which structure parameter is most important for the permeability. In Equation 4.5 this is a solid–liquid interfacial area concentration, but sometimes dendrite arm spacing or grain size is used as such a parameter: (1 – fs)3 d2 _______ , K = ____ 180 fs2

(4.6)

where d is the secondary dendrite arm spacing. In the case of globular equiaxed grains, the secondary dendrite arm spacing in Equation 4.6 can be substituted for Dgr, the grain size [18]. In the case of columnar structures, when the low occurs parallel to the dendrite trunks, the following relationship has been suggested [19, 20]: λ21(1 – fs)3 , K = C ________ fs

(4.7)

where λ1 is the primary dendrite arm spacing and C is a coeficient depending on the fraction solid. This expression can be substituted by a simpler one (accurate within 10% of the experimental data for aluminum alloys) [20]: λ21 . K = _________________ 5000(1 – fs)(2fs – 1)

(4.8)

The applicability of Equation 4.7 to solid fractions larger than 0.35 is arguable due to the complex shape of dendrites and, hence, interdendritic channels [19]. The real dendritic structures are, of course, quite complicated and the melt low through their network can go both along grain boundaries (extradendritic low) and through intradendritic channels (intradendritic low), as suggested by Dobatkin [7]. A rather complex model that describes extradendritic and intradendritic channels in the equiaxed mushy zone has been developed by Wang and Beckermann [21]. They use a concept of grain envelope that separates the inner space of the grain from the exterior. Hence, there is an interfacial area concentration of the grain envelope that characterizes the interface between the grain envelope and the interdendritic liquid. The rest of the liquid exists between dendrite branches inside the envelope. The permeability is then a function of the ratio of extradendritic and intradendritic permeabilities characterized by a factor β and the shape factor ψ [22]: 3fe 2 1 ___ K = __ , 2 ψβ Se

[ ]

(4.9)

where Se is the interfacial area concentration of the grain envelope and fe is the volume fraction of grain envelope. The ratio β can be written as: βd , β = ______________ β 2n 1/2n f ne + ___d β1

[

Copyright 2008 by Taylor and Francis Group, LLC

( )]

(4.10)

Macrosegregation

135

where βd and βl are the intradendritic and extradendritic permeabilities, respectively. In numerical simulations, e.g., the modeling of macrosegregation [23], it is convenient to use permeability depending only on the solid fraction. Thus the so-called Blake–Kozeny equation is adopted: (1 – fs)3 K = k0 _______ , fs2

(4.11)

where k0 is a coeficient that takes into account structure length scale and morphology. For simplicity this coeficient is usually considered constant, which is a rather rough approximation. The comparison of different approaches in the estimation of permeability, including experimental, analytical, and numerical methods, showed that for equiaxed grain structures and in the medium range of solid fractions the Kozeny–Carman relationship (4.5) gives adequate agreement with both experimental data [24, 25] and with numerical simulations of liquid low through an array of close-packed spheres [24]. The Wang–Beckermann model has good agreement with experimental data but requires a detailed knowledge of the real structure. A summary of the results is given in Figure 4.3. Numerical simulation of low behavior in porous structures is a good method to evaluate the permeability in cases where information about the microstructure is available. This information can be in the form of measured quantities from micrographs or in the form of a 3D digitized structure. The parameters required to calculate the permeability are the solid–liquid

AlSi5

−1 Kozeny−Carman

2

Log(KS v )

Own experiments AlCu16GR

AlSi4

−2

Numerical simulations Experiments by Nielsen et al.

AlCu10GR

Calculations with Wang− Beckermann model

−3

−4 0.1

AlCu10

0.2

0.3

0.4

f1 FIGURE 4.3 Comparison of analytical (Kozeny–Carman and Wang–Beckermann) and numerical calculations [24] with experimental measurements of permeability [24, 25]. Index GR = a grainreined sample.

Copyright 2008 by Taylor and Francis Group, LLC

136

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

interfacial area concentration, the liquid fraction or porosity and, in the case of numerical simulations, the viscosity and density of the liquid. Having detailed and adequate information on these quantities is the most important factor in determining the permeability. If more detailed and 3D information on the structures at low liquid fractions were available it could well be possible to simulate melt low in real 3D networks. Nowadays, high-brilliance X-ray radiation produced in a synchrotron provides modern researchers with a tool for studying in situ solidiication, including the possibility of 3D tomography. Information obtained has already been used for numerical assessment of permeability in real alloy structures [9]. When choosing the way to evaluate the permeability of the mushy zone in metallic alloys, it is important to take into account the accuracy required. For most solidiication models the permeability needs to be known accurate to the order of magnitude. Therefore, the simplest approximation, which is the Kozeny–Carman relationship, can be reliably used for 0.6 < fs < 0.9. However, its application to solid fractions approaching unity should be further investigated because of the so-called percolation limit at which the liquid low stops although solidiication has not yet been completed. Now let us look at the main mechanisms of macrosegregation in more detail. 4.1.3

Convection-Driven Macrosegregation

One of the most recognized phenomena behind macrosegregation is thermosolutal convection, or the melt low driven by temperature and concentration gradients. These gradients exist in the liquid (or more correctly, the luid) part of a casting (billet) due to uneven cooling of the whole volume. In the case of a DC-cast billet, the sides cool faster than the bottom, and there is a noticeable temperature difference between the central part and the periphery of the billet sump. Temperature difference determines the difference in density for the liquid. As a result of this thermal convection, the cooler liquid sinks* at the periphery and creates the momentum that forces the liquid in the center to rise. In the liquid part of the sump only thermal convection is active (if we neglect the solutal effects brought by washing out of liquid from the slurry zone, which is discussed later). As soon as the liquidus isotherm is passed, the partitioning of alloying elements starts to produce the difference in composition and corresponding difference in density, causing the solutal convection. Figure 4.4 shows the variation of densities of liquid and solid Al–Cu alloys in dependence on the composition, which relects the change of temperature during solidiication (compare with Figure 2.7) [26]. In aluminum alloys and during DC casting, low direction that resulted from both solutal and thermal convection coincide with each other, giving the overall low pattern shown in Figure 4.5 [27]. * We assume here that the liquid becomes denser (heavier) on decreasing the temperature, which is the case for most alloys. Otherwise, the direction of liquid low is opposite.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

137

3.00

At solidus

2.80 Density, g/cm3

At liquidus

2.60

2.40 0

5

10 15 Cu, wt%

20

25

FIGURE 4.4 Compositional dependence of density for Al–Cu alloys at the solidus and liquidus temperatures [26]. Refer to Figure 2.7 for the correspondence with the temperature changes during solidiication.

Hot

Water cooling

Cold

(a)

(b)

FIGURE 4.5 General thermo-solutal convective low pattern in the sump of a billet during DC casting (a) and a result of simulation of thermo-solutal convection in an Al–4.5% Cu billet (corresponds to curve 1 in Figure 4.6) superimposed on the concentration map (lighter means more copper) and isotherms of liquidus, coherency, and solidus (b). (Reproduced with kind permission of The Minerals, Metals & Materials Society (TMS).)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

138

The main reason why this low may affect the distribution of alloying elements in the billet cross-section is the penetration of this low into the slurry zone and the washing out of the liquid with the composition already changed by the solidiication process. The interaction between the liquid pool and the transition zone of the billet was noted as the main reason for convectiondriven segregation by Tageev in 1949 [28]. In addition, the thermo-solutal low may assist in transporting the solid phase within the slurry region and to the liquid pool (see Figure 3.16 and Section 4.1.5). Figure 4.5 shows that the penetration of the melt low into the slurry zone occurs in the outer quarter of the billet cross-section. Thus, the soluteenriched liquid from this part of the billet is mixed with the bulk liquid and the resultant mixture is brought to the center of the billet. The result is centerline positive segregation, as shown in Figure 4.5b and curve 1 in Figure 4.6. At the billet periphery the melt low is directed toward the surface (Figure 4.5b). Here the liquid of the nominal composition penetrates the mushy zone and dilutes the melt that is enriched there by solidiication. Hence, a negative segregation at the billet periphery is facilitated (Figure 4.5b and curve 1 in Figure 4.6). Generally, we can conclude that the natural thermosolutal convection in DC-cast billets of aluminum alloys enhances normal (direct) macrosegregation. A rather simple experiment clearly demonstrates the dependence of the macrosegregation pattern on the extent and direction of thermo-solutal convection. Figure 4.7 shows the scheme of such an experiment with a screw

0.05 Relative concentration of Cu

1 0

−0.05

3 2

−0.1 4 −0.15 0

0.02

0.04

0.06

0.08

0.1

Distance from the center, m FIGURE 4.6 Simulated distribution of Cu (relative concentration is (Cx – C0)/C0) along the radius of a 200-mm billet of an Al–4.5% Cu alloy: 1 = only thermo-solutal convection is included in the model, linear phase diagram; 2 = thermo-solutal convection and solidiication shrinkage are included in the model, linear phase diagram; 3 = same as (2) but the permeability K is increased two times; 4 = same as (2) but with Scheil approximation of solidiication path [27]. (Reproduced with kind permission of The Minerals, Metals & Materials Society (TMS).)

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

139

Crucible

Height

Screw pump

Width

FIGURE 4.7 An experimental scheme with controlled convection using a screw pump submerged in the melt along the central axis of a crucible. The low is organized either aligned with or opposite to the natural convective low.

pump made of a refractory alloy submerged along the central vertical axis into the crucible illed with liquid Al–4.5% Cu alloy. The melt low forced by this pump depends on the direction of screw rotation and can be either aligned with the natural thermo-solutal low or opposite it. The action of the pump was veriied using computer simulations. It is important that the pump creates the melt low in the axial direction of the sample, thus acting along the same directions as the thermo-solutal convection. During experiments the temperature in the vicinity of the pump was controlled to ensure that the pump operated in the fully liquid state and only inluenced the low pattern in the liquid part of the samples with subsequent penetration of the low into the transition region. The experimental results on macrosegregation proiles are shown in Figure 4.8. It is clear that the enhancement of the low in the direction of thermo-solutal convection facilitates the positive centerline segregation, whereas the low opposite to natural convection suppresses it and effectively eliminates macrosegregation. Real commercial alloys are always multicomponent. Different alloying elements can have opposite contributions to the density of the mixed melt and, as a result, can affect the overall contribution of solutal convection to macrosegregation [29]. Let us take as an example an Al–Cu–Mg alloy. Two cases can be considered: the macrosegregation of Cu without (Case 1) and with (Case 2) taking into account the macrosegregation of Mg. The solutal buoyancy caused by the addition of a second alloying element, which is only present in Case 2, may make some difference in the inal segregation pattern. It is expected that due to the negative contribution of Mg to the density in the liquid phase, which is opposite to the contribution by Cu, the extent

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

140

Relative segregation

Relative segregation

Natural convection

0.4 0.2 0.0 −0.2 −0.4

40 30 20 10 0

30 He igh 20 t, m 10 m

w ible ruc

−20

C

−30

Aligned with natural convection

−40

0.4

40

0.2 0 −0.2 −0.4 30

30 20 10 0 −20

20 t, m

igh

m

10

Cr

th,

id le w cib

−10

He

Relative segregation

,

idth

−10

mm

mm

u

−30

Opposite to natural convection

−40

0.4

40

0.2 0 −0.2 −0.4 30

30 20 10 0 20 igh t, m 10 m

−20

He

Cr

th,

id le w cib

−10

mm

u

−30 −40

FIGURE 4.8 Experimental macrosegregation pattern obtained using the setup shown in Figure 4.7. Three experiments were performed: with idle pump (natural convection), screw pump producing low aligned with the natural convection, and screw pump producing low opposite to natural convection.

of (positive) macrosegregation of Cu in Case 2 would be less than in Case 1. Results shown in Figure 4.9 conirm this estimation. Both Cases 1 and 2 predict positive centerline segregation and in Case 2 the relative segregation of Cu is 0.017, which is about 20% less than in Case 1 (0.020). In general, the low pattern will be similar to that shown in Figure 4.5b but with the following

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

141

Relative concentration of Cu and Mg

0.02 0.01 0 −0.01 −0.02 −0.03

Cal. 1: Cu Cal. 1: Mg Cal. 2: Cu Cal. 2: Mg

−0.04 −0.05 0

0.02

0.04

0.06

0.08

0.1

Distance from the center, m FIGURE 4.9 Calculated segregation proiles for a 200-mm round billet from a 2024-type (Al–Cu–Mg) alloy cast at 120 mm/min for two cases. In Case 1 only macrosegregation of Cu contributes to the solutal buoyancy, hence there is no segregation of Mg; in Case 2 both Cu and Mg contribute to the solutal buoyancy, hence both elements segregate. A Gulliver–Scheil solidiication path is implemented with a mapping technique [29] and no solidiication shrinkage is taken into account. (Reproduced with kind permission of Springer Science and Business Media.)

changes caused by Mg. The low beneath and along the liquidus contour and upward low in the center of the liquid sump in Case 2 is weaker than in Case 1. Figure 4.10 illustrates the contribution of solutal buoyancy to the low pattern in the billet sump by the velocity difference, which is calculated by the velocity ield in Case 1 minus that in Case 2. This velocity difference is more detectable close to the center of the billet, where an upward relative velocity component with a magnitude of 0.8 mm/s appears, corresponding very well to the place where the maximum difference in average concentration occurs. This velocity component brings Cu from the solidifying part of the billet to the liquid part, effectively decreasing the concentration of copper in the center. We can conclude that the addition of alloying elements, in this case Mg, will inluence the inal segregation pattern by its contribution to the solutal buoyancy. It is important to consider the contribution of every solute to the buoyancy force because they may have opposite contributions to the overall low. 4.1.4

Shrinkage-Driven Macrosegregation

Historical accounts summarized in Table 4.2 show that the importance of shrinkage-driven lows for the formation of inverse segregation was realized as early as the 1930s. Inverse segregation is caused by the movement of the solute-rich liquid in the direction opposite to the movement of the solidiication front. In the case of DC casting, that would be a melt low directed from the center to the periphery of a billet (ingot).

Copyright 2008 by Taylor and Francis Group, LLC

142

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

L

S

0.3fs

Z Y

X

FIGURE 4.10 Relative low velocities (velocity ield in Case 1 minus that in Case 2) calculated for a steadystate casting of a 200-mm round billet from a 2024-type alloy at a casting speed of 120 mm/min [29]. Isotherms of liquidus (L), solidus (S), and coherency (0.3fs) are also shown. (Reproduced with kind permission of Springer Science and Business Media.)

Original experiments performed by Livanov et al. [30] and Anoshkin [31] showed unambiguously the existence of melt low in the mushy zone, its dependence on casting speed, and its inluence on macrosegregation. During DC casting of an aluminum alloy, steel partitions were installed into the billet concentrically in two modes: (1) only to the bottom of the sump (acting only in the liquid and slurry zones) and (2) with subsequent freezing into the billet (acting all the way to the fully solid billet). After the end of the casting, the billets were sawed in the horizontal plane and the chemical composition along the diameter and the billet axis was analyzed. There were no changes in chemical composition in both directions in the case when the partitions were acting only in the liquid and slurry zones of the billet, and there were no changes in the axial directions in all cases studied. However, when the partitions were frozen into the billet, a rather dramatic difference was observed in the radial direction of the billet, as illustrated in Figure 4.11. One can see that there is a remarkable difference in copper concentration between the different

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation Cu, % 4.8

143

Undisturbed half

Half with partitions

30 mm/min

A 2

4.4

1

2

4.0 4.6

B

4

C 5

4.2

5

4 3.8 Surface

3

60 mm/min Center

Surface

FIGURE 4.11 Experimentally measured macrosegregation in a 330-mm billet from a 2024 (Al–Cu–Mg) alloy cast at 30 (1, 2) and 60 mm/min (3, 4, 5) [30]. Proiles 1 and 3 show the macrosegregation in the undisturbed part of the billet. Proiles 2, 4, and 5 show the macrosegregation in the part of the billet with partitions A, B, and C, respectively.

sides of the partition. The trend depends on the casting speed. When the casting speed is low, there is a decrease in copper content at the internal part of the partition, indicating that the solute-rich liquid tends to move toward the center of the billet. The inverse situation occurs at a higher casting speed, when the copper concentration systematically increases at the inner side of the partition and is much lower at its external part. This can be interpreted as clear evidence of the solute-rich low toward the periphery of the billet. The driving force for such a melt low in the mushy zone is the solidiication shrinkage. Solidiication shrinkage is a result of density change during solidiication and occurs throughout the entire solidiication range. In the slurry zone, however, the solidiication shrinkage is easily compensated for by the melt low. There is no pressure difference that may result in the additional lows and the solidiication shrinkage does not play any signiicant role in the relative movement of solid and liquid in the slurry zone, the main inluence being exerted by thermo-solutal convection. Deeper into the mushy zone when the permeability is limited and the feeding of the solid phase is restricted, the solidiication shrinkage (assisted closer to the solidus by thermal contraction of the solid phase) causes the pressure difference over the solidifying layer of the mushy zone that creates the driving force for the so-called “shrinkagedriven” low. The low in the mushy zone, despite its small magnitude (velocities are around 10−4 m/s or 6 mm/min as compared to casting speeds of 100 to 200 mm/min), involves the highly enriched liquid, which determines its signiicance for macrosegregation. It is important that this low is directed perpendicular to the solidiication front [11, 32]. Flemings bases modern macrosegregation theory on the interdendritic low in the mushy zone and on the ratio between the velocities of solidiication front and the shrinkage-driven

Copyright 2008 by Taylor and Francis Group, LLC

144

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

low [32]. If the shrinkage-driven low is just enough to feed the solid phase, then no macrosegregation occurs. A low faster than the solidiication rate results in inverse segregation, whereas a low with a speed less than the solidiication rate (or in the opposite direction) results in normal segregation. In DC casting, shrinkage-driven low is generally directed from a hotter part of the mushy zone to a colder one, and as such from the center to the periphery of a billet. There are two components to this low. One is horizontal along the radius directed toward the billet surface, and the other is vertical, along the casting direction, as shown in Figure 4.12 [33]. Although the vertical downward low will dilute the local volume element by bringing less enriched liquid into it, the negative centerline segregation will not form

Vh Vshr

α

Lm

α

Vcast

FIGURE 4.12 Schematic representation of the components of shrinkage-driven low in the mushy zone of a DC-cast billet [33]. Isotherms of liquidus, coherency, and solidus are shown similar to those in Figure 4.10. Angle α characterizes the inclination of the solidiication front to the horizon; Lm is the thickness of the mushy zone; Vshr and V h are the velocity of shrinkage-induced low and its horizontal component. Black arrows show the direction of shrinkage-induced low, whereas gray arrows show the direction of thermo-solutal convection. See Section 4.2.1 for more details. (Reproduced with kind permission of Elsevier/Pergamon.)

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

145

if only this vertical downward low is present because as solidiication proceeds the process eventually reaches the point that less and less solute is taken out and only solute accumulation occurs. It is the horizontal component that takes the solute away from the center to the surface [7, 11], though this solute transport physically occurs very slowly. Step by step, however, an overall solute transfer occurs from the center of the billet to its surface. The depletion in the center cannot be compensated for, as there is no horizontal inlow of the solute from more enriched regions. At the surface, there is a pile-up of the solute because there is no outlow. Because the magnitude of the shrinkage-induced low is dependent on the shrinkage ratio, one may conclude that the corresponding macrosegregation should depend on the dimensions of the mushy zone and the degree of shrinkage, as will be considered in more detail in Section 4.2.1. Computer simulations using a model that includes solidiication shrinkage demonstrate the high potential of the shrinkage-driven low in the formation of inverse segregation during DC casting [27, 29]. Figures 4.6 (curves 2–4) and 4.13 give clear evidence of this; hence, shrinkage-induced low can adequately explain the occurrence of inverse segregation. 4.1.5

Floating Grains and Macrosegregation

Movement and sedimentation/growth of solid grains in the slurry zone is sometimes considered the main mechanism of centerline segregation in DCcast billets and ingots [34, 35]. As already mentioned in Section 4.1.1, the link between the transport of solute-lean grains and inverse segregation is not a 0.06

Relative segregation of Cu and Mg

0.04 0.02 0 −0.02 −0.04 −0.06

Case A: Cu Case A: Mg Case B: Cu Case B: Mg

−0.08 −0.1 0

0.02

0.04

0.06

0.08

0.1

Distance from the center, m

FIGURE 4.13 Calculated segregation proiles for a 200-mm round billet from a 2024-type (Al–Cu–Mg) alloy cast at 120 mm/min for two cases. In Case A permeability is 6.67 × 10−11m2 and in Case B, 5 × 10−10m2. The model includes the contributions of thermo-solutal buoyancy and solidiication shrinkage. A Gulliver–Scheil solidiication path is implemented with a mapping technique [29]. (Reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

146

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

novelty. This idea was already rejected in the 1940s because it contradicted the cases of inverse segregation in castings where no loating grains had been found. However, the applicability of this mechanism of macrosegregation to DC casting was supported by numerous experimental observations of a duplex structure, characterized by a mixture of ine-branched dendrites and coarsecell dendrites (see Section 3.3) near the ingot centerline (Figure 3.17), e.g., Refs. [36–39]. The occurrence of such a duplex structure is taken as an evidence of solid-phase transport (loating) within the transition region. Hence, the mechanism of macrosegregation by loating grains was reinstated in the 1980s. The mechanisms of loating-grain formation have been already discussed in Section 3.3. Let us now look at the reasons why negative segregation in the DC-cast billet/ingot center is attributed by some researchers to the presence and accumulation of the loating dendrites at the bottom of the sump, although their presence is not a necessary condition for the centerline depletion to occur [40–42]. The mainstream concept assumes that if the coarse-cell dendrites are solute poor, ine-cell dendrites are richer in solute and close in their average composition to the nominal alloy composition [34, 43]. Quite an opposite line of argument is proposed by Chu and Jacoby [35], who suggested that the ine-cell dendrites originated from the start of the solidiication in the region of rapid cooling (i.e., dendrites detached and transported from the periphery to the center and frozen into the solidiication front without further growth) and coarse-cell grains grew in situ. Consequently in their opinion, ine-cell grains are solute poor and responsible for the negative centerline segregation, while coarse-cell dendrites are solute rich. Yet another suggestion is made by Glenn et al. [44] and Lesoult et al. [45] based on their examination of 5051-alloy DC-cast ingots. They found that the structure of a nongrain-reined billet contains, along with normal-size dendritic grains, ine grains, which allegedly represent grain fragments brought from the mold walls to the ingot center by turbulent lows. If we assume that loating grains arrive from other parts of the billet and settle in its center, then they effectively bring there more solid phase than should be there at this point in time and space. Because the primary solid in hypoeutectic aluminum alloys is always depleted of the solute, the accumulation of the loating grains has to result in negative segregation. In addition, the different solidiication schedule of coarse- and ine-cell dendrites may result in different bulk compositions of these grains. Obviously, the microsegregation pattern of loating grains would shed light on the extent of their potential inluence on macrosegregation. Quantitative analysis of compositional differences in the grain interior is, however, rarely reported, e.g., Yu and Granger [34] have observed a uniform plateau of copper depletion away from the cell boundary in a coarse dendrite. We performed local composition electron-probe microanalysis (EPMA) measurements on several coarse and ine cells in the duplex structures found in the center of 200-mm billets from a 2024 alloy cast at 80 mm/ min [46]. The nominal composition of this alloy was 3.6 wt% Cu and 1.4 wt% Mg. Figure 4.14 shows the line scans of Cu and Mg concentrations in

Copyright 2008 by Taylor and Francis Group, LLC

Mg Cu

2.5 Concentration, wt%

2.5 Concentration, wt%

Macrosegregation

3.0

3.0

2.0 1.5 1.0 0.5

2.0 1.5 1.0 Mg Cu

0.5 0.0

0.0 0

100

200

300 400 Distance, µm

(a)

500

600

700

0

100

200

300 400 Distance, µm

500

600

700

(b)

Copyright 2008 by Taylor and Francis Group, LLC

147

FIGURE 4.14 Measured microsegregation in coarse-cell (a) and i ne-cell (b) grains found in the center of a 200-mm billet cast at 80 mm/min from a nongrain-rei ned 2024 alloy [40].

148

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

TABLE 4.4 Minimum Concentrations of Alloying Elements in Dendrite Cells Measured by EPMA and Difference between Nongrain-Reined (NGR) and Grain-Reined (GR) Samples from 2024-Alloy Billet Cast at 80 mm/min NGR Element Cu, % Mg, %

GR

Coarse

Fine

Difference

Coarse

Fine

Difference

0.72 0.47

1.1 0.7

0.38 0.23

0.73 0.46

1.1 0.7

0.37 0.24

nongrain-reined (NGR) DC-cast samples, clearly demonstrating that the coarser cells are more solute depleted compared to the “regular” i ner cells. Grain reining (GR) does not change the situation. From these measurements, the minimum Cu and Mg concentrations in the center of dendrite cells are listed in Table 4.4. The contribution of loating grains seems to explain the negative centerline segregation alone, not invoking other mechanisms, e.g., shrinkage-induced low. The latter conclusion is in direct opposition to the conclusion of the previous subsection, where we stated that shrinkageinduced low can adequately explain inverse segregation. Obviously, these two mechanisms are the key factors in the formation of negative centerline segregation. It is important to realize that shrinkage-induced low is a physical phenomenon that is always present in the transition region of a solidifying billet, whereas the transport of solute-lean solid with its accumulation in a certain part of a casting is a conditional incident, the occurrence of which depends on a number of factors such as the design of the mold, structure evolution, temperature regime, and the direction of strong lows in the sump. It is clear that the only contribution that the loating grains can make to the segregation is negative. This can be explicitly demonstrated by experiments with solidiication under constant low, as explained below. Figure 4.15a shows the simulated transport of solid grains in the experiment where the solidiication occurs in a chilled cavity embedded into the bottom of a trough with a constantly lowing melt [47, 48]. The corresponding macrostructure with a clearly visible pile of equiaxed grains at the downstream side of the cavity is given in Figure 4.15b (at low low velocities, the structure appears to be uniform and columnar along the entire length of the sample). The change in the macrosegregation pattern with the increasing melt low velocity is presented in Figure 4.15c. There is an obvious correlation between the negative segregation at the downstream section of the sample and the transport of solute-lean grains by the low. The correct estimation of this contribution in the case of DC casting, however, poses a dificulty because the complexity of phenomena that should be taken into account includes, among other things, the following: (1) microsegregation development as a result of different solidiication times and diffusion

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

149

(a)

(b) 0.10 0.03 m/s Relative segregation of copper

0.1 m/s 0.3 m/s

0.05

0.00

−0.05

−0.10 0

20

40

60

80

100

Longitudinal section of sample, mm (c) FIGURE 4.15 Effect of solid transport by melt low on the macrosegregation in an Al–4.5% Cu alloy: (a) computer-simulated transport of solid upon melt low from the left to the right; (b) corresponding macrostructure obtained at a low velocity of 0.3 m/s; and (c) macrosegregation patterns obtained at different melt velocities showing the increasing negative segregation at the downstream part of the sample [48].

distances; (2) altered solidiication path because of the local introduction of excess solid fraction with corresponding change of eutectic fraction; (3) changed kinetics of dendrite coarsening, recalescence, and back diffusion; and (4) competition between the settling of loating grains under gravity and the resistance to such settling from the solid-phase dragging by viscous liquid. The evaluation of the contribution of coarse-cell grains to the centerline macrosegregation based on experimental measurements will be discussed in Section 4.3.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

150

Taking into account that there is still a great deal of controversy in the reported data on the appearance of duplex structure in DC-cast billets and ingots and the corresponding macrosegregation pattern (Table 4.5), doubts were expressed as to what extent the loating grains could affect macrosegregation [32]. At the present stage of our expertise, we can conclude that the loating grains contribute to inverse segregation but the extent of this contribution should be examined in every speciic case. It is possible that the contradictory results are caused by the varying mobility of the loating grains (or fragments) in diverse amounts, sizes, and morphologies of the solid phase under different convective conditions. Lesoult [49] reported that the intense negative segregation resulted from a moderate lux density of the loating solid phase, whereas the segregation was effectively reduced when a larger density of the loating grains caused their early immobility. Smaller loating TABLE 4.5 Summary of Reported Observations on Duplex Grain Structures and Macrosegregation in DC-Cast Ingots and Billets

Alloy

Ingot Size, Billet Diameter, mm

Not Grain Reined (NGR)*

Grain Reined (GR)*

Macrosegregation

Ref.

Al Al–(1–4)% Cu Al–3.6% Cu Al–4.5% Cu

900 × 300 ∅200

— D*, CF*

D*, CF* —

N/A Inverse

36 37, 38

∅200

Not observed

D, CF

40

∅533.4

Not observed

D, CF

Inverse in both cases Inverse for NGR

2024

∅400

D, CF

2024

2024

Flat ingot, commercial size ∅200

D, CF

D, CF

7075

∅200

Not observed

D, CF

7XXX 5182

1270 × 406.5 1850 × 550

D, FF* D, FF (ine grains)

5182

1050 × 550

6061

∅200

D, FF (ine grains) Not observed

— Not observed, FF (ine grains) FF (ine grains) D, CF





D, CF

Normal in GR Normal at low casting speed Inverse at high casting speed Inverse

Inverse in both cases Inverse in both cases Inverse Inverse in both cases Inverse in both cases Inverse in both cases

41

43

34

72 42 35 45

44 39

* GR = grain reined; NGR = not grain reined; CF = loating grains with coarse cells; FF = loating grains with ine cells; D = duplex structure.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

151

grains in grain-reined billets may be also less prone to sedimentation than larger grains in nongrain-reined counterparts. The change in coherency temperature and permeability with grain reinement can affect the ratio of contributions of different macrosegregation mechanisms and the resultant distribution of alloying elements. Successful attempts to model the contribution of loating grains to the macrosegregation in DC casting include a description of the relative velocity of the solid and liquid phases by a special drag term [23]. A more physical approach includes the tracing of growing solid particles and, hence, multiphase modeling. Application of this methodology to steel ingot casting showed its feasibility for modeling macrosegregation caused by loating grains [49]. 4.1.6

Deformation-Driven Macrosegregation

Volumetric deformation of the dendritic network in the coherent mushy zone caused by thermal contraction (see Section 5.1 of Chapter 5) can cause melt low [50]. Like the solidiication-shrinkage low, it occurs in the mushy zone and its velocity is rather small. It has been shown that the contribution of this low to macrosegregation can be as important as that of shrinkage-induced low [50]. Deformation can be external (in the case of continuous casting of steels) or thermally induced. The latter applies to DC casting of Al alloys. Results indicate [50, 51] that even small volumetric strains (∼2%), which are associated with thermally induced deformations, can lead to segregations comparable to those resulting from solidiication shrinkage. The compression will artiicially increase the amount of the solid phase in the unit volume and squeeze the solute-enriched liquid out of this volume, effectively decreasing locally the concentration of the solute. In this case, the mechanism of negative (inverse) segregation is similar to that caused by the transport of loating grains. In the case of tension that occurs in the center of a DC-cast billet, the situation should be the reverse. The solute-rich liquid penetrates into the volume, resulting in positive segregation. This mechanism is important for the subsurface and surface segregation [52] when exudation occurs (see Figure 3.31). Close to the surface of a billet, relatively high solidiication and cooling rates in a narrow shell result in large thermal gradients. Under such conditions, the rate of solidiication shrinkage and thermal contraction is also high compared to the inner part of the billet. This causes large shrinkage- and deformation-induced lows directed toward the surface of the billet. These lows are further assisted by the aligned convection low (see Figure 3.9), the metallostatic pressure of the melt, compressive strains in the billet outer shell, and weakening of the mushy zone by partial remelting in the air-gap region (Figure 3.14a). As a consequence of all these phenomena, a strong positive segregation forms at the surface of the billet (sometimes extended to liquation or penetration of the liquid melt through the shell) accompanied by a zone of negative subsurface segregation. The latter is a result of incomplete compensation of the surface segregation by the incoming melt and strong compressive strain that squeezes the liquid to the surface.

Copyright 2008 by Taylor and Francis Group, LLC

152

4.2 4.2.1

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Effects of Process Parameters on Macrosegregation during DC Casting Effect of Casting Speed on Macrosegregation during DC Casting

Since the early years of DC casting practice, it has been recognized that the casting speed is the main parameter that determines the extent of macrosegregation, and in some cases even the macrosegregation direction (see Figure 4.11). The latter issue is discussed in Section 4.3. The effect of casting speed on macrosegregation is illustrated in Figure 4.16. The empirical rule says that the casting speed should not exceed the ratio A/D in order to minimize the macrosegregation in a round billet while maintaining acceptable productivity. In this ratio A is an alloy-dependent constant ranging from 25,000 to 33,300 for aluminum alloys and D is the billet diameter in mm [7]. Hence, for a 200-mm billet, the maximum allowable casting speed should be between 125 and 167 mm/min, which gives the maximum relative segregation of 3 to 6% of the nominal composition (Figure 4.16). Dobatkin [53] emphasized two main geometrical parameters of the sump that affect the extent of inverse macrosegregation upon DC casting, namely, the size of the transition region and the slope of the solidiication front (see Figure 4.12). We have already discussed in detail the effect of casting speed on the dimensions of the transition region (see Section 3.1). The fact that the centerline macrosegregation increases with the casting speed gives a hint that there might be a correlation between the macrosegregation and the dimensions of the transition region. Indeed, the macrosegregation proiles observed at different casting speeds, being normalized to the vertical thickness of the transition region, fall onto each other through the most of the billet cross-section, except for the subsurface region, as shown in Figure 4.17 [54]. The peculiar behavior of macrosegregation at the periphery of the billet can be explained by the additional contribution of deformation- and shrinkage-driven lows (see Section 4.1.6). Why does the wider transition region facilitate inverse macrosegregation? It is important to note that the deepening of the sump and widening of the transition regions affect mostly the dimensions of the slurry zone where the main acting mechanisms of macrosegregation are thermo-solutal convection and loating grains. These two mechanisms act in different directions as we discussed in Sections 4.1.3 and 4.1.5. They can also compensate for each other. On the one hand, the larger slurry zone creates a possibility for the convective lows to penetrate more deeply into the transition region and wash out more solute-rich liquid at midradius and bring it to the center of the billet. On the other hand, there are more possibilities for the transport of solid phase (loating grains) in a loose slurry zone and their settling in the center of the billet [38, 54]. The deepening of the sump also changes the geometry of the solidiication front by increasing the slope of the solidiication front and

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

153

Deviation of copper concentration from the average

0.04 Center

0.02 0 −0.02 −0.04 −0.06 120 mm/min

−0.08

160 mm/min −0.1 0

40

120

80

160

200

Cross-section of billet, mm (a)

Deviation of iron concentration from the average

0.06 Center

0.04 0.02 0 −0.02 −0.04 120 mm/min

−0.06

160 mm/min

−0.08 0

40

80

120

160

200

Cross-section of billet, mm

(b) FIGURE 4.16 Effect of casting speed on macrosegregation in a 200-mm billet of an Al–4.3% Cu (0.22% Fe and 0.11% Si as impurities): (a) relative segregation of copper and (b) relative segregation of iron.

affecting the shrinkage-induced low in the mushy zone. This phenomenon tends to produce inverse macrosegregation (see Section 4.1.4). A simple analytical model [33] is suggested to estimate the magnitude of the shrinkage-induced macrosegregation during DC casting in relation to the steepness of the solidiication front. The shrinkage-induced low, which is almost perpendicular to the coherency-fraction contour, can be broken down into two components (Figure 4.12). Let us consider a small volume in the mushy zone as shown in the Figure 4.12 insert. The shrinkage low is directed normally to the coherency isotherm. The coherency isotherm in this region can be approximated by a straight line that

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

154 0.006

Normalized relative segregation of Cu

120 mm/min

200 mm/min

0.002

−0.002

−0.006

−0.01 0

20

40

60

80

100

Distance from the billet center, mm FIGURE 4.17 Relative macrosegregation of copper normalized to the vertical thickness of the transition region (see Figure 3.2c) in a 200-mm billet from an Al–4.3% Cu alloy [54]. (Reproduced with kind permission of Elsevier.)

is inclined with the angle α to the horizon. The total solidiication shrinkage in the direction normal to the coherency isotherm can be written as ε = Lm cos α Sβf l,

(4.12)

where Lm is the vertical dimension of the mushy zone, S is the surface area of the horizontal cross-section of the volume, β is the volumetric shrinkage (can be taken as 0.1 for aluminum alloys), α is the angle between the tangent to the coherency isotherm and the horizon, and f l is the liquid volume fraction. This shrinkage has to be compensated for by the inlow Vshr with the following volume: V = VshrS Lm/Vcast,

(4.13)

where Vshr is the shrinkage low velocity, Vcast is the casting speed, and Lm/Vcast is the average time available for the shrinkage. By equating (4.12) and (4.13) we obtain: Vshr = Vcast f l β cos α.

(4.14)

We assume that only the horizontal component of shrinkage low is responsible for segregation, which is Vh = Vshr sin α = Vcast f l β(sin2α)/2.

(4.15)

The solute lux caused by this horizontal velocity will be F = CLVh = CLVcast f l β(sin2α)/2, where CL is the liquid concentration.

Copyright 2008 by Taylor and Francis Group, LLC

(4.16)

Macrosegregation

155

The integration of Equation 4.16 over time will lead to the overall amount of transferred solute. The evolution of the liquid phase concentration (CL ) and volume fraction ( f l) derived from heat transfer analysis have to be provided to complete this integration. For a irst-order approximation, we can assume a linear evolution of liquid volume fraction during the period of time (t) required to pass the mushy zone, i.e., f l = 1 – (Vcast/Lm)t

(4.17)

and then the liquid concentration in weight percent can be obtained based on the lever rule (for simplicity, though nonequilibrium solidiication path with back diffusion in the solid should be taken into account in a more thorough analysis): C0 CL = ______________________ 100%, f 1 ___________ (1 – K) + K 1 + (1 – f1)β

(4.18)

where C0 is the nominal concentration and K is the partition coeficient. Note that the presence of the shrinkage ratio β here is for the purpose of converting the liquid volume fraction into the weight fraction. The total amount of the transferred solute during the time period of Lm/Vcast or the horizontal solute transfer distance is as follows: Lm/Vcast

Lh = ∫0

CLf1 Vcast β(sin2α)/2 dt Lm/Vcast C0 f 1 _____________________ dt. = Vcast β(sin2α) ∫0 f 1 ___________ (1 – K) + K 1 + (1 – f1)β

(4.19)

Let us take as an example the Al–Cu system [33]. K is equal to 0.171 and the shrinkage ratio β can be set to 0.1 (though the solidiication shrinkage can be as large as 0.117 for an Al–4% Cu alloy). Substituting these values in Equation 4.19, we obtain Lh = 0.78 C0Lm β(sin2α)/2.

(4.20)

The derivative of Equation 4.20 with respect to radial distance from the billet center R will lead to the net eflux and is a measure of the macrosegregation caused by solidiication shrinkage, with (dLh/dR)/C0 relecting the relative segregation. The application of this analytical model to DC casting at 100 and 120 mm/min of a binary Al–4% Cu alloy yielded the following results [33]. The isotherms of liquidus, coherency, and solidus are calculated upon computer simulation of DC casting using the model and software described in detail elsewhere [55] and the representation of results is given in Figure 4.18a,b. Different casting

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

156

(b)

Lh = −0.00089 + 0.0832 R − 0.0128 R2 + 0.0006 R3 − 1.6399 10−5 R4 0.3

0.2

0.1

0 0

2

4

6

8

10

Amount of transferred solute (Lh), wt% cm

Amount of transferred solute (Lh), wt% cm

(a)

Lh = 0.00737 + 0.1676 R − 0.0373 R2 + 0.0036 R3 − 1.5627 10−4 R4

0.3

0.2

0.1

0 0

2

4

6

8

10

Distance from the center of the billet (R), cm

Distance from the billet center (R), cm

(c)

(d)

FIGURE 4.18 Evaluation of shrinkage-induced macrosegregation upon DC casting of a 200-mm billet from an Al–Cu alloy at 100 (a,c,e) and 120 mm/min (b,d,f): (a,b) solidiication proiles with isotherms of liquidus, coherency, and solidus; (c,d) amount of transferred solute along the billet radius; and (e,f) relative macrosegregation (net solute eflux) [33]. (Reproduced with kind permission of Elsevier/Pergamon.)

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

157

0.04

Relative net solute efflux

0.02

0

−0.02

−0.04

−0.06 0

2

4

6

8

10

Distance from the billet center, cm (e) 0.04

Relative net solute efflux

0.02

0

−0.02

−0.04

−0.06 0

2

4

6

Distance from the billet center, cm (f) FIGURE 4.18 (Continued)

Copyright 2008 by Taylor and Francis Group, LLC

8

10

158

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

speeds result in different proiles of the isotherms and different dimensions of the sump and the mushy zone. It can be expected that a shallower sump in the case of casting at 100 mm/min will produce less macrosegregation. Figure 4.18c,d shows the distributions of the amount of transferred solute (or the horizontal solute transfer distance) along the radius of the billet. In this case, we took Lm and α at equal distances from the center to the surface of the billet from the proiles in Figure 4.18a,b, and C0 is taken as 4% Cu. Simple analysis of the curves shows that the maximum solute transfer distance corresponds to the region around midradius of the billet, where, in practice, the segregation is either negligible or positive. It seems logical if we interpret the results as follows. Solute-rich liquid is transported by the shrinkage-induced low from the center to the periphery of the billet (see Figure 4.11). The greater the distance from the center, the more solute is transported. Starting from the distance corresponding to the maximum in Figure 4.18c,d, more solute arrives than goes away, so the accumulation of solute starts, producing positive segregation. The macrosegregation can then be better described by the irst derivative of the amount of transferred solute (with the negative sign) divided by the alloy nominal composition, which relects the relative segregation. Figure 4.18e,f shows the radial distribution of the irst derivative of the polynomial its shown in Figure 4.18c,d. One can easily see a good semiquantitative agreement between the distribution in Figure 4.19d and the real macrosegregation proile shown in Figure 4.16a. It should be noted that the suggested model adequately responds to the change of the casting conditions and corresponding change in the sump proile. Less segregation is predicted for the shallower sump at a lower casting speed, as shown in Figure 4.18c,d. Additional features in the real segregation proile in Figure 4.16a, such as the latter region of slightly positive segregation in the midradius region and the negative subsurface segregation, can be attributed to the effects of thermo-solutal convective lows and surface exudation, which were not accounted for in the model shown. The loating grains may also contribute to the extent of the negative centerline segregation. The proposed model links the macrosegregation to the local slope of the coherency isotherm, the thickness of the mushy zone, the shrinkage ratio, and the solidiication path of an alloy. We can conclude that the main reasons for the dependency of macrosegregation on the casting speed are the changes in the dimensions of the transition region (which affects thermo-solutal convection and transport of solid) and the slope of the solidiication front (which affects shrinkagedriven segregation). In practice, the macrosegregation is negligible when the slope α (Figure 4.12) is less than 30° and acceptable when it is less than 60° [53]. Another factor that may change the geometry of the sump is the water low rate. But, as we already discussed in Section 3.1, its effect is insigniicant if the amount of cooling provided is suficient. Our data on macrosegregation conirm this [54].

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation 4.2.2

159

Effect of Melt Temperature on Macrosegregation during DC Casting

The melt temperature is an important part of any casting recipe. We have already discussed in Sections 3.1, 3.3, and 3.4 the effects of melt temperature on the geometry of the sump, low patterns during casting, and the structure of aluminum billets. The increase in the melt temperature does not greatly affect the geometry of the solidiication front and the dimensions of the mushy zone, but it narrows the transition region by lowering the position of the liquidus in the sump (see Figures 3.8c,d and 3.9). A higher melt temperature also facilitates the strong melt low toward the surface of the billet and the deeper penetration of the melt low into the mushy zone (Figure 3.9). The amount of loating grains decreases with melt temperature at low casting speeds and increases at high casting speeds (Figure 3.24c,d). What is even more important for macrosegregation is that at high melt temperatures the loating grains concentrate in the central part of the billet at low casting speeds and tend to spread more over the entire cross-section at high casting speeds. These observations allow us to forecast the effect of melt temperature on macrosegregation. The thermo-solutal convection in the subsurface area and a generally weaker shell of the billet will facilitate subsurface segregation and, possibly, liquation and bleed-outs (see Figure 3.31). The thermosolutal currents in the slurry and mushy zone that are directed toward the center of the billet bring more solute-rich and heavier liquid to the center and should promote positive centerline segregation. This tendency is opposed by an increased amount of loating grains in the center, which are, however, also distributed more evenly in the billet cross-section. Finally, the contribution of shrinkage-driven segregation should not depend on the melt temperature because the thickness of the mushy zone and the slope of the solidiication front are not affected by the melt temperature. All in all, we can expect, as the melt temperature increases, a higher macrosegregation close to the surface of the billet and no signiicant changes in the rest of the billet. Our experimental observations fully support this conclusion, as illustrated in Figure 4.19. Our results and arguments concur with the suggestions of Dobatkin [53], Tarapore [56], and Reese [57] that macrosegregation should decrease as casting temperature increases. 4.2.3

Dimensions of the Billet: Scaling Rule of Macrosegregation

Examination of numerous data on macrosegregation indicates a distinct dependence of macrosegregation on the dimensions of the billet or ingot. Taking into account the interrelation between macrosegregation and the shape and dimensions of the sump, which are affected by the size of the billet and the casting speed, this dependence should be a function of these two parameters. Livanov et al. [30] studied macrosegregation in 2024-alloy billets in a wide range of billet diameters, from 50 to 470 mm, and of casting speeds, from 30 to 350 mm/min. They showed that the extent and direction of

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Relative segregation of Cu

160

Centerline

0.3 0.2 0.1 0.0

760 200 740

160

Bille

ss-s

ecti

on,

mm

lt Me

40 0 700

°C

p

720

80

re,

tu era

120

t cro

tem

200 180

0.06 0.02

0.06 0.02

0.06 0.02

0.08 0.06 0.02

Billet cross-section, mm

160 0.02 0.02

140 120

−0.02

−0.02

−0.02

−0.02

100 80

Centerline −0.02

60

−0.02

−0.02 0.02 0.02

40 20 0 700

−0.02

0.02 710

720 730 740 Melt temperature, °C

750

760

FIGURE 4.19 Experimentally measured macrosegregation in Al–2.8% Cu billets cast at 200 mm/min and at different melt temperatures [38]. (Reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

161

macrosegregation depends on the average cooling rate in the solidiication range. Macrosegregation is negligible at very low cooling rates (no microsegregation), then the normal segregation occurs. Upon further increasing the cooling rate, normal segregation starts to decrease and, eventually, turns to inverse segregation. There exists a certain cooling rate for each billet diameter when the centerline macrosegregation is negligible. For example, this threshold cooling rate is about 2.3 K/s for a billet diameter of 300 mm and almost 11 K/s for a diameter of 168 mm (see Figure 4.20a). The extent of inverse segregation increases with the cooling rate, and then decreases at very high cooling rates. This dependence resembles the cooling-rate dependence of microsegregation, as shown in Figure 2.9a. Evidently, the cooling rate cannot be a universal criterion for macrosegregation because its critical value depends on the casting dimensions. Livanov et al. made the remarkable conclusion that the similarity of thermal and strain ields should lead to a similarity in macrosegregation patterns. They suggested a similarity criterion [30]: SC = VcastD/a,

(4.21)

where Vcast is the casting speed in mm/min, D is the billet diameter in mm, and a is the constant that is about 12,000 (for normal molds without hot tops) for zero-centerline segregation. This dependence is illustrated in Figure 4.20b. If SC is less than unity, the macrosegregation is normal, hence positive in the center of the billet. If SC is larger than unity, the macrosegregation is inverse, therefore negative in the center. The transition from positive to negative macrosegregation for a 200-mm billet should occur at a casting speed of about 60 mm/min, which is far below the optimum (and commercially applicable) speed. Let us try to summarize the results on macrosegregation in a wider range of casting speeds, dimensions, and alloys, based on references and our own data. In this case we need a report on the billet (ingot) dimensions together with the casting speed and macrosegregation observations; the latter can be qualitative (normal vs. inverse or positive vs. negative). Tables 4.6 and 4.7 present a literature survey of the last 30 years that includes experimental and numerical (calculation) reports [7, 23, 30, 34, 35, 41, 43, 45, 54, 58–63]. All these seemingly diverse results are then plotted on billet diameter–casting speed axes, as shown in Figure 4.21. One can clearly see that it is possible to draw a line that separates the regions of positive (under the line) and negative (above the line) centerline macrosegregation in aluminum-alloy billets [64]. The line for grain-reined alloys (VcastD = 21,774) apparently lies higher than the one for nongrain-reined alloys (VcastD = 12,465), indicating more possibilities for positive centerline segregation in grain-reined billets, which agrees well with the experimental observations of Finn et al. [41] and our own results, shown in Figure 4.22. Our computer simulations of three billets of different dimensions produced at different casting speeds (the four cases shown in Figure 4.23) give results that it well onto the plot in Figure 4.21 (shown by crosses), with one regime

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

162 0.3

70 mm

0.2 Deviation from average composition, wt%

168 mm 0.1

330 mm

0 10

100

−0.1 −0.2 −0.3 −0.4 Cooling rate, K/s (a) 0.3

Deviation from average composition, wt%

0.2 0.1 0 5000

10000

20000

15000

25000

30000

35000

−0.1 −0.2 −0.3 −0.4 −0.5 VcD, mm2/min (b)

FIGURE 4.20 Relationships between the macrosegregation pattern of Cu and (a) cooling rate for different billet diameters and (b) scaling parameter VcastD. These plots are based on experimental data for DC casting of a 2024-type alloy in conventional molds without hot top [30].

producing a positive centerline (normal) segregation, i.e., the 200-mm billet cast at 60 mm/min. These values are in good agreement with the experimental data of Livanov et al. [30]. This is a good indication that the model we used works well and agrees with a set of experimental data.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

TABLE 4.6 Relationship between the Size of a Billet, Casting Speed, and Positive Centerline Macrosegregation [66] D, mm*

Vcast, mm/min

Grain** (perm.)

Comp., wt%

85–75 60 60 38 60

GR, ea GR, ea GR, ea NGR. ea NGR/GR

5.25 Cu 4.5 Cu 4.5 Cu 2024 6.0 Cu

450calc

60

NGR/GR

6.0 Cu

533.4exp 533.4calc

38 38 45–60

GR GR (high) NGR (low) GR, ea

4.5 Cu 4.5 Cu 4.5 Cu Al–Cu–Mg

30–200

N/A

2024

218–282calc 400calc 400calc 400exp 450exp

Standard commercial ingot 55–330exp

Comments***

Ref.

— — Positive without FG; negative with FG Mold with hot top Positive/negative; airslip mold; in NGR only central point is positive Positive overestimated/negative underestimated; airslip mold; K0 two times less than in [23] Casting with loat — Negligible segregation Scatter, at 45.4 mm/min—some points are positive; at 60 mm/min—more negative Mold without hot top

59 60 23 43 61 61 41 58 58 34 30

* exp. – experimental data; calc. – data from computer simulation. ** GR – grain reined; NGR – not grain reined; ea – equiaxed; perm. – permeability. *** FG – loating grains; K0 – permeability constant.

163

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

164

TABLE 4.7 Relationship between the Size of a Billet, Casting Speed, and Negative Centerline Macrosegregation* [66] Vcast, mm/min

Grain (perm.)

533.4exp 533.4calc 195–370exp 1800 × 550

38 38 60–190 60

NGR GR ine (low) NGR NGR; GR

4.5 Cu 4.5 Cu Al–Cu–Mg 5182

150 253calc, exp 200exp

54 N/A 120–200

(High) (High) NGR, ea

4.5 Cu 4.5 Cu 4.5 Cu

N/A 45–60

ea GR, ea

Al-Zn-Mg-Cu Al–Cu–Mg

35–350

N/A

2024

D, mm

406 × 1270 Large commercial ingot 55–330exp

Comp., wt%

Comments

Ref.

— — Mold without hot top Larger in GR; not related to FG that are observed in NGR high viscosity — Fraction of FG increases with Vcast FG have ine DAS FG have coarse DAS, 30 vol.% Mold without hot top

41 58 7 45

62 63 54 35 34 30

* See footnotes for Table 4.6. 400

Positive, GR Positive, NGR Negative, GR Negative, NGR NGR GR

Casting speed, mm/min

300

Positive, our simulation Negative, our simulation

200

100

0 0

200 400 Billet diameter, mm

600

FIGURE 4.21 Centerline segregation in relation to casting speed and billet diameter based on reference data from Tables 4.6 and 4.7 (circles and rhombs) and our computer simulations [66] (crosses). Lines represent a threshold between positive (under the line) and negative (above the line) segregation for nongrained-reined (solid line) and grain-reined alloys (dashed line).

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

165

Relative segregation of Cu

0.06 0.04

Center

80 mm/min

0.02 0 120 mm/min

−0.02 −0.04 −0.06

192

−0.08 100

50

0

150

200

Billet cross-section, mm FIGURE 4.22 Experimental data on macrosegregation of copper in a 200-mm grain-rei ned Al–3.8% Cu billet cast at two casting speeds [40]. Casting at 80 mm/min produces positive centerline segregation in a 200-mm billet.

Relative segregation of Cu

0.1 4

0.05

1

3 0 −0.05 2 −0.1

1 - diameter 500 mm, speed 60 mm/min 2 - diameter 200 mm, speed 200 mm/min 3 - diameter 200 mm, speed 60 mm/min 4 - diameter 100 mm, speed 300 mm/min

−0.15 −0.2 0

0.05

0.1

0.15

0.2

0.25

Radius, m FIGURE 4.23 Computer-simulated macrosegregation proiles (deviation of Cu concentration from the nominal alloy composition Al–4.5% Cu) along the radius (billet center at R = 0) for four test cases [64]. Casting at 60 mm/min produces positive centerline segregation in a 200-mm billet.

Let us now look now at the geometry of the sump in the billet and try to draw some conclusions about the reasons for the scaling dependence given in Figure 4.21. Figure 4.23 summarizes the calculated segregation proiles. One can see that not only the direction of macrosegregation but also its magnitude differs in relation to the diameter/casting speed ratio. Figure 4.24 and Table 4.8 show the sump proiles and relevant sump parameters for the test cases.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

166

Mold region S Sump depth

Distance between liquidus and solidus

L

S

Ang

le

(a)

(b)

FIGURE 4.24 Positions of liquidus (i.e., 0.05% solid) (L) and solidus (S) isotherms in 200-mm Al–4.5% Cu billets cast at 200 (a, inverse segregation) and 60 mm/min (b, normal segregation). Characteristic features given in Table 4.8 are marked in (a). Typical low pattern is shown in (b) with convection lows indicated in the upper part of the transition region and the shrinkage-induced low indicated closer to the solidus.

Regression analysis of data given in Table 4.8 shows that the strongest inluence on the extent and sign of centerline macrosegregation (CS) is exerted by the thickness of the transition region normalized to the billet radius (Th) and the slope of the solidus isotherm (A): CS = –0.06124 – 4.341 × 10 –2 Th + 1.64 × 10 –3A; R = 0.65,

(4.22)

where R is the regression coeficient. Note that the thicker transition region (and the sump depth correlated with it) facilitates the negative segregation, whereas the larger angle (i.e., a smaller slope of the solidiication front and a latter sump) acts in the opposite direction. These observations correlate well with the dependences that we discussed in Section 4.2.1. The scaling dependence in Figure 4.21 and corresponding sump parameters in Table 4.8 show that the increased casting speed at the same billet diameter will result in a deeper sump with a steeper solidiication front, and in the transition from normal to inverse segregation. The analysis of the same scaling dependence tells us that for a given casting speed, a larger billet is more prone to negative (inverse) centerline segregation. A large billet usually means a slower cooling, a deeper sump, and a thicker transition region. The thicker transition region implies a thicker slurry zone. A higher

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

TABLE 4.8 Parameters of the Sump for the Test Cases Shown in Figure 4.23 [64] Casting Parameters Vcast, mm/min

D, mm

300 200

100 200

60

500

60

200

Centerline Segregation (S) – 0.04, negative – 0.16, strongly negative – 0.02, slightly negative +0.01, slightly positive

Distance between Liquidus and Solidus in the Center of the Billet, mm/ the Same Normalized to the Billet Radius (Th in Equation 4.22)

Angle between the Billet Axis and the Tangent to the Solidus Isotherm at Midradius (A in Equation 4.22), degrees

153/3.06 220/2.2

70.5/1.41 117/1.17

15 16

383.5/1.534

82.4/0.33

23

88/0.47

47/0.47

55

Sump Depth, mm/ the Same Normalized to the Billet Radius

167

Copyright 2008 by Taylor and Francis Group, LLC

168

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

ratio between the thickness of the transition region and the billet radius suggests a deeper sump. Flood and Davidson performed a scaling analysis of the low patterns in the sump of a DC-cast ingot and came to the conclusion that the sensitivity of macrosegregation to the ingot thickness and to the casting speed was a result of the changed sump dimensions [65]. They stated that the increase in ingot thickness in the commercially relevant range (400 to 600 mm) had a greater effect on the sump depth than the casting speed. The interpretation of the observations summarized in Table 4.8 and by Equation 4.22 agrees with the suggestions of Anoshkin and Dobatkin [66] that the direction and extent of macrosegregation is determined by a ratio between thermo-solutal convection and shrinkage-induced low, as shown by arrows in Figure 4.24b (see also Section 4.2.4). This relationship, however, is not straightforward, and the inal macrosegregation pattern is a result of a fragile balance between the amount and composition of liquid that is brought to the center of the billet by the convective low and taken from the center by the shrinkage-induced low, and the direction of relative movement of solid and liquid phases in the slurry zone. As we already know, there are several co-related phenomena that can affect the macrosegregation. First of all, the deep sump enhances the shrinkageinduced low and the negative centerline segregation, as we discussed in Section 4.2.1. At the same time, the wide slurry zone with loose solid grains suspended and loating in the liquid matrix provides more opportunities for the thermo-solutal convection low to penetrate into the transition zone and to wash out solute-rich liquid from it (Figure 4.24b). This liquid can go up and mix with the bulk melt or can be transported toward the center along the liquidus isotherm, settle there because of its higher density, and effectively induce the positive segregation. This process is facilitated by grain reining, when the solid phase becomes coherent at lower temperatures (higher fractions of solid). Hence, the effect of the extent of the transition region is a function of the direction of thermo-solutal low in the slurry zone. On the other hand, the “loose” slurry zone creates more opportunities for the loating grains to be generated. These grains can be brought to the central part of the billet and sediment there, contributing to the negative segregation. In this process, iner and thus lighter grains in a grain-reined billet may be less prone to settling and can be carried out more easily by the convective low. This means less negative contribution from loating grains in the grain-reined billet. It is important that the transport of solid be assisted by the convection. Floating grains generally follow the direction of liquid low. At the same time, it is the relative movement of the solid and liquid (settling of the solid phase under gravity, iltration of the solid through the slurry) that actually affects the macrosegregation. At the end, the ratio between positive contribution of thermo-solutal convection and negative contribution of loating grains can be such that (1) each process compensates for the other and the negative centerline segregation results from the shrinkage-induced low; (2) thermo-solutal convection prevails and the segregation pattern depends on

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

169

the ratio between convective and shrinkage contributions; and (3) loating grains intensely accumulate in the center, enhancing the negative segregation caused by solidiication shrinkage. Positive centerline segregation can be achieved if the contribution of thermosolutal convection dominates over the other mechanisms of macrosegregation. This is possible when the sump is shallow and the transition region is narrow, hence at low casting speeds and small billet diameters. Unfortunately, the actual casting speeds that may result in positive or negligible macrosegregation are much lower than those that are currently used in industry. 4.2.4

Effect of Forced Convection on Macrosegregation during DC Casting

Normal

Intensity of forced convection Inverse

Extent of macrosegregation

Anoshkin and Dobatkin [31, 66] suggest that the transition from negative (inverse) to positive (normal) segregation is possible during DC casting, especially in the presence of forced convection (mechanical or electromagnetic stirring). They link this transition to the ratio between the convection in the slurry part of the billet that washes solute-rich melt from the mushy zone and brings it to the center and the shrinkage-induced low that takes the solute-rich liquid inside the mushy zone from the center toward the surface, and also to the solidiication shrinkage of the alloy. This concept is illustrated in Figure 4.25. The higher the intensity of forced convection in the sump and the smaller the solidiication shrinkage, the greater the possibility of normal (positive) macrosegregation.

FIGURE 4.25 Effect of forced convection on the extent and type of macrosegregation upon DC casting; arrows show the tendency in the case of decreasing solidiication shrinkage [31].

Copyright 2008 by Taylor and Francis Group, LLC

170

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Livanov et al. [30] noted that the threshold value of VcastD (Equation 4.21) is larger for billets cast using an electromagnetic mold, when intensive melt lows are created by the electromagnetic ield. Figure 4.26 shows the experimental results on macrosegregation in billets of a 2024-type alloy cast by electromagnetic casting (EMC) (Figure 4.26a) and with mechanical stirring (Figure 4.26b). In both cases a suficient degree of forced convection results in the normal (positive) segregation. This fact indicates the potential of the liquid convection in the formation of macrosegregation patterns. Similar results were obtained upon EMC of a 7075 alloy [30, 67].

4.3

Effect of Composition on Macrosegregation: Macrosegregation in Commercial Aluminum Alloys

The fundamental reason for macrosegregation lies in the partitioning of solute elements between liquid and solid phases during solidiication (see Section 2.2). It is, therefore, not surprising that the extent and pattern of macrosegregation depend on the partition coeficient K, i.e., on its deviation from unity and on whether it is less than or greater than 1. This coeficient is deined as the ratio of the solid composition to the liquid composition. Partition coeficients for some alloying elements in aluminum are listed in Table 4.9. There is no partitioning and, therefore, segregation if K = 1. Most of the alloying elements and impurities (e.g., Cu, Mg, Zn, Li, Mn, Si, Fe) are present in aluminum alloys at hypoeutectic concentrations with K < 1. In this case, the lower the coeficient, the greater the macrosegregation. The overall macrosegregation pattern of such elements is similar to those already discussed in this chapter (see, for example, Figures 4.13, 4.16). Certain elements (Ti, Zr, Cr) that have a peritectic reaction with aluminum (with K > 1) exhibit a segregating tendency that is exactly opposite, i.e., with a positive centerline segregation that increases with the size of the partition coeficient. Figure 4.27 illustrates these phenomena by experimentally measured macrosegregation proiles. Iron has a much more pronounced segregation compared to manganese, and titanium has an opposite segregation pattern. Figure 4.28 graphically illustrates the increasing tendency for segregation with decreasing partition coeficient of the alloying elements with K < 1. The extent of segregation of a particular alloying element depends more on the base alloy itself rather than on the absolute concentration of the alloying element [68]. Most commercial aluminum alloys are grain reined during casting in order to make a ine-grained structure favorable for downstream processing and to improve the alloy castability, including hot-tearing resistance (see Sections 5.3.1 and 5.4.2 of Chapter 5). The effect of grain reining on macrosegregation remains, however, unclear; relatively few reports with controversial

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

171

5.1 EMC DC casting 4.9

Cu,%

4.7

4.5

4.3

4.1

3.9 0

50

100

200

150

250

Billet cross-section, mm (a)

4.8

Strong stirring Moderate stirring Weak stirring

4.6

Cu,%

4.4

4.2

4

3.8 0

40

80

120

160

200

240

280

Billet cross-section, mm (b) FIGURE 4.26 Effect of electromagnetic (a) and mechanical (b) stirring on the macrosegregation of copper in 2024-alloy billets [30, 31, 53].

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

172

TABLE 4.9 Partition Coeficients for Some Alloying Elements in Aluminum Element K

Fe

Si

Cu

Mg

Zn

Mn

Ti

Cr

0.03

0.13

0.17

0.43

0.45

0.82

9.0

2.0

0.06

Relative segregation

0.04 0.02 0 −0.02 −0.04

Mn Fe Ti

−0.06

Center

−0.08 0

50

100

150

Billet cross-section, mm FIGURE 4.27 Macrosegregation proiles of Mn (K = 0.8), Fe (K = 0.03) and Ti (K = 9) in a 200-mm billet from a 2024 alloy cast at 120 mm/min [42]. Data for Ti are divided by 10.

results are available on this subject [34, 39, 41, 42, 44, 45, 68–70]. Grain reining modiies the resistance to low through the solid network (mushy zone), affecting the permeability (see Section 4.1.2). According to Equations 4.5–4.8, grain reinement should decrease the permeability of the mushy zone constructed out of iner grains, though the inal result would depend on the morphology and packing of the solid phase [24]. At the same time, grain reinement lowers the coherency [71] and rigidity temperatures (see Section 5.1 of Chapter 5). For example, the solid fraction at coherency in an Al–4% Cu alloy increases from 0.23 to 0.45 upon addition of 0.05% Ti [71], and the solid fraction at rigidity in a 2024 alloy increases with grain reinement from 0.75 to 0.82 [72] (see Figure 5.9 in Chapter 5). The lower permeability will hinder the shrinkageinduced low, narrow the rigid mushy zone, and prevent the penetration of convective lows into the rigid mushy zone. On the other hand, the lower coherency temperature means a wider slurry zone with more possibilities of convective lows and transport of loating. In addition, the smaller size of these loating crystals impede their settling ability. It is not surprising that the consequences of grain reinement can be quite different, depending on the ratio of the contributions of all these effects.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

173

35 AA 6009 AA 3104

30 Deviation of concentration, %

AA 5182 25 20 15 10 5

Fe

0 0.0

Si Cu

0.2

Mn

Mg

0.4

0.6

0.8

1.0

Partition coefficient, K FIGURE 4.28 Centerline segregation of alloying elements vs. partition coeficient [68]. The deviation of concentration is taken as the sum of the absolute values of the lowest (negative) deviation and the average of the high (positive) deviation points, and characterizes the overall amplitude of macrosegregation.

The reference data given in Tables 4.6 and 4.7 and in Figure 4.21 show that grain-reined alloys are more prone to positive centerline segregation. Hence, grain reinement favors normal segregation. The published accounts are, however, very diverse. Gariepy and Caron [68] examined the effect of grain-reining practices on centerline macrosegregation in sheet ingots of commercial 3104 (Al–Mn–Mg), 5182 (Al–Mg–Mn), and 6009 (Al–Si–Mg) alloys. They concluded that the extent of centerline segregation increases with the increasing Ti content, as demonstrated in Figure 4.29. Segregation tendency is also dependent on the type of grain reiner. For the same feeding conditions, the application of a more potent Al5Ti0.2B master alloy leads to more severe segregation as compared to a 5182 alloy reined with an Al–6% Ti master alloy or not grain reined at all [44, 68]. The detrimental effect of grain reining on macrosegregation is usually ascribed to the nucleation of a larger number of free dendrites in the wider slurry region and consequent sweeping of those to the bottom and center of the sump. We already know that the loating-grain phenomenon cannot always explain the observed macrosegregation pattern (see, e.g., Table 4.5). It is suggested that the tendency for “grain-reining-aided” macrosegregation would depend on the alloy composition, following the “grain reinability” of the various alloy systems (see Section 2.4).

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

174 20

Deviation of Mg concentration, %

18 16 14 12 10 8 COMBO bag

6

Channel bag

AA 3104

Headscreen

4 0.000

0.001

0.002

0.003

0.004

0.005

0.006

Ti, % FIGURE 4.29 Effect of grain reining and melt feeding system on centerline segregation of Mg in commercial 3104 ingots [68]. The deviation of concentration is taken as the sum of the absolute values of the lowest (negative) deviation and the average of the high (positive) deviation points, and characterizes the overall amplitude of macrosegregation.

Joly et al. [45, 69] also reported that grain reining caused more severe centerline segregation in a DC-cast Al–Mg sheet ingot. Their results agree well with the results of Glenn et al. [44] obtained independently on a similar ingot, as shown in Figure 4.30. Remarkably, duplex microstructures (mixture of coarse and ine grains) were observed only in the nongrain-reined ingot, where the segregation is less severe, but not in the grain-reined ingot. Consequently, the negative centerline segregation is not speciically associated with the presence of duplex microstructures (or “loating” grains). It was pointed out that grain morphology was an important factor to be considered [45], which was more dendritic in the nongrain-reined ingot and more globular after grain reining. Combined effects of the changed permeability and movement of equiaxed grains were considered responsible for the experimentally observed segregation patterns shown in Figure 4.30. Contrary to the above observations, there is a report that shows that grain reining induces positive centerline segregation upon casting 530-mm billets of an Al–4.5% Cu alloy [41]. A bi-level feeding system with a central spout and a loating diffuser was used. The nongrain-reined counterpart exhibited the usual pattern of negative centerline segregation. The most interesting observation was that the grain-reined billet exhibited positive centerline segregation despite the presence of loating grains in the central portion of the billet, whereas the negative centerline segregation in nongrain-reined

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

175

Deviation of Mg concentration from average, %

15 10 5 0 −5 −10

Along thickness Centerline

Along width Centerline

−15 0

100

200

300

400

500

600

Distance, mm FIGURE 4.30 Macrosegregation in commercial ingots from a 5182 alloy. Data along the thickness (triangles) are given for a 1850 × 550 mm2 ingot cast at 60 mm/min [69] and data along the width (circles) are given for a 1050 × 550 mm2 ingot [44]. Dashed lines represent nongrain-reined part of the ingot and solid lines show the grain-reined part of the ingot (0.015% Ti in [44] and 0.003 Ti in [69]).

billet occurred without any loating grains noticed. It was suggested, in line with our previous discussion, that the different ratio between the shrinkage-driven low and the advection of the solute-rich liquid caused the observed phenomena. We have observed positive segregation in a grainreined Al–Cu billet cast a low casting speed in our level-pour DC casting system, which turns to negative centerline segregation at a higher casting speed (Figure 4.22). Our studies on various commercial Al alloys showed that addition of an Al3Ti1B grain reiner did not affect the macrosegregation in 200-mm billets cast at 80 to 120 mm/min [39, 42, 70], as illustrated in Figure 4.31. At the same time, we observed duplex structures with depleted loating grains (see Figure 4.14 and Table 4.4). The number of these loating, coarse-cell grains signiicantly increases with grain reinement, from 35 to 65–70 vol.% [70]. The data on the composition of coarse-cell and ine-cell grains may help us understand the phenomena involved in the formation of the inal macrosegregation pattern in the central portion of the billet. The integration over the microsegregation proiles like those shown in Figure 4.14 allows us to estimate the composition of different grains. Let us look at the results in Table 4.10. The bulk composition of coarse-cell grain is conirmed to be less than the nominal alloy composition. Somewhat surprisingly the composition of the ine-cell grain areas appears to be higher than nominal, especially in the grain-reined alloy. This contradicts the usual view that the ine-cell grain structure relects the structure of nominal composition (without macrosegregation) [34]. The real situation is far from that. It seems that the bulk

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

176 0.06 0.04

Center

0.02

∆C

0 −0.02 −0.04 Cu Mg

−0.06

192

−0.08 0

20

40

60

80

100

120

140

160

180

200

Distance, mm (a) 0.06 0.04 Center

0.02

∆C

0 −0.02 −0.04 Cu Mg

−0.06

192

−0.08 0

20

40

60

80

100

120

140

160

180

200

Distance, mm

(b) FIGURE 4.31 Measured relative macrosegregation in 200-mm billets cast at 80 mm/min from a nongrainreined (a) and grain-reined (b) 2024 alloy [70].

composition of ine-cell regions gives us an opportunity to estimate the contribution of different mechanisms to macrosegregation. Using the data in Table 4.10 we can separate the macrosegregation caused by loating grains and by other mechanisms. The results of this analysis are shown in Table 4.11. Indisputably, the coarse-cell, loating grains are able to produce strong negative segregation, especially in the grain-reined alloy. In fact, the structure consisting entirely of loating grains would give centerline segregation about one order of magnitude larger than the observed macrosegregation.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

177

TABLE 4.10 Measured Using EPMA Average Compositions of Grains in the Central Part of Nongrain-Reined and Grain-Reined 2024-Alloy Billets Cast at 80 and 120 mm/min Nongrain Reined, 80 mm/min* Structure Type

Line Scan

Area Scan**

Cu, % Mg, % Cu, %

Coarse-cell grains Fine-cell grains

Grain Reined, 80 mm/min* Line Scan

Area Scan**

Grain Reined, 120 mm/min* Line Scan

Mg, % Cu, % Mg, % Cu, % Mg, % Cu, % Mg, %

2.55

0.96

2.92

1.08

2.19

0.94

3.32

1.13

1.58

0.08

3.9

1.48

3.32

1.13

4.01

1.56

4.38

1.43

3.49

1.51

* Nominal alloys composition: 3.5% Cu, 1.4% Mg. Grain-reined alloys additionally contain 0.006–0.008% Ti. ** Results on area scan are shown for comparison. Although the areas were chosen so as to relect coarse-cell or ine-cell grains, portions of mixed structures were inevitably measured.

TABLE 4.11 Estimated Relative Centerline Segregation in Nongrain-Reined and Grain-Reined 2024-Alloy Billets Cast at a Speed of 80 mm/min; Data from Line-Scan Analysis in Table 4.10 Are Used Nongrain Reined

Grain Reined

Assumption

∆Cu

∆Mg

∆Cu

∆Mg

100% ine-cell structure 100% coarse-cell structure

+0.11 –0.27

+0.05 –0.32

+0.15 –0.37

+0.11 –0.33

In contrast, the structure consisting only of ine-cell grains would give positive (in the case of a grain-reined billet strongly positive) centerline segregation. We know that in the absence of loating grains the main acting mechanisms of macrosegregation are thermo-solutal convection and shrinkage-induced low. Taking into account the shallow proile of the sump at the given, low casting speed, we can conclude that the enrichment of ine-cell grain structure is mostly due to the solute brought to this structure by convective lows. (As shown in Figure 4.18e, the contribution of shrinkage-induced low to the relative segregation of copper should be less than |−0.02|.) Therefore, the effect of coarse-cell grains on the overall macrosegregation is always negative and somewhat larger in the grain-reined structure. The impact of the enriched ine-cell grains and, hence, of thermo-solutal convection is much more pronounced in the grain-reined structure, where it can produce positive centerline macrosegregation. Obviously, the measured values given in Table 4.10 are rather local and, therefore, the speciic numbers may change. But the overall trend in the variation in macrosegregation induced by different grain structures is clear.

Copyright 2008 by Taylor and Francis Group, LLC

178

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Based on our analysis of the mechanisms of macrosegregation and on the experimental evidence presented in Tables 4.4, 4.10, and 4.11 and in the reference literature, we can suggest the following line of logic for explaining some seemingly controversial experimental results. A stronger penetration of the convective lows into the slurry zone of the grain-reined billet in combination with the hindered shrinkage-induced low in the dense mushy zone facilitates the normal segregation (positive segregation in ine-cell structure in Table 4.11). On the other hand, the strongly depleted loating grains appear on the massive scale in the grain-reined slurry zone and may constitute the bulk of the macrostructure (negative segregation in coarse-cell structure in Table 4.11). These processes compete with one another, depending on the scale of the casting, the casting speed, and the corresponding sump proile. The resulting macrosegregation may then seem unaffected by the grain reinement, simply because the convective-induced enrichment and loating-grain-induced depletion compensate for each other (Figure 4.31b). However, if the grain reinement effectively hampers the shrinkage-induced low, while opening the possibilities for more active convective lows in the slurry, then the result can be positive centerline segregation, as observed by Finn et al. [41] and by us, as shown in Figure 4.22. Otherwise, grain reinement can result in more pronounced negative macrosegregation, as has been reported elsewhere [44, 68, 69]. In the nongrain-reined billet (with a columnar grain structure [41] or coarser equiaxed structure in our experiments) the slurry zone turns into mush at a higher temperature, preventing eficient convection, while the shrinkage low is rather strong (less positive segregation in ine-cell structure in Table 4.11). The amount of loating grains is lower than in the grainreined billet, though they still contribute to the negative segregation (see the coarse-cell structure in Table 4.11). The experimentally observed macrosegregation will depend on the ratio between the convective and shrinkage lows as well as on the amount of coarse-cell grains. All the processes can damp one another, producing negligible segregation, as shown in Figure 4.31a. In a billet with a deeper sump because of larger size [41] or higher casting speed (Figure 4.22) the shrinkage-induced low may prevail, shifting the pattern toward the negative centerline segregation. This is true for both grain-reined and nongrain-reined billets. The results shown in Table 4.10 (compare entries for grain-reined structures at 80 and 120 mm/min) conirm that the ine-cell regions become less enriched in the solutes with increasing casting speed. This indicates that the shrinkage-induced low is more active in taking away the enriched liquid that was brought to the center by thermosolutal convection. The amount of loating grains does not change in grainreined billets cast at different casting speeds. So their contribution to the negative centerline segregation can be associated only with their somewhat larger depletion of solute at a higher casting speed, as given in Table 4.10. It is worth noting that the casting parameters (speed and size) in the experiments of Finn et al. [41] were close to the threshold condition for the transition from positive to negative centerline segregation described by Figure 4.21.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

179

The choice of a feeding system is also important, as illustrated in Figure 4.29. Melt distribution through a channel- or combo-bag induces much more variation in the centerline segregation with grain reining than when a headscreen is used [68]. At the same time, the extent of macrosegregation is considerably less in the case of the combo-bag than in other cases studied [68]. The main difference that the feeding system produces is the melt low pattern in the liquid and slurry parts of the sump. The headscreen directs the low along the sides of the ingot and downward, enhancing the natural convection low. The combo-bag and channel-bag are designed for the change in the natural low. In this case the channels or openings direct the low either opposite to the natural convection or at some angle toward it (see also the effects of the screw pump in Figure 4.8 and EMC in Figure 4.26a). When the low is directed downward in the central portion of the sump, then the macrosegregation can be minimized [68]. Similar experimental data were obtained for a 5182 alloy [68]. The experimental, analytical, and numerical data given in this chapter demonstrate unambiguously that the macrosegregation upon DC casting of aluminum alloys is a very complex phenomenon that cannot be explained by a single mechanism. The seemingly controversial and diverse results reported in the literature are the consequences of the intricate interaction between different mechanisms of macrosegregation that can eventually produce any result, from positive to negative. The analysis of macrosegregation should be always performed taking into account all the possible mechanisms that can reveal themselves to varying extents, depending on the experimental conditions. At the present level of our understanding, we can conclude that we have the means (including experimental, analytical and numerical tools) to deal with the macrosegregation mechanisms known to us, individually or in some combination. The development of the model that would include all of the mechanisms in a coherent manner is the challenge of the immediate future. In the past, the revelation of the true nature of a macrosegregation pattern observed under particular experimental conditions has been based largely on guesswork, but more recently, educated guesswork.

References 1. V. Beringuccio. Pirotechnia, 1540. Translated by C.S. Smith, M.T. Gnudi, Cambridge, MA: MIT Press, 1966, pp. 259–260. 2. L. Ercker. Beschreibung allerfürnemisten mineralischen Ertzt und Bergwercksarten, Prague, 1574. Translated by J. Pittus as Fleta Minor. The Laws of Art and Nature, in Knowing, Judging, Assaying, Fining, Reining and Enlarging the Bodies of Conined Metals. Pt. 1, London: T. Dawks, 1683. 3. W.T. Pell-Walpole. Met. Treatment Drop Forging, 1949, Summer, pp. 103–115; Autumn, pp. 171–182.

Copyright 2008 by Taylor and Francis Group, LLC

180

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

4. B.B. Gulyaev. Zatverdevanie I Neodnorodnost Stali (Solidiication and Inhomogeneity of Steel), Moscow: Metallurgizdat, 1950. 5. S.M. Voronov. Z. Metallkde., 1929, vol. 21, p. 310 (cited from [3]). 6. P. Brenner, W. Roth. Z. Metallkde., 1940, vol. 32, pp. 10–14. 7. V.I. Dobatkin, Nepreryvnoe Lit’ye i Liteinye Svoistva Splavov (Direct Chill Casting and Casting Properties of Alloys), Moscow: Oborongiz, 1948. 8. J. Verö. Über den Mechanismus der Blockseigerung, Mitt. Fakult. Berging. Geo-Ing., Sopron, 1936, vol. 8, pp. 194–200. 9. D. Bernard, Ø. Nielsen, L. Salvo, P. Cloetens. Mater. Sci. Eng. A, 2005, vol. 392A, pp. 112–120. 10. M. K. Hubbert. The Theory of Ground-Water Motion and Related Papers, New York: Hefner Publishing, 1969. 11. M.C. Flemings. Solidiication Processing, New York: McGraw-Hill, 1974. 12. Ø. Nielsen, L. Arnberg. Metall. Mater. Trans. A, 2000, vol. 31A, pp. 3149–3153. 13. T.S. Piwonka, M.C. Flemings. Trans. Metall. Soc. AIME, 1966, vol. 236, pp. 1157–1165. 14. D.R. Poirier, S. Ganesan. Mater. Sci. Eng. A, 1992, vol. A157, pp. 113–123. 15. Ø. Nielsen, S.O. Olsen. In: M. Tiryakioglu, P.N. Crepeau (Eds.), Shape Casting: The John Campbell Symposium as Held during TMS Annual Meeting, February 2005, San Francisco, CA, USA, Warrendale, PA: TMS, 2005, pp. 103–112. 16. P.C. Carman. Trans. Inst. Chem. Eng. London, 1937, vol. 15, pp. 150–156. 17. J. Kozeny. Sitzungsberg. Akad. Wiss. Wien, 1927, vol. 136, pp. 271–306. 18. J.F. Grandield, C.J. Davidson, J.A. Taylor. In: K. Ehrke, W. Schneider (Eds.), Continuous Casting, Weinheim: Wiley–VCH, 2000, pp. 205–210. 19. S. Ganesan, C.L. Chan, D.R. Poirier. Mater. Sci. Eng. A, 1992, vol. A151, pp. 97–105. 20. J.F. Grandield, C.J. Davidson, J.A. Taylor. In: J. Anjier (Ed.), Light Metals 2001, Warrendale, PA: TMS, 2001, pp. 895–901. 21. C.Y. Wang, C. Beckermann. Metall. Trans. A, 1993, vol. 24A, pp. 2787–2802. 22. C.Y. Wang, S. Ahuja, C. Beckermann, H.C. de Groh III. Metall. Mater. Trans. B, 1995, vol. 26B, pp. 111–119. 23. C.J. Vreeman, F.P. Incropera. Int. J. Heat Mass Transfer, 2000, vol. 43, pp. 687–704. 24. J.W.K. van Boggelen, D.G. Eskin, L. Katgerman. In: P. Crepeau (Ed.), Light Metals 2003, Warrendale, PA: TMS, 2003, pp. 759–766. 25. Ø. Nielsen, L. Arnberg, A. Mo, H. Thevik. Metall. Mater. Trans. A, 1999, vol. 30A, pp. 2455–2462. 26. A.E. Vol. Handbook of Binary Metallic Systems. Structure and Properties, Vol. 1, Moscow: Gos. Izd. Fiz.-Mat. Lit., 1959 (English translation by Israel Program for Scientiic Translations, Jerusalem, 1966). 27. Q. Du, D.G. Eskin, L. Katgerman. In: Ch.-A. Gandin, M. Bellet (Eds.), Proc. XI Intern. Conf. Modelling Casting, Welding and Advanced Solidiication Processes (MCWASP XI). June 2006, Opio, France; Warrendale, PA: TMS, 2006, pp. 235–242. 28. V.M. Tageev. Doklady Akad. Nauk SSSR, 1949, vol. 67, pp. 491–494. 29. Q. Du, D.G. Eskin, L. Katgerman. Metall. Mater. Trans. A, 2007, vol. 38A, pp. 180–186. 30. V.A. Livanov, R.M. Gabidullin, V.S Shipilov. Continuous Casting of Aluminum Alloys, Moscow: Metallurgiya, 1977. 31. N.F. Anoshkin. Zone Chemical Inhomogeneity in Ingots, Moscow: Metallurgiya, 1976. 32. M.C. Flemings. ISIJ Int., 2000, vol. 40, pp. 833–841. 33. D.G. Eskin, Q. Du, L. Katgerman. Scr. Mater., 2006, vol. 55, pp. 715–718.

Copyright 2008 by Taylor and Francis Group, LLC

Macrosegregation

181

34. H. Yu, D.A. Granger. In: E.A. Starke, Jr., T.H. Sanders, Jr. (Eds.), Int. Conf. Aluminum Alloys—Their Physical and Mechanical Properties, Charlottesville, VA; Warley (U.K.): EMAS.; 1986, pp. 17–29. 35. M.G. Chu, J.E. Jacoby. In: C.M. Bickert (Ed.), Light Metals 1990, Warrendale, PA: TMS, 1990, pp. 925–930. 36. A. Håkonsen, D. Mortensen, S. Benum, H.E. Vatne. In: C.E. Eckert (Ed.), Light Metals 1999, Warrendale, PA: TMS, 1999. pp. 821–827. 37. Suyitno, D.G. Eskin, V.I. Savran, L. Katgerman. Metall. Mater. Trans. A, 2004, vol. 35A, pp. 3551–3561. 38. D.G. Eskin, V.I. Savran, L. Katgerman. Metall. Mater. Trans. A, 2005, vol. 36A, pp. 1965–1976. 39. R.K. Nadella, D. Eskin, L. Katgerman. In: H.R. Muller (Ed.), Continuous Casting of Non-ferrous Metals, Neu-Ulm, Germany, Weinheim: Wiley-VCH, 2005, pp. 277–282. 40. R. Nadella, D.G. Eskin, L. Katgerman. Unpublished work, 2007. 41. T.L. Finn, M.G. Chu, W.D. Bennon. In: C. Beckermann (Ed.), Micro/Macro Scale Phenomena in Solidiication, New York: ASME, 1992, pp. 17–26. 42. R. Nadella, D. Eskin, L. Katgerman. In: M. Sørlie (Ed.), Light Metals 2007, Warrendale, PA: TMS, 2007, pp. 727–732. 43. R.C. Dorward, D.J. Beerntsen. In: C.M. Bickert (Ed.), Light Metals 1990, Warrendale, PA: TMS, 1990, pp. 919–924. 44. A.M. Glenn, S.P. Russo, P.J.K. Paterson. Metall. Mater. Trans. A, 2003, vol. 34A, pp. 1513–1523. 45. G. Lesoult, V. Albert, B. Appolaire, H. Combeau, D. Daloz, A. Joly, C. Stomp, G.U. Grün, P. Jarry. Sci. Techn. Adv. Mater., 2001, vol. 2, pp. 285–291. 46. R. Nadella, D.G. Eskin, Q. Du, L. Katgerman. Progr. Mater. Sci., 2008 (in press). 47. A.N. Turchin, D.G. Eskin, L. Katgerman. Metall. Mater. Trans. A, 2007, vol. 38A, pp. 1317–1329. 48. D.G. Eskin, Q. Du, R. Nadella, A.N. Turchin, L. Katgerman. In: H. Jones (Ed.), Proc. 5th Decennial Intern. Conf. Solidiication Processing ’07, University of Shefield, U.K., Padstow: TJ International Ltd, 2007, pp. 437–441. 49. G. Lesoult. Mater. Sci. Eng. A, 2005, vols. 413–414, pp. 19–29. 50. L.C. Nicolli, A. Mo, M. M’Hamdi. Metall. Mater. Trans. A, 2005, vol. 36A, pp. 433–442. 51. L.C. Nicolli, C.L. Martin, A. Mo, O. Ludvig. Mater. Sci. Forum, 2006, vol. 508, pp. 187–192. 52. H.J. Thevik, A. Mo, T. Rusten. Metall. Mater. Trans. B, 1999, vol. 30B, pp. 135–142. 53. V.I. Dobatkin. Ingots of Aluminum Alloys, Sverdlovsk: Metallurgizdat, 1960. 54. D.G. Eskin, J. Zuidema, V.I. Savran, L. Katgerman. Mater. Sci. Eng. A, 2004, vol. 384, pp. 232–244. 55. Q. Du, D.G. Eskin, L. Katgerman. Mater. Sci. Eng. A, 2005, vols. 413–414, pp. 144–150. 56. E.D. Tarapore. In: P.G. Campbell (Ed.), Light Metals 1989, Warrendale, PA: TMS, 1989, pp. 875–880. 57. J.M. Reese. Metall. Mater. Trans. B, 1997, vol. 28B, pp. 491–499. 58. A.V. Reddy, C. Beckermann. Metall. Mater. Trans. B, 1997, vol. 28B, pp. 479–489. 59. M. Založnik, B. Šarler. Mater. Sci. Eng. A, 2005, vol. 413–414, pp. 85–91. 60. M. Založnik, B. Šarler, D. Gobin. Materiali in Technologije, 2004, vol. 38, pp. 249–255.

Copyright 2008 by Taylor and Francis Group, LLC

182

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

61. C.J. Vreeman, J.D. Scholz, M.J.M. Krane. J. Heat Transfer, 2002, vol. 124, pp. 947–953. 62. B.C.H. Venneker, L. Katgerman. In: J.L. Anjier (Ed.), Light Metals 2001, Warrendale, PA: TMS, 2001, pp. 823–829. 63. L. Katgerman, B.C.H. Venneker, J. Zuidema. In: P. Crepeau (Ed.), Light Metals 2003, Warrendale, PA: TMS, 2003, pp. 815–820. 64. D.G. Eskin, Q. Du, L. Katgerman. Metall. Mater. Trans. A, 2008 (in press). 65. S.C. Flood, P.A. Davidson. Mater. Sci. Technol., 1994, vol. 10, pp. 741–751. 66. N.F. Anoshkin, V.I. Dobatkin. Mater. Sci. Eng. A, 1999, vol. 263, pp. 224–229. 67. B. Zhang, J. Cui, G. Lu. Mater. Sci. Eng. A, 2003, vol. 355, pp. 325–330. 68. B. Gariepy, Y. Caron. In: E.L. Rooy (Ed.), Light Metals 1991, Warrendale, PA: TMS, 1991, pp. 961–971. 69. A. Joly, G.-U. Grün, D. Daloz, H. Combeau, G. Lesoult. Mater. Sci. Forum, 2000, vols. 329–330, pp. 111–120. 70. R. Nadella, D. Eskin, L. Katgerman. Mater. Sci. Forum, 2006, vols. 519–521, pp. 1841–1846. 71. L. Arnberg, L. Bäckerud, G. Chai. Solidiication Characteristics of Aluminum Alloys. Vol. 3: Dendrite Coherency, Des Plaines, IL: AFS, 1996. 72. A. Stangeland, A. Mo, D. Eskin. Metall. Mater. Trans. A, 2006, vol. 37A, pp. 2219–2229.

Copyright 2008 by Taylor and Francis Group, LLC

5 Hot Tearing Casting practice is only too much familiar with various defects occurring in the inal product. One of the most severe and unrecoverable defects is hot tearing, also known as hot cracking or hot shortness. Hot tearing is the formation of an irreversible crack (failure) in the still semisolid casting. The solidiication of alloys always occurs in some temperature range, i.e., the solidiication or freezing range. Unlike pure metals, which solidify at one temperature, alloys transform gradually from liquid to solid over a (wide) temperature interval. During casting there is a considerable time when the alloy consists of both solid and liquid. The semisolid state can be divided into two classes: slurry and mush. Slurry is deined as a liquid with suspended solid particles. Upon decreasing temperature, the solid phase nucleates, grows in a form of grain (usually dendritic in shape), and, starting from a certain point in the solidiication range, these solid grains begin to interact with one another, irst by sensing the presence of neighbors, then by coming into contact and bridging with them, and inally by forming a continuous skeleton of the solid phase. At a certain temperature, when solid grains start to interact with one another, the material irst changes its viscous behavior and later develops a certain strength [1]. The temperature at which the grains start to sense one another is called the coherency point, and the temperature at which the continuous solid network is formed is called the rigidity point. Below the coherency temperature, the material is called a mush. The solid fraction at which this transition occurs varies between 0.25 and 0.6, depending on the morphology of the solid grains [1]. In DC casting, the name “mushy zone” is often applied to the entire transition region between liquidus and solidus (see Figure 3.1a), which is misleading. The upper part of the transition region is actually a slurry because solid grains are suspended and can freely move in the liquid. A real mush is formed only after the temperature has dropped below the coherency temperature. Below the rigidity point, the semi-solid body acquires the main characteristics of the solid phase—retention of the shape and mechanical properties, such as strength and ductility [3]. Because of the strongly different mechanical behaviors of these different semi-solid states, slurries are usually described by viscosity-based models, and mush is usually described by deformation-based models [2]. The viscosity-based models start from the liquid side and are modiied to take into account the effect of the increasing amount of solid particles. The deformation-based models are based on models for hot working, which are modiied to take into account the presence of liquid. The transition from a slurry to a mush remains a complicated problem to model, and a satisfactory model, 183 Copyright 2008 by Taylor and Francis Group, LLC

184

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

which describes the behavior for the complete solidiication range, is yet to be developed. The solidiication process can be divided into four stages, based on the permeability of the solid network [4–7] (see also Section 4.1.2): 1. Mass feeding, in which both liquid and solid are free to move. 2. Interdendritic feeding, in which, after the dendrites have formed a coherent skeleton, the remaining liquid has to low through the dendritic network. A pressure gradient (pressure drop) may develop across the mushy zone from solidiication shrinkage occurring deeper in the mushy zone. However, at this stage the permeability of the network is still large enough to prevent pore formation. 3. Interdendritic separation, in which the liquid network becomes fragmented and pore formation or hot tearing may occur. With increasing solid fraction, liquid is isolated in pockets or immobilized by surface tension. When the permeability of the solid network becomes too small for the liquid to low, further thermal contraction of the solid will cause pore formation or hot tearing. 4. Interdendritic bridging or solid feeding, in which the ingot has developed a considerable strength and solid-state creep compensates for further contraction. At the inal stage of solidiication ( fs > 0.9), only isolated liquid pockets remain and the ingot has considerable strength. Only solid-state creep can now compensate for solidiication shrinkage and thermal stresses. Recently, the mushy zone was modeled with an emphasis on its coherency and feeding characteristics [8]. Its has been shown for an Al–1% Cu alloy that clusters of grains start to form at solid fractions about 0.97. This process involves more and more grains, and at about fs = 0.99 large grain clusters are formed with a few isolated liquid ilms inside. Between solid fractions of 0.99 and 1.0 the solid network is continuous and the liquid phase survives only in separated pockets. This evolution of the mush structure causes the localization of feeding. Starting from the fraction of solid above 0.97 the melt low begins to “prefer” some paths along grain boundaries, while other channels begin to “dry out” despite the fact that they still exist. As a result, the permeability of the mushy zone decreases at higher solid fraction more rapidly than predicted by the Kozeny–Carman relationship (see Equations 4.4–4.11 and Figure 4.3). Such a feeding localization along with solidiication shrinkage induces a pressure difference between metallostatic pressure on the hotter part of the mush and a lower pressure in forming cavities on the cooler part of the mush. This local pressure drop and insuficient feeding ability may result in the formation of cavities and, eventually, cracks. From many studies [9–13] starting already in the 1950s, and reviewed by Novikov [3], Sigworth [14], and Eskin et al. [15], it appears that hot tears

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

185

initiate above the solidus temperature and propagate in the interdendritic liquid ilm. The fracture surface is usually a smooth layer and sometimes shows solid bridges that connect or have connected both sides of the crack [7, 13, 16–21]. During solidiication, the liquid low through the mushy zone decreases until it becomes insuficient to ill initiated cavities so that they can grow further. In this chapter we mainly consider the last two stages of solidiication because in the slurry and upper mush regions the feeding is usually suficient to avoid any casting defects. It is mainly at the “interdendritic separation” stage when the casting is vulnerable to pore formation and hot tearing. Industrial and fundamental studies show that hot tearing occurs in the late stages of solidiication when the volume fraction of solid is above 85–95% and the solid phase is organized in a continuous network of grains. However, a hot crack would not appear if there were not some precondition for its formation. One of the main preconditions for hot tearing is the existence of tensile stresses in certain sections of the casting. During DC casting of aluminum alloys, the primary and secondary cooling causes strong thermal gradients in the ingot and a corresponding uneven thermal contraction of different sections of the casting, which may lead to distortion of the ingot shape (e.g., butt curl, butt swell, rolling face pull-in) and/or hot tearing and cold cracking. Therefore, we may conclude that thermal contraction of the solid phase during solidiication is one of the major physical phenomena involved in the formation of hot cracks.

5.1

Thermal Contraction during Solidification

Hot tearing or hot cracking occurs in the lower part of the solidiication range, close to the solidus, when the solid fraction is more than 0.9 [7, 15]. At this point, the mushy zone is deinitely coherent and consists of interconnected solid grains [1], but the liquid ilm still exists between most of the grains. The term coherency (or coherency temperature) should be used with caution. We have also mentioned in the introductory part to this chapter that the mushy zone is formed at temperatures below the coherency point but the real interconnection of solid grain occurs at much lower temperatures. If the coherency is understood as a temperature at which a continuous dendritic network is formed, and the material starts to develop strength and retain its shape, then this point can be better deined as a rigidity point [1, 3]. At temperatures above the rigidity point, the grains are relatively free to re-arrange themselves with respect to one another and hence do not transfer any forces. Moreover, before the rigidity temperature is reached upon solidiication the liquid phase can still low between grains and, therefore, melt feeding occurs without much dificulty. This is the terminology that we adopt in this chapter. The region where an alloy is most susceptible to hot tearing was dubbed in the 1940s–1950s “the effective solidiication range” [3, 22] or “the vulnerable

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

186

Liquidus

0.00

0.25%

Volume shrinkage, %

−1.00 −2.00 −3.00

Solidus

−4.00 −5.00

3.93% Eutectic 5.3%

−6.00 −7.00 500

550

600

650

700

Temperature, °C FIGURE 5.1 Development of volume solidiication shrinkage in an Al–4% Cu alloy [23]. (Reproduced with kind permission of Maney Publishing.)

part of the solidiication interval” [9]. The upper boundary of this range is the point where the stresses begin to build up [3]; the lower boundary is the solidus (equilibrium or nonequilibrium, depending on the solidiication conditions). Novikov [3] suggested determining the upper boundary of the effective solidiication range by measuring a so-called linear shrinkage. Hence, the upper temperature of this solidiication range is the temperature at which the “linear shrinkage” starts. Let us specify the terms we are using. Solidiication shrinkage is the shrinkage (usually volume shrinkage) that occurs in the solidiication range, from 100% liquid to 100% solid, due to the phase transformation as a result of density difference between the liquid and solid phases (see Figure 4.4). Solidiication shrinkage of aluminum alloys amounts to 6–8 vol.%, as illustrated in Figure 5.1. Thermal contraction is the contraction of the solid phase resulting from the temperature dependence of the solid density. (The liquid phase also contracts according to the temperature dependence of the liquid density, but this contraction does not cause any stresses or strains due to luidity of the liquid.) The thermal contraction is usually described by the linear or volume thermal expansion coeficient. The linear contraction (or shrinkage) is the horizontal change in linear dimensions of a casting during solidiication and usually ranges from one hundredth to one percent [3, 23, 24]. Above the temperature of the linear contraction onset, the semi-solid alloy is “loose” because between opposite walls of the mold there is no continuous network of dendrites. In this stage of solidiication, the solidiication shrinkage of the melt and the thermal contraction of the solid phase cannot manifest themselves as the horizontal contraction of the casting. All volumetric changes

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

187

appear as the decreasing level of the melt in the permanent mold (schematically shown in Figure 5.2a and experimentally demonstrated in Figure 5.2b) or do not appear at all during DC casting, due to the continuous supply of melt to the mold. However, the linear contraction appears, and can be measured, when the luidity of the alloy drops drastically, and the continuous rigid skeleton of the solid phase forms. Starting from that moment, the alloy acquires the capability of retaining its shape, and the volumetric changes display themselves as the linear contraction in the horizontal direction (Figure 5.2c). The understanding of the shrinkage and contraction phenomena occurring in the solidiication range is very important for the analysis of stress–strain development and modeling of hot cracking. The temperature dependence of the linear contraction in the solidiication range is used in some hot tearing criteria, as we will discuss later in this chapter. Below the solidus, the thermal contraction continues and frequently reveals itself as geometrical distortion of the billet shape. In DC casting practice, this phenomenon is known as butt curl (lifting of the billet shell from the starting block). The occurrence of butt curl can reduce the stability of the ingot standing on the starting block and is therefore a potential safety hazard. Besides that, the partial loss of contact between the ingot and bottom block will initially reduce the heat transfer with the possible danger of remelting. In the worst case scenario, butt curl can cause cracks and hot tears. A special technique was developed to measure the linear thermal contraction upon solidiication [3, 23–25]. Several designs of an experimental set-up were suggested, all sharing the following features suggested by Novikov [3]: graphite mold (providing low friction and high thermal conductivity) with one moving wall; water-cooled base (for high cooling rates comparable with those upon DC casting); and simultaneous temperature and displacement measurements (as shown in Figure 5.2c). Figure 5.3 shows a scheme and a photograph of the experimental set-up used in this work along with a sketch of an original set-up proposed by Novikov [3]. The experimental set-up used consists of the following parts: a T-shaped graphite mold (Figure 5.3b,c) with one moving wall; a water-cooled bronze base; and a linear displacement sensor (linear variable differential transformer or LVDT) attached to the moving wall from outside and aligned with the longitudinal axis of the mold. The reason for the T shape, which is narrower than the main cavity, is to make the melt solidify faster there than in the rest of the mold, and so the solidifying sample can be ixed on that side. To attach the solidifying metal in the cavity to the moving wall, we use a metallic rod with a thread (screw) embedded through the moving wall. The metallic rod ixed in the moving head is frozen in the sample immediately after illing the mold with the melt. The cross-section of the main cavity is 25 × 25 mm with a gauge length of 100 mm. Temperature measurement is of great importance for the reliability of results obtained in such experiments. The temperature should be properly correlated with the measured displacement of the moving wall so as to capture correctly the moment of contraction onset. Therefore, the point

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

188

Liquid

Solid−liquid part in the effective solidification range: dendritic network

Liquid−solid part

Linear contraction porosity

Decrease of melt level

Filling the mold

Solidification

(a)

Vertical displacement, mm

50 40

Solidification Mold filling

30 20 10 0 0

200

400

Horizontal displacement, %

0.2

(c)

600

800

Temperature, °C

(b)

Expansion

0 0 −0.2

200

400

600

800

Contraction

−0.4 −0.6 −0.8 −1

FIGURE 5.2 Schematic representation of external signs of solidiication shrinkage and thermal contraction (a) and experimentally measured melt level (b) and horizontal contraction (c) of an Al–4.5% Cu alloy [24]. Experiments were performed in a mold shown in Figure 5.3b. Solidiication shrinkage is revealed by the decrease in melt level (measured by a laser sensor) up to the moment when the rigid solid skeleton is formed and the contraction begins (measured by a linear variable differential transformer). Note the difference in scales on vertical axes. (Reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

189

(a)

(b) nt me ace kage) l p s l di shrin tica n Ver ficatio TC lidi o s (

Screw Moving 25 block

130 35

T-shaped end Screw

25

Horizontal displacement (linear thermal contraction)

57

(c)

(d)

FIGURE 5.3 Sketch of an original set-up by Novikov [3] made of graphite with sand inserts (a), photo of an experimental set-up on a water-cooled chill base used in this work (b), a scheme of the same set-up (c), and a sample after testing (d) [23, 25].

of temperature measurement should be in the location where the coherent solid structures propagating from cooled sides of the mold irst meet. This is the point where the lowest solid fraction of the continuous solid network is located. Experimental studies and heat transfer analysis of solidiication in the experimental mold demonstrated that the coherent structures meet in the central part of the mold close to its bottom, as shown in Figure 5.4a,b [26]. Depending on the design of the experiments, the temperature should be measured as illustrated in Figure 5.4c, either near the side wall [25] or in the center of the mold [24]. In the latter case refractory paint is applied at the walls to delay locally the solidiication. During the experiments the temperature and displacement are recorded simultaneously by a PC-based data acquisition system. The linear thermal contraction is determined as follows: εth = [(ls + ∆lexp − lf)/ls] × 100%,

(5.1)

where ls is the initial length of the cavity, lf is the length of the sample at the solidus temperature and ∆lexp is the pre-shrinkage expansion.

Copyright 2008 by Taylor and Francis Group, LLC

190

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

(a)

(b)

(c) FIGURE 5.4 Development of solid fraction in the central symmetry plane (a) and at the bottom (b) of the experimental mold and the proper location of thermocouples (c) [26]. Parts (a) and (b) are related to the case when no refractory paint is applied. The black thick lines in (c) indicate refractory paint. The application of the paint delays the solidiication at the walls and makes the solidiication front advance in a more straight manner from short sides of the mold [24].

The pre-shrinkage expansion is mainly due to the evolution of gases and the pressure drop over the mushy zone, and it depends on the alloying system, melting, and solidiication conditions [3, 27]. In most cases, liquidus and solidus temperatures can be derived from the cooling curve. Note, however, that we determine the linear contraction in the entire solidiication range which, at the cooling rates (5 to 10 K/s) used, extends to the lowest possible eutectic temperature—nonequilibrium solidus (NES). After acquiring the primary data, temperature, and displacement against time, the cooling curve is processed in order to obtain information about critical temperatures and cooling rates. After that, the data are reconstructed to ind the direct dependence of displacement on temperature. From this dependence the linear pre-shrinkage expansion, the linear solidiication contraction, the temperature of its onset, and the linear thermal contraction coeficient (TCC) can be extracted (see Figure 5.2c). The measured linear thermal contraction in the solidiication range is affected by the friction force applied to the moving block [23] and by the melt level in the mold [24]. The gauge length, however, does not inluence the measured values [24]. Figure 5.5 shows that thermal contraction decreases with larger friction forces (in the range 0.1 N to 0.85 N). At the same time, it has been demonstrated that the temperature of the contraction onset remains unchanged [23]. Another factor indicated in Figure 5.5 is a so-called structure factor SF that is deined as SF = Vc−0.33/Dgr,

Copyright 2008 by Taylor and Francis Group, LLC

(5.2)

Linear contraction, %

Hot Tearing

191

0.2

0.1

0.0 .1 0

4.5 .0 ) 0.2 .3 4 0 0 .4 3.5 × 100 .0 ( Fri 0 0.5 3 ctio or 6 2.5 fact n fo 0. .7 .0 e r 0 rce 2 8 u t ,N 0. .9 .5 uc 0 .0 .0 1 Str 1 1

5.0

FIGURE 5.5 The effect of friction force and structure factor on the linear thermal contraction of an Al–4% Cu alloy [23]. (Reproduced with kind permission of Maney Publishing.)

where Vc is the cooling rate characterizing the dendrite arm spacing (see Section 2.1) and Dgr is the grain size. The structure in this series of experiments was controlled by the cooling rate (7 to 18 K/s) and the melt temperature (700 to 800ºC), which allowed us to vary both DAS and Dgr. The grain size ranged from 120 µm to 220 µm [23]. Note that no grain reinement has been used in these experiments. For such a coarse dendritic structure, the linear contraction increases with the structure factor. In other words, the smaller the grain and the coarser the dendrite arms, the larger the thermal contraction in the solidiication range. As will be shown below, the situation changes when grain reiner is added. The melt level in the mold affects the measured parameters, which is evidently due to the combined inluence of thermal gradients, dimensionality of the solidifying sample (unidirectional solidiication, lat, or full threedimensional casting), and the mechanical and rheological properties of the mushy zone [24]. We know from casting practice that the geometrical changes in the inal size of a DC-cast billet cannot be readily explained as a result of the thermal contraction of a solid sample of the same size. The observed contraction is usually considerably larger. The “real” casting contraction increases with increasing casting speed and varies from the theoretical value (calculated using the linear thermal expansion coeficient) at a zero speed to up to 50% larger values at high casting speeds [28]. The increase of the experimentally observed contraction with respect to the theoretical value originates from thermal gradients in the casting. Evidently, different layers of the casting

Copyright 2008 by Taylor and Francis Group, LLC

192

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

are at different stages of solidiication, contain different amount of the solid phase, and contract at different rates. In addition, the internal, more liquid parts of the casting undergo, along with intrinsic contraction, tension imposed by the external, already solid, and more contracted shell. The difference between the theoretical and observed contraction depends, apparently, on the mechanical properties of the mush, the thickness of the solid shell, and the volume of complete luid slurry and melt (the depth and proile of the sump during DC casting). The “softness” of the liquid interior, which offers only negligible resistance to the contraction of the external shell, contributes to the experimentally observed contraction [26, 28]. Moreover, the pressure drop that appears in the mushy zone due to the solidiication shrinkage and the poor permeability of the mush could cause additional contraction by deforming the very weak solid network. The solidiication of melt in the experimental mold shown in Figure 5.3 b,c is in many aspects (although on the other scale) similar to the solidiication of an ingot or a billet cast from the top and cooled from the bottom (see Figure 5.4a,b). In this case the change of the melt level is similar to the decrease in the sump depth. Let us take a closer look at the contraction behavior in the solidiication range of different alloys. Figure 5.6 shows the temperatures of thermal contraction onset and the total contraction strain accumulated in the nonequilibrium solidiication range of binary Al–Cu and Al–Mg alloys. These data are superimposed on the binary phase diagram. It is quite obvious that the thermal contraction starts at a rather low temperature, close to or even below the equilibrium solidus. There is a maximum of the thermal deformation accumulated in the solidiication range, corresponding to the largest nonequilibrium solidiication range. The thermal contraction in the solidiication range depends on the alloy composition in general and is greater for Al–Mg alloys than for Al–Cu alloys. The volume fraction of solid in the mushy zone corresponding to the beginning of thermal contraction and, in our opinion, relecting the rigidity point, is quite high. At a cooling rate of 3–4 K/s, it ranges from 0.95 at 0.3% Cu to 0.8 at 4% Cu [25]. When the cooling rate is increased to 8–15 K/s, the solid fraction at the contraction onset reaches 0.9 at 4% Cu [23, 25]. The analysis of the contraction during solidiication is quite obvious and straightforward for binary alloys when the phase diagram with all temperatures and concentrations is known. The situation becomes more complicated in commercial, multi-component alloys, when a small shift in composition can dramatically change the solidiication path of an alloy, sometimes in rather unpredictable ways. Grain reining is an additional factor that may affect the contraction behavior of an alloy. As an example we tested a 6061-type alloy containing different amounts of copper (within and above the compositional range of the grade). The effect of grain reiner was also examined. The results are summarized in Figure 5.7.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

193 700

0.4

Contraction onset Contraction to NES

400

0.2 300 0

200 0

1

2

(a)

3

4

5

Cu, wt.% 0.8

700

Temperature, °C

600

0.6

500 Contraction onset

0.4

Contraction to NES

400

Expansion 0.2

300 200

Contraction/Expansion, %

Temperature, °C

500

Contraction, %

0.6

600

0 0

2

(b)

4

6

8

10

Mg, wt.%

FIGURE 5.6 Compositional dependences of thermal contraction (expansion) and the temperature of contraction onset superimposed on the binary phase diagrams for Al–Cu (a) and Al–Mg (b) alloys [24]. NES = “nonequilibrium solidus” shown by dashed lines. (Reproduced with kind permission of Springer Science and Business Media.)

The nonequilibrium solidus assumed was 500°C. The following conclusions can be made: • Contraction maxima in the solidiication range correspond to 0.2–0.3% Cu; • Contraction starts at about 600–610°C and the onset temperature decreases to 585°C with increasing copper concentration from 0.2 to 3.8% in nongrain-reined alloys; • Addition of grain reiner decreases the temperature of contraction onset by 20 to 60°C depending on the composition of an alloy, and thereby delays the contraction to higher volume fractions of solid;

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

194

650

Temperature, °C

550

0.4

500 0.2 450

Contraction/expansion, %

0.6 600

0

400 0

1

2

3

4

Cu, wt.% FIGURE 5.7 Effect of copper concentration and grain reining on the contraction (expansion) behavior of a 6061-type alloy: ◇, ◆ = contraction onset, °C; ❍, ● = contraction to NES, %; and ❑, ■ = expansion [24]. Open symbols refer to an alloy without grain rei ning and solid symbols, to an alloy with grain reining. NES = denotes “nonequilibrium solidus.” (Reproduced with kind permission of Springer Science and Business Media.)

• Contrary to the observations on coarse-grained alloys (Figure 5.5), smaller and more globular grains in grain-reined alloys decrease 2–3 times the total amount of thermal strain upon solidiication; • Grain-reined alloys exhibit much more pronounced pre-shrinkage expansion than nongrain-reined alloys, with the maximum between 0.7 and 1.5% Cu. As we can see, grain reining considerably inluences the contraction behavior of the alloy. At the same time, the pre-shrinkage expansion is much more pronounced in grain-reined alloys, compensating for some portion of the thermal contraction. The much larger pre-shrinkage expansion of grain-reined alloys can be explained in the following way. It is known that the expansion is caused by the evolution of gas (mainly hydrogen) during solidiication [3] and the pressure drop across the mushy zone [27]. In the experimental set-up used in this work, the solidiication starts at the ends of the mold (see Figure 5.4). The temperature gradient in the longitudinal direction causes nonuniform volume shrinkage and, therefore, an uneven pressure drop in the two-phase zone. The pressure in the mushy zone near the ends of the solidifying sample is lower than in the center. As a result there is a pressure-induced low directed from the center toward the ends of the sample. Since the dendrites of the solid phase are mostly separated by liquid ilms (expansion occurs above the rigidity point), they are shifted outward from the center of the sample. This movement is registered as a pre-shrinkage expansion. In the

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

195

case of grain-reined alloys, the structure consists of small, equiaxed grains that become bridged at relatively low temperature (contraction onset). Such a structure facilitates the relative movement of grains and, therefore, expansion. In addition, gas evolves from the melt more readily because ine grains provide more interfacial area and, in addition, allow more time before rigidity. The maximum of expansion frequently corresponds to the maximum of porosity [25, 27]. Therefore, alloys that show less thermal contraction in the solidiication range may exhibit larger interdendritic porosity. The same tendencies are observed in commercial alloys of different alloying groups. Figure 5.8 shows the development of solid fraction upon solidiication of commercial aluminum alloys with an indication of the thermal contraction onset. The corresponding values of the solid fraction at rigidity and contraction/expansion are given in Table 5.1. Obviously, commercial alloys behave differently upon solidiication, depending on the alloying system. Commercial aluminum (1050) does not show thermal contraction, which is no surprise since it has a very narrow solidiication range. Al–Mg (5182) and Al–Mg–Si (6082) alloys also exhibit 528°C/93% 100

563°C/95%

657°C/38%

515°C/87.5%

90 80

525°C/87%

Solid fraction, %

70 60 50 40

Al 1050 Al-Mg-Si 6082 Al-Zn-Mg-Cu 7075 Al-Mg-Mn 5182 Al-Cu-Mg 2024

30 20 10 0 480

520

560

600

640

680

Temperature, °C FIGURE 5.8 Development of the solid volume fraction upon nonequilibrium solidiication of some commercial alloys with the temperatures of thermal contraction onset and corresponding fraction solid being marked [23]. (Reproduced with kind permission of Maney Publishing.)

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

196

TABLE 5.1 Temperature of Thermal Contraction Onset (at Rigidity) (Tth), Fraction Solid at This Temperature ( fsth), Expansion (∆lexp), and Thermal Contraction in the Solidiication Range (εth) of Some Commercial Alloys (Melt Temperature 720–730°C, Cooling Rate 12–18 K/s) [23] Alloy

fsth, %

∆lexp, %

εth, %

657 515 528 563 525

38 87.5 93 95 87

0 0.035 0.08 0.025 0.03

0.03 0.2 0 0.02 0.27

Thermal contraction, % Fraction solid 7075+1% Si

7075+1% Cu

7075+1% Zn

7075+1% Mg

7075 GR

7075 NGR

6061+1% Mg

6061 GR

6061 NGR

2024+1% Mg

2024+1% Cu

2024 GR

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 2024 NGR

Fraction solid at the onset of contraction and thermal contraction in solidification range

1050 2024 5182 6082 7075

Tth, °C

FIGURE 5.9 Effect of grain reining (GR) and additional alloying on the fraction solid at the onset of thermal contraction and the total thermal contraction in the solidiication range of commercial alloys [29]. NGR = alloys without grain reining.

negligible contraction and only starting at temperatures below the equilibrium solidus. The largest thermal contraction in the solidiication range is demonstrated by Al–Cu–Mg (2024) and Al–Zn–Mg–Cu (7075) alloys that are also known, as we will show later, to be prone to hot cracking. But even in these alloys the thermal contraction starts between the equilibrium and nonequilibrium solidus. Similar investigation has been performed using commercial alloys with and without grain reinement and with increased amounts of main alloying elements [29]. The results are shown in Figure 5.9. The fact that the solid fraction at the contraction onset is lower in Figure 5.9 compared to the data

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

197

in Figure 5.8 and Table 5.1 is explained by the inluence of the lower cooling rate in this series of experiments, 2 to 5.5 K/s as compared to 12–18 K/s in the previous series. One can see that grain reinement increases the volume fraction at the thermal contraction onset (at rigidity) and generally decreases the overall thermal contraction, except for a 7075 alloy. This unconventional behavior of a 7075 alloy requires further investigation, especially because this alloy is very susceptible to hot tearing. Additional alloying, as a rule, decreases the solid fraction at rigidity and makes thermal contraction more pronounced. The study of thermal contraction provides important information about two parameters, i.e., the temperature (fraction solid) at which the contraction starts and the total amount of contraction (thermal strain) accumulated in the solidiication range of a particular alloy. The irst parameter makes it possible to establish a constitutive equation for the thermal strain evolution during solidiication [25, 26, 29] and apply it to the modeling of hot tearing [30]. The second parameter can be used for ranking alloys with respect to their hottearing susceptibility [3, 24]. The constitutive equation that relates the macroscopic thermal strain rate ε∙th to the solid fraction fs and temperature T is given by [24, 29]: ∙ 1 ε∙th = __ ψ( fs)βT T I, 3 0 ψ( fs) = fs − fsth _______ 1 − fsth

{(

n

)

for

with fs ≤ fsth

(5.3) ,

fs > fsth

∙ where βT is the volumetric thermal expansion coeficient, T is the cooling rate, I is the identity tensor, and n is a material parameter. The function ψ( fs) ensures that no thermal strain is transmitted through the mushy zone above the rigidity temperature (temperature of thermal contraction onset) or below the corresponding solid fraction. As soon as this critical solid fraction is reached, the macroscopic thermal strain in the mushy zone will tend toward thermal strain in a fully solid body as the solid fraction approaches unity. Computer simulations of experiments on thermal contraction have been used to determine the material parameter n and the shape of ψ( fs). It is shown that the material parameter can be taken as 0 for most commercial aluminum alloys [29]. This means that the ψ( fs) function is in fact a step function and the semi-solid material behaves as completely solid as soon as the mush acquires rigidity. Equation 5.3 reduces to the classical equation for thermal strain rate in fully solid material. In the case of binary alloys, it can be demonstrated that the ψ( fs) function becomes more complex when higher cooling rates or grain reinement are applied to an alloy. Figure 5.10 shows that the increased cooling rate and iner grains in an Al–4% Cu alloy delay the onset of thermal contraction upon solidiication and make the transition from viscous (liquid) to elasto-plastic (solid) mechanical behavior more gradual.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

198

NGR, 4 K/s, n = 0

1 0.8

GR, 4 K/s, n = 0.3

ψ

0.6 NGR, 8 K/s, n = 0.1

0.4 0.2 0 0.7

0.9

0.8

1

fs FIGURE 5.10 Function ψ ( fs) in Equation 5.3 for Al–4% Cu alloys [25]. NGR = without grain reining; GR = with grain reining; cooling rates and the material parameter n are shown.

The same experimental technique of measuring thermal contraction can be used for assessing the linear thermal expansion (contraction) coeficient. The knowledge of this material property is important for the analysis of geometrical changes of a billet in the last stages and immediately after solidiication. The linear thermal expansion coeficients are not readily available in reference books for commercial alloys at high, subsolidus temperatures. Moreover, the linear thermal expansion coeficient is usually determined in a dilatometer on long cylindrical (1D) samples under nearly isothermal conditions (i.e., no thermal gradients in the sample). In addition, these samples are carefully homogenized to achieve the equilibrium state of the alloy. This situation is very different from the real contraction conditions of a just solidiied and cooling bulk sample in which the processes of excess-phase precipitation may well continue. Figure 5.11 demonstrates the linear thermal expansion (contraction) coeficients (LTEC) for binary Al–Cu and Al–Mg alloys with respect to the composition. The LTECs have been obtained using curves like the one shown in Figure 5.2c. The coeficients were estimated in two temperature ranges: within 50°C below the nonequilibrium solidus (adopted as 550oC for Al–Cu alloys and 450°C for Al–Mg alloys) and from 300°C to the solidus. It is worth mentioning here that the results relect the nonequilibrium situation of cooling after solidiication. In binary Al–Mg alloys, the LTEC decreases with an increasing concentration of magnesium, the coeficients being very close at subsolidus and lower temperatures (see Figure 5.11b). This trend is quite distinct from the previously reported results. According to the literature, the linear thermal expansion coeficient in Al–Mg alloys is believed to increase with Mg concentration and temperature [3, 31]. In the given compositional range, the linear thermal

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

199

LTEC, 10−6 K−1

35

500−550°C 300−550°C

30

25

20 0

1

2

(a)

3

4

5

Cu, wt.% 35

LTEC, 10−6 K−1

500−550°C 300−550°C 30

25

20 0 (b)

2

4

6

Mg, wt.%

FIGURE 5.11 Linear thermal expansion (contraction) coeficients (LTEC) obtained for binary Al–Cu (a) and Al–Mg (b) alloys in different temperature ranges [24]. (Reproduced with kind permission of Springer Science and Business Media.)

expansion coeficient, as determined on annealed samples in a dilatometer, increases with magnesium concentration from 33.6 to 35 × 10−6 K−1 at 500°C [31] and from 29.2 to 30 × 10−6 K−1 at 400°C [3]. At the same time the average LTEC (20–500°C) slightly decreases with increasing magnesium, from 29.3 to 29 × 10−6 K−1 [31]. Our experimental results show a much more pronounced change in the TCC (300–450°C), from 29 to 25.5 × 10−6 K−1. The difference can be a result of different sample structures. In the case of reference data, homogenized, single-phase samples were tested. In our experiments, the sample structure was far from equilibrium, consisting of depleted solid solution and nonequilibrium eutectics. Another factor may be the presence of hydrogen in just-solidiied Al–Mg alloys. Afanas’ev et al. [32] reported that the LTEC at 400°C of as-cast Al–Mg alloys decreased with the increasing Mg concentration

Copyright 2008 by Taylor and Francis Group, LLC

200

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

as opposed to the behavior of annealed alloys. These authors noted that this anomalous behavior was a function of hydrogen content in the alloy. The contraction behavior of Al–Cu alloys in the temperature range from 300°C to 550°C complies with the previously reported data. According to Novikov [3] the LTEC at 500°C decreases from 32.4 to 31.2 × 10−6 K−1 upon increasing the copper concentration from 0 to 4.5%. The average LTEC (20– 500oC) decreases from 29 to 28 × 10−6 K−1 [31] in the given range of copper concentrations. In our experiments the average TCC (300–550°C) decreases from 30 to 28.5 × 10−6 K−1. This is in good agreement with the data from Ellwood and Silcock [33] on the subsolidus linear thermal expansion coeficients of binary Al–Cu alloys. They reported that the LTEC decreases from 32.4 × 10−6 K−1 at 0.95% Cu to 28.2 × 10−6 K−1 at 4.97% Cu. Though the estimation of LTEC by using this technique may be less accurate than using a dilatometer, it has two main advantages. First, it gives data for nonequilibrium structure that is close to real structures occurring in practice. Second, it is simple and does not require special equipment and sample preparation. As a result, the suggested technique can be used as a relatively simple means for the estimation of the real contraction coeficient at subsolidus temperatures, especially in cases when these coeficients are unknown. Thermal contraction during and after solidiication along with uneven cooling of a casting creates thermo-mechanical conditions that may facilitate the formation of cracks. However, the mechanical properties that semi-solid material acquires at supersolidus temperatures may be suficient to prevent failure. Let us take a closer look at mechanical properties of semi-solid aluminum alloys.

5.2

Mechanical Properties of Semi-Solid Alloys

In the last 60 years there have been many attempts to determine the properties of metallic alloys in the semi-solid state. Most of these studies were initiated with the aim of examining and studying hot tearing, in other words, the failure of the semi-solid mush under external or internally induced stresses. The latter usually originated from uneven thermal contraction during solidiication. Here we present a summary of the mechanical properties of semi-solid alloys. Interested readers are referred to the extensive reviews on the mechanical properties of semi-solid alloys by Novikov [3] and Eskin et al. [15] for more information. 5.2.1

Testing Techniques

For an alloy to be tested for mechanical properties it must have the ability to retain shape and to transfer load. In a lower part of the solidiication range, the so-called solid-liquid or semi-solid part, alloys meet these conditions and, therefore, can be mechanically tested.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

201

Mechanical testing can be performed in two ways: by heating the sample to a certain temperature above the solidus (remelting) or by cooling the liquid alloy to a certain temperature below the liquidus (solidiication). The structures thus obtained are different, and so also are the results of mechanical tests. Testing methods can be conditionally classiied as tensile, shear (including torque), and compression tests performed upon remelting or solidiication. The main dificulty during tensile tests of semi-solid samples is their very low strength and ductility due to the presence of liquid ilms at grain boundaries. Therefore, the conventional tensile testing machines cannot be used directly for studying the mechanical properties of mushy alloys. The irst tests of the strength of semi-solid materials date from the beginning of the twentieth century when Norton [34] tested an aluminum alloy by suspending various weights to a solidifying sample. Tamman and Rocha [35] and Verö [36] used plane and notched cylindrical samples, respectively, to determine the tensile strength of binary and commercial alloys above solidus. It was shown that the ultimate tensile strength and ductility dropped at a temperature of solidus. A reliable testing machine for tensile tests of aluminum alloys above solidus was designed and used by Singer and Cottrell [37] for studying Al–Si alloys. These authors decided that the sample should be short and completely surrounded by machine clamps in order to prevent shape distortion. The sample was tested in the horizontal direction. The determined property was the ultimate tensile strength, the elongation being almost zero. Researchers who studied hot cracking upon welding used mainly latshaped specimens cut from sheets. Pumphrey and Jennings [38] heated a plane Al–Si specimen in a lead bath and tested it upon tension in the vertical direction at a rate of 25 mm/min. The temperature dependence of the ultimate tensile strength was constructed, but not for the elongation that scattered above solidus between 0 and 3%. By the late 1940s the techniques for assessing the ultimate tensile strength of semi-solid aluminum alloys were well established. However, there were no reliable methods to measure the elongation or other plastic characteristics of alloys heated above their solidus. These techniques were developed in the 1950s mainly due to the efforts of Russian scientists, e.g., see Refs. [3, 39, 40]. The main dificulties with respect to the measurement of elongation in the semi-solid state were (1) poor centering of the sample and, therefore, offset of the load and (2) low accuracy of measurements. The simplest way to solve the irst problem is to use the scheme where the sample self-centers under gravity [3, 39]. Recently, Gleeble 1500 and 3500 thermo-mechanical simulators (testing machines) were used for studying mechanical properties of aluminum alloys upon remelting in the semi-solid state [41]. The signiicant advantage of Gleeble-series testing set-ups is the use of Joule heat for rapid melting of the sample. In combination with modern means of strain–stress and temperature acquisition and control, this increases the accuracy and reliability of measured results and

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

202

600 2

El, %

P, 10−2 N

1 400

200

0.2

0.4

0.6 0.8 l, mm

1.0

T, °C

FIGURE 5.12 Typical stress–strain curve of an Al–6% Cu alloy tested at 570°C at a rate of 1 mm/min (a) and schematic temperature dependences of elongation for as-cast (1) and worked (2) samples tested upon remelting in the semi-solid state [3].

ensures more complete retention of the as-cast structure to the beginning of the test. Figure 5.12a illustrates an example of a tensile curve recorded for an Al–6% Cu alloy tested at 570oC (22oC above the solidus) and at a strain rate of 1 mm/min. The typical feature of stress–strain curves for semi-solid testing is a well-pronounced branch of gradually decreasing stress, despite brittle fracture. Novikov [3] explained this phenomenon by the local deformation of solid bridges and extension of liquid intergranular ilms existing in the semisolid sample. As a result, the effective cross-section of the sample decreases and the stress–strain curve shows the gradual decrease of the stress. However, the corresponding elongation is ictional because the sample is already fractured. Therefore, it is more reliable to determine the actual elongation by measuring the fractured sample with a microscope after the test. It is worth mentioning here that plane specimens cut from rolled or extruded billets are quite useful for studying semi-solid properties and sensitivity to hot cracking of welded joints and heat-affected zones. But deformed samples should not be used for studying semi-solid properties and hot tearing of cast metals because the mechanical properties and, in fact, the structures in the melting range of cast and deformed materials are different. Figure 5.12b illustrates the typical difference in plastic behavior between cast and deformed alloys tested in the semi-solid state. The lower amount of eutectics in semi-solid deformed material decreases its elongation, whereas the iner grain structure in this material should increase its plasticity [3]. The examination of mechanical properties of alloys during the solidiication is, of course, essential for studying the hot tearing phenomenon. The methods for this can be divided into two general groups, namely, tests upon

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

203

continuous cooling when the sample exhibits solid, mushy, and liquid zones simultaneously, and tests under isothermal conditions when the sample is cooled in a controlled fashion to a certain temperature and then tested. Hall [42] reported perhaps the irst attempt of tensile tests upon continuous solidiication. Steel of different compositions is poured in sand molds with bolts on both sides connected with clamps of a tensile testing machine. When frozen into the solidifying melt the bolts transfer the tensile stress onto the casting. Gulyaev [43] determined the tensile strength of solidifying steel by dividing the load at failure by the cross-section area of a ring layer (determined in special experiments by pouring out the rest of the melt). Determined values of tensile strength are ascribed to the average temperature of the ring layer. In the 1980s and 1990s a group of scientists [44–46] developed and extensively used the technique of measuring tensile stresses and elongation of solidifying shells. The idea was to simulate the formation of a solid shell upon DC casting and to gain insight into its deformation behavior. The technique includes the formation of a solidifying shell onto a water-cooled chill and the subsequent tensile testing of this shell. Yet another apparatus for characterizing tensile strength development upon solidiication was developed by Instone et al. [47]. This testing setup provides constraint during solidiication and collects information about strength development, strain accommodation, and hot cracking behavior of the mushy zone. The characteristics of strength and plasticity obtained for solidifying samples under nonisothermal conditions can only be accepted conditionally as mechanical properties of an alloy at different temperatures in the solidiication range. These characteristics relect the averaged behavior of a casting and, therefore, the corresponding tests can be classiied as sophisticated technological probes rather then pure mechanical tests. To obtain real mechanical properties of an alloy in the solidiication range, the test should be performed under isothermal (or nearly isothermal) conditions. In experiments by Pumphrey and Jennings in the 1940s [38] an aluminum alloy was poured into a horizontal, split, stainless steel mold, two halves of which were connected with clamps of a tensile testing machine. The working part of the sample was completely surrounded by the mold. Wedge-shaped heads of the sample (parts of the mold) were opened. Melt was poured into the preheated mold and allowed to cool slowly to the test temperature. The authors managed to determine the tensile strength of Al–Si alloys in the solidiication range, but failed to measure the elongation. In the late 1950s [40] Prokhorov and Bochai used their set-up to examine the mechanical properties of aluminum alloys upon solidiication. A long, lat sample was placed on the copper support in the furnace, melted, cooled to a testing temperature, kept at this temperature for a certain time, and then stretched out to failure. Some Al–Cu, Al–Si, and Al–Mn alloys were

Copyright 2008 by Taylor and Francis Group, LLC

204

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

tested this way, and their tensile strength and elongation were measured. However, this technique is not very reliable due to large corrections required to compensate for the change of specimen sizes during heating–cooling due to the expansion–contraction. Novikov [3] described a tensile machine for testing cylindrical samples in the solidiication range. The melt was cooled in the mold to the testing temperature and then tensile tested. This set-up was used to measure the elongation in the solidiication range of aluminum alloys. However, the same design can be used for studying tensile strength (it should be additionally equipped with a dynamometer or load cell), as was done by Kubota and Kitaoka [48]. Forest and Bercovici [4] developed a set-up allowing the controlled solidiication of an alloy and tensile testing of it under nearly isothermal conditions. However, only ultimate tensile strength was measured. The tensile sample was a rod placed in a centrally situated graphite crucible capable of holding the metal in the liquid or mushy state. The crucible was heated by a focused induction furnace and then water-cooled to a test temperature inside the vertical tensile testing machine. The test was made perfectly uniaxial by means of balls and sockets. The sample temperature and the applied load were recorded throughout the experiment. Similar techniques was used by Suvanchai et al. [49], Suéry et al. [50], and Nagaumi et al. [51]. In situ melting and solidiication tensile tests were performed by Wisniewski [52]. Samples were melted and resolidiied in a stainless steel mold coated with a graphite lubricant. The mold had three sections, each of which was split longitudinally. The upper and lower grip portions were attached to steel pull rods. The center section was designed to align itself with the two ends when a small compressive force was applied. The center was not attached to the end sections and was free to move longitudinally when the compressive force was removed. Stress–strain curves were recorded, allowing one to determine the ultimate tensile strength and the elongation at rupture. Another way of testing mechanical properties in the semi-solid state is to deform the mush in shear [10, 53, 54]. The force–deformation curve can be recorded. One of the valuable features of this type of experiment is the possibility of observing the interaction between dendritic grains as they move relative to one another. The development of strength in solidifying alloys can also be examined by another sort of measurement, by torque upon rotating a vane in the mush [55]. A four-bladed vane is used to measure the fracture strength of the dendritic network during solidiication. The shearing zone is approximated to be cylindrical, deined by the dimensions of the vane. The torque–time curve at a certain temperature exhibits a yield point relecting the moment when the interdendritic connections start to break. After network collapse the torque decreases rapidly. The yield stress can be calculated from the measured maximum torque and the known surface area of the cylindrical yielding surface. The strength development can be evaluated by performing experiments at several points in the entire solidiication range.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing 5.2.2

205

Strength Properties of Aluminum Alloys in the Semi-Solid State

In the mid-1940s, Singer and Cottrell [37] were the irst to systematically study the tensile strength of aluminum. They heated samples of Al and Al–Si alloys to a certain temperature (including temperatures above the solidus) and then tested them in tension at a deformation rate of 6 mm/min. It was shown that the ultimate tensile strength gradually decreases as the testing temperature increases, and abruptly drops at the solidus temperature. This sharp drop in strength was accompanied by an almost complete loss of ductility. Extensive studies of mechanical behavior of semi-solid binary aluminum alloys performed during remelting and solidiication showed the following general trends. A coarser grain structure and internal dendritic structure decrease the tensile strength. This phenomenon can be explained by the decreasing number of solid bridges between coarse grains [3]. Grain reinement shifts the beginning of the strength build-up to higher fractions of solid [3, 10, 55]. Since the dendritic grains in the nongrain-reined alloy are large and highly branched, they are very much entangled in one another, and the structure cannot accommodate much strain by rearrangement of the grains. Instead, strain must be adopted by deformation of the dendrites proper, and extensive dendrite fragmentation can occur. Strength therefore increases rapidly. In the grain-reined alloy, however, the dendrites are much smaller and almost round, which allows them to rearrange more easily by sliding during straining. Deformation is then concentrated in interdendritic areas, and strength develops more slowly. An increase in the cooling rate in the range relevant to DC casting shifts the onset of strength development to higher temperatures and increases the strength level [4]. The strength of semi-solid aluminum alloys increases with the deformation rate [3, 56]. This is interpreted as an effect of the deformation rate on the fracture resistance of solid bridges between grains. Creep and stress relaxation do not occur if the strain rate is suficiently high. It has been also shown that liquid remains in triple junctions at high strain rate whereas it spreads over grain boundaries at lower strain rates [41]. The effect of alloying elements on the strength of aluminum alloys is due to their effect on the structure, temperature range of solidiication, and liquid–solid interface. Novikov [3] showed that an addition of 0.8% Fe to an Al–0.7% Cu alloy considerably increased the strength in the semi-solid state. The study of microstructures of samples quenched from the mushy state shows that the “liquid” ilms are much less continuous in the ironcontaining alloy, and therefore it contains more solid bridges between grains. An addition of 0.2% Fe to an Al–1.5% Mn alloy shifts the entire softening curve to higher temperatures although the solidus decreases somewhat due to the formation of the ternary eutectics. Structure examination demonstrates that the structure with elongated Al6Mn particles at grain boundaries is substituted for the structure with small and compact Al6Mn and Al3Fe eutectic particles located at the grain boundaries, ensuring therefore the existence of numerous solid bridges upon melting/solidiication.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

206

0.15

1

UTS, 0.1MPa

UTS, 0.1MPa

0.3 2

0.2

1 0.1

0.05

0.1

2 600 (a)

620

640 T, °C

540 (b)

560

580 T, °C

600

FIGURE 5.13 Temperature dependence of ultimate tensile strength of (a) Al–1.5% Cu and (b) Al–7% Cu alloys [3]. 1 = Without grain reining and 2 = with grain rei ning. Testing performed upon remelting.

Figure 5.13 demonstrates that the positive effect of grain reinement in a low-copper alloy (a) can be completely superseded by the negative effect of the continuous liquid ilms at grain boundaries in a high-copper alloy (b). The mechanical properties of semi-solid commercial wrought alloys are particularly important because these properties determine to a great extent the hot-cracking susceptibility of these alloys, usually produced by DC casting. Before considering the results of mechanical tests, it is important to note that the properties of semi-solid material depend not only on its chemical composition but also on the initial state. As shown experimentally, as-cast and deformed materials behave differently [3]. This was discussed briely above in connection with ductility in the semi-solid state (Figure 5.12b). Obviously, the deformed and homogenized material has iner, rounder (equiaxed) grain structure with much fewer excess particles at grain boundaries. As a result, the solidus of such a material and the strength in the lower part of the solidiication range are higher (Figure 5.14a). However, closer to the liquidus the situation may change because the bonding (bridging) between developed dendrites is better than that between round small grains (Figure 5.14b). The research in the tensile properties of semi-solid aluminum alloys shows the existence of two characteristic points in the solidiication range, i.e., zero-strength and zero-ductility temperatures. The transition from ductile to brittle behavior is associated with the onset of the formation of continuous liquid ilm at grain boundaries. Different alloys behave differently on approaching the solidus. For example, a 3004 alloy exhibits signiicant hardening after the mush starts to build up strength, suggesting that it cannot fully accommodate thermally induced stresses [57]. Conversely, 2024

Copyright 2008 by Taylor and Francis Group, LLC

0.75 2 0.50 1 0.25

520

(a)

207

UTS, 1.0 MPa

UTS, 1.0 MPa

Hot Tearing

530

540 550 T, °C

560

570

(b)

0.75

0.50

2 1

0.25

550 560 570 580 590 T, °C

600

FIGURE 5.14 Temperature dependence of ultimate tensile strength of cast (1) and hot-rolled and annealed (2) 5456-type (Al–6% Mg) (a) and 7039-type (Al–3.3% Zn–4.2% Mg) alloys (b) [3]. Testing performed upon remelting.

36

UTS, MPa

28 1

2

8

4

440

500

560 T, °C

FIGURE 5.15 Temperature dependence of ultimate tensile strength of a 7075 alloy with (1) 0.1% Fe and 0.1% Si (Fe:Si = 1) and (2) 0.5% Fe and 0.1% Si (Fe:Si = 5) [3]. Testing performed upon remelting.

and 7075 alloys increase the strength slowly on the decreasing temperature until several degrees above the solidus ( fs = 0.8 and 0.95, respectively) when the strength increases abruptly [57]. Small changes in the alloy composition, e.g., concentration of impurities, may considerably affect the mechanical behavior of a semi-solid alloy as shown in Figure 5.15 for a 7075 alloy with different ratios of Fe and Si. These data are

Copyright 2008 by Taylor and Francis Group, LLC

208

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

strong proof for the well-known dependence of the hot-cracking tendency on the Fe:Si ratio in an alloy. The general recommendation is to keep the Fe:Si ratio above unity [58]. The explanation is that the concentration threshold for the occurrence of the low-melting eutectics with the β(Al5FeSi) and Si phases shifts to higher iron and silicon concentrations upon increasing the Fe:Si ratio. This prevents the formation of free silicon that tends to segregate to grain boundaries and form low-melting eutectics, thereby decreasing the strength level in the semi-solid state [59]. The results in Figure 5.15 show clearly that the increase in the Fe:Si ratio indeed signiicantly strengthens the mush in the lower part of the solidiication range. 5.2.3

Ductility of Aluminum Alloys in the Semi-Solid State

Athough strength in the semi-solid state is an important property, many consider ductility of the semi-solid metal to be the property that determines the hot-tearing tendency of the material because the ductility characterizes the ability of the mush to accommodate stresses and strains occurring upon solidiication as a result of solidiication shrinkage and thermal gradients. The deformation ability of the semi-solid metal is frequently used in various criteria and models of hot tearing. However, the experimental data on the ductility of aluminum alloys in the semi-solid state are scarce. The reason is that the values are so low that there is no sense in measuring them because they frequently compare on the same scale with the accuracy of measurements. Most of the data are collected in the works of Novikov et al. [3] who considered ductility (tensile elongation to rupture) to be the main feature, along with linear thermal contraction, of hot cracking susceptibility of metallic alloys. These works are reviewed elsewhere [15]. First, let us consider the main features of the temperature dependence of the ductility in the solidiication range. All alloys exhibit some, albeit low, ductility at temperatures above the solidus. The temperature dependence of the ductility in the semi-solid state has a typical U-shape with an intermediate temperature range where the ductility is very low. In this brittle range, the ductility of alloys different in composition varies only by tenths of a percent, but even these small variations can be crucial for hot tearing susceptibility. Several common types of ductility dependence that can be observed in the semi-solid range are demonstrated in Figure 5.16 [3]. Some alloys, e.g., Al–1.4% Mn, show a sharp drop in ductility above solidus. The ductility remains almost permanent upon heating in some temperature range, and then it sharply increases (see Figure 5.16a). Many commercial alloys, e.g., 7039 and 7075, demonstrate behavior as shown in Figure 5.16b. The type shown in Figure 5.16c is rather rare, but it can be observed in an Al–3% Zn alloy. The range of constant low ductility can be wide or very narrow, as shown by the dashed line in Figure 5.16a–c. Figure 5.16d depicts behavior typical of many binary and complex aluminum alloys with a stepwise change of ductility. Such behavior is observed in 2024, 7064, and 3XX.0-series alloys.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

209

a

b

c

d

El

El

T

T

T

T

FIGURE 5.16 Typical forms of temperature dependence of elongation (El) of semi-solid alloys [3].

Let us look at some factors that inluence the semi-solid ductility. Novikov [3] experimentally conirmed the ideas of Singer and Cottrell [37] and Pumphrey and Jennings [38] that the main deformation mechanism in the semisolid state is the intergranular slide. Taking this into account, the role of grain size and shape becomes obvious. The elongation of a sample due to the intergranular deformation (δig) depends on the number of interfaces per unit length of deformation and equals approximately [3]: δig = 0.6 ∆l/Dgr,

(5.4)

where ∆l is the displacement of two adjacent grains and Dgr is the average grain size. Figure 5.17a shows the effect of grain size on the tensile elongation of an Al–4% Cu alloy. The analysis of these results reveals that the grain reinement is accompanied by a decrease in the intergranular displacement ∆l. At the same time, this example shows that the semi-solid ductility increases with grain reinement, which is the frequently observed case. However, in general, the resultant ductility depends on the ratio at which ∆l and Dgr are changing. The transition from equiaxed structure to columnar can decrease the ductility considerably in the most vulnerable temperature range (see Figure 5.17b,c). This can be explained in terms of less available interfaces of intergranular slide and fewer possibilities for deformation by grain rotation. The grain coarsening or the equiaxed–columnar transition shift the upper boundary of brittle range to higher temperatures (see Figure 5.17). The position of the lower boundary depends on the development of intergranular deformation. In most cases, the development of columnar and coarse equiaxed grains shifts this boundary to higher temperatures because the intergranular deformation occurs more easily in ine-grained material (see Figure 5.17a,c). But in some cases, for example, when the surface of columnar grains is very smooth, the dependence is the reverse (see Figure 5.17b). The structure as we know is strongly affected by cooling rate upon solidiication. The effect of cooling rate on the ductility of semi-solid alloys is illustrated in Figure 5.18. The solidiication conditions are chosen in such a way that the grain size and the structure type remained the same.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

210

2.0

1.0 0.5

Solidus

El, %

2 1.5

1

550

560

570

2.4

2.4

2.0

2.0 1.6

1.6

1

1.2

El, %

El, %

580

590

T, °C

(a)

0.8

1 2

1.2 0.8

2 0.4

(b)

620

630

640

T, °C

650

660

530 (c)

SoL

610

0.4

550 570

590

610

630

650

T, °C

FIGURE 5.17 (a) Effect of grain size (1 = coarse structure; 2 = iner structure) and (b,c) structure type (1 = equiaxed; 2 = columnar) on elongation (El) of semi-solid Al–4% Cu (a), Al–1.5% Cu (b), and Al–5% Cu (c) alloys [3]. Testing performed upon remelting.

Slower cooling shifts the lower boundary of the brittle range to lower temperatures. In low-alloyed materials (Al–1.5% Cu), in which the amount of nonequilibrium eutectic liquid is small and distributes as isolated inclusions, the lower boundary of the brittle range is well above the nonequilibrium solidus (548oC for Al–Cu alloys) (see Figure 5.18a). At a high cooling rate when the dendritic grains are well developed and the liquid inclusions are particularly ine, the lower boundary of the brittle range is close to the temperature of the equilibrium solidus. Such an effect of cooling rate correlates well with the fact that the surface of grains becomes smoother upon decreasing the cooling rate and, hence, the transition from intergranular to transgranular fracture occurs at a lower temperature [3]. In the alloys containing more copper, the amount of remaining eutectic liquid is much larger and the alloys are brittle in the entire solidiication range, the lower boundary of the brittle range being upon slow cooling even lower

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

211

2.7 K/s

0.3 K/s 1.6

El, %

1.2

0.8

0.4

590

600

Tsol

0.17 K/s 610

620

630

640

650

T, °C

(a)

1.6 2.7 K/s

El, %

1.2 0.17 K/s 0.8

0.4

530

550

(b)

570

590

610

T, °C

FIGURE 5.18 Effect of cooling rate upon solidiication on tensile elongation (El) of Al–1.5% Cu (a) and Al–5% Cu (b) alloys tested upon remelting [3].

Copyright 2008 by Taylor and Francis Group, LLC

212

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

than the temperature of the lowest eutectics (see Figure 5.18b). This phenomenon can be explained by very weak Al/Al2Cu interfaces. Although the position and span of the brittle range depend on the solidiication conditions, the ductility itself remains virtually unaffected. Two factors acting in opposite directions compensate for each other: upon decreasing the cooling rate the liquid layers become thicker but simultaneously less continuous. The size, shape, and distribution of liquid phase inclusions at grain boundaries should essentially affect the development of intergranular deformation. So-called liquid-metal embrittlement is discussed in detail elsewhere [3, 46, 60–63], and we refer the reader to these sources for more information. In brief, the liquid-metal embrittlement and thus the ductility and strength are related to the dihedral angle, which, in turn, is a function of the relative surface energies of the solid grain boundaries and the liquid: σ ls/σss = 1/2cos(θ/2),

(5.5)

where σ ls and σss are the surface energies on the solid–liquid and solid–solid interfaces, respectively, and θ is the dihedral angle for liquid inclusions. The smaller the dihedral angle, the greater the embrittlement caused by liquid inclusions. For aluminum alloys the dihedral angle is usually less than 20–25° and decreases with increasing temperature [3, 14]. At such values of the dihedral angle, 1% of liquid fraction is suficient to cover approximately 30% of grain boundaries with liquid, and at 10% liquid nearly 85% of grains are covered with liquid. The amount of liquid at the last stages of solidiication is usually represented by nonequilibrium eutectics that can be formed by the main alloying elements and/or by impurities. It is experimentally shown that impurities present in alloys in amounts of tenths of a percent can considerably affect the ductility and the brittle range proper. The change in the purity of base aluminum does not affect the position of the upper boundary of the brittle range, whereas the lower boundary shifts toward lower temperatures. This is due to the fact that the local melting temperature at grain boundaries is considerably lower in the presence of impurities (low-melting eutectics). At the same time, the ductility itself can be somewhat higher for alloys prepared on low-purity aluminum. This can be interpreted in terms of the increased amount of intergranular liquid, which helps accommodate intergranular deformation at supersolidus temperatures. These effects are illustrated in Figure 5.19. The ductility of semi-solid materials measured upon tensile tests depends on the strain rate. Novikov and Novik [64] showed that a higher deformation rate (~0.004 s−1, ~0.04 s−1, and dynamic loading) does not affect the position of the upper and lower limits of the brittle range but decreases the amount of elongation in semi-solid alloys with suficient amount of liquid, e.g., Al–6.5% Cu, whereas the behavior of less alloyed alloys, e.g., Al–1.5% Cu, remains virtually unaffected. Hence, the effect of the deformation rate depends

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

213

El, %

2 2 1 1

0 560

580

600

620

T, °C FIGURE 5.19 Effect of impurity level on tensile elongation (El) of an Al–3.5% Mg alloy prepared using 99.7% pure Al (1) or 99.99% pure Al (2) [3]. Testing performed upon remelting.

mainly on the amount of liquid phase (thickness of liquid intergranular ilms) in the alloy. Because plastic deformation of semi-solid alloys occurs mostly by grain sliding assisted by intergranular liquid layers, the accommodation of strain by liquid ilms and grain sliding occurs more easily at lower deformation rates. In contrast, Qingchun et al. [56] and Wisniewski [52] reported that the failure strain increased with strain rate above 0.001 s−1. Qingchun et al. [56] mentioned that the observed phenomenon was due to the fact that creep and stress relaxation do not occur at high strain rates. Wisniewski [52] suggested that the liquid-ilm cavitation may contribute to the increased strain-at-failure at high strain rates. It is quite possible that the different testing conditions produce different dependencies of the observed ductility on the strain rate. The “soft” (force control) scheme used by Novikov allowed for creep and strain accommodation by redistribution of liquid, whereas the “rigid” (strain-rate control) scheme employed by Wisniewski caused liquidilm cavitation and increased apparent plasticity at high strain rates. The data on ductility of commercial aluminum alloys are scarce. Most authors did not report data, but only mentioned that ductility was very small and could not be reliably measured. However, some authors reported the results on ductility of commercial aluminum alloys in the semi-solid state [3, 19, 49, 65]. Figure 5.20 shows the results of tensile tests performed upon melting of commercial alloys of 7075 (Al–Zn–Mg–Cu), VAD 23 (Al–Cu–Li), and 2037 (Al–Cu–Mg–Si) alloys. The temperature dependence of tensile elongation given in Figure 5.20c for a 2037-type alloy characterizes the effect of structure coarsening and gas concentration on the ductility of the mush. The samples for these experiments

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

214

50 30 4 10 El, %

El, %

2.0 1.6

3 2

1.2

1

2

0.8

1

0.4 εth 520

440 460 480 500 520 540 560 580

(a)

(b)

T, °C

540

560

590

600

620

T, °C

1.0 El, %

1 2 0.5

450 (c)

480

510

540

570

T, °C

FIGURE 5.20 Temperature dependence of tensile elongation of (a) 7075 alloy; (b) VAD23 alloy with 5.15% Cu, 1.2% Li (1) and 6% Cu, 1.2% Li (2); and (c) 2037 alloy that was held liquid in the furnace for 20 min (1) and 10 h (2) [3]. Testing performed upon remelting.

were prepared from 200-mm-diameter ingots direct-chill cast at a speed of 130 mm/min after holding the melt in the furnace for 20 min (1) or 10 h (2). The holding time coarsens the structure and, as shown experimentally [3], decreases the amount of hydrogen in the melt from 0.30 to 0.19 cm3/100 g. These two factors (grain coarsening and decreased amount of gas) can considerably deteriorate the plastic ability of the semi-solid metal and favor hot tearing [3]. The effect of gas amount has been explained from its effect on linear thermal contraction. The higher the gas concentration in the melt, the lower the linear contraction and, therefore, the higher the “plasticity” reserve of the mush. Experimental results for tensile properties of commercial aluminum alloys are summarized in Table 5.2 [3, 4, 19, 46, 49, 51, 57, 66–68]. There are two

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

215

TABLE 5.2 Tensile Properties of Some Commercial Wrought Aluminum Alloy According to Reference Data Alloy Al–Mg 5456 Al–Mg–Mn 3004 Al–Mg–Si 6061 6063

6110

HS60 HS65

Al–Cu–Mg 2024

2618 2017

2037

UTS, MPa 1.0 at 550°C — — 0.85–1.75 at 610°C (GR) 0.6 at 610°C (NGR) 1.0 at 565°C 0.25 at 607°C (GR) 3 at 607°C —

0.5 – 0.6 at 560°C 1 at fs = 0.9 5 at 520°C 2.6 at 540°C 0.39 at 560°C (fs = 0.9) 1.4 at 520°C —

Al–Zn–Mg–(Cu) 7075 1.0 at 520°C 3.2 at 480°C 6.5 at 470°C 7475 7005 7039

El, %

7064

— — 2.5 at 575°C 3.3 at 570°C —

Al–Cu–Li 5–6 Cu, 1.2 Li

0.4 at 550°C 0.9 at 530°C

0.075–0.18 at 610°C (GR) 0.1 at 610°C (NGR) 0.9 at 565°C 0.6 at 607°C (GR)

Zero Strength

Zero Ductility

Ref.

570



3

fs = 0.81



57

623 ( fs = 0.82) 620 ( fs = 0.9) GR 630 NGR

— —

49 49, 46

fs = 0.7

fs = 0.77

66

0.4 at 607°C —

620 ( fs = 0.69) 623 ( fs = 0.78) to 631 ( fs = 0.63) on incr. [Fe]

610 ( fs = 0.77) 613 ( fs = 0.82) to 617 ( fs = 0.78) on incr. [Fe]

51 67

2.7 at 560

fs = 0.76



19, 57



610°C



3



580°C



4



3

1.8 at 500

0.37 at 510°C 0.27 at 500°C 0.22 at 490°C 0.17 at 520°C 0.1 at 490°C 40 at 480°C 35 at 470°C — — 0.15 at 575°C



490–500

3, 57

600–610 630 600

530–550 610 570

68 68 3

0.5 at 480–500°C 0.1–0.15 at 450–460°C





0.7 at 550°C 0.15 at 530°C





Copyright 2008 by Taylor and Francis Group, LLC

570°C fs = 0.92

3

3

216

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

temperatures that can serve as benchmarks for characterizing the behavior of semi-solid material. A zero-ductility temperature marks the lowest temperature in the semi-solid range when the material acquires ductility. A zero-strength temperature corresponds to the temperatures when the semi-solid material starts to show some strength. Typically, the zero-strength temperature is higher than the zero-ductility temperature, meaning that the semi-solid alloy can start to build up some strength (and transfer stresses) while remaining absolutely brittle (ductility is zero) [51, 66, 67]. In conclusion, there are quite a few ways to measure or evaluate mechanical properties upon melting or solidiication. The strength and ductility of the semi-solid material depend on its initial structure (solidiication conditions, structure modiication by alloying, deformation, etc.), position of an alloy on the phase diagram (amount of eutectics), strength properties of solid bridges, and wetting conditions on the liquid–solid interface.

5.3 5.3.1

Mechanisms and Criteria for Hot Tearing Historic Overview of Hot Tearing Research and Suggested Mechanisms

Foundrymen have been well acquainted with hot cracking for centuries. Compositions of alloys for casting large cannons and bells were empirically chosen in order to decrease solidiication shrinkage and thermal contraction in the solidiication range. In the case of early castings that date from several centuries ago, an alloy of 75–85% copper with 15–25% tin was selected for its castability without cracking (though thermal contraction in the solid state could cause cold cracking). The thermo-mechanical situation during solidiication was also recognized as a vital condition for producing a sound casting. It was not uncommon that solidiication and cooling of a casting took several days or even weeks in an attempt to prevent hot and cold cracks. Hot cracking was a problem that faced the pioneers of DC casting, especially when they tried to cast high-strength alloys and large-scale lat ingots. Livanov [69] mentioned in 1945 that the problems of cracking in DC-cast ingots were so severe that the attempts to cast them were discontinued in the former Soviet Union during World War II, between 1940 and 1945. Only after considerable experience was accumulated from casting round billets did it become possible to return to the technology of lat ingots for rolling. But even then, until the mid-1950s, large-scale ingots from 7075-type alloys were cast with secondary cooling by air in order to reduce thermal stress [58]. The deformation behavior of the mush that was discussed in the previous section is very critical for the formation of hot tears. The link between

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

217

the appearance of hot tears and the mechanical properties in the semi-solid state is obvious and has been explored for decades. This link is relected in a number of hot tearing criteria [3, 15]. Another important correlation between the hot cracking susceptibility (HCS) and the composition of an alloy has been established on many occasions. The large freezing range of an alloy promotes hot tearing because such an alloy spends a longer time in the vulnerable state in which thin liquid ilms exist between the dendrites. Two parameters are important here. First of all, as we showed in Section 5.1, the amount of thermal strain accumulated during the solidiication range depends on the extent of the nonequilibrium solidiication range and, as such, on the alloy composition (Figure 5.6). Another important issue is the distribution of liquid at the grain boundaries. The liquid ilm distribution is determined by wetting of grain boundaries, i.e., by the surface tension between liquid and solid phases (see Equation 5.5). When the surface tension is low and wetting is good, the liquid tends to spread out over the grain boundary surface, which strongly reduces the dendrite coherency, weakens the mush, and may promote hot tearing. Otherwise, the liquid will remain as droplets at the grain junctions so that the solid network maintains its strength. The surface tension of the remaining liquid can produce a Marangoni force that facilitates the removal of precipitating gas from the potential voids and thereby reduces hot tearing [70]. At relatively low fractions of solid, e.g., 80–85%, the interdendritic liquid can provide a bond between grains that creates a tensile strength of the mush [71]. The dependence of hot cracking on the composition of binary alloys is described by a typical “lambda” curve, as shown in Figure 5.21 [72]. One can see a nearly perfect correspondence between the compositional dependences of hot tearing susceptibility in Figure 5.21 and of thermal contraction upon solidiication in Figure 5.6. This is additional proof that thermal contraction accumulated in the solidiication range is a good measure of hot tearing susceptibility of an alloy [24]. The composition dependence of hot cracking susceptibility has been reported for three series of wrought aluminum alloys, namely the AA2XXX (Al–Cu–Mg), AA6XXX (Al–Mg–Si), and AA7XXX (Al–Zn–Mg) series. For other alloys only few data are available. The probable reason is that the hot tear is found mostly in these three alloying systems. In commercial alloys, hot cracking susceptibility is not always related to a characteristic value but is rather represented comparatively, showing the effect of various additions on a certain alloy. Speciic data can be found elsewhere [3, 15]. Here we mention only general tendencies. Al–Cu–Mg–Si. Many commercial alloys belong to this system, e.g., AA2038, AA2017, AA2025, and AA2014. An increase in silicon concentration causes a shift of maximum hot cracking to lower copper and magnesium concentrations and a decrease in the overall hot cracking susceptibility [3]. The addition of iron (up to 0.65%) to an AA2038 alloy increases the hot cracking susceptibility. This is subject to the silicon concentration. At 0.95% Si the

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

218 7 6

Crack length

5 4 3 2 1 0 0

1

(a)

2

3

4

5

Cu, wt.%

Crack length

3

2

1

0 0

2

(b)

4

6

8

10

Mg, wt.%

FIGURE 5.21 Compositional dependence of hot tearing susceptibility in binary Al–Cu (a) and Al–Mg (b) alloys [72].

hot cracking index is very low even at high Fe concentrations. The addition of manganese to an AA2038 alloy initially reduces the hot cracking susceptibility, but at concentrations above 0.5% Mn the susceptibility increases. The maximum HCS an AA2017 alloy corresponds to about 0.2% Si. Al–Cu–Mg–(Ni) –Ti. Several commercial alloys belong to this system, e.g., AA2024 and AA2618 alloys. Addition of Ti signiicantly reduces the hot cracking susceptibility, especially at low Fe concentrations, as illustrated in Figure 5.22. The signiicant effect of increasing magnesium content in Al– Cu–Mg alloys is in the shifting of the maximum of hot cracking from 0.5% Cu at 0.5% Mg to 1.5% Cu at 4% Mg [3].

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

HCS, %

219 100 90 80 70 60 50 40 30 20 10 0

2% Mg 0% Ti

0.2% Ti

0

2

4 Cu, %

HCS, %

(a) 100 90 80 70 60 50 40 30 20 10 0

0% Ti

0.2% Ti

0

HCS, %

8

4% Mg

1

2

3

4

5

6

Cu, %

(b) 100 90 80 70 60 50 40 30 20 10 0

0.2% Fe

0% Ti

0.2% Ti 0

0.1

0.2

0.3

0.4

0.5

0.6

Si, %

(c)

HCS, %

6

100 90 80 70 60 50 40 30 20 10 0

0.3% Fe 0% Ti

0.2% Ti 0

0.1

(d)

0.2

0.3

0.4

0.5

0.6

Si, %

FIGURE 5.22 Effects of composition and grain rei nement on hot tearing susceptibility in Al–Cu–Mg (a,b) and Al–4.5% Cu–1.5% Mg–0.6% Mn (b,c) alloys [3].

Copyright 2008 by Taylor and Francis Group, LLC

220

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Al– Mg–Si– Fe–Ti. AA6XXX series alloys such as AA6060, AA6005, and AA6063 belong to this system. The maximum hot cracking susceptibility is at 0.3–0.4% Fe for an alloy containing 0.5% Mg, 0.5% Si, and 0.15% Ti [3]. Al– Mg–Si–Cu– Fe–Ti. This system includes such commercial alloys as AA6151, AA6351, and AA6111. The addition of Ti has been proved to decrease the hot cracking susceptibility of such alloys. The minimum HCS of an AA6111-type alloy (0.5% Mg, 1.2% Si, and 0.2% Fe) is at 0.7–0.8% Cu [3]. These data correlate well to our measurements of thermal contraction, shown in Figure 5.7. Al–Mg–Zn–Cu. High-strength commercial alloys, e.g., AA7039, AA7079, and AA7008, belong to this system. The hot cracking diagrams of Al–Zn–Mg alloys demonstrate that the addition of copper increases hot cracking susceptibility, with the maximum HCS shifting to lower Zn concentrations upon increasing the copper content. For example, alloys without Cu have a peak at about 1% Mg and 7% Zn, and alloys with 0.5% Cu show two peaks: at about 1.5% Mg and 4% Zn, and at about 1.5% Mg and 7–10% Zn [3]. The addition of manganese typical of AA7039, AA7079, AA7075, and AA7050 alloys is quite harmful, especially in high-copper and low-zinc alloys [3]. Warrington and McCartney [73] compared the hot tearing susceptibility of AA7050 and AA7010 alloys. The effect of grain-rei ning addition on both alloys was also reported. They conclude that AA7050 is marginally more sensitive for hot cracking than AA7010. They argue that the difference can be attributed to a larger volume fraction of iron-rich intermetallic phases in the AA7050 alloy. A grain rei ner can signiicantly affect the hot cracking susceptibility. The effect depends on the alloy, with the hot tearing being minimum at 0.005% and 0.3% Ti for AA7010 and AA 7050, respectively. Considerable effort has been made to understand the hot tearing phenomenon. Several mechanisms of hot tearing and conditions that may lead to it have been suggested in the literature [7, 9, 14, 22, 28, 36, 46, 63, 74–93]. Some of these mechanisms are outlined in Table 5.3 and summarized in Figure 5.23 [94]. We can see that, over the years, much more effort has been made to investigate and understand the conditions required for hot tearing occurrence rather than the mechanisms of crack initiation and propagation. And when it comes to the nucleation and propagation of hot tears, an educated guess frequently replaces experimental proof. Figure 5.23 also shows that the conditions for and causes of hot tearing can be considered on different length scales, from macroscopic to microscopic, and some of these conditions are important on both mesoscopic and microscopic scales. It is important to note that most of the existing hot tearing criteria deal also with the conditions rather than with the mechanisms of hot tearing. Different macroscopic and mesoscopic parameters, such as stress and strain, were considered critical for the development of hot tearing.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

221

TABLE 5.3 Summary of Hot Tearing Mechanisms [94] Mechanisms and Conditions Cause of hot tearing Thermal contraction Liquid ilm distribution Liquid pressure drop Vacancy supersaturation Nucleation Liquid ilm or pore as stress concentrator Oxide bi-ilm entrained in the mush Vacancy clusters at a grain boundary or solid–liquid interface Propagation Through liquid ilm by sliding By liquid ilm rupture By liquid metal embrittlement Through liquid ilm or solid phase depending on the temperature range Diffusion of vacancies from the solid to the crack Conditions Thermal strain cannot be accommodated by liquid low and mush ductility Pressure drop over the mush reaches a critical value for cavity nucleation Strain rate reaches a critical value that cannot be compensated for liquid feeding and mush ductility Thermal stress exceeds rupture or local critical stress Stresses and insuficient feeding in the vulnerable temperature range Thermal stress exceeds rupture stress of the liquid ilm

Suggested and Developed by*

Ref.**

Heine (1935); Pellini (1952); Dobatkin (1948) Verö (1936) Prokhorov (1962); Niyama (1977) Fredriksson et al. (2005)

74, 9, 28 36 75, 76 77

Patterson et al. (1953, 1967); Niyama (1977); Rappaz et al. (1999); Braccini et al. (2000); Suyitno et al. (2002) Campbell (1991)

78, 79, 76, 80, 81, 82 7

Fredriksson et al. (2005)

77

Patterson (1953); Williams and Singer (1960, 1966); Novikov and Novik (1963) Pellini (1952); Patterson (1953); Saveiko (1961); Dickhaus (1994) Novikov (1966); Sigworth (1996) Guven and Hunt (1988)

78, 83, 84

Fredriksson et al. (2005)

77

Pellini (1952); Prokhorov (1962); Novikov (1966); Magnin et al. (1996)

9, 75, 3, 87

Niyama (1977); Guven and Hunt (1988); Rappaz et al. (1999); Farup and Mo (2000) Pellini (1952); Prokhorov (1962); Rappaz et al. (1999); Braccini et al. (2000)

76, 86, 80, 88

Lees (1946); Langlais and Grizleski, (2000); Lahaie and Bauchard (2001); Suyitno et al. (2002) Bochvar (1942); Lees (1946); Pumphrey and Lyons (1948); Clyne and Davies (1975); Feurer (1977); Katgerman (1982) Saveiko (1961)

89, 90, 63, 82

9, 78, 85, 46 3, 14 86

9, 75, 80, 81

22, 89, 72, 91, 92, 93 85

* The list of the authors is by no means complete. The references have been chosen to represent the development of ideas. ** References are given in the same order as the authors in the second column.

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

222

Hot tearing Conditions and causes

Mechanisms Nucleation Liquid film or pore; Oxide bi-films; Inclusions; Vacancy clusters

Propagation Liquid fim rupture; Liquid metal embrittlement of solid bridges; Through liquid film; Vacancy diffusion

Macroscopic

Nonuniform thermal contraction

Mesoscopic Thermal stress; Insufficient feeding; Critical pressure drop over the mush; Thermal strain; Strain rate; Liquid film distribution

Microscopic

Thermal stress; Local pressure drop; Vacancy supersaturation

FIGURE 5.23 Mechanisms and conditions of hot tearing suggested to date.

Pellini [9] suggested a hot tearing theory based on the strain accumulation with the following main features: (1) cracking occurs in a hot spot region, (2) hot tearing is a strain-controlled phenomenon that occurs if the accumulated strain of the hot spot reaches a certain critical value, and (3) the strain accumulated at the hot spot depends on the strain rate and time required for a sample to pass through a ilm stage. The most important factor of hot tearing based on this theory is the total strain on the hot spot region. The total strain is the additive of strain over a period within which the hot spot exists. Taking into account that the highest strain accumulates in the liquid ilm, Metz and Flemings [10] explain the increase in hot tearing caused by the segregation of a low-melting component from the viewpoint that this addition increases the time of the liquid ilm existence. Although Pellini mentions a critical value for the accumulated strain, it is not clear whether it is ductility or another entity. The Pellini theory is a basis for a hot tearing criterion proposed by Clyne and Davies, who paid more attention to the time spent by an alloy in the mushy zone [6, 91]. Novikov and Novik [84] have reported that at low strain rates the grain boundary sliding is the main mechanism of deformation of a semi-solid body. The load applied to the semi-solid body will be accommodated by a grain boundary displacement that is lubricated by liquid i lm surrounding the grain. Prokhorov [40, 75] proposed a model for deformation of the semi-solid body. If two tangential forces τ 1 and τ 2 are applied to the equilibrium semi-solid body, the response of the body manifests itself as the grain movement and at some point the grains will touch one another. The liquid covering the grain will circulate to the lowest pressure point. Further deformation will be possible if the surface tension and resistance to liquid low are suficient to accommodate the stress imposed. If not, a brittle intergranular fracture or hot tearing will occur. In relation to this theory, Prokhorov postulated that (1) an increase in ilm thickness increases the fracture strain; (2) a decrease in grain size increases the fracture strain; and (3) any non-uniformity of grain size decreases the fracture strain. Based on this theory, the main measure for hot tearing is

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

223

the ductility of the semi-solid body. A hot tear will occur if the strain of the body exceeds its ductility. Today, the mesoscopic strain rate is believed to be the most important factor, and some modern models are based on it. The physical explanation of this approach is that semi-solid material during solidiication can accommodate the imposed thermal strain by plastic deformation, diffusion-aided creep, structure re-arrangement, and illing of the gaps and pores with the liquid. All these processes require some time, and the lack of time will result in fracture. Therefore, there exists a maximum strain rate that the semi-solid material can endure without fracture during solidiication. Prokhorov [75] was the irst to suggest a criterion based on this approach. More recently, an elaborate, strain-rate-based hot tearing criterion was proposed by Rappaz et al. [80]. A theory of shrinkage-related brittleness divides the solidiication range into two parts. In the upper part the coherent solid-phase network does not exist. Cracks or defects occurring at this stage can be healed by liquid low. As the solidiication progresses and the solid fraction further increases, at a certain stage or a certain solid fraction a coherent network is formed. This stage is considered to be the start of linear shrinkage. From this moment, the shrinkage stress is imposed onto the semi-solid body. Fracture or hot tearing occurs if the shrinkage stress exceeds the rupture stress [3]. On the mesoscopic and microscopic levels, the important factor is believed to be the feeding of the solid phase with the liquid. In this approach, the hot tear will not occur as long as there is no lack of feeding during solidiication. Niyama [76] and Feurer [92] use hindered feeding as a base for their porosity and hot tearing criteria. The feeding depends on the permeability of the mush, which is largely determined by the structure. Later, a two-phase model of the semi-solid dendritic network [88], which focuses on the pressure depression in the mushy zone, was suggested to describe the hot tear formation. This approach treats the semi-solid material as solid and liquid phases with different levels of stress and strain. The pressure drop of the liquid phase in the mush is considered to be a cause of hot tear. An extension of the two-phase model that includes plasticity of the porous network was also reported [81]. These models do not distinguish between the pore formation and the crack initiation, the void being considered as the crack nucleus. The pores should not, however, necessarily develop into a crack. The crack may nucleate or develop from another defect and then propagate through a chain of pores. Logically, bridging and grain coalescence, which determine the transfer of stress and limit the permeability of the mush, are the other important microscopic factors in the development of hot tearing [95]. The hot tearing theories that operate with macroscopic mechanical behavior or mesoscopic and microscopic phenomena such as feeding and porosity formation do not take into account the mechanism of crack nucleation and propagation and, thus, are intrinsically weak. The other

Copyright 2008 by Taylor and Francis Group, LLC

224

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

approach to the description of the hot tearing phenomenon is the application of fracture mechanics that describes the initiation and propagation of cracks. The liquid ilm surrounding the grain at late stages of solidiication is considered as a stress concentrator of the semi-solid body [78, 83]. In this theory, a liquid-illed cavity acts as a crack initiator. The propagation of the crack is determined by the critical stress [83]. The critical stress can be estimated using the Grifith energy balance approach modiied by taking into account the plasticity. Recently, a theory of hot tearing as a phenomenon related to microporosity has been proposed [96]. In this approach the porosity and the hot tearing are considered sequential events. As a result, there is a possibility of predicting simultaneously the occurrence of microporosity and hot tearing. The model uses the feeding dificulties at the last stage of solidiication as a starting point of cavity nucleation. The nucleus then grows and becomes at the end of solidiication either a micro-pore or a hot tear as determined by the Grifith model for brittle crack growth. The fracture-mechanical concept of hot tearing can be enriched with the application of a liquid-metal-embrittlement mechanism [14]. It is obvious that the actual hot tearing mechanism includes phenomena occurring on two scales: microscopic (crack nucleation and propagation, stress concentration, structure coherency, wet grain boundaries) and mesomacroscopic (lack of feeding, stress, strain, or strain rate imposed on the structure). Figure 5.24 illustrates these scales during equiaxed dendritic solidiication.

Melt flow to compensate solidification shrinkage and thermal contraction

Free flow, no hot tears Slurry

Feeding paths

Healed cracks Coherency Bridging

Liquid film rupture Mush

Wet grain boundaries

Plastic deformation of bridges Liquid metal embrittlement High-temperature creep

Soild Macroscopic stress Eutectics FIGURE 5.24 Different length scales of equiaxed solidiication showing hot tearing mechanisms [94].

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing 5.3.2

225

Hot Tearing Criteria

Hot tearing criteria can be conditionally divided into two categories: mechanical and nonmechanical. The former type of criteria involves critical stress [46, 63, 83, 90, 96], critical strain [3, 87], or critical strain rate [75, 76, 80, 81, 88]. The latter deals with vulnerable temperature range, phase diagram, and process parameters and is represented by the criteria of Clyne and Davies [6], Feurer [92], and Katgerman [93]. Let us consider these criteria in more detail. 5.3.2.1

Stress-Based Criteria

The stress-based criteria of hot tearing consider that a semi-solid body will fracture if the applied or induced stress exceeds the strength of the body. The irst type of these hot tearing criteria takes into account that a material has a stress limit at which it fails. This approach can be subdivided into two sets of criteria. The irst set is based on the strength of a liquid ilm trapped between grain boundaries and the second set is based on the strength of bulk material. Within the second type of stress-based criteria, the hot tearing susceptibility is based on the assumption that the material contains defects or weak points, and whether they become a fracture or not depends on such parameters as stress, defect dimensions, coniguration of a casting, and the like. This approach is more comprehensive in describing the hot tearing phenomena. The effects of some parameters involved in hot tearing can be incorporated in these criteria, for example, grain diameter, viscosity, and liquid fraction. Although it is rather dificult to obtain certain material property that is comparable to the fracture toughness of solid materials, the idea of applying fracture mechanics theory is interesting, particularly within the framework of micromechanics. A stress-based hot tearing criterion that uses the strength of liquid trapped between grain boundaries was reported by Saveiko [85], Novikov [3], and Dickhaus et al. [46]. They refer to the stress needed to pull apart two parallel plates separated by a thin liquid ilm as the strength of semi-solid metals. The assumptions that are applied in this model are the uniform distribution of liquid and no inluence of sliding on the fracture strength. This criterion is expressed as: σfr = 2γ/b,

(5.6)

where σfr is the fracture stress, γ is the surface tension, and b is the ilm thickness. The constraints of this model are the negligible viscosity and wetting angle. To overcome this constraint, Dickhaus et al. [46] promoted a model proposed by A. Healy in 1926 that involved the viscosity. The separation force of the

Copyright 2008 by Taylor and Francis Group, LLC

226

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

parallel plates of radius R separated by a liquid ilm of thickness b is given as follows: 3πµR4 1 1 FZ = ______ __ − __ , 8t b21 b22

(

)

(5.7)

where FZ is the force required to increase the ilm thickness from b1 to b2, µ is the dynamic viscosity, R is the radius of a plate, and t is the time required to increase the ilm thickness from b1 to b2. The equation for the ilm thickness can be written as: (1 − fs ) Dgr (5.8) b = __________, 2 where fs is the fraction of solid and Dgr is the average thickness of a solidifying grain. Lahaie and Bouchard [63] proposed a hot tearing model that was an extension of Equation 5.7. This model involves solid fraction fs, strain ε and microstructure parameter m and is written as: fsm 4µ ε σfr = ___ 1 + _______ 3b 1 − fsm

( (

))

−1

.

(5.9)

The equation can be applied to equiaxed and columnar grain structures by adjusting the microstructure parameter m, which is 1/3 for equiaxed and 1/2 for columnar structure. The fracture stress calculated using this model decreases with grain coarsening and sharply increases at a high volume fraction of solid, which agrees well with the experimental data (see Section 5.2). The calculated fracture strain, however, decreases linearly with increasing solid fraction and shows no response to the grain size variation. This trend is totally different from the experimental data that show a U-shaped dependence of the ductility on the solid fraction (see Section 5.2). The fracture stress calculated using this criterion could explain the low tendency of tear for a ine-grained structure. However, the predicted negligible effect of the grain diameter on the fracture strain is questionable. Langlais and Gruzleski [90] dei ned the hot tearing susceptibility based on the bulk strength of the semi-solid material. The susceptibility is expressed as the inverse of the maximum tensile stress. They reported the result of this criterion for Al–Si alloys. Using this criterion a typical lambda curve is reproduced. However, the silicon concentration corresponding to the maximum point of the lambda curve is different from some experimental data [15]. The stress-based hot tearing criteria that apply a fracture mechanics approach was proposed by Williams and Singer [83]. They modify Grifith cracking criteria for application as a hot tearing criterion. The original Grifith criterion considers a defect or a small crack as a stress concentrator and, therefore, the initiator of fracture. On transition to hot tearing phenomenon, the volume of liquid in the inal stage of solidiication substitutes the defect as the crack initiator in the Grifith equation. Williams and Singer [83]

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

227

argue that the inal liquid is the weakest point and the stress concentrator of the semi-solid body. The modiied Grifith model is written as: _____________

σfr =



8Gγ ____________ , π(1 − ν)AV½L

(5.10)

where σfr is the fracture stress; G is the shear modulus; γ is the effective fracture surface energy; A is a constant dependent on the grain size and the dihedral angle; VL is the volume of liquid; and ν is Poisson’s ratio. This equation has been also modiied for the contribution of grain size and grain boundary sliding that facilitates the liquid crack growth. This model predicts well the variation of fracture stress with alloy composition but fails to adequately describe the effect of grain size, predicting higher strength for coarser structures. Recently, a formulation of hot tearing as a phenomenon related to microporosity is proposed [82]. In this approach both phenomena are considered in sequence. When the coherency solid fraction is reached, liquid becomes increasingly isolated in separate locations of the mush and, eventually, as a result of solidiication shrinkage, thermal contraction and low permeability cavities may form at triple junctions of grain boundaries. Three stages, each possibly the inal stage, are possible. First, the solidiication shrinkage and thermal contraction are completely compensated for by liquid low, shrinkage low and, later, high-temperature creep. As a result, potential cavities are illed and a fully dense structure is formed with no porosity. In the second stage, feeding possibilities are limited and the cavities develop into pores. The interplay between feeding melt low and shrinkage/contraction determines the transition from the irst to the second stage. The feeding rate is deined as Ps fe = K____ , 2 µLm

(5.11)

where Ps is the effective feeding pressure, K is the permeability as deined by Equation 4.6, and Lm is the length of the mushy zone. The shrinkage/contraction rate is as follows: ρs · ρs ∂f1 __ ___ fr = __ ρ1 − 1 ∂t + ( ρ1 )ε,

(

)

(5.12)

where ρs and ρl are the densities of the solid and liquid phases, respectively, f l is the liquid fraction, and ε· is the strain rate. The pore is formed if fr > |fe|

(5.13)

fr − |fe| ≥ ζcr,

(5.14)

and will grow if

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

228

where ζcr is the critical feeding rate that has the same form as Equation 5.11 only with Pcr substituting for Ps. This critical feeding rate is rather small and can be taken as zero in practical simulations. The third stage is when the cavity reaches a critical size suficient for the development of a crack under particular thermo-mechanical conditions. The Grifith criterion of brittle fracture is used in this case: a > acr,

(5.15)

where a = Cd with C is the shape factor and d is the diameter of a spherical pore, and the critical defect size is deined as follows: E acr = 4γ ____ πσ ,

(5.16)

fr

where E is Young’s modulus and γ is the surface tension of the liquid phase. Hot cracking susceptibility is deined as HCS = a/acr .

(5.17)

As a result, it is possible to predict simultaneously the occurrence of microporosity and hot tearing, as shown in Figure 5.25a. This model also reacts adequately to the change in the cooling rate and in the grain size. The latter is illustrated in Figure 5.25b [97]. We can see that grain reinement increases the critical stress, and that below a certain grain size no hot cracking is possible.

10−02 10−03

2.0 1.8

C B A

10−05 10−06 10−07

10−08 0.70 0.75 0.80 0.85 0.90 0.95 1.00 fs

, MPa

., s−1

10−04

1.6 1.4 1.2

4

1.0 0.8 0.6 0.4 0.2 0.0 0.80

3 2 1 1′,2′,3′,4′ 0.85

0.90

0.95

1.00

fs

FIGURE 5.25 (a) Three typical regions that can be predicted by a model by Suyitno et al. [82] with no porosity (A), only porosity (B), and hot cracks (C); and (b) sensitivity of this model to the grain size: 1’, 1, 600 µm; 2’, 2, 300 µm; 3’, 3, 100 µm; and 4’, 4, 30 µm (1’–4’ show the stress accumulated in the mush while 1–4 show the critical stress) [97].

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing 5.3.2.2

229

Strain-Based Criteria

Drawing on his extensive studies of mechanical properties and thermal contraction in the solidiication range, Novikov suggested a hot shortness criterion based on the ductility of semi-solid nonferrous (including aluminum) alloys [3] (see also Sections 5.1 and 5.2). He proposed a characteristic called a “reserve of plasticity in the solidiication range” pr. The “reserve of plasticity” pr is actually the difference between the average integrated value of the elongation to failure (εp) and the linear thermal contraction (called linear shrinkage by Novikov) (εth) in the brittle (or effective, or vulnerable) temperature range (∆Tbr) (Figure 5.26a). The hot tearing susceptibility is then given by: S pr = ____, ∆Tbr

(5.18)

where S is the area between the εp and εth curves in the brittle temperature range ∆Tbr. If the εth curve intersects the εp curve (see Figure 5.26b,c) then: S1 − S2 pr = _______ ∆Tbr

(5.19)

Of course, the situation when the semi-solid ductility drops below the accumulated thermal strain (contraction) is speciically dangerous in terms of hot tearing, especially when the temperature range of such brittleness is high as in Figure 5.26c. Magnin et al. [87] used a similar model to predict hot tearing in a DCcast billet of an Al–4.5% Cu alloy. They suggest that cracking occurs when the maximum principle plastic strain at the solidus temperature exceeds the experimentally determined fracture strain at a temperature close to the solidus. It has been demonstrated that hot cracking is more likely to occur at higher casting speeds, which is in line with industrial experience. A rather sophisticated strain-based criterion was suggested recently based on a two-phase model of hot tearing [98]. In this criterion, the development of porosity under conditions of insuficient feeding and thermalcontraction-induced deformation is considered a potential trigger for hot tearing. However, no hot cracking occurs if (1) the liquid pressure onto the mushy zone is suficient for feeding (compensating) the solidiication shrinkage and, hence no porosity is formed or (2) the liquid pressure is below the critical value and pores do form but do not reach the critical size, or (3) the porosity starts to form at a very high fraction of solid when there is already no continuous liquid ilm existing between the grains and the mush is suficiently strong to withstand or suficiently ductile to accommodate thermal stresses and strains, respectively. Thus the effective tearing strain is a measure of hot tearing possibility in the vulnerable temperature range that is limited by the rigidity solid fraction at the upper end and by the “no-fracture” solid fraction at the lower end. It is important that the effective tearing strain

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

230 2

 p = El

El,  th, %

1.6 1.2 ∆Tbr

0.8 0.4

S

εth 0 500

520

540

560

580

600

620

600

620

600

620

T, °C

(a) 2

 p = El

El,  th, %

1.6 1.2 ∆Tbr

0.8 0.4

S2

εth

0 500

520

S1

540

560

580

T, °C

(b) 2

 p = El

El,  th, %

1.6 1.2 ∆Tbr

0.8 0.4

S1

εth 0 500

S2 520

(c)

540

560

580

T, °C

FIGURE 5.26 Illustration of Novikov’s strain-based hot tearing criterion using experimentally measured elongation (El, εp) and thermal contraction (εth) in the solidiication range of Al–5% Cu–2.65% Li (a), Al–3.8% Cu–0.9% Li–0.45% Mn (b), and Al–3.8% Cu–0.9% Li–1.4% Mn (c) alloys [3].

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

231

is the strain accumulated in the vulnerable section of the mushy zone, and its value depends on the structure parameters and mush evolution, number of pores, and distribution of liquid phase [98]. The critical pore size that may develop to a hot crack is several times larger than the average grain size in a grain-reined material [98]. This criterion reacts adequately to the grain reinement and casting practice. One of the consequences of the criterion is that a large number of distributed pores may effectively decrease hot tearing susceptibility by relaxing applied deformation. In concept this criterion is close to the stress-based criterion of Suyitno et al. [82, 97], but it does not employ fracture mechanics for crack propagation. 5.3.2.3

Strain Rate-Based Criteria

During solidiication, alloys are in a two-phase state: liquid and solid phases are present though in a constantly changing proportion. The liquid–solid phase transformation leads to the solidiication shrinkage and the solid phase then contracts, resulting in the strain development affected by geometrical coniguration of the solidifying body. The more complex the shape of the casting and corresponding thermal gradients, the higher the strain. Based on this fact, hot tearing criteria that are based merely on the comparison of the ductility of semi-solid alloys and the accumulated thermal strain (thermal contraction) cannot be used for hot tearing prediction of complex castings. Prokhorov [75] proposed a hot tearing model that includes the coniguration of a solidifying body. During solidiication, a solidifying alloy passes through a low-ductility range that is called the brittle temperature range (∆Tbr) (see Figure 5.26). The highest temperature of the brittle temperature range is known as the upper limit of ∆Tbr (Bupper), and the lowest one is known as the lower limit of ∆Tbr (Blower). Figure 5.27 shows a diagram illustrating the concept. It is very similar to the previously discussed approach of Novikov.* However, Prokhorov focuses not on the thermal strain but on the thermal strain rate, considering the latter to be a determining parameter. The strain relecting the appearance of a hot tear is deined as the limitation of strain capacity in the semi-solid alloy, as indicated by the intersection of the strain curve (ε) and the ductility curve (El). The intersection is determined by the slope of the strain curve (strain rate), the ∆Tbr span, the shape of the El curve within the brittle range ∆Tbr, and the minimum value of elongation Elmin. Because the strain in the solidifying body is determined by the thermal contraction and the geometrical coniguration (thermal gradient), the strain under balance conditions can be written as follows: εapp = εfree + εint ,

(5.20)

where εapp is the actual (apparent) strain in the solidifying body; εfree is the free thermal contraction strain; and εint is the internal strain from the * Novikov considered his criterion a simpliied version of Prokhorov’s criterion [3].

Copyright 2008 by Taylor and Francis Group, LLC

232

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Elongation, Thermal strain

Blower

Bupper

∆Tbr

EI

EImin

∆εres ∆εapp ∆εfree

Solidus

Liquidus

Temperature FIGURE 5.27 A diagram illustrating the parameters of Prokhorov’s hot tearing criterion.

restricted shrinkage and thermal stresses as a result of the coniguration of the solidifying body. The reserve of technological strain in the semi-solid state (∆εres) can be approximately written in the following form: ∆εres = Elmin − (∆εfree + ∆εapp),

(5.21)

with the symbols explained in Figure 5.27. The expression is divided by ∆Tbr and written as: Elmin − (∆εfree + ∆εapp) ∆εres ____________________ _____ . = ∆Tbr ∆Tbr

(5.22)

Taking into account that ∆εres ∙ ε· = _____T, ∆Tbr

(5.23)

· where T is the cooling rate, one can write Equation (5.22) in terms of strain rate: ε·res = ε·min − ε·free − ε·app.

(5.24)

Prokhorov [75] postulated that hot tearing occurs when ε·res ≤ 0, and therefore when the total strain rates of free contraction and apparent strain exceed

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

233

the strain rate that can be accommodated by the ductility of the semi-solid material: ε·min ≤ ε·free + ε·app .

(5.25)

The strain rate in Equation 5.25 varies with the solidiied body coniguration or the design factor. The left-hand side of the equation includes factors that depend on the alloy composition, and it is the hot tearing criterion. The value of ε·app is considered a function of the solidiied body coniguration. The higher the difference (ε·min − ε·free ) at which the hot cracking occurs, the worse the solidifying body coniguration in relation to hot tearing. As already mentioned, this criterion is similar in principle to the criterion proposed by Novikov [3]. The major differences are (1) Prokhorov’s criterion uses the strain rate as a measure for hot tearing tendency, while Novikov’s criterion uses the strain; (2) Prokhorov considers both the coniguration of the solidifying body and the thermal contraction as factors of hot tearing, while Novikov’s strain criterion uses only the thermal contraction; and (3) Prokhorov uses the difference between the lowest ductility of the semi-solid body and the actual strain as the reserve plastic strain, while Novikov’s criterion uses the area between the ductility curve and the strain curve in the entire effective solidiication range. The most recent hot tearing criterion based on strain rate, derived by Rappaz et al. [80], is widely known as the RDG criterion from the names of the three co-authors of the original paper. This criterion is based on the critical pressure drop over the mushy zone that results in the formation of a void that then necessarily develops into a crack. The reason for the void formation is primarily a pressure drop over the mushy zone that can be written as: ∆P = ∆Psh + ∆Pmech − ρgh,

(5.26)

where ∆Psh is the pressure drop due to the solidiication shrinkage, ∆Pmech is the pressure drop due to the deformation-induced luid low, ρ is the liquid density, h is the melt level, and g is the gravitation acceleration. The pressure drop is given by the following expression: 180 µ∆T (1 + β)ε·∆T __________ ∆Psh + ∆Pmech = ________ AV β + B , cast Gλ22 G

[

]

(5.27)

where G is the thermal gradient, Vcast is the casting speed, λ2 is the secondary dendrite arm spacing, β is the volumetric solidiication shrinkage, ε·is the viscoplastic strain rate, µ is liquid dynamic viscosity, and A and B are the parameters accounting for the solidiication path, coherency development, and mass feeding.

Copyright 2008 by Taylor and Francis Group, LLC

234

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

If the pressure locally drops below the critical value ∆Pcr (taken, for example, as 2 kPa for an Al–1.4% Cu alloy [80]), a pore is formed. In this case, the hot tearing criterion appears as ∆P > ∆Pcr.

(5.28)

On the other hand, the formation of a void or pore occurs only if the mush cannot accommodate the imposed strain at a given strain rate. Hence, there exists a critical strain rate at which cavitation starts. The appearance of the irst pore is considered a hot tear. This model emphasizes the nucleation rather than the propagation of a hot crack. The hot cracking sensitivity (HCS) is then described as: 1 HCS = ____ , ε·max

(5.29)

where ε·max is the maximum deformation rate sustainable by the mushy zone. Grandield et al. [71] extended the RDG model to equiaxed microstructures. Braccini et al. [81] complemented the model of Rappaz et al. [80] with the rheological behavior of the mushy zone. The RDG criterion is intrinsically weak because it does not take into account the accumulation of thermal strain over the entire coherent mushy zone [88]. 5.3.2.4

Criteria Based on Other Principles

A well-known and widely used hot tearing criterion proposed by Clyne and Davies [6] is based on the idea that in the last stage of freezing, the liquid cannot move freely so that the strain applied during this stage cannot be accommodated by mass feeding. This last stage of solidiication is considered the most susceptible to hot tearing. On further decreasing the liquid fraction, however, bridging between adjacent dendrites is established so that interdendritic separation is prevented and the mush acquires some strength, e.g., at solid fractions above 0.99. The cracking susceptibility coeficient proposed by Clyne and Davies is formulated as the ratio between the vulnerable time period (hot tearing susceptibility in the range between solid fraction 0.9 and 0.99) tV and time available for stress relief (mass feeding and liquid feeding in the range of solid fractions between coherency, e.g., 0.4, and rigidity, e.g., 0.9) tR: t0.99 − t0.90 tV _________ HCS = __ t = t −t , R

0.90

0.40

(5.30)

where t is the time at the solid fraction denoted by indices. If the time available for feeding the solidifying solid phase is great and the time spent by an alloy under conditions of restricted feeding is short, then the alloy is less susceptible to hot tearing. This criterion works well in predicting the inluence of composition on hot cracking, i.e., reproducing the lambda curve in Figure 5.21.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

235

Adequate feeding of the forming solid phase with the liquid is essential for the continuity of the solid phase and, hence, plays an important role in hot tearing phenomenon. Feeding-based criteria consider that hot tearing occurs due to the lack of feeding, which is related to the dificulties of luid low through the mushy zone as a permeable media. Let us consider the hot tearing criteria suggested by Niyama [76] and Feurer [92]. In the Niyama derivation, a pore is nucleated if a critical strain rate is achieved at which the liquid cannot ill the intergrain gap formed by the deformation. In this model, the formation of a cavity (gap) can be considered the mechanism of hot tearing. In fact, this model is very similar to the model that Rappaz et al. [80] used for their hot tearing criterion. Feurer’s model focuses on the inluence of alloy composition and solidiication conditions on the dendrite arm spacing, feeding shrinkage, and hot tearing properties of aluminum alloys. In this approach hot tearing occurs due to the lack of feeding that is a result of competition of accumulated solidiication shrinkage and decreasing permeability. Feurer proposes two parameters, SPV and SRG. SPV is the maximum volumetric low rate per unit volume (feeding term). SRG is the rate of volumetric solidiication shrinkage. SPV is formulated as follows: f12λ22Ps SPV = _________ 24πc3ηL2m

(5.31)

where f l is the liquid volume fraction, Ps is the effective feeding pressure, Lm is the length of porous network, and c is the tortuosity constant of dendrite network. SRG is given by the following expression: __

∂ρ ∂lnV 1 ___ SRG = _____ = −__ ρ ∂t ∂t

(5.32)

__

where ρ is the average density and V is the volume element with the constant mass in the mushy zone. SRG depends on the cooling rate and the alloy composition. The Feurer criterion says that hot tearing is possible if: SPV < SRG

(5.33)

The comparison of SPV and SRG in graphical form is shown in Figure 5.28. A model suggested by Katgerman [93] combines the assumptions of Clyne and Davies and Feurer. The model is speciically developed for hot tearing during DC casting of aluminum. In the model, the effects of casting speed, ingot diameter, and alloy composition are considered. The hot tearing index is deined as follows: t0.99 − tcr HCS = ________ t −t , cr

0.40

(5.34)

where t0.99 is the time when fs = 0.99 is reached, t0.40 is the time when fs = 0.40, and tcr is the time when the afterfeeding becomes inadequate. The time tcr is

Copyright 2008 by Taylor and Francis Group, LLC

236

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys 10−1

SPV

SPV and SRG, S−1

10−2

Sufficient-

Insufficient feeding

SRG

10−3

10−4 620

610

600

590

580

Temperature, °C FIGURE 5.28 Dependence of SPV and SRG parameters in Feurer’s criterion on temperature during solidiication (an Al–5% Si alloy, cooling rate 4 K/s) [92].

determined using Feurer’s criterion and relects the intersection of SPV and SRG curves in Figure 5.28. 5.3.3

Application of Hot Tearing Criteria to DC Casting of Aluminum Alloys

Different casting processes impose speciic requirements on the application of hot tearing criteria. That is why some criteria work better for shape casting whereas others are more suitable for DC casting. There is no doubt that a good hot tearing criterion for DC casting should correctly respond to the casting parameters, e.g., casting speed, ramping rate, and alloy composition, and predict the vulnerable section of a billet or an ingot, e.g., the center of a round billet. Most of the existing criteria have been tested for composition sensitivity by calculating the hot tearing susceptibility of several binary alloys with an attempt to reproduce the so-called lambda curve showing the maximum susceptibility at a certain composition. Most existing criteria can do this successfully. However, dynamic parameters such as casting speed

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

237

and strain rate are usually kept constant upon such testing. Therefore, the compositional sensitivity of a hot tearing criterion does not ensure its successful application to a particular casting technology. The basic phenomena that lead to hot cracking are well established and understood, but a generic criterion that will predict hot cracking under varying process conditions does not exist. Although the earlier simple criteria based on the thermal history of the casting have been extended and improved to include shrinkage and deformation, they are still unable to give reliable predictions under all process conditions. As we have shown in the previous section, most of the existing hot tearing criteria do not incorporate the nucleation and propagation of a hot tear, focusing instead more on the macro-, meso- and microscopic conditions that may result in rupture (see Figure 5.23). The ultimate hot cracking criterion should combine aspects of thermal history, shrinkage and porosity formation, and constitutive behavior together with the evolution of the semi-solid microstructure. Current research efforts are aimed, in particular, at the quantitative description of structure evolution (see Chapter 3) and its correlation with cracking [21, 82, 95, 96, 99]. Recently, several mechanical and nonmechanical hot tearing criteria were evaluated by implementing them into a thermo-mechanical model of DC casting [15, 96]. The criteria show different results in predicting hot tearing susceptibility, as shown in Table 5.4 and Figure 5.29. The criteria of Clyne and Davies and Novikov give results that are inconsistent with casting practice, and do not show any sensitivity to the casting speed and position within the billet volume. These criteria are, however, very successful in predicting the compositional dependence of hot tearing and are frequently used for shape casting. The criteria of Feurer, Katgerman, Magnin et al., Prokhorov, Rappaz et al., Braccini et al., and Suyitno et al. respond correctly to the casting parameters, demonstrating that the increasing casting speed results in an increasing hot tearing susceptibility TABLE 5.4 Sensitivity of Hot Tearing Criteria to the Casting Parameters and Practice upon DC Casting [94, 96]

Criterion Clyne and Davies [6] Katgerman [93] Feurer [92] Novikov [3] Magnin et al. [87] Prokhorov [75] Rappaz et al. [80] Braccini et al. [81] Suyitno [97]

Hot Tearing Increases with Casting Speed

More Hot Tears in the Billet Center

Ramping Casting Speed during Start-Up of the Casting Reduces Hot Tearing

Correlation with Actual Cracking Observed in Practice

No Yes Yes No Yes Yes Yes Yes Yes

No Yes Yes No No Yes Yes Yes Yes

No No No No No No Yes No Yes

N/A N/A N/A N/A No No No N/A Yes

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

238

CL

CL

CL

CL

CL

CL

CL

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

ur

Fe s

ie

av

D

v

.

al

.

al

an

m

&

er

ne

ly

C er

et

et

ro

ov

n

ho

tg

Ka

ok

Pr

ik

ni

z

pa

ag

ov

N

M

ap

R

100 200 Casting speed, mm/min

FIGURE 5.29 Comparison of some hot tearing criteria in the case study of DC casting of an Al–4.5% Cu alloy with the casting speed ramped up and down during the casting as shown in the right-hand part of the diagram [15, 96, 99]. In the left-hand part, a hot tearing susceptibility is shown in half of a 200-mm billet. CL = the centerline of the billet. Lighter gray shades correspond to a higher hot tearing probability.

in the center of billet, which is in accordance with casting practice (see Section 5.4). However, most of the tested criteria, except those by Rappaz et al. [80] and Suyitno et al. [82, 97], are not sensitive to the ramping of casting speed during the start-up phase of casting (which is a usual practice to prevent hot cracking). When confronted with casting practice, the criteria of Prokhorov, Magnin et al., and Rappaz et al. predict the occurrence of hot cracks whereas no cracks have been found in billets cast under given conditions. Only Suyitno et al.’s criterion adequately responds to all tested parameters, i.e., casting speed, ramping rate, grain size, position in a billet, and casting practice. This is illustrated in Figure 5.30 where the results of implementation of Suyitno et al.’s criterion in thermomechanical simulation of DC casting of an Al–4.5% Cu alloy is shown [97]. One can see that the hot tearing susceptibility decreases with grain reinement (Figure 5.30a), at a lower casting speed (Figure 5.30b) and upon slower speed ramping (Figure 5.30d), which agrees with casting practice (see Section 5.4). It is important to note that the values of HCS are well below unity, which means, according to this criterion (Equation 5.17), that no hot cracks are formed in the billet, which again agrees with experimental results presented in Section 5.4. The RDG criterion calculated in terms of the pressure drop (see Equation 5.28) shows in the same case studies qualitatively similar results [97], but the pressure drop under most

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

239

0.15 0.12

500 µm

0.06

HCS

HCS

0.09 300 µm

0.03 100 µm 0.00 0

20

(a)

40

60

80

100

Distance from center, mm

(b)

0.25

140

4 1 2

110

100

3

4

0.15 0.10 1

100

0.05

90 80

2

0.00 0

(c)

80

0.20

130 120

60

Distance from center, mm 0.30

3

HCS

Casting speed, mm/min

160 150

0.50 0.45 180 mm/min 0.40 0.35 0.30 150 mm/min 0.25 0.20 0.15 0.10 120 mm/min 0.05 0.00 0 20 40

0

50 100 150 200 250 300 350 400 450

Distance from bottom, mm

(d)

100 200 300 Distance from bottom, mm

400

FIGURE 5.30 Some results of implementation of Suyito et al.’s hot tearing criterion in thermomechanical simulation of DC casting of a 200-mm billet from an Al–4.5% Cu alloy: (a) effect of grain size on HCS in the radial direction; (b) effect of casting speed on HCS in the radial direction; (c) case studies on casting-speed ramping in the start-up phase of casting; and (d) corresponding distribution of HCS along the centerline of a billet [97].

conditions exceeds a critical value of 2 kPa. As a result the RDG criterion predicts the occurrence of hot tearing in situations when hot cracking does not occur in reality. The sensitivity of the criterion of Suyitno et al. is, however, a function of correctly chosen values of properties such as Young’s modulus of the mush, surface tension between liquid and solid, and permeability of the mush. These parameters are rarely available and must be determined experimentally, while the existing experimental techniques are not yet reliable. We can see that there are two problems. First, the application of a hot tearing criterion is not straightforward in DC casting. In practice, only those criteria that incorporate dynamic parameters such as strain rate or pressure build-up are, at least qualitatively, applicable to DC casting. Second, most of the existing criteria can forecast only the probability of hot tearing and not its actual occurrence. Hence, the quest for a new hot tearing criterion is high on the research agenda.

Copyright 2008 by Taylor and Francis Group, LLC

240 5.3.4

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys Quest for a New Hot Tearing Criterion

To date, hot tearing has been considered as a phenomenon linked to casting and welding processes, and the existing theories, models, and criteria are biased by their applicability to solidiication, e.g., by direct link to the solidiication shrinkage and limited feeding. Hot tearing is, however, just another example of material failure. Therefore, it should be treated adequately, using the welldeveloped methodology of fracture mechanics. The challenge nowadays lies not in the adequate description of macroscopic and microscopic stress–strain situations and their correspondence to the parameters and properties of the mushy zone, but rather in inding real factors causing the nucleation and propagation of hot cracks. In fact, some existing theories and models of hot tearing describe in part these factors with the crack initiator presented as a cavity illed with liquid or a pore [80, 82], or an oxide bi-ilm [7] and with the crack propagation path through the liquid ilm covering grain boundaries [14]. What is lacking is the completeness of the model. A comprehensive model and a corresponding hot tearing criterion should include nucleation and propagation of hot tears and connect these processes (1) to the microstructure evolution during solidiication of the semi-solid material; (2) to the macroscopic and microscopic thermo-mechanical situation in the mushy zone; and (3) to the mechanical (or fracture-mechanical) properties of the mushy zone. The last two components are well covered by a large body of publications, though many mechanical properties still need to be determined and the fracture mechanics potential has not been fully exploited. The correspondence between the structure evolution during solidiication and the crack nucleation and propagation is studied in much less detail. Let us now consider the possible mechanisms of crack nucleation and propagation (see also Table 5.3). The nucleation of hot cracks is a virtually unexplored phenomenon. It is obvious that under any stress-strain conditions, there should be a certain critical size of a defect (law, nucleus) that would enable crack growth. The problem of the crack initiation is today solved by an educated guess, as only recently experimental observations on the crack nucleation upon natural hot tearing have started to emerge, irst for transparent analogs [21] and then for real metals [100]. Usually, the development of a hot crack is studied on samples with a notch, hence with the artiicial crack initiator [95, 101]. Based on these observations and “post-mortem” examination of hot tear surfaces, the following crack nuclei have been suggested: (1) liquid ilm or liquid pool; (2) pore or series of pores; (3) grain boundary located in the place of stress concentration; and (4) inclusions that can be easily separated from the surrounding liquid or solid phase, e.g., intermetallic particle or oxide ilm. It should be noted that pores that are frequently cited as potential hot tear nuclei can originate from gas precipitation [7], solidiication shrinkage [7], or vacancy supersaturation [77]. Figure 5.31 summarizes some of the possible crack initiators. The mechanism of hot tear formation and propagation can be elucidated from observations of fractures. Unfortunately, the reports on hot crack fractures in metallic materials are rare. Most such reports describe the cracking of

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

241

Pore

Mesoscopic flaw Plastic deformation

Wet boundary

Low-melting phases

Local deformation Intergranular fracture/separation Microscopic flaw/oxide film

Transgranular fracture/cleavage

FIGURE 5.31 Possible initiators and propagation paths for hot cracking, as compiled in [94].

semi-solid alloys at relatively large fractions of liquid, when grain boundaries are completely covered with liquid. In this case the mechanism of crack propagation—through liquid ilm by grain separation—is obvious. An example of such alloys is the classic Al–4% Cu alloy. However, alloys with high fractions of liquid in the vulnerable solidiication range are in practice not susceptible to hot tearing [15, 24]. Cavities and gaps between grains, which may form in the mushy zone of such alloys due to solidiication shrinkage, presence of not wetted inclusions, thermal contraction, or external tension, are easily illed with liquid due to the adequate permeability of the mushy zone and the suficient amount of available liquid that is represented in the inal structure by nonequilibrium eutectics [24]. Much more important is the mechanism of crack formation and propagation in alloys containing little solute that are most susceptible to hot tearing. But the information on semi-solid fracture in such alloys is only starting to emerge. If we summarize the indings available to date, it would appear that the bridging of grain boundaries is an essential feature of the fracture surface. Moreover, the closer the semisolid material gets to the temperature range of its maximum vulnerability to hot cracking, i.e., 90–95% solid, the greater the fraction of grain boundaries connected to each other or coalesced [102]. In this case, there is no chance of crack propagation through a continuous liquid ilm because such a ilm does not exist. Recent observations [103, 104] show that a hot tear apparently propagates through the liquid ilm in more alloyed materials and through solid bridges in less alloyed materials. Although in both cases the fracture surface appears to be brittle, we can suggest that different cracking mechanisms are acting. In real castings, cracks appear both above and below the solidus. In the latter case, these cracks are called “cold.” There are reports, however,

Copyright 2008 by Taylor and Francis Group, LLC

242

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

that these cracks, appearing at high temperatures in the fully solid material, can originate in areas of local remelting caused by intense local deformation [105]. The liquid thus formed also assists in crack propagation. These observations make such “cold” cracks in fact similar to hot tears. Despite a large body of literature on hot tearing, little research has been done on the mechanism of hot tear propagation. We have already discussed the application of the Grifith criterion for brittle fracture [82, 97]. Another mechanism is vacancy-diffusion-controlled growth [77]. The methodology for the description of crack propagation is well developed within fracture mechanics for various situations, including those resembling hot cracking, i.e., liquid ilm rupture, pore coalescence, high-temperature creep, and liquid-assisted fracture [106]. It is clear that hot crack propagation in the potential presence of liquid (above or below the solidus) should involve the following aspects: liquid feeding (involves permeability and, inevitably, structure evolution), pore coalescence, stress transfer by solid bridges, plastic deformation and creep of solid bridges in the absence of liquid, and brittle fracture of solid bridges in the presence of liquid. It is also obvious that hot crack, like any other crack, can develop catastrophically (which is usually assumed), have sustained growth, or stop growing. Liquid feeding plays a dual role. Firstly, adequate feeding of the shrinking material with liquid does not eliminate the causes of hot tearing but rather “patches” the consequences, which is relected in the term “crack healing.” We can say that a semi-solid alloy containing enough liquid at the last stage of solidiication, having a microstructure that enables adequate permeability of the mush, and subjected to tensile stresses, is a self-healing material. On the other hand, the development of solid bridges between grains at high solid fractions in the absence of any liquid would build up enough strength and ductility to prevent any brittle rupture. Hence, the liquid feeding should be just enough to supply some liquid to solid bridges that enables their liquid embrittlement, otherwise only the mechanisms of ductile fracture, e.g., high-temperature creep and pore coalescence will be active [7]. Evidence of plastic deformation during hot tearing has been noted in direct observations [101] and upon examination of fracture surfaces [3]. We can suggest the following approach to treating the nucleation and propagation of hot cracks. Several distinct mechanisms are operational in different temperature or compositional ranges or, in other words, at different fractions of solid. Here we consider only the decreasing temperature as the factor affecting the solid fraction, though the composition is obviously the other factor that acts in a similar manner. It is important to note that the microstructure, e.g., grain size and morphology, affects the critical temperatures and fractions of solid. In most cases, the local stress–strain situation is crucial. Therefore, the stress concentration in speciic locations produces the conditions for crack nucleation. The crack initiator in all cases considered below could be represented by pore, liquid pool or ilm, interface with an intermetallic particle, or non-metallic inclusion. At relatively low fractions of solid below the coherency temperature, the permeability of the mushy zone enables adequate feeding of the solidiication

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

243

shrinkage, and most of the precipitating gas bubbles can loat to the liquid part of the sample. In this case, the tensile stresses caused by nonuniform thermal contraction of the coherent dendrites may cause the formation of cavities and gaps that are immediately illed with liquid, or “healed.” With further temperature decrease, the tensile stress builds up to such an extent that the liquid ilm separating grains ruptures and the gap formed cannot be illed with liquid due to the increasing capillary pressures required to ill ever narrowing openings between grains. On further cooling, the bridging between solid grains replaces former entanglement and touching of grains, and the stress can now be transmitted over larger distances through the rigid solid skeleton, hence the semi-solid body acquires macroscopic strength. This critical “rigidity” temperature can be determined experimentally as discussed in Section 5.1. It is worthwhile to recall that the rigidity temperature strongly depends on the structure. Usually the bridging is referred to as the bridging between (Al) dendrite branches. Recently, Nagaumi et al. [67] made a very interesting observation on the effects of impurities on hot tearing susceptibility of aluminum alloys. They showed that large intermetallic particles, e.g., α(AlFeMn), that were formed early in solidiication, just after the aluminum solid solution, could bridge (Al) dendrite arms. As a result, the semi-solid alloy acquires some strength at a lower solid fraction, well above the zero-ductility temperature (see Table 5.2, entry HS65). Therefore, the hottearing susceptibility increases. In addition to the effect on the rigidity, these large intermetallic particles can effectively block the feeding channels in the mushy zone, decreasing the possibility of crack healing. Generally, the decreased permeability of the mush in combination with solidiication shrinkage leads to the local pressure drop over the semi-solid region, which, in combination with the evolving gas, promotes the formation of voids at available interfaces, mainly on grain boundaries and interfaces with inclusions. The nonuniform thermal stress causes rather signiicant strains in the semi-solid material that may or may not be sustained by the solid bridges. Even limited access of the liquid to the solid bridge will result in its brittle fracture by the mechanism of liquid-metal embrittlement. This is partly relected in the proposal of van Haaften et al. [107] to use the fraction of grain boundaries covered with liquid rather than the fraction of liquid in the constitutive equation for the mechanical behavior of semi-solid aluminum alloys. The same mechanism may act at subsolidus temperatures when some amount of nonequilibrium liquid is present at grain boundaries or other stress concentrators because of nonequilibrium character of solidiication [7] or local remelting [105]. There is also a possibility that the semi-solid material fails macroscopically in a brittle manner (because of ilm rupture and liquid-metal embrittlement) with ductile rupture of some solid bridges on the microscopic level [101]. Finally, the fraction of bridged grain boundaries becomes so overwhelmingly large and the remaining liquid is so scattered in the solid network that the semi-solid material behaves like completely solid material and fails in a ductile manner by ductile pore coalescence and high-temperature creep.

Copyright 2008 by Taylor and Francis Group, LLC

244

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

TABLE 5.5 Possible Mechanisms Acting in the Hot Tearing Phenomenon [94] Temperature Range/ Fraction of Solid Between coherency and rigidity temperatures; 50–80% solid Below rigidity temperature; 80 to 99% solid

Close to the solidus; 98–100% solid

Nucleation of Crack* Grain boundary covered with liquid, shrinkage or gas pore Pore, surface of particle or inclusion, liquid ilm or pool, vacancy clusters

Pore, surface of particle or inclusion, segregates at grain boundary, liquid at stress concentration point, vacancy clusters

Propagation of Crack

Fracture Mode

a. Liquid ilm rupture b. Filled gap

a. Brittle, intergranular b. Healed crack

a. Liquid ilm rupture; liquid metal embrittlement of solid bridges b. Plastic deformation of bridges a. Liquid metal embrittlement

a. Brittle, interganular

b. Plastic deformation of bridges, creep

b. Ductile failure of bridges possible a. Brittle, intergranular, transgranular propagation is possible b. Macroscopically brittle or ductile, intergranular; transgranular propagation is possible

* The crack initiator should be located in the place of stress concentration.

The outline of these mechanisms is given in Table 5.5. Figures 5.23, 5.24 and 5.31 illustrate the correlation between these mechanisms and the development of the structure during solidiication. A criterion that can predict not the probability but the actual occurrence and extent of hot tearing should be based on the application of multi-phase mechanics and fracture mechanics to the failure of semi-solid materials, which is today limited by the lack of knowledge about the actual nucleation and propagation mechanisms. The mechanisms outlined in Table 5.5 are based on the common sense and interpretation of very few experimental observations. What is needed is a thorough and systematic study of fractures occurring in solidifying materials with the aim of singling out the nature and the critical dimensions of defects or structure features that can cause the nucleation of hot cracks. It is also necessary, in our opinion, to acknowledge that different mechanisms of crack propagation are possible at different fractions of solid. Therefore, different models should be applied to the development of hot tears in ingots, billets, and castings depending on the alloy composition, structure, and the level of stresses that are present. For example, coarse-grained material with a large solidiication range and high coherency temperature is likely to fail due to liquid ilm rupture. In the case of alloys with a considerable amount of eutectics, e.g., foundry

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

245

Macro- and mesoscopic parmeters: solidification shrinkage; thermal contraction; liquid feeding; stress−strain conditions

Alloy composition: nonequilibrium solidification

Distribution of pores and other defects: size, volume fraction

Local (microscopic) parameters: solidification range; fraction solid at coherency, rigidity and eutectic temperature; stress, strain, strain rate; critical size of a defect under given stess−strain conditions

Solid fr. < Coherency No cracks/healing

Coherency < Sold fr. < Rigidity Brittle fracture with Griffith-type criterion if feeding is insufficient

Sold fr. > Rigidity Brittle and/or ductile fracture with Kc or J- intergral type criterion

FIGURE 5.32 Flow chart for the development of a new hot tearing model and criterion [94].

alloys of the Al–Si system, the healing of cracks is most probable. In contrast, a ine-grained alloy that develops coherency late in solidiication and does not contain much eutectics, e.g., a commercial wrought alloy, will undergo complex failure involving liquid-metal embrittlement and plastic deformation of solid bridges with a resultant mixed brittle/ductile fracture. Figure 5.32 shows a low chart for the development of a new hot tearing criterion. In summary, the development of a new hot tearing criterion faces two main challenges. First, we lack knowledge of the actual causes of crack nucleation, that is to say, we do not know exactly what defects or structure defects can act as crack initiators under particular temperature–stress conditions. Second, there is a possibility that different mechanisms of crack propagation and inal failure depend on the fraction of solid at which the fracture occurs and on the alloy structure. The application of multi-phase mechanics and, eventually, fracture mechanics to the phenomenon of hot cracking looks quite promising.

5.4.

Effects of Process Parameters on Hot Tearing during DC Casting

During DC casting, alloy composition and casting parameters are critical for the formation of structure and solidiication defects, as we have already shown in Chapters 3 and 4.

Copyright 2008 by Taylor and Francis Group, LLC

246

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

The compositional dependence of hot tearing susceptibility was discussed in Section 5.3.1. However, the results that we considered previously were obtained from special hot-tearing tests, and not from real-scale DC casting. Grain reinement is another aspect of compositional effects in hot cracking, particularly important for DC casting. Practice shows that, in general, structure reinement and homogeneity decrease the vulnerability of an alloy to hot tearing. We will look more closely at this aspect. We also know that casting parameters play an important role in the occurrence of casting defects [99, 108, 109]. The most important casting parameter that affects hot tearing is the casting speed [99, 110]. Water low rate has much less impact [110], and the literature data on the inluence of melt temperature are controversial [16, 110, 111]. Casting speed and other process parameters, i.e., casting temperature and water low rate, are also known to affect the structure formation during solidiication. This is because of their inluence on cooling conditions, melt low, and geometry of liquid and semi-liquid parts of the billet (see Chapter 3). It is logical to suggest that the effects of casting parameters on structure formation and hot tearing are correlated. And, of course, the thermo-mechanical situation in the billet (ingot) is of paramount importance because the distribution and magnitude of stresses and strains create the external conditions for the formation of hot cracks. 5.4.1

Thermo-Mechanical Behavior of a DC-Cast Billet (Ingot)

Temperature and cooling-rate gradients and thermal contraction of the solid phase are the main reasons behind the uneven stress distribution across the billet or ingot. As a result, the regions of tension and compression are formed in different sections of a casting. The sections that are subjected to tensile stresses are most vulnerable to cracking. Hot tearing is likely to occur if tensile stresses spread to the semi-solid region of the billet (ingot). We did not differentiate previously in this book between round-shaped extrusion billets and lat-shaped rolling ingots. In the case of stress distribution it is, however, important to distinguish these two basic shapes of DCcast products. The thermo-mechanical behavior of billets is much simpler than that of ingots. In the latter case, the width-to-thickness ratio plays a role in stress distribution. The shape distortions are also more diverse in the case of ingots. Figure 5.33 demonstrates schemes of stress distribution for a round billet (a) and a lat ingot (b) [112]. For the billet, the pattern is simple—a region of tensile stresses is located in the center of the billet, while the periphery experiences compression. Accordingly, hot cracks usually appear in the central section of the billet. Stress distribution in the ingot is more complicated with the zones of tension concentrated closer to the short-side surface (along a′–c′ direction in Figure 5.33b) and corners. There is still a possibility of tensile stresses

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

247

b

Compression Tension

b′

(a)

b a′

c′

l

a c

(b)

FIGURE 5.33 Distribution of stresses in vertical cross-section of a round billet (a) and a lat ingot (b) in a steady-state regime of DC casting. (After Livanov [112].)

in the central section of the ingot but the magnitude of these stresses is usually considerably less than those at the short-side periphery (area c–c′–a–a′ in Figure 5.33b). As a result of this stress distribution, hot cracks appear closer to the short side of the ingot. The situation is further complicated by the deformation of the wider (rolling) face of the ingot (along l direction in Figure 5.33b). As the compression stress is higher at the surface than in the interior, and is greater in the middle of the rolling face than closer to the edges (where it in fact changes to the tensile stress), the whole wider face of the ingot bends inward, exhibiting socalled pull-in distortion (Figure 5.34). Consequently, the part of the solid shell closer to the mush experiences tension and is prone to cracking. Hot tear can appear if this tension region expands to the mush, as shown in Figure 5.34. In practice, modern molds are designed in a convex shape so as to compensate for the pull-in of the rolling face and, therefore, reduce scalping of the ingot before rolling.

Copyright 2008 by Taylor and Francis Group, LLC

Liquid

Mush

Pull-in

Solid

Butt swell Butt curl Starting block

FIGURE 5.34 Shape distortions typical of DC-cast ingots. A view from the short side of the ingot.

248

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

Quite obviously, the increase in the casting speed that results in the deepening of the sump and thinning of the shell will make the pull-in more pronounced with the consequence of more hot tearing in the subsurface layer. In addition to “pull-in,” two other shape distortions are typical of DC casting, namely butt curl and butt swell (see Figure 5.34). These defects occur in the initial stage of casting close to the bottom end of the casting, which is nicknamed “butt.” Both ingots and billets are prone to butt curl and swell. Rapid cooling of the bottom end of the casting immediately after the exit from the mold—both from the starting block and from the direct water jet— produces excessive thermal stresses that lift up the corners (edges) of the ingot (billet), forming the butt curl. This defect can lead to dangerous consequences as the casting loses the irm stand on and the thermal contact with the starting block. As a result, the casting can deviate from the verticality, the bottom part may remelt with subsequent bleed-out of the melt, and the water may enter the formed gap and vaporize with ingot “bumping” on the starting block. Cracks may form due to the stress concentration. The butt swell is a result of lesser apparent linear thermal contraction of the bottom part of the ingot (billet). This part is formed in the mold closed by the starting block when the sump is shallow. As soon as the casting exits the mold and the direct cooling is applied, the sump deepens and the thermal contraction of the shell becomes more pronounced because the liquid interior does not offer much resistance [28]. Therefore, the bottom appears to be swollen whereas, in fact, the rest of the casting is more contracted. The butt swell leads to the removal of the ingot (billet) butt prior to the downstream processing. It is clear that the distribution of stresses and strains in the billet (ingot) is of paramount importance for the occurrence of cracks. Nonuniform cooling across the thickness of the billet (ingot) plays the main role in the development of thermo-mechanical situation upon DC casting. The surface is cooled much faster than the core and, in addition, there is a difference in cooling rates in the subsurface layer that is formed sequentially as the casting passes through the regions of primary cooling, air-gap and secondary cooling (see Figures 3.14a,b and 3.15). In addition, there is a frequently observed acceleration of cooling in the center of the billet (ingot), a phenomenon that we discussed in detail in Section 3.2 (Figure 3.14c,d). Figure 5.35 shows the experimentally measured distribution of cooling rates in the vertical centerplane of a 2024-alloy billet cast at two casting speeds. The measurements were performed by thermocouples traveling downward with the billet as we described in Section 3.2. The cooling rates that are shown in the igure are the instant cooling rates (tangents to the cooling curve) taken at different times of solidiication. Therefore, these data do not relect the total solidiication time but rather the instantaneous cooling situation in the speciic location of the billet. Two conclusions can be drawn from the cooling-rate distribution shown in Figure 5.35. First, there are two distinct regions of high cooling rate, at the surface in the point of water impingement and in the center of the billet. Second, the cooling rates and the gradient of cooling rates in the radial direction of the billet increase as the casting speed increases.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

249 30 mm/min 0.7

Distance from the melt level in the mold, mm

50 100

60 mm/min

L L

1.2

Mold bottom

3.5 2.25 1.7 1.6 1.4

150

3.9

S 2.25 1.8

3.2 2.5

200 1.4

250

1.6

2.2 1.6

1.2 0.8

S

300 2.5

350 3.8

3.2

400 2.2 1.6

450 FIGURE 5.35 Experimentally obtained distribution of cooling rates (isolines marked in K/s) in a 470-mm billet of a 2024 alloy cast at 30 (left) and 60 (right) mm/min [109]. L and S denote liquidus and solidus isotherms, respectively. Mold without hot top has been used.

As a result of uneven cooling and uneven solidiication progress in the radial direction of the billet (ingot), a peculiar thermo-mechanical situation is created in DC castings, as shown in Figure 5.33. Let us look at the rather simple situation in a round billet. The solid shell has already cooled suficiently and undergone most thermal contraction, when the inner core of the billet enters the solidiication domain and, in particular, passes the isotherm of rigidity. This inner core wants to contract but the solid shell is robust enough to offer a considerable resistance to the shape change. As a result, the contraction of the inner core is constrained and tensile stresses start to build up there, whereas the shell remains under compression. The build-up of stresses starts from the beginning of casting, passes a transient stage, and then stabilizes in the steady-state regime of casting. The start-up of casting is the most critical stage in the process because the development of stresses is a function of the sump evolution (see Figures 3.3 and 3.4) and the casting recipe (how the casting speed changes in the transient stage of casting). Three stresses and strain components can be distinguished in the round billet, i.e., axial, radial, and circumferential. The combination of these three components when they are in tension may result in the failure of the semisolid material with crack propagation in the plane normal to the strongest

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

250

tensile stress. Suyitno et al. [97, 113] performed a thorough analysis of stress and strain distribution during steady-state and transient (start-up) regimes of DC casting. Some of the results are shown in Figure 5.36. It can be immediately seen from Figure 5.36a that the maximum tensile stresses at temperatures above the solidus are concentrated in the center of the billet with radial (CR) and circumferential (CC) components prevailing. At the periphery of the billet (10 mm from the surface) only the circumferential component (PC) is in tension. This stress distribution may help explain why the hot crack in the round billet spreads along the axis (normal to radius) and in radial (normal to circumference) directions. 200

25 CA

Stress, MPa

CR; CC

Solidus center Semi-solid Solidus periphery

100 50

Circumferential stress, MPa

150

PR

0 Semi-solid

−50 PA

20 3

15 4 1 2

10

5

−100 PC

−150

(a)

0

50

100

150

0

200

0

50 100 150 200 250 300 350 400 450

(b)

Time, s

Distance from bottom, mm

Circumferential visco plastic strain

0.010

0.008

0.006 3 4

0.004

1

2

0.002

0.000 0

(c)

50 100 150 200 250 300 350 400 450 Distance from bottom, mm

FIGURE 5.36 Calculated development of stresses (a,b) and strains (c) during DC casting of a 200-mm billet of an Al–4.5% Cu alloy: (a) steady-state regime at a casting speed of 120 mm/min, axial (second index = A), radial (R) and circumferential (C) stresses in the center (i rst index = C) and at the periphery, 10 mm from the surface (P). Solidus line indicates the position of nonequilibrium solidus in the billet, semi-solid region is on the left of this line; (b) and (c) start-up stage of casting, circumferential stress and strain, respectively, numbers correspond to different startup regimes as shown in Figure 5.30c [97, 113]. (Reproduced with kind permission of Springer Science and Business Media.)

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

251

In a transient regime of casting, in particular during the critical start-up stage, the evolution of stresses and strains depends on the casting regime. There is an obvious correlation between the development of tensile stresses and strains at the centerline of the billet (Figure 5.36b,c), the hot tearing susceptibility (Figure 5.30d), and the depth of the sump and the thickness of the mush (Figure 3.4b,c). The most rigid casting recipe (3 in Figure 5.30c and Figure 5.36b,c) shows the highest tensile stress and strain in the center of the billet, which also reach their maxima at the maximum sump depth (Figure 3.4b). The most gentle start-up mode (2 in Figure 5.30c and Figure 5.36b,c) produces more moderate stresses and strains, which results in less hot tearing susceptibility (Figure 5.30d). A realistic start-up regime (4 in Figure 5.30c) suficiently reduces the maximum tensile stress and strain (Figure 5.36b,c) and, correspondingly, the hot tearing possibility (Figure 5.30d). 5.4.2

Effects of Composition and Casting Speed on Hot Tearing during DC Casting

It is interesting to know whether the compositional dependence of hot tearing (see Figures 5.21 and 5.22) is still valid in DC casting, and not overwhelmed by the speciic thermo-mechanical situation that is caused by cooling conditions. In order to check this, several binary Al–Cu alloys with compositions given in Table 5.6 were tested upon casting 200-mm billets in a pilot DC caster [99]. An experiment scheme with ramping up and down the casting speed was chosen, as illustrated in Figure 5.37a. The billets were checked for the occurrence of hot cracks in horizontal sections relecting different casting speeds and in the vertical centerplane of the billet. The large sections were cut and polished, and the hot crack susceptibility (HCS) was quantiied as the area affected by cracks in the horizontal cross-section (Acrack) divided by the area of the billet cross-section (Atotal), as shown in Figure 5.37b. Hot cracks were observed only in alloys 1–3 from Table 5.6. Cracks appeared in the central part of the billet and had a typical spider-type shape in the billet cross-section, as shown in Figure 5.38. In most cases cracks had the appearance of spreading from the center in radial directions. TABLE 5.6 Chemical Composition of Tested Alloys as Determined by Spectrum Analysis

Alloy No. 1 2 3 4

Cu, wt%

Fe, wt%

Si, wt%

Other Impurities, Total wt%

1.03 1.98 2.93 4.49

0.16 0.18 0.19 0.19

0.04 0.06 0.10 0.06

0.04 0.03 0.04 0.03

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

252

Casting speed, mm/min

Casting parameter

200 Water flow rate, l/min 150 100 50 0 0

200

600

400 Billet length, mm

(a)

Atotal Acrack

(b) FIGURE 5.37 Process chart of DC casting experiment with ramping the casting speed (a) and a scheme for hot tearing quantiication (b) [99]. (Reproduced with kind permission of Springer Science and Business Media.)

FIGURE 5.38 Experimentally observed hot crack in the centerplane of a 200-mm billet from an Al–2% Cu alloy cast according to the schedule shown in Figure 5.37a. Billet bottom is on the left. Horizontal slice in the center corresponds to the maximum casting speed.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

253

5

HCS

0.0

0.04 0.03

0 −1 2 /min 1 − m 14 n, c 6 − dow 1 − eed 8 −1 ng sp 0 i −2 Cast 18 in m / , cm

0.02

0. 0. 0 5 1. 0 1. 5 2. 0 2. 5 3. 0 3. 5 4. 0 4. 5

0.01 0.00

Cu

,%

10

16 p 14 ed u pe s g

12 stin Ca

FIGURE 5.39 Effect of composition and casting speed on hot tearing susceptibility of DC-cast Al–Cu alloys [99]. The casting schedule is shown in Figure 5.37. (Reproduced with kind permission of Springer Science and Business Media.)

The dependence of hot cracking susceptibility on the casting speed and copper concentration in the examined billets is illustrated in Figure 5.39. Hot cracking depends strongly on the casting speed. The casting speed required for the initiation of crack is higher than the casting speed at which the crack stops or heals. The concentration dependence of hot tearing shows the existence of the compositional range of maximum hot cracking susceptibility, between 0.5 and 1.5% Cu. It should be noted that almost no hot tearing at any given casting speed (the range was extended up to 220 mm/min) was observed in alloys that contained more than 4% Cu (alloys containing up to 5.6% Cu were studied). The larger hot tearing sensitivity at lower copper concentrations can be explained by a thicker mushy zone in these alloys (region with hindered feeding or low permeability), less residual liquid available for feeding (represented by the amount of nonequilibrium eutectics in Figure 3.27e), more time spent in the vulnerable temperature range [6], and larger deformations induced by thermal contraction [24]. We have already discussed that there is a correlation between the dimensions of the sump and mush and the development of stresses in the billet (Section 5.4.1). Figure 5.40 shows the effect of copper concentration on the vertical distance between the liquidus and solidus (transition region) in the billet cast at two speeds in the examined range. These data are obtained by computer simulation where nonequilibrium solidiication has been assumed, with the equilibrium liquidus and the nonequilibrium solidus, the latter was taken as a temperature of the binary eutectic, 548°C. The transition region becomes wider with the increasing casting speed and decreasing copper

Copyright 2008 by Taylor and Francis Group, LLC

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys Distance between liquidus and solidus, mm

254

60 50 40 30 20 10 0 20

Bill

et c

0 16

0 12 ect io

s-s

80

n, m

(a)

Distance between liquidus and solidus, mm

ros

4.0 40

m

0 1.0

2.0

5.0

3.0 , % Cu

60 50 40 30 20

10 0 20 60 Bil let 1 20 cro ss- 1 80 sec (b) tion 40 ,m m

4.0

0 1.0

2.0

5.0

3.0 , % Cu

FIGURE 5.40 Effect of copper on the width of the transition region in a 200-mm billet cast at (a) 100 mm/min and (b) 200 mm/min [99]. (Reproduced with kind permission of Springer Science and Business Media.)

concentration, which agrees well with the hot tearing tendency. Another observation is that the transition region tends to narrow in the central part of the billet. This feature becomes more pronounced at higher casting speed and copper concentration. Casting experiments with ramping the casting speed up and down exhibit one peculiar feature. The maximum of hot tearing corresponds not to the maximum casting speed but to a certain speed at the deceleration stage of casting (Figures 5.37a and 5.39). Such behavior has been also reported by Commet et al. [114] and explained in terms of the thermal inertia in the development of the sump. We have considered this phenomenon in detail in Chapter 3 (see Figure 3.5). This observation proves that the most important parameter for the development of hot cracking is the sump depth rather than the casting speed proper.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

255

(a)

(b)

FIGURE 5.41 Distribution of strains (positive values mean tensile strains) in the vertical cross-section of a 200-mm billet cast at (a) 100 mm/min and (b) 200 mm/min [99]. TL and TS are the temperatures of liquidus and solidus, respectively. Surface is on the left and the centerline (CL) is on the right of each image. The position of the liquidus isotherm differs from those shown elsewhere in this book (e.g., Figures 3.6, 3.16) because in the simulations shown in this igure luid low has not been taken into account. (Reproduced with kind permission of Springer Science and Business Media.)

Thermomechanical simulation of the billet shows that tensile stresses and strains are concentrated in its central part, which is in line with our previous discussion in Section 5.4.1. What is more important is that these tensile stresses and strains start to appear already in the mushy zone, and this “penetration” of tensile stress and strain into the mushy zone strongly increases with increasing casting speed, as demonstrated in Figure 5.41 for strains (the pattern for stresses looks similar). The stress–strain situation is a necessary but not suficient condition for the appearance of hot cracks. This becomes clear when we compare alloys with different copper concentration (Figure 5.39). The feeding of the solidifying material with the melt should be inadequate for compensation of the

Copyright 2008 by Taylor and Francis Group, LLC

256

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

solidiication shrinkage and thermal contraction [15, 80] (see also Sections 5.3.1 and 5.3.2). Hence, the amount of liquid available at the crucial stage of solidiication, when the semi-solid material is most vulnerable and the hot cracking susceptibility is high, is extremely important. Figure 3.27 clearly demonstrates that the amount of eutectics (last liquid to solidify) is decreasing toward the center of the billet, with this tendency more pronounced at higher casting speeds and lower copper concentrations. This observation its well into the mechanism of hot tearing, proving that the amount of available liquid at the end of solidiication is lower in the most critical part of the billet and in the most crack-sensitive compositional range. The amount of eutectic structure constituent, therefore, can be an important structure indicator of the vulnerability of the alloy to hot cracking at the particular location in the billet. One of the common practices to improve the structure and properties of ascast material is grain reinement (Section 2.4). Apart from making the grain size smaller and its distribution more homogeneous across the billet, grain reinement results in a lower thermal contraction in the solidiication range (Figures 5.7, 5.9, and 5.10) through the delay in the rigidity development; improved tensile strength (Figure 5.13a) and ductility (Figure 5.17a) of a semisolid material; and lower permeability of the mushy zone at high volume fractions of solid (Equation 4.6). The complex interaction of these phenomena generally results in a decreased hot tearing susceptibility (Figure 5.22). Recently the positive effect of grain reinement on the development of hot tears in commercial 1050, 3104, and 5182 alloys was demonstrated by Lin et al. [104]. First of all, let us look at the effect of grain reinement on the development of hot tears during DC casting. We studied the structure of a 200-mm billet from a commercial 7075 alloy [115, 116]. In the irst experiment, part of the billet was cast without grain reinement, then grain-reining Al–Ti–B rod was fed to the melt in the launder. The transition from coarse to ine structure is obvious on the macrostructure, as shown in Figures 5.42a, 3.21b, and 3.22. The crack readily appears from start-up of casting at a speed of 80 mm/min and propagates along grain boundaries (Figure 5.42b). The crack, however, vanishes when the grain reining is started and the grains become smaller, as vividly demonstrated in Figure 5.42a,c. The crack again re-appears when the casting speed is increased to 120 mm/min. In the second experiment, the grain reiner was added in the furnace. In this case, no cracks were found throughout the billet irrespective of the casting speed. The grain size, as shown in Figure 3.21b, decreases dramatically with the grain rei ner added (from 1080 to 165 µm), but depends only slightly on the casting speed (165 and 95 µm at 80 and 120 mm/min, respectively) and does not depend on the way the reinement was performed. Figure 3.21d demonstrates that the dendrite arm spacings in the center of the billet are close for grain-reined and nongrain-reined billets, 22–23 µm. We may conclude that grain reinement can eficiently hinder the development of hot crack in high-strength aluminum alloys. The fact that the crack re-appeared at a higher casting speed in the

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing

257

1000 µm

(a)

(b)

(c)

FIGURE 5.42 Crack propagation and healing during DC casting of a 7075 alloy: (a) macrostructure showing the transition region from nongrain-reined to grain-reined structure; (b) propagation of hot crack along grain boundaries in nongrain-rei ned part of the billet; and (c) microstructure of the zone where hot crack stopped [117].

Copyright 2008 by Taylor and Francis Group, LLC

258

Physical Metallurgy of Direct Chill Casting of Aluminum Alloys

irst experiment and does not appear at all in the second experiment may be due to different localization of stresses and strains in the billet that has been already weakened by a crack. There are, however, some experimental data showing that structures with very ine grains exhibit somewhat enhanced hot tearing [73]. Several phenomena may play a role here [117]. Grain reinement can affect permeability in a nonlinear manner. The structure modiication may include the transition from columnar dendritic to equiaxed dendritic and then to cellular, nondendritic grains. As a result, the structure parameter that determines the permeability changes from dendrite arm spacing to grain size (Equation 4.6). The coherency temperature and, therefore, the region where stress development and feeding take place becomes narrower upon grain reining or, in other words, corresponds to higher volume fractions of solid. There is also a change in the liquid distribution between the grains with associated different capillary pressures and stress localization. The decrease of permeability associated with the grain reinement, especially for grain sizes below 300 µm, can be fully compensated for by the delay of thermal strain development and reduction of liquid ilm thickness [117]. As a result, the hot tearing susceptibility decreases with grain reinement as long as the grain remains dendritic [104]. This analysis agrees well with our results, as demonstrated in Figure 5.42 and with casting practice [73, 118]. If the grain reinement is accompanied by the transition from dendritic to nondendritic grain morphology, the analysis based on interplay between permeability and stress development shows the hot tearing susceptibility may increase again, especially for cellular grain size below 50 µm [119]. At the same time, we see that for grain sizes above 100 µm the hot tearing susceptibility of ine grains is almost always less than that of coarser grains, irrespective of their dendritic or globular morphology [119]. There are only a few examples of commercial DC-cast billets or ingots with nondendritic grain structure. One of the technical means to achieve the nondendritic structure in commercial-size billets and ingots from aluminum alloys is ultrasonic (cavitation) melt treatment [120]. There are no direct reports on the change of hot cracking susceptibility upon transition to nondendritic structure. However, nondendritic structure with grain sizes between 20 and 80 µm has been obtained in large ingot and billets from high-strength 2XXX and 7XXX series alloys cast with cavitation melt treatment [120]. No cracks, either hot or cold, have been found in these billets and ingots, whereas billets and ingots with dendritic grain structure cast under the same conditions but without ultrasonic treatment always had cracks. These practical data prove that analytical models cannot encompass the entire complexity of the nature but, at the same time, they can be very useful in the analysis of contributions of individual phenomena that frequently counteract one another in reality.

Copyright 2008 by Taylor and Francis Group, LLC

Hot Tearing 5.4.3

259

Effect of Melt Temperature on Hot Tearing during DC Casting

In contrast to casting speed, the melt temperature has not received much attention with regard to its inluence on hot tearing in DC-cast billets. Only few experimental results are available on the subject, which can be explained by technological dificulties accompanying high pouring temperatures during DC casting. Grün and Schneider [121] examined billets of grain-reined commercial aluminum obtained by level pour DC casting at different pouring temperatures, 700 to 740°C. They concluded that the increase in melt temperature resulted in the development of feathery columnar grains and higher temperature gradients closer to the mold. Tarapore [122] and Reese [123] reported the increased depth of the sump, higher temperature gradients in the liquid bath, and a thinner surface liquation layer corresponding to a higher melt temperature upon DC casting. The effects of melt temperature on the structure, hot tearing, and quality of experimental and shape castings have been studied and reported on several occasions, e.g., [3, 16, 72, 124]. Under conditions of moderate superheating (