You are on page 1of 19

Modelling of macrosegregation: applications and future needs

C. Beckermann
of alloy species due to the relative movement or ow of segregated liquid and solid during solidi cation. There are numerous causes of uid ow and solid movement in casting processes: d ow that feeds the solidi cation shrinkage and the contractions of the liquid and solid during cooling d buoyancy induced ows due to thermal and solutal gradients in the liquid; the thermal and solutal buoyancy forces can either aid or oppose each other, depending on the direction of the thermal gradient and whether the rejected solutes cause an increase or a decrease in the density of the liquid d forced ows due to pouring, motion of gas bubbles, applied magnetic elds, stirring, rotation, vibration, etc. d movement of free (equiaxed) grains or solid fragments that have heterogeneously nucleated in the melt, separated from a mould wall or free surface, or melted oV dendrites; the solid can either oat or settle depending on its density relative to the liquid d deformation of the solid network due to thermal stresses, metallostatic head, shrinkage stresses, or external forces on the solid shell such as those from the rolls in continuous casting of steel. All eVorts to prevent macrosegregatio n are aimed at controlling uid ow and solid movement. Examples include adjustments to the alloy composition or thermal gradients to induce a stable density strati cation in the liquid; application of nozzles, baZes, porous materials, centrifugal forces, or electromagnetic elds to redistribute the ow; changes in the riser or mould design to in uence the cooling and thermal convection patterns; addition of inert particles (as in composites) to change the eVective viscosity of the melt; controlled deformation such as soft reduction during continuous casting of steel to squeeze highly enriched liquid away from the centreline of slabs; and modi cations to the grain structure (e.g. grain re ners) that change the resistance to ow through the solid network or the prevalence of equiaxed grains. Macrosegregation models are generally aimed at understanding the basic mechanisms involved, quantitatively predicting the occurrence and severity of macrosegregation , and performing parametric studies for control and improvement of casting processes. Such models are very complex and require large computing resources in their solutions, because they often must consider many aspects of a solidi cation process simultaneously. Macroscopic phenomena to be considered include heat transfer, solute transport, uid ow, solid movement, and solid deformation at the scale of the casting. In addition, one must account for phase equilibrium, nucleation, structure formation, segregation, and ow at various microscopic scales. Any factors that aVect the ow and the microstructure also in uence macrosegregation, and vice versa.
2002 Vol. 47 No. 5 243

Modelling and simulation of macrosegregation has experienced explosive growth since the pioneering studies of Flemings and co-workers in the mid-1960s. This paper presents a review of recent macrosegregation models, with particular emphasis on their application to selected industrially relevant casting processes. The successes and shortfalls of the models in predicting measured macrosegregation patterns are noted. Although it is still early in their development, advanced macrosegregation models that include a detailed consideration of the solidification microstructure are also reviewed. Recent studies involving direct numerical simulation of macrosegregation related phenomena on a microscopic scale are highlighted. Important issues deserving future research attention are identified. IMR/378
The author is in the Department of Mechanical and Industrial Engineering, 2412 SC, The University of Iowa, Iowa City, IA 52242, USA (becker@engineering. uiowa.edu). © 2002 IoM Communications Ltd and ASM International. Published by Maney for the Institute of Materials, Minerals and Mining and ASM International.

Introduction
Producers of virtually all types of cast metal continue to struggle with macrosegregatio n defects. The rejection of single crystal superalloy turbine blades due to the presence of freckles is only one of the more pressing examples. Not surprisingly, modelling of macrosegregatio n is still a very active research eld,1 – 3 since the pioneering contributions by Flemings and co-workers beginning in the 1960s.4 – 1 5 Flemings and co-workers discovered the importance of convection within the mushy zones of solidifying alloys, and derived the basic equation describing macrosegregation due to interdendritic ow. Over the years, this research has led to numerous improvements in industrial casting processes, as well as to a better understanding of convection in the mushy zones of the earth’s core, solidifying magmas, and sea ice. The phenomena involved in alloy solidi cation that lead to macrosegregatio n are so complex that their quantitative prediction for industrially relevant casting processes is only beginning. The purpose of the present work is to review the current state of the art in modelling of macrosegregation, and to provide an outlook of where the eld is going. The emphasis in this review is on the application of models to quantitatively predict macrosegregatio n in cast alloys. The basic derivations of the model equations are adequately reviewed elsewhere.1 , 2 Before proceeding, it is useful to recall that the cause of macrosegregatio n is the long range advection
DOI 10.1179/095066002225006557

International Materials Reviews

related and increasingly important work on microscale modelling of convection in mushy zones is highlighted. implying no macrosegregation. Then. Both ow in the same direction as the shrinkage ow. Consequently. k is the partition coeYcient. for example at an impenetrable chill face. v T =vT ä Ñ T . they arrived at the following ‘local solute redistribution International Materials Reviews 2002 Vol. The subsequent sections review work in this area that has been performed in the last 10 to 15 years.e. C. 2.244 Beckermann Modelling of macrosegregation 1 Potentially important interactions in modelling macrosegregation in castings Figure 1 illustrates some of these interactions in casting processes. but with a speed less than the shrinkage velocity. result in n C D . Equation (1) reduces to the Scheil equation.5 – 7 They considered the ow of interdendritic liquid through a xed dendritic solid network. 2. Then. n The physical signi cance of equation (1) can be understood as follows (for k <1): 1. v is the liquid velocity vector. 4. vn =­ vT b (1 ­ b ) = vs . Flow in the single phase bulk liquid region and movement or deformation of solid were not treated explicitly. vn <vs . 3. this is the so called inverse segregation often observed in casting of aluminium alloys. 5 where f l is the volume fraction liquid. since a model of such complexity would hardly be useful in practice. this review focuses on examples of research where some of these interactions are taken into account in order to address industrially relevant macrosegregatio n problems. 47 No. 2. Flemings. respectively. vn =vä Ñ T. .e. The review concludes by discussing future research needs and directions. relative to the normal isotherm velocity. the liquid is not undercooled) and (ii) there is no solute diVusion in the solid. and ç T / ç t is the time rate of temperature change. Flemings and co-workers assumed that: (i) the liquid owing through it is well mixed and in local equilibrium with the solid (i. positive macrosegregatio n will result. The next section is a brief review of early work (up to the early 1980s) on modelling of macrosegregation. vs < vn < vT . in the direction of decreasing temperature) . Flow in the same direction as the shrinkage ow (i. . towards regions of lower solid fraction). The review is structured as follows. .e. MIT) equation’ (LSRE)5 – 7 d fl =­ d Cl (1 ­ (1 ­ b) vä Ñ T fl 1+ k) ç T /ç t Cl Basics of macrosegregation modelling A physical understanding of macrosegregation and its modelling is best obtained by reviewing the pioneering work of Flemings and co-workers in the latter half of the 1960s. Equation (1) also reduces to the Scheil equation when the liquid velocity is just that required to feed solidi cation shrinkage. Ñ T is the temperature gradient. Today. If the liquid velocity alone is zero.e. 2 Schematic of liquid flow through fixed dendritic solid skeleton (courtesy M. i.e. when b and v both vanish. it is certainly not desirable to generate a macrosegregatio n model that includes all the eVects listed in Fig. b= (rs ­ rl )/rs is the solidi cation shrinkage. The latter quantity can also be expressed in terms of the isotherm velocity vector vT and the temperature gradient as ç T / ç t =­ vT ä Ñ T. 4 while accounting for the diVerent densities of the solid and liquid. results in negative macrosegregation. but the shrinkage b is nite (>0). as shown in Fig. i. 1. as well as ow in the opposite direction (in the direction of increasing temperature.2 6 By combining the equations in a special way. 5. These are the same assumptions as made in the derivation of the Scheil equation. (1) . n where vs is the shrinkage ow velocity perpendicular to the isotherms. In deriving the mass and species balances for the ‘mush’ volume element in Fig. C l is the solute concentration in the liquid in weight per cent. Six examples are presented of modern macrosegregation models applied to selected casting processes. the last term in the square brackets in equation (1) can be viewed as the local liquid velocity perpendicular to the isotherms. Following this. in which rl and rs are the densities of the liquid and solid. It is simply intended to provide a physical understanding of macrosegregation modelling to a non-specialist reader. i. . a review of macrosegregatio n models that are coupled to microstructural evolution models is provided. but with a speed greater than the shrinkage velocity.

in order to predict typical macrosegregation patterns and verify them experimentally. axisymmetric ingots. In the numerical solution of the equations. equation (2). The review by Worster2 7 illustrates the strong interest of the uid mechanics community in the same subject. Mehrabian et al. . Felicelli et al. 6.1 4 to macrosegregatio n in multicomponent low alloy steels. The rst macrosegregation model that accounted for the coupling of the ow between the mushy and bulk liquid zones was reported by Ridder et al. . solved the coupled set of equations given by Darcy’s law. The numerical results obtained in these early studies illustrate the importance of the coupling of the ow between the mushy and liquid zones.1 9 Beckermann and Viskanta. They are the result of buoyancy driven convection through the columnar dendritic zone in the same direction as. These examples are intended to illustrate the present state of the art in modelling macrosegregatio n in a few selected casting processes and to highlight some of the shortcomings. Piwonka and Flemings9 and Apelian et al. later. (2) momentum. It is emphasised that macrosegregatio n models have been applied to casting processes other than those covered in the examples below. can be derived from the single domain models as a limiting case. v > vT . The pencil-like shape of the A segregates is related to the localised nature of the thermosolutal convection patterns.1 6 – 1 8 were derived by Bennon and Incropera. formed early in the solidi cation process and relatively poor in solute. Mehrabian et al.e. 2 6 Beckermann and Wang.and shrinkage-driven interdendritic uid ow during the nal stages of solidi cation. equation (1). xed numerical grid. If the ow velocity in the direction of increasing temperature is larger than the isotherm velocity. which are based on mixture theory or volume averaging.2 5 presented the rst direct numerical simulations of freckles or A segregates. and Kou et al. and discuss laboratory scale experimental validation. 2 2 . Example 1: Steel ingot The rst example involves prediction of the classical macrosegregatio n pattern in a heavy steel ingot.1 2 predicted the conditions for freckling in rotating. This is the cause of the open channels in the mush that lead to A segregates or freckles. solved the coupled equation set given by the LSRE and Darcy’s law. Because they considered a quasisteady system. mushy. 5 Applications of macrosegregation models Modelling of macrosegregatio n made signi cant progress in the mid-1980s and early 1990s when so called single domain models. as shown schematically in Fig. . Articles by Beckermann and Viskanta. Hence. In particular.1 5 in the early 1980s. Instead of providing another comprehensive summary of this research. . 2 and Combeau et al. . A shortcoming of all the above studies is that they neglect ow in the (fully melted ) bulk liquid region ahead of the mushy zone.8 proposed that the interdendritic ow velocities could be calculated from Darcy’s law for ow in porous media v=­ K where K is the permeability. a two domain approach was employed. and microstructural parameters such as dendrite arm spacings. the remainder of this section is devoted to reviewing six recent applications of macrosegregatio n models to industrially relevant casting processes. which causes local remelting (see item 6 in the list below equation (1)). the isotherm velocity. and solute conservation equations that are equally valid in the solid. present the detailed derivations of the governing equations. m fl (Ñ P ­ rl g) . . the grid did not evolve with time. The momentum equations are a generalisation of Darcy’s law.Beckermann Modelling of macrosegregation 245 positive macrosegregation . energy.3 thoroughly review the very substantial body of literature on this subject. 47 No. Predicted macrosegregation patterns compared favourably with experimental measurements. Ñ P is the pressure gradient. the equations can be solved using a single domain. 2 3 These models consist of a set of mass. The LSRE of Flemings and co-workers.2 1 and Poirier and co-workers. the term in the square brackets in equation n n (1) becomes negative and local remelting results. . m is the viscosity. Ridder et al. . Fujii et al. and liquid regions of alloy solidi cation systems.. Solutal convection in the bulk liquid was neglected.3 1 Negative segregation appears as a zone in the bottom third of the ingot. and each region was meshed separately. arises partially from buoyancy. and reveal the existence of complex thermosolutal convection patterns in the melt. and the LSRE in the mushy zone. highly enriched melt from colder regions in the mush streams through the still open channels into the bulk liquid ahead of the International Materials Reviews 2002 Vol. It is usually associated with equiaxed dendrites.2 0 Voller et al. Positive segregation near the centreline. Examples include centreline and under-riser segregation in steel casting. where boundary conditions were explicitly satis ed at the boundary between the mushy and bulk liquid zones. that have settled in this region. Voller and co-workers2 8 – 3 0 studied issues associated with numerical solutions of the governing equations. . and g is the gravity vector. and the momentum and energy equations in the fully liquid region.1 1 predicted the conditions for freckling. 3.1 Prescott and Incropera. rl is the density of the liquid as in equation (1). For example.1 0 were the rst researchers to perform experiments to measure the permeability of mushy zones as a function of the liquid fraction. Flemings and co-workers applied the above model to various alloys and casting situations. and particularly at the top (hot top segregation) . the energy equation. The A segregates are pencil-like chains of equiaxed crystals that are highly enriched in solute. and reduce to the single phase Navier– Stokes equations in the liquid region. Bennon and Incropera2 4 and. Single domain continuum or volume averaged models have been further developed and used by numerous researchers. but the temperature eld was taken from measurements. Early work by Mehrabian and Flemings1 3 on ternary alloys was extended in the late 1970s by Fujii et al. i. During solidi cation. but at a faster speed than.

Hence.246 Beckermann Modelling of macrosegregation 3 Schematic of macrosegregation pattern in steel ingot (after Ref. 4. Strong buoyancy driven ow persists throughout the solidi cation process. 5 International Materials Reviews . c 9000 s. due to absorption of carbon from the insulation material. Results for the predicted evolution of the solid fraction and liquid velocity distributions are shown for four diVerent times in Fig. Measured and predicted macrosegregation pro les at the vertical ingot centreline are compared in Fig. The highly coupled advection–diVusion equations in the model were discretised on a two-dimensional (2D) rectangular grid using the nite volume method. 31) mushy zone. the banding pattern along the sidewalls of the ingot is believed to be due to unsteady heat transfer or ow early in the solidi cation process. Fissures then open up along shear planes oriented in a V pattern. b 3000 s. leads to an underprediction of the measured positive carbon a 500 s. The mould geometry and thermal boundary conditions were patterned after an experimental ingot. 47 No. The V segregates in the centre of the ingot arise from equiaxed crystals settling in the core and forming a loosely connected network that can easily rupture due to metallostatic head and liquid being drawn down to feed solidi cation shrinkage. the top centre portion). the increase of carbon content in the melt. cast at Lukens Steel Company. d 15 000 s 4 Predicted velocity vectors (largest vector represents 2·5 cm s – 1) and solid fraction contours (in 20% increments) during solidification of 2·55 m (height) steel ingot32 2002 Vol.e. ultimately resulting in strong positive macrosegregatio n in the last region of the ingot to solidify (i. 5 for carbon and sulphur. and are lled with enriched liquid. the bulk liquid is continually enriched due to the formation of A segregates. While the level and variation of the macrosegregatio n is generally predicted well. Finally. Gu and Beckermann3 2 recently developed a single domain macrosegregation model for a steel ingot. The model considers eleven alloying elements in the steel. about 2·55 m in height. including their diVerent segregation behaviours and eVects on the liquid density.

e. respectively macrosegregatio n near the very top. on a scale of a few millimetres) over an ingot about 2·5 m in height and solidifying over more than 10 h will probably be unavailable for at least the next decade. An earlier numerical study using a ner grid and a smaller domain readily predicted A segregates for a low alloy steel. This enabled them to distinguish between the eVects of melt ow and of sedimentation of crystals on macrosegregation. three-dimensional (3D) eVects and uncertainties in the mushy zone permeability may contribute to some of the disagreement between the measured and predicted macrosegregatio n patterns. Example 2: Continuous casting of steel Many of the macrosegregatio n patterns in continuous casting of steel are similar to ingot casting.3 3 in their simulation of macrosegregation in a steel ingot. Gu and Beckermann3 2 report a CPU time of ‘several weeks’ on an HP J200 single processor workstation for a grid of only 54 (height)Ö 38 (width) control volumes and time steps of the order of several seconds. This may contribute to the underprediction of the measured negative macrosegregatio n over the middle half of the ingot. Similar observations were made by Vannier et al. 47 No. The issue of macrosegregation in the presence of undercooled liquid is addressed in more detail below in the section ‘Coupled microstructure–macrosegregatio n models’. no such disagreement exists for sulphur. They claim to have achieved a mesh independent solution with 20 (height)Ö 12 (width)Ö 12 (thickness) control volumes. Furthermore. The total computational time was on the order of 5 h on a SGI workstation (200 MHz R4400 processor) .3 2 the reader is referred to an earlier study by Olsson et al.3 6 They used experimental information from two steel ingots where the composition was varied to obtain diVerent convection patterns in the mushy zone. They also noted that the LSRE. This can be directly attributed to the coarse numerical grid used in the simulations. this study3 2 revealed a number of shortcomings in macrosegregatio n modelling for a large steel ingot that are not easily overcome.3 5 With respect to the neglect of sedimentation of crystals in the model of Gu and Beckermann. The computational resources required to resolve ow patterns associated with A segregates (i. A corresponding underprediction of the positive macrosegregatio n can be observed near the hot top junction. b sulphur 5 Comparison of measured and predicted macrosegregation variations along vertical centreline of steel ingot:32 Cmix and Cin are local mixture concentration and initial concentration in ingot. They developed a simple equation that predicts the negative macrosegregation due to sedimentation of crystals as a function of the relative height of the sediment zone and the fraction of solid in that zone. might not be valid for large steel ingots because the distance between the free crystals is so large that the assumption of a well mixed liquid in equilibrium with the solid does not hold. Olsson et al. if time is International Materials Reviews 2002 Vol. 32 was obtained. 3) observed in the experimental ingot. A three-dimensional model of macrosegregatio n in steel ingots (considering carbon and sulphur only) was recently reported by Sundarraj and Miller.3 6 also developed a phenomenological model of macrosegregatio n due to A segregates and obtained good agreement with measurements. 5 . equation (1). The neglect of the movement of equiaxed crystals in the model prevents the prediction of V segregates along the vertical centreline. Not shown here is the fact that the model also failed to predict the oV-centreline A segregates (see Fig. In summary.Beckermann Modelling of macrosegregation 247 a carbon.3 4 The failure to predict A segregates will also contribute to the underprediction of the positive macrosegregatio n near the hot top junction.3 5 and much better agreement with the measured macrosegregation patterns of Ref.

where the ferrostatic pressure is suYciently large to cause creep of the solid crust. and Lee et al. and species conservation equations. as already mentioned in connection with stationary ingot casting. Aboutalebi et al. A sulphur print showing a typical macrosegregatio n pattern in a continuously cast steel slab is shown in Fig.4 0 Hence. Yang et al. 6. All three studies included thermosolutal convection and a k–e turbulence model. thermal contraction of the solid as it cools has also been identi ed as an important mechanism in macrosegregatio n formation in continuous casting of steel billets and blooms.4 1 (2D axisymmetric). they concentrate on the thermal strains due to diVerential contractions during cooling. instead. in addition to solidi cation shrinkage ows. momentum.e. with permission. although less pronounced. focused on binary Fe–C alloys. the liquid zone in Fig. This mush must be compressed as the second roll is approached. the conservation equations were not fully coupled. Lee et al. Figure 7 illustrates how bulging between two successive rolls of a continuous casting machine induces the necessary relative liquid–solid velocities. and thus causes positive macrosegregation near the centre of the slab. This centreline segregation is often thought to be caused by bulging of the slab due to inadequate roll containment close to the bottom of the liquid pool. like during compression of a sponge. 7). in addition to the centreline macrosegregation . promising attempts are reviewed below. As the solid network compacts. In addition to bulging. 5 species.3 8 Heat extraction results in growth of the mushy zone between the rolls. Choudhary and Gosh4 5 measured considerable macrosegregatio n at the location of the CET. Several studies modelling macrosegregatio n formation in continuous casting of steel have been reported for situations where bulging and solid deformation are not important. This indicates a strong correlation between the grain structure and macrosegregation. however. melt convection. may be seen in longitudinal sections of continuously cast steel. Interesting theoretical approaches to the problem of macrosegregatio n due to solid deformation were developed by Lesoult and Sella3 9 and Fredriksson and co-workers. the strains and deformations in the solid shell and of the solid inside the mushy zone must be taken into account.) 7 Schematic illustration of how bulging and compression of mush between two successive rolls during continuous casting of steel causes relative liquid–solid flow and. In these studies. energy. None of these studies considers bulging. and El-Bealy and Fredriksson solved their modi ed LSRE only in one dimension across the slab thickness. equation (1). whereas Aboutalebi et al. in order to predict macrosegregatio n in continuous casting of steel. Lesoult and Sella3 9 and Raihle and Fredriksson4 6 were able to explain the formation of positive centreline segregation due to contractions. 2 due to interdendritic strains and deformation of the solid skeleton. Settling of free equiaxed grains can also contribute to macrosegregation. . Yang et al. None of the studies accounts for the presence of a columnar to equiaxed transition (CET) in the structure of the castings simulated. and Lee et al. the only obvious feature is the narrow band of strong positive macrosegregatio n along the centreline. El-Bealy and Fredriksson4 7 and El-Bealy4 8 obtained excellent agreement between measured and predicted uctuating macrosegregatio n patterns near the cast surface due to unsteady heat extraction in the primary and secondary cooling zones of a continuous slab caster. 38) measured with respect to a reference frame that moves with the strand. accounted for multiple International Materials Reviews 2002 Vol.4 6 . 47 No.4 3 (3D) employed standard single domain macrosegregatio n models that solve the coupled mass. from Fig. to account for changes in the volume of the mush element in Fig. macrosegregation (taken. motion of solid below a certain critical solid fraction (using a relative viscosity concept4 4 ). in continuously cast steel billets.4 2 (3D). All three studies claim good correspondence between measured and predicted macrosegregatio n patterns. 1 in Ref.3 9 The liquid ows toward regions of lower solid fraction (i.3 7 A and V segregates. 4 7 They modi ed the LSRE. and sedimentation of free crystals. hence.4 3 also included solidi cation shrinkage ow. Such a comprehensive model does not exist.248 Beckermann Modelling of macrosegregation 6 Sulphur print showing centreline segregation in continuously cast steel slab: segregated centreline is no more than few millimetres in width (courtesy of Ipsco Inc. enriched liquid is expulsed from the mush. and the eVect of soft reduction. Here.

whereas shrinkage alone results in an opposite behaviour. the dendrites are assumed to move with the same velocity as the solid shell. The most advanced model to date of deformation induced macrosegregatio n in the continuous casting of steel has been presented very recently by Kajitani et al. for the grain re ned ingot with an equiaxed structure. Example 3: Aluminium direct chill casting The next example deals with macrosegregatio n in direct chill (DC) casting of aluminium alloys. as illustrated in Fig. while shrinkage ows alone cause a slight reduction (Fig. the bulging of the slab between pairs of support rolls was rst simulated using the Abaqus code. Miyazawa and Schwerdtfeger obtained good qualitative agreement with macrosegregatio n pro les observed in practice. Together. No direct com- a along centreline (distance between rolls is 400 mm). at each pair of support rolls). indicate that bulging alone results in strong positive segregation near the centre and negative segregation in the peripheral regions of the slab. where the strand bulges. The deformation behaviour of a mush consisting of equiaxed grains can be expected to be diVerent from that of a columnar structure.Beckermann Modelling of macrosegregation 249 A more complete model of macrosegregatio n in continuously cast steel billets and blooms that considers thermal deformations of the solid only (and not bulging) was developed by Janssen et al. Finn et al. In the region where the strand is squeezed together (i. has been used in combination with a two-dimensional. Between pairs of support rolls. single domain mixture model by Ohnaka and Shimazu5 1 to predict measured macrosegregatio n pro les in continuously cast steel slabs.e. The solid velocities were obtained by requiring that the product of the solid density and velocity is a constant and taking the solid density as a function of temperature. 5 . negative segregation in the periphery. b across slab thickness after six rolls (fs =0·37 at centreline. 9.. 8b. The solid velocities across the deforming mushy zone were then obtained by using the bulging pro le as input and assuming a linear variation between the solidus isotherm and the centreline in the compression areas. A similar approach. the mushy zone is more permeable and allows for advection ( by buoyancy driven ow) of solute rich liquid toward the centreline. Whereas Lesoult and Sella3 9 express their governing equations in a Lagrangian form. 8a). shown in Fig. Typical macrosegregatio n patterns have been measured by Finn et al. where the velocity of the solid phase due to bulging is prescribed through simple algebraic relations. showed that. and a mass conservation equation that accounts for shrinkage were solved together iteratively. and the International Materials Reviews 2002 Vol. and a second negative zone adjacent to the positive centreline segregation. the solid phase is assumed to move with a horizontal velocity that depends linearly on the solid fraction.5 2 for Al–4·5 wt-%Cu cylindrical ingots solidi ed both with and without grain re ner. two phase. the less permeable mush reduces such advection. The solute conservation equation. Only solid deformation in the thickness direction across the mushy zone was taken into account. maximum bulging of 0·1 mm) 8 Predicted macrosegregation patterns during continuous casting of steel: Co is initial carbon content (with permission from Ref. thus producing positive centreline macrosegregation . 38) parisons with experiments were oVered by Kajitani et al.4 9 use a more common Eulerian formulation. 47 No.3 8 Based on two-dimensional heat ow calculations.4 9 within the framework of a two phase computational uid dynamics code (Phoenics). particularly when the solid fraction is high and coherency is increased by coalescence of neighbouring columnar dendrite branches. The macrosegregatio n pro les after six rolls across the slab thickness. while neglecting thermal contractions. Janssen et al.3 8 owing in part to convergence problems at low and high solid fractions. It was noted that improved descriptions are needed for the deformation behaviour of the solid skeleton in the mushy zone. On the other hand. in the case without grain re ner (columnar structure) .3 8 It can be seen that the solute content along the centreline increases with each bulging event. bulging and shrinkage produce strong positive segregation at the centre. Miyazawa and Schwerdtfeger5 0 developed an early model of macrosegregatio n due to bulging and contractions in continuously cast steel slabs. Figure 8 shows typical carbon macrosegregatio n pro les computed by Kajitani et al. by solving combined mass and species conservation equations together with Darcy’s law. The model should also be improved by accounting for the inner equiaxed zone usually present in continuous casting of steel. Darcy’s law. They obtained good qualitative agreement between calculated and measured radial macrosegregatio n pro les for carbon and phosphor in a bloom.

respectively 9 Simulation of direct chill (DC) casting of Al–4·5 wt-%Cu cylindrical ingot53 remaining shrinkage driven ow produces negative centreline segregation. c predicted relative velocities and solid fraction contours (from 0·0 to 1·0 in intervals of 0·2) inside mushy zone near mould for simulations with and without grain refiner. it can be seen that there exists a shrinkage driven ow towards the eutectic front. As measured in the experiments.250 Beckermann Modelling of macrosegregation a schematic illustration of simulation domain (h is heat transfer coefficient). the simulations showed that the buoyancy driven ow in the grain re ned case produces positive centreline macrosegregatio n (Fig. the grain sizes and dendrite arm spacings needed in the permeability model were taken from the experiments. In both cases.5 4 – 5 8 The above comparison shows that the macrosegregation model predicts the correct macrosegregation trends and magnitudes. In the grain re ned case (Fig. the shrinkage driven ow.52 b predicted profile in simulation without grain refiner 10 Radial macrosegregation profiles in DC casting of Al–4·5 wt-%Cu cylindrical ingot53 . b. and illustrates the strong eVect of the microstructure . 9b and 9c for the grain re ned and non grain re ned cases. Results from numerical simulations of these experiments are shown in Figs. in the non grain re ned case. 9b). the extent of positive inverse segregation at the ingot surface is somewhat underpredicted. However.. On the other hand. 10a). together with the weakness of the buoyancy driven ow. results in negative centreline macrosegregation (Fig. 5 a comparison of measured and predicted profiles in grain refined experiment of Finn et al. A truly predictive model would also need to predict the International Materials Reviews 2002 Vol. Because exudation was neglected. 47 No. 10b). with a more permeable mushy zone. strong positive macrosegregatio n due to shrinkage driven ow and exudation was measured at the ingot surface.5 3 Close-ups of predicted relative liquid velocities and solid fraction contours inside the mushy zone are shown in Fig. a relatively strong thermosolutal buoyancy driven ow towards the ingot centreline is present in the lower solid fraction region of the mush. In both cases. respectively. Surface macrosegregation in aluminium DC casting has been modelled successfully by Mo and co-workers. 9 and 10.

The novel feature of this model is that phase equilibrium was calculated at each time step. simulation of the transport phenomena during solidi cation of the complex multicomponent alloys used in actual DC casting processes. using a thermodynamic subroutine for multicomponent nickel base superalloys. F. The lling of these channels with equiaxed grains. which are then observed as freckle-like chains of equiaxed crystals in the otherwise single crystal columnar structure. has recently been simulated by Gu et al. a number of light elements (such as aluminium and titanium) are rejected into the liquid. The channels are con- 12 a predicted velocity vectors (largest vector represents 6·3 mm s – 1) and solid fraction contours (in 20% increments) and b macrosegregation pattern (Ti concentration normalised by initial concentration in equal intervals between 0·87 and 1·34) showing freckle formation during upward directional solidification of single crystal Ni base superalloy in 5Ö 15 cm rectangular domain65 tinually fed by melt ow through the surrounding mush. Furthermore. United Technologies Research Center) microstructure. 12. have recently been reported by Joly et al. 6 2 Further discussion regarding the modelling of macrosegregatio n in the presence of moving equiaxed dendrites can be found in a subsequent section. leading to open channels in the mush through which liquid streams upwards into the superheated region of the mould.3 using a nite volume method. 59). leading to strong solutal buoyancy forces in the mushy zone. and by Combeau et al. The remelting necessary to generate open channels occurs according to the mechanism mentioned in the sixth item discussed in connection with the LSRE. None the less. Ref.Beckermann Modelling of macrosegregation 251 11 a freckles in single crystal Ni base superalloy prototype blade and b higher magnification of single freckle: freckles are about 1–2 mm wide (courtesy A. these channels are lled with dendrite fragments. and some heavy elements (such as tungsten) are preferentially incorporated into the solid. Preliminary simulations with a two phase model support this conclusion. Giamei. leading to the appearance of freckles in a fully solidi ed casting.6 1 .6 3 concurrently with the prediction of grain structure development. 47 No. it has been argued that the transport of broken/detached solid fragments and equiaxed grains from the mould region to the ingot centre tends to cause negative centreline macrosegregation in aluminium ingots (see e. The open channels in the mush through which enriched liquid streams upward into the bulk melt are predicted realistically. Typical predictions of freckling in upward directional solidi cation of a superalloy from a twodimensional micro-/macrosegregatio n model6 5 are presented in Fig. suitable for validation of such future models. these buoyancy forces can trigger convection cells.6 0 The eVect of free oating dendrites on macrosegregatio n in DC casting of aluminium alloys has also been modelled recently by Vreeman and co-workers.6 6 Simulations of freckling in directional solidi cation of superalloys have also been performed by Felicelli et al. Although three-dimensional calculations have been reported by Neilson and Incropera7 0 and Felicelli International Materials Reviews 2002 Vol.6 4 Example 4: Single crystal nickel base superalloy casting A particularly severe and technologically important macrosegregatio n phenomenon is freckling in directional solidi cation of single crystal nickel base superalloys.g. Despite the presence of a stabilising thermal gradient. Later. Figure 11 shows photos of such freckles in an experimental superalloy casting. Coupled macrosegregatio n and grain structure measurements for a commercial DC cast 5182 aluminium alloy. Components with freckles are rejected. equation (1). is still lacking.. During solidi cation of nickel base superalloys.6 7 – 6 9 using a nite element method. 5 .

sample simulation results from the study of Feller and Beckermann8 0 are shown in Figs. Figure 14 shows how additional simulations can then be used. whereas the particles in the single phase liquid region are generally mobile. 16 and 17. In this International Materials Reviews 2002 Vol. Continued in ltration then forces the remaining solute rich liquid past the bres that already have a solid layer on them.7 2 extension of such simulations to actual cast components is diYcult. The macrosegregatio n pattern predicted by Mortensen and co-workers was in good agreement with their experimental measurements. The critical Rayleigh number derived from experiments agrees with the value corresponding to freckle initiation in the simulations. In order to illustrate the eVect of the particles on macrosegregation . 75 still need to be reconciled.g. Simulations were conducted for solidi cation of an Al–7 wt-%Si alloy with SiC particles inside the cavity shown in Fig. and 20 vol. a simpler freckle predictor was developed that relies on the calculation of a local mush Rayleigh number in casting simulations that consider heat transfer only. 15. the alloy that rst solidi es on the bres is low in solute. For this reason. It is also assumed that the particles are entrapped by the advancing dendrites. with the largest positive macrosegregatio n observed at the last in ltrated point in the mould. in turn.7 3 Very recently. . Auburtin et al. 80 in Fig. instead of experiments.7 5 performed experiments on freckle formation in superalloy castings at various angles to the vertical. The modelling results of Ref. respectively 14 Simulation results for dependence of critical mush Rayleigh number on inclination of domain with respect to gravity for freckle initiation in directional solidification of single crystal Ni base superalloys73 et al. 5 process.252 Beckermann Modelling of macrosegregation 13 Variation of mush Rayleigh number with local thermal conditions in experiments of Pollock and Murphy 74 on directional solidification of single crystal Ni base superalloys: no freckles form below critical mush Rayleigh number of 0·25 (Ref.-% particles with a diameter of 14 mm (case II ). The symbols in Fig. This is illustrated for the rectangular domain modelled in Ref. 73 and the experimental data of Ref. Feller and Beckermann8 0 investigated macrosegregation and particle redistribution during solidi cation of metal matrix particulate composites (MMPCs). 73): G and R are temperature gradient and isotherm speed. 7 4 Based on their measurements of freckle occurrence. resulting in solute enrichment downstream of the path followed by the in ltration front.-% particles with a diameter of 100 mm (case III ).7 1 . leads to macrosegregation in the matrix of the cast composite.2 7 Figure 13 shows the variation of the Rayleigh number with the local thermal conditions (temperature gradient and isotherm speed). and also developed a freckle criterion that shows good correlation with their data. due to the large computer resources required to resolve such small macrosegregation features over the scale of the entire component. Example 5: Casting of metal matrix composites Macrosegregation is also of concern in the casting of metal matrix composites (MMCs). to determine the variation of the critical Rayleigh number with the inclination of the casting with respect to gravity. This. a critical Rayleigh number for freckle formation can be identi ed.. 20 vol. The three phase (i. Mortensen and co-workers 7 6 – 7 9 developed a model and performed experiments for unidirectional pressure in ltration of a brous preform by a binary Al–Cu alloy. 47 No. SiC or graphite) are dispersed into the liquid alloy (typically aluminium alloys) before introduction into the mould and solidi cation. 13 correspond to the experiments of Pollock and Murphy. 15.7 3 This Rayleigh number is based on the mean permeability and the liquid density inversion over the height of the mush zone. Three cases were considered: no particles (case I). The particles may be segregated to diVerent regions of the casting due to settling/ oating. the reinforcing particles (e. In casting of MMPCs. as well as due to interactions with the advancing mushy zone. solid–particle–liquid) model developed by Feller and Beckermann assumes that the solid matrix is rigid and stationary at all times.e. and that no particle rejection occurs at the interface between the mushy zone and the liquid region.

with strong positive segregation present along the insulated walls of the cavity ( bottom and the right-hand walls). c case III (large particles) 16 Predicted bulk solid fraction (particle plus solid matrix) distribution and liquid velocity vectors during solidification of metal matrix particulate composites in domain shown in Fig.Beckermann Modelling of macrosegregation 253 15 Schematic illustration of domain and boundary conditions used in simulation of solidification of Al–7 wt-%Si metal matrix composite with SiC particles: different degrees of shading indicate particle concentration at intermediate time80 Figure 16 shows the computed bulk solid fraction. without particles. 16c contains a packed bed of particles that settled before the mush could entrap them. 15 (Ref.8 0 For the unreinforced alloy in case I. 17 for all three cases. The computed macrosegregatio n patterns (of silicon) at the end of solidi cation are shown in Fig. the lower third of the cavity in Fig. macrosegregatio n is widespread. and several channel segregates present in the centre. very little particle settling occurs in case II. de ned as the sum of the solid and particle volume fractions. and melt velocities in this region are of a magnitude similar to those in case I. as witnessed by the uniformity of the bulk solid fraction in the solid free region. with the small particles. Consequently. This can be explained by the fact that the melt– particle mixture viscosity is more than twice as high as the melt viscosity without particles. because the relative velocity between the particles and the liquid is very small for the 14 mm particles. The upper half of the cavity is virtually free of particles in case III. Furthermore. 80) International Materials Reviews 2002 Vol. On the other hand. and liquid velocity elds at an intermediate time during solidi cation. are much smaller than in case I. 47 No. b case II (small particles). It can be seen that the liquid velocities in case II. for cases II and III a case I (unreinforced). This macrosegregatio n is caused by thermosolutal convection. the large size of the particles (100 mm) results in much greater relative velocities. In case III. 5 .

Figure 18b shows the predicted carbon macrosegregatio n pattern in the fully solidi ed casting. The pressure–velocity coupling was handled using either a modi ed version of the Sola-VOF algorithm or the Simplec scheme. with particles). 8 3 and (v) rheological behaviour of the particle–melt mixture. This can be explained by the reduction in the convection velocities due to the presence of the particles.8 2 (iv) microstructural changes due to interactions between the matrix and the reinforcement. especially if crystals nucleate on moving particles. Some limited experimental validation was presented for a simple block casting with a central top riser. While the previous work succeeds in predicting a number of important phenomena relevant to macrosegregation in MMCs.254 Beckermann Modelling of macrosegregation a case I (unreinforced). the solid fraction as a function of temperature was speci ed as model input. The governing equations were discretised using a control volume based nite diVerence scheme. depending on which scheme required less iteration in the previous time steps. A three-dimensional model was International Materials Reviews 2002 Vol. 34. shrinkage driven ow and solid movements were neglected. Convection in the low solid fraction portion of a mushy zone is the primary cause of macrosegregation in unreinforced alloys. 85 leads to some inaccuracies in the macrosegregation predictions. 8 4 Certainly. In order to reduce the computational eVort. particularly at the riser/casting junction where . 80) (i. Example 6: Complex steel casting The nal example illustrates that macrosegregatio n due to interdendritic ow can now be simulated for the complex shaped castings commonly encountered in foundry practice.8 5 Figure 18 shows an example of macrosegregatio n calculations for a large (96 300 kg.e. maximum dimensions 2·65Ö 2·1Ö 3·7 m) production steel casting poured in a green sand mould with an insulated riser and water cooling lines run through the hole in the centre of the casting. 47 No.8 5 One half of the casting was enmeshed with 113 760 control volumes of which 21 420 were metal volumes. These macrosegregation predictions have not yet been compared to measurements. 15 (Ref. but also strongly decrease the permeability in that portion of the mush where the solid fraction is low and the permeability would be high in the absence of particles. macrosegregation can be prevented by the addition of small particles to the melt. Hence. It can be seen that the ow has created carbon rich regions around the hole and in the centre portion of the riser. Figure 18a is a plot of the predicted convection patterns in the bulk liquid and mushy zones during solidi cation. Some of the issues still to be addressed include: (i) the eVect of a nonstationary solid phase. 5 developed8 5 based on the formulation for macrosegregation formation in multicomponent steel of Ref. macrosegregatio n is negligible when reinforcement is present.8 1 (iii) macroscopic particle movement within the low solid fraction portion of the mushy zone. Since the upper right-hand portion of the cavity is particle free in case III due to settling. instead of calculating it from the temperature–concentration coupling in the mushy zone through phase equilibrium. It can be expected that the neglect of shrinkage driven ow in the model of Ref. b case II (small particles). A typical simulation took less than one day on a medium range computer workstation. additional experiments are needed to clarify these and other issues. (ii) particle rejection at the mush/liquid interface when the solid microstructure in the mush is relatively dense. The particles not only increase the viscosity. In addition. especially when it is nonNewtonian. c case III (large particles) 17 Predicted final macrosegregation patterns for solidification of metal matrix particulate composites in domain shown in Fig. there are numerous issues that deserve further research attention. the extent of macrosegregation in that region is similar to case I. Particularly strong ow exists around the hole in the casting.

the dendrite arm spacings are the primary microstructural parameters that aVect the permeability.9 7 in their measurements of the permeability of equiaxed Al–Cu alloys. Even though much work has been performed to investigate the dependence of the permeability on the microstructure. In the equiaxed case. 47 No. has a profound in uence on macrosegregation. such as during eutectic and peritectic reactions or transformations . For example. since macrosegregatio n changes the local average alloy composition. 85. In reality. the coupled prediction of the microstructure together with convection and macrosegregatio n is still lacking in many respects. 5 . and can now eVectively be dealt with. resulting in macrosegregatio n by settling (or otation). Since diVerent amounts of solute are rejected into the interdendritic liquid depending on the microsegregation behaviour. where there is negligible solute diVusion in the dendrite arms. in the continuous casting of steel.e. most castings ( both continuous and static) feature zones of equiaxed grains. microsegregation) processes that. 9 6 also investigated this dependency of the permeability on the structure of equiaxed grains. Conversely. Further progress in macrosegregatio n modelling can only be made if the microstructure is predicted as well. The situation is further complicated by the fact that small dendrite fragments or equiaxed grains can move in the melt before forming a coherent solid. Arnberg et al. however. Furthermore. DiVerent microstructures also result in diVerent microscopic solute rejection (i. Coupled microstructure– macrosegregation models The examples in the previous section illustrate that the microstructure of castings. where both the solid and liquid are solutally well mixed and in equilibrium on the scale of the dendrite arms. the melt must undercool before nucleation and growth of solid microstructures can proceed. 34. Many macrosegregatio n features cannot be predicted without detailed consideration of the microstructure.9 3 . when the dendrites start to impinge and form a continuous solid network. and much work has been performed to measure or calculate the permeability of such systems. even for multicomponent alloys with diVerent mass diVusivities for each solute. It is well known that for a given alloy a nely dispersed microstructure promotes Lever rule type microsegregation behaviour. the resolution of small scale macrosegregatio n features (such as A segregates or freckles) in large castings is also beyond the current capabilities of the model of Ref. 1 1 3 estimated that. As in the steel ingot example. with less interfacial area per unit volume. it will induce diVerent microsegregation behaviours. Duncan et al. in turn. This was shown theoretically by Wang et al. columnar or equiaxed). 65. On the other hand.9 9 measured the coherency point in equiaxed solidi cation. accounted for the large in uence of coarsening of the microstructure . there is an undercooled liquid zone several metres in height. In directional solidi cation (i. They also noted that coarsening rates were enhanced by the melt ow. columnar dendritic growth).9 2 and veri ed experimentally for aluminium alloys by Nielsen and co-workers.e. implying that the microstructural length scale that controls microsegregation is not a constant. diVerent macrosegregatio n patterns will result in the presence of ow. such as the dendrite arm spacings and the type of grain (i..9 2 provided a correlation for the interfacial drag in equiaxed solidi cation that covers the entire range from a single particle to a packed mush. The most obvious example of this interdependency is the variation of permeability with microstructure. A International Materials Reviews 2002 Vol.1 0 6 – 1 1 1 None the less.. 56. or are fully equiaxed. The drag coeYcients of single equiaxed dendrites have recently been measured and correlated. all the previously reviewed macrosegregation models assume that the liquid is in equilibrium and locally well mixed (as in the Scheil or Lever rule models). b calculated final C macrosegregation pattern where concentrations range from 0·21 wt-%C (dark shade) to 0·25 wt-%C (light shade) 18 Simulation of macrosegregation formation in large (2·65 m height) steel casting85 the shrinkage ow velocities are largest. all castings with equiaxed structures feature correspondingly large undercooled liquid regions during solidi cation. tends to result in Scheil type microsegregation behaviour. M’Hamdi and co-workers1 1 2 . a coarser microstructure. Some micro-/macrosegregatio n models have been developed for situations where multiple phases are forming. The situation is further complicated by the fact that the dendrites in the mushy zone undergo considerable coarsening during solidi cation. there are often large uncertainties in the prediction of the microstructure that underlies the micro-/ macrosegregatio n processes during solidi cation.Beckermann Modelling of macrosegregation 255 a predicted liquid velocity vectors (largest vector represents 1 cm s – 1 ) in mush (light shade) and liquid (dark shade) regions at time when casting is 50% solidified. 100–105). both the grain density and the structure of individual grains in uence the permeability.8 6 – 9 1 However. 9 4 Poirier and co-workers9 5 . 9 8 Wang et al. The topic of coupled micro-/macrosegregation has recently received much research attention (Refs.e. in uence macrosegregation . This section highlights recent eVorts to develop coupled microstructure– macrosegregatio n models.

The eVects of grain movement.1 2 5 Gandin et al. including the derivation of the governing equations. etc. the diVerence between the average and microscopic value of a variable). a reduced grain density. interfacial area concentration. such as the average solute concentration in the solid in the REV. with simultaneous melt convection and grain movement. Once the superheat is dissipated.1 1 4 – 1 2 0 Solute rejection from equiaxed dendrites growing and moving in an undercooled melt has been investigated experimentally by a few researchers. and hence larger grains. An example of a macrosegregation model that accounts for the growth of equiaxed grains in an undercooled melt is presented later in this section. etc. solute diVusion-dispersion tensor. During growth of dendrites or other structures in an undercooled melt. in fact. 47 No. but there has. The averaging results in separate macroscopic conservation equations for each phase and averaged interface conditions that can be solved using a single domain approach for the entire casting. on macrosegregatio n are illustrated in Fig. this point may seem unimportant. 5 form of some of the equations. the microscopic (or point) conservation equations for each phase are formally averaged over a representative elementary volume (REV). While it is beyond the scope of this work to present the details of the coupled microstructure–macrosegregation models that have been developed during the past decade.1 2 7 However. drag coeYcients.256 Beckermann Modelling of macrosegregation large undercooled zone has also been predicted for DC casting of aluminium. local diVusion lengths. Therefore. permeability. No such models are available for other structures. The application of the multiphase/multiscale model of Wang and Beckermann to predict macrosegregation during dendritic alloy solidi cation. and terms accounting for the latent heat. Several models of this eVect have been developed for purely diVusive conditions where ow can be neglected. 19. permeability. The reader is referred to other reviews1 . several (inverted) A segregates can be observed in the lower central region. such as columnar dendrites. and of diVerent nucleation rates. 1 2 9 – 1 3 1 and the extended continuum model of Ni and Incropera. etc. 2 for a thorough description of the models. the remainder of this section brie y reviews their origin and then presents a single example of model predictions. This approach has recently been used9 1 . and no channel segregates are predicted. The key advantages of a formal averaging procedure are: d The macroscopic variables. because the cold.1 3 0 An Al– 4 wt-%Cu alloy solidi es in an equiaxed fashion inside a rectangular cavity cooled from the left sidewall.1 2 1 – 1 2 4 Severe solute pluming was observed in the wake of the crystals. its consideration is crucial when modelling coupled microstructure–macrosegregation development.. 2.1 3 3 In the derivation of the models by Beckermann and co-workers. grain density. this can be attributed to the continuous sedimentation of grains out of this region.1 3 7 can be derived for the uctuating variables (i. Because the local nucleation rate was set to a constant value.e. containing freely moving and growing grains.6 0 Even in purely columnar growth. which can then be solved for a small representative domain with periodic boundary conditions. The interface between the bed and the overlying region of undercooled melt. This . are exactly de ned in terms of the microscopic pro les. Free grains nucleate rst near the cooled sidewall. Overall. 2 3 mentioned in the previous section. In addition. dendrite tip growth. which in turn in uences macrosegregation. Modelling of microstructure formation in castings has been summarised by Rappaz. corresponding to the grain density plots of Fig. they settle and form a xed bed of equiaxed crystals at the bottom of the cavity.1 3 5 d The averaged conservation equations explicitly contain microscale parameters such as the phase volume fractions.1 3 8 to determine various eVective transport properties (e. all spatial variations in the grain density are directly due to grain movement. Figure 20b shows the macrosegregatio n pattern for a simulation with moving grains. is characterised by a relatively sharp gradient in the grain density. can be observed in the upper part of the cavity. Later.e. Although the level of undercooling may be no more than a few degrees in traditional casting processes. nucleation rate. eVective thermal conductivity. Upon complete solidi cation. The grain density in the bed itself is invariant. been quite a controversy over the correct International Materials Reviews 2002 Vol. the microscopic solute rejection behaviour (and also the solid fraction variation) can be very diVerent from that during solidi cation with a well mixed interdendritic liquid in equilibrium. 20. Thermosolutal convection causes strong positive macrosegregation along the bottom wall and right sidewall of the cavity. This bed continues to grow in height as more free grains settle downward.1 3 4 and some confusion over how to best incorporate. as well as of those by Poirier and co-workers2 2 . for example. the macrosegregatio n is less severe. is illustrated in the following example. d Each term in the averaged equations has a clear origin. columnar dendrite tip undercooling in macroscopic models. and are swept into the central part of the cavity by the thermosolutal melt ow.1 3 2 . the number of grains per unit volume).) for a columnar mushy zone. there is a potentially large undercooled liquid region in front of the primary dendrite tips. Figure 20a corresponds to a simulation where a stationary solid phase was assumed. solute rich melt is heavy and ows downward during solidi cation. especially when the thermal gradient is low (such as in sudden cross-section expansions in casting of columnar turbine blades). the only models to date that consider concurrent microstructure and micro-/macrosegregatio n development (including melt convection and grain movement ) are the two phase model of Ni and Beckermann. the grain density would have been uniform in the absence of grain movement.g. such as that shown in Fig. Figure 19 shows the predicted evolution of the grain density (i. when the grains have grown to a suYcient size. the melt surrounding the free equiaxed grains becomes undercooled and the grains will grow.1 2 6 and Rappaz et al. mass exchange coeYcients.1 2 8 the multiphase/multiscale model of Wang and Beckermann. naturally arise from the averaging procedure. d Local ‘closure problems’1 3 6 .

Beckermann Modelling of macrosegregation 257 a 10 s. d 100 s 19 Predicted evolution of grain density during equiaxed dendritic solidification of Al–4 wt-%Cu alloy with grain movement inside 5Ö 10 cm rectangular cavity cooled from left sidewall130 a stationary grains (low nucleation rate). 19). 47 No. c 50 s. b 30 s. c moving grains (high nucleation rate) 20 Effect of grain movement and of different nucleation rates on predicted macrosegregation patterns in equiaxed dendritic solidification of Al–4 wt-%Cu alloy with grain movement inside 5Ö 10 cm rectangular cavity cooled from left sidewall130 International Materials Reviews 2002 Vol. 5 . b moving grains (low nucleation rate: corresponds to Fig.

Concentrated research eVorts are underway to clarify these and International Materials Reviews 2002 Vol. resulting in less relative motion between the solid and liquid phases than in the case of a stationary solid. such as that shown in Fig.1 3 1 While the multiphase/multiscale model appears to provide a suitable framework for coupled microstructure–macrosegregatio n modelling. grey shades indicate Cu concentrations in liquid (10 equal intervals between 4·861 and 4·871 wt-%Cu) and solid (4 equal intervals between 0·682 and 0·6805 wt-%Cu)147 can be attributed to the fact that both the solute rich liquid and the solute poor solid ow downward. 139).1 4 7 They used the phase eld method to model the ow through. many uncertainties still exist. only approximate models are available for the eVects of ow on grain generation (including fragmentation) . 20b and c). where a higher nucleation rate was employed in the simulation. results in an increase in the permeability. much progress has been made in the direct numerical simulation of solidi cation microstructure development. In particular.1 4 0 – 1 4 2 Beckermann et al. in turn. Direct numerical simulation During the last decade. the solute rich melt is heavier. One important goal would be to predict macrosegregatio n in castings that feature a columnar to equiaxed transition. 47 No. Refs. dendrite tip growth. 121–124. it is important to realise that the ripening dynamics are in uenced . This. etc. but much work remains. dendrite tip growth.258 Beckermann Modelling of macrosegregation a–d simulation under purely diffusive conditions. because such investigations are diYcult to perform experimentally owing to the small scale of the phenomena involved. and ripening of. However. 2.1 4 4 – 1 4 6 microsegregation. Ripening (or coarsening) causes the solid/liquid interfacial area to decrease with time. 21. with and without ow. An example of a microstructure simulation that has a direct bearing on modelling of macrosegregatio n is the recent study by Diepers et al. two phase rheology. Even less macrosegregatio n is observed in Fig. resulting in less severe overall macrosegregatio n (Fig. the phase eld method has been used to predict various complex growth patterns from rst principles. A comparison of the multiphase/multiscale model with experiments conducted using the transparent model alloy NH4 Cl–H2 O revealed good agreement between the measured and predicted ow patterns and crystal settling rates during solidi cation. For example.1 4 3 have recently extended the phase eld method to include melt convection. Representative two-dimensional simulation results. The higher nucleation rate results in smaller grains that closely follow the liquid motion until they pack. The classical macrosegregatio n pattern in a steel ingot (Fig. 3) is an example of that eVect. in the Al–Cu system of Figs. e–h simulation with pressure drop of 0·2 N m – 2 applied in north–south direction (arrows are liquid velocity vectors) 21 Phase field simulations of evolution of microstructure due to Ostwald ripening in Al–4 wt-%Cu alloy mush where solid fraction is constant at 0·207: domain is 0·81 mm square with periodic boundary conditions.g. 5 other questions (see e. However. 19 and 20. fragmentation. and other ow related issues that have hampered coupled microstructure–macrosegregatio n modelling in the past. 20a) is partially ‘cancelled’ by the settling of solute poor solid. are shown in Fig. 20c. an adiabatic Al–4 wt-%Cu alloy mush volume element. the settling of free grains causes negative macrosegregation at the bottom and positive macrosegregatio n at the top of the solidi cation domain. black contour lines show solid/ liquid interface. in alloys where the solute rich liquid is lighter and the solid is more dense than the melt of the original composition. It should be noted that. This modelling work is of particular importance. and the positive macrosegregation at the bottom of the cavity due to melt ow (Fig. This new capability will allow for the detailed investigation of the eVects of ow on the structure of the mush.

but the ow also in uences the evolution of the microstructure. wang: in ‘Annual review of heat transfer VI’. 1998. B. The writing of this review was made possible. all the microscopic phenomena would be directly taken into account.Beckermann Modelling of macrosegregation 259 by convection. c. greater eVorts should be made to reduce numerical inaccuracies and to resolve the highly three-dimensional. and found that the permeability normalised by the square of the interfacial area per unit solid volume is constant in time. combeau. PA. Begell House. Vol. direct numerical simulations of solidi cation with ow on a microscopic scale using. Acknowledgements The author would like to thank his present and former graduate students and collaborators who have contributed to some of the work reviewed in this paper. they were able to correlate the permeability with the various microstructural parameters present.1 4 7 also point the way toward the future of macrosegregatio n modelling through numerical simulations on the microscopic scale. and g. on the other hand. Warrendale. Based on the present review. A few of the outstanding research issues are: d nucleation in a convective environment d fragmentation and transport of fragments originating in the mushy zone or at outer surfaces d eVects of ow on the growth rates of the dendrite tips. and most simulations reported in the literature are restricted to two dimensions. Direct numerical simulations of ows through a unit cell inside a mushy zone in order to determine permeabilities have also been reported in Refs. in those studies the microstructure was taken to be static. lesoult: in ‘Modeling of casting. for example. incropera: in ‘Advances in heat transfer’. TMS. Beckermann). Some of these issues are already being addressed by various research groups (Refs. It has become clear that numerous macrosegregation phenomena cannot be predicted without detailed consideration of the evolving microstructure. G. The resolution of these issues will not only require even more complex models. Tien). j. appolaire. While some successes have been reported in predicting measured macrosegregation patterns in industrially relevant casting processes. further progress in macrosegregation modelling can only be made if the microstructure and grain structure transitions are predicted as well. the basic phenomena that should receive increased research attention can be summarised as follows: d macrosegregation in multicomponent alloys. Furthermore. 149–155) as well as by developers of commercial casting simulation codes. 70–72. Diepers et al. 156. In the presence of ow. These advances in modelling are especially important in light of the diYculties in performing accurate experimental measurements on a microscopic scale. 148. However. macrosegregatio n modelling requires a quantitative understanding of the convective interactions on a microscopic scale. welding and advanced solidi cation processes VIII’.g. The simulations of Diepers et al. In other words. 1996. the resolution of melt ow patterns in numerical solutions of macrosegregatio n models has often been inadequate. they would be realistic for scales of the order of several millimetres (requiring of the order of 109 to 101 2 grid points in three dimensions). columnar to equiaxed) d macrosegregation due to deformation of the solid or mush d macrosegregation due to movement of solid d micro-/macrosegregation in the presence of undercooled. 88–91. the phase eld method. beckermann and c. transient. but also careful validation using specially designed experiments. 1995. b. New York. By performing simulations with diVerent volume fractions of solid inside the element. and concurrent coarsening or solidi cation was not modelled. In this way. While research in the area of solidi cation microstructures has traditionally focused on the prediction of phases and growth patterns in a diVusive environment. Thus. (ed. International Materials Reviews 2002 Vol. 5 . Academic Press. p. d eVects of ow on the evolution of dendrite arm spacings and other microstructural length scales d two phase rheology of a melt laden with equiaxed grains. 2. C. (ed. through the support of NASA under Contract no. in part. Conclusions and future research needs Modelling of macrosegregatio n has experienced major advances since the pioneering work of Flemings and co-workers in the mid-1960s. Poulikakos). and sometimes turbulent ow patterns present in often complex shaped castings. However. in accordance with classical ripening theories. such microscopic simulations of macrosegregatio n formation cannot be performed on the scale of an entire casting. taking into account the formation of multiple phases d macrosegregation in the presence of grain structure transitions (e. the microstructure of the mush governs the ow. With the availability of increased computing power and more eYcient numerical techniques. D. with some evolution in computing power. 157). convecting liquid. Future research on any of the above issues will increasingly rely on rst principles. 245–256. h. CA. also measured the permeability of the mush volume element from the simulation results. 231–338. 6. p. the interfacial area decreases with the cube root of time when there is no ow. 47 No. the interfacial area decreases with the square root of time. in many cases. veri ed that. L. While knowledge of the permeability of an evolving microstructure is important for present macrosegregation models. there are still numerous areas where further development is required. NCC8–94. y. References 1. Presently. 3. Thomas and C. prescott and f. San Diego. the simulations of Diepers et al. etc. eutectic front. 115–198. However. (ed. and the results could be used to assess and improve models that are based on averaging (such as those reviewed in the previous two sections). the use of such microscale simulations to predict macrosegregation on the scale of an entire casting will probably not be an option until at least the year 2050 (Refs.

welding and advanced solidi cation processes IX’. h. 123–133. W. c. el-bealy: Scand. a. 57. 1533–1542. w. bennon and f. B. c. 145–161. tirard-collet: Metall. daloz. 711–719.. hong: in ‘Modeling of casting. p. M ater. drew: A nn. 1997. J. d. felicelli. a. T rans. T rans. (ed. dubus. T rans. c. 21–35.. 1990. f. e. T rans. j. Shaker Verlag. 1983. d. Merton C. c. 1998. Beckermann). heinrich: Metall. A . 83. T rans. M at er. 875–898. Sparrow). Fluid M ech. s. 1997. B. (ed. zhang. m. j. Iron and Steel Institute of Japan. beckermann.). T rans. 1997. m. c. w. Mater. T rans. 1. 1997. 21. beckermann. x. h. e. T rans. 28B. 71. j. 338–345. a. 23. r. 1517–1531. j. A . 1994. reddy and c. a. 1. s. B. thevik and a. thevik and a. cornelissen. j. m. 1989. 1987. haug. 1999. 1978. 10. 2533–2537. 35. feller and c. 32. keane. 38. r.. ASME. 78. thevik. 23. heinrich: J. s. A IM E. 239. (ed. 5. A . r. Mater. 1992. ohnaka and t. c. and w. j. 331–339. a. d. and c. 62. 22. 1999. j. 41. 1998. T rans. 236. 46. and w. and g. T rans. viskanta: A ppl. mehrabian and m. m. ganesan and d. 2161–2170. on ‘Solidi cation and materials processing’. 30. mo. 1449– 1461. 5. keane. 1996. A . v. P. i. 1970. j. c. 1992. and m. p. m. 217–229. j. 46. 4. miller: Proc. g. poirier. PA. gu. r. 1999. felicelli. 11. worster: A nn. hasan. 173–181. flemings: Metall. T rans. fredriksson: Scand. s. r. flemings: M etall. AIME. l. c. 28A. j. 91–122. fredriksson: Can. 37. 12B. murphy: Metall. j. and j. J. and a. 21B. M. 82. n. r. h. l. r. A173. c. 1997. and r. 833–841. A . han and j. 52. 18. Mech. T rans. thevik: Metall. 1993. sundarraj and g. 23A. 693–699. 167–178. e. m. j. 1–27.. bennon and f. (3). A . heinrich: J OM . 45. 66. R. and m. R es. 1995. B. flemings: ISI J Int.. kou. j. v. T rans. 140–150. and p. c. T rans. h. flemings: Metall. beckermann: Metall. c. 48. T rans. 1994. mortensen: Metall. d. poirier. r. 1995.. incropera: Int. j. gu. A . rabenberg: A ppl. m. d. 30. deng. k. y. 1990. 20. 879–888. michaud. and g. and m. el-bealy and h. voller: in ‘Advances in numerical heat transfer’. heinrich. A ppl. t. P. M ater. 847–859. hunt: ISIJ Int. J. 311–316. c. flemings: in ‘Modeling of casting. 41–49. M etall. g. r. 207–227. q. w. R ev. rusten: Metall. and d. poirier: Metall. Heat M ass T ransf . felicelli. d. 28B. loper. 1. janssen. 2189–2194. fredriksson: Scand. f.. V. rappaz: M etall.. 3–4. T rans. 49. gu and c. 2000. 1993. T rans. Growth. J. a. 7. aboutalebi. s. 6. 1997. and a. 68. heinrich. neilson and f. 17.. 40. M ater. prakash: Int. mo and h. and h. 1967. Mater. a. c. d. c. v. 55. beckermann: M etall. Sci. 29. Thomas and C.). T rans. and u. raihle and h. B. beckermann: Metall. t. m. B. 75. c. v. 681–688. 1995. 58. e. j. Vol. (ed. d. s. 1986. Cryst. TMS. mortensen and v. 24. 1992. 51. 1983. 60. B. 56. jarry. flemings and g. 2055–2065. 36. zhao. 1991. voller. lee. 79. cockcroft.. Forum. nereo: T rans. t. g. mortensen and i. a. 731–744. p. Fluid. chu. bennon: in ‘Micro/macro scale phenomena in solidi cation’. c. 1479–1491. poirier: Metall. 1970. M ech. 47. j. 63. r. s. brent. r. roberts: Q. Nagoya. 1990. 12. d. Heat M ass T ransf. Heat M ass T ransf. p. Fluid M ech. M ater. j. 1209–1220. New York. boettinger: Metall. m. c. N umer. s. m. incropera: Int. e. 10. M. AIME. joly. B. a. p. Mater. m. 49. gosh: ISIJ Int. g. 1553–1563. Mater. poirier. h. 171–189. schneider and c. C. m. m. d. and r. j. 889–900. mo: Int. m. 26A. g. Eisenhuttenwes. 677–686. A . felicelli. and c. vreeman. r. 21A. j. 23. 35. a. c. 1165–1183. 1995. m. and m. 1993. u. 25B. B. i. 43. 2373–2388. l. h. 1979. d. M ater.. T rans. 29. welding and advanced solidi cation processes VIII’. and b. Warrendale. Growth. 1990. 39–50. c. dawson: in HTD-Vol. 9. voller: N umer. B. j. d. Voller et al. Mater. A. beckermann and r. 13. a. 329. e. voller. T rans. prakash and v. 40. west. 15. 1968. 19. A .. poirier: J. c. felicelli. 106–120. r. m. kou. N umer. s. l. Beckermann et al. 17–26.. c. 1992. TMS. thomas: Int. c. z. J. c. 16. a. j. 31. 30. 2000. t. a. New York. 801–811. poirier. Sci. piwonka and m. d. Heat Mass T ransf. d. Rohatgi). flemings: Metall. mortensen. J. p. s. incropera: M etall.). d. B. v. T rans. (ed. m. r. v. B. 2263–2280. ASME. Taylor & Francis. 47 No. d. poirier. c. giamei. jin: Int. T rans. 177. 9B. 24. combeau. oldenburg and f. j. 242. R ev. and r. 28A. nereo: T rans. J. guthrie: Metall. 1983. mitchell: M etall. (ed. flemings: T rans. g. chung. r. mehrabian. Heat T ransf. t. t. 67. schwerdtfeger: A rch. choudhary and a. 50. 43. m. w. 4335 –4340. 1988. lesoult and s. 36. 341–375. krane. lesoult: Mater. 22B. viskanta: PhysicoChem. a. and r. 1081–1094. bart. Heat T ransf . T rans. Eng. 31A. r. 3238–3241. 26B. 27. j. c. nandapurkar. Mater. 1994. 155–162. B. McGraw-Hill. 32A. r. and j. incropera: N umer. apelian. 31A. M ater. PA. finn. Abbaschian et al. hills. PA. reddy and c. 135–142. 76. s. i. 10B. 28. Tokyo. m. r. d. 847–855. c. shimazu: Proc. J. gu. r. prantil and p. T rans. a. PA. Warrendale.. 1998. 74. 30B. (ed. 25. flemings: M etall. Q. 5 . PA. c. 1991. J. A . c. Rev. 1989. ridder. sella: Solid State Phenom. 81. T rans. chu and j. 38. w. mo: M etall. T rans. 34. drezet. G.. 1966. vreeman and f. 648–655.. trivedi: in ‘Solidi cation of metal matrix composites’. c. h. j. j. 1990. 53. r. 1993. Warrendale. 52. 665–669. 1995. ¨ 1981. 36. C. flemings. J.. T rans. 1970. 242. a. Mater. auburtin. Heat M ass T ransf . 111–119. poirier. Washington. 89–102. beckermann: M etall. 29B. v. M ath. vafai: J. spera: N umer. s. lesoult: M ater. r. and s. T rans. 1. d. miyazawa and k. and t. Bickert). R ev. c. giamei: M etall. 33. Japan. Eng. mehrabian: M etall. g. DC. 22B. 1990. m. J. 611–616. (ed. 2281–2289. kattner: M etall. 39. Metall. and f.. a. 2000. Sci. schneider. 1990. thevik: Int. beckermann: in ‘Materials processing in the computer age II’.-m. 2000. 1–13. B. beckermann and r. 26. 27A. 8. s. nereo: T rans. New York. J. 47–54. Iron and Steel Congress. 21–25. jacoby: in ‘Light metals 1990’. k. 72. c. 1981. 1968.. and m. mo: Int. c. li. r. 479–489. 195–213.). mo. 15. 1157– 1165.. r. 1718–1731. d. r. 2059–2072. B. pollock and w. 34. TMS. m. m. T rans. 6th Int. d. 317–321. M ethods Fluids. Mater. 2545–2557. g. s. 2000. 70. 52. i.. swaminathan. 42. Cryst. 455–464. 435–447. A IME. 1345–1356. J. and d. k. michaud and a. Flemings Symp. 104–112. 505–539. 261–291. 1995. 30A. flemings. Mater. boettinger. 29B. R. mehrabian. Hydrodyn. r. 1–20. c.. p. and h. 1991. 59. Warrendale. s. 43. 1987. fujii. m. T rans. p. Warrendale. Minkowycz and E. 2001. and y. 65. 1974.. v.260 Beckermann Modelling of macrosegregation 44. yang. 235–244. p.. beckermann. p. 1357 –1366.. and g. Heat M ass T ransf . Heat M ass T ransf . 191. 14. grun. fredriksson: M etall. flemings: ‘Solidi cation processing’.. j. sekhar and r. 64. 40. 32. 1997. poirier: Int. Aachen. J. m. 2000. 1994. c. and j. michaud: Metall. mehrabian: M etall. kajitani. 27.. 1974. 50–55. m. 21. d. 101–128. combeau. and j. h. 73. 30. 2000. 925–930. B. 1998. M at er. 69. flemings and g. 29. Heat T ransf. felicelli.. 687–704. d.. 29A. TMS. B. 2001. 1988. incropera: Int. a. g. 1997. wang. 37. c. 28B. M etall. 23A. m. c. 1998. j. 415–422. 15. giuliani and k. mehrabian. ganesan: Metall. Methods Eng. 80. TMS. Metall. 18B. s. R. olsson. 15. d. p. 77. 1997. A . r. 61. j. a. 121. p. A . 1992. heinrich. 54. Sahm et al. 1998. gan: M etall. l. T rans. International Materials Reviews 2002 Vol.. h. p. vannier. B. and d. 1996.

T rans. u. Warrendale. 1999. Aachen. 27A. 154. chai. and g. IV. k. j. 135. a. karma: Phys.. 99. 107. (ed. l. th. Cryst. 199–211. 1997. h. and s. 187. m. gandin. B.. 87. m’hamdi. T rans. j.. incropera: Int. 349–361. M ater. 1995. 91. smith: in ‘Modeling of casting. (ed. A . h. Comput . T rans. 2000. and c. (ed. 44. TMS. 1990. 143. 745–750. mo. 3827–3835. T rans. and m. rappaz: in ‘Modeling of casting. rappaz and ph. m. karma. combeau and a. han. a. 148. A . viswanathan: M etall. and i.. B. beckermann: Int. c. wang. r. flood and j. PA.. 2765–2783. A . m. B. n. A151. p. y. o. and c. 1999. 153. s. 1986. A . m. a. goyeau. 1994. poirier. hunt: A ppl. welding and advanced solidi cation processes IX’. m. Mater. 1993. m. d. mo: M etall. M at er. d. 1996. c. 25B. welding and advanced solidi cation processes VIII’. Heat Mass T ransf. c. Heat T ransf. 150. h. 245–255. 1285–1296. 41–61. 2001. beckermann. 1999. k. Porous M edia. 129. 134. b. appolaire. j. 42. beckermann: M etall. A. j. beckermann. 142. R. wheeler. 3129–3135. el-bealy and h. gerardin. arnberg. cornie. 329. 96. incropera: Int. 98. 38. 1996. Mater. 5 . J. rappaz. 93–123. d. 1993. h. p. y. heinrich: M etall. x. 26B. mo. j. and d. T rans. 92. A . Eng. rappaz: Metall. 29A. T rans. sung. de groh iii. h. beckermann: M etall. and x. Heat Mass T ransf . heinrich. p. 89. 468–496. M ater. r. Warrendale. A173. 35. 27A. tong. c. r. Beckermann). 120. wang: M etall. heinrich: M etall. y.. 613–622. and d. 108. R49–R52. c. Beckermann). ocansey and d. r. s. 1996. T rans. goldenfeld. 2000. 410.. Shaker Verlag. h. b. l. wang and c. and d. goyeau. bhat. j. d. 146. quintard: M etall. A . nagelhout: Scr. wang and c. kobayashi: Physica D. B. M ater. 1998. A .. beckermann: Int. kurz: Z . t. rappaz. A . c. and metals’. T rans. p. 113–123. p. felicelli: Int. M etallkd. t. kaempfer and m. moon. 34. p. T rans. rappaz. and s. wang and c. 27A. 339–344. d. lesoult. 749–753. 1987. B.. desbiolles. j. schneider. T rans. 141. g. 155. c. 40. 639–644. t. r. 1995. 122. 113. 27–42. ahuja. diepers. Eng. lipinski. 25A. l.. 265–273. R ev. prescott. Shaker Verlag. PA. J. 1996. k. T rans. Mater. 47. 47 No.-w. Sahm et al..). A144. 1989. r. m. kim. J. welding and advanced solidi cation processes VI’. c. bhat. 115. voller: Int. R ev. Sci. y. x. poirier: N umer. Commun. Phys. sung. Cross). b. R es. poirier. 112. n. felicelli: M etall. v.-m. v. a. thevik: Metall. 109. J. c. 1992. poirier. 136. Sci. 263–270. 1487–1497. 397–407. 189–198. bhat. 2001.. and m. 2000. d. Heat T ransf. stomp: Metall. q. combeau. 1995. y. Warrendale. A . j.. A157. and l. T rans. r. 7. arnberg: M etall.. h. 45. 2000. Eng. 2305–2313. a. 1994. o. 2784–2795. bennon: Int. 77. 28A. zakhem. J. 1049 –1056. 597–605. 041602. r. 21A. 154. beckermann: Metall. 46. J. c. and d. i. 1573 –1581. 1992. tong. T rans. beckermann. 123. poirier: Mater. jalanti. 39.. x. 1994. TMS. p. fredriksson: M etall. f. c. 363–374. A . incropera. s. m. 3663–3678. 18B. B. and s. G.. h. ganesan. 34. combeau. Comput . M ethods Fluids. 86. lesoult: Acta M ater. Thomas and C. beckermann and m. 100. a. beckermann and c. 152. p. beckermann: M etall. Heat M ass T ransf . p. M ater.. quintard and s. wang and c. d. and m. 31A. diepers. combeau. boettinger. P. 64. 139. 94. 21. c. c. and w. 1081– 1093. Sci. c. 2001. Aachen. 24B. 102. 149. s. 57. T rans. Heat Mass T ransf . d. a. 2000. v. Mater. r. 117. 140. l. y. voller and f. 357–370. M ater. f. Warrendale. 1998. 27B. 26B. tong and c. 209–222. combeau. Methods Heat Fluid Flow. 88. Beckermann). o. 2000. 1996. 119. 1998. c. 111. j. d. ni and f. 2754–2764. m.-a. 138. c. R ev. 2001 TMS Fall Extraction and Process Metallurgy Meeting: ‘Computational modeling of materials. i. c. 2705– 2714. prescott and f. A171. karma and w. flemings: M ater. and s. and m. 93. c. (ed. R. M etall. S. desbiolles. gobin. 156. 145. jeong. B. 40. and d. 289–302. beckermann: M ater. r. B. N umer. 31.. c. gobin: in ‘Modeling of casting. J. 1993. 253–265. c.Beckermann Modelling of macrosegregation 261 84. 63. h. dantzig: Phys. d. 1999. 151. E. G. rappaz: Int. Mater. B. 617–630. 1999. poirier: Metall. m. and s. 1998. 121. 1992. c. 1995. d. 101–103. 35. 1997. 179. 104. Heat Mass T ransf. incropera. whitaker: T ransp. m. incropera: J. j. quintard. and g. 38. p. duncan. 633–638. T rans. m. m. Heat Mass T ransf. j. appolaire. chan. j. mccarthy: A cta M etall. ganesan: Mater. r. j. 1993. 133. 11. frueh. welding and advanced solidi cation processes VIII’. j. krane and f. Eng. and g. 1993. Sci. c. Piwonka et al. T rans. wang and c. T rans. b. poirier: M ater. p. benihaddadene. PA.. a. r. A . poirier. c. Eng. Mater. schaefer: in ‘Modeling of casting. h. 97. and j. Sci. 26B. T rans. T rans. nielsen. rappaz: in ‘Modeling of casting. j. and r. (ed. 2455–2462. r.. voller: Int. welding and advanced solidi cation processes IX’. 1996. 95. Heat M ass T ransf. j. bousquet-melou. felicelli: Metall. 3455–3473.. m. li: Phys. sundarraj and v. 132. R ev. nielsen and l. 103.. 116.. r. h. m’hamdi. M at er. Mater. 1995. b. 63. 1998. TMS. TMS (in press). 061601. 1995. 131.. 181–199. minerals. krane. combeau. lesoult: ISIJ Int. paradies.. m. l. (ed. N umer. 5851–5862. 616–623. Warrendale. 1997. Sci. jalanti. A171. ni and f. J. f. 1993. 38. J. 30A. m. M ultiphase Flow. c. a. M ethods Heat Fluid Flow. Thomas and C. and w. 124. 90. 114. m. rappaz and v. 31A. 15. 1994. 1997. backerud: M at er. 7. G. 111–119. sundarraj and v. 127. T rans. de groh iii: M etall. 97–105. 85. s. B. t. A. p. porte-agel: J. mcfadden: Phys. 749–753. M at. Mater. 126. beckermann: M etall. and h. d. International Materials Reviews 2002 Vol. 1994. krane and f. 157. arnberg. Sci. 9. 128. Growth. m. c. incropera: Int. combeau. B. voller: Metall. lesoult: J. TMS. 1997. 106. 1993. w. m. 1993. m. 1991. tong: J. 1991. d. 137. s. Mater. r. Forum. 1300– 1307. PA. gaskell: M etall. A . welding and advanced solidi cation processes VIII’. Mater. y. 144. 1992. 3837–3847. 698–703. karma. R ev. m. A . E. 2787– 2802. c. 2001. 35. steinbach. d. J. gipouloux. schneider: ISIJ Int. s. a. g. drezet. r. lesoult: Int. Phys. 36. huang. Porous M edia. beckermann: Scr. 2001. thevoz: Acta M etall. 843–853. dustin and w. beckermann. whitaker: T ransp. R ev. 389–396. 2001. gandin: Mater. T. and a. Mater. c. ch. and h. barrere. n. 105. 1271–1284. 1998. s. and j. N umer. beckermann: Metall. 716–724. p. d. (ed. 375–382. T rans. Eng. 2351–2359. 29A. 1987. h. 24A. T rans. 101. M ater. g. 231–240. r. j. 999–1014. j. h. e. 32A. 1995. c. s. 29. r. 7424. incropera: Int. fichot. ni and c. P. j. wang and c. 110. T rans. 1998. beckermann. Sahm et al. j. Phys. 125.. PA. albert. 640–647. 296–317. 1987. 1998. beckermann: J. mo: M etall. 143–150. s. ahuja. 19. m. d. combeau. B. schneider and c. Sci. and g. E. 27A. A .. T rans. and g.). M. rappel: Phys. 1993. 1999. a. 30B. y. weidman. 4323. c.). and j. bobadilla. 1995. 61. 202–207. d. poirier and s. a. 3149 –3153. ramani and c. p. 31–49. beckermann. 1999. f. 1991. heinrich. 30A. and g. M ater. 130. 117. a. 2002. j. j. steinbach: Acta Mater. 1091 –1092. poirier. lesoult: in ‘Modeling of casting. 257–264. and q. E. m. glicksman. j. poirier. 309–316. h. Thomas and C. m. A . v. albert. j.. 22B. 147. and c. voller: Proc. ahmad. 118.