You are on page 1of 180

COMPUTER METHODS IN MATERIALS SCIENCE

Informatyka w Technologii Materia ł ów
Publishing House AKAPIT

Vol. 13, 2013, No. 2

Contents Spis treści

Anthony R. Thornton, Thomas Weinhart, Vitaliy Ogarko, Stefan Luding MULTI-SCALE METHODS FOR MULTI-COMPONENT GRANULAR MATERIALS .............................. 197 Marta Serafin, Witold Cecot NUMERICAL ASPECTS OF COMPUTATIONAL HOMOGENIZATION .................................................... 213 Liang Xia, Balaji Raghavan, Piotr Breitkopf, Weihong Zhang A POD/PGD REDUCTION APPROACH FOR AN EFFICIENT PARAMETERIZATION OF DATA-DRIVEN MATERIAL MICROSTRUCTURE MODELS ..................................................................... 219 Marek Klimczak, Witold Cecot LOCAL NUMERICAL HOMOGENIZATION IN MODELING OF HETEROGENEOUS VISCO-ELASTIC MATERIALS ....................................................................................................................... 226 Balbina Wcisło, Jerzy Pamin NUMERICAL SIMULATIONS OF STRAIN LOCALIZATION FOR LARGE STRAIN DAMAGEPLASTICITY MODEL ...................................................................................................................................... 231 Andrzej Milenin, Piotr Kustra, Dorota J. Byrska-Wójcik THE MULTI-SCALE NUMERICAL AND EXPERIMENTAL ANALYSIS OF COLD WIRE DRAWING FOR HARDLY DEFORMABLE BIOCOMPATIBLE MAGNESIUM ALLOY .............................................. 238 Piotr Gurgul, Marcin Sieniek, Maciej Paszyński, Łukasz Madej THREE-DIMENSIONAL ADAPTIVE ALGORITHM FOR CONTINUOUS APPROXIMATIONS OF MATERIAL DATA USING SPACE PROJECTION ........................................................................................ 245 Wacław Kuś, Radosław Górski PARALLEL IDENTIFICATION OF VOIDS IN A MICROSTRUCTURE USING THE BOUNDARY ELEMENT METHOD AND THE BIOINSPIRED ALGORITHM ................................................................... 251 Krzysztof Muszka, Łukasz Madej APPLICATION OF THE THREE DIMENSIONAL DIGITAL MATERIAL REPRESENTATION APPROACH TO MODEL MICROSTRUCTURE INHOMOGENEITY DURING PROCESSES INVOLVING STRAIN PATH CHANGES ....................................................................................................... 258 Ewa Majchrzak, Bohdan Mochnacki IDENTIFICATION OF INTERFACE POSITION IN TWO-LAYERED DOMAIN USING GRADIENT METHOD COUPLED WITH THE BEM .......................................................................................................... 264

195 – 196

ISSN 1641-8581

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

Agnieszka Cebo-Rudnicka, Zbigniew Malinowski, Beata Hadała, Tadeusz Telejko INFLUENCE OF THE SAMPLE GEOMETRY ON THE INVERSE DETERMINATION OF THE HEAT TRANSFER COEFFICIENT DISTRIBUTION ON THE AXIALLY SYMMETRICAL SAMPLE COOLED BY THE WATER SPRAY ............................................................................................... 269 Michael Petrov, Pavel Petrov, Juergen Bast, Anatoly Sheypak INVESTIGATION OF THE HEAT TRANSPORT DURING THE HOLLOW SPHERES PRODUCTION FROM THE TIN MELT ..................................................................................................................................... 276 Sławomir Świłło AN EXPERIMENTAL STUDY OF MATERIAL FLOW AND SURFACE QUALITY USING IMAGE PROCESSING IN THE HYDRAULIC BULGE TEST ..................................................................................... 283 Szymon Lechwar SELECTION OF SIGNIFICANT VISUAL FEATURES FOR CLASSIFICATION OF SCALES USING BOOSTING TREES MODEL ............................................................................................................................ 289 Stanisława Kluska-Nawarecka, Zenon Pirowski, Zora Jančíková, Milan Vrožina, Jiří David, Krzysztof Regulski, Dorota Wilk-Kołodziejczyk A USER-INSPIRED KNOWLEDGE SYSTEM FOR THE NEEDS OF METAL PROCESSING INDUSTRY ........................................................................................................................................................ 295 Stanisława Kluska-Nawarecka, Edward Nawarecki, Grzegorz Dobrowolski, Arkadiusz Haratym, Krzysztof Regulski THE PLATFORM FOR SEMANTIC INTEGRATION AND SHARING TECHNOLOGICAL KNOWLEDGE ON METAL PROCESSING AND CASTING ........................................................................ 304 Jan Kusiak, Gabriel Rojek, Łukasz Sztangret, Piotr Jarosz INDUSTRIAL PROCESS CONTROL WITH CASE-BASED REASONING APPROACH ........................... 313 Krzysztof Regulski, Danuta Szeliga, Jacek Rońda, Andrzej Kuźniar, Rafał Puc RULE-BASED SIMPLIFIED PROCEDURE FOR MODELING OF STRESS RELAXATION ..................... 320 Sławomir Świłło EXPERIMENTAL APPARATUS FOR SHEET METAL HEMMING ANALYSIS ....................................... 326 Sławomir Świłło, Piotr Czyżewski AN EXPERIMENTAL AND NUMERICAL STUDY OF MATERIAL DEFORMATION OF A BLANKING PROCESS ................................................................................................................................. 333 Piotr Lacki, Janina Adamus, Wojciech Więckowski, Julita Winowiecka MODELLING OF STAMPING PROCESS OF TITANIUM TAILOR-WELDED BLANKS ......................... 339

COMPUTER METHODS IN MATERIALS SCIENCE

Andrzej Woźniakowski, Józef Deniszczyk, Omar Adjaoud, Benjamin P. Burton FIRST PRINCIPLES PHASE DIAGRAM CALCULATIONS FOR THE CdSe-CdS WURTZITE, ZINCBLENDE AND ROCK SALT STRUCTURES ........................................................................................ 345 Andrzej Woźniakowski, Józef Deniszczyk PHASE DIAGRAM CALCULATIONS FOR THE ZnSe – BESE SYSTEM BY FIRST-PRINCIPLES BASED THERMODYNAMIC MONTE CARLO INTEGRATION ................................................................. 351 Michal Gzyl, Andrzej Rosochowski, Andrzej Milenin, Lech Olejnik MODELLING MICROSTRUCTURE EVOLUTION DURING EQUAL CHANNEL ANGULAR PRESSING OF MAGNESIUM ALLOYS USING CELLULAR AUTOMATA FINITE ELEMENT METHOD ................. 357 Bartek Wierzba THE MIGRATION OF KIRKENDALL PLANE DURING DIFFUSION ........................................................ 364 Onur Güvenc, Thomas Henke, Gottfried Laschet, Bernd Böttger, Markus Apel, Markus Bambach, Gerhard Hirt MODELING OF STATIC RECRYSTALLIZATION KINETICS BY COUPLING CRYSTAL PLASTICITY FEM AND MULTIPHASE FIELD CALCULATIONS ............................................................. 368

– 196 –

COMPUTER METHODS IN MATERIALS SCIENCE
Informatyka w Technologii Materia ł ów
Publishing House AKAPIT

Vol. 13, 2013, No. 2

MULTI-SCALE METHODS FOR MULTI-COMPONENT GRANULAR MATERIALS
ANTHONY R. THORNTON1,2, THOMAS WEINHART1, VITALIY OGARKO1, STEFAN LUDING1 Multi-Scale Mechanics, Department of Mechanical Engineering, University of Twente, 7500 AE Enschede, The Netherlands 2 Mathematics of Computational Science, Department of Applied Mathematics, University of Twente, 7500 AE Enschede, The Netherlands *Corresponding author: a.r.thornton@utwente.nl
Abstract In this paper we review recent progress made to understand granular chutes flow using multi-scale modeling techniques. We introduce the discrete particle method (DPM) and explain how to construct continuum fields from discrete data in a way that is consistent with the macroscopic concept of mass and momentum conservation. We present a novel advanced contact detection method that is able of dealing with multiple distinct granular components with sizes ranging over orders of magnitude. We discuss how such advanced DPM simulations can be used to obtain closure relations for continuum frameworks (the mapping between the micro-scale and macro-scale variables and functions): the micro-macro transition. This enables the development of continuum models that contain information about the micro-structure of the granular materials without the need for a priori assumptions. The micro-macro transition will be illustrated with two granular chute/avalanche flow problems. The first is a shallow granular chute flow where the main unknown in the continuum models is the macro-friction coefficient at the base. We investigate how this depends on both the properties of the flow particles and the surface over which the flow is taking place. The second problem is that of gravity-driven segregation in poly-dispersed granular chute flows. In both these problems we consider small steady-state periodic box DPM simulations to obtain the closure relations. Finally, we discuss the issue of the validity of such closure-relations for complex dynamic problems, that are a long way from the simple period box situation from which they were obtained. For simple situations the pre-computed closure relations will hold. In more complicated situations new strategies are required were macro-continuum and discrete micromodels are coupled with dynamic, two-way feedback between them. Key words: coupled multiscale model, multi-component granular materials, Navier-Stokes equation, discrete particle simulations
1

1. INTRODUCTION Granular materials are everywhere in nature and many industrial processes use materials in granular form, as they are easy to produce, process, transport and store. Many natural flows are comprised of granular materials and common examples include rock slides than can contain many cubic kilometers of material. Granular materials are, after water, the second most widely manipulated substance on the

planet (de Gennes, 2008); however, the field is considerable behind the field of fluids and currently no unified continuum description exists, i.e. there is no granular Navier-Stokes style constitutive equations. However, simplified descriptions do exist for certain limiting scenarios: examples include rapid granular flows where kinetic theory is valid (e.g., Jenkins & Savage, 1983; Lun et al., 1984) and shallow dense flows where shallow-layer models are applicable

197 – 212

ISSN 1641-8581

g. is an active area of research and many steps forward have recently been made. and. For problems COMPUTER METHODS IN MATERIALS SCIENCE Fig. steady and accelerating material. 2012. 1989. 1. surface contact properties. Weinhart et al. discrete particle simulations are used to determine unknown functions or parameters in the continuum model as a function of microscopic particle parameters and other state variables. Solid lines indicate the main steps of the method and dashed lines the verification steps.. 2012). Bokhove & Thornton. it is not possible in this fashion to create a unified model capable of describing a new scenario. Once these averaged parameters have been tuned via experimental or historical data.INFORMATYKA W TECHNOLOGII MATERIAŁÓW (e. more research on a continuum description is required. variations in density. 1012). here. In simple situations. flow obstacles and constrictions. arbitrary shapes. coexistence of static. continuum and micro-scale models are dynamically coupled with two-way feedback between the computational models. Thornton et al. complex (nonspherical) shape. thus reducing computational expense and allowing simulation of complex granular flows. it is possible to pre-compute the relations between the particle and continuum (micro-macro transition method). but. Savage & Hutter. (a) shows the idea for the micro-macro transition and (b) for two way coupled multi-scale modelling (CMSM). however. these models can be surprisingly accurate. Illustration of the modeling philosophy for the undertaken research. complex basal topography.g. For CMSM. For the case of quasi-static materials the situations is even more complicated and. With the recent increase in computational power it is now possible to simulate flows containing a few million particles.g. in arbitrary geometries. Therefore. nonuniform shape. Gray. breaking and cohesional particles. but. in more complicated situations twoway coupled multi-scale modelling (CMSM) is required. The coupling is done in selective regions in space and time. Discrete particle methods (DPMs) are a very powerful computational tool that allows the simulation of individual particles with complex interactions. Continuum methods are able to simulate the volume of real industrial or geophysical flows. however. How to capture these elaborate interactions of sintering. but have to make averaging approximations reducing the properties of a huge number of particles to a handful of averaged quantities. in size and density. DPM is very powerful and flexibility tool. 2013). For the micro-macro transition. it is computationally very expensive. – 198 – . for 1 mm particles this would represent a flow of approximately 1 litre which is many orders of magnitude smaller than the flows found in industrial and natural flows. DPM can be used to obtain the mapping between the microscopic and macroscopic parameters allowing determination of the macroscopic data without the need for a priori knowledge. these mappings are referred to as closure relations (e. Flows in both nature and industry show highly complex behaviour as they are influenced by many factors such as: poly-dispersity.. by solving Newton's laws of motion for each particle (e. 2003. by the contact model. a model tuned for one flow configuration often has no prediction power for another setup.

2007). The relative distance between two particles i and j is rij = ri-rj. see e.vj . f ijn . and the overlap is:    ijn  max 0. the tangential force yields – 199 – COMPUTER METHODS IN MATERIALS SCIENCE Two particles are in contact if their overlap is positive. In this paper we discuss the approach we are taking. 2. and tangential.  1  d i + d j   rij    2  n ˆij  n ˆij vij   vij  n t ˆij  n ˆij  ωi  bij  ω j  b ji vij  vij   vij  n The total force on particle i is a combination of the normal and tangential contact forces fijn  fijt from each particle j that is in contact with particle i. and external forces. Firstly. Different contact models exist for the normal. mass mi. we will not take this approach. Our approach is illustrated in figure 1. Each particle i has diameter di. The problem is fully determined by specifying the forces and torques between two interacting particles. The normal and tangential relative velocities at the contact point are given by: . Newton's laws are solved for both the translational and the rotational degrees of freedom. and iterative approach can be used. when the tangential-tonormal force ratio becomes larger than a contact friction coefficient. For each contact model. where a continuum solver is used everywhere. the branch vector (the vector from the centre of particle i to the contact n ˆ / 2  2 . often called the discrete element method. c. position ri. where the unit point) is bij   d i  dij n   ˆ ij  ri  r j / rij . We will focus on scenarios where simplifying approximations are made which lead to continuum models (still containing undetermined quantities) that are valid only in certain limits. the efficiency of DPM becomes an issue and a new algorithm will have to be considered. and §7 future prospects and conclusions. §3 how to construct continuum fields from discrete particle data (how to perform the micro-macro transition). however.g. For large complex regions or globally complex problems. It is then expected that this closed continuum model is able to simulate the flow of the same particles in more complex and larger systems. one can use a localised hybrid approach where a particle method is applied in the complex region and is coupled through the boundary conditions to a continuum solver (Markesteijn. the contacts can be treated as occurring at a single point. For the second problem. gravity-driven segregation of poly-dispersed granular material. (Weinan. INTRODUCTION TO DPM In the discrete particle method. Here. §5 collision detection. We then perform small scale periodic box particles simulations and use this data to determine unknowns in the continuum models (i. mig. The outline for the rest of the paper will be: §2 introduction to DPM.1. we will briefly review three commonly used contact laws for almost spherical granular particles: linear spring-dashpot. Outline It is possible to apply CMSM or micro-macro methods to completely general three-dimensional Cauchy mass and momentum equations and use the DPM to determine the unspecified constitutive relations for the stresses. §5 DPM simulations with wide particle distribution. The validity of this closed model will be investigated by comparing its results with both computationally expensive large-scale simulations and experiments. The ultimate aim of this research would be to determine the unknowns (material/contact properties) in the contact law from a few standard experiments on individual particles. we will consider shallow granular flows (of major importance to many areas of geophysics) and secondly. Since we are assuming that particles are almost spherical and only slightly soft. 1. §6 micro-macro transition for segregating flows. we will use the computationally more expensive two-way coupled models to simulate the flow.INFORMATYKA W TECHNOLOGII MATERIAŁÓW that contain only small complex regions. and small particle simulations are run each time the closure relations need to be evaluated. and the relative velocity normal is n is vij = vi . f ijt . §4 the micromacro transition for shallow granular flows. forces. velocity vi and angular velocity i. The idea is to obtain the particle material properties from small (individual) particle experiments and use this information to determine the parameters of the contact model for DPM simulations. Hertzian springs and plastic models. which for this investigation will be limited to gravity.e to close the model). review the current steps we have made and discuss the future directions and open issues with this approach. For situations were this one-way coupled micro-macro approach fails. 2011).

we give a short overview. We integrate the resulting force and torque relations in time using Velocity-Verlet and forward Euler (Allen & Tildesley.e. 1995). For a more detailed review of contact laws. where tc is the collision time. Goldhirsch. δij . The elastic tangential displacement. (2012b) as this is still valid within one coursegraining width of the boundary. and tangential. Walton & Braun. or coarse-graining spatially and temporally (Babic. Shen & Atluri. 1950. 2010). (iii) the results are even valid for single particles as no averaging over groups of particles is required. kt and damping coefficients t n.INFORMATYKA W TECHNOLOGII MATERIAŁÓW and the particles slide. The coarse-graining method has the following advantages over other methods: (i) the fields produced automatically satisfy the equations of continuum mechanics. This model is designed to model particles that are elastic. There are many papers in the literature on how to go from the discrete to the continuum: binning micro-scale fields into small volumes (Irving & Kirkwood. Schoeld & Henderson. 2004.. one of the biggest challenges is how to obtain continuum fields from large amounts of discrete particle data. (Silbert et al. 2001). for enk1n / k2 f ijn( sd ) . 1986). (ii) it is neither assumed that the particles are rigid nor spherical. 1982. Luding. since oscillations on the unn ) branch do not dissipate enerloading/reloading ( k2 (2) with spring constants kn. Weinhart. in the normal direction a different spring constant is taken for loading and – 200 – .g. For the spring-dashpot case (Cundall & Strack.. 1989) with a time step t = tc = 50. Unlike n (Luding. the first term is the relative tangential velocity at the contact point. Walton & Braun. 2001). Finally. 2012a) for details. (Luding. However. hence: n ˆij   n vij f ijn( sd )  k n ijn n . we use the coarse graining approach described by Weinhart et al. for the plastic case (designed to capture small plastic deformation) we modify the normal force using the (hysteretic) elastic-plastic form of Walton and Braun e. we refer the reader to (Luding. 1986). the contact area can be defined to be zero at the initial time of contact.g. 2008): fijn( p ) ˆ ij  k1n ijn n  n e ˆ ij   k2  ij n  0  if if if n e k2  ij  k1n ijn n e k1n ijn  k 2  ij  0 n e k2  ij  0 (5a) t t f ijt ( p )  f ijt ( sd )   k t δij   t vij tc   kn   n  mij   2mij     2 (1) (5b) e n max n n max n with  ij   ij   ij 1  k1 / k2 and.: fijt   c fijn . For the Hertzian case we modify the interaction force with: f n/t ij ( Hertz )   ijn d t fijn(/sd ) (4) see e. Luding et al.. Here. the maximum overlap during the contact.. (Luding.  ij is  max  ij   with the reduced mass mij = mimj/(mi + mj). Here. forces are modelled with linear elastic and linear dissipative contributions. in general. even near the flow base. 2008).e. and the t second term ensures that δij remains normal to nij. t t fijt( sd )   k t ij   t vij during contacts the dissipation is smaller than in the spring-dashpot case. is gy. 2012a) the normal. f ijt( sd ) . 1997. therefore. see e.g. t. This models follows from the theory of elasticity and takes account of the full non-linear elastic response. so that the normal coefficient of restitution is given by n = n . we take k 2 to be constant. 2008. averaging along planes (Todd et al. and. 3. . 2008. then review in more detail the approach we prefer. and its rate of change is given by: d t t t δij  vij  rij1  δij  vij  nij dt (3) COMPUTER METHODS IN MATERIALS SCIENCE In equation (3). see (Weinhart. but dissipated with clearly defined coefficient of restitution. THE MICRO-MACRO TRANSITION For all multi-scale methods. 1979. The only assumptions are that each particle pair has a single point of contact (i. the particle shapes are convex). 2008. Luding. and we truncate the magnitude of the tangential force as necessary to satisfy unloading/reloading of the contact and no dash-pot is used i. 2004.

Bold vector notation will be used when convenient. It is clear that as w → 0 the macroscopic density defined in (8) reduces to the one in (7).:  p   V V      t   g   r r r (13) . The following definition of the macroscopic density of the flow is used: It is straightforward to confirm that equations (7) and (10) satisfy exactly the continuity equation:   r . mic. at a point r at time t is defined by: Next. or surface traction density.:  mic  r  ri  t     r .2. as described in detail in (Weinhart et al.INFORMATYKA W TECHNOLOGII MATERIAŁÓW replaced by a contact point (i. j. 3.t) is then defined as the ratio of momentum and density fields. Mass balance Vectorial and tensorial components are denoted by Greek letters in order to distinguish them from the Latin particle indices i. In general. The coarse-grained momentum density vector p(r. According to Goldhirsch (2010). we will consider how to obtain the other fields of interest: the momentum density vector and the stress tensor. t dr ' mic i (9) where  is the stress tensor.e.t  = R3  W  r  r  t     r '.3. and the width w is the key parameter. t. 3. respectively. Since we want to describe boundary stresses as well as internal stresses. As before. Assume a system given by Nf owing particles and Nb fixed basal particles with N = Nf + Nb. the microscopic mass density of the flow. Expressions (10) and (11) for the momentum p and the velocity V have already been defined. i. Notation and basic ideas 3. the desired momentum balance equations are written as:  r. we will calculate macroscopic fields pertaining to the owing particles only. see (Weinhart et al. the particles are not too soft). the boundary interaction force density. t  = p  r . we arrive at the following expression for the stress: – 201 – COMPUTER METHODS IN MATERIALS SCIENCE with width or variance w. and reach closure. Momentum balance W  r  ri  t      r  r t  2  i  exp   3 2 2 w   2 w   1  (8) Finally. the coarse-grained fields depend only weakly on the choice of function.1. t  =  m v   t W  r  r  i 1 i i i Nf (10) where the vi ‘s are the velocity components of particle i. 2012b) for details. The coarsegraining function can also be seen as a convolution integral between the micro and macro definitions. and g is the gravitational acceleration vector. has been included. but the Gaussian has the advantage that it produces smooth fields and the required integrals can be analysed exactly.. t    r. Then after some algebra. t    mi  i 1 Nf (6) V  r . t    mi W   r  ri  t    i 1 Nf (7)  p  0 t r (12) thus replacing the Dirac delta function in (6) by an integrable ‘coarse-graining’ function w whose integral over space is unity. t  (11) where (r) is the Dirac delta function and mi is the mass of particle i. t) is defined by: p  r . We will take the coarsegraining function to be a Gaussian: with the Einstein summation convention for Greek letters. Since we are interested in the flow. The next step is to compute their temporal and spatial derivatives. The macroscopic velocity field V(r.. and that collisions are not instantaneous. we will consider the momentum conservation equation with the aim of establishing the macroscopic stress field. Other choices of the coarse-graining function are possible. 2012b).e. From statistical mechanics. all macroscopic variables will be defined in a way compatible with the continuum conservation laws.

(2012b). . the boundary interaction force density: t   fikW i 1 k 1 Nf Nb  r  cik  (15) is active. 4. 1989.. Moreover. . i 1 Nf i (14) Equation (14) differs from the results of Goldhirsch. as detailed by Weinhart et al. 1997). Originally these models were derived from the general continuum incompressible mass and momentum equations. Here. while the contribution from the interaction of particles i with a fixed particle k is distributed along the line from ri to the contact point cik = ri + bik. The height of the flow is h = s . y-axis across the slope and the z-axis normal to the slope. The contribution to the stress from the interaction of two flow particles i. Further details about the accuracy of the stress definition (14) are discussed by Weinhart et al. using the long-wave approximation (Savage & Hutter. 2007). 2003. respectively..b and velocity components are u = (u.v. (2012b). u the depth-averaged velocity and the source term is given by:   u S x  gh cos   tan     u2 v2   Before these equations can be solved.y) and z = b(x. they are now used as a geological risk assessment and hazard planning tool (Dalby et al.... wedges (Hakonardottir & Hogg. and are thus an effective tool in modelling geophysical flows. closure relations need to be specified for three unknowns: the velocity shape factor. The strength of this method is that the spatially coarse-grained fields by construction satisfy the mass and momentum balance equations exactly at any given time. 2007) and contractions (Vreman. Our philosophy was to determine these unknown relations using Shallow-layer granular continuum models are often used to simulate geophysical mass flows. including snow avalanches (Cui et al. the ratio of the two diagonal stress components.y). we refer the interested reader to (Babic. K. dense pyroclastic flows. SHALLOW GRANULAR FLOWS for shallow variations in the flow height and basal topography. 2005. 2008). we will only consider flows over rough at surfaces where b can be taken as constant.INFORMATYKA W TECHNOLOGII MATERIAŁÓW      fij rij  W  r  ri  sbik  ds i 1 j 11 b ik ik  0 Nf Nf 1   f b i 1 k 1 Nf Nf W  r  r  sb  ds   m v  v W i ik 0 i 1 . and lahars (Williamset al. block and ash flows (Dalby et al. or determined from experiments or DPM simulations. Such shallowlayer models involve approximations reducing the properties of a huge number of individual particles to a handful of averaged quantities. i i . 2008). Gray. Depthaveraging the mass and momentum balance equations and retaining only leading and first order terms (in the ratio of height to length of the flow) yields the depth-averaged shallow-granular equations. 2007). 2003). debris flows (Denlinger & Iverson. 2012) – 202 – . Background where g is the gravitational acceleration. measuring the force applied by the base to the flow. 2008). and the friction. shallow granular equations have been applied to analyse small-scale laboratory chute flows containing obstacles (Gray. the x-axis downslope. There. shallow-layer models tend to be surprisingly accurate. It has nonzero values only near the basal surface and can be introduced into continuum models as a boundary condition.Gray. Bokhove & Thornton. The expression for the energy is also not treated in this publication. 2001). showing good quantitative agreement between theory and experiment. irrespective of the choice of coarsegraining function. (e. In general. j is spatially distributed along the contact line from ri to rj. which in one-dimension are given by: h     hu    hv   0 t x y (16a) (16b)   g 2  2  hu      hu  K h cos    S x t x  2  COMPUTER METHODS IN MATERIALS SCIENCE 4.. In addition to these geological applications.1. Gray. the free-surface and base locations are given by z = s(x.w)T. 2003). Despite the massive reduction in degrees of freedom made. We will consider flow down a slope with inclination .g. between the granular materials and the basal surface over which it flows.(2010) by an additional term that accounts for the stress created by the presence of the base. These closure relations can either be postulated (from theory or phenomenologically).

2012a). F   tan 1   tan  2   tan 1  h 1 Ad  F    (20) Fig. The main conclusion were:  The law (18) holds for the spring-dashpot.  The geometric roughness  is more important than the contract friction in the interaction law. and  2 is the maximum angle at which steady uniform flow is possible.  The coefficient of restitution of the particles only affects.  1. inclination  = 24 and the diameter ratio of free and fixed particles. not the friction angles. The base was created from particles and roughness was changed by modifying the ratio of the size of base and flow particles. Silbert et al. In (Pouliquen & Forterre. below which the flow arrests. the particle size ratio  and even the type of contact law. at time t = 2000. Weinhart et al. 2012. where  1 is the minimum angle required for flow. 2004.. K. c. First the K was found to be linear in the inclination angle and independent of  (for all but the smooth base case of  = 0):  tan  2   tan   1     2 tan    tan 1  (18) where d is the particle diameter and A a characteristic dimensionless scale over which the friction varies. From these relations you can show that the friction closure is given by:   h.and y-directions. and above which the flow accelerates indefinitely. 2002). scaled linear with this curve: (17) F with d0 = 132 and d1 = 21. In the early models a constant friction coefficient was assumed (Hungr & Morgenstern. Pouliquen & Forterre. In this case flow particles are monodispersed.3. The chute is periodic and of size 20d10d in the xydirections. DPM simulation for Nf/200 = 17:5. Pouliquen.  = tan . Initially. The domain is periodic in x. steady uni- This experimentally determined law has previously been shown to hold from DPM simulations (e. 1999. that is. 2. is the ratio of two stress component K = xx/zz. (2001). 2001). In the z-direction.. gravity direction g as indicated. 1989). 2012a) we investigate how the parameters A. plastic and Hertzian contact models  The properties of the basal particles have very little affect on the macroscopic friction. i. fixed particles (black) form a rough base while the surface is unconstrained.. Colours indicate speed: increasing from blue via green to orange.  2. . h   1     2 hstop   (19) where  and  are two experimentally determined constants. Closure for  By far the most studies closure relation for shallow granular flows is the basal friction coefficient . only a weak effect on and.INFORMATYKA W TECHNOLOGII MATERIAŁÓW small-scale periodic box simulations DPMs similar to the ones used by Silbert et al. with inclination . For these models. and  change as a function of the contact friction between bed and owing particles. Detailed experimental investigations (GDR MiDi. In (Thornton. .2. Closure for K form flow is only possible at a single inclination. 4. we consider only the springdashpot contact model and looked at the closures across different basal surfaces (Weinhart et al.  1 <  < 2.g. where  is a fixed basal frictional angle. 2002) for the flow over rough uniform beds showed that steady flow emerges at a range of inclinations. see figure 2. was fitted to: hstop   Ad For the shallow layer theory presented in (16). They also observed that the Froude number F = K fit  1  o   d1 d0 o u / gh cos   . Savage & Hutter. Below some of the main findings are summarized. 1984. – 203 – COMPUTER METHODS IN MATERIALS SCIENCE . 4.30 .e.  = 1. the measured height hstop () of stationary material left behind when a owing layer has been brought to rest.

5. Bagnold profile  = 1:25 and lines profile  = ever. 5  hz  u  u 1    3    h  9. This comparison and verification set is represented by the dashed lines in figure 1. and fit (h. We observe a Bagnold velocity profile. Weinhart et al. plug flow  = 1.42.4. 4. van der Vegt). see figure 4 for details.  1. These were then fitted separately and from these fits the Fig. For comparison. in the bulk. howlines). This was done in two steps: firstly. This is the shape of the velocity profile with height and is defined as shape factor  was computed. The results of this closed model are compared with computationally expensive full-scale DPM simulations of the same scenarios. hpGEM (Pesch. b This closed continuum model has recently been implemented in an inhouse discontinuous Galerkin finite element package.  2. A series of test cases and currently being investigated include complicated features with contractions and obstacles. 2007.and microscopic parameters can be found in (Thornton. .. and  as a function of both macro. it was observed that the vertical flow velocity structure contained three parts. Shape factor  from simulations (markers). 2012a). 3. for case  = 1.) (dotted fine for simple flow scenarios. Flow velocity profiles for varying height for 4000 particles and  = 1. a linear profile near the surface and a convex profile near the base [z < b1hstop()/h) with b1 =   4.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Full details of the values of A. The values of  as a function of height and angle. It is anticipated that this closured continuum model will work Fig. 2012. Future directions We have now established closure relations for shallow-granular flows and the natural question of (21)   bs h the range of validity of closure relations derived udz  from this small steady periodic box simulations. we consider the closure the velocity shape factor. Closure for  Finally. 3/2   . particle free regions and static mate2 u dz 1 s COMPUTER METHODS IN MATERIALS SCIENCE – 204 – . is shown in figure 4. for complex flow containing 1:5. 4.

For this situation. s  s   h   h  (22) The present algorithm is designed to determine all the pairs in a set of N spherical particles in a ddimensional Euclidean space that overlap. it has been utilized since the beginning of particle simulations. 2012. For a given number of levels and cell sizes. 1993. 2011. L. and that is used for all the simulations in this paper. the hierarchical grid cells are defined by the following spatial mapping. according to the particle size distribution. Every regular grid is associated with a hierarchy level h  1. and is easily implemented in parallel codes (Form. however. The first d components of a(d + 1)-dimensional vector c represent cell indices (integers). Mapping phase The d-dimensional hierarchical grid is a set of L regular grids with different cell sizes. 2004. which include both the contact model and the contact detection algorithm. Stadler et al. 1993. M. It must be noted that the cell size of each level can be set independently. 1997).. Grids are ordered with increasing cell size so that h = 1 corresponds to the grid with smallest cell size. 1981. a fully twoway coupled code will have to be developed. of points r  R d to a cell at specified level h:   r1   rd    M : (r. Each level h has a different cell size sh  R. and  = amin/amax is the extreme size ratio.. This advanced contact detection algorithm is already implemented in Mercury (Thornton et al.2) for every particle in the system the potential contact partners are determined. where L is the integer number of hierarchy levels. This can effectively be addressed by the use of a multilevel contact detection algorithm (Ogarko & Luding.. sh < sh + 1.1. The most commonly used method for contact detection of nearly mono-sized particles with shortrange forces is the Linked-Cell method (Hockney & Eastwood. 2000) where the cell sizes are taken as double the size of the previous lower level of hierarchy. 2011).1. 1999). How to do this is explained by (Ogarko & Luding. every particle p can be mapped to its cell: .1. Algorithm The algorithm is made up of two phases.. hence sh + 1 = 2sh. The level of insertion h(p) is the lowest level where the cell is big enough to contain the particle p: – 205 – COMPUTER METHODS IN MATERIALS SCIENCE where [r] denotes the floor function (the largest integer nor greater than r). COLLISION DETECTION The performance of the DPM computation relies on several factors. The flexibility of independent sh allows one to select the optimal cell sizes. 1989). In the second ‘contact detection phase’ (subsection 5.. i.1. 2012). Ogarko & Luding. 5. The latter is limited whereas the former are not. An extensive review of various approaches to contact detection is given in Munjiza. Every particle is characterized by the position of its centre rp and its radius ap.e. the Linked-Cell method is unable to efficiently deal with particles of greatly varying sizes (Iwai. Thatcher. 2012). which will be the case in the next problem considered. to improve the performance of the simulations. amin and amax denote the minimum and the maximum particle radius. cp  M  rp . 5.). Raschdorf & Kolonko. If you were to undertake the task of collision detection in a naive fashion you would have to perform N2 checks were N is the number of particles. Allen & Tildesley. 1999). the open-source code developed here. 2. In the first ‘mapping phase’ all the particles are mapped into a hierarchical grid space (subsection 5. this becomes impractical even for relatively small systems.1. The performance difference between them is studied in (Muth. h  p    (23) where h(p) is the level of insertion to which particle p is mapped to.1). and the last one is the associated level of hierarchy. More discussion of problems associated with the development of this code can be found in chapter 7. respectively. 2007. where the cells are d-dimensional cubes. 5. in contrast to contact detection methods which use a tree structure for partitioning the domain (Ericson. For differently-sized spheres.INFORMATYKA W TECHNOLOGII MATERIAŁÓW rials it is likely to fail.…. and the geometrical intersection tests with them are made. Raschdorf & Kolonko. Due to its simplicity and high performance. The collision detection of short-range pair wise interactions between particles in DPM is usually one of the most time-consuming tasks in computations (Williams & O'Connor. h)  c =  . Nevertheless.. Using the mapping M. which we review here.

1. one more layer of cells (which also includes particle A). Any particle q from level h. to avoid testing the same particle pair twice. The search for potential contacts is performed in every cell c = (c1. To test two particles for contacts. where  is Euclidean norm. Then. and the small particles which are located in those cells are dark (green).5sh and ei is the standard basis forRd. and s2 = 2amax = 8 (a.6. two different sizes.e. cp. The cells where the cross-level search for particle B has to be performed from (1. However. 1989). performing such test first usually increases the performance.3. In figure 5.. Figure 5 illustrates a 2-dimensional two-level grid for the special case of a bi-disperse system with amin = 3/2. i. 2. Define the cells cstart and cend at level h as c start : M  rc . one searches for potential contacts only at levels h lower than the level of insertion: 1  h < h(p). and in its neighbour (surrounding) cells. … . 5.) and its position is rB = (10. first. Performance test COMPUTER METHODS IN MATERIALS SCIENCE Fig. i. cd.e. 2). particle B is mapped to the second level to the cell cB = (1..1) are marked in grey. The first step is the contact search at the level of insertion of p. Note. h) for which cistart  ci  ciend for all i  1. For a given particle p. the cell sizes of the two-level grid can be easily selected as the two diameters of each particle species. h  (25) where a search box (cube in 3D) is defined by rc = rp    d i 1 i e . 14. each particle which was mapped to one of these neighbour cells is tested for contact with particle p. The second step is the cross-level search. as will be considered for the segregation case in the next section. 1988) are tested for overlap.e. with centre xq outside this box can not be in contact with p.INFORMATYKA W TECHNOLOGII MATERIAŁÓW h  p  c p  min h : sh  2rp 1 h  L   (24) In this way the diameter of particle p is smaller or equal to the cell size in the level of insertion and therefore the classical Linked-Cell method (Allen & Tildesley.u. Correspondingly.u. i.. the level h = 1 cells where that search has to be performed (for particle B) are marked in grey. d  . and cd 1  h  h  p  (26) where ci denotes the i-th component of vector c. The search is done in the cell where p is mapped to. 5. 5. thus avoiding double checks for the same pair of particles. the axisaligned bounding boxes (AABB) of the particles (Moore & Wilhelms. For such systems. and cend : M  rc . half of the surrounding cells are searched. and s2 = 8. In other words. The first level grid is plotted with dashed lines while the second level is plotted with solid lines.4). Contact detection phase The contact detection is split into two steps. h(p). 1). for every particle pair which passed this test.2. that in the method of Iwai et al. The cross-level search for particle p (located at h(p)) with level h is detailed here: 1. the exact geometrical intersection test is applied (Particles p and q collide only if r p  rq < ap +aq. h  . (1999) the search region starts at cell (1. Since the overlap test for AABBs is computationally cheaper than for spheres. 2.3. two hierarchy levels are sufficient here. h(q) = h.). and cell sizes s1 = 3. The radius of the particle B is aB = 4 (a. 2.2. according to equations (23) and (24).. i. and the search is done by looping over all particles p and performing the first and second steps consecutively for each p. with  = ap + 0. size ratio  = 8/3. particle A is mapped to the cell cA = (4. 1).1) to (5.1. A 2-dimensional two-level grid for the special case of a bi-disperse system with cell sizes s1 = 2amin = 3. Only In this section we present numerical results on the performance of the algorithm when applied for bi-disperse particle systems. 1989) can be used to detect the contacts among particles within the same level of hierarchy. This implies that the particle p will be checked only against the smaller ones. for some – 206 – . since the diameter of the largest particle at this level can not exceed sh.e.). Since the system contains particles of only two different sizes. using the classical Linked-Cell method (Allen & Tildesley. Therefore.

We consider a bi-disperse size distribution with the same volume of small and large particles.e. 0. 2012).e. i. 1970. which is the ratio between small and large particle radius. the speed-up is increasing. Namely.631. migrate towards the bottom of the flow and lever the large particles upwards due to force imbalances. 0.7. The speed-up S of the two-level grid relative to the single-level grid (Linked-Cell method) for bidisperse systems with different size ratios . see (Ogarko & Luding. The motion of particles is governed by Newton’s second law with a linear elastic contact force during overlap. This distribution can be characterized by only one parameter. This was termed squeeze expulsion by Savage and Lun (1988). Their derivation has two key assumptions: firstly. 2011). The basic idea is: as the grains avalanche down-slope. the local void ratio fluctuates and small particles preferentially fall into the gaps that open up beneath them. We will focus on dense granular chute flows where kinetic sieving (Middleton. every particle undergoes only translational motion (without rotation) and gravity is set to zero.6.7.3 and N = 768 000 for  = 0. decreasing . Savage & Lun. 1983).. respectively. the volume fraction of systems with  = 0. it follows that a higher percentage of the small particles will be falling and. For more details on numerical procedure and preparation of initial configurations see (Ogarko & Luding.3.642. however. 2012).e. segregation due to size-differences is often the most important (Drahun & Bridgewater.2. in this convention 0 <   1. Since the number of voids available for small particles to fall into is greater than for large particles. Gray and Thornton (2005) developed a similar model from a mixture-theory framework. i. secondly.INFORMATYKA W TECHNOLOGII MATERIAŁÓW situations this may be not as efficient as the use of the single-level Linked-Cell method.1.4.665. 0. 0. but poorly understood. Later. role on the flow dynamics Iverson. the small and large particles) and. as they are more likely to fit into the available space than the large ones.2.723.. Figure 6 shows the speed-up S of the two-level grid relative to the single-level grid (Linked-Cell method). segregation plays an important. using a statistical argument about the distribution of void spaces.4. as the different particles percolate past each other there is a Darcystyle drag between the different constituents (i. The small particles. 6.635. Three independent runs were performed for every  and the average CPU time values are used for the calculation of S.2. The number of particles used is N = 128 000 for  > 0.. .682. 6. This is due to the overhead associated with cross-level tests. 0. 0. How the algorithm performs for polydisperse systems and how to select optimal cell sizes and number of levels for such systems is shown in (Ogarko & Luding.2 is   0. With increasing difference in particle size. particles falling into void spaces do not support any of the bed weight. 1988) is the dominant mechanism for particle-size segregation. 0.703.e.  > 0. For similar sizes of particles.652. The considered systems have volume fraction close to the jamming density. The maximum speed-up of 22 is achieved in the case of the lowest considered  = 0. There are many mechanisms for the segregation of dissimilar grains in granular flows. N = 256 000 for  = 0. as we show below. not sup- – 207 – COMPUTER METHODS IN MATERIALS SCIENCE . segregation effects can be observed in granular materials. For the influence of the volume fraction on the performance of the algorithm.9. For simplicity. 0. 0. 0. 2003. 1999).5. The first model of kinetic sieving was developed by Savage and Lun (1988). MICRO-MACRO FOR SEGREGATING FLOWS 6. Background Fig.8. hence. 0. i. We use homogeneous and isotropic disordered systems of colliding elastic spherical particles in a unit cubical box with hard walls. therefore.. Much higher speed-up is expected for  < 0.7 the speed-up exceeds unity and the use of the twolevel grid becomes advantageous. For  > 0. Shinbrot et al. In both natural and industrial situations. the use of the two-level grid slightly (within 40%) slows down the performance of the algorithm. Except for the very special case of all particles being identical in density and size. 0. 0. 0.

0 = 0:5 (to within the volume of one large particle). In general. especially considering there is only one degree of freedom. The simulations are performed with 5000 flowing small particles and the number of large particles is chosen such that the total volume of large and small particles is equal. Here are summarized the main results of (Thornton. In recent years. We limit our attention to small-scale DPM simulations. 2006) the effect of diffusive remixing (Gray & Chugunov. and the generalization to multi-component granular flows (Gray & Ancey.INFORMATYKA W TECHNOLOGII MATERIAŁÓW porting any of the bed load. where the ratio of these parameters was determined from DPM simulations.  is used to indicate a partial derivative. shear-rate. however. such as: size-ratio. periodic in the x and y-directions. The Micro-Macro transition     ˆ    v ˆ    w ˆ   u ˆ ˆ ˆ tˆ x y z        Sr 1       Dr   ˆ ˆ ˆ z z z  (27) COMPUTER METHODS IN MATERIALS SCIENCE where Sr is the dimensionless measure of the segregation-rate. v ˆ and w ˆ are the base coordinate. y ˆ. The fit is performed using non-linear regression as implemented in MATLAB. i. inclined at an angle of 25 to the horizontal. Therefore. with this definition only values between 0 and 1 are possible. y ˆ . which ranges from 1 to infinity. particle roughness.21 is the saturation constant and Pmax = 7.as a function of the spatial ˆ. Ps will be a function of the particle properties. The base was created by adding fixed small particles randomly to a flat surface.3ds long. which is periodic in x and y.2. – 208 – . is 5ds wide and 83. this segregation theory has been developed and extended in many directions: including the addition of a passive background fluid (Thornton et al. we will present the results in terms of  -1. whose form in the most general case is discussed in Thornton et al. we are inter- Figure 7 shows a series of images from the DPM simulations at different times and values of  -1. respectively. ˆ and z x The conservation law (27) is derived under the assumption of uniform porosity and is often solved subject to the condition that there is no normal flux of particles through the base or free surface of the flow. From these plots it is possible to obtain Ps as a function of  -1 and this was found to be given by: 1  Ps  Pmax 1  exp   k   1    (29) where k = 5. The simulations take place in a box. Furthermore u the dimensionless bulk velocity components in the ˆ. 6. at z ˆ and tom) and 1 (the top). however. This solution represents a balance between the last two terms of (27) and is related to the logistic equation. and ˆ is the downthe ‘hat’ a dimensionless variable. Ps. material properties. It should be noted that  has been defined such that it is consistent with the original theory of Savage and Lun (1988). The t is reasonable in all cases. In (27). 2006). (2006) and Dr is a dimensionless measure of the diffusive remixing. and we will use DPM to investigate the dependence of Ps on the particle size ratio  = ds/dl. and investigate the final steady-states. We will use the two-particle size segregation-remixing version derived by Gray and Chugunov (2006).e. and time tˆ . y ˆ directions. 2006) takes the form of a nondimensional scalar conservation law for the small particle concentration . x ˆ normal to ˆ the cross-slope and z slope coordinate. The bi-dispersed segregation remixing model contains two dimensionless parameters.. These in general will depend on flow and particle properties. ˆ and z coordinates x ested in a steady state solution to (27) subject to noˆ = 0 (the botnormal flux boundary condition.. One of the weaknesses of the model is that it is not able to predict the dependence of the two parameters on the particle and flow properties. by using phenomenological arguments. etc. Figure 8 shows a fit of equation (28) to the small particle volume fraction for several cases. The two-particle segregation-remixing equation (Gray & Chugunov. 2011). that is independent of x ˆ . it should be noted that Dolgunin and Ukolov (1995) were the first to suggest this form. Therefore.35. Gray and Chugunov (2006) showed that such y a solution takes the form:   1  exp  0 Ps    exp 0  z  Ps 1  exp    1  0  Ps   1  exp  0 Ps    exp 0  z  Ps (28) where Ps = Sr/Dr is the segregation Peclet number and 0 is the mean concentration of small particles. slope angle. 2013).

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

Fig. 7. A series of snapshots from the DPM simulations with large (orange) and small (blue) particles. The rows correspond to distinct particle sizes and columns to different times. Along the top row  -1 = 1.1, middle row  -1 = 1.5 and bottom row  -1 = 2; whereas, the left column is for t = 1, middle t = 5 and right t = 60.

simulation data and the blue lines are the fit to equation (28) produced with MATLAB’s non-linear regression function. For the fit only Ps is used as a free parameter. Dotted lines shows the 95% confidence intervals for the fit.

6.3.

Future directions

For the segregation case the micro-macro transition has been shown to be useful in establishing the relations between the parameters that appear in the continuum descriptions and the material parameters of the particles. Additionally, in this case, we highlighted a discrepancy between the particles simulations and theory, see figure 8, i.e. the inflection point near the base in the simulation concentration profiles. Further analysis of the simulation data has shown that this discrepancy arises because the particle diffusion is not constant with depth, as assumed

by the model. Therefore, for this situation the model has to be improved to capture the full dynamics in these situations. From a modeling point of view one of the opentopic at the moment is the determination of segregation shallow-water models, see e.g. (Woodhouse, 2012), but this is beyond the scope of this review. 7. CONCLUSIONS AND FUTURE PERSPECTIVE Here, we have shown that continuum parameters such as the macroscopic friction can be accurately

– 209 –

COMPUTER METHODS IN MATERIALS SCIENCE

ˆ f . The black lines are the coarse-grained DPM Fig. 8. Plots of the small particle volume fraction  as a function of the scaled depth z

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

extracted from particle simulations. We have shown that the micro-macro transition can be achieved using small particle simulations, i.e., we can determine the closure relations for a continuum model as a function of the microscopic parameters. Here, this one-way coupling from micro- to macro-variables was achieved for steady uniform flows, but can in principle be used to predict non-uniform, timedependent flows, provided that the variations in time and space are small. Comparisons with large-scale experiments and large DPM simulations are needed to determine the range of parameters for which the steady uniform closure laws hold, as indicated in figure 1a. However, for strongly varying flows, such as arresting flows, avalanching flows, flow near boundaries or near abrupt phase changes (dead zones, shocks), no closure relations in functional form are known. Ideally, the full 3D granular flow rheology could be determined in the full parameter space and then introduced into a pure continuum solver. However, since the parameter space is just too wide and situations can and will occur that are not covered by a systematic parameter study, other strategies and approaches can be thought of. For such interesting situations, where the rheology enters unknown regimes, or where changes are too strong and/or fast, a two-way coupling to a particle solver is a valid approach. If these complex regions are small, one can use a two-way boundary coupling, where a particle solver is used in the complex region and a continuum solver in the remaining domain, with an overlapping region where both solvers are used and where the two methods are coupled by using suitable boundary conditions (Markesteijn, 2011). Alternatively, if the complex regions are too large to be solved by particle simulations, one can use a continuum solver where a small particle simulation is run each time the closure relations need to be evaluated (Weinan et al., 2007). This particle simulation is two-way coupled to the continuum solution in the sense that it has to satisfy the parameters set by the continuum solution (such as height, depth-averaged velocity and depth-averaged velocity gradient and boundary conditions) and return the closure relations (such as friction and velocity shape factor). Both alternative strategies provide plenty of unsolved challenges in communicating between discrete and continuous "worlds" concerning nomenclature, parameters, boundary conditions and their respective control.

The next versions of both the in-house continuum solver hpGEM (van der Vegt et al.) DPM code Mercury (Thornton et al.), are designed such that they can be easily coupled and hence used to form the basis of a granular two-way coupled code. Acknowledgements. The authors would like to thank the late Institute of Mechanics, Processes and Control, Twente (IMPACT) for the primary financial support of this work as part of the research program “Superdispersed multiphase flows". The DPM simulations performed for this paper are undertaken in Mercury-DPM, which was initially developed within this IMPACT program. It is primarily developed by T. Weinhart, A. R. Thornton and D. Krijgsman as a joint project between the Multi-Scale Mechanics (Mechanical Engineering) and the Mathematics of Computational Science (Applied Mathematics) groups at the University of Twente. We also thank the NWO VICI grant 10828, the DFG project SPP1482 and B12 the Technology Programme of the Ministry of Economic Afairs (STW MuST project 10120) for financial support. REFERENCES
Allen, M.P., Tildesley, D.J., 1989, Computer simulations of liquids, Clarendon Press, Oxford. Babic, B., 1997, Average balance equations for granular materials, Int. J. Eng. Science, 35 (5), 523-548. Bokhove, O., Thornton, A.R., 2012, Shallow granular flows, in: Fernando, H.J., ed., Handbook of environmental fluid dynamics, CRC Press. Cui, X., Gray, J.M.N.T., Johannesson, T., 2007, Deflecting dams and the formation of oblique shocks in snow avalanches at flateyri, JGR, 112. Cundall, P.A., Strack, O.D.L., 1979, A discrete numerical model for granular assemblies, Geotechnique, 29, 47-65. Dalbey, K., Patra, A.K., Pitman, E.B., Bursik, M.I., Sheridan, M.F., 2008, Input uncertainty propagation methods and hazard mapping of geophysical mass flows, J. Geophys. Res., 113. Denlinger, R.P. Iverson, R.M., 2001, Flow of variably fluidized granular masses across three-dimensional terrain, 2. numerical predictions and experimental tests, J. Geophys Res., 106 (B1), 533-566. Dolgunin, V.N., Ukolov, A.A., 1995, Segregation modelling of particle rapid gravity flow, Powder Tech., 83 (2), 95-103. Drahun, J.A., Bridgewater, J., 1983, The mechanisms of free surface segregation, Powder Tech., 36, 39-53. Ericson, C., 2004, Real-time collision detection (The Morgan Kaufmann Series in Interactive 3-D Technology), Morgan Kaufmann Publishers Inc., San Francisco. Form, W., Ito, N., Kohring, G.A., 1993, Vectorized and parallelized algorithms for multi-million particle MD-simulation, International Journal of Modern Physics C, 4(6), 10851101.

COMPUTER METHODS IN MATERIALS SCIENCE

– 210 –

INFORMATYKA W TECHNOLOGII MATERIAŁÓW GDR Midi, 2004, On dense granular flows, Eur. Phys. J.E., 14, 341-365. de Gennes, P.G., 2008, From rice to snow, in: Nishina Memorial Lectures, vol. 746 of Lecture Notes in Physics, Springer, Berlin/Heidelberg, 297-318. Gilbert, L.E., Ertas, D., Grest, G.S., Halsey, D., Levine, T.C., Plimpton, S.J., 2001, Granular flow down an inclined plane: Bagnold scaling and rheology, Phys. Rev. E., 64 (051302). Goldhirsch, I., 2010, Stress, stress asymmetry and couple stress: from discrete particles to continuous fields, Granular Matter, 12 (3), 239-252. Gray, J.M.N.T., Ancey, C., 2011, Multi-component particle-size segregation in shallow granular avalanches, J. Fluid Mech., 678, 535-588. Gray, J.M.N.T., Chugunov, V.A., 2006 Particle-size segregation and diffusive remixing in shallow granular avalanches, J. Fluid Mech., 569, 365-398. Gray, J.M.N.T., Cui, X., 2007, Weak, strong and detached oblique shocks in gravity driven granular free-surface flows, J. Fluid Mech., 579, 113-136. Gray, J.M.N.T., Tai, Y.C., Noelle, S., 2003, Shock waves, dead zones and particle-free regions in rapid granular free surface flows, J. Fluid Mech., 491, 161-181. Gray, J.M.N.T., Thornton, A.R., 2005, A theory for particle size segregation in shallow granular free-surface flows, Proc. Royal Soc. A, 461,1447-1473. Hakonardottir, K.M., Hogg, A.J., 2005, Oblique shocks in rapid granular flows, Phys. Fluids. Hockney, R.W., Eastwood, J.W., 1981, Computer simulation using particles, McGraw- Hill, New York. Hungr, O., Morgenstern, N.R., 1984, Experiments on the flow behaviour of granular materials at high velocity in an open channel, Geotechnique, 34 (3). Irwing, J.H., Kirkwood, J.G., 1950, The statistical mechanical theory of transport processes, iv. the equations of hydrodynamics, The Journal of Chemical Physics. Iverson, R.M., 2003, The debris-flow rheology myth, in: Rickenmann and Chen, eds, Debrisow hazards havards mitigation: Mechanics, prediction and assessment, Millpress, 303-314. Iwai, T., Hong, C.W., Greil, P., 1999, Fast particle pair detection algorithms for particle Simulations, Int. J. Modern Physics C, 10(5), 823-837. Jenkins, J.T., Savage, S.B., 1983, A theory for the rapid flow of identical, smooth, nearly elastic, spherical particles, Journal Fluid. Mech., 130, 187-202. Luding, S., 2004, Micro-macro models for anisotropic granular media, in: Vermeer, P.A., Ehlers, W., Herrmann, H.J. Ramm, E., eds, Micro-macro models for anisotropic granular media, Balkema A.A., Leiden, 195-206. Luding, S., 2008, Cohesive, frictional powders: contact models for tension, Granular Matter, 10 (4), 235-246. Luding, S., Laetzel, M., Volk, W., Diebels, S., Hermann, H.J., 2001, From discrete element simulations to a continuum model, Comp. Meth. Appl. Mech. Engng., 191, 21-28. Lun, C.K.K., Savage, S.B., Jeffrey, D.J., Chepurniy, N., 1984, Kinetic theories for granular flow: inelastic particles in couette flow and slightly inelastic particles in a general flow field, J. Fluid Mech, 140, 223. Markesteijn, A.P., 2011, Connecting molecular dynamics and computational fluid dynamics, Ph.D. Thesis, University of Delft. Middleton, G.V., 1970, Experimental studies related to problems of flysch sedimentation, in: Lajoie, J., ed., Flysch Sedimentology in North America, Business and Economics Science Ltd, Toronto, 253-272. Moore, M., Wilhelms, J., 1988, Collision detection and response for computer animation, Computer Graphics (SIGGRAPH '88 Proceedings), 22 (4), 289-298. Munjiza, A., 2004, The combined finite-discrete element method, John Wiley & Sons Ltd. Muth, B., Mueller, M.K., Eberhard, P., Luding, S., 2007, Collision detection and administration for many colliding bodies, Proc. DEM07, 1-18. Ogarko, V., Luding, S., 2010, Data structures and algorithms for contact detection in numerical simulation of discrete particle systems, Proc. 6th Word Congress on Particle Technology WCPT6, Nuremberg. Ogarko, V., Luding, S., 2011, A study on the influence of the particle packing fraction on the performance of a multilevel contact detection algorithm, in: Onate, E., Owen, D.R.J., eds, II Int. Conf. on Particle-based Methods Fundamentals and Applications, PARTICLES 2011, Barcelona, Spain, 1-7. Ogarko, V., Luding, S., 2012, A fast multilevel algorithm for contact detection of arbitrarily polydisperse objects, Computer Physics Communications, 183 (4), 931-936. Pesch, L., Bell, A., Sollie, W.E.H., Ambati, V.R., Bokhove, O., Vegt, J.J.W., 2007, hpgem a software framework for discontinous galerkin finite element methods, ACM Transactions on Mathematical Software, 33 (4). Pouliquen, O., 1999, Scaling laws in granular flows down rough inclined planes, Phys. Fluids, 11 (3), 542-548. Pouliquen, O., Forterre, Y., 2002, Friction law for dense granular flows: application to the motion of a mass down a rough inlined plane, J. Fluid Mech., 453, 131-151. Raschdorf, S., Kolonko, M., 2011, A comparison of data structures for the simulation of polydisperse particle packings, Int. J. Num. Meth. Eng., 85, 625-639. Savage, S.B., Hutter, K., 1989, The motion of a finite mass of material down a rough incline, Journal Fluid. Mech., 199, 177-215. Savage, S.B., Lun, C K.K., 1988, Particle size segregation in inclined chute flow of dry cohesionless granular material, J. Fluid Mech., 189, 311-335. Schofield, P., Henderson, J.R., 1982, Statistical mechanics of inhomogenous fluids, Proc. R. Soc., 379, 231-246. Shen, S., Atluri, S.N., 2004, Atomic-level stress calculation and continuum molecular system equivalence, CMES, 6 (1), 91-104. Shinbrot, T., Alexander, A., Muzzio, F.J., 1999, Spontaneous chaotic granular mixing, Nature, 397 (6721), 675-678. Stadler, J., Mikulla, R., Trebin, H.R., 1997, IMD: A software package for molecular dynamics studies on parallel computers, Int. J. Modern Physics C, 8 (5), 1131-1140. Thatcher, U., 2000, Loose octrees, in: deLoura, M., ed., Charles River Media. Thornton, A.R, Weinhart, T., Krijgsman, D., Luding, S., Bokhove, O., Mercury md. http://www2.msm.ctw.utwente.nl/ athornton/MD/. Thornton, A.R, Gray, J.M.N.T., Hogg, A.J., 2006, A three phase model of segregation in shallow granular free-surface flows, J. Fluid Mech., 550, 1-25. Thornton, A.R, Weinhart, T., Luding, S., Bokhove, O., 2012, Modelling of particle size segregation: Calibration using

– 211 –

COMPUTER METHODS IN MATERIALS SCIENCE

granular temperature. Problemy te są odległe od omówionych wcześniej rozwiązań dla krótkich okresów stacjonarnych. Thornton. From discrete particles to continuum fields near a boundary.W. Granular Matter. 2008. T. C. A. Segregation induced fingering instabilities in granular avalanches. 52(2). A. van der Vegt. Journal of Rheology.R.. S.. Kokelaar. WIELOSKALOWE METODY DLA WIELOSKŁADNIKOWYCH MATERIAŁÓW ZIARNISTYCH Streszczenie W artykule przedstawiono postęp w zrozumieniu przepływu przesypywanych materiałów ziarnistych. 233-269. B. Vreman. JFM. J. van Sint Annaland. 2007. Bokhove. Luding. Friction dependence of shallow granular flows from discrete particle simulations. Williams. E.. 709. O’Connor. Vanden-Eijnden. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 212 – .. Closure relations for shallow granular flows from particle simulations. Thornton. 14.. jaki został ostatnio osiągnięty dzięki technikom modelowania wieloskalowego.. Al-Tarazi. D..F. A.. Evans.R.. Rev. A..R. Sheridan.R. Phys. Phys. C. Engquist.msm.J. Luding. Williams.. Kuipers. M. Walton. 2012b. J. 1999. 2012 Received in a revised form: November 4.. 1240014. Pokazano jak zaawansowane symulacje DPM mogą być zastosowane do uzyskiwania zależności dla modelu ciągłego (mapowanie zmiennych i funkcji między skalami mikro i makro). Bokhove.. 30 (949). O.INFORMATYKA W TECHNOLOGII MATERIAŁÓW the discrete particle method. JFM. Received: September 20.. M. Bokhove. 2007. Archives of Computational Methods in Engineering.M.J. frictional disks.. A. Na początku omówiono metodę dyskretnych cząstek (ang... Mod. Granular Matter. W pracy badano jak ten współczynnik zależy od własności przepływających cząstek i powierzchni. aby model był spójny z makroskopową zasadą zachowania masy i pędu. Weinhart. wzajemnie dopełniające się zależności opisujące ciągłe własności materiału. 2012 Accepted: November 21.R... R.ctw. bardziej zaawansowane rozwiązania. Phys. 543-580. dla których te zależności były otrzymywane. J.. 2012. E. 2013. Drugim analizowanym problemem jest segregacja przy zsypie polidyspersyjnych cząstek.. 367450.. 2012a.. W. Gray. Communications in Computational Physics. Pierwszym jest płynięcie w płytkim zsypie.N. 177.B.. 2 (3).. 531-552. Discrete element simulation and the contact problem. W obydwu analizowanych problemach rozważono krótkie okresy stacjonarne w symulacji DPM.. Woodhouse. 279-304. Thornton. Letts.R. E. A. 289-294. J. Johnson. Bokhove.L.R. która może być wykorzystana do wieloskładnikowych materiałów sypkich o rozmiarze cząstek zmieniających się w zakresie rzędów wielkości. hpgem. Bokhove. gdzie główną niewiadomą w modelu kontinuum jest makro współczynnik tarcia. Submitted to Eurp.. Weinan. P. 578. względem której cząstki przepływają. 6 (4). 1995. Ren. O.... Stinton. W pracy omówiono również problem dokładności i poprawności opisanych wzajemnie dopełniających się zależności dla złożonych problemów dynamicznych. J. To umożliwiło rozwój modeli kontinuum zawierających informację o mikrostrukturze materiałów sypkich bez potrzeby robienia dodatkowych założeń.J.. S. A. w których makrokontinuum i mikro dyskretny model są połączone w sposób dynamiczny ze sprzężeniem zwrotnym. Todd. Pressure tensor for inhomogeneous fluids. Li.nl/athornton/hpGEM/. O. P. M. R.DPM) i wyjaśniono w jaki sposób należy konstruować ciągłe pole na podstawie dyskretnych danych tak. O. Supercritical shallow granular flow through a contraction: experiment.. Luding. O. D. S.. W bardziej skomplikowanych sytuacjach potrzebne są nowe.. Thornton. Discrete particle method . 1986. Braun..M. 14.. M.. Thornton. T. Submitted to Int.. T. O. aby otrzymać realistyczne.. Evaluation of the Titan 2D two-phase flow model using an actual event: Case study of the vazcun valley lahar. Viscosity.R. X. Heterogeneous multiscale methods: a review. Przejście mikro-makro przedstawiono dla dwóch problemów płynięcia materiału sypkiego.T.utwente. and stress calculations for shearing assemblies of inelastic. Weinhart. http://www2.. Dla prostych przypadków zastosowanie zdefiniowanych wzajemnie dopełniających się zależności dawało poprawne wyniki. Weinhart. A.J.. Journal of Volcanology and Geothermal Research. 23(8). 760-766.W. Zaprezentowano też nową metodę wykrywania kontaktu. theory and simulation. Daivis.

where homogenization cannot be used and the hp-mixed FEM discretization details. The paper presents application of one of the most efficient numerical techniques. 2012) and it may be used both at the micro and macro-levels. In the micro-level it is required to guarantee accurate representation of heterogeneity and at both scales the optimal number of degrees of freedom should be used. mixed FEM 1. 13. Warszawska 24. The stability of this approach is a difficult task (Arnold et al. In order to increase reliability of the results the error of homogenization is estimated (Cecot et al. 2012). Since stresses are of primary interest we decided to use the mixed FEM. Kouznetsova et al.pl Abstract Computational homogenization enables replacement of a heterogeneous domain by an equivalent body with effective material parameters.. Approach that we use is based on two-scale micro/macro analysis. but also large enough to contain all information about material heterogeneity.. We propose assessment of homogenization error by additional analyses in selected subdomains with boundary conditions determined by the homogenized solution. which are small enough to satisfy separation scale condition.edu. rather than by derivatives of displacements.e. 2002). It is well known that the method gives the fastest convergence for linear problems. 2003. 2004) without loosing its accuracy. which are associated to independent RVE. INTRODUCTION Even though numerical homogenization speeds up solution of real-life problems for heterogeneous materials the time of computation may be very large. We have confirmed that one may expect the same situation for inelastic problems (Serafin and Cecot. approximation of solution in the whole domain is performed. In the macro-scale the material is assumed as a homogeneous with the effective material parameters obtained during RVE analysis..COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol.pk. in which stresses are approximated directly. Key words: homogenization. Even though such a homogenization significantly reduces the time of computation. 2013. 2007) but efficient and stable hp-mixed FEM is possible if appropriate weak formulation and shape functions are used. that is crucial for determination of parts of the body. ul. Numerical analysis is performed by the automatic hp-adaptive version of FEM (Demkowicz et al.. especially if nonlinearity is accounted for. the efficiency and accuracy of the analysis are still not trivial issues. 31-155 Kraków *Corresponding author: mserafin@L5. automatic hp-adaptive FEM that enables a user to obtain error-controlled results in rather short time. assessment of homogenization error. In the micro-scale heterogeneous properties are collected in so-called representative volume elements (RVE). The coupling between both scales is provided at the selected macro-level points. Therefore. No. representative volume element. i. Residuum of the differential equa- 213 – 218 ISSN 1641-8581 . adaptive finite element method. 2 NUMERICAL ASPECTS OF COMPUTATIONAL HOMOGENIZATION MARTA SERAFIN*. WITOLD CECOT Cracow University of Technology. 2006. we discuss in this paper certain numerical aspects that should increase efficiency of computational homogenization (Feyel. Gitman. Then.

N . which is a place of lower solution regularity. It means that for the coarse mesh the optimal (h. proposed by Demkowicz et al. denoted here for the sake of brevity by that serves as a substitute for the exact solution. for coercive problems by the Cea's lemma. In this paper convergence of hp adaptation strategy for elastic-plastic problems is examined for twoscale modeling and some modifications of the algorithm are proposed. (2002). ∏ . Nopenetration boundary conditions were assumed on the left and bottom sides. which is a good upper bound of the best approximation error that in turn. a fast two-grid solver may be used to guide the optimal hp-refinement. Comparison of meshes obtained by original and modified hp – 214 – . Material parameters are collected in table 1. superior to h and p adaptation techniques. is the upper bound for the actual approximation error. The strategy proposed by Demkowicz et al. After additional h-refinements of elements.INFORMATYKA W TECHNOLOGII MATERIAŁÓW tion for heterogeneous body may also be used as another criterion for detection of subdomains with large discrepancy between the exact and homogenized solutions. Material parameters Young modulus (GPa) Poisson ratio Yield strength (MPa) inclusion 300 0. However. ∏ . which contained both elastic and plastic parts. 2007. are considered. (2002) is based on the interpolation error estimate. Material parameters. p or hp) refinement is determined by maximizing the following expression both h and p is performed and the optimal mesh is selected by maximization of the function ω defined by equation (1). denote H projection-based interpolants on the current and optimal meshes. inelastic deformations should be accounted for in a special way in a-posteriori error estimates in order to obtain appropriate stress approximation accuracy. According to literature (Barthold et al. Boundary conditions. Zero and constant loading (220 MPa) were applied at the top and right edges. Additional h-refinement is proposed along elastic-plastic interface. This automatic mesh adaptation was successfully used for various linear problems. 2. Gallimard. Table 1. Its key point is an appropriate strategy of anisotropic h.. . ADAPTIVE FINITE ELEMENT METHOD In this paper automatic hp-adaptive finite element method. only partially convergent solution obtained by e. RVE. specified in figure 1. Its main adventage is exponential convergence. where only algebraic convergence may be obtained. A quarter of the domain was considered (figure 1). 1. The maximization is performed by search over a suitable subset of all possible hp refinements for every coarse mesh element. Fig.3 300 30 matrix 100 0. elastic-plastic zone was successfully detected. is used for numerical analysis. ω M .3 200 10 COMPUTER METHODS IN MATERIALS SCIENCE ω ∏ ∏ . ). the refinement in To verify the efficiency of automatic hp-adaptive FEM for inelastic problems RVE with cylinder-like inclusion was analyzed. (1) Hardening coefficient (GPa) with additional assumption that the mesh is oneirregular. Thus. where M . respectively. and . Such an algorithm is called here modified (automatic) hp-refinement. respectively. ). normality rule with kinematic hardening and boundary conditions. Both materials underwent elastic-plastic deformations. The difference between and its interpolant approximates the interpolation error and the optimal anisotropic mesh refinement is determined in such a way that the reduction of the interpolation error per number of additional degrees of freedom is maximal. More precisely plane strain problem with the Mises yield condition. 1998. max ω M . M . the algorithm of adaptation approach starts with the solution of the problem on the current (coarse) mesh ( / . Cecot. Then. The aforementioned interpolation error is estimated making use of a fine mesh solution ( / . elastic and plastic zones. For large problems computation of the fine mesh solution may be time-consuming. 1996). Such an ''exact'' solution is interpolated locally by the possible new hp-refined meshes.g. N are the numbers of degrees of freedom in optimal and current meshes. p or hp mesh refinement. denote arbitrary and the optimal new meshes.

The problem has the following form: find ∈ L Ω. Qiu & Demkowicz. respectively. stable mixed finite elements for solid mechanics are very difficult to construct. One may also use formula obtained by Piola transformation that gives the following stress shape function possessing the same properties x. ds v.u dΩ Ω τ.002GPa and 200GPa.ξ x. V .divσ dΩ Ω Ω v. Simulation of tension was performed and this way effective material parameters were computed. 2007.ξ y. Ω. However.3). in which stress tensor symmetry is enforced in a weak way (Arnold et al. 2009). Mesh after 40 steps of original and 20 steps of modified hp-refinements (grey scale indicates order of approximation). 3.ξ y. examples one may conclude that the original automatic hp algorithm works well for inelastic problems. even though elastic-plastic interface is not detected in accurate way. Fig.INFORMATYKA W TECHNOLOGII MATERIAŁÓW algorithm is shown in figure 2.p dΩ ∂Ω τn. rather than all the stress components. Convergence of effective Young modulus obtained by classical FEM (displacement formulation) and mixed method (displacement-stress formulation) is compared in figure 6. ∀ ∈ L Ω. The example of a tensor shape function that enables obtaining continuous tractions at every point of element interfaces may have the following form x.η y. J stands for Jacobi matrix of transformation between those elements. Comparison of traction components continuity of approximation by equations (3) and (4) is shown in figure 4. ∀ ∈ L Ω.ξ /detJ (4) σx = 0 0 The main difference between both formulas is in definition of degrees of freedom (traction in normal/tangent or x/y directions). The RVE with square-like inclusion in plane strain state was considered as a benchmark for the proposed mixed approximation. RVE. x and y are coordinates of physical element.η /detJ (3) σx = -x. τ:C-1 σ dΩ Ω Ω divτ. Such approximation of stresses is useful in multiscale computations. RVE. Ω. M . M . 2.ξ x. A quarter of the domain was taken into account (figure 5) with appropriate boundary conditions. K is the space of skewsymmetric tensors. K ∈ H div.b dΩ (2) q. V and ∈ L Ω. Convergence of error norm. One may use a modified Hellinger-Reissner principle. Deformations were only in elastic range.. Poisson ratio for both materials: 0. One may observe. but not necessary symmetric tensors. that if we use both methods with small number of degrees of freedom we are NDOF = 21048 NDOF = 54052 Fig. 3.η where ξ and η denote coordinates of master element. such that: – 215 – COMPUTER METHODS IN MATERIALS SCIENCE . where homogenization is based on evaluation of stress in the micro-scale.η -y. They have to provide symmetry of stress tensor and continuity of only traction across interelement boundaries.ξ y. not presented here. In this example inclusion was much weaker than the matrix (Young modulus: 0. MIXED FINITE ELEMENT METHOD Mixed method enables independent approximation of at least two fields. On the basis of this and other.σ dΩ 0 Ω ∀ ∈ H div. Convergence history is presented in figure 3. K where M is the space of second order.

for critical zones. where the error should be estimated and consider heterogeneous material in this subdomain. it may happen that the homogenization should not be used for certain part of the domain. Boundary conditions. mation analysis of macrostructures. 4. 2) select a part of the body.INFORMATYKA W TECHNOLOGII MATERIAŁÓW able to evaluate effective value with a good accuracy as an average of both solutions. No-penetration boundary 3) solve the boundary value problem for cut-off conditions were assumed on the left and bottom sides. Here. Such an error estimate consists of a few steps: 1) compute the homogenized solution u0 in domain Ω. (1996). The proposed algorithm of homogenization error estimation may be stated in the following way: 1) compute effective material parameters for homogenized domain. assumed on the basis of homogenized solution. obtain the solution uI and consider it in a smaller truncated part of the selected heterogeneous domain . Zero and heterogeneous subdomain A with boundary conconstant loading were applied at the top and right edges.mains with boundary conditions tions evaluated for stress fields of adjacent elements along this edge. 1978). Subsequently. – 216 – . it is not able to capture local error. Thus. RVE. Traction components continuity. Effective Young modulus. 4. Fig. However. are identified based on a postprocessing step on the homogenized solution. 5. One is based on the solution of heterogeneous problem in selected subdoFig. 4) estimate the error between solution uI and homogenized one u0 in subdomain . MODELING ERROR ESTIMATION Replacement of a heterogeneous body by a homogenized one with effective material parameters introduces an error. exact microstructural representation is introduced. COMPUTER METHODS IN MATERIALS SCIENCE Fig. A global explicit estimate using the homogenized (coarse) elasticity tensor and the actual fine-scale elasticity tensor was proposed by Zohdi et al. A scale adaptation strategy developed by Temizer and Wriggers (2011) was used to account for loss of accuracy for the finite defor- ditions resulted from homogenized solution. The arrows along the common edge denote trac. 1987. without intermediate models. by analogy to the explicit residual error indicator for FEM solution (Babuska & Miller. 6. related to incomplete information about the microstructure. Another possibility is based on residuum. other possibilities of modeling error estimation are presented. In that method the adaptation zones that correspond to regions with high strain-gradients. Babuska & Rheinboldt. since this issue is essential for reliability of the results.

A part of the domain (reentrant corner of Lshaped domain – figure 8) was selected and estimation by subdomain solutions was used. 12. I.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 2) solve homogenized problem with effective properties to obtain u0 .e. Mech. automatic hp mesh adaptation and mixed approximation for inelastic two-scale analysis. 1987. Mathematics of Computations. 8. 7. CONCLUSIONS The paper presents application of two efficient numerical techniques. Boundary displacements for cut-off heterogeneous domain A resulted from homogenized solution. R. Residuum in each macroelement of homogenized body was calculated and error indicators were evaluated. as well as material distribution. 3) compute residuum R of equilibrium equation div for heterogeneous body in each macro-scale finite element. REFERENCES Arnold. Part 1. Assumed admissible error enables selection of subdomains. Winther.. body forces. 9. distributed uniformly. The L-shaped domain in plane strain state was considered to perform numerical tests. i. Meth. Comp... Numerical improvements of computational homogenization.13% The residual modeling error estimate was also used for this example. Num.B 0.B ‖u0 ‖0. where denotes stands for stress tensor (in fact. I. 1978. J. Int. 1597-1615. its norm. m2) interfaces. Appl. In order to obtain reliable results the error introduced by homogenization was estimated giving information about quality of the results. Fig. A. components ( where ∂ denotes common edge of adjacent elements and ∂ stands for material (m1. W.N.. New material distribution after excluding. L-shaped domain with homogeneous and heterogeneous subdomains.C. This research was supported by the National Science Center under grant 2011/01/B/ST6/07312. Engng. Engng. may be bounded by the norms of regular part of and jumps of the first derivatives of the solution). 5) wherever ηR h‖R‖0 h1/2‖Je ‖0 h1/2‖Jm‖0 is large. Babuska. Miller. L-shaped domain. A feedback finite element method with a posteriori error estimation. Fig. on the basis of residual error estimate.. Boundary conditions and material distribution. Rheinboldt. Error estimates for adaptive finite element computations. is a distribution. L-shaped domain. will be used in further applications of the approach in modeling of elastic-plastic composites Acknowledgment. 2007. which should not be homogenized (figure 9). 4) compute jumps of tractions at interfaces of finite on ∂ ) and material elements ( on ∂ ⊂ ∂ ). Assumed boundary conditions.. D. Prospects of both approaches were positively verified by solution of selected numerical examples. 5. Babuska. Meth. Estimated error for displacements is as follows ‖u u0 ‖0. presented in this paper. 76. The metal matrix was reinforced by cylinder-like inclusions. Mixed finite element methods for linear elasticity with weakly imposed symmetry. Fig.. 1-40. homogenization should not be used. which we are interested in. R. Automatic hp refinements enables obtaining a mesh that captures all the details of heterogeneity in selected subdomains. two subdomains from homogenization and appropriate FE mesh (gray scale indicates order of approximation). – 217 – COMPUTER METHODS IN MATERIALS SCIENCE . 1699-1723. Falk. 61. are shown in figure 7.

Int. Geers... Engng. Klimczak. J. J. Numer. F. I. Serafin. hp-adaptacyjną wersję metody elementów skończonych.. W publikacji zawarto również możliwości oszacowania błędu homogenizacji. 2006. jednak należy zagwarantować poprawność uzyskanych wyników. Temizer. Ph.. Pelle. W. 1996. Reliability of computational homogenization. Gitman. 200. Gallimard. Methods Appl... 1998. International Journal for Multiscale Computational Engineering. 201 COMPUTER METHODS IN MATERIALS SCIENCE – 218 – . Representative Volumes and Multi-scale Modelling of Quasi-brittle Materials... Schmidt. Cecot. Transfer informacji między skalami dokonywany jest w wybranych punktach skali makro. który jest na tyle mały. Engrg.. Mixed hp-finite element method for linear elasticity with weakly imposed symmetry. Engrg. W.I. Engrg.. t. która pozwala na uzyskanie wiarygodnych wyników w stosunkowo krótkim czasie oraz sformułowanie wielopolowe pozwalające uzyskać możliwie dokładną aproksymację naprężeń. Mech.. 3233-3244. Feyel. Rodin.J. Error estimation and adaptivity in elastoplasticity... Appl. International Journal for Numerical Methods in Engineering (submitted). PhD thesis. Self hp-adaptive FEM for elasticplastic problems. W skali makro zakłada się materiał jednorodny z efektywnymi parametrami materiałowymi otrzymanymi z analizy RVE. Demkowicz. 2639-2661.. Serafin. Comp..INFORMATYKA W TECHNOLOGII MATERIAŁÓW Barthold. 1996.. J. W. L. M. W skali mikro rozważa się materiał niejednorodny w tzw.. 2007. 2011.P.. Adaptive FEM analysis of selected elasticvisco-plastic problems. 39. Demkowicz. Ladeveze. 17. będących głównym celem obliczeń.. International USPoland Workshop: Multiscale Computational Modeling of Cementitious Materials. 273-298. Engng.. W. 196. Mech. żeby zapewnić separację skal. T. A multilevel finite element method (FE2) to describe the response of highly non-linear structures using generalized continua.. 3682-3701. 2012. Mech. 198. W pracy zastosowano dwie efektywne techniki numeryczne. Stein. Cecot. M..j. aby informacje o wszystkich niejednorodnościach zostały w nim zawarte. 225-238. 2009. Meth.. Size of a representative volume element in a second-order computational homogenization framework. Methods Appl. 183-194. Computational Mechanics. 2003. An adaptive multiscale resolution strategy for the finite deformation analysis of microheterogeneous structures. 138. 3859-3870. 127-155. Appl. Rachowicz. niezbędnego do wyznaczenia obszarów. P. 2004. Devloo. 2002. 192. Mech. NUMERYCZNE ASPEKTY HOMOGENIZACJI OBLICZENIOWEJ Streszczenie Homogenizacja komputerowa pozwala na zastąpienie materiału niejednorodnego przez ośrodek jednorodny z efektywnymi parametrami materiałowymi. W ten sposób redukowany jest czas obliczeń. Comp. 189-217.. Comput. Methods Appl. M. W.T. Podejście to bazuje na analizie w dwóch skalach – mikro i makro. 575598. Comput. Error indicators and mesh refinements for finite-element-computations of elastoplastic deformations. Brekelmans. Zohdi.. 2012 Accepted: November 9.. Engng. Meth. Meth. Oden. Hierarchical modeling of heterogeneous bodies. L. I. W skali mikro niezbędne jest dokładne odzwierciedlenie mikrostruktury. Mech. P.. Qiu. Comput. 22. 2012. Delft University of Technology. A fully automatic hp-adaptivity. a w obu skalach optymalnej liczby stopni swobody. L. E. 2. powiązanymi z niezależnymi RVE. Następnie dokonywana jest aproksymacja rozwiązania w skali makro. Wriggers. w których homogenizacja nie powinna być stosowana ze względu na zbyt duży błąd. 2012 Received in a revised form: October 31. reprezentatywnym elemencie objętościowym (RVE). Received: September 20. G. równocześnie na tyle duży. F. ISBN 978-83-7242-667-3. W. Cecot. V.. Kouznetsova.. M. M.. Journal of Scientific Computing.

China *Corresponding author: piotr.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. PIOTR BREITKOPF1*.2. This analysis may be performed in order to answer questions such as: how does microstructure affect the structural behavior? What particular microstructure yields the desired performance? How do the inherent material uncertainties propagate at the structural level? Given a set of 2D/3D geometrical instances (snapshots) of the Representative Volume Element (RVE) generated from a priori known information about feasible microstructure topology.fr Abstract 1 The general idea here is to produce a high quality representation of the indicator function of different phases of the material while adequately scaling with the storage requirements for high resolution Digital Material Representation (DMR). material uncertainties 1. This problem. one must first construct models of the microstructure variations to be used as input in the subsequent parameterized analysis. On the other hand. each snapshot pixel/voxel matrix is decomposed into a linear combination of tensor products of 1D basis vectors. UTC. the analysis of the dimensionality of the resulting nonlinear space yields the minimal set of parameters needed in order to represent the microstructure with sufficient precision. Next a common basis is determined for the entire set of microstructure snapshots.breikopf@utc. UMR 7337 UTC-CNRS. the progress in material science allows us to control the material microstructure composition to an unprecedented extent. we propose a three-stage reduction algorithm combining Proper Orthogonal Decomposition (POD) and Proper Generalized Decomposition (PGD). which could be directly usable in multi-scale finite element procedures. homogenization. We showcase this approach by constructing a low-dimensional model of a two-phase composite microstructure. voxel approaches. and multiphase polymers. The material microstructure varies in a much smaller length scale than the actual structural size. BALAJI RAGHAVAN1. No. Typical examples include polycrystalline materials. Finally. requires the right projection space in which the set of snapshots generates ISSN 1641-8581 219 – 225 . INTRODUCTION The constant increase of computing power coupled with ever-easier access to high-performance computing platforms enable the computational investigation of realistic multi-scale problems. a reducedorder parameterized representation is formulated. it becomes essential to include the effects of the microstructure variation when modeling the structural behavior. 2013. Xi’an. To this end. 2 A POD/PGD REDUCTION APPROACH FOR AN EFFICIENT PARAMETERIZATION OF DATA-DRIVEN MATERIAL MICROSTRUCTURE MODELS LIANG XIA1. To perform a multi-scale analysis involving these materials. storage costs. Compiègne. In order to accurately predict the performance of structures employing such new materials.first. WEIHONG ZHANG2 2 Laboratoire Roberval. functionally graded materials. porous media. 13. NPU. similar to those encountered in computer vision image processing and statistical data analysis. Key words: parameterization of microstructure. France Engineering Simulation and Aerospace Computing (ESAC).

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

a low-dimensional manifold that can then be fitted by a parametric hyper-surface. Basically, the parameterized representation of the material microstructure may be split into two major steps: 1. Low-dimensional model construction of various material properties within the RVE, 2. Using this model as an input to the Finite Element analysis at the RVE level. The work introduced in this paper utilizes the available sources of information about the variability of the microstructure to construct a set of possible realizations of the material internal structure and mechanical properties. This initial data may be provided by numerical models (Voronoi diagrams, cellular automata,…) or may be experimentally obtained by using imaging techniques: computer tomography (CT), magnetic resonance imaging (MRI), etc. (Ghosh & Dimiduk, 2011). This set of instances – called snapshots, is mapped into a lowdimensional continuous space that spans all the admissible variations permitted by the experimental data. By exploring this low-dimensional equivalent surrogate space, one is essentially sampling over the parameterized space of material topology variations that satisfy the (simulated) experimental data. This low-dimensional representation is subsequently used as an input model in the finite element analysis either via a mesh or a voxel-based variant of finite element model at microstructural scale. The major advantage of this approach is a significant reduction in complexity due to the analysis in a lowdimensional space and results in a drastic reduction in the critical memory requirement. This representation may then be used in FE2type multi-scale analysis (Feyel, 2003), in continuous optimization procedures (Sigmund & Torquato, 1997) or within a stochastic framework to include the effects of the input uncertainties at the material level in order to understand how they propagate and affect the performance of the structure (Velamur Asokan & Zabaras, 2006). The literature reveals little investigation into developing parameterized material representations. Ganapathysubramanian and Zabaras (2007) used Principal Component Analysis (PCA) to represent data-driven representation of the property variations of random heterogeneous media. A general framework has been proposed by Raghavan et al. (2010) for POD-based shape representation by separating the space variables and the design variables in the space optimization context. In this paper, an exten-

sion of this technique for material microstructures is made by separating the space variables. The overall goal is to linearly scale the storage requirements in order to cope with the ever-increasing resolution of microstructural snapshots. The paper is organized in the following manner: section 2 presents the general description of the overall problem. POD and PGD-POD approaches are introduced in Section 3 and 4, respectively. Section 5 compares the reconstruction errors obtained in using the two approaches based on the approximation on a defined two-phase periodic microstructure media. The paper ends with concluding comments and suggestions for future work. 2. PROBLEM DESCRIPTION Consider a material sample defined by a realvalued continuous N × N or discrete density map s  s  x, y , v defined over a square, periodic domain  = [0,1] × [0,1] and depending on a set of (possibly unknown) parameters (design variables) v p . The problem addressed is how to identify a smallest set of design variables given a set of M snapshots (instances, realizations, samples or images) of the microstructure. The snapshots are given by N × N matrices S k , k  1 M such that

S k  i , j  s x n i , y n  j , v k  , i, j  1 N with xn , y n
defining a regular grid of data points. 1. POD MODEL OF THE MICROSTRUCTURE Consider a set of discrete 2D snapshots Sk, k = 1…M. Each snapshot matrix is stored in a column vector sk of length N2. The full set of snapshots is stored in an N2 × M matrix s1  s  s M  s  , cenM tered around the mean snapshot s   k 1 s k M . The

COMPUTER METHODS IN MATERIALS SCIENCE

interpolation may be performed using standard 2D finite element shape functions   x, y .

  x, y, v k    T  x, y s k s

(1)

The snapshot matrix may be decomposed by Singular Value Decomposition (SVD)

s1  s  s M  s   UDVT

(2)

with the U and V matrices containing respectively the left and right singular vectors. Taking a (reasonable) assumption that M  N 2 , we define a projection basis composed of the first m

– 220 –

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

left singular vectors   1 m   U 1: N ;1: m . An arbitrary centered snapshot is approached by

 k  s   k ,  k   T s k  s  s

(3)

The relative Frobenius norm of reconstruction error of the whole matrix of snapshots is

 k  s k  s k   1 s k  s   kM  T  k 1 s k  s  s k  s  M M 2 i i i m1 i  i  m 1  i F  M M i i i 1 i i 1 i2 F
M T

(4)

(7)

  x, y, v   T  x, y s   i i v  s
i 1

m

The dimension of Ek is now mx × my, decided by the dimensions of the two basis: mx of  and my of  , the process of basis extraction is given afterwards (equation10-13). For an arbitrary point x , y of the snapshot, we can rewrite equation (6)

s  x, y   i  x, y  i v
i 1
T T

m

(5)

  x, y, v k   s  x, y + T  x Ek  T   y  s
s  x, y +   Ei , j v k i  x j  y
i 1 j 1 mx m y

(8)

with s x, y   x, y s and i x, y   x, yi .
The storage requirements for this approach are m  N 2  m  M which may be a problem when the resolution of the grid increases. This is even more critical when extending the approach to 3D with the

where s  x, y   T  x S  y , i  x   T  x  i and

 j  y   T  y j . For arbitrary value of design and

– 221 –

COMPUTER METHODS IN MATERIALS SCIENCE

so for dropped off modes i, i = m + 1,…,M the error is given explicitly by the sum of corresponding diagonal entries i of D, squared. It is thus possible to build a mapping giving for each microstructure instance generated by a set of design variables v p , a unique image   m . There are however two problems to be solved:  an arbitrary   m does not necessarily yield an v p ,  the dimension of p is not a priori known. Both problems are to be addressed here by building a manifold of admissible microstructures. This is done locally for a snapshot by analyzing the local dimensionality of the space spanned by the coefficient vectors  k . The interpolation of coefficients is then formulated as a minimization problem constrained by manifolds of admissible shapes. Combining the above with the spatial interpolation functions, we express s = s(x,y,v) for an arbitrary value of design variables not belonging to the initial set and at any point in  , possibly not on the grid. The bi-level representation allows us to separate the space variables x and y from the design variables v. Assuming that the basis vectors, defined at the grid points, are constant and that only the coefficients  k depend on the design variables, we rewrite equation (1)

storage becoming m  N 3  m  M , which is definitely not scalable as for N = 1024 and M = 100 some 800GB (considering 8 byte floating point numbers) are necessary for the modes only and for N = 4096 and M = 1000, with ~50TB memory which goes well beyond the capacity of current workstations. The second problem concerns the sensitivity of the modes with respect to the order in which matrix terms are rearranged into a vector. Therefore, there is a clear need for an approach not requiring renumbering of the matrix entries and scaling better with increasing resolution. The proposed algorithm is presented in the next section. 2. “PGD-POD” MICROSTRUCTURAL MODEL WITH SEPARATED SPACE VARIABLES In the previous section, an interpolation form separating design variables from the space coordinates is proposed. In this section, a further separation is performed to the individual space dimensions x and y. Given an N  N grid of sampling points and a matrix snapshot of the density map S k  i , j  s xn i , yn  j , v k  , i , j  1 N , the spatial interpolation of s x, y , v k  is

  x, y, v k    T  x S k  y s

(6)

by means of the standard 1D finite element shape functions  x and   y . Instead of rearranging the matrix snapshots Sk into a column vector sk and performing SVD directly on the data set, the idea is to transform each snapshot Sk to matrix Ek of reduced dimensions in terms of two separate basis  and  .

 k  S  E k  T S

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

space variables, the continuous model may now be expressed as

  x, y, v  s    Ei , j vi  x j  y s
i 1 j 1

mx m y

(9)

Such a 1D approximation in each direction thus transforms every N × N snapshot Sk into a mx × my compressed matrix Ek. The process of extraction of the two separate basis  and  is introduced in the following. We start by the truncated SVD decomposition of each individual snapshot

this approach into 3D, the storage requirement would decrease drastically from m × N3 + m × M to mx  m y  mz   N  m  mx  m y  mz  m  M . The reconstruction error of all snapshots in this approach is calculated in a similar way as in equation (4). Note that, two factors, mx and my in the basis extraction and m in SVD reduction of [e1…eM], actually influence the reconstruction error in this approach. 5. PARAMETERIZATION OF A TWO-PHASE RVE In this section, a comparison is given for both proposed approaches on a commonly analyzed periodic two-phase microstructure pattern as shown in figure 1. Snapshots of such a pattern could be utilized to model various types of materials and we have a list of them in table 1. The periodic snapshot is defined by two parameters controlling the radii of two groups of circular inclusions. 500 snapshots of resolution 256×256 are randomly generated for a local approximation.
Table 1. List of possible material types. Material Type Porous Aluminum Reinforced Alloys Fiber Composites Reference Kouznetsova et al. 2001 Ghosh et al. 2001 Zeman and Sejnoha 2001 Nguyen et al. 2010

S k  S  U k Dk VkT

(10)

where only the first mk left and right singular vectors corresponding to the first mk largest singular values are calculated. Next, we arrange all sub-matrices U k 1: N ;1: mk  and Vk 1: N ;1: mk  into two
N  m1    mM  matrices

U*  U1 1: N ;1: m1 U M 1: N ;1: mM  and V*  V1 1: N ;1: m1 VM 1: N ;1: mM 

(11)

and we apply SVD separately to U* and V *
T T U*  UU DU VU and V*  UV DV VV

(12)

The two separate basis  and  are composed of the first mx and my left singular vectors from UU and from UV , respectively.

  1 mx   UU 1: N ;1: mx  and

   1  m y   UV 1: N ;1: m y 

(13)

Quasi-brittle Materials

Thereafter, the matrices Uk and Vk in equation (10) may be now be approximated in terms of the two separate basis  and 

COMPUTER METHODS IN MATERIALS SCIENCE

 k  A k , A k   T U k and V  k  B k , B k   T Vk U (14)

Substitute above into equation (10) and we have equation (7) where matrix Ek  A k Dk BT k is the only term depending on the design variables. Once Ek are obtained, a similar approximation approach is followed to that in section 3. Transforming Ek to column vector ek of length mx × my, SVD reduction is performed on a full set mx × my × M matrix [e1…eM]. Then a new mapping connection is built between each microstructure v p and the coefficients   m calculated by SVD on [e1…eM]. The storage requirements for this approach are mx  m y   N  m  mx  m y  m  M which is significantly less than m × N2 + m × M. When extending

Fig. 1. Two-phase, two-parameter microstructure snapshots.

5.1.

POD approach

SVD is performed directly on the data set. From figure 2, it can be seen that the α’s form a set of twodimensional manifolds rather than a cloud of points in 3D space regardless of the particular triplet of modes used, clearly indicating that the design domain is parameterized by two parameters t1, t2, as is known in priori. This means that α1 = α1(t1, t2), α2 = α2(t1, t2).... The surfaces formed by α’s could be interpreted as the set of all possible “constraints” (direct geometric constraints, technological con-

– 222 –

The microstructure can be duced matrix E. Note that. 99. 2D α-Manifolds for the data set. the RVE considered here is anisotropic. POD is performed on the reduced data set of E’s. now within a reduced assume mx = my and the errors versus mx and my are storage requirement. but an original error of 1% is introduced at the same time (when mx = my = 256). For clearer visualization of the reparameterized again by taking surface coordinates t1 construction error in the basis extraction process. 3. Reconstruction errors versus mx and my in case of square marked curve in figure 3. By setting mx = my = 180. etc. mx doesn’t have to equal my. 4. Therefore. Only the case retaining Fig. The curve with data points marked using circles gives the error when retaining 100% projection energy. are considered and obviously. The curve with the data points marked using – 223 – COMPUTER METHODS IN MATERIALS SCIENCE . The manifolds formed by the β’s (see figure 5) is similar to 5. m1 = m2 = … = mM = 256. Reconstruction errors versus mx  m y in three cases. When 99% projection energy is retained. fests the fact that PGD maintains the interrelationeach snapshot matrix S is firstly transformed to reship among snapshots. PGD-POD approach that in the previous approach. Such a reduction makes the SVD in equation (13) and (14) a little bit more effective as shown in the curve marked with triangles.9% projection energy and the average number of retained modes is about 147.9% projection error is considered for each snapshot SVD. which indicates the microstructure has two parameters and also maniBy extracting basis vectors in both directions. Figure 4 shows the reconstruction error versus mx and my chosen independently..2. new microstructure snapshot could be generated parametrically in a reduced dimension by taking the surface coordinates t1 and t2 as design variables. In the next step. Fig..INFORMATYKA W TECHNOLOGII MATERIAŁÓW straints.e. especially when anisotropic materials Fig. that are difficult to be expressed mathematically). the data set of 256×256 matrices S’s is transformed to reduced 180×180 matrices E’s with an introduced error of 4%. This means that a further reduction in storage requirement may be achieved by choosing mx and my independently. Three curves correspond to cases that different number of modes are retained after the SVD on each snapshot. i. the average number of modes is reduced to 122. squares is the result of retaining 99. we and t2 as design variables. 2. The dimension of the data set is reduced from 2562×500 to 1802×500. given in figure 3.

Methods Appl. Comput. F.J. Eng. Comparison of the reconstructed snapshots. 5. Figure 6 plots the two error curves against the number of retained modes in both approaches. 5. 7.A. 6. error in PGD-POD approach. Lee. N.. O.. Comput... Dimiduk. 326-353. CONCLUSIONS AND PERSPECTIVES A three-stage model reduction scheme combining Proper Orthogonal Decomposition (POD) and Proper Generalized Decomposition (PGD) has been developed to build a reduced order model for the efficient parameterization of material microstructures. 2001.P.. 192. 23352385. With a reduced order model of this type.. the PGD-POD approach could achieve a – 224 – . from left to right: Original Snapshot. Modeling diffusion in random heterogeneous media: Data-driven models. 3233-3244. Sluys. S. 38. W. Springer.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig.T.. A comparison of the reconstruction errors obtained using the two approaches is given in this section. Ghosh. 226. Solids Struct.. Considering an error of 20%. varying in a similar trend except for the red curve converging to zero while the blue curve converges to a value of 4% due to the reduction in the PGD process. Eng.. we retain the first 5 modes for both approaches.. 2010. 2003.3. additional investigations may be conducted into the prediction and optimization of material properties using microstructures. Acknowledgement. Labex MS2T and of the China Scholarship Council. Mech. Fig.M. stochastic collocation and the variational multiscale method. Brekelmans.. 27.. New York. 199. 2011. 2001. A multi-level computational model for multi-scale damage analysis in composite and porous materials. The authors acknowledge the support of OSEO in the scope of the FUI OASIS project F1012003Z... D. Ganapathysubramanian. Computational Methods for Microstructure-Property Relationships. J. is calculated by equation (4). The proposed model maintains the high quality of the reconstructed microstructure snapshots with a significantly reduced storage requirement compared to the traditional POD model. K. On the existence of representativevolumes for softeningquasi-brittlematerials – A failure zone averaging scheme. A multiscale finite element method (FE2) to describe the response of highly non-linear structures using generalized continua. Zabaras. M. 2D β-Manifolds for the reduced data set. Raghavan.. Mech. If the expected reconstruction error is no less than 10%. Ghosh.. The curve marked with squares. Thereafter.P. Baaijens. is calculated similarly with a presupposition of retaining mx = my = 180 in the PGD process. B. The result of the reconstruction is shown in figure 7. Comput. POD reconstructed and PGD-POD reconstructed. V. Comput. The curve marked with triangles. 6. Comparison of the reconstruction errors similar result in a much less storage requirement compared to POD approach. where there is no obvious difference between the two reconstructed snapshots. 30283038. An approach to micro-macro modeling of heterogeneous materials. Lloberas-Valls... Phys. then the numbers of modes needed for the two approaches are close to each other. P. Mech. J.. 37-48. Nguyen. Stroeven. L. 2007. REFERENCES Feyel F. error in POD approach. Int. V. The reconstruction errors of the two approaches. COMPUTER METHODS IN MATERIALS SCIENCE Fig. Methods Appl. Figure 6 shows that the two curves match each other. S. Kouznetsova..

J. 218. macierz obrazu reprezentatywnego elementu objętościowego jest rozłożona na liniową kombinację tensorowych produktów jednowymiarowych wektorów bazowych. Solids. 2012 Received in a revised form: November 8. J. analiza wymiaru otrzymanej rozmaitości topologicznej daje minimalny zestaw parametrów potrzebnych do reprezentowania mikrostruktury z odpowiednią dokładnością. 2012 – 225 – COMPUTER METHODS IN MATERIALS SCIENCE . B. A stochastic variational multiscale method for diffusion in heterogeneous random media. łączący cechy dekompozycji POD i PGD. Eur. PARAMETRYZACJA CYFROWEJ REPREZENTACJI MIKROSTRUKTURY MATERIAŁU POPRZEZ REDUKCJĘ POD/PGD Streszczenie Ogólna idea pracy polega na automatycznej generacji precyzyjnej funkcji reprezentującej topologię i geometrię poszczególnych faz materiału celem uzyskania modelu obliczeniowego o minimalnej liczbie parametrów. 2001.. Phys. Jako przykład podajemy budowę niskowymiarowego modelu dwufazowej mikrostruktury kompozytu.. Torquato. Zabaras.. an a posteriori grid parametrization method for shape optimization. Design of materials with extreme thermal expansion using a three-phase topology optimization method. Numerical evaluation of effective elastic properties of graphite fiber tow impregnated by polymer matrix. Następnie budujemy wspólną bazę dla całego zbioru obrazów mikrostruktury. Received: October 16. 2006.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Raghavan. Mech. O.. Mech.. Breitkopf.. 49.. S. Villon. Velamur Asokan.. N. Zeman.. Sigmund. W tym celu proponujemy trzystopniowy algorytm redukcji obrazu. Phys. 10371067. Solids. 654-676. POD-morphing. 69-90. P. M. Comput.. 1997. 671-697. Sejnoha. 2010. J. W pierwszym etapie. 45. J. Comput. B. W trzecim etapie. 19. Mech. Phys. J. 2012 Accepted: November 23.. P.

g. 226 – 230 ISSN 1641-8581 . 2011). The approach we propose can reduce the computational cost of the analysis without deterioration of the modeling reliability. numerical modeling of heterogeneous materials is a challenging task. Therefore. 2008). No. Before their implementation they are thoroughly tested. INTRODUCTION Most of new materials are composites of different kinds. which accounts for all details of the material heterogeneity. However. Integration of local numerical homogenization with Burgers model should improve modeling of heterogeneous visco-elastic materials. ul. ‘Brute force’ FEM analysis. 2003). binders or asphalt mix). Warszawska 24.. we make use of the local numerical homogenization not based on RVE concept (Jhurani. Local numerical homogenization is one of the computational homogenization methods.pl Abstract The main objective of this paper is to present the prospects of application of local numerical homogenization to visco-elastic problems. 2013. Because of rheological effects a transient analysis is necessary. In such a way one obtains a coarse mesh approximation of the time consuming fine mesh solution. The coarse element stiffness matrix is computed on the basis of the fine element matrices. Its main advantage is that it can be used in the case of modeling of heterogeneous materials with neither distinct scales separation nor periodic microstructure.edu. 31-155 Kraków *Corresponding author: mklimczak@L5. various approaches to evaluation of effective material properties and composite response are proposed (e. In the 1D case the Burgers model is interpreted as a combination of a spring and a dashpot and it is mainly used for bituminous materials (e. In this paper we use the Burgers model to describe inelastic deformations. which is presented briefly below. 2009.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol.. 2003. however any other constitutive equations may be applied. Geers et al. 2009. One of them is the computational homogenization. We present numerical results of 1D and 2D analysis for selected problems that provide comparison between the ‘brute force’ FEM approach and local numerical homogenization in application to modeling of heterogeneous visco-elastic materials in order to validate the technique. It is used to bridge ‘neighboring’ analyses scales by the concept of representative volume element (RVE).g.. Jhurani & Demkowicz. Klimczak & Cecot. proposed by Jhurani in 2009 for linear problems. The main idea of the approach is to replace of a group of many small finite elements by one macro element. However. Mang et al. Burgers model 1. 13. WITOLD CECOT Cracow University of Technology. Key words: local numerical homogenization. Numerical tests can significantly reduce the cost of design process by eliminating some laboratory or ‘in situ’ experiments. 2 LOCAL NUMERICAL HOMOGENIZATION IN MODELING OF HETEROGENEOUS VISCO-ELASTIC MATERIALS MAREK KLIMCZAK*.pk. This approach was developed for example by Geers and his collaborators (Geers et al. especially in the case of inelastic deformations and non-periodic material microstructure. visco-elasticity. is either highly time consuming or even impossible.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

2. LOCAL NUMERICAL HOMOGENIZATION Local numerical homogenization is one of computational homogenization methods. It was proposed by Chetan Jhurani (Jhurani, 2009) for linear problems. Mainly linear elasticity was discussed. Unlike other computational homogenization methods, local numerical homogenization is not based on the concept of RVE. The main advantage of this method is that no separation of scales condition has to be fulfilled. It means that the ratio of microscale characteristic dimension and macroscale characteristic dimension does not have to be much smaller than the unity. Moreover, periodicity of the material is not required. Therefore this method is suitable to model asphalt pavement structures, which is the subject of our interest. Local numerical homogenization is thoroughly presented in Jhurani’s dissertation (Jhurani, 2009). We would like to give an overview of this method and its main steps in context of linear problems.

where K † denotes the Moore – Penrose pseudoinverse of K and u 0 is an arbitrary vector in the null space of K . Analyzing the same problem at macroscale level we use effective stiffness matrices of K  R MxM (M≤N) and coarse scale load vector defined in terms of f as f  AT f ( A  R NxM - a chosen interpolation operator for a respective element). The coarse – scale solution is expressed in the following way:
^
^

u  K f  u0
^

^

^ † ^

^

(3)

where K † denotes the Moore – Penrose pseudoinverse of K , and u 0 is an arbitrary vector in the null space of K . The difference between (2) and (3) is equal to:
^ ^

^

Algorithm of the method consists of the following steps:
 assume ‘trial‘ effective material properties of the analyzed heterogeneous domain,  solve the auxiliary coarse mesh problem (it is advised to use adaptive FEM but it is not obligatory),  refine the coarse mesh within its every element in order to match all the heterogeneities (fine and coarse meshes are naturally compatible then),  find the coarse mesh element effective matrices knowing the fine mesh element matrices,  assemble coarse mesh element effective matrices,  solve the coarse mesh problem. The core of the algorithm is evaluation of effective coarse mesh element matrices. Let us focus on a single coarse mesh element of the analyzed domain. Then we refine the mesh within this coarse mesh element to capture all details of the heterogeneity. , and load vector For a stiffness matrix K  R N f  R , the fine mesh local FEM equation is:
NxN

u  A u  ( K †  A K AT ) f  (u 0  A u 0 )

^

^ †

^

(4)

Thus, we can express , up to a constant, the error

e  R N as:

e  ( K †  A K AT ) f

^ †

(5)

Finally, minimization of the above expression (enhanced with the regularization term) with respect to K † (Jhurani, 2009) leads to the effective coarse mesh element stiffness matrix K . This routine needs to be repeated for every coarse mesh element. Then the coarse mesh problem can be solved in the standard way. 1. BURGERS MODEL Visco-elastic Burgers model is commonly used for modeling of bituminous materials. Its 1D scheme is presented in figure 1. It is a material model, which efficiently simulates all of the most important response characteristics of bituminous materials, i.e. elastic, viscous, and visco-elastic. Additionally, it is may be easily implemented numerically. The total strain increment (  ) in Burgers model is the sum of the elastic
^

Ku  f

(1)

Its solution is equal to :

u  K † f  u0

(2)

– 227 –

COMPUTER METHODS IN MATERIALS SCIENCE

^

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

(  E ), visco-elastic (  VE ) and viscous strain increments (  V ) then:
   E   VE   V

(6)

Fig. 1. Burgers model.

All of the above increments are presented in details by Collop et al. (2003) for both 1D and 3D case. Algorithm for time integration of the Burgers model is as follows:  evaluate elastic solution as a trial one for a time step ti ,  calculate inelastic strain increments,  update load vector considering the ‘impact’ of inelastic strain increments,  solve the problem and calculate total strain increment,  if the difference between updated total strain increment and a trial  is negligible and the difference between updated solution and a trial one is also negligible, go to the next time step; otherwise – go to the first iteration step. 2. INTEGRATION OF LOCAL NUMERICAL HOMOGENIZATION WITH BURGERS MODEL Modeling of heterogeneous visco-elastic materials also requires time-consuming transient analysis. In this chapter we present the prospects of local numerical homogenization in application of viscoelastic Burgers model. Analysis becomes much more complex as we have to ‘homogenize’ at every time step.

Algorithm of the proposed approach for known load history and known constituents characteristics is as follows (for each time step):  solve the elastic problem using local numerical homogenization according to the routine presented in chapter 2,  consider each coarse mesh element to be an independent problem: refine the mesh within this element, assume boundary conditions on the basis of elastic solution and solve this local problem at time ti,  update the coarse mesh load vector considering inelastic contribution,  assemble coarse mesh element matrices and updated load vectors,  solve the coarse mesh problem. Whole routine requires then solving several local visco-elastic problems instead of the global problem. 3. NUMERICAL RESULTS In this chapter we present preliminary results of 1D and 2D numerical tests for visco-elastic materials. In figure 2 analyzed 1D domain is presented. All material data (for 'white' material) are the same as for the test performed by Woldekidan (2011). 'Black' material is characterized by two times weaker parameters. Cross-sectional area is equal to 50 cm2. Analysis period is equal to 60 s. Load P is equal to 1.5 kN for t ≤ 15 s, then it is removed. Results for an arbitrary time step ti are presented in figure 3. The whole domain was discretized by 10 fine mesh elements and 5 coarse mesh elements. Thus, two fine mesh elements were homogenized into one coarse mesh element. Distribution of inclusions is periodic for the sake of simplicity. 2D analysis was carried out for the domain presented below in figure 4. It is a 2m by 4m square analyzed in plane strain state. Its bottom edge is fixed, left and right hand side edges can displace only in the vertical direction. Uniformly distributed tensile load (1 kN/m2) is applied to the upper edge. Material data were assumed in the same manner as for the 1D test. The Poisson ratio both for the inclusion and the matrix is equal to 0.3.

COMPUTER METHODS IN MATERIALS SCIENCE

Fig. 2. Analyzed heterogeneous visco-elastic 1D domain.

– 228 –

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

Fig. 3. 1D example. Displacements at arbitrary time ti.

Fig. 4. Analyzed 2D domain with randomly distributed inclusions.

Vertical displacements of the upper edge are presented in figure 5.

Fig. 5. 2D example. Vertical displacements along the upper edge.

– 229 –

COMPUTER METHODS IN MATERIALS SCIENCE

1. Brekelmans. Cecot. Ostatecznie wystarczy rozwiązać zadanie w obszarze zdyskretyzowanym siatką rzadką.. iż może być wykorzystana do modelowania materiałów niejednorodnych. Hofstetter...edu/research/reports/). C. Eberhardsteiner. 371-386. Technische Universiteit Delft. de Bondt. M . Integracja lokalnej homogenizacji numerycznej z modelem Burgersa może poprawić sposób modelowania niejednorodnych materiałów lepkosprężystych. J h u r a n i . PhD thesis... Received: September 20.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 6. zamiast siatką gęstą. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 230 – . J.. Kouznetsova.. A. ICES Reports 0934÷09-36. E. W. Kasbergen. . Transportation Research Record. A. 108. Proponowane przez nas podejście może ograniczyć czas obliczeń bez pogorszenia wiarygodności modelowania. CONCLUSIONS Summing up we can conclude that for the tests presented in the paper integration of local numerical homogenization with visco-elastic material models:  significantly reduced the computational cost of numerical analysis. Czasopismo Techniczne. REFERENCES Collop. Development and finite element implementation of a stress dependent elasto-visco-plastic constitutive model with damage for asphalt. K . W przypadku jednowymiarowym model Burgersa jest interpretowany jako kombinacja sprężyn i tłumików. Bituminous Mastic and Mortar. The Eleventh East Asia-Pacific Conference on Structural Engineering and Construction. wynikami obliczeń wykonanych za pomocą MES przy pełnym uwzględnieniu mikrostruktury materiału. Klimczak. R. Możliwe jest jednak zastosowanie innego równania konstytutywnego opisującego zachowanie materiału w czasie. . 2008. Lackner. Taipei. Conf.. C. Jhurani w roku 2009 do zagadnień liniowych. W. Hackspiel. W niniejszym artykule wykorzystujemy model Burgersa do opisania deformacji lepkosprężystych.. Proc.A. Multiscale Modeling Using Goaloriented Adaptivity and Numerical Homogenization. że proponowane przez nas podejście może być z powodzeniem wykorzystane do modelowania niejednorodnych materiałów lepkosprężystych. B. Wydawnictwo Politechniki Krakowskiej. The University of Texas at Austin. Multi-scale first-order and second-order computational homogenization of microstructures towards continua. Hellmich..  did not introduce significant additional error to the solution. Pichler.ices. C .. C. 2012 Accepted: November 21. Scarpas. 1-B. 2011. These results encourage us to perform further tests and obtain an effective algorithm for analyses of heterogeneous visco-elastic materials. International Journal for Multiscale Computational Engineering. 2012 Received in a revised form: November 4. (http://www. M. Kluczowym krokiem algorytmu jest zastąpienie grupy elementów siatki gęstej jednym elementem siatki rzadkiej. C .. 87-94. Lokalna homogenizacja numeryczna jest jedną z metod homogenizacji komputerowej. K . 2003. Local homogenization in modeling heterogeneous materials. J h u r a n i . Stürzenbecher. The University of Texas. Prezentujemy wyniki zadań 1D oraz 2D dla wybranych zagadnień. 2009. z. tj. 2003. Wykorzystywany jest głównie do modelowania zachowania materiałów bitumicznych (np. LOKALNA HOMOGENIZACJA NUMERYCZNA W MODELOWANIU NIEJEDNORODNYCH MATERIAŁÓW LEPKOSPRĘŻYSTYCH Streszczenie Głównym celem niniejszego artykułu jest prezentacja możliwości wykorzystania lokalnej homogenizacji numerycznej do zadań lepkosprężystych. Aigner. L . H. F. Ch. D e m k o wi c z . K. Geers. . A. gdyż nie wprowadza znacznego dodatkowego błędu do rozwiązania obniżając jednocześnie koszt wykonywanych obliczeń. Austin.. Rezultaty porównań powyższych metod pokazują.. które nie wykazują wyraźnej rozdzielności skal ani nie charakteryzują się periodycznością mikrostruktury. 2009. M.. Response Modeling of Bitumen. Została zaproponowana przez Ch. 2011. Mang. Jej główną zaletą jest to. Porównane zostały one z wynikami podejścia “brute force”. Scheiner. D & E Drawing and Editing Services Company. lepiszcza asfaltowe lub mieszanki mineralno-asfaltowe).. V. Computational Multiscale Analysis in Civil Engineering. Powoduje to znaczne wydłużenie czasu obliczeń ze względu na konieczność rozwiązania zadania w każdej chwili czasu oraz iteracyjny charakter algorytmu. . R. 96-104. St.. Główną cechą tego podejścia jest wykonanie homogenizacji po dyskretyzacji analizowanego obszaru. T. 1832. Woldekidan. 3 – 14... PhD thesis.utexas. Ze względu na reologię zagadnienia niezbędna jest wykonanie analizy niestacjonarnej. R.

g. large strain. due to damage of the materi- al) the equilibrium equations lose their ellipticity in the post-peak regime.edu.. The reason for the discretization-dependence in the computational tests is that the localization is simulated in the possibly smallest material volume which depends on the assumed mesh. JERZY PAMIN Institute for Computational Civil Engineering. To avoid pathological mesh sensitivity of numerical test results the gradient averaging is applied in the damage model. This leads to an ill-posed boundary value problem that entails a pathological mesh-sensitivity in the numerical solution. Although the softening and the localized deformations are visible in the macroscopic material response. Particularly. The application of standard continuum models to these problems fails to provide an objective description of the phenomena.g. Selected results are presented for the standard and regularized continuum. plasticity. Key words: strain localization.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. plasticity) are included. microcracking or nonassociated plastic flow).g. gradient-enhancement. Cracow University of Technology. Poland *Corresponding author: bwcislo@L5. No. 2 NUMERICAL SIMULATIONS OF STRAIN LOCALIZATION FOR LARGE STRAIN DAMAGE-PLASTICITY MODEL BALBINA WCISŁO*. they have the physical origin in the evolution of the microstructure. 1996). The material description is implemented within the finite element method and numerical simulations are performed for a uniaxial tensile bar benchmark. 13. necking of metallic bars) or material instabilities (e. The localization phenomenon associated with the softening response can be properly reproduced using enhanced continuum theories which have a non-local character and take into account higher deformation gradients in the constitutive description (Peerlings et al. concrete or composites) is strain localization which is closely related to the softening of the material.pk. 2013. The paper includes the description of the material model which incorpoISSN 1641-8581 231 – 237 . The strain localization in the analysed model has a twofold source: geometrical effects (necking) and softening due to damage of the material. 31-155 Cracow. AceGen package 1. but also geometrical nonlinearities (large strains) are taken into account. For the descending stressstrain relationship (e.pl Abstract This paper deals with the phenomenon of strain localization in nonlinear and nonlocal material models. Warszawska 24. damage. INTRODUCTION One of the features of materials with microstructure (e. The gradient averaging involves an internal length scale which is an additional material parameter coming from the microstructure.g. The localization means that from some point the whole deformation concentrates in a narrow zone while a major part of the structure experiences unloading. The parameter is usually associated with the width of the localization band and is determined for instance by an average grain size. The strain localization has a twofold source: geometrical effects (e. in the description of the material not only nonlinear constitutive relations (damage.

In the following numerical simulations the exponential model adopted from Mazars and Pijaudier-Cabot (1989) is applied:  ( )  1  0 1     exp(  (   0 )   (5) where  and  are model parameters. The application of the automatic code generator AceGen significantly simplifies implementation of the elaborated models and due to automatic computation of derivatives allows one to avoid an explicit derivation of the tangent matrix for the Newton-Raphson procedure. The Kirchhoff stress tensor  is related to the elastic left Cauchy-Green tensor be with the formula: 1  Nbe  Lv be   2 (8) – 232 – . The damage condition takes the form: ~ . The plastic sor which is computed as:  regime is defined through the yield function Fp which is an isotropic function of the effective Kirchˆ and the plastic multiplier : hoff stress tensor  ˆ .  ).1 is a deformation measure which governs damage and 0 is a damage threshold. a scalar measure of plastic flow  and a scalar damage parameter  : 2  e e b be (3) Damage is understood in the described model as a degradation of the elastic free energy function in the form:  e . ) = f(  ˆ)– Fp(  COMPUTER METHODS IN MATERIALS SCIENCE 2 / 3 (y .INFORMATYKA W TECHNOLOGII MATERIAŁÓW rates hyperelasticity. The free energy function in the presented model is assumed to be an isotropic function of the elastic left CauchyGreen tensor b e  F e F eT . 2009). The simulations are performed using Mathematica–based package AceGen/AceFEM (Korelc. and which is computed from the damage growth function. The associative flow rule is assumed in the form: e   1 1   1/3 e b )  3 ( J be  1)  ln( J be )    tr( J be  2 2 2  2 (2) where Jbe is the determinant of the elastic left Cauchy-Green tensor and  and  are material parameters. The function q represents the isotropic 2 linear hardening as: q( ) = -h. where  ~ = variable calculated as max(  0 det(F) . damage is not directly coupled with plasticity thus depending on the assumed material parameters the model can also reproduce hyperelasticity-plasticity or hyperelasticity-damage. The implicit gradient model is incorporated.q( ))  0 (7)   (1   ) e (b e )   p ( ) (1) ˆ )is assumed to be the HuberThe function f(  The constitutive relations of hyperelasticity are expressed through the elastic part of the free energy function which is assumed in the following form (Simo and Hughes. The analysis is performed with the assumption of large deformations and isothermal conditions. 1998): Mises-Hencky equivalent stress f  2 J 2 . where h is a hardening modulus.  is a history ~ . ) =  ~ -0 Fd (  (6) For Fd < 0 there is no growth of damage. 2. The plastic part of the presented model is described in the effective stress space which means that it governs the behaviour of the undamaged skeleton of the material. which depends on the second invariant of the deviatoric part of the effective Kirchhoff stress tensor t : 1 J 2  t 2 . plasticity (with or without hardening) based on Auricchio and Taylor (1999) and gradient enhanced damage.d  (1   ) e (4) where  is a scalar damage variable which grows from zero for the intact material to one for a complete material destruction. Thus the following formulation takes into account the effective Kirchoff stress tenˆ = /(1 . However. 1998). which is reflected in an additional partial differential equation to be solved. The paper presents the results of a computational test of localization in a tensile bar. The material description is developed with the assumption of isotropy and isothermal conditions and is based on a classical multiplicative split of the deformation gradient into its elastic and plastic parts: F  FeF p (Simo and Hughes. plasticity and damage and takes into account large strains.). SHORT PRESENTATION OF MATERIAL MODEL The following simulations are performed for a material model which involves hyperelasticity.

The application of the gradient averaging to the material model including large strains is additionally difficult due to the distinction of the undeformed and deformed configuration. Firstly. 1999) or the plastic strain measure (Żebro et al. the material averaging is chosen for the following simulations. v = 0.INFORMATYKA W TECHNOLOGII MATERIAŁÓW where Lv is the Lie derivative of b e (Bonet and Wood. In the described model. The results of the computational test performed for two FE discretizations with linear interpolation 20x2x2 and 40x4x4 are depicted in the first graph of figure 2. see e..3. The simulations of the material model including hyperelasticity coupled with local damage are performed for a tensile bar with imperfection presented in figure 1. In the literature different variables to be averaged are taken into account. 2009). In this paper a gradient regularization is applied. 2008). The simulation confirms that for the gradient model the behaviour of the sample does not depend on the adopted mesh. Geometry and boundary conditions of a bar with imperfection. Finally. .l22  ~ = ~  tion due to geometrical softening in plasticity are discussed. Thus. the complex model of hyperelasticity-plasticity coupled with gradient damage is considered. The enforced displacement and the boundary conditions preserve the uniaxial stress state. In the second subsection the results for strain localiza- Fig. the stored energy function (Steinmann. The internal length parameter is assumed to be l = 3 mm. 0 = 0. especially for the medium and the fine mesh.  = 1. Areias et al.99. 2008) and N is a normal to the yield hypersurface. All model variants have been implemented in the Mathematica-based packages AceGen and AceFEM (Korelc.011. The next test is performed for the same sample but the hyperelastic model is coupled with gradient damage. Based on the results obtained by Steinmann (1999) and Wcisło et al.g. NUMERICAL SIMULATIONS OF STRAIN LOCALIZATION In this section the numerical examples of strain localization in hyperelasticplastic-damage model are presented. (2012) which show that the spatial averaging does not fully preserve the numerical results from the dependence on the discretization. e. In the central zone the damage threshold is assumed to be 0 = 0. The former is an automatic code generator used for the preparation of finite element code whereas the latter is a FEM engine. The computations for the finest mesh 80x8x8 fail for the displacement control (snap-back occurs). Hyperelasticity coupled with damage (9) 4. 3. the results for material softening due to damage are presented for standard and regularized continuum. the local strain measure governing ~ is replaced with its non-local counterpart damage  ~ which is specified by the averaging equation:  ~ . GRADIENT-TYPE REGULARIZATION A variety of approaches can be applied to preserve numerical results from the pathological meshsensitivity observed for a standard continuum model reproducing the behaviour of materials exhibiting softening.g. The material parameters applied in the simulations are: E = 200 GPa. The deformed mesh with the damage variable distribution is presented in figure 4.  = 0. the averaging equation and the internal scale l can be specified in the initial or current configuration.01. The parameter l appearing in equation (9) is a materialdependent length parameter called the internal length scale. 1. which is not only a computationally convenient approach but it is also motivated by micro-defect interactions.1. It can be noticed that diagrams are close to one another. – 233 – COMPUTER METHODS IN MATERIALS SCIENCE with homogeneous natural boundary conditions. 4. The width of the damage zone which is related to the internal length parameter is similar for each discretization. We can observe in figures 2 and 3 that the results significantly differ for each discretization and that the zone of strain localization covers only the middle rows of elements. The introduction of the gradient enhancement into the material description requires the choice of a nonlocal parameter and the formulation of a corresponding averaging equation. (2003). The reactions sum diagram is presented in the second graph of figure 2.

can be observed that in the plastic regime the dia- COMPUTER METHODS IN MATERIALS SCIENCE Fig. For each discretization the number and the arrangement of the strain localization zones is different. Deformed mesh and distribution of damage variable  for three discretizations (hyperelastic–gradient–damage model).INFORMATYKA W TECHNOLOGII MATERIAŁÓW a) b) Fig.2. – 234 – . The dimensions and the boundary conditions of the sample are the same as in the previous section and three meshes are taken into account: 20x2x2. 3. 2008). It gram is descending although ideal plasticity is assumed. Hyperelasticity coupled with plasticity Firstly. 2. σy = 300 MPa (perfect plasticity). 4. Evolution of Green strain Exx for meshes 20x2x2 and 40x4x4 (hyperelastic–damage model). the test is performed for an ideal bar with a constant square cross-section along the length. It is caused by taking into account the change of a cross-section during deformation.bar approach is incorporated to avoid locking (de Souza Neto et al. v = 0. Fig. 40x4x4 and 80x8x8 elements. whereas for the medium and the fine mesh the necking occurs at the same time. Sum of reactions vs. The material parameters applied in the test are as follows: E = 200 GPa.. For all discretizations the loss of stability can be observed: for the coarse mesh the phenomenon is observed the earliest. Figures 6 and 7 present the deformed sample with the final accumulated plastic strain distribution and the evolution of the Green strain Exx along the bar length respectively. 4. It can be noticed that the loss of stability manifests itself in multiple necking. displacement for a) hyperelasticity– damage model and b) hyperelasticity–gradient–damage model. The Huber-MisesHencky yield criterion is applied and the F .3. The graph of the reactions sum vs end displacement is depicted in the first diagram of figure 5.

The reaction diagrams are close for each discretization and present the plasticity regime with hardening and reduction of the reaction forces due to damage.5%E. that the same results are obtained for hyperelasticity-plasticity with hardening where the hardening modulus has sufficiently small value. h = 1%E. even in the absence of damage. the material softening is reproduced properly due to gradient regularization and geometrical softening does not occur because of a sufficiently large value of the hardening modulus. v = 0. The last test is performed for the material model which includes both plasticity and gradient damage. for example 0. Fig. In the analysed test. a) Fig. Hyperelasticity-plasticity coupled with gradient damage Fig. the less stiff the model is.0002.99. Figures 9 and 10 present selected results of the simulation. The imperfection is prescribed as the reduction of the yield stress to the value σy = 290 MPa. Final accumulated plastic strain for three meshes and ideal bar (hyperelastic–plastic model). σy = 300 MPa. We can observe that the results depend on the adopted finite element mesh and the finer the discretization is.  = 1. 0 = 0. 5. b) 4. – 235 – COMPUTER METHODS IN MATERIALS SCIENCE It should be also mentioned.  = 0. Evolution of Green strain Exx along the bar for three meshes and ideal bar (hyperelastic–plastic model). The sum of reactions diagram is presented in the second graph of figure 5. regularization is necessary for large deformations and ideal plasticity or small hardening. 6. The imperfection is as in the previous subsection.3.3. It seems that.INFORMATYKA W TECHNOLOGII MATERIAŁÓW The test for the same sample but with the assumed imperfection in the middle of the bar is also performed. The tested sample is the bar with imperfection with the following material parameters: E = 200 GPa. . In figure 8 it can be noticed that the strains localize in the middle part of the sample where the imperfection is assumed. displacement for a) the ideal bar and for b) the bar with imperfection – hyperelastic–plastic model. Sum of reactions vs. 7.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. 1999. 2003. Acknowledgments. The hyperelastic-damage – 236 – . The simulations are performed for different variants of the model and exhibit geometrical or material softening. financed by the Ministry of Science and Higher Education. The research has been carried out within contract L-5/66/DS/2012 of Cracow University of Technology. L. Cambridge. Kowalczyk-Gajewska from IFTR PAS. R... For a sample with imperfection one zone of large strains can be predicted in contrast to the ideal sample where multiple necks are formed. Warsaw. allows one to properly reproduce the material behaviour. 15. The considered model is briefly described and selected numerical results are presented. incorporating an internal length parameter.. D. model exhibits material softening which can cause mesh-sensitivity observed in the presented simulation results. J. displacement for hyperelasticity– plasticity–gradient–damage model. Cambridge University Press. C. A gradient model for finite strain elastoplasticity coupled with damage. A return-map algorithm for general associative isotropic elasto-plastic materials in large deformation regimes. Final accumulated plastic strain for imperfect bar and three discretizations (hyperelastic–plastic model). Deformed mesh and damage variable  distribution for discretizations 20x2x2 and 40x4x4 (hyperelasticity– plasticity–gradient–damage model). F. 8.. Sum of reactions vs. Auricchio. 1191-1235. 39. Cesar de Sa. CONCLUSIONS In the paper the problem of strain localization for a material model including geometrical and material nonlinearities with the applied gradient regularization has been outlined.. Plasticity.. Poland. R. 2008. 5. Nonlinear continuum mechanics for finite element analysis. The gradient averaging procedure. 10. J. the application of the regularization of the plastic part of the model should be considered in the future work. and Conceicao. Fig. Int. To prevent the numerical results from a pathological mesh-dependence. P. 9.. The authors acknowledge fruitful discussions on the research with Dr K. The numerical tests reveal that for a model incorporating ideal plasticity in large strain regime the strain localization might occur. Bonet. 13591378. COMPUTER METHODS IN MATERIALS SCIENCE Fig. REFERENCES Areias. Moreover. Wood. Taylor. J. Finite Elements in Analysis and Design. the work is planned to be extended towards thermo-mechanical coupling.

J. Aby uniknąć patologicznej zależności wyników testów numerycznych od dyskretyzacji należy zastosować odpowiednią regularyzację. K. Peerlings. W. – 237 – COMPUTER METHODS IN MATERIALS SCIENCE . 44.INFORMATYKA W TECHNOLOGII MATERIAŁÓW de Souza Neto.application to concrete. Mazars. de Borst. 631-649. 3391-3403. 2009. SpringerVerlag. J.. T. R. J. 757-779.. Technical Transactions. ASCE J. Wcisło. R. W niniejszej pracy zastosowano uśrednianie gradientowe.. P. 46. W artykule przedstawiono zwięźle opis analizowanego modelu sprężysto-plastycznego sprzężonego z uszkodzeniem przy dużych odkształceniach oraz zastosowanej regularyzacji gradientowej. 7.. 2008... B. Engng. Meth. Gradient-enhanced damage for quasi-brittle materials. Korelc. Kowalczyk-Gajewska. G. który może określać szerokość strefy lokalizacji odkształceń. K.... w których można zaobserwować lokalizację odkształceń zarówno związaną z osłabieniem materiału jak i osłabieniem geometrycznym. T. Computational Inelasticity. ale również nieliniowości geometryczne (duże odkształcenia). Arch. Simo.. W takim przypadku odkształcenia lokalizują się w najmniejszej możliwej objętości materiału. 345-365. gdy zależność między naprężeniami a odkształceniami wchodzi na ścieżkę opadającą.. Int. C. 2012 NUMERYCZNE SYMULACJE LOKALIZACJI ODKSZTAŁCEŃ DLA MODELU USZKODZENIA SPRZĘŻONEGO Z PLASTYCZNOŚCIĄ W DUŻYCH ODKSZTAŁCENIACH Streszczenie Artykuł dotyczy zjawiska lokalizacji odkształceń w nieliniowych i nielokalnych modelach materiałowych. Lokalizacja odkształceń w analizowanym modelu ma dwojakie źródło: efekty geometryczne (szyjkowanie) oraz osłabienie spowodowane uszkodzeniem materiału. która w symulacji numerycznej określona jest przez rozmiar elementu skończonego. J. Kowalczyk-Gajewska. New York. de Vree. Automation of primal and sensitivity analysis of transient coupled problems. 2012. Chichester.. w którym istotną rolę odgrywa wewnętrzna skala długości. Int.. Continuum damage theory . plastyczność). 2012 Accepted: October 26. Computational Mechanics. R. 1989. Numer. Jest to dodatkowy parametr materiału związany z jego mikrostrukturą. E. przedstawiony opis materiału zawiera nie tylko nieliniowe związki konstytutywne (uszkodzenie. Mech.. Owen.. Hughes. Pamin. Pamin. Pijaudier-Cabot. 1996. Computational methods for plasticity. J. John Wiley & Sons. 1998. Formulation and computation of geometrically non-linear gradient damage. Interdisciplinary Applied Mathematics Vol. 2012 Received in a revised form: October 18. Received: October 16. Peric. 1999. 2008. (to be published). W pracy zaprezentowane są wybrane wyniki symulacji rozciągania pręta dla różnych wariantów przyjętego opisu materiału... J. Steinmann. Ltd. D. Eng. Theory and applications. A geometrically nonlinear model of scalar damage coupled to plasticity. Brekelmans. Engng. Meth.. J. 115. J. Mech. D. W szczególności. 20/3-Ś. Model ten został oprogramowany w pakiecie AceGen w środowisku Mathematica oraz przetestowany przy użyciu pakietu AceFEM. Zastosowanie standardowych modeli continuum nie prowadzi do poprawnej symulacji zachowania materiałów z osłabieniem. Gradient-enhanced model for large deformations of damaging elastic-plastic materials. Numer.. UK. J. 251-262. Żebro. Spowodowane jest to utratą eliptyczności równań równowagi.. 39.

. The first of them is according to the results of simulation allowed the appearance of microscopic cracks.pl Abstract The problem of determination the drawing schedule of the cold drawing of thin (less than 0. The reasons of this fact are explained in the (Milenin et al. These microcracks considerably make worse the restoration of plasticity using annealing. identification of the fracture model and its implementation into the FEM model of wire drawing. Solution of the problem is proposed in the works (Milenin et al. The simulation of this state in the wire drawing process and development on this basis regimes of wire drawing is the purpose of paper. As shown in the previous works (Kustra et al. Mickiewicza 30. Solution of problem required the development of the fracture model of alloy in the micro scale. Experimental verification is executed in laboratory conditions on the specially developed device. 2013. multi-scale modeling.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. 2011. it had high plastic characteristics and allowed further wire drawing. these alloys contain lithium and calcium supplements. The second regime made it possible to carry out 7 passages without the fracture..075 mm did not contain surface defects. 1999. 2010a). magnesium alloys 1. is proposed to consider it as optimum from the point of view of the restoration of plasticity with the aid of the annealing. It is experimentally proven that the microscopic cracks during the tension tests occurs long before the complete fracture of samples. 2001. The production of thin surgical threads to stitching tissues may be an example of application of these alloys (Seitz et al. Two schedules of wire drawing are examined. a technological ductility of these alloys during cyclic processes based on a combination of a cold deformation and annealing is significantly lower than for most known magnesium alloys. It is ISSN 1641-8581 238 – 244 . after which the fracture of wire occurred. No. Thomann et al. Milenin & Kustra. Feature of these alloys is a low technological ductility during cold forming.1 mm. The annealing was carried out before each passage. 2 THE MULTI-SCALE NUMERICAL AND EXPERIMENTAL ANALYSIS OF COLD WIRE DRAWING FOR HARDLY DEFORMABLE BIOCOMPATIBLE MAGNESIUM ALLOY ANDRZEJ MILENIN*. PIOTR KUSTRA. Poland *Corresponding author: milenin@agh. 2011) that there are fractures on the grain boundaries long before the fracture of the sample in the macro-scale in these alloys during cold deformation. Thus. 30-059 Kraków.. DOROTA J. INTRODUCTION This paper is devoted to the new magnesium alloys used in medicine as a soluble implants (Heublein et al. al. BYRSKA-WÓJCIK AGH University of Science and Technology. 2009). The cracks on the grain boundaries were observed in this case on the surface of wire. The special feature of the alloy Ax30 is the mechanism of fracture on the grain boundaries.edu. 2009. the obtained wire with a diameter of 0.. 2010b). which directly precedes the appearance of these microscopic cracks. Haferkamp et al. 13. The second regime was designed so that the microscopic cracks would not appear during wire drawing. In the first regime it was possible to realize only 2-3 passages.. The initial diameter of billet was 0.1 mm) wire from the hardly deformable magnesium alloy Ax30 with the aid of the multi-scale mathematical model is examined in the paper.. Milenin et al. 2010). The state of metal. the validation of the developed multi-scale model is executed for two principally different conditions of deformation. Key words: drawing process. 2010b.. Typically. Milenin et al.

An example of cracks for alloy Ax30 is shown in figure 1 in macro-scale (figure 1a) and micro-scale (figure 1b). The figures show that in these alloys microcracks appear much earlier than in the typical magnesium alloy Az80. the effectiveness of annealing is much reduced and reaching of large deformation in a multi-pass process is impossible. The multi-scale model of wire drawing process was proposed to solve this problem in the works (Milenin et al. The porosity values in the stretching sample characterize the technological plasticity during multi-pass drawing. the annealing allows restoring the plasticity (Milenin et al. the increase of porosity in the early stage of deformation is a fundamental difference between considered alloys and known magnesium alloys.8% Ca 99. used in mechanical engineering. The experiment is described in detail in the works (Milenin et al. 96. 3. Another disadvantage is that the biocompatible lubricant cannot be used what becomes to be important in medical application. b) in micro-scale Fig. which allows to easily simulate of the fracture of grain boundaries. The technique of research of the fracture mechanism is based on stretching sample in microscope’s vacuum chamber. The test shown that these alloys crack mainly on grain boundaries.. A porosity in sample appears long before the moment of fracture in macro-scale. During the process of stretching changes of microstructure and microcracks nucleation are monitoring.0%Al. 2. In figure 2 the values of porosity in a center of sample during tensile test of MgCa0. It is proved that if the microcracks do not appear in a current pass. are also shown for comparison purpose.. Obtainment a thin wire is difficult because of the strong sensitivity to the velocity of drawing process. 2010a.1 mm) wires by using a multi-scale modeling of wire drawing process and experimental verification of the results.2% Mg) were selected as a material for the study.8 (0. It follows from this that the development of the drawing technology should be made in such a way that in an every pass the material does not have microcracks. 2011). Thus the solution of listed problems requires in-depth study of cold drawing process for these alloys.. Otherwise. 2. Milenin et al. MECHANISM OF FRACTURE The MgCa0. Studies show that this method is effective for more than 0. The values for Az80 alloy. 1.8 and Ax30 alloys are shown. 2011). 2010a.1 mm wire diameter. The examples of microcracks during tensile test of Ax30 alloy: a) on the surface of the sample in macro-scale.. For micro-scale modeling of the fracture processes the boundary element method (BEM) was used. Milenin et al.INFORMATYKA W TECHNOLOGII MATERIAŁÓW based on drawing by a hot die.. 2011). The porosity dependence on a total sample elongation – 239 – COMPUTER METHODS IN MATERIALS SCIENCE .8%Ca. Thus. a) b) Fig. The aim of this work is to determine the parameters of the cold drawing of thin (less than 0.2%Mg) alloy and its modification Ax30 (0.

 – yield stress of material in grain.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 3. which refers to change of elastic-plastic properties due to the various orientations of grains. 1969). E – Young modulus. The proposed criteria of crack initiation are based on the theory by L. The macro-scale and micro-scale models are coupled in a such way that the results of simulation on macro-scale. Milenin et al. 2002) for modeling of grain boundary cracking in the case of the deformation of the polycrystals. but the most important parameter in micro-scale simulation is the damage parameter D. 2010a. (4) 2 2 . M.  ij – the increment of strain components. The digital representation of the microstructure in a micro-scale model in proposed BEM code is considered as a two-dimensional representative volume element (RVE) which is divided into grains (figure 3).. The crystallographic orientation is included in the developed program by a random parameter k. are the boundary conditions at the micro-scale. N. 1983) for the two-dimensional tasks and incompressible material. The solution of boundary problem is based on the Kelvin’s fundamental solution (Crouch & Starfield.. σn – tensile (positive) component of normal stress at the boundary between two grains. 3.. The microscale model of deformation is based on boundary element method. The Saint-Venant-Levy-Mises theory is used for relation between stresses and increments of strains for plastic deformation:  ij   ij 0  2  ij 3 (2) where  ij – the Kronecker delta. Photo of microstructure (a) and BE mesh (b) where: D – damage parameter. The solution of boundary problem and fracture criteria are described in detail in previous works (Milenin et al. This theory was successfully used in (Diard et al. The effective plastic modulus of the material for each grain is calculated as follows: E eff  k   (1) where: k is the random parameter. Rabotnov (Rabotnov. According to the equation (3)-(5). 2011). The model at the micro-scale includes the BE mesh generation based on images of a fragment of real microstructure and numerical solution at the micro-scale level. At the micro-scale the displacements. the damage parameter is computed at micro-scale for all boundary elements and depends on the material and stress – 240 – . Kaczanov and Y. which is explained below.   – increment of mean equivalent strain in grain. especially stress and displacements. This model was modified to describe the crack initiation at the grain boundary: COMPUTER METHODS IN MATERIALS SCIENCE a)  d  1 . D  D 0  (3)  eq   b D 1   E    b2 1  D b 3 . 2005).  eq   n  b0 S (5) b) Fig. b0-b3 – empirical coefficients. strains and stresses on grains boundary are computed. σS – shear stress at the boundary between two grains.  0 – the mean stress. THE MULTI-SCALE MATHEMATICAL MODEL OF A DRAWING PROCESS The macro-scale numerical model of drawing process is based on finite element method (FEM) and described in paper (Milenin.

a) b) Fig. in the variant 1 the cracks on the grains boundaries did not appear. the following coefficients of equations (4) and (5) for alloy Ax30 are received: b0 = 0. As a result of experimental data processing. However. that stresses and strains in variants 1 and 2 are significantly different.20).1230.09120. b3 = -0. All passages in each variants was geometrically similar. in spite of a possible fulfillment of condition. Variant 2: 0. this refers to alloys without high propensity to microcracks in the early stages of deformation. which are shown in figure 2.43. This suggests that in this case the ductility restoration for alloy after pass will not be possible and the number of passes before the fracture of the wire will be less then in variant 1. in variant 2 there is the emergence of microcracks (figure 6b).0955 0. These parameters are used as boundary conditions for the microscale simulation of microstructure deformation. Distribution of triaxility factor: a – for variant 1. this may be the proof of theoretical conclusions about the major impact of micro cracks on technological plasticity. Thus. which was installed before the device for drawing.0870. The drawing speed was 10 mm/s and was chosen in such a way that the annealing could be done in a furnace.1350. 4. When the value of parameter D reaches the value 1 for boundary element. simulation only first passage for each variant was performed. The maximum value of the parameter D is reached for passage amounted to 0. The outer boundaries of the domains were assigned to boundary conditions and. when the experimental verification of numerical simulation finds that preferable to variant 1. 2004). The determination of empirical parameter of fracture model at the micro-scale is based on inverse analysis of experimental data. b1 = 0.07940.02.0758 (elongation per pass 1. Diameters of wires in variant 1: 0. The present data shown. 2012). The crack initiation is allowed only for the internal boundaries in the developed model. For this reason.89.08310.112 (elongation per pass 1.096). The purpose of this analysis is to minimize the difference between the empirical and calculated moment of crack initiation and the empirical and calculated porosity in microscale of sample during deformation (Milenin et al. they cannot be destroyed.50. However. so results of stress and strain for all calculated passages are close.1470. the fracture criteria is met. The value of parameter D varies from 0 to 1..10. b2 = 0. From the point of view of experience in a drawing of magnesium alloys and based on the results of the simulation in macro-scale the variant 2 is preferred because in this case deformation is more homogeneous and value of tensile stresses is lower (Yoshida.30. As can be seen from the results (figure 6). Results of simulation in micro-scale are shown in figure 6. .INFORMATYKA W TECHNOLOGII MATERIAŁÓW state. b – for variant 2 – 241 – COMPUTER METHODS IN MATERIALS SCIENCE In figure 5 the distribution of strain in the drawing direction and vertical stresses along the centre line of the deformation zone are shown. In figure 4 the results of simulation (triaxility factor) of the first pass for variant 1 (figure 4a) and variant 2 (figure 4b) are shown. 4. Angle of die in each pass was 50. THE MULTI-SCALE MODELING OF TWO VARIANTS OF DRAWING PROCESS For the purpose of proposed technique validation two variants of wire drawing process were simulated.1620.

Further attempts of annealing and drawing were unsuccessful. b) and variant 2 (c. Study of mechanical properties – 242 – . The hairline fractures on the grains boundaries after passage 2 on the surface of the wire can be observed using an optical microscope. d) Fig. wire diameter is 0. variant 2 Much higher wire quality (figure 8) and mechanical properties which allow further drawing were achieve in variant 1. 5. The surface of the workpiece does not contain defects observed on the optical microscope. The received wires were fragile and crumble and after 2 pass tie the knot is impossible. Network of cracks after 4 pass. 2010). 6. The distribution of strain in the direction of drawing (b. d) and vertical stresses along the centre line of the deformation zone (a. In the variant 2 only 4 passage was perform. Results of simulation (effective strain in grains) in microscale for variant 1 (a) and for variant 2 (b) c) COMPUTER METHODS IN MATERIALS SCIENCE the billet by hot wire drawing process is presented in the work (Milenin & Kustra. The methodology of receiving Fig. The Ax30 alloy was used. Developed network of cracks after 4 pass is shown in figure 7. 7.INFORMATYKA W TECHNOLOGII MATERIAŁÓW a) a) b) b) Fig.112 mm. d) 5. As a lubricant sunflower oil was proposed and the temperature of drawing was 30o С. c) for variant 1 (a. THE EXPERIMENTAL VALIDATION OF RESULTS The experimental validation of the results of calculations was performed in the context described above.

Rohde. Advanced Engineering Materials. M. Werkstofftech... Starfield. S. 207-214. von der Hoh.. N. Byrska.8 alloy. Kustra. Kaese. Byrska. 110. 11. J. spec. Mathematical Model of Warm Drawing Process of Magnesium Alloys in Heated Dies. Kustra. Milenin. 11. 2004. Annual Symposium "Transcatheter Cardiovascular Therapeutics". G. REFERENCES Crouch. The Manufacture of Resorbable Suture Material from Magnesium – Drawing and Stranding of Thin Wires. Meyer-Lindenberg. Milenin. Mat. Computational Materials Science. P. Milenin. J. CONCLUSIONS 1. A. C... Wulf.W. eds.Rm = 252. 116-120. 2005. J. 73-84. 2010 b. The multi-scale physical and numerical modeling of fracture phenomena in the MgCa0. O. A. 1983. GEORGE ALLEN & UNWIN London. Pietrzyk.0758 mm . D.. 32. Production of thin wires of magnesium alloys for surgical applications. u. 2001. H. Computer Methods in Materials Science..0758 mm 6. Phan-Tan. Phillip. D. Grydin. 1251–1254. New York. 863866... 2012. R.. Sydney. 2. L.. B. Schaper. A.. Schaper. 81. który bezpośrednio poprzedza pojawienie się mikropęknięć. D. Multiscale modeling and interpretation of tensile test of magnesium alloy in microchamber for the SEM... 199-202.. O.. Comparison of the resorbable magnesium alloys LAE442 and MgCa0.. 1999. V. K.INFORMATYKA W TECHNOLOGII MATERIAŁÓW in INSTRON machine showed that the tensile strength Rm of wire for all passages is not significantly different (diameter 0.2002. Niemeyer. D. Cechą charakterystyczną stopu Ax30 jest mechanizm pękania po granicach ziaren. Rabotnov. M. Kaese. 40.150 is gratefully acknowledged.. diameter 0. Rousselier... Proc.. zgodnie z wynikami obliczeń. 89. Cailletaud.. Heublein. M. A. 2009. Ch.. A. O... 72. Thomann. W. Rohde. N. K. Yoshida.Rm = 250. 18. G. Heublein. The physical and numerical modeling of intergranular fracture in the Mg-Ca alloys during cold plastic deformation.. Klose. Milenin. 1087-1095. Bach. jest uznany za optymalny pod względem możliwości odzyskania plastyczności za pomocą wyżarzania. E. O. Fr. Cold drawing of magnesium alloy wire and fabrication of microscrews. Pierwszy z nich. Steel Research International. 1038-1049.. B. 100-104 (in Polish). Based on the developed schema of drawing the wire diameter 0... wire diameter 0.. D. Werkstofftech. Bormann. Surface of wire after drawing according variant 1. Program komputerowy Drawing2d – narzędzie do analizy procesów technologicznych ciągnienia wielostopniowego. Wire & Cable Technical Symposium. G.. J. Murray. Hausdorf. 14th international conference on Metal Forming. Kustra. Byrska-Wójcik. 13..... C. Milwaukee. Acknowledgements.. Milenin. V.. O. 2010. Dwa przypadki procesu ciągnienia zostały zbadane. Głównymi celami pracy są symulacja takiego stanu materiału oraz opracowanie procesu ciągnienia na tej podstawie.8 concerning their mechanical properties.. Bormann. Distribution of normal stress at grain boundaries in multicrystals: application to an intergranular damage modeling.. 2011. Udowodniono eksperymentalnie. their progress of degradation and the bone-implant-contact after 12 months implantation duration in a rabbit model. Bach. 8. A. Leclercq. Creep Problems in Structural Members. Haferkamp. Kraków.11.2011.. Seitz. A. A.1969. Grydin. Boston. Boundary element methods in solid mechanics.. Exploration of Magnesium Alloys as New Material for Implantation. A. 2.... Mat. Utermohlen.. 82-87. prowadzi do powstania mikro- – 243 – COMPUTER METHODS IN MATERIALS SCIENCE . The experimental research and the numerical modeling of the fracture phenomena in micro scale.. Hartung. Krause.. Degradation of Magnesium Alloys: A New Principle in Cardiovascular Implant Technology. Rozwiązanie przedstawionego problemu wymaga opracowania modelu pękania stopu w skali mikro. WIELOSKALOWE NUMERYCZNE MODELOWANIE ORAZ ANALIZA EKSPERYMENTALNA PROCESU CIĄGNIENIA NA ZIMNO TRUDNO ODKSZTAŁCALNYCH BIOZGODNYCH STOPÓW MAGNEZU Streszczenie Praca poświęcona jest opracowaniu procesu ciągnienia na zimno cienkich (o średnicy mniejszej niż 0. A.. Wire Ass. 2009.. Seitz.. S.0758 mm for Ax30 alloy by cold drawing could be reached. Milenin. M. 2. M. Hutnik-Wiadomości Hutnicze.. że mikropęknięcia w trakcie próby rozciągania pojawiają się na długo przed pęknięciem próbki w skali makro. Haverich. Milenin. ed.wiss. 61-70. H.-M. M. 2010 a. M. Windhagen. Diard. Schaper.0955 mm ... Inc. M. P.-W.7 MPa. Niemeyer. F. Röcken. Fig. The prediction of the microcracks using multiscale model coincided with the results of the experiment. Stan metalu.9 MPa).-wiss. Financial support of the Ministry of Science and Higher Education project no.. P...-M. 61-68. M. identyfikacji parametrów pękania oraz implementacji modelu w skali mikro do modelu MES procesu ciągnienia. Steel Grips. T. eds. u. Amsterdam/London. R. It is shown that microcracks on grains boundaries have influence on parameters of wire drawing technology of thin wire from Mg-Ca alloys. J.. Grydin. Computers and Structures.1mm) drutów z trudno odkształcalnego biozgodnego stopu magnezu Ax30 przy wykorzystaniu wieloskalowego modelu numerycznego. Kusiak. 2. Y. J. Grydin.. D. Computer Methods in Material Science. 4131/B/T02/2009/37 and project no.. Int. North-Holland Publishing Company.

2012 Received in a revised form: October 29. po których w materiale wystąpiły pęknięcia. Wyżarzanie było wykonywane przed każdym przepustem. by nie pojawiły się mikropęknięcia w ciągnionym drucie. Received: September 22. 2012 Accepted: November 9. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 244 – . Drugi rozważany schemat ciągnienia pozwolił na przeprowadzenie 7 przepustów bez pojawienia się pęknięć.INFORMATYKA W TECHNOLOGII MATERIAŁÓW pęknięć. Eksperymentalna weryfikacja wyników obliczeń została przeprowadzona w warunkach laboratoryjnych w specjalnie do tego celu opracowanym narzędziu.1 mm. Tak więc. Drugi rozpatrywany schemat ciągnienia został dobrany tak. przeprowadzono walidację modelu na dwóch zasadniczo różnych przypadkach procesu ciągnienia. Początkowa średnica drutu wynosiła 0. otrzymano drut o średnicy 0. W pierwszym przypadku możliwe było przeprowadzenie 2-3 przepustów.075 mm bez defektów na powierzchni o plastyczności pozwalającej na dalsze ciągnienie. W tym przypadku pęknięcia po granicach ziaren były obserwowane na powierzchni drutu.

The continuous data approximation. to create an approximation of a generic bitmap in the finite element space. 2 THREE-DIMENSIONAL ADAPTIVE ALGORITHM FOR CONTINUOUS APPROXIMATIONS OF MATERIAL DATA USING SPACE PROJECTION PIOTR GURGUL*. Due to the crystallographic nature of polycrystalline material. 2010. The projection operator is applied iteratively on a series of increasingly refined meshes. namely the H1 projection. particular features are characterized by different properties that significantly influence material deformation. morphology of the digital material representation (DMR) during FE analysis of material behavior under deformation and exploitation conditions (Madej et al. Demkowicz & Buffa. 2005). In order to capture FE solution gradients properly. specific refined meshes have to be created. Demkowicz et. 2013. 2011.. MARCIN SIENIEK. concluding that hp-adaptivity for three dimensional approximation of non-continuous data loses its exponential convergence. specific refined meshes have to be created.. al. A proof of concept for a limited set of applications has been presented in earlier author’s works: (Gurgul et al. Key words: adaptive finite element method. 2010. Sieniek et al. which can be used in diverse applications including finite element (FE) analysis (Demkowicz. digital material representation 1. 2006). Finally conclusions with the evaluation and discussion of the numerical results for an exemplary problem and convergence rates obtained for described problem are described. INTRODUCTION Space projections constitute an important tool. To properly capture FE solution gradients which are the results of mentioned material inhomogeneities. Some examples may involve:  satellite images of topography of the terrain.pl Abstract The concept of the H1 projections for an adaptive generation of a continuous approximation of an input 3D image in the finite element (FE) framework is describe and utilized in this paper. The operator can be applied iteratively on a series of increasingly refined meshes. Paszyński et al.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. for example.g. al. It might be used.. ŁUKASZ MADEJ AGH University of Science and Technology.. A developed algorithm for linking image processing to the 3D FEM code is also presented within the paper. resulting in an improving fidelity of the approximation.. 2004. Poland *Corresponding author: pgurgul@agh. 2004. Such bitmaps can represent e. along with a corresponding FE mesh. 13. is necessary in case of a noncontinuous input data representing continuous phenomena. No. 2011.. space projections. can be interpreted and used as an input data for FE solvers. Such an approximation.  input data obtained by using various techniques representing temperature distribution over the 245 – 250 ISSN 1641-8581 . resulting in an improving fidelity of the approximation. 30-059 Krakow. Gurgul et al. Madej. 2012).edu. when we have a non-continuous bitmap data representing rather continuous terrain. In particular we compare hp-adaptive algorithm with h-adaptivity. Mickiewicza 30. MACIEJ PASZYŃSKI.

x3 that are produced by the function r(x).0)=u0. We conclude that for three dimensional H1 projection of continuous data the hpadaptivity loses its exponential convergence and thus h-adaptation is enough. 2006). For a given X = (x1. where u represents temperature. we find the closest existing (integer) coordinates for x1. . This is done using selfcontaining spaces corresponds . the method above considers only the function itself for minimization of the error. H-adaptive algorithm restricts the mesh refinement process to breaking selected finite elements with the fixed polynomial order of approximation. it is rather necessary to perform H1 projection of the u0 to get the required regularity of u.. . α f x (6) u needs to be minimal. . constiAn tutes the solution to this system.….. However. where the approximation will be performed. for which the desired precision is achieved. A number of adaptive algorithms for finite element mesh refinements are known. ∈ Given an arbitrary function . compare equation (7). 2011). x3). where the temperature is rather continuous phenomena. find such that . We can include information about derivatives and though minimize not only the error of function's value. the coefficients of this linear combination. COMPUTER METHODS IN MATERIALS SCIENCE Given an arbitrary function such that . we differentiate the equation with respect to the coefficients and compare them to zero in equation (1): ∑ This leads to a linear system (2): M U F (2) Ω 0 (1) 3. with initial conditions u(x. based on relative error rate in each step. to find the minimum. as it results in an exponential convergence with the number of degrees of freedom (Demkowicz et al. the application of the second case concerns the solution of time dependent problems with input data representing initial conditions. – 246 – . Then. we compute the approximation of .. x2. x2. we have to determine . PROJECTION OPERATOR projection onto the space V may be exA pressed as the following minimization problem: where: . HP-adaptation is one of the most complex and accurate. In this work we compare hpadaptivity with h-adaptivity for the H1 projection of non-continuous data. find ∈ is minimal. where to the initial mesh and is the first mesh. Given ‖f x ‖ Ω.e. When we solve the nonstationary problems of the form u = f. ( … ). where u0 is represented by a noncontinuous input data. The exemplary application of the first case may involve the flood modeling (Collier et al. Since the material data is not continuous in our case. Ω Ω (3) (4) projection onto the space V. ADAPTIVE ALGORITHM USED FOR SOLUTION Quality of the approximation depends on the choice of the space . and it results in algebraic convergence only. but projecalso its gradients.INFORMATYKA W TECHNOLOGII MATERIAŁÓW material. are basis functions for V (i.. Let:  uV be a solution in the space V. 2. (5) Thus the equation (6) f x u Ω Ω. There is no efficient way to determine precision of a given a priori and a workaround here is to refine space V iteratively. This method is called tion and can be expressed very similarly to projection. (7) ∑ Since where . we need to approximate the partial derivatives in the gradient f by finite differences. The hpadaptation process breaks selected finite elements into smaller ones and modifies the polynomial order of approximation locally.

← _ . 2. ← . Algorithm presented in figure 1 is being performed in iterations until the stop condition (usually the desired precision) is met.1. Two methods of adaptation are being considered in the present work: 3.1. Choice of an optimal mesh for the following iteration of the adaptive algorithm. 18: end for 19: return . ← ←t+1 _ do such that | 8: ← compute a projection of 4: while 9: | | 5: . In FEM terms. approximating space with the 12: end if 13: end for 14: add all basis functions from on 15: 16: 17: if max_ end if to .2. with supports The quality of the interpolation can be improved by the expansion of the interpolation base. 5: foreach element 6: ← ⊂ | Major steps of described algorithm are presented in figure 1. These steps are presented in figure 2. 1: ← ∞ 7: foreach ⊆ | | over element 0 do onto 2: ← 1 3: ← initial space corresponding to a trivial mesh > . where element refinements have been optimally chosen by comparing Vt fine and Vt . More functions in the base means smoother and more accurate solution but also more computations and the use of highorder polynomials.  Vt opt be a space corresponding to a mesh.  4: _ ←0 in coarse mesh do Vt w be any space such that Vt  Vt w  Vt fine . where all elements have been refined by one order with respect to Vt . One approach is to increase order of the basis functions on the elements where the error rate is higher than desired. ← . Algorithm _ . this could be done thanks to some kind of mesh adaptation. HP Mesh refinements and its role in projection-based interpolation . P-adaptation – increasing polynomial approximation level. 6: 7: ← perform listing 1 on 10: if 11: then 8: end while 9: return Fig. – 247 – COMPUTER METHODS IN MATERIALS SCIENCE 3.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 1: 2: 3: ← solve the minimization problem in ← ← solve the minimization problem in  Vt fine be a space corresponding to a mesh. _ Fig. 1.

2006). This idea arose from the observation that the domain is usually non-uniform and in order to approximate the solution fairly some places require more precise computations than others.4 63.nor the h-adaptation guarantees error rate decreases in an exponential manner with a step number.1. NUMERICAL RESULTS Presented projection algorithm was tested on one three dimensional example. This can be achieved by combining together mentioned two methods under some conditions. 4. where the acceptable solution can be achieved using small number of elements. h-adaptivite counterpart that delivers similar convergence to the one presented in table 2. This may represent the initial distribution of temperature over one ball shape material inside another material. Still. into four parts (both horizontally and vertically on one side). Another way is to split the element into smaller ones in order to obtain finer mesh. The refinement process is fairly simple in 1D but the 2D and 3D cases enforce a few refinement rules to follow. – 248 – .3 62.4.or p-refinement or not was developed. 3D balls problem: mesh after the sixth iterations of hp adaptive algorithm and solution over the mesh.1. which are not necessarily satisfied in the present case. The crucial factor in achieving optimal results is to decide if a given element should be split into two parts horizontally.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 3.3 66. and thus it is reasonable to replace it with its cheaper. 3. with simpler implementation and longer execution time.9 57. into eight parts (both horizontally and vertically on the both sides) or not split at all.. in an analogical manner to the algorithm for Finite Elements adaptivity described by (Demkowicz et al.03 COMPUTER METHODS IN MATERIALS SCIENCE Fig. This temperature distribution may constitute the starting point for some non-stationary time dependent heat transfer simulation. into two parts vertically. Table 1. Iteration 1 2 3 4 5 6 Mesh size 125 2197 5197 12093 22145 41411 Relative error in H1 norm 71. Convergence rate for the problem of H1 projections of 3D balls with hp-adaptivity. Automated hp-adaptation algorithm Neither the p. It decides if a given element should be refined or it is already properly refined for the satisfactory interpolation. The example concerns the approximation of the input data representing the ball shape distribution of data.7 51. H-adaptation – refining the mesh. in order to locate the most sensitive areas at each stage dynamically. The numerical results presented in table 1 obtained by the hp-adaptive solution show that the algorithm utilized for the three dimensional H1 projections does not deliver exponential convergence. That is why the automated algorithm that decides after each iteration for the element if it needs h. and improve the solution as much as possible.3. with hp-adaptive (see figure 3) and h-adaptive (see figure 4) algorithms. the self-adaptive algorithm can be applied. 3.

It is desirable to experiment with more 3D images as well as with various input parameters (e. Taylor and Francis. continuous interpolation of given arbitrary data. boundary conditions or image conversion algorithms). Computing With Hp-adaptive Finite Elements. M. CRC Press. Iteration 1 2 3 4 5 6 Mesh size 125 729 4913 11745 32305 68257 Relative error in H1 norm 72. Dalcin.. Sieniek. M.. Ł. Gurgul. Buffa. K. Table 2. DEC-2011/03/N/ST6/ 01397.. P. The University of Texas in Austin. which is significantly easier to implement. N.. 2006. Zdunek. Paszyński. Application of multi-agent paradigm to hp-adaptive projection-based interpolation operator.1 52.. Sieniek.1 68. The described method allows for generation of a smooth. H..04.. A. Convergence rate for the problem of H1 projections for 3D balls with h-adaptivity. 2004. H1. L. accepted to Journal of Computational Science. Magiera. The work of the second author was supported by Polish National Science Center grants no. ICES-Report 04-24.. L. Demkowicz. Skotniczny.. The work of the third author was supported by Polish National Science Center grant no. CONCLUSIONS AND FUTURE WORK This paper presents a way of incorporating wellestablished H1 projection concept into an adaptive algorithm used to prepare material data. In this paper we compared three dimensional hpadaptivity with h-adaptivity. NN519447739. Subactivity 4. 2011.. The work of the first author was partly supported by The European Union by means of European Social Fund. Demkowicz.1: Improvement and Development of Didactic Potential of the University and Increasing Number of Students of the Faculties Crucial for the National Economy Based on Knowledge. Acknowledgements. Gurgul.9 49. nr 820/N-Czechy 2010/0. Besides. The applicability of this methodology for non-stationary finite element method solvers will be tested in our future work. No. 2011. PO KL Priority IV: Higher Education and Research... H(curl) and H(div) – conforming projection-based interpolation in three dimensions.. Pardo. A.01. ICES Report 04-03.1. Madej. 2012.. The work of the fourth author was supported by grant no. accepted to Journal of Computational Science. P. Collier. 3D balls problem: mesh after the sixth iterations of the h-adaptive algorithm and solution over the mesh.. M.44 the quality of projection of the initial state to the further stability of the non-stationary simulation. Kurtz. M. L. and it may be reasonable to utilize just h-adaptation algorithm. Paszyński. with uniform polynomial order of approximation. Two dimensional hp-adaptive algorithm for con- – 249 – COMPUTER METHODS IN MATERIALS SCIENCE REFERENCES . UDA – POKL. Time Adaptivity in the Diffusive Wave Approximation to the Shallow Water Equations. 5.1: Improvement of the Didactic Potential of the AGH University of Science and Technology ``Human Assets''. N. D.3 68. Projection-based interpolation. 2004.01-00-367/08-00. M. Demkowicz. more sophisticated digital material representations should be investigated . concluding that the hpadaptive algorithm does not deliver the exponential convergence in the case of non-continuous approximation of data. Activity 4. alongside with an initial pre-adapted mesh suitable for further processing by a non-stationary FE solver.. The University of Texas in Austin. L...g.2 62. In particular we plan to test the influence of Collier. J..INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig.. Radwan. 4.

Development of the modeling strategy for the strain localization simulation based on the Digital Material Representation. L. 1. Madej. Paszyński. Omówiony został również sposób oszacowania i redukcji błędu aproksymacji. Artykuł przedstawia teoretyczne podstawy mechanizmu projekcji wraz z porównaniem algorytmów hp. L. L. 2011. 56. 2012 Accepted: October 26. 661-679. Romkes.. Procedia Computer Science. 2010... K..... Madej.. Cybułka P. Wilson. L.G. DSc dissertation.. Paszyński. Archives of Civil and Mechanical Engineering. 393-399. Digital Material Representation as an efficient tool for strain inhomogeneities analysis at the micro scale level. J. ICES-Report 05-38. 2011. Ł. Meiring... M. C. Archives of Metallurgy and Materials.. P. One the modeling of Step-and-Flash imprint lithography using molecular static models. AGH University Press. Gurgul. Demkowicz. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 250 – .. Ł. Pietrzyk. 2012 Received in a revised form: October 23. TRÓJWYMIAROWY.adaptacji i h-adaptacji użytych do iteracyjnego generowania kolejnych aproksymacji. może być interpretowana jako ciągła reprezentacja danych wejściowych dla solwerów metody elementów skończonych (MES). Analysis of the stress concentration in the nanomultilayer coatings based on digital Representation of the structure. Agent-based parallel system for numerical computations. Collister. A.. Received: September 26. Perzyński. Przedstawiony został ponadto przykład obliczeniowy ilustrujący działanie opisywanych metod. Taka aproksymacja. M. 1971-1981.. M. K. Sieniek. razem z odpowiadającą jej siatką. Major. ADAPTACYJNY ALGORYTM DO APROKSYMACJI CIĄGŁYCH DANYCH MATERIAŁOWYCH Z WYKORZYSTANIEM PROJEKCJI PRZESTRZENNYCH Streszczenie Celem niniejszego artykułu jest opis i pokazanie praktycznego wykorzystana koncepcji projekcji H1 do adaptacyjnej generacji aproksymacji ciągłej wejściowego obrazu w 3D w bazie elementów skończonych. 2005. Rauch. Kołodziejczyk. accepted to Computer Science. Madej. M..INFORMATYKA W TECHNOLOGII MATERIAŁÓW tinuous approximations of material data using space projections. 11... E. Krakow. 2010. Perzyński.. P.

The time needed for an evaluation of a single objective function depends on a boundary value problem usually solved by numerical methods. The scalability tests of the algorithm are performed using a server consisting of eight floating point units. As a result of using the hierarchical structure of the identification algorithm and the BEM. Institute of Computational Mechanics and Engineering. INTRODUCTION The bioinspired algorithms are very efficient optimization tools for single and multimodal objective functional problems (Michalewicz.kus@polsl. The solution of boundary value problems by the BEM and the determination of effective material properties by numerical homogenization method are also parallelized. identification. No. RADOSŁAW GÓRSKI Silesian University of Technology. A three-dimensional unit-cell model of a porous microstructure is modelled and analyzed by the boundary element method (BEM). a significant computation speedup and the accuracy are achieved.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. a complex microstructure may be represented for instance by means of a representative volume element (RVE) or a unit cell and can be 251 – 257 ISSN 1641-8581 . bioinspired algorithms. A recent increase in a computational power gives a possibility of studying different materials using a numerical homogenization approach. The main drawback of these algorithms is a large number (hundreds or thousands) of objective function evaluations. By using this technique. like the finite element method (FEM) or the boundary element method (BEM). 1996). The computation of the compliance matrix for a porous microstructure is shown.pl Abstract The problem of identification of the size of a void in a microscale on the basis of the homogenized material parameters is studied in this work. The algorithm used for identification is characterized by a hierarchical structure which allows for parallel computing on three different levels. Konarskiego 18A. boundary element method. 44-100 Gliwice. The method is very accurate and for the considered problem requires discretization only the outer boundary of models. 2 PARALLEL IDENTIFICATION OF VOIDS IN A MICROSTRUCTURE USING THE BOUNDARY ELEMENT METHOD AND THE BIOINSPIRED ALGORITHM WACŁAW KUŚ*. 2013. micromechanics. The matrix is used to formulate the objective function in identification problem in which the size of a void is searched. Key words: parallel computing. numerical homogenization 1. the numerical homogenization methods can be performed instead. The parallel algorithm is used for evolutionary computations. Poland *Corresponding author: waclaw. numerical simulations play today an important role in the prediction of a behaviour of new materials of a complex structure. 13. 2008). The overall wall time of identification can be shortened when the parallel algorithms are used (Kuś & Burczyński. Apart from the analytical models and experimental testing. Since the direct modelling and analysis of most of engineering structures made of heterogeneous materials is computationally very demanding.

for which the above integral equation is applied. Uij and Tij are fundamental solutions of elastostatics.j = 1. The studies in the literature concern the homogenization of different materials of complex microstructures.INFORMATYKA W TECHNOLOGII MATERIAŁÓW modelled and analyzed on two or more different scales. Identification of material parameters of a bone by using a multiscale modelling and a distributed parallel evolutionary algorithm is presented by Burczyński et al. for example composite materials. (2010a). The stress-strain relationships for an orthotropic material are presented. an evolutionary algorithm (an optimization tool). Boundary integral equations for a general threedimensional (3D) isotropic body are shown. It is assumed that the material is macroscopically orthotropic and that a macro model of a structure made of this material is subjected to small deformations. The external boundary of the body is denoted by . In the present paper the identification of voids in microstructures modelled by the BEM is presented. the developed system is build of three programs. In order to solve the problem. Araujo et al. Düster et al. Fang et al. First. the FEM and the BEM are the most frequently used. the relation between the loading of the body and its displacements can be expressed by the boundary integral equation (known as the Somigliana identity) in the following form (Gao & Davies. (2010) have modelled and analysed threedimensional composite microstructures by the BEM and the parallel algorithm. – 252 – . a computational homogenization module (evaluation of objective function) and the BEM program (a boundary value problem solver).x  u j  x  d  x     U  x  .e. Chen and Liu (2005) have analysed composites reinforced by spherical particles or short fibres by the advanced BEM. analytical homogenization procedures are out of the scope of this paper and will not be discussed. the parameters defining voids in the material on a microscale are determined on the basis of orthotropic parameters in a macroscale. COMPUTATIONAL HOMOGENIZATION BASED ON THE BEM In this section. 2. The important feature of the method is a possibility of discretizing of complicated microstructures in a fully automatic way. As a result. Multiscale analysis by coupling the molecular statics and the BEM is presented by Burczyński et al. Because the main emphasis in this work is put on parallel bioinspired computations assisting numerical homogenization.2. The numerical homogenization of an orthotropic material is shown for which the homogenized properties are determined. (2010b). heterogeneous tissue scaffolds and other. They computed the effective mechanical properties of different scaffolds materials and pore shapes. The body is statically loaded along the boundary  by boundary tractions tj. The summation convention is used in the equation (the indices for a 3D problem are i. Difficulties in dealing with nearly-singular integrals during modelling of composites with closely packed fillers have been resolved by new and improved techniques. The porous microstructure is modelled and analyzed by the BEM. displacements of the body are denoted by uj. Optimization of macro models analyzed by the FEM using parallel algorithms is shown by Kuś and Burczyński (2008).3). The properties are used to formulate the objective function depending on the quantities of a macro and micro model in order to identify the size of a void in the unit cell model of the material. the macroscopic homogenized properties of the material are determined on the basis of analysis of the unit cell models in a micro scale. x  t  x  d  x  ij j  (1) where x’ is a collocation point. the idea of a numerical homogenization is described within the framework of a linear elastic material characterized by a periodic microstructure containing voids. consider a 3D body (a macro model) made of a homogeneous. Three-dimensional unit-cell models (which play the role of a RVE) of microstructures with voids are considered. 2002): COMPUTER METHODS IN MATERIALS SCIENCE cij  x  u j  x    Tij  x. (2012) have shown a new approach for the numerical homogenization of heterogeneous and cellular materials using the finite cell method. Assuming that the body forces does not act on the body. x is a point on the external boundary . Among the numerical homogenization methods. (2005) for instance have studied homogenization of porous tissue scaffolds by the FEM and by two other approaches. a constant cij depends on the position of the point x’. i. cellular materials. the identification problem is solved by the parallel hierarchical algorithm. isotropic and linear elastic material. In order to speed up the computations. As a result of the numerical homogenization by the BEM.

For the considered orthotropic material in a macro scale. 13. 12 are engineering stresses. 6 numerical tests are performed using the unit cell in figure 1. In the RVE or the unit cell analysis. the relation between strains and stresses is formulated in an av- – 253 – COMPUTER METHODS IN MATERIALS SCIENCE . In the developed computer program the outer boundary of the body is divided into 8-node quadratic boundary elements. representative sections (volumes) of a material are analyzed in order to calculate the homogenized properties. In order to compute the elements of the compliance matrix. 13. i. a2. in which the strains and stresses in equation (3) are defined. A unit cell of this material (a micro model) representing its porous microstructure contains a single rectangular prism of arbitrary side lengths (i. which are perpendicular to the three symmetry planes. which is successively applied for all collocation points. but only 9 are independent. 3 tensile tests and 3 shear tests. In this work. The resulting BEM equations can be expressed in the following matrix form: H uG t (2) erage sense in order to determine the homogenized (the effective) macroscopic properties. the representative volume elements (RVE) representing a microstructure of this material should be rather used than the unit cell models.e. the material in a macro scale is referred to as orthotropic. In order to determine the homogenized macroscopic properties represented by this matrix. Using the engineering notation. Along the external boundary the variations of coordinates.e. the compliance matrix is specified in the coordinate system defined by these axes. When the first stress state is applied. 2. More comprehensive definitions of the RVE can be found elsewhere. a3) as shown in figure 1. The walls of the unit cell in figure 1 align with the x1. 1. A unit cell model of an orthotropic material If a material has a non-regular and non-uniform microstructure. Repeating an analysis five more times for the remaining unit stress vectors allows determining all columns of the compliance matrix. the Fig. The coupling of the macro and micro levels is based on the averaging theorems. H and G are coefficient matrices dependent on the boundary integrals of fundamental solutions and shape functions. 12 are engineering strains. the strain-stress relationships for an orthotropic material have the following form (Kollár & Springer.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Numerical BEM equations are obtained after discretization of the boundary integral equation (1). x2 and x3 axes. for instance in Kouznetsova (2002). a1. 3 and 23. 1. respectively. for the tensile test in the x1 direction only the traction in this direction (1 stress) is prescribed and the remaining are zero. then the resulting strains are obtained from the strain-stress relationships for an orthotropic material and the first column of the compliance matrix in equation (3) is determined. homogeneous static boundary conditions are applied. The unit tractions are prescribed to the unit cell models. 2003): where u and t are displacement and traction vectors. displacements and tractions are interpolated using quadratic shape functions.  ε1   S11  ε  S  2   12   ε3    S13   γ23   0  γ13   0      γ12    0 S12 S22 S23 0 0 0 S13 S23 S33 0 0 0 0 0 0 S44 0 0 0 0 0 0 S55 0   σ1   σ   2    σ3     0   τ 23  0   τ13    S66    τ12   0 0 0 (3) where the compliance matrix S has 12 nonzero elements. 2. The means for obtaining the elements of the compliance matrix S for an orthotropic material by using the numerical homogenization concept and the BEM are presented below. Thus. Consider now a heterogeneous material with a periodic microstructure with voids in the form of rectangular prisms. 3 and 23. For instance. The mechanical properties of a linear elastic material are characterized by the compliance matrix S or by the stiffness matrix C. Because there are three mutually perpendicular symmetry planes with respect to the void aligned along the x1. The compliance matrix S is symmetrical for an elastic material (Sij = Sji) and it is the inverse of the stiffness matrix (S = C-1). x2 and x3 axes. 1. and their elements are integrated numerically using the Gauss quadratures.

The fitness function defined by equation (4) is obtained by solving six boundary value problems with the use of the BEM and the homogenization procedure. On each level different program of the developed system consisting of three modules is applied. S66}. 4. S12. the value from the database is used. The loop of the algorithm is repeated until the end condition is fulfilled (expressed e. These strains provide the relevant terms in the compliance matrix S and thus the effective properties. The physical memory installed in a server should be taken into account during parallelization of the algorithm to prevent swapping memory to disk which may lead to much longer wall time of computations. a2.g. The parallelization of the evolutionary algorithms is quite easy (Kuś & Burczyński.INFORMATYKA W TECHNOLOGII MATERIAŁÓW relation between strains and stresses is formulated in an average sense. S22. the randomly chosen chromosomes and their genes are modified by using evolutionary operators. PARALLELIZATION OF IDENTIFICATION ALGORITHM The aim of parallelization of the identification algorithm is to obtain the results as fast as possible. The efficiency of using a parallel algorithm is high especially for problems for which evaluation of a fitness function is long (from seconds to hours or in some cases days). In the considered process of identification solving of a boundary value problem is the most time consuming task. on the second the computational homogenization procedure and on the third the BEM program for the solution of a boundary value problem (parallel system of equations solver . The procedure saves much time because solving of a boundary value problem is usually the most expensive operation in terms of time during the evolutionary process. 3. The maximum number of parallel ˆi F   si  s i 1 n (4) where si are computed homogenized material properties. The design variables (the side lengths a1. The same value of the objective function may be obtained for different number and other parameters of voids. the island (also called distributed) version of the evolutionary algorithm is proposed in the present work. 2008).e. FORMULATION OF IDENTIFICATION PROBLEM The identification problem consists in finding the side lengths a1. In order to improve the evolutionary process of the algorithm and speed up the computations. i. The memory usage by the algorithm is important because the methods used in the paper increase memory requirements. S33. Identification problems belong to a class of illdefined problems and the uniqueness of the solution is not guaranteed. a3) are coded in genes of each chromosome which is a potential solution of the problem. In the next step. The new generation is created on the basis of the offspring population during the selection process. from experiments). The identification problem is solved by the evolutionary algorithm. S13.e. a2. If this is the case. The parameter nLx is a number of threads used by a program on level x. a3 of the void in the unit cell model in figure 1 by minimization the following functional F dependent on the elements of the compliance matrix: phase. in which a population of chromosomes are processed in each iteration. Two factors are taken into account in the parallelization strategy: wall time of computations and memory consumption. Then the values of the objective function (fitness function) for all chromosomes are calculated. the elements of the compliance matrix si ˆi are ref={S11. i. It uses few subpopulations of chromosomes which evolve separately. S55. n is a number of independent material parameters for an orthotropic material (n = 9 in this case). the initial population of chromosomes is generated randomly. on the first level the evolutionary algorithm.PSS). It prevents from the evaluation of fitness function for chromosomes which have the same genes. s erence homogenized material properties related to a macromodel (e.g. The developed algorithm uses a database containing an information about evaluated chromosomes and their fitness function. The parallelization is hierarchical and the total number of parallel threads is equal to the multiplication of the parameters nLx for all three levels. S23. as a maximum number of iterations). At the beginning. The parallelization of the identification algorithm can be performed on at least three levels as shown in figure 2. Another improvement concerns the evaluation of fitness function. The chromosomes between subpopulations can be exchanged during a migration COMPUTER METHODS IN MATERIALS SCIENCE – 254 – . The average strains are computed on the basis of displacements obtained from the BEM analysis by their integration over the boundaries of the models. S44.

The homogenization procedure runs 6 BEM analyses therefore the number of parallel tasks should be 1. Homogenized properties of the reference material Material parameter Value [GPa-1] S11 1. For each test the prescribed traction to a wall of the unit cell is p = 1 MPa. probability of uniform mutation is 10%. number of iterations is 50.027 GPa-1. Each outer wall of the unit cell and inner wall of the void is divided into 16 quadratic boundary elements. The maximum number of parallel tasks is 6 for a 3D problem. 50. The parallelization increases the memory requirements for computations. The most important is parallelization of solving of the system of equations. the side lengths a1. The parallel approach is realized with the use of an Intel MKL library. The server contains two processors AMD Opteron 6272. The boundary value problem is solved with the use of the BEM on third level of parallelization.319 8.0 GPa and Poisson’s ratio  = 0.3 Table 1. of 0.763 5. The parallelization on level 2 is related with the parallel computational homogenization. i. The parameters of the evolutionary algorithm are as follows: number of genes is 3. probability of simple crossover with Gaussian mutation is 90%.300 0. In all tests. Actual and found void side lengths Void parameter Actual Found Error % . The elements of the reference compliance matrix were obtained for an actual void and its side lengths shown in table 2. 2. The constraints on design variables.076 S12 -0. 2.3.318 S13 -0.200 0. are imposed and each is within the range The tests were performed with the use of a server Dell PowerEdge R515.317 S33 1. The parallel evolutionary algorithms that use the floating point representation operate on small populations of chromosomes.346 0.400 0.217 0. The corresponding value of the objective function is F = 0. a2.319 S22 1.5 13.85 mm. The results of identification and an error with respect to the actual void are shown in table 2. 20. The linear elastic material properties of the microstructure are as follows: Young’s modulus E = 1. The unit cell size is 111 mm. The memory consumption is nL1 times larger than in the case of a sequential algorithm. resulting in 192 elements for the whole model. NUMERICAL TESTS A geometry of the considered RVE is presented in figure 1. Table 2.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. The homogenization procedure consists of boundary value problems solved in a parallel way and sequential algorithm which computes homogenized material properties.780 S66 2. number of chromosomes is 20. Several steps of the BEM algorithm can be parallelized. The memory amount is a sum of memory requirements for the evaluation of a fitness function for each chromosome.729 S55 2. The orthotropic properties of the reference material are shown in table 1. each with 16 cores (8 floating point units). 3 or 6 in order to use all cores equally.057 S44 2. In the BEM the full matrices are created thus the standard algorithms like LAPACK can be used in order to solve a system of algebraic numerical equations.050 S23 -0.05 to 0. for example 10. parameters of the evolutionary algorithm are the same as in the previ- – 255 – COMPUTER METHODS IN MATERIALS SCIENCE threads nL1 is equal to the total number of chromosomes.e. a3. A hierarchical parallel structure of identification algorithm a1 a2 a3 0.5 6.

J. Times of identification for different number of tasks Test 1 2 3 4 5 6 7 8 9 10 11 12 13 nL1 1 1 1 1 1 1 2 2 1 6 2 16 8 3 6 nL2 1 1 1 1 1 2 1 2 6 1 6 1 2 6 3 nL3 1 2 4 8 16 1 1 1 1 1 1 1 1 1 1 Number of threads 1 2 4 8 16 2 2 4 6 6 12 16 16 18 18 Time [s] 21 220 18 136 17 983 17 112 16 449 11 937 12 039 6 861 5 975 5 802 3 426 3 213 2 806 2 549 2 539 Speedup 1 1. Kuś. Comput. Appl. parameters defining geometry of a void were successfully identified. Sun. Int. 2. Z. Burczyński. Kollár.C. Cambridge University Press.29 1.... and finite element approach.. T. 1285-1292.. Brodacka. W. Yan. Chen. Numerical homogenization of heterogeneous and cellular materials utilizing the finite cell method.J.17 1.. Kuś. Boundaryelement parallel-computing algorithm for the microstructural analysis of general composites. Molecular statics coupled with the subregion boundary element method in multiscale analysis. ABBI. Springer-Verlag.36 tification.. Fang. Liu. A three-dimensional unit-cell model of a porous material is modelled and analyzed by the boundary element method (BEM).. Gao. the homogenization allows 6 parallel threads. Genetic algorithms + data structures = evolutionary algorithms.. Theor.S.. 413-431. A. A. W. The advantages are valuable and can be exploited in more complex problems dealing for instance with numerical homogenization and optimization or iden- – 256 – .. 855870. Boundary element programming in mechanics.09 3. the total number of cores which may be used in computations is equal to 108. Anal. Eng.. In order to solve the problem.. Kuś. PPAM 2007. Burczyński. X.. REFERENCES Arújo.. W..INFORMATYKA W TECHNOLOGII MATERIAŁÓW ous example...56 8. Gray. Multiscale Comput. H. Shokoufandeh. 2005.. LNCS.. 48.76 3. X. 2003... Eindhoven.60 7. Dongarra. The main advantage of using the BEM in analysis is its high accuracy and that it requires discretization only the outer boundary of the considered models. Berlin. Eng. 2008. d’Azevedo. W.. Górski R. E. 2010b. F. Kouznetsova. Sehlhorst. Davies.G. Acknowledgements. The times of identification for all tests are presented in table 3.J. the hierarchical parallelization of the algorithms was developed. COMPUTER METHODS IN MATERIALS SCIENCE 15 16 6. Springer. Cambridge New York. 2002. 88. Wyrzykowski.G. T.. Mech. In the future the authors plan to use a cluster with more number of cores to check the scalability of the presented approach. 319-331. 513-523. 2010. The parallel evolutionary algorithm and the homogenization procedure are characterized by a similar efficiency (tests 7-16). PhD thesis. T..32 8.. The results of numerical tests with wall time measurements for different number of cores are shown. G..F.L. CONCLUSIONS The identification of the size of a pore in a micro scale model on the basis of parameters in a macro scale is considered in this work. The parallelization on the level 3 (tests 1-6) is not efficient due to a partial parallelization of the BEM algorithm.. Wasniewski. R. Regli. Z. Elem. E. Computational homogenization for the multi-scale analysis of multi-phase materials. Mech.55 3. Comput. 29.. W. An advanced 3D boundary element method for characterizations of composite materials. C. L.78 1. V..66 6. Parallel Bioinspired Algorithms in Optimization of Structures. Homogenization of heterogeneous tissue scaffold: A comparison of mechanics. Multiscale modeling of osseous tissues. Y. Burczyński. The speedup is computed in a reference to results for the test 1.. Michalewicz.. Rank. Technishe Universiteit Eindhoven. 2010a. Mechanics of composite structures. A..19 6. The maximum number of parallel evolutionary algorithm threads is 18. 1996. 2002. 8. The scientific research has been financed by the Ministry of Science and Higher Education of Poland in years 2010-2012. J. L. eds. The maximum value of the nL1 parameter is 18 (which corresponds to a maximal number of chromosomes evaluated in each iteration of the evolutionary algorithm). 17-29. Table 3. 2012. respectively. Karczewski. Cambridge University Press. 50.. The time of computations is reduced from about 6 hours to about 40 minutes when one core and the parallel approach is applied. Mrozek A. J.P. In numerical examples. 2005.18 1. J. Struct.W. Düster. 4967. Bound. asymptotic homogenization. 773-784. K. T.24 1.. Cambridge...

Received: October 11.INFORMATYKA W TECHNOLOGII MATERIAŁÓW RÓWNOLEGŁA IDENTYFIKACJA PUSTEK W MIKROSTRUKTURZE Z WYKORZYSTANIEM METODY ELEMENTÓW BRZEGOWYCH ORAZ ALGORYTMU INSPIROWANEGO BIOLOGICZNIE Streszczenie W pracy przedstawiono zagadnienie identyfikacji rozmiaru pustki w skali mikro. Pokazano sposób wyznaczania macierzy podatności mikrostruktury porowatej. na podstawie zhomogenizowanych parametrów materiałowych. Wykorzystano równoległy algorytm do obliczeń ewolucyjnych. 2012 Accepted: October 29. Metoda jest bardzo dokładna i dla rozważanego zadania wymaga jedynie dyskretyzacji zewnętrznego brzegu modeli. Przeprowadzono testy skalowalności algorytmu z użyciem serwera zawierającego osiem jednostek zmiennoprzecinkowych. Zrównoleglono także rozwiązywanie zadań brzegowych za pomocą MEB oraz wyznaczanie zastępczych własności materiałowych metodą numerycznej homogenizacji. Jako rezultat zastosowania algorytmu o budowie hierarchicznej oraz MEB uzyskano znaczne przyśpieszenie i dokładność obliczeń. 2012 Received in a revised form: October 22. Trójwymiarowy model komórki jednostkowej mikrostruktury porowatej modelowany i analizowany jest metodą elementów brzegowych (MEB). Macierz jest wykorzystana do sformułowania funkcji celu w zagadnieniu identyfikacji. 2012 – 257 – COMPUTER METHODS IN MATERIALS SCIENCE . w którym poszukiwany jest rozmiar pustki. Zastosowany algorytm do identyfikacji charakteryzuje się hierarchiczną budową pozwalającą prowadzić obliczenia w sposób równoległy na trzech różnych poziomach.

Existing numerical tools are powerful and offer various possibilities. 2 APPLICATION OF THE THREE DIMENSIONAL DIGITAL MATERIAL REPRESENTATION APPROACH TO MODEL MICROSTRUCTURE INHOMOGENEITY DURING PROCESSES INVOLVING STRAIN PATH CHANGES KRZYSZTOF MUSZKA*. multiscale modelling. 30-059 Kraków. however there are still limitations in the modelling of processes that are characterised by non-linear and non-symmetrical deformation modes. precipitation and phase transformation kinetics during hot deformation of steels. Due to a combination of the multiscale finite element model with the DMR approach. High local strain accumulation leads to significant grain refinement and significantly improves strength of the material but in some cases (severe plastic deformation methods) decreases ductility of the material. No. both theoretically and experimentally (Davenport et al. Strain path changes applied during cold deformation also play important role in the control of strain and microstructure inhomogeneity. High strain gradients. Computer modelling needs to be involved in order to learn how to control the microstructure and ISSN 1641-8581 258 – 263 . Understanding of the strain path in the light of aforementioned problems is therefore of paramount importance. 13. Proper control of those parameters is very difficult and can be effectively optimised only if the numerical tools are involved. Jorge-Badiola & Gutierrez. It has been found that this processing parameter significantly retards recrystallization. metal may be subjected to complex strain path changes that introduce high level of both deformation and microstructural inhomogeneity and make the prediction of material behaviour extremely difficult. 2013.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. Faculty of Metals Engineering and Industrial Computer Science.edu. material undergoes complex loading history that introduces significant inhomogeneity of the strain.pl Abstract The present paper discusses possibilities of application of the 3D Digital Materials Representation (DMR) approach in the light of the multiscale modelling of materials subjected to the complex strain paths. Problem of strain path change on the microstructure evolution and mechanical behaviour has been widely studied. Key words: 3D digital material representation. In some metal forming processes. Poland *Corresponding author: muszka@agh. Mickiewicza 30. The 3D Digital Materials Representation approach is presented and introduced in the present paper into a multiscale finite element model of two metal forming processes characterised by high microstructural gradients: the cyclic torsion deformation and the Accumulative Angular Drawing (AAD). INTRODUCTION During manufacturing processes. 2004). lead to high inhomogeneity of microstructure and make the prediction of the final material’s properties especially complicated. detailed information on strain inhomogeneities was obtained in both investigated processes. 1999. in turn. ŁUKASZ MADEJ AGH University of Science and Technology. strain path changes 1.

Flow curves recorded during cyclic torsion deformation of Fe-30wt. Since in Fe-30wt%Ni systems austenite phase is stable down to room temperature and they are characterised by similar Stacking Fault Energy and high temperature flow behaviour as low carbon steels. It can be metry. Due to nonlinearity and lack of symcal microscopy are shown infigure1 b. The first case study involves cyclic torsion deformation of the FCC structure. mulstructure has been subdivided into well-developed tiscale modelling approach should also be considlamellar structures separated by high angle grain ered. they are widely used to model the austenite phase of those materials.51Mo-0. In the present paper. 4 cycles of forward/reverse with strain of 0. Additionally. the original shape of austenstrain path changes requires 3D models to be creatite grains has been restored. structure. Accumulative Fig.%Ni.INFORMATYKA W TECHNOLOGII MATERIAŁÓW deformation inhomogeneity during complex loading Deformed microstructures observed using optiprocesses. Fig. Torsion test was carried out using servo-hydraulic torsion rig at 840ºC with strain rate of 1/s. BCC structure is modelled. Two deformation routes were applied.-c) respectively. In the case of 2-pass ed. The present study confirmed that strain path effect represent one of the most important processing parameters characterising hot metal forming processes. where occurrence so called Bauschinger effect – due to microstructure is explicitly represented by properly rearrangement of the substructure upon reversal the divided heterogeneous finite element a) b) c) mesh. however.. Proper representation of the microstructural boundaries (Sun et al. 2. 2-passes of deformation with the strain of 1 per pass and only one reversal were applied. As most of the microstructural phenomena durdeformation. 2011).25 per pass (8-passes in total) were applied. both with the same equivalent total strain of 2. In the second case. simulation of deformation involving complex seen that in both cases.19Si (in wt.. The recorded flow features can be effectively done with utilisation of curves are summarized in figure 2. optical microstructures of the deformed samples using deformation route Angular Drawing (AAD) process of the 1 and 2 -b). the 3D Digital Materials Representation approach is presented and incorporated into a multiscale finite element model of two metal forming processes characterised by high microstructural gradients. The initial microstructure of the studied material represented by EBSD map is shown in figure 1. Computer modelling of such problems can put some new insight into understand- – 259 – COMPUTER METHODS IN MATERIALS SCIENCE . %). whereas in 1. Solid bar torsion specimens with gauge length of 20mm and gauge diameter of 10mm were machined out of the solution treated plate. Various austenite state as an effect of different strain path in steel is crucial since it affects the subsequent phase transformations and thus its products what in turn has an effect on the properties of the final materials. In the first case.1. In both cases. 2011) technique. c. Forward/reverse torsion test The effect of strain reversal on austenite subjected to strain path reversal was studied in torsion using model alloy system with a chemical composition of 0.092C-30.67Mn-1. Electron Backscatter Diffraction (EBSD) map of the initial austenite microthe second case study. the initial austenite microing deformation take place at various scales.3Ni-1. EXPERIMENTAL INVESTIGATION that this effect has been multiplied in the 8-pass test. the the recently developed Digital Materials Representastrain level upon reversal is lower what suggests tion (DMR) (Madej et al. 2. dislocation density in the reversed structure is lower. based on the flow curves it can be seen 2.

The main aim of the present work was to study whether combination of the multiscale finite element modelling with 3D DMR approach can be used to effectively model complex deformation processes that were described in the previous chapter. i. Accumulative Angular Drawing (AAD) process to bending/unbending process. It is a combined effect of: reduction of the area. σ0 is the yield stress at zero plastic strain and Q∞ and b are material parameters. 2. αk is the backstress. the present study was concentrated on the stepped die positioning. and an isotropic hardening component describing the change of the equivalent stress defining the size of the yield surface. in the areas near the surface. substantial grain refinement was achieved in the transverse section of the wires. The dependence of grain shape. The evolution law of this model consists of the two main components: a nonlinear kinematic hardening component which describes the translation of the yield surface in stress space through the backstress α: Optical and electron microscopy observations have shown high level of microstructure inhomogeneity. The presented work confirmed that the strain path applied in the AAD process affects directly the microstructure and texture changes in the final product.27Si//0.. NUMERICAL INVESTIGATION In order to carry out the study of the AAD process. strain accumulation in the outer part of the wire due  0   0  Q 1  e  b    (2) where. 3.37Mn/ 0. with homogenous equiaxed ferrite microstructure and the mean grain size of 15μm. Although the AAD design allows various combinations of die positioning to be used.e. Initial microstructures of the studied material taken in the longitudinal –a) and transverse –b) cross-section. in which the offset from the drawing line between the successive dies was equal to 15°.07Nb/0. The refinement of the microstructure is localised in the near-surface layers. was used in this study. Again numerical modelling can be a valuable support to the experimental research on these effects.5mm diameter wire rods were drawn down to the diameter of 4 mm through the set of three dies (in two passes of drawing) with the total strain of 0.INFORMATYKA W TECHNOLOGII MATERIAŁÓW ing and optimisation of the processes carried out with strain path changes. Grains were also elongated along the wire axis. 3.97.2. Euler angle maps of the deformed wires taken near the surface –b) and in the centre –d) of the longitudinal cross section of the deformed wire. material behaviour was described using elasto-plastic model with combined isotropic-kinematic hardening (Lemaitre & Chaboche. a special die was designed such that an ordinary drawing bench could be used (Wielgus et.009N) supplied as a wire rod. as a function of plastic deformation: c) d) COMPUTER METHODS IN MATERIALS SCIENCE Fig. size and distribution on the transverse cross section on the processing route is clearly seen in figure3. however. al. Microalloyed steel (0. In both cases. 1990). is the number of backstresses.     k 0  k 1 1 N (1) where. N. The 6. 2010). The model is – 260 – . and desired shear deformation. σ0 the equivalent stress defining the size of the yield surface and Ck and γk are material parameters. a) b)  k  Ck      pl   k  k  pl . Calculations were performed using Abaqus Standard/Explicit package.07C/1. with various intensities.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW based on the two major model parameters Ck (the initial kinematic hardening moduli) and γk (rate at which the kinematic hardening moduli decrease with increasing plastic deformation). Equivalent von Mises stress distributions in the global model and in unit cell during the first forward/reverse cycle of torsion test are presented in figure 6a. Fig. Due to presented advantages authors decided to apply the same approach to model the AAD process. Multiscale model of the cyclic torsion test.1. Tools were meshed with quad-dominated discrete rigid elements (R3D). The multiscale model of the torsion test was designed as seen in figure 4. twist angle. Multiscale model of the AAD process . or can be obtained based on the test data from a stabilized cycle (when the strain-stress curve no longer changes shape from one cycle to next). Submodelling technique was used to bridge different scales. Furthermore. Multiscale model of the AAD process due to its complexity requires two steps of submodelling as presented in figure 7. second submodel was generated using the 3D DMR approach and calculations were performed again. Its Fig. – 261 – COMPUTER METHODS IN MATERIALS SCIENCE 3. Comparison of the measured and calculated torque vs. It can be seen that the divergence of the model and experiment is good what proves the accuracy of the applied methodology. A unit cell (100 μm 100 μm  100 μm) with 37 grains was created to capture the effect of the process on inhomogeneity of both strain and microstructure.5mm was modelled. Parameters of the model have been identified using inverse approach based on data from cyclic torsion test that was performed on studied materials at both deformation temperatures. It can be seen that the application of the multiscale modelling approach and its combination with 3D DMR approach resulted in much higher accuracy of the results compared to simulation using only the global model (figure 6a). These parameters can be specified directly. global model with 42000 eight-node hexagonal reduced integration elements with hourglass control (C3D8R) was realized using Abaqus Explicit. the submodel was generated using the DMR approach and calculations were performed again. Next.1. calibrated based on a halfcycle test data (unidirectional tension or compression). It can be seen that the first pass of cyclic deformation caused grain rotation. Additionally 3D DMR properly captured not only inhomogeneites in stress or strain state but also grain shape changes – as an effect of strain reversal (figure 6c). angle data calculated using calibrated hardening model and data measured experimentally (cyclic torsion test) are presented in figure 5. however macro scale model neglects inhomogeneities occurring along microstructure features. The parameters of the combined material hardening model applied in the submodel were additionally diversified using the Gauss distribution function to reflect differences in the crystallographic orientations. Multiscale model of the cyclic torsion test original position was then restored after strain reversal and application of the second pass of deformation with the same strain level applied in the opposite torsion direction. Drawing of 300 mm long wire with the initial diameter of 6. Example of the comparison of the torque vs. Finally. Global model of strain gauge was prepared and analysed using Abaqus Standard code. b. Macro scale model is unable to provide such detailed results. 3. Global material response obtained from both models can be similar to some extend. the analysis was replicated on the smaller cylindrical area (10mm long) subdivided from the global model using Abaqus Standard and much finer mesh was used. 5. 4. First.

– 262 – . Equivalent plastic strain in drawn wire after 1st pass of drawing. Fig. Examples of calculations. 8. 7. COMPUTER METHODS IN MATERIALS SCIENCE using Abaqus Standard. Von Mises stress distributions in global –a) and submodel –b) Equivalent plastic strain distribution in the selected grain –c) Fig. Multiscale model of the AAD process. 6. Set of 5 unit cells (100 μm x100 μm x 100 μm) containing 37 grains each was created to capture the effect of the process on inhomogeneity of both strain and microstructure. Obtained global equivalent plastic strain distributions on the surface and on the transversal cross section of the drawn wire after the first pass are presented in figure 8.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. Global model and unit cells attached at various positions of the wire’s cross-section.

2012 Accepted: December 11... Wynne. Effect of strain path on mechanical properties of wire drawn products. Muszka. Sellars. C. localisation and distortion across subsequent grains resulting from the AAD process. gdy zostanie wsparta narzędziami numerycznymi. Packo. 333-341. Received: October 17. J. Mechanics of Solid Materials. Acknowledgements. J. Philosophical Transactions of the Royal Society of London A. CONCLUSIONS Two complex loading cases with high local strain accumulation were simulated using multiscale FEM model combined with 3D Digital Materials Representation approach. która charakteryzuje się dużą niejednorodnością odkształcenia. ZASTOSOWANIE TRÓJWYMIAROWEJ CYFROWEJ REPREZENTACJI MATERIAŁU DO MODELOWANIA NIEJEDNORODNOŚCI MIKROSTRUKTURY W PROCESACH CHARAKTERYZUJĄCYCH SIĘ ZMIENNĄ DROGĄ ODKSZTAŁCENIA Streszczenie W pracy przedstawiono możliwości wykorzystania trójwymiarowej Cyfrowej Reprezentacji Materiału do wieloskalowego modelowania materiałów odkształcanych w warunkach zmiennej drogi odkształcania. 64.L. Steel Research International. P.J. Jorge-Badiola. D. R.. 5. Financial support the Polish Ministry of Science and Higher Education (grant no. Based on the presented modelling results it can be concluded that the applied modelling strategy was able to catch most of the important phenomena accompanying processes with complex deformation modes with reasonably good accuracy. 11. Higher strain accumulation near the wire surface was also predicted by the computer model (figure 8d. Future research will focus on application of the crystal plasticity model integrated with DMR what will even more extend predictive capabilities of the proposed methodology. Archives of Civil and Mechanical Engineering. Perzynski. 357. L.B. 661-679.. 280-283.L. 52.. FEM calculations were realised at ACK AGH Cyfronet Computing Centre under grant no: MNiSW/ IBM_BC_HS21/AGH/075/2010. 81. 2012 Received in a revised form: November 26.. Sun. J. Majta. Madej.. e.. J. Scripta Materialia. Rauch L. Cambridge University Press. Study of the strain reversal effect on the recrystallization and strain-induced precipitation in a Nb-microalloyed steel. Digital Material Representation as an efficient tool for strain inhomogeneities analysis at the micro scale level. REFERENCES Davenport.. W procesach przeróbki plastycznej materiał poddawany jest złożonej historii odkształcania.INFORMATYKA W TECHNOLOGII MATERIAŁÓW It can be noticed that the inhomogeneity of strain that is characteristic for this deformation process was properly captured by the applied model. Wielgus. The 3D DMR approach show different levels of strain inhomogeneity. f). N N508583839) is gratefully acknowledged. K. Higginson.M. Duży gradient odkształcenia prowadzi z kolei do niejednorodności rozwoju mikrostruktury i powoduje. It can be seen that application of the 3D DMR approach for the modelling of AAD can be an effective support of the experimental research. Gutierrez. 1645-1661. Palmiere. much more detailed information regarding strain localisation and inhomogeneities can be extracted from the submodels in comparison to macro scale model predictions.. że przewidywanie własności wyrobu finalnego staje się szczególnie skomplikowane. że połączenie wieloskalowego modelu MES wraz z trójwymiarową Cyfrową Reprezentacją Materiału wpływa na znaczą poprawę dokładności uzyskiwanych wyników w przypadku modelowania niejednorodności odkształcenia w rozpatrywanych procesach przeróbki plastycznej... 2011.. E... The effect of strain path reversal on high-angle boundary formation by grain subdivision in a model austenitic steel. 490-493. K. Lemaitre. I. M. Again. W pracy wykazano.. 2011. Łuksza. Acta Materialia. P. The effect of strain path on material behaviour during hot rolling of FCC metals. 2012 – 263 – COMPUTER METHODS IN MATERIALS SCIENCE .. Podejście przedstawione w niniejszej pracy procesu zostało zastosowane do modelowania dwóch procesów przeróbki plastycznej charakteryzujących się zmienną drogą odkształcania: procesu cyklicznego odkształcania na drodze skręcania oraz procesu Kątowego Wielostopniowego Ciągnienia (KWC). B. 1990. 2010. S.Chaboche. 1999. 2004. Cybulka.P... Odpowiednia kontrola tych parametrów jest utrudniona i może być efektywnie optymalizowana jedynie w przypadku. L.

majchrzak@polsl. boundary element method 1 1. y )  Ta  n (1) where index e corresponds to the respective subdomains. Poland *Corresponding author: ewa. y )  T ( x . y )  λe  e 2    0. in particular where Γex is the external surface of domain marked in figure 1. y ) 2  1 (3) 264 – 268 ISSN 1641-8581 . at the same time the position of internal interface is unknown. Bankowa 8. y )  T ( x. respectively.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. In the final part of the paper the examples of computations are shown. 2 ( x. gradient method. 2013.pl Abstract The non-homogeneous domain being the composition of two sub-domains is considered.40-007 Katowice. INTRODUCTION The following boundary value problem is considered ( x. Silesian University of Technology. This approach is especially convenient in the case of boundary element method application (this method is used at the stage of numerical algorithm construction). To solve the inverse problem the gradient method is used. The steady temperature field in domain considered is described by two energy equations (the Laplace equations). α is the heat transfer coefficient. continuity condition given on the contact surface and the boundary conditions given on the external surface of domain. On the surface between sub-domains the continuity of heat flux and temperature field is assumed. y )   c :  T1 ( x. Key words: heat transfer. The equation (1) is supplemented by the typical boundary conditions. BOHDAN MOCHNACKI2 Institute of Computational Mechanics and Engineering. y )   λ2 2 λ1 n n   T ( x.  y2   x e  1. this means ( x. From the practical point of view the points X should be located at the external surface of the system. No. The additional information necessary to solve the identification problem results from the knowledge of temperature field at the set of points X selected from the domain analyzed. x. inverse problem. T. y )   e :   2T ( x. 13. The sensitivity coefficients appearing in the final form of equation which allows one to find the solution using a certain iterative procedure are determined by means of the implicit approach of shape sensitivity analysis. 2 IDENTIFICATION OF INTERFACE POSITION IN TWO-LAYERED DOMAIN USING GRADIENT METHOD COUPLED WITH THE BEM EWA MAJCHRZAK1*. y )  α T1 ( x. y )  2Te ( x. 44-100 Gliwice. ∂T1 /∂n denotes the normal derivative. y )   ex : (2)  λ1 T1 ( x. Poland 2 Higher School of Labour Safety Management. Ta is the ambient temperature. y denote the temperature and spatial co-ordinates. Konarskiego 18a. λe is the thermal conductivity.

. T1ex . . )   (10)  for sub-domain Ω2 q c 2   q3  4  2   G2   q in    H c 2 2  4   q2   Tc 2   T3    2 H4 2  T2in   4   T2   e  T (.  Tc1 . n = [nx. figure 1) the Dirichlet condition is taken into account Te* (. x. )   e : * e B (. Domain considered T24 . .f. while the position of Γc is unknown (Ciesielski & Mochnacki. As is well known. (x. x. y) and * qe (. SOLUTION OF DIRECT PROBLEM BY MEANS OF THE BOUNDARY ELEMENT METHOD The boundary integral equations corresponding to the Laplace equations (1) are the following (Brebbia & Dominguez. x. y) qe ( x. thermographs). 2. 2001) of functions T and q at the boundary Γ3Γ4Γin of domain Ω2 and then one has  for sub-domain Ω1 1 ex   G1 G1 2 G1 2 H1 (.g. 2010). q c 2 are the vectors of functions T and q on the contact surface Γc between sub-domains Ω1 and Ω2. 2012. y)de = e  qe (. . q1 1. Romero Mendez et al. ny] and Te*(. 2001) external surface ex  G eq e  H e Te (9)   c   Now.f. when the thermophysical and geometrical parameters appearing in the mathematical model of the process considered are given then the direct problem is formulated and the temperature distribution in the domain  can be found.. . The inverse problem considered here bases on the assumption that the temperature distribution at the boundary Γex is known (e. y ) d  (7) n 2r 2 (8) while d  ( x  ) nx  ( y  )n y In numerical realization of the BEM the boundaries are divided into boundary elements and the integrals appearing in equations (5) are substituted by the sums of integrals over these elements. After the mathematical manipulations one obtains two systems of algebraic equations (Majchrzak. q c1 . 1992 . Fig. y )  1 1 ln 2 e r (6) ( x. Tc 2 . 1) is the coefficient connected with the local shape of boundary. q 3 q4 q in 2. y )  Tb (4) On the remaining parts of boundary the no-flux condition can be accepted. x. y) = - e Te (x. y where r is the distance between the points (. y )  in : T2 ( x. . y) Te ( x. qe(x. x. the following notation is introduced (c. 2 are the vectors 2. y)de * (5)  G c 2 G3 2 G in 2 H3 2 H in 2 where B(. y )   e Te* (.INFORMATYKA W TECHNOLOGII MATERIAŁÓW On the internal surface in (c. x. y)/n. 2 ex are the vectors T12 . )(0. figure 1)  T11 .  x in of functions T and q at the boundary Γ1Γ2Γex of domain Ω1. )Te (. y) is the fundamental solution (11) The continuity condition (3) written in the form q c1  q c 2  q   Tc1  Tc 2  T (12) – 265 – COMPUTER METHODS IN MATERIALS SCIENCE  q1  1  ex  q1  H1 H ex G c1  1   q2     1 1    qc1    T11   ex  T1  H c1    T2  1   T   c1   . ).  T23 . () is the observation point. Majchrzak. q1 q1 . 1. T2in .

N (16) where A is the main matrix of the system of equations (13). …. the system of equations (14) should be differentiated with respect to parameter bj and then A Y B Y+A  b j b j b j or (20) – 266 – . j 1 U ik. one obtains  T11   ex  T1   T12    ex 0   T   αG 1 Ta      4  in  G 2 Tb  H 2  q   3  T2   q in   2 4   T2     H1 1  0  ex ex αG1  H1 0 2  H1 0  H c1 H c 2 G c1 G c 2 0 H3 2 0 G in 2 (13) or AY  B (14) S 2   bn M  (T i 1 M i  Td i )  Ti  0. . yn). In this paper the real measurements are substituted by the temperatures Ti obtained from the direct problem solution for arbitrary assumed position of points (xn. n  1. bN )  1 M  (T i 1 M i  Td i ) 2 (15) where Tdi .. 1997. Burczyński. Here the implicit differentiation method.missing’’ boundary values. .. So. . Majchrzak et al. N (18) where  T U ik. . From the system of equations (18) the values bjk+1 are calculated. Knowledge of nodal boundary temperatures and heat fluxes constitutes a basis for computations of internal temperatures at the optional set of points selected from the domain considered. (11). . b2. 2. bN which correspond to the co-ordinates yn shown in figure 2. 3. The criterion which should be minimized is of the form (Kurpisz & Nowak. Taking into account the remaining boundary conditions finally one obtains Using the necessary condition of optimum.. SOLUTION OF INVERSE PROBLEM USING GRADIENT METHOD COUPLED WITH THE BEM The function Ti is expanded into the Taylor series taking into account the first derivatives N  T Ti  Ti k    i  j 1   b j  (b k 1  b k  j ) (17)  k j b j  b j where bj0 is the arbitrary assumed value of parameter bj. To determine the sensitivity coefficients the methods of shape sensitivity analysis are used (Kleiber. n (bk  bk j j )  COMPUTER METHODS IN MATERIALS SCIENCE The inverse problem considered here bases on the assumption that the temperature distribution at the boundary Γex is known. are the sensitivity coefficients. 2. j   i   bj     k b j  b j (19) S (b1 . while the position of Γc is unknown. n  0. Freus et al.. which belongs to the discretized approach.. . The aim of investigations is to determine the values of shape parameters b1.. basing on the differentiation of algebraic boundary element matrix equations (14) is applied. Introducing (17) into (16) one has  U i 1 j 1 M M N k i. . 2003)  T i 1 di  Ti k U ik. The system of equations (14) allows one to find the . 2012). yn).. N. Y is the vector of unknowns and B is the free terms vector. The surface Γc is defined by the set of points (xn. .  bn n  1. . 2011. n = 1. respectively. …. 1995. while for k > 0 it results from previous iteration. . 2. Ti are the temperatures known from the measurements and calculated ones..INFORMATYKA W TECHNOLOGII MATERIAŁÓW allows one to couple the equations (10). bn .

2011). 29 shape sensitivity coefficients corresponding to the y co-ordinate of nodes from 29 = 68 to 47 = 50 (figure 2) has been distinguished.011 m (the different start points allow ones to observe the course of iteration process). boundary temperature Tb = 37oC (c. (22) where (L. 2.f. At first. λ2 = 0. 3. 0. condition (4)). 8 L( x  L) 2  y p x ( x  2 L) L2 Fig.INFORMATYKA W TECHNOLOGII MATERIAŁÓW A  Y  B A   Y b j b j b j (21) It should be pointed out that the derivatives of the boundary element matrices are calculated analytically (Majchrzak et al. while figure 4 illustrates the course of temperature at the external surface. It is visible that – 267 – COMPUTER METHODS IN MATERIALS SCIENCE .02 m. the direct problem described in the chapter 1 has been solved. In figures 5 and 6 the results of computations are shown.02 m) shown in figure 1 has been considered.012 m) is the tip of parabola. Temperature distribution in the domain considered . 5. condition (2)). Fig. RESULTS OF COMPUTATIONS The domain of dimensions 2LL (L = 0.. 4. yp) = (0. 29 additional problems connected with the determination of sensitivity functions have been formulated. heat transfer coefficient α = 10 W/(m2K). To solve the inverse problem.015 m or yp = 0. ambient temperature Ta = 20oC (c. The discretization of boundaries using the linear boundary elements is shown in figure 2. equation (20).2 W/(mK).1 W/(mK). So. Temperature distribution at the external surface Fig. Discretization of boundaries In figure 3 the temperature distribution in the domain considered is presented. The following input data have been introduced: thermal conductivities λ1 = 0.f. Fig.f. The shape of internal surface Γc has been assumed in the form of parabolic function (in this place the optional shapes can be taken into account) y ( x)  0. 4. Results of identification – variant 1 The identification problem has been solved under the assumption that the temperatures at the nodes from 5 to 23 (figure 2) are known and the initial position of internal boundary is described by function (22) where yp = 0. The nodes 28 and 48 are fixed – c.

S. K. 1. 2003.. J.J. 151-162. Beskos. Czestochowa (in Polish). Model matematyczny procesu tworzy układ równań eliptycznych (równań Laplace’a). 21-38. Freus.. warunki idealnego kontaktu na powierzchni kontaktu i warunki zadane na powierzchniach zewnętrznych. Nowak. Austria. Computational Mechanics Publications. M.. ner. Ciesielski. które jest szczególnie efektywne w przypadku zastosowania metody elementów brzegowych (tę metodę wykorzystano na etapie konstrukcji algorytmu numerycznego).N. eds. Rozwiązanie zadania uzyskano metodą gradientową.Wiley & Sons Ltd. przy czym położenie powierzchni granicznej nie jest znane. Dominguez J. A. W końcowej części artykułu zamieszczono wyniki obliczeń numerycznych. Analytical solution of a Pennes equation for burn-depth determination from infrared thermographs. 27. Freus. The algorithm proposed allows one to identify the complex shapes of internal boundary (the co-ordinates yn are estimated separately). J. 2012. Received: October 2. Implicit approach using boundary element method. CD-ROM Proceedings of the 6th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS 2012). G.. Identification of internal boundary position in two-layers domain on the basis of external surface temperature distribution. a współczynniki wrażliwości występujące w układzie rozwiązującym wyznaczono wykorzystując niejawne podejście analizy wrażliwości. In a similar way the 3D problems can be also solved.. 2012 Received in a revised form: October 22. Shape sensitivity analysis. in the medical practice (estimation of wound shape on the basis of surface temperature distribution). E. B. Results of identification – variant 2 6. Vienna University of Technology.. Fig. Scientific Research of the Institute of Mathematics and Computer Science. Boundary elements. J.. E. T. Scientific Research of Institute of Mathematics and Computer Science.. of the Czestochowa University of Technology. M. 2012 COMPUTER METHODS IN MATERIALS SCIENCE REFERENCES Brebbia. Sensitivity analysis. Maier. 1. D. H.. 6. M..J. Boundary element advances in solid mechanics. among others.G. 2001. Vien.. CONCLUSIONS The algorithm proposed can be useful.. WYKORZYSTANIE METODY GRADIENTOWEJ I MEB DO IDENTYFIKACJI KSZTAŁTU GRANICY MIĘDZY PODOBSZARAMI W DWUWARSTWOWYM NIEJEDNORODNYM OBSZARZE CIAŁA STAŁEGO Streszczenie W pracy rozpatruje się niejednorodny obszar ciała stałego będący złożeniem dwóch podobszarów. 2012. 2010. Rammerstorfer. Publ. Vienna.A.. Dodatkową informacją pozwalającą rozwiązać sformułowane w ten sposób zadanie odwrotne są wartości temperatury w punktach X wyróżnionych w rozpatrywanym obszarze.INFORMATYKA W TECHNOLOGII MATERIAŁÓW for exact input data the exact position of boundary is obtained and the iteration process is convergent. Freus. Freus. Springer Verlag. Mochnacki. K.. Chichester. eds.. In the future the detailed research of iterative procedure convergence should be also done. CMP... Bohm. K. an introductory course.J. C. 245-307.. Kurpisz. F. 1995. 2012 Accepted: November 9. 2011. 15-22. SouthamptonBoston.. Burczyński..Boundary element method in heat transfer. Sen. Inverse Thermal Problems. Numerical analysis of interactions between skin surface temperature and burn wound shape. but taking into account the practical applications it seems that the scientific research in this scope should be continued.. E. 1992. Z praktycznego punktu widzenia punkty przyłożenia sensorów powinny być zlokalizowane na powierzchni zewnętrznej pozostającej w kontakcie z otoczeniem.. Majchrzak... Romero Mendez. London. Parameter sensitivity. B. Eberhardstei- – 268 – . Kleiber. Majchrzak. Jimenez-Lozano.. New York. Mathematical Medicine and Biology. R. optimization and inverse problems. 1997. S. Majchrzak. Mochnacki. It should be pointed out that both from the mathematical and numerical points of view the problem is rather complicated. F. McGraw-Hill Book Company. Gonzalez.

COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. Malinowski et al. 2 INFLUENCE OF THE SAMPLE GEOMETRY ON THE INVERSE DETERMINATION OF THE HEAT TRANSFER COEFFICIENT DISTRIBUTION ON THE AXIALLY SYMMETRICAL SAMPLE COOLED BY THE WATER SPRAY AGNIESZKA CEBO-RUDNICKA*. The proper cooling rate affects the final mechanical properties of products which strongly dependent on microstructure evolution processes. ISSN 1641-8581 269 – 275 . TADEUSZ TELEJKO Faculty of Metals Engineering and Industrial Computer Science. the comparison between two different models employed to determine the heat transfer coefficient over the cooled sample surface have been presented. 13. INTRODUCTION In the metal industry the water cooling is widely used to control the product temperature variation in the production process. The first one. No. boundary inverse problem. The heat transfer boundary condition in case of water cooling is defined by the heat transfer coefficient (HTC).pl Abstract The paper presents the results of the heat transfer coefficient determination while the water spray cooling process. the direct measurements of the HTC by such methods as mass transfer or transient method that uses liquid crystals to measure the surface temperature cannot be used in the case of steel industry processes (Mascarenhas & Mudawar. two finite element models have been tested in the inverse determination of the heat transfer coefficient. Numerical simulations can be used to determine the water flux which should be applied in order to ensure desired product temperature. Also. The boundary condition models differ in description of the function that has been employed to approximate the heat transfer coefficient distribution over the cooled surface in the time of cooling. The main goal in these case is to ensure sufficient heat transfer from the ingot surface to achieve a proper solidification structure. 2012). Department of Heat Engineering and Environment Protection. Mickiewicza 30. finite element method 1. 1985. BEATA HADAŁA. 2010. which simplifies the sensor geometry to a cylinder and the second one. 30-059 Kraków. In the case of strip rolling mills the main cooling system is situated at run-out table to ensure the required strip temperature before coiling (Tacke et al. Continuous casting lines are equipped with the water spray secondary cooling zones. The industrial hot rolling mills are equipped with systems for controlled cooling of hot steel products. Because of the specific setup of the sensor used in investigations. ZBIGNIEW MALINOWSKI.. Also.edu. Poland *Corresponding author: cebo@agh. In the investigations the axially symmetrical sample has been used as a cooled object. that describes the real shape of the sensor. heat transfer coefficient. To determine the boundary condition over the metal surface cooled by water spray the inverse heat conduction problem has been used. Al. Key words: water spray cooling. Due to complex nature of the cooling process the existing heat transfer models are not accurate enough in the case of high temperature processes common in metal industry.. 2013. AGH University of Science and Technology.

1. HTC can be determined as a function of cooling parameters and product surface temperature. where: pi is the vector of the unknown parameters to be determine by minimizing the objective function. flange and housing has been presented. . temperature measured by the sensor m at the time τn. The space between the cylinder and the housing has been filled with air. However. c – specific heat. In the paper the results of the inverse calculation of HTC have been presented. Fig. In inverse algorithm various heat conduction models and boundary condition models can be implemented. NiCr . that allows to reduce the heat losses to the surrounding. The experimental tests have been performed for two materials. The experimental investigations have been conducted for two materials: inconel and brass. The calculations have been performed on the basis of temperature measurements inside selected points of axially symmetrical sample cooled by water spray. qv – internal heat source. . Descriptions of the model with linear shape functions has been presented in the paper of Gołdasz et al.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Liu et al. 2012). the flange which allows to join the cylinder and housing. . The heat transfer boundary condition on the cooled surface of the metal cylinder has been expressed as a function of surface radius and time: . (2009). as well as housing have been made from the same material. . 4 and 6 mm from the cooled surface. Schematic illustration of the experimental setup employed for the determination of the heat transfer boundary condition. λ – thermal conductivity. 2. τ – time. In the first model (model A) average HTC over the cooled surface as a function of the time of cooling and average sample surface temperature has been determined. The sample has a form of a cylinder 20 mm in height and 20 mm in diameter. Np –number of the temperature measurements performed by one – the sample sensor in the time of cooling. Thermocouples have been placed in the symmetry axis of the cylinder in the distance of: 2. ρ – density. h – heat transfer coefficient. . r. causes that the sample temperature field is not perfectly one dimensional. The top surface of the cylinder has been cooled by the water sprays.NiAl thermocouples have been used. 3. z – cylindrical coordinates. There. In the second model (model B) HTC distribution over the cooled surface has been approximated by the witch of Agnesi type function with the expansion in time of the HTC parameters Cebo-Rudnicka et al. PROBLEM FORMULATION In the present study the boundary condition over the surface of the metal sample cooled by water spray has been sought. which differ substantially in thermal – 270 – . – heat flux. The cylinder with flange. . (3) where: Ts – cooled sample surface temperature. The cylinder has been completed with a flange 30 mm in diameter and 1 mm thickness and it has been placed in cylindrical housing. For such processes the best way to determine the HTC is to formulate the boundary inverse heat conduction problem (IHCP). In figure 1 the schematic illustration of the experimental setup which consists of the cylindrical sensor. (2012). Nt – number of the temperature sensors. In the finite element model employed to solve equation (2) linear shape functions have been used. To measure the temperature inside the cooled sample three fast response. . Ta – cooling water temperature. BOUNDARY INVERSE MODEL The HTC on the cooled surface of the cylinder can be determined from the inverse solution to heat transfer problem by minimizing the objective function defined as: ∑ ∑ (1) Variation of the heat transfer coefficient h at the metal surface in time of cooling has been approximated by two HTC models. – the sample temperature at the location of the sensor m at the time τn calculated from the finite element solution to the heat conduction equation: . COMPUTER METHODS IN MATERIALS SCIENCE 0 (2) where: T – temperature..

low heat conductivity of inconel causes that heat transfer to the flange is low and does not influence the average HTC in these two boiling regimes. 4. The heat transfer in radial direction to the flange increases while the surface temperature decreases. Because of the sensor construction two finite element models have been tested in the inverse determination of the heat transfer coefficient. Below 250°C (figure 3) the heat transfer process changes to the nucleate boiling. In such a case heat transfer to the flange is much more important and the exact description of the cooled sample geometry plays an important role in the HTC identification. In case of water spray cooling of inconel sample.INFORMATYKA W TECHNOLOGII MATERIAŁÓW conductivity. In figures 2 to 5 the comparison between HTC variations in the cooling process calculated for simplified and exact description of the sample geometry in the finite element model have been presented.795 7.735 5. Inconel and brass samples have been selected for the study. two boundary condition models have been employed in equation (3) in order to determine the heat transfer coefficient on the sample surface. The average difference between the calculated and measured temperatures has been equal to 7. The water spray pressure in both testes was 1 MPa and water temperature was equal to 20°C. The first finite element model simplifies the sample geometry to the perfect cylinder (the simplified model).953 3. The thermal conductivity of the brass is much higher than the inconel one. Further. the simplification in sample geometry description to the perfect cylinder does not effects the average HTC for the mean sample surface temperature from 730°C to about 250°C (figure 3). Table 1. RESULTS OF INVESTIGATIONS The results of the inverse calculations have allowed to determine the influence of the sample geometry description on the heat transfer coefficient identification. The temperature measurements logged while experimental tests have been assumed as an input data in inverse calculation of HTC. °C 7. The figures present variations in the average values of HTC (boundary condition model A) versus the time of cooling (figures 2 and 4) and versus the average sample surface temperature (figures 3 and 5). That results in significant increase in the HTC values. Therefore the sample geometry simplification in the finite element model to the perfect cylinder results in the HTC values about 10 percent higher if compared to those obtained with the real sample geometry description (with flange) in the finite element model of heat transfer (figures 2 and 3). Inconel sample Case of study Average difference in temperatures. The initial temperature to which the materials have been heated up was 730°C for inconel and 517°C for brass.6 kg/(m2·s) while cooling inconel sample and 1 kg/(m2·s) for cooling the brass sample. The average difference between measured and calculated temperatures at the thermocouples locations. During that processes the vapor film is formed on the cooled surface and it limits the heat transfer between the cooled surface and the cooling water. Neglecting the flange in the definition of the sample geometry has caused that the calculated values of HTC in the whole spray – 271 – COMPUTER METHODS IN MATERIALS SCIENCE . These to processes affect the inverse determination of HTC.95°C and has not decreased for the better definition of sample geometry (table 1).735 3. This range of temperature corresponds to the film and transition boiling regimes that take place on the sample surface while water spray cooling process.418 Average HTC over the cooled surface calculated for simplified definition of the sample geometry in the finite element model Average HTC over the cooled surface calculated for exact definition of the sample geometry in the finite element model Radial distribution of HTC over the cooled surface calculated for simplified definition of the sample geometry in the finite element model Radial distribution of HTC over the cooled surface calculated for exact definition of the sample geometry in the finite element model 7. The water flux was 38.953 Brass sample Average difference in temperatures. °C 4. Additionally.752 The inverse calculations performed on the basis of temperature measurements obtained for the spray cooling of brass sample have indicated a significant influence of the sample geometry description on the average HTC values (figures 4 and 5).924 7. In the case of the second model the cylindrical sample and the adapter ring (flange) have been described by the finite element mesh (the exact model). Simultaneously heat conduction in radial direction is more significant.

20000 perfect cylinder cylinder with flange 16000 HTC. Brass sample. Brass sample. Such a possibility gives the second boundary condition model. Exact description of the sample geometry (with flange) in the finite element model has resulted in lower of about 10% values of HTC (figure 8). Due to axially symmetrical problem only radial variation of HTC in the time of cooling has been modeled.K) 12000 20000 8000 10000 4000 0 0 0 4 8 Time.K) perfect cylinder cylinder with flange 30000 HTC.oC 500 600 Fig. Inconel sample. s 12 16 0 5 10 15 Time. 20000 perfect cylinder cylinder with flange 16000 HTC.K) 30000 HTC. 40000 perfect cylinder cylinder with flange Fig.INFORMATYKA W TECHNOLOGII MATERIAŁÓW cooling process have been greater for about 25 percent than the ones calculated by using the model with the flange (the exact geometry model) (figure 4 and 5). W/(m2. The results of the inverse calculation of HTC distributions as functions of the sample radius and the time of cooling have been presented in figures to 9 for inconel sample and in figures 10 to 11 for the brass sample. W/(m2. The analysis has also been performed for two materials: inconel The inverse solution to HTC distribution along the cooled sample radius performed for the inconel with simplified definition of the sample geometry in finite element model has not indicated a visible differences in HTC along the radius of the cooled sample surface (figure 6 and 7). 4. 2. – 272 – . In practice it is expected HTC distribution over the cooled surface.oC 800 100 200 300 400 Temperatrure. The average HTC variations as a functions of sample surface temperature obtained for the simplified and exact definition of the sample geometry in the finite element model. The comparison of the average HTC variations in the time of cooling obtained for the simplified and exact definition of the sample geometry in the finite element model. Moreover. The comparison of the average HTC variations in the time of cooling obtained for the simplified and exact definition of the sample geometry in the finite element model.K) 12000 20000 8000 10000 4000 COMPUTER METHODS IN MATERIALS SCIENCE 0 0 0 200 400 600 Temperatrure. 40000 and brass. 5. in the case of brass sample the exact definition of the sample geometry (with flange) has lead to the lower difference between measured and calculated temperatures at the thermocouple locations (table 1). 3. The discussed above boundary condition model gives only average HTC over the cooled sample surface. Fig. simplified and exact description of sample geometry in the finite element model has been considered. The average HTC variations as a functions of sample surface temperature obtained for the simplified and exact definition of the sample geometry in the finite element model. W/(m2. W/(m2. Further. Inconel sample. s 20 25 Fig.

HTC variation versus surface temperature for selected locations along the cooled surface radius calculated for exact definition of the sample geometry in the finite element model. 9. HTC model B. HTC model B.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 40000 40000 r = 1mm r = 5 mm r = 7 mm r = 10 mm 30000 30000 HTC. 7. HTC variation versus surface temperature for selected locations along the cooled surface radius calculated for simplified definition of the sample geometry in the finite element model. In the case of simplified definition of the sample geometry to the perfect cylinder in the finite element model the HTC values along the radius of the sample decrease. In the case of cooling brass sample the diversification of HTC along the cooled surface radius has been observed both for simplified and exact description of the sample geometry in the finite element model (figures 10 and 11). oC 400 600 800 Fig. 40000 30000 20000 10000 r = 1mm r = 5 mm r = 7 mm r = 10 mm 0 0 4 Time. W/(m2. It can be explained by the high water flux applied in the cooling of inconel sample. HTC variation versus time of cooling for selected locations along the cooled surface radius calculated for simplified definition of the sample geometry in the finite element model.K) r = 1mm r = 5 mm r = 7 mm r = 10 mm 20000 10000 10000 0 0 4 0 Time. Same difference of HTC distribution versus sample surface temperature for the HTC model B has been observed only at the sample and flange connection (r = 10 mm in figure 9). Inconel sample. 6. 8. HTC model B. s 8 12 16 0 200 Temperature. Implementation of the HTC model which allows for the distribution of heat transfer coefficient resulted in very similar solutions to the average HTC model. 30000 20000 10000 0 0 200 Temperature. oC 400 600 800 Fig. 40000 r = 1mm r = 5 mm r = 7 mm r = 10 mm Fig. In these case the difference between the maximum values of HTC is equal to about 17 kW/(m2·K) (fig- HTC. Inconel sample. W/(m2. In such a case sample surface is cooled uniformly. The greatest difference between the maximum HTC value in the cylinder axis and at the distance of 10 mm from the symmetry axis is equal to about 9 kW/(m2·K) (figure 10).K) .K) – 273 – COMPUTER METHODS IN MATERIALS SCIENCE HTC. Implementation of the exact definition of the sample geometry and the HTC variation over the sample surface in the finite element model has allowed to illustrate both the influence of thermal conductivity of sample material as well as the influence of the cylinder flange on the heat transfer between cooled sample and water spray. Inconel sample. Implementation of the exact definition of the sample geometry results in much higher diversification of HTC values. W/(m2. Inconel sample. HTC model B. HTC variation versus time of cooling for selected locations along the cooled surface radius calculated for exact definition of the sample geometry in the finite element model. s 8 12 16 Fig. It can be explained by the better description of the sample temperature near the flange by the exact geometry model.K) 20000 HTC. W/(m2.

oC 400 600 Fig. Cebo-Rudnicka. 2012. Gliwice – Wrocław. Institute of Thermal Technology. 54. The developed definition of the boundary condition is capable of identification both constant and variable heat transfer coefficient over the cooled surface of the cylindrical sample. HTC model B. Hadała. 35-54. HTC variation versus surface temperature for selected locations along the cooled surface radius calculated for exact definition of the sample geometry in the finite element model. 2009. 10. 2010. 2012... 53.INFORMATYKA W TECHNOLOGII MATERIAŁÓW ure 11). G. E.. Identification of HTC performed for two boundary condition models has shown that allowing for the HTC distribution over the cooled surface results in more accurate determination of the heat transfer boundary condition. Telejko T.A. 5871-5883. W/(m2. Heat and Mass Trans. and Raquest. Implementation of the axially symmetrical and three dimensional finite element models to the determination of the heat transfer coefficient distribution on the hot plate surface cooled by the water spray nozzle. Sothwick. REFERENCES 10000 5000 0 0 5 10 Time. H. N. 504-506.. Pennsylvania. Brass sample. Mascarenhas. Correcting lateral heat conduction effect in image-based heat flux measurements as an inverse problem.. R. Wang... Z. 25000 r = 1 mm r = 2 mm r = 3 mm r = 4 mm r = 5 mm r = 6 mm r = 7 mm r = 8 mm r = 9 mm r = 10 mm 20000 HTC. Study of heat balance in the rolling process of bars. Hadała. In both considered cases of cooling the metal samples exact definition of the sample geometry (with the flange) in the finite element model have resulted in lower average differences between measured and calculated temperatures (table 1).. B. Sullivan.. 685-694. Rubal.1244-1258. Nowak. A.. s 15 20 25 Fig. P. Grant No NR15 0020 10. J. CONCLUSIONS The conducted analysis has allowed to determine the influence of the exact definition of the cooled sample geometry in the finite element model on the solution to the heat transfer process. T. 2012. Liu..D.. Int. 11. J.. Heat and Mass Trans. 5.. Proceedings of Numerical Heat Transfer 2012 International Conference.. Z. The work has been financed by the Ministry of Science and Higher Education of Poland. Z. Hadała. Pittsburgh. Malinowski. HTC model B. 30000 r = 1 mm r = 2 mm r = 3 mm r = 4 mm r = 5 mm r = 6 mm r = 7 mm r = 8 mm r = 9 mm r = 10 mm HTC..P.. J. The Metallurgical Society of AIME. KEM. – 274 – . It has been shown that simplification of the sample geometry to the perfect cylinder in the finite element model results in about 10 percent growth in the heat transfer coefficient determined by the inverse method in case Cebo-Rudnicka. 58-67.... Malinowski. Analytical and computational methodology for modeling spray quenching of solid alloy cylinders.K) 20000 10000 COMPUTER METHODS IN MATERIALS SCIENCE 0 0 200 Temperature. 1985. eds.K) 15000 of material which is characterized by low heat conductivity (inconel) and about 25 percent growth in HTC value in the case of high conductivity materials (brass). Proceedings of a Symposium on Accelerated Cooling of Steel.. Tacke. Mudawar. A. Implementation of the finite element model with linear and Hermitian shape function to determination of the heat transfer coefficient distribution on the hot plate cooled by water spray. T. Investigation into the Efficiency of Cooling System for Wide-Strip Hot Rolling Mills and Computer-Aides Control of Strip Cooling. Malinowski.. A. Silesian University of Technology. Białecki. I. Telejko. Acknowledgements. B. Litzke. Archives of Metallurgy and Materials. Int. 54. A. B. Gołdasz.. eds. B. Brass sample.J. 1055-1060. W/(m2. J... HTC variation versus time of cooling for selected locations along the cooled surface radius calculated for simplified definition of the sample geometry in the finite element model.

2012 Accepted: October 29. Ze względu na specyficzną budowę czujnika wykorzystanego w badaniach. Pierwszy model upraszczał geometrię próbki do postaci „zwykłego” walca.INFORMATYKA W TECHNOLOGII MATERIAŁÓW WPŁYW GEOMETRII PRÓBKI OSIOWOSYMETRYCZNEJ NA WYZNACZANIE ROZKŁADU WSPÓŁCZYNNIKA WYMIANY CIEPŁA PODCZAS CHŁODZENIA NATRYSKIEM WODNYM Streszczenie W pracy przedstawiono wyniki obliczeń współczynnika wymiany ciepła wyznaczonego na podstawie badań eksperymentalnych. Badania eksperymentalne przeprowadzono dla próbki osiowosymetrycznej. 2012 – 275 – COMPUTER METHODS IN MATERIALS SCIENCE . W pracy testowano również dwa modele aproksymacji warunku brzegowego. w algorytmie metody odwrotnej przetestowano dwa modele elementów skończonych opisujące geometrię próbki. Do wyznaczenia warunku brzegowego na powierzchni metalu chłodzonego natryskiem wodnym wykorzystano rozwiązanie brzegowego odwrotnego zagadnienia przewodzenia ciepła. Received: September 17. 2012 Received in a revised form: October 24. drugi model opisywał rzeczywisty kształt próbki.

2013. The main controlling operations can be also implemented to obtain the spheres outer geometry. The metallurgical method exhibit all known methods in intelligence. spherical shells. The property of the product changes through the microstructure development results from the thermal energy consumption during the primary crystallization and further heat treatment under different regimes. Current investigation was completed by the determination of the whole temperature fields on the side surface of the form nozzle. Department „Electrical. thermography. Russia 3 Moscow State Industrial University (MSIU). to determine the preferable boundary conditions. CFD. Fluent. heat transport 1 1. Department „Car Body Making and Metal Forming“. Avtozavodskaya street 16. 09596 Freiberg. To one of them belongs the investigation of the temperature fields during the formation process. At the end the data verification of the obtained simulation results with the measurements on the laboratory stand and theoretical calculations were carried out. Russia *Corresponding author: petroffma@gmail. 30. 13. JUERGEN BAST1. 115280 Moscow. 107023 Moscow. Institute of Machine Elements. Key words: hollow spheres. fusion targets.com Abstract The present paper reveals one of the energy efficient ways of the units (hollow spheres) production for cellular structures for their further application in light weight constructions. Germany 2 Moscow State University of Mechanical Engineering (MAMI).COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. For these purposes the CFD. No. Design and Manufacturing. The heat transport from the tin melt into the semi-solid tin shell influence the nucleation process so the solid wall should be formed before the gas starts to form the inner hollow space. strength/ductility and inner geometry. Heating and Hydraulic Engineering and Energy machines“. Pure tin shells were produced intransient (thixotropic) state of materials by elevated temperatures (close to the melting point of the pure tin) and several simulation steps were used. PAVEL PETROV2. FEM. Solid Works. illustrated in figure 1.(Computational Fluid Dynamic) and FEM-commercial codes such as FLUENT and Solid Works Simulation Package respectively were taken. These small hollow spheres made from different materials could change the weight of a construction part essentially. tin melt. INTRODUCTION Generally any production route consists of one or several production. Leipzigerstr. ANATOLY SHEYPAK3 Technical University Mining Academy of Freiberg. Otherwise the semi-solid shell will be broken by the gas pressure or the inner hollow space does not occur. Once the products were sorted the structure as- 276 – 282 ISSN 1641-8581 . They can be used as a unit cell for big parts and alone filled with an inert gas. used as acoustic and thermal insulation and also as protection against vibrations. not involving any organically core and preparation of the powder and slurries. treatment and controlling operations which are connected together through automation devices. Semenovskaya street 38. For hollow spheres production the metallurgical technique. was used because of its high effectiveness and low production costs. B. 2 INVESTIGATION OF THE HEAT TRANSPORT DURING THE HOLLOW SPHERES PRODUCTION FROM THE TIN MELT MICHAEL PETROV1. because it based on the own physical properties of the used materials and boundary conditions of the process.2. e.g. which was obtained from a thermogram captured with the help of an infrared camera (IRC). realized through the metallurgical procedure.

protecting gas against oxidization procedure). Prepared model for simulation (a). To perform the numerical simulation the CAD-models of heating devices inclusive crucible were optimized before meshing (fasteners connections were closed. To enable the simulation the following boundary conditions were assigned and are presented in the table 1.INFORMATYKA W TECHNOLOGII MATERIAŁÓW sembling occurs depending on the application cases. Temperature distribution in the system During the simulation setup of the temperature distribution in the whole system (furnace – heating plates – nozzle) three main heat transport mechanisms (thermal conductivity. The lifetime of the bridge corresponds to the formation time of a single shell. To obtain adequate results different size of the mesh elements were applied: the biggest of 0. – 277 – COMPUTER METHODS IN MATERIALS SCIENCE . 2012). 2. dosing and controlling of the solidification process occur in a special heating unit. 2.2 mm for the heating furnace with the common side ratio of volume tetraeder elements of 1. The whole metal melting process runs in three stages: 1) preparing the metallic melt (optionally: alloying. tempered by heating rods. The simulated hollow sphere has a transient heating bridge. sharply corners were rounded and the fiber thermal insulation was proposed as a material with homogeneous properties).5. 1. It was assumed that the heating bridge elevates the temperature in the spherical shell. convection and radiation) were activated. 3) process initiation: forming gas is fed to the tin melt through the gas needle and forms the spherical shells at the nozzle. which was made in a „sandwich” like assembly from copper and aluminum plates. EXPERIMENTAL EQUIPMENT Designed and manufactured labor stand for metal hollow spheres and shells production from the tin melt was investigated.5 mm (d). Also many fundamental aspects on the heat transport given by e.1.2. Petrov. 3. Dorogotowcev & Merkulyev (1989) several differences and resulted setups of the heat transport problems were not still numerical investigated. but the temperature difference stays the same. Baehr & Stephan (2006) should to be coupled to the manufacturing route and equipment.g. 2010. 2) switch on the heating devices (furnace. shown in ume(c) and hollow sphere of 3 mm in diameter and the wall thickness of 0.9 mm for the nozzle and 22. It is expected that an additional heat amount will transferred into the spherical shell. The current paper is focused on the first two steps in this line (metal melting and hollow sphere/shell solidification). 3. heating unit). degassing. figure 3. NUMERICAL SIMULATION Although the similar production technique was discussed by Kendall (1981). 3. Model preparation Fig. The melting. which connected the sphere with the main tin melt volume in the nozzle area as it is shown in figure 2d. Further description of the equipment could be found elsewhere (Petrov & Bast. model of the crucible (b) with the tin melt vol.bution in the crucible. The simulation results delivered a homogeneous temperature distriFig. General process route of hollow spheres/shells production. It allows adjusting the temperature in a narrow range and performing the nozzle tempering with a high precision.

e. In the aluminum upper and transported heat energy is constant through a period lower blocks the gravure of the nozzle was milled. Giving the heat energy from the heating a „sandwich” like assembly and consists of an upper rods to the colder parts the main amount of it riches and lower blocks. that the amount of the cast aluminum alloy. 4. according tion reduce the calculation error (compared to the the route in figure 4. From this analysis it was stated. measured value from the table 2) up to 8. placed in the upper copper plate. So the theoretically. the biggest element of 10. Temperature distribution in the heating tetraeder elements.2 mm. plates (mesolevel) side ratio of 1. at the nozzle. i. ϑ – temperature difference.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 1. the in situ measurements were carried out to justify the temperature distribution in the plates was calculated proper choice of the boundary conditions. was to determine the expected temperature fields on obtained with the help of thermocouple Ni-CrNi the microlevel. To perform the simulation Fig. Temperature fields in the longitudinal cross section of the nozzle (a) and heating equipment (b). Construction stages of the heating plates. of time and enough to guarantee the forming process After that the plates were mounted together. At the same time the capacity of the of the fact. The results from the transient heat transport fluids and inside the heating system was investigated simulation were compared with the temperature on and represents the mesolevel of the system. Theoretical calculation was based on the total capacity of the heating rods.5. Element Furnace Heating plates Inductor Heat source initial temperature on the refractory lining initial temperature on each heating rod or capacity per rod initial temperature on the refractory lining nects the volume and thermal material properties of the plates (Petrov.3. The heating unit is (type K). t t (1) where cCu and cAl – specific heat capacity of copper COMPUTER METHODS IN MATERIALS SCIENCE and cast aluminum alloy. that this heating unit will be placed in the heating rods determine also the heating time. 3. Several space with a variable environmental temperature. It was stated. the mesh parameters were defined as follows: volume 3. The goal the heating rod. mCu and mAl – mass of the copper and aluminium plates. P mCu  cCu   m Al  c Al    . 2012). after that compared with the simulation additional heat transfer mechanism. As a merit of the numerical simulation accuracy the duration of the heating stage was The simulation of the heat transfer between the taken. produced from pure copper and the nozzle. Fig. The equation (1) con– 278 – .4%. Because discontinuity. namely convecresults and validated with the help of IRC. that the thermal resistance (temperature drop or heat loss) due to plates stacks and heat loss due to convection and a little amount of the radiation does not exceed 5%. Boundary conditions in the simulated system.

defined as   1    with ξ′ as a drag coefficient. and together with the energy equation: . obtained for different simulation cases Heating stage. (2) cp cV – isentropen exponent. 5. The aim of the test is to find out the cone height from the simulation results which corresponds to the real cone height from the verification test under the same boundary conditions and for the same time point. showed in figure 5b and 5c. HR + HC + T+C where TR – transient heating process. during the simulation and theoretical investigations followed to a temperature of 16 – 25°C and 13 – 24°C respectively. that even the test chamber temperature of 232°C (melting point of the pure tin) by investigated gas flow rates of an average value of 750 liter per hour does not allow to eliminate the undercooling effect on the displacement of 1. HR – heat radiation. under the assumption. HR + HC + T TR. This undesirable effect leads to the process discontinuity due to metal solidification between two periods of hollow spheres formation. Measurements with IRC has shown. Heating time.th where    po   Ti  p    i  1  . „o“ for outside and „th“ for theoretical) The true temperature due to gas expansion was calculated from the equation (3). where the greater value corresponds the greater heat loss in the system. cp(v) – isobare (isochore) specific heat capacity and po – critical pi pressure ratio (po – pressure value after gas expansion and pi – pressure value in the pipeline). obtained in Solid Works. To . b – gas distribution in the tin melt. obtained with high speed camera. the gas distribution in the tin melt was still unknown. – 279 – COMPUTER METHODS IN MATERIALS SCIENCE An obtained temperature distribution for mesolevel could be applied also for a microlevel. Temperature distribution at the nozzle (microlevel) Moreover once the temperature influence was cleared. that the gas velocity exceeds the value of the velocity coefficient. Loycyanskiy (2003). 3.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 2. The temperature can be influenced by the forming gas expansion due to the differences of the cross sections of the pipeline and the feeding needle. Sheypak (2006) and other fundamentalists. obtained in FLUENT).4.5 mm from the needle top of the nozzle presented in figure 5a. The theoretical problem description can be found by Bohl (1991). To carry out the case study a special CFD-simulation and simple verification test were developed. C – convection. The rash temperature drops of millisecond duration results strong tin melt undercooling. T – constant temperature.c – simulation of the gas distribution in the tin melt. Because of a new fluid phase (forming gas) the new problem had to be defined. Through the simulation it could be shown. that the temperature difference between the heating plates and the melt at the nozzle orifice stays by 16°C. min : sec Experiment Simulation 14:30 16:55 15:30 Boundary conditions TR. T– temperature (index „i“ for inside. Because of the fact that the forming gas expands into the certain melt volume the expected temperature in the cross section could be calculated from the equation (2) as: Fig. °C from 0 up to 257 Time. Phase and temperature distribution during the forming gas injection (a – numerical simulation of the temperature distribution after the gas expansion in the hot chamber. 100 fps. HC – heat capacity. The principle of the test is the periodic gas injection (frequency of 1 Hz) into the tin melt and measurement of the height of the invader cone zone.

5 mm from the nozzle. The area above the distance of 6. r 1 i r .INFORMATYKA W TECHNOLOGII MATERIAŁÓW To  Ti   2 Ti  To . After the simulation the obtained temperatures from the middle radius of the shell were compared with the calculated from the equation (4). Temperature fields: temperature – distance diagram for determining undercooling effect (a).5 mm is out of the gas distribution zone and not intended into the investigation.5. 6. r – arbitrary radius of the sphere. Temperature distribution in the sphere’s wall (microlevel) The temperature distribution in the spherical shell occurs also in micro level of the investigated system and can be calculated on hand the simple equation (4) with sphere radius as argument.th  . Following the equation (4) the thicker walls propose the smaller temperature derivation from the linearity in The hollow spheres in figure 7 were previously meshed with volume tetraeders’ elements with the biggest size of 1 mm and side ratio of 1. (3) So for an environmental temperature (ET) of 17°C the temperature differences stay by 7°C for experiment and 6°C for numerical simulation. Taking the height of the involved cone from the figure 5b of 6. The expected temperature drop up to 6 – 7°C by initial ET of 17°C gives the distance of 5. – 280 – . shown in figure 6a. 3.5.5 mm on the diagram temperature-distance. „o“ for outer sphere surface and „th“ for theoretical). temperature distribution in the shell’s wall with different radius and thickness (b and c) and slope of the linear temperature distribution curves (d). figure 6c as thinner walls in figure 6b. For an initial ET of 232°C the same distance will give the temperature decrease up to 45°C. (4) Tth  T (r )  Ti  Ti  To  ri 1 ro where T – temperature (index „i“ for inner sphere surface. COMPUTER METHODS IN MATERIALS SCIENCE Fig. Self the line for the thin wall has a greater slope as the line for the thick wall that means that the temperature gradient for thin walls under the same boundary conditions should be greater to transport the heat amount from the melt to activate the solidification process.5 mm and over this value. two areas can be pointed out: up to the distance of 6.

3. λC and λL – thermal conductivity of the crystal and melt respectively. The necessary solidification rate (SR) of a solidification process can be calculated from the equation (5): where T – temperature. ISO-surfaces in 900 seconds of calculation time for a temperature distribution from 313°C and higher (b) and from 314°C and higher (c).2°C. differ from the temperatures on the other surfaces not more than on 1°C. 4. 2000 seconds (c). From the figures 6b and 6c the importance of the wall thickness in the heat transport problem is obtained. ϑ – temperature difference on the outer and inner surfaces of the hollow sphere. that the solidification front moves from the top of the spherical shell to the nozzle orifice and the temperature difference does not exceed 0. 7. 200 seconds (b). From the balance of heat fluxes described by Gottstein (2004) the heat of solidification can be obtained on the base of the equation (6): C   dx  dT   dT    L    hS   dt  dx  C  dx  L  с   . heating plates and nozzle. x – distance (wall thickness). The results go back on the micro-. For thinner walls the temperature distribution shows greater deviation from a linear characteristic as for thicker walls. t cal (5) where tcal – calculation time. ISO-surfaces also show.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. CONCLUSIONS AND OUTLOOKS In the present paper the heat transport during the hollow spheres production from the tin melt was investigated in the numerical. there temperature could be better described by a linear function.2°C. presented in figures 8b and 8c. Simulation of the temperature fields in the spherical shell at the time point of 20 seconds (a). Fig. Solidification process From the simulation results presented in figure 8 it is clearly seen. meso.  (6) – 281 – COMPUTER METHODS IN MATERIALS SCIENCE . 8. Hollow spheres/shells could be produced directly from the tin melt if the boundary conditions were properly defined. represented by simulation of the temperature distribution in heating system. Temperature fields on the outer surface of the spherical shell in 1000 seconds (a). that at the same time point the temperature in the spherical shell distributes very fast. hS – heat of solidification. tC – time of the crystal growth. because the temperature changes here during the whole processing time does not exceed 0. that non-uniform wall thickness of the shell SR   . The temperature fields in the sphere’s wall due to small dimension of the sphere changes very quickly and need to be investigated separately with finer finite elements. theoretical and experimental way.6. From the equation (4) follows.and macrolevel numerical problems of the investigated system.

Both the information about the temperature distribution in the metallic melt and the gas distribution were used by the design and construction of the nozzle: nozzle placing. Można je stosować jako komórki jednostkowe dla większych elementów lub jako elementy samodzielne wypełnione gazem obojętnym. Physical Lebedev Institute Publishing. Puste kulki o małych średnicach wykonane z różnych materiałów mogą znacząco zmniejszyć wagę elementów konstrukcyjnych wykorzystywanych jako izolatory akustyczne i termiczne. The prediction of the undercooling can be made on the hand of temperature-distance diagram in figure 6a and calculated from the equation (5) and (6). 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 282 – .A. M. Opisany proces metalurgiczny produkcji kul wykorzystuje fizyczne własności zastosowanych materiałów oraz warunki brzegowe procesu bez wprowadzania proszków czy zawiesin. The results can be used by the further microstructure prediction by function with two arguments.. The methods of the hollow microspheres production. A. Gottstein. This paper was prepared in scope of the state contract № 16. funded by Ministry of Education and Science of the Russian Federation. Moscow (in Russian) Petrov. W pracy do symulacji wykorzystano komercyjne programy FLUENT oraz pakiet SolidWorks. Berlin. Vogel Buchverlag. 2012 Received in a revised form: November 4.und Stoffübertragung. K. H... Scientific reports on resource issues. Moscow (in Russian). Technische Strömungslehre... Physical foundations of materials science.. Hydraulic and hydraulic drive systems. BHT (Berg. Springer Verlag. Drebenstedt. M. Dorogotowcev. LeCroisette. C. orifice diameter. MSIU Publishing. Loycyanskiy. Volume 3. 9th edition.11. Powłoki z czystej cyny produkowane są w temperaturze bliskiej temperatury topnienia materiału (w stanie tiksotropowym). The minimal diameter could be obtained from the Young-Laplace equation for hollow microspheres up to 1 mm in diameter and for bigger microspheres up to 3 – 4 mm in diameter the technique given by Petrov (2012) can be applied. Consequently the cooling rate will differ and the microstructure development in the wall will vary. Petrov. 1981. 2012. ZAGADNIENIE TRANSPORTU CIEPŁA PODCZAS PRODUKCJI PUSTYCH KUL Z CYNY Streszczenie Praca przedstawia jedną z energooszczędnych metod produkcji zespołów struktur komórkowych (pustych kul) stosowanych w lekkich konstrukcjach. TU Bergakademie Freiberg. Transport ciepła w procesie formowania powłok z cyny od stanu płynnego do stanu pół-stałego ma wpływ na proces zarodkowania.. Untersuchungen zur Hohlkugel. New York (in German). Moscow (in Russian). H. New York. 2006.M. Also a problem of undercooling effect due to expansion of not preheated forming gas at the nozzle was formulated and investigated. Merkulyev. 7th edition. Pasadena..G. Wärme. Acknowledgement. Springer Verlag. Proceedings of the second international colloquium on drops and bubbles.0744. 2010. NASA JPL. REFERENCES Baehr.. Proceedings of the 61. Received: September 20. namely the outer shell diameter and wall thickness. Ponadto badania uzupełniono o wyznaczenie pola temperatury na powierzchni dyszy formującej dzięki termogramowi zarejestrowanemu kamerą na podczerwień (IRC). Badano między innymi rozkład temperatury podczas procesu produkcji. 1991. G. 2006.. J. dlatego ścianki będące w stanie stałym powinny być formowane zanim gaz rozpocznie kształtowanie wnętrza powłoki. 5th edition.und Schalenherstellung direkt aus der metallischen Schmelze zu ihrer Anwendung in Leichtbaukomponenten. W. Bast.. D. Berlin. Würzburg (in German). W przeciwnym wypadku pół-stała powłoka pęknie pod wpływem ciśnienia gazu lub w ogóle nie powstanie. Heidelberg. W ramach pracy wykonano szereg symulacji komputerowych. J. L. 5th edition. Hydrodynamic performance of an annular liquid jet: production of spherical shells. Stephan. 2003.740. a także jako ochrona przed wibracjami. W.INFORMATYKA W TECHNOLOGII MATERIAŁÓW will cause different temperatures.und Hüttenmännische Tagung). eds. 1989. PhD thesis.. 343-350 (in German). Sheypak. Freiberg (in German). Heidelberg. eds. temperature fields around the nozzle etc. 7987. 2004. These essential parts of the work make possible to increase the production capacity of the laboratory equipment. 2012 Accepted: November 21.. Wyniki symulacji zostały porównane z wynikami doświadczeń oraz obliczeń teoretycznych. Mechanic of fluids and gases. Entwicklung einer Anlage zur Hohlkugelherstellung. J. Freiberg. Bohl.. Kendall. D. Also the optical surface quality of the produced microsphere varies between rough for small cooling rates and smooth shiny for greater cooling rates. które umożliwiły określenie właściwych warunków brzegowych procesu produkcji.

all the contemporary methods – 3D reconstruction of objects or measurements of deformations . yet cost-effective solutions. 1996. This involves a number of tests that should be carried out to find optimal. The second method is based on the analysis of surface patterns (Swillo. Warsaw University of Technology. thus providing fast and accurate results for material testing and process analysis. In this group numerous solutions are available. However. Warsaw/Poland *Corresponding author: s. However. and checking if they do not exceed certain limit resulting in the loss of stability. the sheet metal forming is a very complex manufacturing process. a new method for the 3-dimensional geometry and strain measurement based on laser scanning technique is presented. The hydraulic bulge test was identified as a method most applicable in these investigations. 13.pl Abstract The paper presents a method for the surface shape and strain measurement applied in the determination of metal flow and product quality. sheet metal forming. The first of them is the projection method described in detail by Swillo and Jaroszewicz (2001). Especially. 2 AN EXPERIMENTAL STUDY OF MATERIAL FLOW AND SURFACE QUALITY USING IMAGE PROCESSING IN THE HYDRAULIC BULGE TEST SŁAWOMIR ŚWIŁŁO Faculty of Production Engineering. No. hydraulic bulge test 1. In this study the author discusses the investigations that are focused on better understanding of the strain values and their distribution in a product. First. Sirkis. 2013. The third method is laser scanning. three methods are widely used by different authors. These techniques seem to be very useful in the field of metal forming because they are very effective when strain values have to be determined by the analysis of surface patterns. The 3D shape and strain analysis presented in this paper offers a valuable tool in the metal products quality control. Key words: strain analysis. and its success depends on many factors. compared with the solutions existing so far. Accurate determination of these characteristics in the sheet metal forming operation is extremely important. The method commonly used in studies of the kinematics of the sheet metal forming process is 283 – 288 ISSN 1641-8581 . along with a complete testing equipment for maximum strain calculation just before cracking.swillo@wip. where both theoretical and experimental analysis was conducted.are based on a system of image recording by means of a CCD camera. 2001. INTRODUCTION Methods for the measurement of surface shape or deformation (displacement) generate solutions that are currently and generally applied in various scientific fields. Second. the image acquisition process and digital image correlation (DIC) are presented to recognize and analyze the objects taken from camera.pw. 1990). and their classification depends on the technique by which the displacements are measured.edu. The computer measurement system is directly connected to the hydraulic bulge test apparatus. the presented solution in the field of reconstruction of three-dimensional (3D) objects significantly expands the capabilities of the analysis. Next. Koga & Murakawa. especially in automotive applications.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol.

According to the current experimental investigation. One of such solutions is a system that enables automatic measurement of lines perpendicular to each other and lying at a distance of 1-5 mm from each other. At the same time. enabling analysis of the local state of strain in industrial sheet metal forming. Schematics (section view) of the sheet metal forming: a) using metal punch. due to which the processed element assumes the shape of a spherical cap. These methods are based on either correlation or analysis of the geometry of regular coordination grids. Image analysis of the deformed sample area is done by a commercial ARAMIS system made by GOM (Hijazi et al. All these solutions are based on the identification of grids applied to the surface of sheet metal. The stand with this device is capable of performing a fully automated control of the bulging process using a metal punch The solution uses a digital image correlation technique based on stochastic grids. It is believed that. 2010). Therefore the goal is to develop a major parameter that can be used in an assessment of the quality of automotive parts after sheet metal forming. originally proposed by Marciniak (1961). are much more efficient as regards the accuracy of the obtained results than the regular grids. Fig. or at least controlled. 1. a uniform biaxial stretching occurs. studies are carried out to detect the crack onset.INFORMATYKA W TECHNOLOGII MATERIAŁÓW stretching of the sheet metal surface with a metal punch (figure 1a). during laboratory tests. we can find. This solution is based on the use of a metal punch operating on samples with different geometry (figure 1a). Owing to this characteristic of the deformation limit of any sheet material. provide comprehensive information about the state of the material (for example: Swillo et al. The process is performed on a specially designed testing machine. The Company has developed a system called AutoGrid used for the analysis of deformation during bulge test. all available information to predict the quality of the final products is not sufficient. 2004). In the last several years. which. depending on the nature of the measurement.. As complementary to the created solutions of automated strain measurement. minimized. various technical improvements in the method of sheet metal pattern recognition took place. 2. In the present study. combined with calculations of the deformation occurring in the examined products. where appropriate optical system with two cameras can record the run of the forming process. In this operation. methods that serve the determination of forming limit curves which.. On the example of bulging test. The. Currently. a method of bulging with fluid under pressure the sheet metal disc clamped at the edges has been applied (figure 1b). method of grids coupled with the image processing has gained numerous solutions. in the authors opinion. 2006. EXPERIMENTAL APPARATUS Solutions presented in this paper are referring to the three dimensional cases. The proposed method of the strain measurement in a bulging process is based on the numerical image processing and three-dimensional object reconstruction. Studies on the design of systems for automatic strain analysis are carried out by Vialux Company (Feldmann & Schatz. The possibility of plotting the forming limit curves also enables predicting what are the chances for further plastic forming of the sheet metal. most of the final product defects and limitations can be eliminated. These techniques seem to be very useful for the metal forming analysis because they are very effective when strain values have to be determined by the analysis of stochastic surface pattern. a new possibility in the field of image processing techniques is demonstrated. The grids can COMPUTER METHODS IN MATERIALS SCIENCE have a regular or stochastic shape. to test the plastic forming process. the collected information allows us to determine if the strain values in areas of the largest product deformation are approaching the limit values. 2007). used for many years. Feldmann. b) using hydro-bulging. many authors have been evaluating the sheet metal quality based on the forming limit curves (FLC). A specially designed exper- – 284 – . among others. and Marciniak & Kuczyński (1967). Another option is a system for the analysis of deformation based on the bulging test using a steel punch (Liewald & Schleich. by careful designing of process operations. 2000).

The use of the stand allows running two types of the measurements: basic and complex.INFORMATYKA W TECHNOLOGII MATERIAŁÓW imental apparatus for bulging process analysis was assembled (figure 2a). proposed in the measurement model. The deformation in the forming process Fig. fully automatic. 1961). The developed solution comprises a computerized. a biaxial uniform stretching occurs. method of outline reconstruction using laser light bases on a temporary outline searched for the examined element subjected to deformation.. or curling of the sheet metal. 2012). which enables the reconstruction of a 3D image of the measured sample and determination of the size of deformation. The first group of measurements includes recording the run of the plastic forming process in terms of pressure and displacement. each lasergenerated section is identified by one camera only. The proposed mathematical model to solve this problem has been based on the author's own research (Swillo et al. The central measurement system in the test stand is an optical system. In this example of the sheet metal forming process. Owing to some simplifications cannot reach any arbitrarily large values. motorized test stand equipped with optical and vision systems to acquire the data. because some limiting phenomena will occur at a certain stage of the process. producing specimens where the dome is spherical in shape. In this process. high accuracy (sub-pixel) verification of the geometry of the distorted elements. Full description of the bulging process should enable further materials research and development of process control mechanisms. The. while second group includes measurements of the process kinematics and of the shape of the bulged samples. ANALYTICAL MODELING OF THE BULGING PROCESS The aim of the investigations of the sheet metal forming process is to deepen our knowledge about the factors (phenomena) that restrict this process. The main problems include strain localization. The elements of such control can include. 3. it was necessary to use measurements based on vision control. cracking. the formed object is obtained by mapping its shape on a sheet of metal using a punch and a die (figure 1b). In the described study. continuous (pixel-based). an option for the automatic monitoring of the process run and the possibility of its interruption at a strictly determined stage of deformation. In addition to the measured contour. during which the material undergoes plastic deformation by drawing. Schematics of the experimental apparatus for hydro-bulging process: a) the testing stand. a front view of the object (with stochastic grid) is recorded. proposed for studies of the kinematics of the shaped objects. To obtain rapid (in real-time). while the assumed axisymmetrical shape of the examined object allows its 3D reconstruction. In this operation.die contact surface (Marciniak. numerous solutions are possible as regards the description of the process kinematics and study of the test conditions under which the loss of stability occurs due to the absence of friction on the tool . disturbing or disrupting even its further course. The state of biaxial stretching occurring in many – 285 – COMPUTER METHODS IN MATERIALS SCIENCE . among others. tests of the plastic forming were performed using a method of bulging the sheet metal discs (fixed at the edges) with fluid under pressure. 2. whose task is to allow a 3D reconstruction of the sheet metal formed (figure 2b). extrusion or redrawing. b) an optical system to control the bulging process.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW forming operations makes the bulge test very useful in this analysis (mainly due to the mere nature of the state of stress). 4. – 286 – . 3. b) vision inspection (crack localization). The run of this process is usually considered in terms of the strain occurring in a perfectly flexible thin membrane. the extreme values of the internal pressure p is given by: (2) where: R is the actual dome radius and g is the thickness at the top of the dome. where the experimental results obtained on a stand for the test forming using oil were compared with computations made for the membrane theory. the relationship (2) can be solved by differentiation and simplified to the final form of: This relationship allows us to determine in a graphical manner the strain value  at which the pressure p will reach its maximum for the known hardening curve. Fig. Figure 3 shows the results of calculations. Using information from the second camera. The high consistency of the obtained results confirms that the theoretical solutions used to determine the polar limit strain values are correct theoretical assumptions for this area of the metal forming technology. Displacement and strain calculation: a) global strain calculation using DIC. 1961). According to a mathematical formula relevant to this case. b) stress-strain hardening curve. The deformation  occurring in the center of the dome follows the relationship given below: 2 1 (1) (3) where: h is the actual dome high and a is the blank radius. the product of the yield stress multiplied by the actual thickness of the membrane (p*g) is constant (Marciniak. 4. c) comparison for the local strain calculation (micro-strain results) and vision inspection. STRAIN MEASUREMENTS ON STOCHASTIC GRIDS As a result of the performed calculations. In the pure biaxial case where the bulge is a perfect bowl (figure 1). the profile of the bulged sample was determined (according to previous de- COMPUTER METHODS IN MATERIALS SCIENCE Fig. Since we have the biaxial uniform stretching. the coordinates for the projection of nodal points in a twodimensional space were obtained. A comparison of experimental and analytical results of the bulging test: a) strain distribution.

. Proc. 5. The method consists in adding up two values of the deformation. uniform stretching occurs. The described example of the process of the sheet Fig. The method of image correlation used for this purpose is a highly efficient tool for an accurate measurement of these parameters. In this operation. Zurich. FLC. metal forming allows obtaining a number of solutions as regards the description of the process kinematics and study of the test conditions under which the loss of stability occurs due to the absence of friction on the tool . the introduction of a number of assumptions to the design of a test stand using this method. The large scatter in the experimental measurements is due to the lack of a more refined technique for taking precise strain measurements by the method of correlation. in accordance with the proposed 3D reconstruction algorithm. The second value results from the determination of strain that occurs in the crack onset zone (figure 4b). based on experimental calculations. for which the strain localization occurs (figure 4a). the hardening curve for DC04 was plotted. extended to the identification of three-dimensional objects with the possibility of deformation measurements. Conf. 5. resulting in the formation of objects in the shape of a spherical cap. M. a) comparison of experimental results with the hardening curve: b) bulge-samples. The use of this approach in the study of the kinematics of the forming process is a solution that has required the development of a mathematical model. which is a key factor in the calculation of plastic properties. The first value of the deformation is calculated as a result of the identification of the position (image). Effective Evaluation of FLCTests with the optical in-process strain analysis system AutoGrid. The method commonly used in the study of the kinematics of the sheet metal forming is stretching of the sheet metal surface with a metal punch. The accurate measurement of the forming limit is one of the major issues in plotting of the forming limit curves. Acknowledgements. as well as the development of methods to process images recorded during plastic forming. Figure 4c shown the comparison of the calculated strain localization and vision inspection. As an operation complementary to the strain calculation. REFERENCES Feldmann. P. N N508 390637). the test method used for plastic forming has been bulging with a fluid under pressure of the sheet metal discs. ed.. In this study. comparing the results with the measurements by the method of uniaxial stretching and with information about the equation of a curve given in the literature (figure 5a). 2006. Finally. the third coordinate was specified. SUMMARY The method proposed by the author consists in the identification of an outline of the examined axisymmetrical object.INFORMATYKA W TECHNOLOGII MATERIAŁÓW scription) and. 69-73.. Strain measurements were carried out according to the described method of calculations for the main deformation direction and directional displacement gradient measurements (figure 4a). Hora P. micro-deformations in the zone of the crack onset during bulging process were determined. Large number of images generated during measurements is an obstacle in precise determination of the deformation history. a biaxial.die contact surface. fixed on the edges. – 287 – COMPUTER METHODS IN MATERIALS SCIENCE . Scientific work financed as a research project from funds for science in the years 2009-2011 (Project no. Schatz.

Gen. and Jaroszewicz. & Majta J. 1961. Int. 2004. 13. Swillo.1175-1178.P. Automatic of strain measurement by using image processing. 2012 Received in a revised form: November 21. Parsaei. Conf. Sirkis. International Deep Drawing Research Group. Madhavan V. Dokładna charakterystyka tych wielkości w procesach kształtowania blach jest niezwykle ważna. Las Vegas. System response to automated grid methods. Determination of forming limit curves using 3D digital image correlation and in-situ observation. S. Determination of the forming limit curve by using stereo image processing. Pietrzyk M. Proc. Advanced Technology of Plasticity. H. Lisok. Received: October 15.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Feldmann P. Conf. Szeliga D. 545-550. Opt. Columbus.. Conf. Long Beach. 2001. Kraków. T. Metal Forming 2000.R. & Pillinger I. Application of strain analysis system AUTOGRID for evaluation of formability tests and for strain analysis on deformed parts. Koga.. Tisza M.. 3. 2010. jak również dla identyfikacji miejsc potencjalnych pęknięć. Conf. Kuczyński. Conf. Świłło. A... M. 483–490. Schleich R. P. Kusiak J.. Wong J. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 288 – . Engineering Design & Automation. S. 2000. M... Archives of Mechanics. gdzie przedstawiono dwa rozwiązania doświadczalne i analityczne. H. Automatic shape measurement on base of static fiber-optic fringe projection method.. H. A. Z... II. 9.. Limits strains in the processes of stretch-forming sheet metal.R. eds. Leep. Kocańda. Parsaei. H. Altan. and Hartley P. Marciniak. S. Wong J.. N. K.. Eng. J. S.. Liewald. Journal of Material Forming. Zaproponowana został metoda hydro-wybrzuszania. 2012. Czyżewski. Engineering Design & Automation..R. An experimental study for hydro-bulging process using advanced computer technique. S. Las Vegas. Int. 1411-1414. 476-481. Dlatego wykorzystywanych jest wiele testów. W pierwszej części przedstawiono rozwiązania w zakresie rekonstrukcji trójwymiarowej. które znacząco rozszerzają możliwości w stosunku do tradycyjnych metod optycznych analizy kształtu. and Weinheim. L.. 791-803.P. Journal of Mechanics Science.. Następnie zaprezentowano wyniki analizy obrazu z wykorzystaniem korelacji. Conf. Hijazi... ed. Kraków. SAMPE. Vol. Proc. 1990... 1967. eds... A. 2012 Accepted: December 5. M.. M.. Metal Forming 2012.. and Piela. Leep. 609-612. Marciniak Z. 2001.. Proc... Development of an Anisotropic Failure Criterion for Characterising the Influence of Curvature on Forming Limits of Aluminium Sheet Metal Alloys. Conf. w celu określenia charakterystyki materiału. Dzięki sprzężeniu urządzenia testującego z układem komputerowym możliwe jest szybkie i precyzyjne przedstawienie końcowych wyników pomiarów własności materiału. eds. 1996. 743-752 (in Polish). W przedstawionym opracowaniu autor koncentruje się na lepszym zrozumieniu rozkładu odkształcenia i jego koncentracji prowadzącej do utraty stateczności. Proc. Proc. 29. Jakkolwiek jednak jest to proces niezwykle trudny i uzależniony od wielu czynników. Majta J. Influence of the sign change of the load on the strain hardening curve of a copper test piece subject to torsion. Świłło. Gen.R. Yardi N. Proc. szczególnie w przemyśle samochodowym. 2007. 1485-1493 Swillo. Gyor. wskazując na znaczne ułatwienia w rozwiązywaniu problemów wyznaczania odkształceń i pomiaru kształtu dla przedstawionego przykładu.R. ed. Proc. 272-277. vol. eds. Application of visioplasticity to experimental analysis of shearing phenomena. ANALIZA DOŚWIADCZALNA PŁYNIĘCIA MATERIAŁU I KONTROLA JAKOŚCI POWIERZCHNI W PROCESIE WYBRZUSZANIA Z WYKORZYSTANIEM OBRÓBKI OBRAZU Streszczenie W artykule przedstawiono metodę pomiaru geometrii i odkształceń w odniesieniu do pól przemieszczeń i końcowej jakości produktu. and Murakawa. Kusiak J.... 571-574.

No.Automatic Surface Inspection System). global competition and rising customer’s requirements are becoming increasingly important in production of high quality products. ul. Steel industry is no exception to that rule. hot rolling mil 1. Combination of these factors often proves to be very difficult or even impossible to achieve with the use of common production methods.com Abstract The subject of this paper is to design and implement an efficient model for various kinds of scales recognition at the Hot Rolling Mill (HRM) in Kraków. as well as the method of interpretation of the data obtained from this system. to detect local variations in contrast on its surface and to classify individual irregularities. In this way a map can be obtain. as well as reliable online classification. Ujastek 1. The purpose of the system is to take pictures of produced material. 13. This paper describes the basics of automatic surface inspection system (ASIS) which was used as a source of entry data. 2013. To make it possible to build reliable classifier of surface defects. This type of system brings a significant reduction of visual inspections performed by human inspector. which will be subsequently used in building reliable classifier for scale recognition.lechwar@arcelormittal. Person. It was done by use of data mining techniques. Subsequently. This problem will be solved in following studies by creating offline post processing rules. In an ordi- 289 – 294 ISSN 1641-8581 . One way to achieve this goal is the application of automatic surface inspection of rolled sheets (ASIS . each plant puts strong emphasis on the automation of its process and the maximum costs reduction.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. which could distinguish them from the other defects. INTRODUCTION In today's industry. stored in online production database. Each picture taken by the system is digital and converted into grey scale pixels. The ASIS provided numerous features describing single image. cold rolling process) requires production of higher quality steel while reducing costs. The objective of this paper was to supply information regarding the most important visual attributes. However. 2 SELECTION OF SIGNIFICANT VISUAL FEATURES FOR CLASSIFICATION OF SCALES USING BOOSTING TREES MODEL SZYMON LECHWAR ArcelorMittal Poland. Key words: automatic surface inspection system. ASIS needs to be taught. The reason behind that was the lack of unique features. which supplies information regarding defective material in terms of various defects and pseudo defects. Nevertheless. 30-969 Kraków *Corresponding author: szymon. Direct customers and subsequent treatment processes (ex. At the moment an extensive knowledge regarding the reasons of scale occurrence is gathered. which was considered as a defect. boosting trees. the real challenges nowadays seem to be measuring techniques of those phenomena. The result was a set of measurement data. should create sets of defects that will be used to “teach” software provided by the manufacturer. At the same time. some kinds of scales could not be recognized efficiently. data mining. the model and its most important variables can be used to describe and distinguish different kinds of scales. who is an expert in certain classification. Hot Rolling Mill in Kraków.

The research was concentrated on scale defects produced at hot rolling mill in Kraków. This study focused on scale defects.1. Classes will be treated as a reference data based on which data analysis will be carried out. 2006). strip thickness and technological mill settings. which provided information regarding classification of the currently implemented classifier were removed. 2003. were removed as they did not carry any valuable information. From the pool of 26000 candidate’s images some of them were isolated as real scale defects.. In this approach it is impossible to distinguish specific types of defects using results given by the ASIS.. about 30 different real defects could occur on the hot strip. Any transitions and transformations were not carried out during data cleaning stage. Preliminary data analysis COMPUTER METHODS IN MATERIALS SCIENCE The first step in data mining analysis (Hand et al. which had been significantly reduced due to correctness of reference data manual classification. In the end. that will be used in creation of reli- – 290 – . 2006) focused on data preparation and cleaning. Variables. In total. Different defects occurring at hot rolling mill in Kraków production line were sorted in order to conduct the study. DISTRIBUTION OF SCALES AT HOT ROLLING MILL IN KRAKÓW ASIS is monitoring whole coils production at the hot rolling mill. 2006.. raw data consisted of 744 variables. user cannot build additional rules in third party software to assist classification of similar defect classes. were analysed by the manufacturer software to find local variations in contrast. Images. In order to seek the most important features describing categorical variable "type of scale" input data was analysed by the two data mining (Statsoft.. Furthermore. Features. Han & Kamber. returning as a result map of irregularities in the contrast detected on the surface of hot strip. analyse visual features of each defect class. 2. 3. set of 3300 scale defects were gathered and hand-classified. with variance below 10-10. At most. received a number of features that describes their visual characteristics in numeric manner. Defects images were taken by ASIS from production line. Such areas. 2007) and expert knowledge it was decided to select and distinguish 10 scale classes. as they supply unnecessary data at this stage of study. This approach cuts user's knowledge about visual characteristics of the images. 2002. Bakker et al. As a result. On this basis. the end user (expert in the field) makes selection of images that he believes belong to specific classes of defects and arrange them in the program supplied by the manufacturer.2001. In this study it was decided to deal with this issue in more detailed manner.2. 2004). Choice of the best set of features describing variable "type of scale" It was decided to build Boosting Trees model in order to supply information regarding the most important features. which finds the best predictors for each dependent variable.INFORMATYKA W TECHNOLOGII MATERIAŁÓW nary approach. 2006) modules:  Decision trees C & RT (Classification and Regression Trees)  Variables selection and analysis of the causes. which was passed to the further analyses. Not all scale classes will be presented in this paper.. ASIS is trained to detect and classify all of them. called regions of interest (ROI). 3. This was followed by manual classification of images based on expert knowledge and reference materials (Melfo et al. which has an influence on construction of a scale classification model and decides which features could be used in future work to build a reliable scale classifier. SELECTION OF THE MOST RELEVANT VISUAL FEATURES OF THE SCALE DEFECTS USING DATA MINING METHODS Creation of a rich set of reference data provided opportunity to explain visual features of scales in detail. 145 variables were selected for further analysis. Sun et al. Sun et al. Therefore. It does not contain any missing fields or repeated observations.. examines the characteristics of images and selects the rules by which it will be possible to classify newly emerging images. 3. Interactions between predictors are not taken into account. the software creates models. Based on ArcelorMittal internal defects catalogue (Breitschuh et al. taken by ASIS. It was decided to manually select a set of reference defects. The aim of the work was to find visual characteristics of defects that best serve to build a model (Webb. It was assumed that reference data does not contain any unusual values or outliers. The number of possible defects that might be produced during production varies depending on steel grade.

First two items from the table 2 describe decomposition of scale on the strip. Number of variables Erase of redundancy data Fast variables selection Minimum cardinality of node Minimum cardinality of node of descendant Maximum number of levels Maximum number of nodes A priori probability 33 No No 123 1 10 13 Equal The second case (figure 2) gave much more promising results at 33 variables. position of defects throughout the strip will be ignored. Only one parameter. different parameterisations were analysed. the best model reached 93. Nevertheless. Fig. Correctness of the Boosting Trees model depending on number of variables – without the use of redundant variables.34% of correctness on the test sample. The rest of the parameters remained at default values. 3. Two variants were tested. Finally.65% of correctness on the test sample. Table 1. gave better model (figure 3). Table 2 presents collected variables. proper work of descalers is only the cause of defects that should be eliminated. In order to find the most efficient model. i. The change of a priori probability. Correctness of the Boosting Trees model depending on number of minimum cardinality of node. Parameterisation was continued with the use 33 variables as they gave the best model at this point of study. model defined the most relevant attributes that are the case of this study. maximum number of levels and minimum cardinality of descendant did not change – 291 – COMPUTER METHODS IN MATERIALS SCIENCE The benefit from construction of this model was selection of 33 variables. in this paper. without the use of redundant variables (those with correlation between variables exceeding 0. 1. .8) and with redundant variables.INFORMATYKA W TECHNOLOGII MATERIAŁÓW able classifier. First step covered testing of the model depending on number of variables. along with their significance and short description. Besides classification. efficiency of the model. maximum number of nodes. It is most often associated with abnormal work of descalers. minimum cardinality of node at 123. Input data was divided into learning sample and the validation sample with 80% to 20% ratio. with 94. which describe characteristics of the scale defects. Parameters of best Boosting Trees model. Fig. Fig. Therefore. The core of the work was to find visual features that could be used in creation of efficient scale classifier. Correctness of the Boosting Trees model depending on number of variables – with the use of redundant variables.84% of correctness on the learning sample and 80. 2.26% of correctness on the learning sample and 78. because they remove only a portion of scale covering the slab. First case (figure 1) shows that the best model can be obtained for 21 variables.e. Table 1 contains gathered parameters of the model.

straightforward use of the attribute to classify one of these scales. Subsequently.3 and 0. Figure 7 shows decomposition for “V” scale. Its shape resembles “V” letter. which originates from the first stand of finishing train at the end of rolling campaign. which . 5.DISTINCTION OF SCALES THROUGHOUT VISUAL FEATURES OF ITS IMAGES Most relevant features. is not possible. COMPUTER METHODS IN MATERIALS SCIENCE Fig. whenever it lies between -0.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 2. that. – 292 – . Figure 6 shows decomposition of the feature for line scale. is “maximum difference between horizontal and vertical dark segment length”.5 or -0.in the opposite to previous one . The other type of visual feature. One example of feature. could be implemented as classification support is “horizontal to vertical difference of gradient range”. Although. Horizontal to vertical difference of gradient range – Single strip scale. together with necessary logic. Significance of visual features. were used to distinguish scale classes. which is defect originating from a finishing train. these features along with sufficient logic will support automatic classifier (build by default within supplier software). It describes the numerical differences between horizontal and vertical gradient ranges (in grey scale) of the defect.could be used in direct classification. RESULTS . The feature could support final classification decision between these two classes. that had been isolated from a wide variety of attributes given by ASIS.1. Horizontal to vertical difference of gradient range – Line scale. Figure 5 shows decomposition of the feature for single strip scale defect formed due to malfunction of the descalers.3 and 0. 4. Figure 4 shows decomposition of the feature for line scale. Fig. 4. It is possible to create additional classification rules both in C++ language and T-SQL stored procedures (ASIS database). The attribute inform the classification system what is the biggest difference between the horizontal and vertical lengths among all dark segments in the defect (segments composed of dark pixels).

76-83. Jiang.. It will depend on manual selection of the scale defects and their assignment to proper class. 2006. Proc. Smyth. Lu. C. W.. Data mining: concepts and techniques... Cambridge. Difference between horizontal / vertical of maximum dark segment length – “V” scale. Acknowledgements.. Hand. A.. Tieu.. England. Statsoft. Z.K. C. Jiang.. Diaz-Alvarez.... Koschack. Deléglise... Dallemagne. P. Statistical Pattern Recognition (2nd Edition). called postclassification. In-Situ Study of Scale Formation under Finishing-Mill Operating Conditions. Journal of Education and Work. Zhu. In this case a threshold could be set at 0. Maximum difference between horizontal and vertical dark segment length – Line scale. Valcarcel. Webb.M.. 33 most relevant attributes for the model were selected. University of Illinois. R. P.. Chichester. To improve its classification decision some additional rules have to be created. the features can be used in building reliable classifiers for scale recognition.. U. study assumes that creation of the best possible classifier could be hard to obtain using manufacturer software. a wybrane cechy posłu- – 293 – COMPUTER METHODS IN MATERIALS SCIENCE . 2007.. Ltd. John Wiley & Sons. Hoyles. Nevertheless. Subsequently.. Improving Work Processes by Making the Invisible Visible... Tieu. Second part of the paper describes process of Boosting Trees model creation for the scale classification. which could be used in classifier building. Fig. W. Schroyens K. 2006.. A. CONCLUSIONS AND PERSPECTIVES In the paper scale defects occurring at hot rolling mill in Kraków were divided into unique classes. J. A. Sun. Fig. Along with the model. L.2. Principles of data mining. dostarczył automatyczny system kontroli powierzchni (ASIS). Breitschuh.. Fiori. C.. Journal of Materials Processing Technology. 2006.INFORMATYKA W TECHNOLOGII MATERIAŁÓW and Transact SQL stored procedures. Melfo. 2002. 140. This kind of rule might be used in supporting logic to distinguish those two scale defects. Sun. R. 25-35. Kamber. II. Mannila.. were presented in this paper. Ohio.. H. którego podstawy działania przedstawiono w pracy. 7. Z.. J.. Urbana-Champaign.. Elektroniczy Podręcznik Statystyki PL. The rules can be written in C++ programming language DOBÓR NAJISTOTNIEJSZYCH ASPEKTÓW WIZYJNYCH ZJAWISKA WYSTĘPOWANIA ZGORZELINY Z UŻYCIEM WZMACNIANYCH DRZEW KLASYFIKACYJNYCH Streszczenie Przedmiotem badań jest zaprojektowanie i wdrożenie skutecznego modelu klasyfikującego wszystkie rodzaje zgorzeliny występujące w walcowni gorącej w Krakowie. Dippenaar.. High temperature oxide scale characteristics of low carbon steel in hot rolling. Otrzymane dane pomiarowe zostały przeanalizowane z wykorzystaniem metod selekcji cech. Hemmerlin. W. ArcelorMittal internal defects catalogue. 2004. 2001. S. Di Fant. Kraków (in Polish). AISTech 2006. The author would like to express his gratitude to Mr Witold Dymek. Surface characteristics of oxide scale in hot strip rolling. Danych pomiarowych. 1307-1312. dotyczących aspektów wizyjnych pojedynczych obszarów pasma. W ramach pracy postanowiono zająć się techniką pomiarową oraz wykorzystaniem danych pomiarowych do budowy optymalnego klasyfikatora wad tego zjawiska. These numerical visual features were used to describe each scale class by decomposition of its values. Cleveland.. M. Journal of Materials Processing Technology. Noss. Reid. A. D. Kent. 19. Only part of the scale visual features. The MIT Press. J. They will be used in the next classification process. Han. M. M. G. Association for Iron & Steel Technology. 2003.. 2006. The next step in the study will be scale classifier implementation with the use of manufacturer software. 5. REFERENCES Bakker. W.. This step will be assisted by knowledge described in this paper. Model oraz jego kluczowe zmienne mogą opisać i rozróżnić poszczególne typy zgorzeliny. Lu. H.. who supported this work at Hot Rolling Mill in Kraków and Professor Maciej Pietrzyk and Dr Łukasz Rauch for their enthusiastic encouragement and useful critiques of this research. 155-156. A. Crowley. 6. 343-361.K. M.

zostały szczegółowo opisane w niniejszym artykule. 2012 Accepted: November 16. wraz z otrzymanymi wynikami.INFORMATYKA W TECHNOLOGII MATERIAŁÓW żyły do budowy klasyfikatora dla wad powierzchni typu zgorzelina. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 294 – . 2012 Received in a revised form: November 11. Klasyfikator zaimplementowany został z wykorzystaniem metod eksploracji danych. Received: October 27. które.

. Poland Department of Automation and Computer Application of Metallurgy. Herlinga-Grudzińskiego .pl Abstract This article describes the works related with the development of an information platform to render accessible the knowledge on casting technologies. The observed disappointment in this class of systems was due to some difficulties related with collection and application of a sufficient number of rules in knowledge bases.. including fuzzy logic. 2010. still very poorly developed area remains that of the technological decision support tools. Cracow. databases integration. technology platforms to share domain knowledge. Ostrava-Poruba. VŠB – Technical University Ostrava 17. Mickiewicza 30. 2 A USER-INSPIRED KNOWLEDGE SYSTEM FOR THE NEEDS OF METAL PROCESSING INDUSTRY STANISŁAWA KLUSKA-NAWARECKA1*. On the other hand.. ZENON PIROWSKI1. 2013. The System is used as a coupling link in the diagnosis of defects in castings. No. 708 33 CZECH 3 AGH University of Science and Technology. MILAN VROŽINA2. Kluska-Nawarecka et al. Studies are carried out on possibilities to apply modern knowledge engineering formalisms as it has been written in (Jančíková et al.. the software market offers “knowledge" systems for computer-aided design and simulation processes (CAD / CAM) and also knowledge management tools and industrial information of the ERP / MRPII type. The guide facilitating the use of the platform is a "virtual handbook". Kluska-Nawarecka et al. The initial part presents the results of a survey on the preferences of potential users of the platform regarding areas of the used knowledge and functionalities provided by the system. JIŘÍ DAVID2. rough sets. The second part contains a presentation of selected modules of the knowledge with attention focussed on their functionalities targeted at the user needs. Poland *Corresponding author: nawar@iod. Kluska-Nawarecka et al. 2010).COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. tools for integration of knowledge from distributed and heterogeneous sources. Švec et al. knowledge. 2011c. Currently. 2007. Zygmunt et al.krakow. decision tables and the ISSN 1641-8581 295 – 303 . INTRODUCTION Currently. ontologies 1 1. 2011a. 13. Poland 4 Andrzej Frycz Modrzewski University. technology platforms. In recent years. the development of algorithms for automated knowledge acquisition has aroused a new interest of science centres in inference systems what has been described (Adrian et al. while ontological module serves the purpose of knowledge integration when different sources of knowledge are used. Zakopiańska 73. expert systems. 2012. DOROTA WILK-KOŁODZIEJCZYK4 2 Foundry Research Institute. 2012). 2011. ZORA JANČÍKOVÁ2.e. 2009.. Cracow.. distributed and heterogeneous sources. KRZYSZTOF REGULSKI3. i..1 Cracow. listopadu 15. Key words: artificial intelligence. Mrzygłód et al.. the interest in expert systems supporting diagnosis and technological decisionmaking process has been subject to some fluctua- tions . Spicka et al..

 visuals: pictures. handbooks including descriptions of the possible types of treatment  norms. User preferences on the types of shared knowledge. including ontologies enabling modelling of domain knowledge. it was considered necessary to refer to the needs of potential users of the system. unavoidable is the research on the methods of knowledge acquisition and integration. data on production volume in a given sector. Respondents were asked to indicate the types of knowledge they believe are most commonly demanded by the manufacturing plants. balance sheets. ISO. were selected from the circles of the scientists cooperating with plants processing different types of metals.irregularities in the process. limited liability companies). and on the substantive content of the system with determination of the functionality of each of its modules. The article outlines the future plans and gives selected results of work aimed at building a system platform with the task of creating and sharing the knowledge from the area of foundry technologies. statistical yearbooks. databases  expert knowledge in case of defects . In the questionnaire they were given the following options:  literature knowledge about processes. request for expert opinion  Branch stats. and microstructures  reports and studies of own production. namely the area rendering available the knowledge components.  photographs of castings and defects. the main objective of the first stage of the work was considered to be an interview with the industry and scientific communities in Poland and abroad dealing in some way with the casting practice. and chemical compositions of materials. volume of production etc. Therefore. COMPUTER METHODS IN MATERIALS SCIENCE Fig. Surveys covered a specific range of the system utility. The interview was conducted in the form of questionnaires and discussions carried out with the representatives of industry and research centres. 2. studies. document templates. ready compilations  specification of requirements and properties. also for diagrams and visuals in the – 296 – . 1. Polish Standards. standards. computer simulations.  tables.  market analysis  marketing data. The selected companies represented the large. and ultimately the creation of semantically structured systems. data from Chief Statistical Office (CSO) The results of survey are presented in the form of a diagram in figure 1.INFORMATYKA W TECHNOLOGII MATERIAŁÓW use of multimedia techniques to render this knowledge accessible to users. certifications. experts It is clear that professionals reach most often for multimedia resources in the form of photographs and simulations. ANALYSIS OF USER NEEDS When the work was started on the development of a concept of the structure. descriptions of processes and applied technologies. With the current rush of knowledge and data. foresight  industry statistics. to determine the need for different types of functionality of the future system rendering the knowledge accessible. small and medium-sized enterprises (joint stock companies. etc.

As a general conclusion from the survey and discussions held. the above mentioned sources will have to be integrated and made ready for processing (e. Databases. 6. standards and certifications.  specification of requirements and properties. databases. tools to make reports based on production data.a platform to share the collections of documents with descriptions of processes and technologies. To achieve this. the specifications of requirements and properties.  literature knowledge and handbooks. catalogues. chemical composition of materials. and microstructures. statements and reports on e. Electronic standards. E-learning. production volume. catalogues of materials e-learning virtual training in advanced manufacturing technologies.  market analysis and marketing studies. 3. indexed for convenient search). KNOWLEDGE-SHARING PLATFORM When the concept of a knowledge-sharing platform on casting technologies has been created.g. Of minor importance was considered industry statistics.  expert knowledge when it is necessary to have an expertise performed. access to publications visuals: pictures. The responses obtained allowed establishing the following ranking of other functionalities: 1. microstructures electronic standards. will also ensure a constant supply of current information and knowledge from the area of the casting practice. simulations photographs of castings and defects. databases. At the same time. tables. consumption of materials. as well as databases. Definitely the highest rated was the proposed "virtual handbok" . As important the following ones were also identified:  norms. certifications Polish standards. The proposed list of functional features is presented in table 1. Poland. the need arises to design a knowledge base in such a way as to make it interactive. while giving the user an easy to handle interface. severity of defects 3. it was assumed that the platform should include all major solutions developed in the course of previous works on the computer-aided manufacturing processes. 2. Also highly rated were the visuals: photographs. Potential functionalities of information system virtual handbook descriptions of processes and applied technologies. List of potential functionalities of an information system. targeted at meeting the user's preferences as regards the application of new trends and opportunities that arise from the The list of potential functionalities of the system was based on the conducted research and currently available capabilities of a system developed as a result of cooperation between the team of workers from the Foundry Research Institute and Knowledge – 297 – COMPUTER METHODS IN MATERIALS SCIENCE Engineering Team from the Department of Applied Informatics and Modelling at the Faculty of Metals Engineering and Industrial Computer Science. etc. etc. At the same time it should be enriched with new modules and functionalities. simulations. Cracow. catalogues. not only derived from the literature. catalogues of materials – all of them reflecting the most common needs. but also from all other sources such as databases. handbooks including description of possible treatments. reporting tools. chemical compositions of materials. interactive courses in casting techniques expert systems discovering the causes of defects. Table 1. the respondents were requested to assess the individual potential functionalities of the future information system operating in their companies. Expert systems. chemical compositions of materials. University of Science and Technology. 5. Probably the reason is effective circulation of such information and needs satisfied by the already existing tools and management systems. photographs of castings and defects.INFORMATYKA W TECHNOLOGII MATERIAŁÓW form of charts and diagrams. and reports and studies of own production.g. detection of process irregularities tools for classification determination of defect types and class/grade of material from which the product should be made. catalogues specifications of requirements and properties. or knowledge obtained algorithmically from the process data. In order to clarify the need for different types of knowledge.4 Classification tools. certifications. . pictures. it can be stated that such knowledge sharing is needed that. to enable user to get through a dialogue with the system just this knowledge that is necessary for solving of a problem. ISO standards. and handbooks and publications in electronic form. costs.

Specification of functional requirements for the knowledge module „Virtual Handbook”. 2011. 3.1. and as a result can find out which variant in the scenario will lead him to a solution of the problem.2. Consequently. 2010). where individual modules can operate independently. one can mention the Infocast system (including databases on publications. Olejarczyk et al.. Below are outlined the results of the implementation of additional modules specific for the operation of the whole platform and which received most attention from the users.. The approximate logic is a tool to model the uncertainty arising from incomplete knowledge resulting from the granularity of information. Czech and French. the user gets a general idea about the type of knowledge provided by the platform. 3. The main application of approximate logic is in classification. whose schematic diagram is shown in figure 2. linkining together the functionalities of other modules of the platform. The databases can be expanded in the future to include other standards. and catalogues) and the Castexpert system designed to serve as a tool for the diagnosis of casting defects assisted with knowledge presented in the form of multimedia. A casual scenario of the use of the handbook is presented as an example in table 2. COMPUTER METHODS IN MATERIALS SCIENCE Fig. 2.INFORMATYKA W TECHNOLOGII MATERIAŁÓW development of methods and technologies based on computer science what has been said in (David et al. Virtual handbook The virtual handbook. The degree and manner of this integration depends mainly on the scenario of actions taken by the user (when using the system in an interactive mode). Using this tool. System for decision support and classification of defects in castings The RoughCast system allows the use of approximate logic to enable a classification of casting defects according to international standards: Polish. and the results of their actions are subjected (if necessary) to the process of integration. the platform has a multi-module structure. standards. – 298 – . Among the modules already existing and available on the Internet. is a kind of clause.

The conducted research allowed developing a methodology for the creation of decision tables to serve the knowledge of casting defects. Abstract class is the smallest unit in the calculation of rough sets. elementary sets can be determined in the array.) 4. The most important step is to determine the upper and lower approximations in the form of a pair of precise sets. Knowledge Engineer       Artificial intelligence Data Knowledge engineering Sources Data Mining Internet Short description Preliminary conditions Final conditions Preparing a specialised Virtual Handbook. The main flow of events 1. expert knowledge. Cataloguing of data: a) Taken from the Internet by means of the Data Mining methods b) Statistical data using rule induction algorithms c) Data from sub-item 4. Sets determined in this way represent the division of the universe in terms of the indiscernibility relations for the attribute distribution. The user writes in the subject 3. Depending on the que- – 299 – COMPUTER METHODS IN MATERIALS SCIENCE . Algorithms of artificial intelligence are preparing relevant data for display in the handbook (semantic analysis with the use of ontologies) 7.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 2.b and alternative sources of knowledge 3. The user must have access to a computer and specified topic of the handbook Handbook ready to display. The user opens the Virtual Handbook interface (on-line) 2. An example of the use of virtual handbook. Database updated and saved . A fragment of the array is presented in table 3. Based on rough set theory. Using this methodology. Name: Actors Shareholders/Stakeholders: Virtual handbook End User. Saving in XML files 6. a decision table was developed for the selected defects in steel castings. Displaying the appropriate page of the handbook Special requirements Device with access to Internet. The system analyses the subject a) Finds data on the Internet and in the documents and databases in natural language b) Collects statistical data c) Searches alternative sources of knowledge (literature.c using decision tables 5. etc. source materials. Table 3. Expert. Fragment of decision table for defects in steel castings as with this logic it is possible to build models of approximation for a family of the sets of elements for which the membership in sets is determined by the attributes.

The process of full-text cluster analysis was used to create the task category (conceptual clustering) as a method of unsupervised learning. the representation of documents using the selected (small) number of words will reduce the quality of classification. The difficulties associated with the classification of documents written in natural language are related to. In the task of conceptual grouping. The knowledge base created for the needs of a RoughCast system is the embodiment of an information system in the form of decision-making tables. and therefore the query interpreter has been equipped with only this one logical operator. polysemy. susceptible to subsequent modifications or development. The dimension of feature space in the document classification tasks. The main reason for the difficulties is the semantic complexity of natural language. Owing to this method of the formulation of queries. It has been assumed that the most convenient way to build queries in the case of casting defects is by selection from a list of attributes required for a given object (defect). and should be. and not with the hypothetical tuples. Each of the resulting aggregations has its centroid. The system operates on decision-making tables this structure of the data allows the use of an inference engine based on approximate logic. and German textbook of defects. there is no need for the user to know the syntax and semantics of queries in the approximate logic. For example. will carry out the task of clustering the documents by thematic classification based on data mining methods . The user chooses which characteristics (attributes) the selected defect has. However. is also a difficulty. these processes will be supported by a thesaurus (a structured set of key- – 300 – . French directory of casting defects. etc. The system maintains a dialogue with the user asking questions about the successive attributes. each article is included into one of the resulting aggregations. as presented in figure 3. The classification of text documents is a very complex problem. etc. whose main objective is to classify documents (articles) by a thematic classification. or it may also mean a set of specifically arranged data. terms having many different meanings. Czech studies. As a result of this process. Articles will be grouped in clusters on the basis of the probability of adjustment determined by the number of occurrences of the common words. It contains tabulated knowledge on the characteristics of defects in steel castings taken from Polish Standards. This module is fully compatible with the directly cooperating document repository systems and databases.INFORMATYKA W TECHNOLOGII MATERIAŁÓW COMPUTER METHODS IN MATERIALS SCIENCE ry. the training set Ω is a collection of articles provided with a compatible system in the form of a knowledge base. The aim was to design a system operating efficiently. the upper and lower approximation is calculated by summing up the respective elementary sets. related to the number of possible words in a natural language (usually the order of tens of thousands of words). This approach is consistent with the situation occurring every day when the user has to deal with specific cases of defects.e. The process of cluster analysis will compare the sets of articles (with each other) in search for common words excluding irrelevant words (or.). Using this formalism it becomes possible to solve a number of difficulties arising from the foundry knowledge granularity in the form of indistinguishable descriptions created with attributes and inconsistent classifications from various sources (as in the case of standards for steel castings). Thus set up queries are limited to conjunctions of attributes. The implementation of task module was developed based on the full-text clustering method. To improve the quality of cluster formation and classification of new articles. Cluster Analisys General design requirements apply to the data mining system. which. i. also. 3. the term 'table' may refer to either a piece of furniture. the first step of the task of conceptual clustering is to reduce the articles from the knowledge base to the basic grammatical forms. to make this dialogue possible without the need to build a query by the user in the language of logic. and on this basis the aggregation centroid will be created. numbers.cluster analysis. On the other hand. Therefore. supported by the use of a thesaurus.3. the system was equipped with a query interpreter in a semantics much more narrow than the original Pawlak semantics. to the greatest extent possible. but. which represents a concept associated with this aggregation. while the task of the aggregation analysis is to split these objects into categories (aggregations) and construct a description of each category (aggregation centroids). The user can select the answer (the required attribute) in an intuitive manner. based on the documents provided (in the correct format and in accordance with the established standards and norms). enabling the efficient classification of new articles. among others.

p2) p1<>p2 implies p1. Constraints can be formulated in OCL for the class Aggregation_Centroid as exemplified is given below. context Aggregation_Centroid::Generate_Aggregation_Centroid() : Boolean pre: self.Aggregation_Centroid_Path<>'' context Aggregation_Centroid::get_ID : Inteager post rezult = self.Aggregation_Centroid_Path The use case diagram is shown in figure 4. Diagram of classes in the module of document aggregation analysis.ID_Centroid and Aggregation_Centroid.Aggregation_Centroid_Path<> p2. 4.ID_Centroid context Aggregation_Centroid::get_PATH : String post: rezult self. 3. context Aggregation_Centroid inv self. In table 4 there is a description of one of possible utilizations discussed algorithm. – 301 – COMPUTER METHODS IN MATERIALS SCIENCE . Fig. Use Case Diagram.allinstance ->forall (p1.ID_Centroid <> p2.INFORMATYKA W TECHNOLOGII MATERIAŁÓW words).Aggregation_Centroid_Path<>'' and Aggregation_Centroid.p2) p1<>p2 implies p1.ID_Centroid >=1 and self.allinstance ->forall (p1. which will also eliminate the problem when compared to the sets that do not have the same keywords in the text.Aggregation_Centroid_Path = '' post self.Aggregation_Centroid_Path Fig. and yet belong to the same category.

. R. It seems that studies carried out to create this context make an interesting contribution to the contents of this article. I.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 4. C... Obtaining edited documents and information in the form of a knowledge base Thematically grouped forwarded database of articles. Dobrowolski. eds. Wilk-Kołodziejczyk. Ružiak.... B.. A. 820/N-Czechy/2010/0. 317-323.. 341-346.. the knowledge base must be delivered. 67-70 (in Polish). Marcjan. since the results of such surveys are not often disclosed in the presentations of different expert and decision-making systems. 3. N... J. I. Nguyen. Application of Bayesian network in the diagnosis of hot-dip galvanising process.. 7. G. Mrzygłód.. 2011a.. Kopal. Proc. K. Vrožina. Currently work is underway on the implementation of modules of the automatic acquisition of knowledge from the Internet and on the analysis and classification of text documents which is presented (Kluska-Nawarecka et al. to expose the way by which they will be adapted to the declared user needs. J. COMPUTER METHODS IN MATERIALS SCIENCE – 302 – . D. Nguyen. Wilk-Kołodziejczyk.. System sends the created database.. R. The Influence of Rubber Blend Aging and Sample on Heat Transport Phenomena. Jančíková.G. Kluska-Nawarecka. 359-369. FINAL REMARKS The article describes solutions developed for selected modules of a platform for sharing the knowledge of foundry technologies in the context of the preferences expressed by users.. and all these articles are characterised by a high degree of fit with each other. 2011b. Mrzygłód. S. 312-315. Defect and Diffusion Forum. 2011c. B. Kluska-Nawarecka.T.. Rough Sets Applied to the RoughCast System for Steel Castings. It has been the intention of the creators and promoters of the platform to offer a system that will have an evolving nature. Z.. Korea. Kluska-Nawarecka.. 2012. Springer. Scientific work financed from funds for the scientific research as an international project. 1. System creates a new database of articles grouped thematically based on cluster analysis carried out. Name: Actors: (Stakeholders/Interests) Automatic clustering of documents System The system manages the service module of the operations of the database. P.. D. Janiak. Kluska-Nawarecka. T. Determination of crystallizer service life on continuous steel casting by means of the knowledge system. Transactions On Circuits And Systems. eds. ACIIDS 2011. A.. 9. S. Hutnik Wiadomości Hutnicze. Semantic Methods for Knowledge Management and Communication. on the one hand. Automatic clustering of documents in terms of thematic fit. Practical aspects of knowledge integration using attribute tables generated from relational databases. C. according to the emerging needs... Conf.. 1. E. Nawarecki. and will be gradually enriched with new modules... 13-22. For an alternative flow of events to occur. S. 11. Kluska-Nawarecka.. Inteligent Information and Database Systems. Regulski. REFERENCES Adrian. 107-112. 2. Decision No. J.... 2. Smolarek-Grzyb. 2011. Durak. Structuralization of knowledge about casting defects diagnosis based on rough sets theory. Selecting modules of the platform described in the article.. it was attempted. Heidelberg. N. the cluster analysis classifies it into one of the topics. Dobrowolski. Chiu. 10. 4.F. M. K. Third International Conference. A. Adrian. Jonšta. Katarzyniak. 2009. to show their diversity and. Description of the case of use. Z.. D. When new article is downloaded. Jančíková. Wilk-Kołodziejczyk. It performs import operations on documents from a knowledge base and export operations on the resulting database to ontology classes. D. the artificial intelligence system must initiate the delivery of a new article. 4. DziaduśRudnicka. 2. S.T. 2011. 2007. Wilk-Kołodziejczyk. 1. Berlin. Kim. Daegu. then a new topic will be created to which these articles will be assigned. Hong.. 183-186. 52-61. Regulski. on the other. The system also manages the module of document clustering by the method of Data Mining. 2011b). Archives of Foundry Engineering. Computer Methods in Materials Science.. Computer Methods in Materials Science. When there is a limit to the number of articles classified to various topics. Receiving knowledge base System performs cluster analysis for the resulting knowledge base.. S. G. Model strukturyzacji wiedzy dla systemu wspomagania decyzji..F. For the main flow of events to occur. A. Acquisition of technology knowledge from online information sources. Short description: Pre-conditions: Post-conditions: Main flow of events: Alternative flow of events: Special requirements: 1. David.

. Conf. The mathematicalphysical models and the neural network exploitation for time prediction of cooling down low range specimen. J. H. zaś moduł ontologiczny. Warszawa. P. Zygmunt. Archives of Metallurgy and Materials. Filipkowski. Švec. Adrian. A. W. system Rought Cast służy do sprzęgania diagnostyki wad odlewów. 2010.. Proc. 55. Algorithm for controling of quench hardening process of constructional steels. Rolę przewodnika ułatwiającego korzystanie z platformy pełni „wirtualny poradnik”.. Analiza otwartych zrodel internetowych z zastosowaniem metodologii sieci spolecznych. 2012 Received in a revised form: November 28. B.. 2012 Accepted: December 7. International Conference on Metallurgy and Materials. Heger. eds: W. J.. 155-159. I.. 921-926. Koštial.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Olejarczyk.. E. C. J. Archives of Metallurgy and Materials. Tanger. Mrzygłód. 2012 – 303 – COMPUTER METHODS IN MATERIALS SCIENCE .. Jančíková. Część druga zawiera prezentację wybranych modułów wiedzy ze zwróceniem uwagi na ich funkcjonalności ukierunkowane na potrzeby użytkowników. Implementation of neural networks for prediction of chemical composition of refining slag. 2012. Mądrzejowski. 2010. Z.. Bialy wywiad: otwarte zrodla informacji – wokol teorii i praktyki. pozwala na integrację wiedzy pochodzącej z różnych źródeł. W części początkowej przedstawiono rezultaty sondażu dotyczącego preferencji potencjalnych użytkowników platformy odnośnie obszarów wykorzystywanej wiedzy oraz funkcjonalności udostępnianych przez system. 55. 197-221 (in Polish). służącej od udostępnienia wiedzy z zakresu technologii odlewniczych. Beck. Nawarecki.. Koźlak.. Metal 2010. SYSTEM UDOSTĘPNIANIA WIEDZY INSPIROWANY POTRZEBAMI UŻYTKOWNIKA Z ZAKRESU PRZEMYSŁU PRZETWÓRSTWA METALI Streszczenie Artykuł dotyczy prac związanych z realizacją platformy informatycznej. I. M. Spicka... P.. Received: October 16. 2010. A. Melecký. Adrian. Franz.H. 171-179.

30-059 Cracow. 2002). which is a bibliographic casting database containing abstracts of over 38. casting 1. EDWARD NAWARECKI2. German. the Institute launched a SINTE database. or for knowledge in the form of rules or guidance on the characteristics of materials and physico-chemical properties. 2009.. Mickiewicza 30. The platform is expected to serve the needs of the metals processing industry. 2013. English.. Kluska-Nawarecka et al.pl Abstract 1 This paper presents the concept of knowledge sharing platform. No. Poland *Corresponding author: regulski@tempus. ARKADIUSZ HARATYM2. Together with NORCAST. that is. Nawarecki et al. Czech. In 1997. decision support. Key words: knowledge integration. Ukrainian). and will make the search for properties (relations). knowledge base. French. KluskaNawarecka et al.metal. 2005).. at the Foundry Research Institute in Cracow. and R&D works written by the staff of the Foundry Research Institute. 2007. 2 THE PLATFORM FOR SEMANTIC INTEGRATION AND SHARING TECHNOLOGICAL KNOWLEDGE ON METAL PROCESSING AND CASTING STANISŁAWA KLUSKA-NAWARECKA1. An open platform model using Semantic Media Wiki in conjunction with the author's script parsing the domain ontology will be presented. proceedings of conferences. Zakopianska 73. which uses an ontological model for integration purposes. a decentralised decisionmaking information system which is intended to support the casting technology both in industry and in scientific and research work (Marcjan et al. 2002. Górny et al.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. GRZEGORZ DOBROWOLSKI2.agh. not just the keywords. 13. construction tools for integrated knowledge bases have been developed (Dobrowolski et al. Poland AGH University of Science and Technology. 2012. This is particularly true when the user does not know in advance what kind of resources will be of interest to him. Russian. which will allow the semantic search supported by domain ontology. Cracow. KRZYSZTOF REGULSKI2* 2 Foundry Research Institute. CASTSTOP and a CASTEXPERT diagnostic system. possible. whether he is looking for publications on the subject indicated..000 articles published in various casting journals (American. In thus extended knowledge base it becomes increasingly difficult to reach to the information searched.edu. semantic search. metal processing. Semantic search will resolve the difficulties encountered in the class of Information Retrieval Systems associated with polysemy and synonyms. 304 – 312 ISSN 1641-8581 . or for the information on standards and certificates. CONTEXT OF THE RESEARCH WORKS For many years. and its immediate purpose is to build an integrated knowledge base. the SINTE database forms a part of the INFOCAST system. Slovenian... The studies carried out at present are aimed at improving the information retrieval systems (IR systems) in such a way as to make the collection and sharing of documents easier and more functional from the user’s point of view.

However. taking also into account the possible upgrading process. which is a set of descriptors arranged in hierarchies. Problems associated with polysemy (variety of meanings depending on the context) or synonyms (if the word does not appear in the document. Such information goes beyond the scope of standardised material properties. Regulski et al. 2012. Additionally. The platform aims at knowledge sharing for non-routine tasks that are difficult to predict in the normal course of production. and also in a way transparent to him. and thus cannot be easily compared for different national or foreign suppliers. However. 1999). To achieve this. also in this situation. Categorising is done with the help of a structured thesaurus. usually based on keywords. so that they can be explicitly used (shared) by the individual modules and also remain ready for reuse by other computerised systems. namely it does not include knowledge of the material properties that can be obtained by thermal or mechanical treatment. Knowing the technical requirements of the designed product. can be the semantic search based on ontological model.. such as the SINTE database. such as databases. information and knowledge. THE PLATFORM FOR SEMANTIC INTEGRATION AND SHARING KNOWLEDGE ON METAL CASTING – CASTWIKI The knowledge sharing platform should act as an intermediary between the user and heterogeneous sources of knowledge. but using them to create an integrated information and decision-making system enriched ontologically would provide additional functionalities and make it easier for the user to use the functionalities already existing. These descriptors allow us to describe the content of the document and make the search of the database possible. as well as components of the knowledge transferred in the system are described using metadata in accordance with the shared ontological model. The solutions and implementations presented in the previous section comprise the knowledge modules that can operate independently (as evidenced by the experience). Mrzygłód & Regulski. the database offers the possibility to search for Polish counterparts of foreign standard alloy grades. It allows the selection of cast materials based on their physico-chemical and technological properties. malleable cast iron. The system should allow the design of a knowledge base in such a way that it is interactive and makes the codification of expert technological knowledge possible. a CASTSTOP system also appears (Połcik. such as grey cast iron. Using appropriate knowledge formalisms. An attempt to create an algorithm which will enable searching the material by its physico-chemical or mechanical parameters. as it happens in the library catalogues. the system has some limitations. although the document is closely connected with that word) pushed researchers towards semantic description of documents using models in description logic (Ciszewski & Kluska-Nawarecka. The task of the proposed system will be the semantic integration of the collected data. The platform is to be conceived as an ontological tool for integration of various subsystems in a semantic network. Zn and Mg. The answer and the way to solve this problem. it aims at integration of resources in such a way that going through a variety of sources is done with a significant benefit to the user. Total content of the database includes information on more than 1000 alloys. and cast alloys of Al. or knowledge obtained algorithmically from the process data. 2008). as the Platform is an open system. 2. there are internal tools for cataloguing the content. 2004. where individual packets of information and data.INFORMATYKA W TECHNOLOGII MATERIAŁÓW In extended IR systems. also in this case. operating via a system. this way of cataloguing the content has some important limitations. Cu. Among the knowledge bases used in the Foundry Research Institute. the user can select the appropriate cast alloy meeting these requirements. It likewise gives no possibility for defining the rating of response (documents) in relation to the request. but also from all other sources. coming not only from the literature. requires the use of a knowledge model of the treatment processes. The aim will be to provide the end user with a transparent access to the integrated knowledge base based on – 305 – COMPUTER METHODS IN MATERIALS SCIENCE . The search based on keywords does not allow accounting in the results for the natural semantics of the measure of distance between the query and a set of documents. such as description logic-based ontologies. Recent studies done at the Foundry Research Institute among experts from the world of industry led to the conclusion that the goal should be to create a platform for knowledge sharing that by giving the user an easy-to-use interface would provide him with a steady supply of current information and knowledge in the field of the metalcasting practice. these sources will need to be properly integrated and ready for processing (at least indexed for easier search).

overcoming the problem of differences in the systemic. often preventing a clear identification of the concept. The integration tool will be ontologies. Ontology is not a database schema. Alloying_elements. has been the biggest challenge for computer tools. cast steel. at the same time. understandable by both computers and humans.INFORMATYKA W TECHNOLOGII MATERIAŁÓW heterogeneous resources. Properties. syntactic. and therefore a very important issue was to create simultaneously such a language that would allow the symbolic language of logic to be written in the form of computer code. For example. and constructors. Such language for the description logic has proved to be OWL (Web Ontology Language). The basic element of representation are unary predicates corresponding to a set of objects and binary predicates mapping relationships between objects. ontologies allow integrating a wide variety of distributed sources of knowledge in a given field. which can be used to represent a domain in a formalised and structured and. Symbolic representation with directed graph of a fragment of domain ontology in the field of cast iron. HotCrack). and therefore. COMPUTER METHODS IN MATERIALS SCIENCE Fig. 1. HotCrack Hence it follows that concepts can be built from atomic ideas (casting. Directly under the parent class there are 5 main categories: Treatment. The descriptive logic allows creating definite descriptions that depict the domain using concepts (unary predicates) and roles (binary predicates). computer-processable manner. defective = having a defect). Creating a kind of shared formal language. and semantic areas which. Carbon_form. Using logic one can create definitions of concepts: Casting CastSteel Product  cast. – 306 – .CastSteel (2 defective)  defective. finding information related to this concept. The Department of Applied Informatics and Modelling at the AGH Department of Metals Engineering and Industrial Computer Science has developed a domain ontology for different cast iron grades and changes in their properties under the influence of treatment presented in figure 1. ontology is that what the entity diagram is for a database. in the DL record may look like this: Casting  made. a steel casting with hot cracks in two places. a model of a field of knowledge. but a simplification can be used: for knowledge repositories. it is a diagram. All together. 3. so far. on which the reasoning in CastWiki Platform will be based. accumulate more than 90 general terms.Mould or axioms: Alloy („everyCastSteel is Alloy”) Description logic was created for the ontology. and Cast Iron. ONTOLOGIES – MODELS IN DESCRIPTION LOGIC – APPLICATION The Descriptive or Description Logic (DL) is a subset of First Order Logic (FOL). atomic roles (made = made of.

Regulski et al. but discussion leading to the development of a final version of the problem is an integral part of the entry in Wiki.wikipedia. Thus. inherit from Wikipedia several advantages:  Wiki tools are a popular source of information and knowledge. that we allow the discussion to be started on a given topic. A user who was not a knowledge engineer might have great difficulties in understanding the operating principle of the system. and is available to everyone.org/) shows that it is possible to create a system in which each user has the ability to edit and add content. Although the OntoGRator system was solving problems of a semantic description of the area which the processing of metals is.. which the user had access to (Adrian et al. operating through the Jena API programming interface. 2007). and user had no access to them via the network interface. and millions of users around the world every day add new definitions and edit the existing ones. recording of experience and results of the creation of new (experimental) knowledge by free editing of entries. expanded by the data from external sources. Much more functional has proved to be the system that provided the ability to insert the content directly. a wiki-type tool successfully meets the most important demands of the knowledge management: it allows the codification of knowledge. This subsystem provides the data. such as databases and documents from servers located in the Internet. 2008). attempts to use ontologies for integration of knowledge in the field of metal processing have already been going on for several years in cooperation with the AGH Department of Computer Science (Dobrowolski et al. However. giving the opportunity to generate the phenomenon of externalisation of knowledge. Without basic knowledge of the ontology. The ability to create a "stub article". incomplete and uncertain. Participation in editing Wikipedia is voluntary and unpaid. The example of Wikipedia (http:// en. among others. Systems operating in this way. but being the sole repository of a company.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 3. Wikipedia's success has inspired software developers to implement industrial systems operating on the same principle. which is only a draft definition allowing for the extension. the system was becoming incomprehensible. 2007.. but still giving some information about the topic. The lack of success was due to. the mere idea of the system assumed its ability to integrate structured knowledge resources. which it has integrated. and the ability of other users to introduce their own amendments.engine that integrates data from multiple heterogeneous sources. including information on the problem area contained in domain ontology. Other . a very complex structure of the system. by creating first a short description. and at the same time a high level of quality of the accumulated knowledge is maintained through supervision and control. known as content management systems (CMS) or idea management systems.  OntoGRator Web . and by maintaining the history of discussion on a given topic can also be personalised. and at the same time contains a unique URI which is an effective way to identify concepts in the knowledge model. which consisted of the two main subsystems:  OntoGRator Engine .application in J2EE technology presenting in the form of web pages the integrated ontologies available through the Jena API programming interface. which most of internet users have already encountered and become familiar with. The OntoGRator system developed at that time allowed describing in a strict and formal manner the area that the integrated data were related with and also specifying the semantics of integrated resources. ontology. it did not enjoy the sympathy of its potential users. CastWiki Platform – 307 – COMPUTER METHODS IN MATERIALS SCIENCE At the Foundry Research Institute in Cracow..  Wiki keeps the knowledge resources constantly updated through editing. it requires minimal skills to edit and add new content. is a key aspect here. Additionally.  Wiki structure provides a description of the concepts in natural language. this assumption turned out to be ahead of its time the actual databases often did not provide any API. instead of placing the URL / URI for each resource. supports discussion on specific concepts.  Wiki technology is as simple as possible. too short to serve as a definition of the encyclopaedic nature.2.1. in the form of new. OntoGRator System 3. It is precisely in this way.

MediaWiki software was applied as a platform used by Wikipedia and made available under the GPL licence. Ontology editing tools such as Protégé permit the placement of descriptions in text form directly in the description of the OWL ontology. and photographs to CastWiki knowledge base. which the definitions in natural language are. That is why it was necessary to create a separate platform.  graphic files. L200HNM cast steel. being a component of other knowledge systems (e. They also allow user to place references in the form of a URL. Each concept (article) in CastWiki acquires its unique URI / IRI. shared conceptualisation. which can also serve as a reference to the class description in ontology. However.g.  links to all the resources available in the network (documents. animations. thus greatly facilitating the work of CastWiki editors. The system called CastWiki schematically presented in figure 2 is designed to provide a platform for the exchange of knowledge and saving the casting knowledge by specialists in the field of metal processing. it is necessary to integrate duplicate articles and create redirection of individual entries to the integrated article. it allows the integration of knowledge already stored in digital form. which forces the – 308 – . For each class in the ontology. CastExpert + etc). Description of classes in CastWiki. so-called. This gives the possibility to transfer the unstructured knowledge.g.INFORMATYKA W TECHNOLOGII MATERIAŁÓW users can participate in determining the definition of the concept. Ontology facilitates the analysis of overlapping terms through the rdf: SeeAlso property. Cocreation by multiple users makes the situation when for the same substantive term there are several articles under various entries (e. Such a scheme of action allows the collective creation of knowledge resources. a description in natural language can be added. Ontology also solves the problem of homonyms: the model itself cannot have two classes with the same name. which allows placing adequate terms directly in ontological description. which will be a list of words that share the same spelling and pronunciation with a short note about the context of each word. Wiki mechanism allows the inclusion of such types of content as:  descriptions in natural language. adding fragments of the description. Creating the "stub articles" is a voluntary activity. In this way. then mapping their structure to the underlying ontology components. Wikipedia as a public system is not an acceptable solution for companies that need to restrict access to their knowledge. the aim of which is to liberate the externalisation of knowledge by encouraging discussion on a given topic. using standard Wiki specifically for the needs of foundry plants. which is also G200CrMoNi4-3-3 cast iron).  hypertext links to other concepts in CastWiki. Knowledge in industrial plants is valuable. but also highly specialised. catalogues. The problem in Wiki-class systems is page duplication and redundancy of knowledge. images. In this situation. photographs. Another problem that is solved due to this structure of the knowledge model is the problem of COMPUTER METHODS IN MATERIALS SCIENCE Fig. The integration of these data and knowledge resources (as well as those added during the use of sources) consists in describing the resources by homonyms. INFOCAST. 2. Concepts with the same name require the creation of an additional article in CastWiki. terms (concepts) of ontology. databases) and having its own URL.

The way in which SMW represents knowledge was partly inspired by solutions such as Web Ontology Language. for which the basic Wiki version has no place in a separate article. and attributes describing logical relationships between instances. In the present work. Image space). or links to specific procedures to prevent defects).they have a counterpart in the basic mechanisms of MediaWiki as a category. it is also a tool to validate the correctness of the Wiki engine "reasoning" in respect to the annotations introduced earlier. Ontology also provides the ability to create a more structured knowledge base than the traditional wiki approach. It allows users to view the most important facts about the subject. It is worth mentioning that the SMW user interface does not require a formal interpretation of OWL DL. OWL distinguishes the relationships between data (assigning numerical value to ontological element) and between objects (the relation of two ontological elements). knowledge about how to prevent the casting defect would have to be included in the description of a particular type of defect. Namely. but also for relationships (object properties). but can be stored in some other spaces. the MediaWiki platform uses the namespaces to group the pages by content. In order to easily browse the semantic annotations found on the Wiki page. which allows performing an easy transcription from one format to another. Wiktionary and Wikinews. Semantic MediaWiki simplifies this division by aggregating all types of relationships in the namespace called Property. valuation. marking and publishing of Wiki content. quickly became the main tool for the classification of documents in Wikipedia and other Wikimedia Foundation projects. this application provides the basic mechanisms for the performance of semantic issues. Creating one’s own CastWiki system is a way to avoid this limitation. classes which are aggregates of instances with the same characteristics. a factbox is used. Category pages are grouped in the namespace of the same name. From a technical point of view. For example. Usually they are grouped in the main namespace.) by the mechanisms to improve extraction. Wikipedia does not allow the definition of relationships. not just its member classes. CastWiki allows us to create an article integrated with the preventive_means relationship so that the user can easily create a document that contains basic information about how to prevent defects. the user navigating across the ontology can easily find a definite description of relationship. cannot have a representation in the form of a Wikipedia entry. The in- – 309 – COMPUTER METHODS IN MATERIALS SCIENCE . The concept. OWL DL ontology structures can be divided into instances that represent individual elements of a particular domain. Pages of this type account for a significant portion of the data contained in Wiki.these elements are represented as regular articles. At the same time. The category system.the relationships between ontology elements have no counterpart in the MediaWiki engine. Therefore. in an ontology there is a relationship preventive_means. This form of knowledge collection provides hypertext structure of the system and ease of navigation across the resources offered to the user who needs no preliminary knowledge of the conceptual model of the system.INFORMATYKA W TECHNOLOGII MATERIAŁÓW ontology engineer to extend the class name in such a way as to give the context in accordance with the namespace. Semantic CastWiki Semantic MediaWiki (SMW) is a complex semantic extension of the MediaWiki platform (a free Wiki-type solution licensed by open source. It also provides a platform for software development. Users gain the ability to create descriptions not only for classes and instances in the ontology. which has been an integral part of the MediaWiki platform since 2004. 3. too (People space.  OWL properties . Semantic annotations that have been developed for SMW are designed in such a way as to enable a faithful export of ontologies in OWL DL format. and are supplied with the extension of SMW. This mechanism was also used in the clustering of ontology elements:  OWL individuals . search.  OWL classes . nor does it impose restrictions on expressiveness. which makes it the fastest growing project of this type in the network.3. They can be organised hierarchically in a similar way as it is done in OWL ontologies. which is a non-autonomous object. which is a basis for most of the projects such as Wikipedia. developed by the Wikimedia Foundation. For those who are supporting and complementing Wiki. while collecting all the known resources of knowledge on this subject (including specification of defects which are related in the article.

depending on the user needs. in turn. In the header of the fact table there is an eye icon. its value in different units of measurement. the search engine can also provide pages with an approximation of this value. the user will receive a list of all the sites that meet this requirement. Below the list there is a form in which one can specify any desired attribute-value pair. The script applied in SemaWiki to load the ontology was implemented using the Python Wikipediabot (pywikipedia) programming platform. contains attributes used on the page (e. allows a relatively easy optimisation. Parsing of domain ontologies – implementation of the script The recommended method of entry of the ontology into the Wiki structure is by creating one’s own Wiki parsing script.g. 3. 3. the fact table may include. Each attribute name is also a link to its site. SMW initially circulated such a possibility. From the point of view of the Wiki platform. Fact table – a description of links. Author's script. use).INFORMATYKA W TECHNOLOGII MATERIAŁÓW formation is displayed in two columns: the first. robot is a normal user with specific access rights.g. 000). but a multitude of ontological formats and change in the approach of the authors of the application to the construction of semantic structures in Wiki caused giving up the idea of a development of this tool (in most of the recently launched scenes it has been removed completely). Next to each attribute there is a special magnifying glass icon which is a link to a simple semantic search engine. 340. where one can usually find basic information about this attribute (meaning. starting from the left. crawler. if a web page is labelled [[is an alloy::Iron]]. which allows quick browsing of all the semantic annotations as presented in figure 3. It is a set of tools to automate the work on the pages of MediaWiki and other popular Wiki engines using web COMPUTER METHODS IN MATERIALS SCIENCE Fig. Taking into account the available libraries working with MediaWiki it also gives the possibility of parsing most of the popular ontology formats. If the attribute value is numeric. – 310 – . the population). for example. For example. while the second one stores the values assigned to them (e.4. Depending on the design of the attribute.

which will generate a list of all pages in the category "Cast Iron" with value "nickel" for the attribute "has alloying element". i. Each resulting subject-object pair will be represented in CastWiki as a category page. can also easily share new resources and integrate the ones already existing. attributes are not grouped under one and the same parent attribute. Of course. would be insufficient in the long run. however. Semantic search for which the attribute "has alloying element" assumes the value "Nickel". give easy access to proven best practices that exist in the company. At the same time it is possible in the process of implementing CastWiki to fill a basic knowledge base with the resources accumulated previously. the natural class name is extracted. the introduced phrases can be much more advanced.1%]] [[has content of C :: <2%]]). For example. For this task.can significantly affect the competitiveness of the casting plants.INFORMATYKA W TECHNOLOGII MATERIAŁÓW The first step is to import the necessary libraries that allow us to edit the ontology while preserving the logic graph. The resulting pages are tested under four conditions. This process is not much different from the completion of a database content in simple Wiki. so its name must begin with the keyword "Category". but in the long run it could decide about the survival and competitiveness of an industrial plant. the user can start adding pages. Such a condition. a rdflib package is used. – 311 – COMPUTER METHODS IN MATERIALS SCIENCE . Additionally.ontologies . The platform provides employees with complete information about all the resources of knowledge that are available in the organisation. to obtain the required transparency. support knowledge management and reduce the barriers of entry for companies wishing to expand the range of products. The syntax allows creating queries based on the logic of sets such as [[Category: Cast Iron]] [[has alloying element ::! Nickel]]. does not guarantee the extraction of all classes from the ontology. but it is cheap and easy to use. For this purpose. The first checks if parent category is not a class Thing (highest class in OWL. Implementation of integrated knowledge management systems. the application will ask whether to create a page to an earlier question. as well as decision support systems requires long-term investments. Such a system can be built at minimum cost . which allows taking full advantage of the knowledge accumulated in the company. In the case of attributes taking numerical value. and facilitate the development and transfer of knowledge.e. Every employee can participate in the development of a knowledge base. If the condition is not met. Then are defined namespaces used in parses ontology. The implementation of such a system can prevent employees from repeating the same job many times. It is worth noting that during transcription of ontology. or with information from other purchased systems. especially experts’ time. CastWiki must be extended by the staff. in contrast to classes. actual parsing begins. SUMMARY The designed platform is a complete. An algorithm in each iteration of the loop finds a subject-object pair joined with verb such as rdfs: subClassOf. In this way. The algorithm takes subject-object pairs combined by a relationship: rdfs: subPropertyOf. Therefore CastWiki also provides the ability of search based on formal questions. functional tool that allows for the creation in an enterprise of new channels of communication and knowledge transfer. The proposed tools . no reasoning is carried out. the child class becomes the parent class. but still not catalogued. After verifying that the robot correctly loaded the ontology and logged on to the designated Wiki. similar to the solutions used in the same tags.5. If it has not been already included in the base. Simple search by attributes. With ready-to-action ontology. This is due to the fact that. the query [[has alloying element :: Nickel]] will generate all pages. no class shall be neglected. 3.the cost is actually just the time. which takes the time. 4. Thus shaped system has a huge advantage over the dedicated systems with a ready knowledge base. all classes declared in any ontology are subordinate to the Class Thing). First in line is the class structure. there is the additional possibility of declaring the search ranges ([[has content of C :: > 0. a special syntax has been designed. Entering attributes to Wiki structure is done in the same way. it shall not be reflected in the ontology. It is above all much cheaper. from each URI. The easiest way is to enter into a standard MediaWiki search engine the name of a specific term. if a relationship is not defined directly. with the continuously expanding knowledge bases of a Wiki type.

. 2009.: Online expert system supporting casting processes in the area of diagnostics and decision-making adopted to new technologies. 83-88. A few years ago.. G. K. The role of knowledge engineering in modernisation of new metal processing technologies. ISSN 1641-8581 . Heidelberg: Springer-Verlag.fm) successfully use their own semantic solutions. From pasive to active sources of data and knowledge: decentralised information-decision system to aid foundry technologies. T. ..: Integration of the INFOCAST system databases by means of agent technology. 251–261. Issue 2.: Multiaspect character of the man-computer relationship in a diagnostic-advisory system. Centrum Informatyczne TASK (in Polish).. 2004 t. 98). K. 3.. Marcjan. V. Juliusz L. K. Computer Methods in Materials Science : quarterly / Akademia Górniczo-Hutnicza. Marcjan. 85-102. Berlin: CODATA. S.. N. Germany. S.. S. Zdzisław S. Górny. 2007.. . Kluska-Nawarecka. S. G. 2012. 2008... G. 2012 Accepted: December 12. Marcjan.. April-June. Today. Kluska-Nawarecka. Acknowledgement.Kraków : Instytut Odlewnictwa. Marcjan. S. R. Pysz. Kraków : Akademia Górniczo-Hutnicza w Krakowie. INFOBAZY’99. Jerzego J. Teresa Mroczek. R.. 2012 Received in a revised form: December 5. Cz. Kluska-Nawarecka. Kluska-Nawarecka. Kulikowski. Arhipovoj. S. 285-289. last. S. Kluska-Nawarecka. a także umożliwi wyszukiwanie pod względem właściwości (relacji). E. Scientific work financed from funds for the scientific research as an international project. ISSN 1867-5662.. K. Polish Academy of Sciences. I.2007 vol. 385390. Przedstawiony zostanie model wykorzystujący otwartą platformę Semantic Media Wiki w połączeniu z autorskim skryptem parsującym ontologię dziedzinową. mainly owing to the fact that they are based on commonly used technologies. Moskva: Rossijskij gosudarstvennyj gymanitarnyj universitet. several ontology-based Wiki platforms have been created. Centrum Informatyczne TASK. Instytut Oceanologii PAN [Polskiej Akademii Nauk]. Nawarecki... 2005. Pt.Gdańsk: CI TASK. They are justified by the relatively simple solutions that do not require a lot of time to assimilate. Kluska-Nawarecka. Połcik. OntoGRator – an intelligent access to heterogeneous knowledge sources about casting technology. Kul'by. R. 169-172 Ciszewski. 2005. Horst Kremers. S.. R. Hutnik Wiadomości Hutnicze: czasopismo naukowo-techniczne poświęcone zagadnieniom hutnictwa: organ Stowarzyszenia Inżynierów i Techników Przemysłu Hutniczego w Polsce .: Baza znormalizowanych gatónków stopów odlewniczych. Document driven ontological engineering with applications in casting defects diagnostic. Apple) and most popular social networking sites (Facebook. Among them. p. Dobrowolski.. Nawarecki. S. ed. 7 no. Dobrowolski. Połcik. the main problem associated with the implementation of similar techniques was little interest from serious investors. Z. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 312 – . Kluska-Nawarecka..: materiały konferencji / Politechnika Gdańska. R. Computer Methods in Materials Science: quarterly / Akademia Górniczo-Hutnicza. Human – computer systems interaction: backgrounds and applications 2. V. H. Gdansk 1999. S. Decision No. which were used as knowledge bases not only in IT-related companies.. S. E.: Knowledge management in casting industry processes. eds. Sobczaka. S. Regulski. A. cop. Kluska-Nawarecka. . E. 148-151 (in Polish). Within the last few years.. Katedra Informatyki. cop. many multinational companies driving development of information technology (Google. 2002.. Nawarecki. dekabr' 2008 / red. Bazy Danych dla nauki. S. Hippe.2012 R.. (in Polish).. H.: Information-decision system aiding scientists and engineers: Lecture Notes in Information Sciences. Wyszukiwanie semantyczne pozwoli rozwiązać trudności spotykane w systemach klasy Information Retrieval Systems związane z polisemią i synonimami. a nie tylko po słowach kluczowych. 820/N-Czechy/2010/0 and 0R0B0008 01 REFERENCES Adrian. B. Rossijskaâ Akademiâ Nauk [et al. INFOBAZY' 2002 – Bazy danych dla nauki: III [trzecia] krajowa konferencja naukowa: Gdańsk 24 czerwca–26 czerwca 2002 r. 56-64. Platforma ma służyć celom przemysłu przetwórstwa metali: budować zintegrowaną bazę wiedzy. Regulski. ISGI 2005: International CODATA Symposium on Generalization and Information: Berlin. 1. Wójcik. Dobrowolski. 324-328. 79 nr 3 s. 4 nr 1– 2 s. w której możliwe będzie wyszukiwanie semantyczne wspierane przez ontologię dziedzinową.Berlin. RMES\^{Z}08: problemy upravleniâ bezopasnost'û sloznyh sistem: trudy 16 mezdunarodnoj konferencii: Mockva. Regulski.(Advances in Intelligent and Soft Computing . Received: November 16. E.. As regards conversion into semantic knowledge. Marcjan. Nawarecki. pod red. Regulski.. Archives of Foundry engineering.. 2 s.. information used in foundry practice is in no way different from other data.: Application of description logic in the modelling of knowledge about the production of machine parts. E. Nawarecki. September 14–16. Innowacje w odlewnictwie. a foundry plant could find its place without any major obstacles.]. PLATFORMA SEMANTYCZNEJ INTEGRACJI I UDOSTĘPNIANIA WIEDZY TECHNOLOGICZNEJ Z ZAKRESU PRZETWÓRSTWA METALI I ODLEWNICTWA Streszczenie Artykuł prezentuje koncepcję platformy udostępniania wiedzy wykorzystującą w celach integracji model ontologiczny. .INFORMATYKA W TECHNOLOGII MATERIAŁÓW olutions proposed in this article greatly improve the search process and data distribution. S. 2002. 35/2. Kluska-Nawarecka. Mrzygłód. Volume 7. S.

The main idea of the system is processing of the production data in order to find some registered past cases of production that are similar to the present production period. Mickiewicza 30. Presented in (Sztangret et al.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. Faculty of Metals Engineering and Industrial Computer Science. Faculty of Non-Ferrous Metals.4%. The oxidizing roasting process was modeled using artificial neural networks. The paper presents preliminary results of the implementation of the CBR system to industrial control of the oxidizing roasting process of sulphide zinc concentrates. INTRODUCTION Preparation of zinc form sulfide concentrates is currently realized in the industry mainly through hydrometallurgical processes.pl Abstract The goal of presented work is an attempt to design an industrial control system that uses the production data registered in the past during the regular production cycle. Key words: industrial process control. and that is close to the observed output of the modelled object.edu.. what is presented in (Sztangret et al. Poland 2 AGH University of Science and Technology. From a point of view of optimization the oxidizing roasting process is nonlinear and multidimensional process. oxidizing roasting process of sulphide zinc concentrates. ŁUKASZ SZTANGRET1. 2 INDUSTRIAL PROCESS CONTROL WITH CASE-BASED REASONING APPROACH JAN KUSIAK1. the aim is to obtain a minimum content of sulphide sulfur in the composition of the product. 30-059 Kraków. GABRIEL ROJEK1*. Department of Applied Computer Science and Modelling al. A. Such approach is consistent with the core assumption of Case-Based Reasoning.6% and 0. During the roasting process. If the found production past case fulfills the requirements of the given quality criterion. namely that similar problems have similar solutions. 2011). The first stage of this technology is transformation of metal sulfides to oxides. By production of the roasting of sulphide concentrates of zinc heat and gases are obtained. the registered control signals corresponding to that case are considered as the pattern for the actual control. The goal of the artificial neuron network is to generate the proper output signal that depends on the input signals.. multi-agent system 1 1. Poland *Corresponding author: rojek@agh. case-based reasoning. No. PIOTR JAROSZ2 AGH University of Science and Technology. This model is based on artificial neural networks (ANN). that are processed further in the sulfuric acid plant installation. Mickiewicza 30. As the result of roasting of zinc sulfide concentrates in the fluidizedbed furnace zinc oxide (ZnO) is obtained in two fractions: fine and thicker dust of maximum content of sulphide sulfur content 0. 13. 30-059 Kraków. 2013. A. Department of Physical Chemistry and Metallurgy of Non-Ferrous Metals al. 2011) results of modelling of the oxidizing roasting process using artificial neural networks show usefulness of this apISSN 1641-8581 313 – 319 . which is called the roasting process and is carried out in fluidized bed furnaces.

which are relevant for the new case. a new experience is retained in order to be available for future reasoning concerning future problem situation. The similarity measure should induce a preference order in the case base taking into account the new. The main point of technically different known CBR systems is the CBR cycle. This phase can consist of two tasks: evaluation of solution and fault repair. what takes place at the reuse process. which is collection of past made and stored experience items. music). Every time a new problem is solved. As a result of the retain process a CBR system gains new experience due to and together COMPUTER METHODS IN MATERIALS SCIENCE – 314 – . when the solution is returned unchanged as the proposed solution for the current case. planning and human entertainement (computer games. As it is presented in (Rojek & Kusiak. 2009). 2. retrieve process is to find k-nearest-neighbor considering a specific similarity measure. The CBR cycle is common algorithm of every CBR application and consists of 4 sequential processes (or phases) (Aamodt & Plaza. instead of using any general knowledge of a problem domain. This task is usually performed outside the CBR system and makes necessary to link the CBR system with the real world domain. This process can be very simple. An example of implementation of CBR approach is optimization of autoclave loading for heat treatment of composite materials. sustained learning. should be compared and referred to other approaches to process control. recommender system in e-commerce. however there are domains. Main task in the first. Fault repair involves detecting of errors of the current solution and using failure explanation to modify the solution in a way to improve it in a way errors do not to occur. currently solved problem. where airplain parts are treated in order to get the right properties (Aamodt & Plaza. The evaluation task uses the results from applying of the suggested solution to the real environment. Reuse the information and knowledge in that case in order to solve the problem 3. Every time a new problem is solved. especially used together with evolutionary techniques in order to optimize the industrial process control. Retrieve the most similar case or cases 2. as presented in (Aamodt & Plaza. knowledge and experience management. medical applications. 2012b) Case-based reasoning (CBR) seems as one of possible techniques. 1994). Retain the parts of this experience in order to use it for future problem solving The CBR cycle starts when there is a new problem to be solved. however due to the complex nature of the modeled process. CASE-BASED REASONING The main paradigm of case-based reasoning (CBR) is reasoning by reusing of previous similar situations by solving a current problem. as it is presented in (Bergmann et al. (2) reuse the past method that constructed the solution. There are two main ways to adapt retrieved past cases to the current problem: (1) transform the past case. design. 1994).INFORMATYKA W TECHNOLOGII MATERIAŁÓW proach. A decision system with CBR approach uses the case-base. which concerns the solved problem. technical diagnosis. that can be used at industrial control. applications in image processing. At the revise process the solution generated at the reuse process is evaluated and in the case of undesired evaluation there is possibility to repair the case solution using domain-specific knowledge. applications in law. which require adaptation of solution. The preference order should enable to select one or a small number of cases. 1994): 1. the revised solution becomes available for reuse at future problem solving. When one or several similar cases are selected in the retrieved process. From the general point of view the CBR approach is relaying on experienced made in the past during solving of concrete problem situations. or cases.. called past cases. Thanks to this adding. The retain process at the CBR cycle concerns learning by retaining of current experience. a past case relevant to present problem has to be selected in the case-base and next this selected case has to be adopted to current situation. This system uses relevant earlier situations in order to give advise for the current load. Presented here research concerns analysis and implementation of CBR approach to control of the industrial process of the oxidizing roasting process. The retaining of made experiences enables incremental. what can happen by asking a teacher or performing the task in the real world. Other application areas of CBR approach are help-desk and customer service. The similarity measure can be inverse Euclidean or Hamming distance or can be specific modeled according the knowledge of the domain. what usually occurs by simply adding the revised case to the case base. Revise the proposed solution 4. the solution contained in these cases is reused to solve the current problem.

in other words. The problem is specified by measured independent signals (chemical composition of the input concentrate). Fe and S in the input concentrate) are measured only once per day. . All input signals of this process can be divided into three main groups: (1) independent signals – chemical composition of the input zinc sulphide concentrate. so the control function can be described by dependent and controllable signals registered during past production. DESIGN OF THE CBR SYSTEM FOR CONTROL OF INDUSTRIAL PROCESS The implementation of the CBR approach to the industrial process control is illustrated with example of the oxidizing roasting process of sulphide zinc concentrates as an example of typical industrial process. Controllable signals are set with the equal frequency to the frequency of dependent signals measure. what enables to evaluate the production characterized by measured values of independent signals and used control function (in the form of dependent signals and controllable signals). Presented above discussion lets to define a case as the triple problem-solution-evaluation. (2) dependent signals – measured only signals. Considering the oxidizing roasting process of sulphide zinc concentrates it is possible to state. This control function should take into consideration values of measured dependent signals and on this basis propose values of controllable signals. After the end of production day an 3. that consist of:  Problem – single values of independent signals for the whole considered production day. The goal of control of this process is to achieve the minimal concentration of sulphide sulphur in the roasted products.INFORMATYKA W TECHNOLOGII MATERIAŁÓW with regular solving of current problems. Reassuming. Such case-base should enable for designed system to imitate the manual control considering quality results which were obtained during different production days.g.g. what is a case. a case is the data structure. The retrieve phase In the domain of control of presented industrial process the main goal of the retrieve phase is to find a past case. A case relates to one single problem solved by a CBR system. In domain of presented industrial process it is possible to use past data related to manual control done in the past.  Solution – the description of used control function in the form of values of dependent and controllable signals registered during considered production day. The evaluation is represented by the average measure of concentration of sulphide sulphur in the made products during the period of using the control function specified at the solution. which concerns similar problem to the current problem and contained in this case solution is evaluated as desirable.2. the control function used to production characterized by specified independent signals. Every CBR system needs a knowledge represented in the case-base in order to propose solution for current problem. Because the chemical composition of the input concentrate is known only ones per a production day (at the beginning of a day) it is assumed that the whole day of production should be controlled in the same manner – using one single control function. From the general point of view a case represents one day of production. temperature inside the furnace. In some domains of applications continuous increase of the case base cause by the retain process causes continuous decrease of efficiency of the retrieve phase. The quality criterion is minimal concentration of sulphide sulphur in the roasted products. and (3) controllable signals – signals that can be set e.1. The control function takes as the input values of dependent signals and results in values of controllable signals. Similar previous solved problems to the current one are cases representing past production days with similar values of measured independent signals (what means similar com- – 315 – COMPUTER METHODS IN MATERIALS SCIENCE average quality measure is known. All independent signals (concentration of Zn.  Evaluation – average quality measure in the form of average concentration of sulphide sulphur in the products made during considered production day. but dependent signals are measured several time per minute. that the solved problem can be presented in the form of question: how to control the process knowing independent signals (chemical composition of the input concentrate) in order to obtain minimal concentration of sulphide sulphur in the made products. air pressure after blower. Pb. 3. The solution is. This concentration is measured several times a work day (e. that influence the nature of the process e.g. A case in the domain of industrial control The fundamental problem having the goal to design a CBR system for any domain of its use is defining. 3. 5 or7 times a day).

3. so it is not possible any fault repair process concerning present production period.INFORMATYKA W TECHNOLOGII MATERIAŁÓW position of input materials). Having goal to reuse the solution represented by the selected case. The complete functioning of the CBR system is partitioned into individual agents. that is measured as the Euclidean distance between values of independent signals measured for the current problem and the solved problems represented by the previous cases. The feedback in the case of industrial control domain is in the form of evaluations of real products made during current period of production.3. During the second step all values of dependent and controllable signals should be saved in order to be used during the retain phase. It is assumed. The past case contains description of solution in the form of dependent values and values of controllable signals. which is evaluated with the most desirable average value of quality measure. what can happen in two steps: 1. In the retain phase the current case is just add to the case base and becomes one of past cases representing experience items concerning control of the industrial process. 2. The revise phase The retain phase enables learning in the CBR cycle. Wooldridge. which was presented among others in (Weiss. It is proposed to choose first a small number of past cases representing similar problems (using k-nearest neighbor algorithm) and next to select among them only one that has best evaluation. description of the applied solution in the form of values of dependent and controllable signals saved during reuse phase and the evaluation in the form of average value of quality measure. The current case contains already the description of the problem. Select among chosen cases only one.  the Control Agent. which performs the CBR operations concerning resolving solution for control of the present production period.5. that is control of the industrial process. the control function used at the selected case has to be approximated and next has to be used in the control of present case of production. Two main types of agents are functioning in the system:  the Past Episode Agent. The prepared model of control function should be used at solving of the current problem. 2001).1. The reuse phase that is solved. In the case of industrial control of the oxidizing roasting process of sulphide zinc concentrates the quality measure is done after production time. 1999. when the current problem was solved and the evaluation of this solution is known. The retain phase COMPUTER METHODS IN MATERIALS SCIENCE In the reuse phase the solution represented by the selected past case in the previous phase should be applied to the current problem. In the first step the artificial neuron net has to be learned with values of dependent and controllable signals that are contained in the selected past case.4. that one Past Episode Agents represents one past case. 3. 3. Due to the fact. what can be done in two steps: 1. Control of oxidizing roasting process The revise phase assures feedback of the applied in the reuse phase solution into the current problem Presented above analysis concerning CBR system in the domain of industrial control of the oxidizing roasting process of sulphide zinc concentrates is implemented using agent technology. IMPLEMENTATION OF THE CBR SYSTEM 4. A model of the control function relevant for the selected past case should be prepared with the use of values of dependent and controllable signals. 2. The Past Episode Agent can receive messages concerning represented past case and answer to such questions providing information – 316 – . In the second step the learned net has to be used in order to predict values of controllable signals on the base of presently measured dependent signals. which represents one past case. An artificial neuron nets can be used by modeling and using of control function. what enables the currently ended present case to be available for reuse in future problem solving process. This phase starts. that controllable signals are function (named control function) of dependent values. This evaluation has to be made outside computer system and is usually equivalent of quality measure (made by human). 4. the number of Past Episode Agents is equal to the number of past cases contained in case-base. Choose a number of previous cases from the case base with the highest similarity rate.

Every member of the swarm searches in its neighbourhood but also follow the others. that are successively resolved and stored in order to be used in future. second from those five selected only one is chosen. PSO method is inspired by the behaviour of swarms of birds. The elaborated ANN model is based on the architecture of MultiLayer Perceptron (MLP). After the ANN was trained. As a optimization method. Main problems of shown in (Sun. 7 neurons. Implementing this approach to the considered oxidizing roasting process. one case. All neurons are located in 4 layers composed of 9. composed of the measurements of the roasting process and the resulting concentration. Due to mentioned problem with evaluation of made products under control of presented system. that represents the best evaluated production (as presented in subsection 3. In the retrieve phase similar past cases to the current one are searched with the same method. The number of past cases is changing due to new problems. For this step supervised learning method is used. that represent. it can be used as the model. 2008) implementation of the CBR methodology to control of combustion control of blast furnace stoves seems analogous to presented above work. The used ANN first should be trained in order to predict the concentration. Using of agent technology allows to overcome many development problems. as in previous works concerning industrial control presented in (Rojek et al. Presented in section 3 work concerns a case as a whole day of production. The reuse phase is much simpler due to the fact. Such constriction of case-base can be simple transferred to multiple agents. 2008) very shortly. In the algorithm based on this behaviour. usually the best member of the swarm. as it was proposed. 2008) application are related to definition of representation of a case. 4. that is relevant for the current production period. which is used by a optimization procedure. that represent the most similar production periods concerning independent signals. insects or fish shoal looking for food or shelter. neuron nets) in order to obtain temporary control decisions. By the modeling step a supervised learning is used with usage of data represented by the selected in previous phase case. that appear by implementation of CBR systems. In the reused phase the Control Agent uses an artificial neural network (as presented in subsection 3.2.INFORMATYKA W TECHNOLOGII MATERIAŁÓW – 317 – COMPUTER METHODS IN MATERIALS SCIENCE concerning description of problem. This evaluation would be not a problem. solution or evaluation represented by the past case. case-base and all phases of implemented CBR phases. Combustion control of blast furnace stoves Presented in (Sun. The goal of optimization is to obtain values of control signals which provide minimal concentration of sulphide sulphur in a roasted ore. was not a handicap at implementation of the presented system. Lack of evolution. that cases are evaluated later and added to case-base for future problem solving. the swarm is consid- . Rojek & Kusiak.3). the artificial neural network is used as a model for prediction of concentration of sulphide sulphur in roasted ore. the revise and the retain phases are not implemented. that a case represents only one moment of time. when implemented system will be really a control system for the real control system. This neural network is a multilayered perceptron composed of neurons with sigmoid function. The CBR methodology relays on using of information. 2012b). It is assumed. 5. what induces using approximation methods (e. The solution represented in the relevant case is just taken directly as the final control decision.g. The Control Agent performs CBR operations. what is similar to research presented in subsections 3. This selection is made by agents communicating in the system: first 5 Past Episode Agents are selected. The dataset used at training contains records. however. as presented in subsection 3. 13. that is solution for control of the present production period.5. particle swarm optimization (PSO) is used.4-3. that is just set of distributed cases concerning past situations. By implementation of the presented system JAVA and JADE (framework for agent systems) are used. The main problem that we were faced by implementation of the software concerns the revise and retain phase. ARTIFICIAL NEURAL NETWORK BASED CONTROL SYSTEM Other approach to control of an industrial process uses a model and a optimization procedure (figure 1).2). which require the real evaluation of proposed solution. This transformation maintains natural decomposition of problem through use of the agent technology.2 (k-nearest neighbor algorithm and the Euclidean distance). The revise and retain phase are presented in (Sun.. 2011. 11. In the retrieve phase this agent finds one past case. that aim to control of oxidizing roasting process. that is contained in case-base.

1999. P. 2010. Szeliga D. Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence. Ao. Rojek. 400–408. Rauch. Proc. 2012a. Bio-inspired optimization strategies in control of copper flash smelting process. vol. L. simultaneously is gaining its experience by adding current problems and its solutions to the case base. Kusiak. Industrial control system based on data processing. AICom . From general point of view. which is used at currently solved problems. Stanisławczyk. G. I. Plaza. Rojek. John Wiley & Sons. Scheme of control system using an ANN model and a optimization method. Sztangret.. work of such complete system can be analogous to work of an worker. Althoff.. Modelling of the oxidizing roasting process of sulphide zinc concentrates using the artificial neural networks.. Małecki S. Such CBR system will not only use its experience. and System Approaches.11.. 1. 2009. Conf. More accurate description of this method can be found in (Sztangret et al. J. Computer Methods in Materials Science. D. if those decision bring desirable results. 2009.. który w trakcie bieżącego cyklu produkcyjnego wykorzystuje zarejestrowane w przeszłości dane produkcyjne.. 122-127. Kunstliche Intelligenz: Special Issue on Case-Based Reasoning.. 12..Artificial Intelligence Communications. Sztangret. Sztangret Ł.... eds. that gains and uses experience according to made decisions. Case-Based Reasoning: Foundational Issues. Case-Based Reasoning – Introduction and Recent Developments. K. 6. 2012b. Analiza wrażliwości jako metoda wspomagająca optymalizację parametrów procesów metalurgicznych. Minor.. IMECS 2008. J.. J. Such implementation involves many design decisions according to representation of a case. that is information of previous made decisions. 297-302. REFERENCES Aamodt... 721-725 (in Polish). Ł.. ICAISC 2012.. 2008. Głównym założeniem systemu jest przetwarzanie danych produkcyjnych w celu znalezienia zarejestrowanych przeszłych przypadków produkcji. Rutkowski. Acknowledgment. Hutnik – Wiadomości Hutnicze. 9. eds. E. I. The experience is in the form of case-base. K. USA. J.INFORMATYKA W TECHNOLOGII MATERIAŁÓW ered as particles representing single solutions. Introduction to Multiagent Systems. Presented in this article research shows. R. MIT Press Cambridge. Inc. Particles move through decision space and remember the best position they ever had.. 39-59. USA. Kusiak. Conf. Kusiak.. M. M. Financial assistance of the MNiSzW (Działalność statutowa AGH nr 11. Agent-based information processing in a domain of the industrial process optimization. New York.. Weiss. 5-11. 2010). 25-28. as it is presented as the main fea- COMPUTER METHODS IN MATERIALS SCIENCE STEROWANIE PROCESÓW PRZEMYSŁOWYCH Z PODEJŚCIEM OPARTYM NA WNIOSKOWANIU EPIZODYCZNYM Streszczenie Celem prezentowanej pracy jest próba zaprojektowania systemu sterowania przemysłowego. The CBR system uses experience.. submitted and accepted to Computer Methods in Material Science.. CBR Application in Combustion Control of Blast Furnace Stoves. 2009) and (Sztangret et al. Kusiak. Reichle. Hong Kong. 23. 1994. Kusiak J.. Future works should be oriented to implementation of CBR approach to real control of industrial process. 2011. The decision is made on the basis of previous made decisions. that it is possible to implement the CBR methodology to the domain of control of industrial process. construction of case-base and development of the whole CBR cycle. Only such implementation will enable to obtain real evaluation of made decisions according to control of production. 2011. Rojek. J.110. 11. Ł. 7. G. Computer Methods in Materials Science. Each particle is characterized by its own position and the velocity.. A. G.. S.. Bach. Fig. the revise and retain phases will be possible to realize and finally it will be possible to close the CBR cycle. które są podobne do bieżącego okresu produkcji.. M. G. Sun. J. Zakopane. All that decision have to relate to the real industrial process. 2001.. Computer Methods in Materials Science. A.. Wooldridge. Jeśli znaleziona w bazie przeszłych – 318 – . 502-510. Bergmann. If the evaluation of control made by CBR system will be known. Kusiak. ture of Case-base reasoning.. Ł. Methodological Variations. 11. CONCLUSIONS Case-base reasoning approach enables to make decision in the case of unknown model of domain of implementation.. Proc. but also will gain experience. Case-Based Reasoning Approach to Control of Industrial Processes. Ł.085). Sztangret... Jarosz. that is controlled. The CBR system together with solving problems.

że podobne problemy mają podobne rozwiązania. uważane są za wzór dla bieżącego sterowania.INFORMATYKA W TECHNOLOGII MATERIAŁÓW przypadków produkcja spełnia wymagania danego kryterium jakości. 2012 – 319 – COMPUTER METHODS IN MATERIALS SCIENCE . którym jest proces utleniającego prażenia koncentratów siarczkowych cynku. Takie podejście jest zgodne z podstawowym założeniem wnioskowania epizodycznego (ang. które odpowiadają znalezionemu przypadku. zarejestrowane wartości sygnałów sterujących. którym jest stwierdzenie. W pracy przedstawiono wstępne wyniki wdrożenia systemu CBR do sterowania procesu przemysłowego. Received: December 4. 2012 Accepted: December 28. 2012 Received in a revised form: December 18. CaseBased Reasoning).

(Rutkowski. welding 1 1.edu. 2005). Poland. DANUTA SZELIGA1.metal. No. The procedure of simplified decision support model formulation – including expert knowledge externalization. knowledge base.045. 2013.pl Abstract A case study of rule knowledge base on developing stress relaxation in welded components is presented in the paper. and finally during model evaluation. Inconel 718 and Steel 410 are considered in the numerical simulation of welding and heat treatment processes for manufacturing of turbine engine. but also with the knowledge engineering.Rzeszów" S. Poland 2 WSK "PZL . Hetmańska 120. stress-relief annealing. RAFAŁ PUC2 AGH University of Science and Technology. The aim of this research is to propose a set of control rules aiming at decreasing the stress in welded components on the basis of WSK Rzeszow spe- cialists’ technology knowledge. such as: Inconel 625. The design of the model requires the determination of a model scope. Rzeszów. Mickiewicza 30. not only at the stage when the model is defined. formalization of knowledge. Developed application can be implemented in one of the reasoning system shells. 2 RULE-BASED SIMPLIFIED PROCEDURE FOR MODELING OF STRESS RELAXATION KRZYSZTOF REGULSKI1*. JACEK ROŃDA1. A decision tree was used to support creating and visualizing of the model. SOURCES OF TECHNOLOGICAL KNOWLEDGE UTILIZED IN THE MODEL DEVELOPMENT 2. the selection of variables and identifying domain of attributes is described. ANDRZEJ KUŹNIAR2. decision support. This process also requires a collaboration of experts from various areas of expertise not only from manufacturing. the decision criterions and appropriate selection of independent variables. Cracow. etc.agh. Various materials.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. An inference stress relaxation model is developed in collaboration with welding experts from WSK Rzeszow. CHARACTERISTICS OF THE RESEARCH WORK The inference and the development of control models proceeds in several steps and is an iterative process where specialists perform the role of experts. computer science.  inference rules are proposed. Key words: rule-based system. 30-059 Krakow. 35. but also on the stage when  variables of a process are selected. 13.1. 2. Literature review The knowledge base for the assessment of weldment integrity can be based on the measurement ISSN 1641-8581 320 – 325 . The result of the work is a set of rules constituting a simplified model to predict the stress relaxation parameters of stress-relief annealing applied to welded components of PZL-10W helicopter engine produced by WSK Rzeszów. Poland *Corresponding author: regulski@tempus.A.  a variable domain is determined..

Kluska-Nawarecka & Regulski. Mrzygłód & Regulski. 3.. Nawarecki et al. Annealing the single parts of a case and components is carried out in a furnace with a fixture for maintaining shape and dimensions of elements in the risk of deformation due to thermal dilatation. e.  The diagnostic model: determining the causes of defects in the process of heat treatment. 2. after literature study and discussions with engineers from WSK Rzeszów the major tasks were defined and the objectives were identified with selection of decision making criterions and appropriate variables. Each of the engine elements has the exact specification in the manufacturing process output. Cooling a furnace with a specific rate up to the minimum temperature and further cooling a case up to the ambient temperature. The most important parameters of stress-relief annealing are: an alloy composition. 2011.. The annealing temperature lowering could lead to insufficient stress relaxation. 2007. but inner specifications are confidential. Two of the standards: American (AMS) and international (ISO) are generally available. Heat treatment of turbine engine case is carried out after welding. Kluska-Nawarecka et al. Szeliga. There are several variants of such study based on:  The prediction model: determining the quality of the residual stress after heat treatment on the basis of welding. Stress relief annealing leads to some stress relaxation and also can restore ductility in brittle zones. complexity of the shape and size of a product. 4. 2.2. 2012. Internal procedures and standards Authors’ preliminary knowledge in the area of rule system implementation is priceless in the preparatory phase of a research project (KluskaNawarecka et al. Thermal annealing reduces the yield strength at elevated temperatures that results in the occurrence of plastic deformation in areas. Issues related to the relaxation of stress after welding has been discussed in numerous publications. Those confidential specifications are available for the study but the knowledge base and decision making models derived from that specification are still the property of the WSK Rzeszów.g. 2009. Securing of required level of a vacuum. so called inference objects.. – 321 – COMPUTER METHODS IN MATERIALS SCIENCE . annealing parameters and workpiece data. 2007. Knowledge base and experience of executive team  The decision-making control model: determination of heat treatment parameters on the basis of the workpiece.INFORMATYKA W TECHNOLOGII MATERIAŁÓW of residual stress and density of micro-cracks. During the problem formulation. industrial practice and observations recorded by technologists that can be collected together and formally coded in the form of data bases. i. A stress-relief annealing is the most common method for removing internal stress.  The decision support model: determination of heat treatment parameters on the basis of the expected properties of the material after annealing process. by Tasak (2008). It may also be used during engine overhaul when a case is repaired by welding.. 2012. Szeliga et al. Typical heat treatment of a case consists of a vacuum annealing conducted following the procedure: 1. Finally authors followed the fourth option. a repaired element must be rewelded. annealing temperature and cooling rate. and Pilarczyk (1983). That situation is usually preferred in the case of stress-relieving elements that are required to have good strength properties. The specifications are established according to materials treatment standards. Continuous heating of a chamber to the annealing temperature and maintenance of this temperature for a specified period. 2.3. Heating of a furnace to a required temperature. Rules supporting a decision making process will be identified as the data base motor. The same device is use in a case maintenance procedure. When the control procedure shows that weld cracks exceed security limits. heating rate. Some depreciation of this vacuum state is permissible during the process. The first step in developing a decision making support system is to determine the inference object. decision tables. Following that. where the second invariant of internal stresses exceeds a local yield limit. decision making control model.e. Increasing the annealing temperature reduces the limits of yield and the strength of steel. 2012). decision trees and the rules are presented here without confidential details. The manufacturing knowledge base available in the WSK Rzeszow existing in non-formal documentations.

whether the fixture usage is necessary during a heat treatment? This decision is important from economical point of view. the introduced formalism should be clearly understood by a production engineer who can evaluate and verify the knowledge. Usually.  heating rate.1. An attachment and detachment in a strengthening should be easy and unambiguous. Several methods were used to perform that process by developing the following objects: decision trees. other sources of the knowledge play only auxiliary functions. 3. A production engineer is making a decision on which surfaces a fixture should applied. To eliminate thermal deflection after a heat treatment various repair operations are used. Selection of inference parameters tionale. A fixture strengthening a workpiece should be used for flexible parts. The base surface is often the outer cylindrical surface and a cylinder base. determining which of them will be used in relations and which will be evaluated in the inference process. Variables can be selected according to the future application. For example: a variable specification can be a result of conclusion of the inference in the first phase and later may appear as a variable in further rules. such as straightening operation for bars and tubes or spinning for the cylindrical parts. To develop the control model of stress relaxation. 3. In the later stages of the work. Such information is already included among others in the variable COMPUTER METHODS IN MATERIALS SCIENCE – 322 – . variables in rules for concluding. which ranges to several kilograms. Moreover. Production engineering experts may selected the following variables in the decision making process:  specification of materials. In this model some of variables could appear both as a result of conclusion or a ra- One of steps of complete decision-making model is to answer the question related to maintenance of a work piece shape. it is impossible to develop a control model and propose rules of inference. the knowledge of production engineers were codified following a cycle of interviews and the model assessment. the oven to reach the same annealing temperature must be heated longer than for the case of heat treatment without such device. Non-formalized knowledge based on industrial experience is called by managers „tacit knowledge” and the know-how. A domain of each parameter/variable is determined as a field attribute. The major task of the engineer of knowledge is the externalization of the experience of experts. Without the knowledge of the practical aspects of decisionmaking in an industrial environment. decision tables. The objective of coding is to represent a production engineering knowledge in such a way that it can be implemented into the inference process. absorbs heat and therefore. WSK base of knowledge The information available from production engineers appointed by WSK Rzeszów is valuable source of knowledge for the development of process control model. and rules of inference.  number of stops during the heating period for a supply of argon. Variables consisting so called decision table does not distinguish between types of support surfaces. The final heat treatment process is marked by the acronym. cylindrical or flat. and there is high risk of deflection then to secure the shape within strictly prescribed tolerances a strengthening fixation must be used. Some parts could be supported simultaneously on various surfaces.e. Dimensional tolerances and geometrical tolerances are controlled after heat treatment. because mass of a fixture. KNOWLEDGE CODING FOR RULEBASED EXPERT SYSTEMS The knowledge coding is the next step in the design process of the control model for welding.  necessity of fixture usage during stress-relief annealing.2. The number of variables can be proposed. Decision making tables The first step in the design of the decision making system is the identification of parameters i. When such operations are final. The cycle was repeated several times that helps in avoiding modeling errors and identification of decision rules.g. HT. regardless of whether they will be used or not in the final model.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 2. without familiarity with the most common dilemmas emerging technology.4. 3.e. e. those are outer surfaces of a workpiece. Diameters of a workpiece are considered together with dimensional tolerances and shape requirements such as flatness and roundness. i.

then a strengthening fixation must be applied. It can be read as follows:  scenario 1 – when a heat treatment is in interoperation stage and tolerances of the supporting surface are loose. These rules are shortened to include only the necessary conditions for achieving an inference. this parameter could be omit– 323 – COMPUTER METHODS IN MATERIALS SCIENCE IF material = “Inconel 718” ing_stage = “final” AND treat- . and a deflection may could exceed a tolerance. Rules of inference Inference control rules can be generated on the basis of a decision tree. the value of the parameter 'specifications' is presented only by the appropriate acronym or symbol (see figure 1). the authors decide to omit a number of steps leading to the refinement of relations and inference rules. The entire decision tree makes depicts rules in a comprehensive manner. To describe the idea of the decision tree only a small portion of a full model is presented here only for Steel 410. For example.3. ted. For NO NO YES YES NO YES the sake of simplicity. selected few rules are the following: THEN specification = “Inc718a” IF treating_stage = “final” AND tolerance = “tight” AND deformation = “significant” THEN apparatus = “yes” IF specification = “Inc718a” AND heating_speed = “6°C/min” THEN number_of_stops = 0 Fig. This redundancy will be eliminated following the future process analysis. 1. shown as the table 1. 3. The decision table.INFORMATYKA W TECHNOLOGII MATERIAŁÓW named: "deformations". assigns a number of decision rules to the heat process scenarios. internal final However. Construction of decision trees Table 1. A part of the decision tree to determine the heat treatment parameters on the basis of a workpiece-decision-making control model. 3. The decision table showing a requirement for a strengthening fixation during stress-relief annealing. results of such acquiloose tight loose tight sition are gathered collectively YES NO YES YES NO YES in the decision-making table. IF specification = “625” AND heating_speed = “8°C/min” AND exploitation_treatement = TRUE THEN number_of_stops = 2 This rule can be avoided. Possible scenarios: TREATING STAGE TOLERANCES OF SUPPORTING SURFACES ANTICIPATED DEFORMATIONS Use of stress-relief annealing apparatus source: own study The process of knowledge acquisition from experts is much more laborious than it would 1 2 3 4 5 6 appear following this study. as in this particular scenario the information on thermal deflections is redundant. Since the whole tree exceeds one page and the information is confidential. because a decision is made on the based on the information about the heat treatment stage and appropriate tolerances.4. Therefore. The rationales for each decision making should be described for each manufacturing process scenario.

Marcjan. utilization of strengthening fixture during annealing.. Rębiasz B. Archives of Foundry Engineering.  codification of the expert knowledge on the present-day process experience and ways for decisions of heat treatment parameters for selected materials and components. The final step of an “expert” system development consists of the derivation of rules for the control of a stress relaxation process. Regulski.. Knowledge management in material technology support systems. It implements an engine of inference and give an opportunity of developing a knowledge base.. titled SPAW no. Stawowy A. the presented scheme should be implemented in one of inference systems. Business rules management : perspectives for application in technology management. Knowledge base consists on several dozens of variables and tens of rules. 2007. 169-172.. Madej Ł... Adrian. ZPB/33/63903/IT2/10 . Set of such rules would be used in future by process engineers for supporting decisions for the design of stress-relief annealing.  codification and implementation of manufacturing know-how in a decision table. Śliwa Z. Research team decided to apply forward inference. S.. AGH— ISBN 978-83-932904-0-6.. Puka R.. SUMMARY The task was performed with the following works:  acquisition of the knowledge from the best practice of production engineering in WSK regarding e. Stawowy A. Duda J.2010-06-01.g... The dialogue can be advanced as follows: MODEL: What kind of parts made of? USER: Steel 410 MODEL: Treating stage is final or internal? USER: Final MODEL-CONCLUSION: You should use Specification of parameters sign Spec410a MODEL: Is tolerance of dimensions tight or loose? USER: Loose MODEL: Is the risk of deformation significant? USER: Yes MODEL-CONCLUSION: You should use elements geometry sustaining apparatus during stress-relief annealing MODEL-CONCLUSION: You should apply heat rate at 6 celsius degrees per minute MODEL-CONCLUSION: You should plan 1 stop for argon application a material are the COMPUTER METHODS IN MATERIALS SCIENCE 4. Paradowska W. Gaweł B. 2007. — Kraków. 3. For further application. 7.. ReBIT is the Business Rules Management Systems which combines the capabilities of the rule-based decision support system with the expressiveness available in algorithmic programming languages. Acknowledgements. The system was developed in the Department of Applied Computer Science of the Faculty of Management AGH in Krakow. Baster B.. Kluska-Nawarecka. Those rules are also ready to be implemented in one of expert system shells.. The model was sent to the WSK Rzeszow for evaluation. 2011.. The role of knowledge engineering in modernisation of new metal processing technologies. 2011). 5. stress relaxation after welding of engine case and its components. Nowak J. Wrona R.5. Decision variables are constructed from: specification (which is a set of 16 – 324 – . Macioł A. A. and number of breaks application of an argon dose. Developed knowledge base is a functional model of stress relaxation control. eds. K... and control rules of inference. Jędrusik S. heating speed and number of stops. Jankowski R...INFORMATYKA W TECHNOLOGII MATERIAŁÓW Using this type of rules production engineer can decide about indicated variables such as: specification of materials.. Example of use case parameters of stress-relief annealing). The whole system with developed knowledge base was successfully implemented in WSK Rzeszów. REFERENCES Banet E. a requirement for a strengthening fixation. The mayor result of this paper is an attempt to codification of previously informal expert knowledge and a proposition of inference rules. Prob- Reading a decision tree or applying some rules of inference in a daily routine can be presented in a form of dialogue with a user. Kluska-Nawarecka... Macioł A. This research was carried on within the research project of NCBiR. heating rate. Paliński A. APPLICATION OF KNOWLEDGE MODEL Presented in the paper rules was implemented in the system ReBIT (Banet et al.. S.. R.. Macioł P. System acts as expert system shell. Pilch A. decision tree.

(in Polish). Kuziak. Warszawa. 451–467. S. Wydawnictwo Naukowe PWN. Tasak.. Kluska-Nawarecka. B. Jako narzędzie wspomagające proces tworzenia i wizualizacji modelu wykorzystano także drzewo decyzyjne. D. Pilarczyk. Regulski. Brno. Model taki może być w przyszłości implementowany w systemach wnioskujących. Kraków: Instytut Odlewnictwa. vol. 1983. Warszawa (in Polish)... E. 2009. K. Podvysotskyy. Hutnik Wiadomości Hutnicze. 2012. (in Polish). Application of description logic in the modelling of knowledge about the production of machine parts. Pysz. K. Opisana jest procedura powstawania uproszczonego modelu wspomagającego podejmowanie decyzji – obejmująca eksternalizację wiedzy ekspertów. Poradnik inżyniera – Spawalnictwo. 2005. Innowacje w odlewnictwie. Methods and techniques of artificial intelligence. UPROSZCZONA PROCEDURA MODELOWANIA RELAKSACJI NAPRĘŻEŃ OPARTA O SYSTEM REGUŁOWY Streszczenie Artykuł ma na celu przedstawienie studium przypadku tworzenia regułowej bazy wiedzy w zakresie relaksacji naprężeń powstających w elementach spawanych. R. dobór zmiennych i określanie dziedzin atrybutów. Górny. Rheological model of Cu based alloys accounting for the preheating prior to deformation.. Advances in Intelligent and Soft Computing. Regulski... Regulski. Rutkowski.INFORMATYKA W TECHNOLOGII MATERIAŁÓW lems of Mechanical Engineering and Robotics. 2012.. 2012. 21st International Conference on Metallurgy and Materials. 2012 – 325 – COMPUTER METHODS IN MATERIALS SCIENCE . 73– 86. Sobczak J. CD ROM. (in Polish). S. Krakow (in Polish). Wydawnictwo: Naukowo – Techniczne. 2012 Accepted: November 21.. Archives of Civil and Mechanical Engineering.. 11.. 2008.. Z. (in Polish). Kluska-Nawarecka. S... ed.— 251– 261. 2011. 148–151. Szeliga. 36. Nawarecki. D. Design of the continuous annealing process for multiphase steel strips. Received: September 20... 2012 Received in a revised form: November 4. 79. Metalurgia spawania. Mrzygłód. M. Wynikiem prac jest zestaw reguł stanowiący uproszczony model. formalizację wiedzy. który na podstawie wartości kilku zmiennych zdefiniowanych przez użytkownika określa parametry procesu wyżarzania odprężającego stosowanego w WSK Rzeszów do produkcji części składowych silników śmigłowcowych PZL-10W. Szeliga. J. On-line expert system supporting casting processes in the area of diagnostics and decision-making adopted to new technologies.3. Pietrzyk. V. K. Multi-aspect character of the man-computer relationship in a diagnostic-advisory system. L. Wydawnictwo: Biuro Ekspertyz i Doradztwa Technicznego "Techmateks". E...

each stage of the hemming process can effectively eliminate or at least minimise the majority of defects caused by metal deformation. sheet metal hemming. a new method proposed by the author for the reconstruction of a 3D outline of the hemmed sample was disclosed along with a technique to calculate the value of strain on its surface. remembering that it affects the final product quality. a test stand designed for the practical implementation of a three-step hemming process was presented. the whole range of parameters governing the process of forming should be subject to very carefully monitoring. Warsaw/Poland *Corresponding author: s. the mechanism of the hemming process is much more complex than it might be judged from the description of a pure bending process. Among the different vision systems available. warp and recoil described by Livatyaliet ISSN 1641-8581 326 – 332 . i. In the introduction. It is a well known fact that properly designed and performed. Key words: strain analysis. Warsaw University of Technology. or folds and fissures. experimental analysis. the end part of the sheet rolled over to the inside onto itself forms an angle of 180 degrees with the remaining base part of the sheet. information was provided about the need to use machine vision systems to solve problems that occur in the process of hemming. One of the example of the hemming process includes the door hinges. No. followed by pre-hemming and final hemming classified by Muderrisoglu et al. a portable measurement system for quality control of the hemmed surface edges was shown for the industrial application. An optical system for studies of the measurement technique and a method of image analysis used in the described example of the plastic forming process were presented. 2 EXPERIMENTAL APPARATUS FOR SHEET METAL HEMMING ANALYSIS SŁAWOMIR ŚWIŁŁO Faculty of Production Engineering.swillo@wip. and is used for two purposes: (i) to join together two parts of the sheet metal.pl Abstract A new portable system for experimental investigation in the process of sheet metal hemming was developed.pw. where different deformation mechanisms are operating. Then. In this process. and it is analysed next to obtain information about an outline of the deformed line. (1996). the test sample image is recorded. Finally. For this reason. Further in the text.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. the rollin. INTRODUCTION The subject of the paper relates to the proposed complex solution for quantitative and qualitative control of a three-step hemming process (figure 1). 13.e. where one sheet fills a gap between the bent edges of another sheet. and (ii) as a finishing operation by which the raw sheet edge is hidden inside the item shaped. Yet. 2013. mainly in terms of gaps existing between the rolled over edges of adjacent components. vision based measurement 1.edu. first. This process is applied in the final stage of the car body production. Hemming is the sheet metal forming process. which consists in flanging. At the final stage of the process. a method using laser light scanning for reconstruction of the geometry of the examined hemmed sample was selected. roll-out.

Experimental apparatus for hemming process: a) schematic of the designer apparatus. guide sleeves fixed in the top plate. This solution allows precise determination of the forming process parameters. the use of punch guides in the device allows precise control of the punch travel with the possibility to measure the clearance between the die and the punch. (2005 and 2006). a comprehensive understanding of the process itself is necessary to which hitherto not much attention has been paid. 1. Fig. and therefore has been used in the studies of a numerical modelling of the hemming process. the proposed method using a vision system will allow an immediate analysis of the finished product. 2. In contrast to the timeconsuming and less accurate methods of assessment based on an optical system. 2. to achieve better insight into the forming process. The forming elements include a die with an option allowing changes in the bending radius and a punch with an option allowing changes in the pre-hemming angle. b) hemming tool. (2000). This can be achieved by improving the already existing techniques or developing the entirely new ones. The device consists of the following main parts: two columns fixed in the bottom plate. The material used in the hemming test was aluminium sheet (A1050). the process kinematics. to minimise errors occurring when the process of hemming is designed. b) pre-hemming. the process run and data acquisition from the force sensor. and a forming tool. Figure 2b shows the measurement stand for the hemming tests equipped with two systems: a vision system for recording of Fig. a special hydraulic press (applied previously in stud– 327 – COMPUTER METHODS IN MATERIALS SCIENCE . and the geometry and surface quality obtainable in the operation of sheet metal hemming on the basis on research performed by Swillo et al. Due to this it will be possible to evaluate the results of an inspection through comparison of model results with the experimental data generated by a specially developed vision system. The ultimate goal is to determine experimentally the process limit parameters. Therefore. To study the hemming process (curve surface and curve edge). and a displacement sensor.INFORMATYKA W TECHNOLOGII MATERIAŁÓW et al. Moreover. an option has also been provided for the quick setup of the device (figure 2a). The studies will enable quick and accurate analysis of the process of hemming for any geometric and material-related parameters found in the selected segments of a car body. EXPERIMENTAL APPARATUS A special column-like shaped device was designed and built to perform the hemming process. c) final-hemming. Schematics of the three steps hemming process: a) flanging. The concept of the device for the hemming process assumes an easy replacement of the forming elements.

where the bending angle is between 20 to 180 degrees. The studies discussed above.  change in the shaped product geometry. 3.1. b) stationary vision system. 2000 and Graf & Hosford. METHODOLOGY OF HEMMING ANALYSIS Based on the industrial experience and numerous research results. geometry and process parameters. The methods used commonly in the industry to assess the quality of such products are either visual methods or methods using simple de- was to design and manufacture of a portable vision measurement system to allow inspection of the chosen three parameters. as reported in many works on this subject. Due to its high stiffness. Used in combination with the displacement sensor attached to the press. is based on a previously proposed solution by Swillo et al. Therefore. 3. by Świłło et al. This statement leads to the conclusion that it is necessary to develop a vision system based on an advanced method of measurement and control of all process parameters. (2011. co-called ALM (Angle Line Method). and thus indirectly the quality and functionality of the whole car. 2012) was used. 3. Thus. This method COMPUTER METHODS IN MATERIALS SCIENCE – 328 – . the maximum deformation will be concentrated in an area of the size of micrometres. the selected method to measure deformation should be characterised by both high resolution and high accuracy with the ability to determine the deformation in different variations of the hemming process (the curved surface and the curved hemmed edge . 1994) clearly indicate the need to search for proper relationships between the product quality. It should be remembered that all of the above indicated features directly affect the final evaluation of the product quality and functionality. All these parameters can be used in a comprehensive assessment of the hemming process quality referred to the structural parts of a car body design. The measurements were carried out in a Matlab/ Simulink environment that allows for block construction of the performed measurement tasks based on image processing and data acquisition originally proposed by Higham and Higham (2005). the press allowed obtaining similar forming conditions as the conditions used during industrial hemming of the sheet metal. is a serious problem due to high localisation of the non-linear in nature distribution of deformation. Consequently. (1999). and a number of other related works (Livatyali et al. that is:  deformation during forming. it enabled full control of the hemming force in function of the punch position.  quality of the shaped surface. and Chen et al. vices such as a feeler gauge. determined by the presence or absence of such defects as cracks or folds in the sheets. Geometry and deformation measuremen The measurement of deformation in a bent sample. the aim of the project is to use all the three main parameters of the process. it becomes obvious that the final quality in the hemming process depends on a complex interaction between the material properties.. Therefore. The repeatability in such cases is unsatisfactory. In samples with a thickness of 1 mm. Vision based measurement for hemming: a) schematics for the strain and geometry analysis. The use of these three parameters demands further development and implementation of advanced measurement techniques allowing for their full identification.3D). (2005). the strain measurement algorithm. and the geometry of the hemming process. the final task of the running project Fig. presented by the author in this study. The hemming test was carried out using a force measurement system in the form of an axial strain gauge mounted on the model press. large working space.INFORMATYKA W TECHNOLOGII MATERIAŁÓW ies of the flat surface and straight edge. the maximum pressure of 40kN and low punch operating speed during forming.

Then. 3. Next. Świłło et al. which can be written with just any pen. – 329 – COMPUTER METHODS IN MATERIALS SCIENCE . Fig. while maximum deformation covered the area of a width not exceeding 50 m. where the strain was measured in the hemmed sample in an area of 0. which in the case of the proposed method has no major restrictions on account of the pattern geometry used. grid and ALM. Micro-crack formation measurement for the hemmed surface. In particularly. Next a grid circle where several objects (ellipses) was recorded and analyzed. where the discretisation of measurements is imposed only because of the image resolution. An example of this method in its practical embodiment has been described in detail in research performed by Świłło and Czyżewski (2011).6mm. In addition. Figure 3 shows in detail a schematics of the stationary measurement as well as real experimental equipment.2. (2012) and Świłło (2012). a numerical image processing is used. Strain measurement using this method involves the application of a simple pattern-line onto the examined object in the area of the expected deformation. a single line pattern with no visual defect such as broken parts. The proposed method is based on an angled laser line examined element subjected to displacements. 5. involving a simple formula. b) grid pattern images for three methods of strain measurement. (2011). so the strain calculation cannot be calculated (figure 4b). Surface quality measurement Another parameter previously proposed by Swillo et al. For the specimen with the square pattern the grid shape is unrecognized. 4. The advantage of this method over the traditional techniques using different mesh geometry is the process of the measurement discretisation. an experimental study of grid pattern limitation is demonstrated on figure 4a. to identify the line. another advantage of ALM is its simplicity in use.INFORMATYKA W TECHNOLOGII MATERIAŁÓW allows a continuous strain determination in the examined sample. The common practise of optimizing assessing of the hemming quality is based on human inspection of the exposed hemmed surfaces. allow user simultaneously calculate geometry as well as deformation from a single CCD camera. Strain measurement for hemming: a) direct results comparison using three results: circle. the circle shape recognition by using image processing due to such deformations could be difficult to predict. A selected circle shows strong grid defects (cracking) that make strain measurement for this case very difficult. In addition. as an extension of this method a new optical configuration has been proposed by Świłło et al. allowing full automation of the strain determination technique. This inconsistency in the quality Fig. and the line should be applied at a certain angle. Finally. (2006) is used to judge a surface quality for hemming process evaluation. could be easily recognized and analyzed in case of hemming strain measurement (figure 4b).

Figure 6b shows the result of profile calculation for an arbitrarily chosen location within hemmed surface. The first reported application of the vision method for fracture analysis was by McNeill et al. which favours rather fusion of the micro-cracks already existing than the formation of new ones. using the portable vision system with data on the average length of micro-cracks and crack initiation conditions. a portable. Methodology for the measurement using hand-held vision system: a) schematics of the system. Many researchers used various non-optical measurement and image processing methods to study and describe fractures and cracking ( Epstein. For that reason. and ultimately has an impact on the whole process of elimination.. 6. The occurrence of 4. the commonly used method for the surface quality assessment is a visually adopted roughness reference level at which the product should be rejected. PORTABLE MEASUREMENT SYSTEM As a final result of the research investigation in the area of the hemming process analysis. To find a quantitative criterion of the surface deformation. In a perfectly run hemming process. This method of elimination leads to a lack of regularity and repeatability in the process of product elimination. 1993). i. b) profile and strain measurement. conferring considerable roughness to this surface.e. a manual capturing of the images takes place to provide information on the hemmed surface shape. in the developed technique of scanning along the edge of the inspected part. the cumulative length of micro-cracks is determined and referred to the number of these micro-cracks. surface inspection. the average length of micro-cracks relevant to the described maximum has been adopted. In the hemming process. c) surface inspection. The portable vision system is applicable in the analysis of the surface edge hemming under production conditions. local maximum in the graph proves that the maximum number of micro-cracks has been formed and their fusion has started taking place. Therefore. geometry reconstruction and strain measurement. it was necessary to propose an alternative route based on vision control. the laser line generator is located on the top of the portable system rotated relative to the camera by a specifically selected angle. First.INFORMATYKA W TECHNOLOGII MATERIAŁÓW control methodology results in undesirable quality variation in the hemmed parts. All the three above described techniques have been implemented. Second. The size of the roughness is a function of the size of the deformation and changes gradually from the surface smooth to very rough. As a crack formation index. (1987). Fig. The lack of objectivity is a fundamental error. (Livatyali et al. there is a large strain concentration at the edge of the hemmed surface. The proposed method of image analysis of the hemmed surface takes into account the deformation mechanism through statistical analysis of the microcrack formation (figure 5). hand-held vision-based measurement system has been developed (figure 6a). eventually giving rise to the formation of local cracks. Since then many variations of this approach have been developed and implemented to determine crack length or displacement field in the region around the crack. 2004). the surface remains smooth. In practice. This means that there has been the localisation of deformation. we are able to analyze and characterize the hemming quality for any given material and COMPUTER METHODS IN MATERIALS SCIENCE – 330 – .

INFORMATYKA W TECHNOLOGII MATERIAŁÓW

processing conditions. Figure 6c demonstrates the inspection measurement technique for micro-crack evaluation technique based on the use of co-axial illumination system. The third measurement takes place only in the situation when a simple pattern, i.e. an angled single line, is applied to the sheet surface in the region of the anticipated hemming deformation. To identity that pattern, the, improved by the author, ALM solution based on digital image processing techniques is used to provide more reliable solution, more accurate strain measurements and full automation. To summarize the whole, figure 6 demonstrates the portable vision measurement system capabilities: (a) the surface quality characterization by micro-cracks evaluation, (b) measurement of the surface geometry with profile, and (c) continuous strain measurement with improved ALM.

REFERENCES
Chen, K., Giblin, P., Irving, A., 1999, Mathematical exploration with MATLAB, Cambridge University Press. Epstein, J.S., 1993, Experimental techniques in fracture, New York, VCH Publishers. Graf, A. and Hosford, W., 1994, The influence of strain-path changes on forming limit diagrams of A1 6111 T4 , International Journal of Mechanical Sciences, 36/10, 897910. Higham, D.J., Higham, N.J., 2005, MATLAB guide, Society for Industrial and Applied Mathematics. Livatyali, H., Müderrisolu, A., Ahmetolu, M. A., Akgerman, N., Kinzel, G. L. and Altan, T., 2000, Improvement of hem quality by optimizing flanging and pre-hemming operations using computer aided die design, Journal of Materials Processing Technology, 98/1, 41-52. Livatyali, H., Laxhuber, T. and Altan, T., 2004, Experimental investigation of forming defects in flat surface–convex edge hemming, Journal of Materials Processing Technology, 146/1, 20-27. McNeill, S. R., Peters, W.H. and Sutton, M. A., 1987, Estimation of stress intensity factor by digital image correlation, Eng. Fract. Mech. 28/1, 101-112. Muderrisoglu, A. M., Murata, M., Ahmetoglu, M. A., Kinzel G. and Altan T., 1996., Bending flanging and hemming of aluminum sheet - an experimental study, Journal of Materials Processing Technology, 59/1-2, 10-17. Swillo, S. J., Hu, S. J., Iyer, K., Yao, J., Koç, M., Cai, W., 2005, Detection and characterization of surface cracking in sheet metal hemming using optical method, Transactions of the North American Manufacturing Research Institute of SME, 33, 49-55. Swillo, S., Iyer, K., and Hu, S. J., 2006, Angled Line Method for Measuring Continuously Distributed Strain in Sheet Bending, ASME Journal of Manufacturing Science and Engineering, 128, 651-658. Świłło, S., Kocańda, A., Czyżewski, P., Kowalczyk, P., 2011, Hemming Process Evaluation by Using Computer Aided Measurement System and Numerical Analysis; Proc. Conf. Technology of Plasticity 2011, eds. Gerhard Hirt, A. Erman Tekkaya, (Wiley-VCH Verlag GmbH & Co. KGaA) Weinheim, Aachen, 633-637. Świłło, S., Czyżewski, P., 2011, Analiza procesu zawijania z wykorzystaniem pomiarów wizyjnych i obliczeń numerycznych (MES), Zeszyty Naukowe, 238, Oficyna Wydawnicza Politechniki Warszawskiej, 93-98 (in Polish). Świłło, S., Czyżewski, P., Kowalczyk P., 2012, An experimental study of deformation load for hamming process, Przegląd Mechaniczny, 5, 34-37 (in Polish). Świłło, S., 2012, Hemming process strain measurement using Angled Line Method (ALM), 2012, Hutnik, 6, 443-446 (in Polish).

5. SUMMARY
In this paper the author presents several solutions that have been developed in the area of the hemming process experimental analysis. Since, the hemming deformation area is concentrated in a small corner area, advanced vision-based methods were applied with key parameters such as: strain, geometry and quality measurement. As a result of using the surface quality evaluation method, the hemming quality could be analyzed and characterized for any given material and processing conditions. Next, a successful result of using the strain measurement method for large deformation continuous (high increment resolution) strain distribution, maximum strain (strain peak localization and value), were computed. Finally, a geometry reconstruction was performed by scanning laser method. As a final results of the research investigation, a specially design portable vision–based measurement system has been developed to conduct all the experiments instead of a previously used stationary solutions. Currently, the hemming experimental investigation confirms that the surface strain distribution is a major factor including the solution to the problem of hemming diagram representation. By calculating more accurately the strain distribution using a new hand-held system and including the history of deformation, a new model of hemming limit diagram representation can be created. Acknowledgements. Scientific work financed as a research project from funds for science in the years 2009-2011 (Project no. N N508 390737).

APARATURA POMIAROWA DO ANALIZY PROCESU ZAWIJANIA Streszczenie Nowy, przenośny system doświadczalny został zaproponowany do analizy procesu zawijania. We wstępie przedstawiono informację o potrzebie wykorzystania systemów wizyjnych w analizie problemów występujących w procesie zawijania. Następnie, przedstawiono konstrukcję narzędzi do realizacji trzystopniowego procesy zawijania. Pośród licznych wizyjnych urządzeń pomiarowych stosowanych do pomiarów i rekonstruk-

– 331 –

COMPUTER METHODS IN MATERIALS SCIENCE

INFORMATYKA W TECHNOLOGII MATERIAŁÓW cji 3D elementów zawijanych, zaproponowana została przez autora metoda skanowania. W przedstawionym artykule, odniesiono się do problemów zaproponowanych technik pomiarowych i procesu obróbki obrazu w odniesieniu do przykładów doświadczalnych. Po pierwsze, zapisany obraz próbki jest analizowany pod kątem jej geometrii. W dalszej części, przedstawiono szczegóły odnośnie zaproponowanej nowej metody rekonstrukcji geometrii wraz z pomiarem wartości odkształcenia. Na koniec, przedstawiono, system przenośny kontroli jakości w ujęciu przemysłowym.
Received: October 28, 2012 Received in a revised form: December 4, 2012 Accepted: December 13, 2012

COMPUTER METHODS IN MATERIALS SCIENCE

– 332 –

COMPUTER METHODS IN MATERIALS SCIENCE
Informatyka w Technologii Materia ł ów
Publishing House AKAPIT

Vol. 13, 2013, No. 2

AN EXPERIMENTAL AND NUMERICAL STUDY OF MATERIAL DEFORMATION OF A BLANKING PROCESS
SŁAWOMIR ŚWIŁŁO*, PIOTR CZYŻEWSKI Faculty of Production Engineering, Warsaw University of Technology, ul. Narbutta 85, 02-524 Warszawa, Poland *Corresponding author: s.swillo@wip.pw.edu.pl
Abstract An experimental and numerical investigation is carried out in order to determine a material deformation of a blanking process. A highly localized, large strain distribution during the process at the last stage of a complete martial separation, has influence on the final surface quality product. Commonly using method in simulation of the blanking process is based on numerical approach. However, due to the large plastic element deformation is highly recommended to use remeshing procedure and other estimation solutions to simulate the last stage. To verify the final results and theoretical model other method are required. The paper present some implementation of experimental investigation in the field of displacement and strain measurement using a digital image correlation technique (DIC). The authors presents an experimental results of 1 mm thick specimen at the planar blanking process, where different clearances were used in the designed, fully automated apparatus. Finally, the experimental results were compared to the FEM simulation model with a good agreement. Key words: vision system, blanking process, correlation method, strain measurement, FEM

1. INTRODUCTION Currently, the technology of making numerous electronic components and equipment, such as engine rotors or transformer cores, is based on the use of a punching process. The above mentioned components are assembled by packeting a group of cut out elements. Hence, very important for the overall quality of the electrical assembly operation is the quality of single components in a packet. The limiting factor is too large burr on the cutting edge, which causes inaccurate adhesion of the sheet metal in a packet and serious deterioration of the quality of electrical assemblies. Finding a solution to this problem is one of the key issues in this technology of making components, and one of the methods currently applied is an experimental analysis of the cutting process.

Experimental analysis of the blanking process is a very complex issue and for a long time it lacked a solution due to the occurrence of large and irregular deformations near the die edge and punch. For a long period of time, the method used for the analysis of displacements was that of visioplasticity (Sutton et al. 1986), unfortunately giving less accurate results and requiring time-consuming calculations. Difficult to identify patterns of the grids, the blurred images of which were subjected to image processing, did not allow obtaining satisfactory results. Hence the need has emerged to search for new solutions in the field of numerical analysis. An outcome of this search was the development of a method based on Fourier transform, applied in the analysis of displacement increase between the individual stages of a cutting process (Leung et al. 2004). By

333 – 338

ISSN 1641-8581

the results obtained in this way were mainly based on tests carried out with the aid of specially designed instruments. However. adequate lighting and vi- precise punch location for each step of deformation (figure1b).INFORMATYKA W TECHNOLOGII MATERIAŁÓW comparing the image of a certain stage of deformation with the image following immediately this stage. plastic flow in the initial phase and crack formation next. with COMPUTER METHODS IN MATERIALS SCIENCE Fig. b) vision system configuration. taking into account the conditions adequate to vision measurements. strips of 1. getting results even when operating on millimetre samples. The authors’ proposal for the study of the cutting process relates to vision measurements taken under the real conditions of the punching operation. since the proposed method required visualisation of even the smallest displacements of the material. the submitted studies lacked any conclusions regarding the tool wear behaviour and analysis of the crack formation tendency. the authors were able to gain control of the image recording at a resolution relevant to the size of material displacement. the process is carrying out using a stepper motor. From the sheet with overall dimensions of 100x80 millimetres and 1. many authors (Stegeman et al. Thus gathered information is transformed to the computer memory and subjected to further numerical analysis. The goal was achieved by designing a special unit to carry out the punching process. certain conditions had to be satisfied to make such calculations possible. 2008. These conditions demand taking into account the external factors such as vibration. – 334 – . and Hambli 2001). which enabled the deformation size to be determined. For tests an albumin plate was used. Since that time. Hence. mainly due to the rapid development of various vision techniques and gaining access to solutions offering rapid cameras of high-resolution. All these factors make the analysis of the deformation process of the die-cut materials a great experimental challenge in the field of measurement techniques for both experimental and numerical methods (Makich et al. Initially. Specially designed illumination system allowing for the small measurement area and diversity of material structures enables taking a sequence of images captured by the vision system with specially chosen lens and a digital camera recording in the memory hundreds of photos per second. i. 2. Tests were carried out using a specially designed blanking apparatus.5 millimetres thick. 1998. With the punching process performed under static conditions.e. a distribution of displacement was obtained. with several elements such as: base. where the sheet metal is pressed between plate and die. EXPERIMENTAL SET-UP The schematic representation of a measurement stand is shown in the attached figure 1. However. taking into consideration two stages of the process shown in figure 1a. where all the systems such as: optical. Currently. both these issues being quite fundamental in control of the tool performance and monitoring of the process run.5x7x35 mm dimensions were cut out. the blanking process was performed using only a hand holder. Brokken et al. material texture was examined. instead of the typical markers applied to the surface in the form of mesh. This was due to the fact that. sion access to the area of material deformation. Schematic of the experimental set-up: a) blanking apparatus. 1999) have tried to solve this problem. A final configuration of the experimental set-up is demonstrated in figure 1c. possible to achieve only with a sufficiently large image resolution. illumination and vision systems are presented. 1. side walls and upper connecting element and bearing shell as well as sliding element with clamping of the upper cutting surface. but irrelevant to the real conditions under which processes of this type are performed.

since the DIC method is sensitive for the texture marks. several solutions are applied. For this reason. a method for the strain measurement has been proposed that demands the use of combined advanced solutions in the field of machine vision based on the correlation of two images. 3. any possible inaccuracies in calculations were compensated by optical zoom and limiting the analysis to a small area with the highest strain concentration (figure 3).INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. a high surface specimen quality is required. however. a texture patterns need to be related to small group areas. involving the use of a simple natural pattern resulting directly from the texture of the material. the maximum strain will be concentrated in an area not larger than tens of micrometers. An advantage of this method over the traditional techniques using different mesh geometry is the process of measurement discretisation. Specimen surface quality under different illumination and after surface machining. An additional advantage of the proposed solution is its simplicity. where material texture combine with their illumination are the major parameters as shown in figure 2. – 335 – COMPUTER METHODS IN MATERIALS SCIENCE . In the tests carried out. STRAIN MEASUREMENT The measurement of deformation in a punched sample is a serious problem due to high localisation of the non-linear strain distribution. Next the optical magnification is an important parameter. To perform an analysis of the recorded images of the surface of the cut out material. Methodology of the planar blanking process: a) blanking apparatus. require that the test area was adequately illuminated and the process of numerical calculations referred to small displacements of the material. The numerical process of comparing two images is performed for each pixel in the examined area. advanced numerical solutions of the image processing based on digital image correlation have been used. To accurately measure the material flow on the surface of an object. b) initial step. Due to this. which enables high-accuracy determination of the measured parameters. Figure 4 shows results of the surface discretization for the sample before and after process using virtual grid pattern. Therefore. To meet this requirement. high resolution and high accuracy. 2. However. b) final step (just before cracking). such as: dividing the examined area into small sub-groups or using high resolutions. 3. the process of the measurement discretisation is imposed solely because of the image resolution. although a low-resolution camera has been used (640x480 pixels). which in the case of the proposed solution is not restricted by the geome- Fig. Assuming that the die-cut sheet has a thickness of 1 mm. The measurement conditions. the strain measurement method should be characterised by try of the pattern used.

Figure 7 shows that the relation between the punch penetration and clearance is most likely linear. showing a good agreement to the FEM results. Δs(n) is a distance between the points xi( n ) and xi in the direction. 2. 5. That prediction could be successfully implemented in numerical calculation.3 (1) where:  2ui( n ) is the second order increment of the displacement vector. The punch penetration for each case were obtained every time up to fracture. The solution of the equations is available when at least two directions are investigated and the least square method is used. b) FEM.485mm of clearance (figure 6). DISCUSSION OF THE RESULTS Next. b) final stage. On the basis of directional derivative of the displacement increment vector we get: Finally. Fig. Let us choose some point xi(n) from the neighborhood of the point xi. The next surrounding in relation to the grid is conceived as neighboring points. 4. v j ( n) expresses COMPUTER METHODS IN MATERIALS SCIENCE vector directional cosines of the n direction. Results of the True Strain for the planar blanking process: a) experiment. 4. the total logarithmic strain was calculated as presented in figure 5. additional experimental investigation was conducted for a planar blanking process to determine an influence of the clearance for the material fracture. In the numerical simulation a large grid pattern was used intentionally (similar to the virtual grid – figure 4) in order to verify the image processing procedure that has been performed based on the correlation method. The displacement increments Δuij are unknown in this equation. Surface discretization using digital image correlation method (virtual grid pattern applied to the real surface: a) initial stage. In a mathematical formulation it means the use of a directional derivative of the gradient of an increment of displacement. The upper braced index denotes the chosen direction. the assessed quantity of which depends upon the position of the analyzed node. Ten sets of experiments were conducted in the range of 0. and by taking into consideration the impact of the near-by surrounding (Swillo 2001).035mm to 0. – 336 – . j  1.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. j   2ui( n ) s ( n ) for i. As for the kinematics calculation a method of analysis of grid has been applied. During these experiments the clearance between the material and the punch was measure and the blanking process was recorded in computer memory.  (j n ) ui .

. Y... G. An effective process of strain measurement for severe and local- DOŚWIADCZALNA I NUMERYCZNA ANALIZA PROCESU CIĘCIA W POMIARACH DEFORMACJI Streszczenie Doświadczalne i numeryczne badania zostały przeprowadzone.A. referring to the two dimensional displacement analMakich.R. Peters. 143-150. An exa valuable tool for verifying numerical methods. Techn. Leung. and are Stegeman Y.M.T. 2001.. Application of an optimized digital correlation method to planar deformation analysis. Set of experimental results for different clearance (0.M. International Journal of Mechanical Sciences.Y. Ch. Chan. Wong J. Proc. W. Automatic of strain measurement by using image processing. Swillo.. Mingqi. Duże wartości odkształcenia połączone z ich koncentracją na niewielkich obszarach prowadzące do procesu rozdzielenia – 337 – COMPUTER METHODS IN MATERIALS SCIENCE perimental and numerical study of a planar blanking process. 83. 1998.. 266-276. H. J. Brekelmans.. Brokken D. Sutton.R. 6.485 mm). Conf.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. 43. The DIC method was used to numerically control the defamation and the FE method was compare to experimental results. 272-277. L. 5. the deformation in a planar blanking process was monitored up to fracture. w celu określenia wielkości deformacji w procesie cięcia. 7. Chao Y. Engineering Design and Automation 2001. Hambli. Mater. Image and Vision Computing. Roizard...P.. Acknowledgements. Nevada. W. Parsaei. Proc. H. Form. 2008.P.. The results indicate blanking operation parameters (blanked material – wear that there are ample possibilities in the field of exof the punch). REFERENCES Brokken. monitored up to fracture.. Influence of clearance on the material punch penetration. J. Govaert. 192– 199. M.W. 3. Lee. M. Leep.. F.. 1986. International Journal of Machine Tools and Manufacture. H..T. W. yses (according to the function (1)) and strain measJ.. Scientific work financed as a research project from funds for science in the years 2011-2013 (Project no. . 1243-1246.M. 2769–2790. perimental analysis of the material flow. S..J.C.P. Gen. Chambert. L. 87. S. Carpentier. Comparison between Lemaitre and Gurson damage models in crack growth simulation during blanking process. 1. Las Vegas..correlation with urements using grid method. Monteil. Numerical modeling of the metal blanking process. X.. 7-8. McNeill. The proposed automatic vision system enabling the realization of measurements and calculations in a quick and precise manner for the blanking process even for the small Fig.R. N N508 628140).. Goijaerts A. 669-676. T. (less than 1 mm) thick. Baaijens F.A... eds.A. R. C.. 2001. 1999.. Brekelmans. Baaijens. 2004.035-0. D.C. Tang...E. Journal of Materials Processing Technology. The experimental examples presented in this paper ized plastic deformation.C. Mat.. SUMMARY In the currently ongoing project.H... L... Metrology of the burr amount .. Int.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW materiału mają duży wpływa na końcową jakość wyrobu. Zaprezentowano wyniki obliczeń doświadczalnych dla procesu cięcia dla różnych wielkości luzów. Wyniki pomiarów doświadczalnych zostały zestawione z symulacja komputerową MES. 2012 Accepted: November 5. W artykule przedstawiono możliwość zastosowania numerycznej obróbki obrazu z zastosowaniem korelacji w pomiarach pól przemieszczeń i odkształceń. Received: October 17. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 338 – . Tradycyjne metody analizy tych zjawisk opierają się wykorzystaniu metod numerycznych. 2012 Received in a revised form: October 22. Z uwagi jednak na dużą koncentrację odkształceń zalecane jest weryfikowanie tych wyników innymi metodami doświadczalnymi.

13. Poland *Corresponding author: piotr@lacki. 2 MODELLING OF STAMPING PROCESS OF TITANIUM TAILOR-WELDED BLANKS PIOTR LACKI*. 2000). No. titanium sheet 1. A three-dimensional model of the forming process and numerical simulation are performed using the ADINA System v. whose direction and mag- 339 – 344 ISSN 1641-8581 .8 mm are examined. WOJCIECH WIĘCKOWSKI. sheet-metal forming. The results obtained in the numerical simulations show some difficulties occurring in forming of welded blanks and provide important information about the process course. and consequently decline in the demand for tools. Sinke et al.8. Key words: TWB blanks. 2008. They might be useful in design and optimization of the forming process. Meinders et al. 2007. INTRODUCTION Tailor-Welded Blanks (TWB) become more popular in industrial applications in these sectors where reduction of weight and manufacturing costs are important. Grade 2 and Grade 5 (Ti6Al4V) titanium sheets with thickness of 0. It is estimated that application of TWB blanks can reduce the number of required parts to 66% and reduce the weight by half (Qiu & Chen. FEM modelling.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. 2010. and decrease in number of parts needed to produce component. Babu et al.. Schubert et al.6. Forming the spherical caps from the uniform and welded blanks are analysed. 2013. ul. Reduction of production costs for elements made of TWB blanks results from limitation to material usage and number of required forming operations. based on the finite element method (FEM). Application of welded blanks for products manufactured with use of stamping process requires solving many problems.pl Abstract In the paper some numerical simulation results of sheet-titanium forming of tailor-welded blanks (TWB) are presented. especially in case of forming hard-to-deform sheets. JANINA ADAMUS.. JULITA WINOWIECKA Częstochowa University of Technology. They are of particular interest in automotive and aircraft industry where there is the growing demand for shell parts (drawpieces) meeting specific functional properties which include low fuel consumption and sufficient strength of elements responsible for usage safety (Hyrcza-Michalska & Grosman. The preliminary experimental studies confirm correctness of the assumptions in the presented numerical forming process.. Drawability and possibilities of plastic deformation are assessed based on the comparative analysis of the determined plastic strain distributions in the drawpiece material and thickness changes of the cup wall. 42-201 Częstochowa. Dąbrowskiego 69. The presence of the weld (its geometric parameters) of different (generally lower) plasticity compared to the base material and heterogeneity of stamped blank lead to change in material deformation scheme in comparison with the deformations that occur in a homogeneous material. 2010. such as alpha and beta titanium alloys.com. This is due to weld dislocation. It also allows for reduction of discards from cutting and blanking. An analysis of the mechanical properties and geometrical parameters of the weld and its adjacent zones are based on the experimental studies.. Application of TWB blanks allows for achieving in one operation drawpieces characterized by mixed strength and functional properties. 2001).

Experimental studies are designed to confirm the validity of the assumptions made in the numerical model of the process (figure 1a). Microstructure of the border zone between the joint penetration and HAZ in Grade 2 is more evolute than the border zone be- COMPUTER METHODS IN MATERIALS SCIENCE Fig 1. in order to evaluate its drawability and formability in traditional stamping processes. Lisok & Piela. b) microstructure of electron beam welded joint. it is necessary to carry out several studies.. heat affected zone (HAZ) in Grade 5. 2006. 2007. while width of HAZ in Grade 5 together with zone of joint penetration is of ~553 m. Kinsey et al. Więckowski et al. 2. Titanium Grade 5 has a globular. Microstructure of HAZ in Grade 2 shows big recrystallized grains of  phase. 2010. uniform sheets Grade 2 and Grade 5 were performed. a) Drawpiece obtained during experimental research. Rectilinear shape of grain boundaries is typical for recrystallized grains. 2003. 2011.Grade 2. titanium Grade 2 sheets have good drawability however produced drawpieces are characterized by low strength.INFORMATYKA W TECHNOLOGII MATERIAŁÓW nitude depend on differences in mechanical properties and thickness of welded materials (HyrczaMichalska & Grosman. 2000). In order to evaluate suitability of welded blanks for the forming processes. Generally. 2004. 2010. zone of joint penetration. Additionally calculations for the tween HAZ in Grade 5 and the joint penetration zone. Rojek. Babu et al. On the other hand titanium Grade 5 sheets have higher strength than titanium Grade 2 sheets and they have low propensity to plastic deformation and this limits their application in forming processes (Adamus. 2009 a. – 340 – . 2010. 2000).. Zimniak & Piela. 2007.. Hyrcza-Michalska et al. 2000. HAZ in Grade 2 is wider than HAZ in Grade 5.. Microstructure of electron beam welded joint is shown in figure 1b. In this study the numerical simulation of drawing spherical cap from welded sheets made of titanium Grade 2 and Grade 5 of the same thickness was performed. for structural elements with specific functional properties leads to a growth of interest in sheet-titanium forming. among others in aircraft industry. Higher magnification shows a change of globular microstructure into lamellar one on the transition of HAZ in Grade 5 into the joint penetration zone. 2010. 2008.  phase grains with lenticular grains represent the microstructure of base material Grade 2. including numerical simulations of the process. EBW causes some changes in material microstructure. fine-grained structure. Grains of  phase with separation of  phase on the grain boundaries are visible. Meinders et al... Grade 2 and Grade 5 materials were joined using electron beam welding (EBW) technology. that will allow for prediction of sheet behaviour in consecutive stages of the forming process (Ananda at al. Analysis of the joint microstructure shows occurrence of 5 zones – from the left: base material – Grade 5. 2009 b). Babu et al. The zone of microstructure changes has a width of less than 3 mm. The increase in demand. using numerical simulation and experimental verification of changes in the wall thickness distribution in the drawpiece. heat affected zone in Grade 2 and base material . GOAL AND SCOPE OF THE WORK A goal of the paper is evaluation of changes in deformation and displacement scheme of TWB blank material in consecutive stages of the forming process. Qiu & Chen.. Its width is of ~2282m.

Some important geometric parameters of the model are presented in table 1. FEM geometry model is shown in figure 2. M2) .figure 2. for performing calculations. 2. which are required ness. two heat affected zones (HAZ) located symmetrically on both sides of the weld (HAZ1. each other was realised by immobilising die and applying the displacement to the punch in the direcTable 1. tion of X axis.0 mm Discretization of the blank material (TWB blank-holder force Fd 3000 N blanks) for the stamping process. linear description of material hardening and the conMutual displacement of tool elements against tact between the tool and the forming blanks.1 was set for contact surfaces between the die. and the friction coefficient . Parameters assumed in FEM model for the stamping process of TWB blanks.8 mm value of the blank-holder force was determined weld width W 1. blank and Fig.8 mm.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 3. was performed using four-node shell elements of specified thickThe mechanical properties. HAZ2) and two zones representing base materials (M1. In the analysed case the weld was located in the drawpiece centre. An optimal die fillet radius rm 4 mm blank thickness g 0. which equals to thickness of the welded blanks – 0. A contact interaction between the tool and the blank material plays an important role in the forming process (Adamus. count different material properties in the weld vicinity.6. In the case of the blank-holder its progressive motion was limited by a hold-down Parameter value force Fd. of base material. In the numerical calculations a friction coefficient  . NUMERICAL MODEL A three-dimensional model of the stamping process was developed. was set In the calculations all elements corresponding to for the contact surface “punch – deformed material the tool were assumed to be perfectly rigid and for (blank)” without lubrication.9 mm based on the preliminary numerical simulation of the heat-affected zone width HAZ1 1. a punch and a blank-holder. The model comprises material of the welded blank and the stamping tool consisting of a die. where the working with the specified 5-zone model of the welded blank. and it also has a significant impact on the punch radius rs 16 mm distribution of drawpiece wall thickness. surfaces were lubricated. 2010. heat-affected zone width HAZ2 1. heat Modeling of the welded blank material required affected zone and weld zone were determined based distinction of appropriate zones and taking into acon the uniaxial tensile test as well as on the basis of – 341 – COMPUTER METHODS IN MATERIALS SCIENCE . which allows for non(bilinear plastic material model). A discrete model of the forming process of a spherical cup made of TWB blank blank-holder. in the form of punch path hs 20 mm a disc with diameter dk = 60 mm.7 mm stamping process. In the calculations a constant thickness of the weld and heataffected zones. elements corresponding to the deformed sheet an Calculations were performed using the ADINA isotropic elastic-plastic material model was applied System v. Measurements of the zones were performed during observation of the weld cross-section structure. based on FEM. 8. In the presented model 5 zones were distinguished: weld zone (W). 2009 a). A proper selection of the blank-holder blank diameter d 60 mm k force prevents wrinkling of the flange material (figclearance between punch and die l = dm-ds 2 mm ure 3). was assumed.

Tensile strength Rm [MPa] 316. The assumed mechanical properties are summarized in table 2. b) blank-holder force 3000N.37 0. b) material thinning [mm]. and HAZ material.5 Yield strength R0. Experimentally determined material properties of Grade 2 and Grade 5 titanium. and is accompanied by uniform thinning of the drawpiece wall (figure 4b).4 442.37 0. Fig.INFORMATYKA W TECHNOLOGII MATERIAŁÓW changes in hardness distribution within the weld cross-section.3 747.37 4. In this area it is seen a considerable material thinning (figure 5a and 5b). Test specimens were prepared using TIG welding method.8 964. The numerical calculation results of plastic strain  [-] and thinning of the drawpiece wall as a result of stamping process are shown in figures 46. In the case of stamping process of uniform Grade 2 blank it can be observed that the plastic strain distribution in the blank material is uniform and circular (figure 4a).0 Young’s modulus E [GPa] 110 110 110 110 110 Poisson’s ratio  0. The numerical simulation of TWB blank forming shows that the weld moves in the direction of Grade 5 material as punch hollows into the deformed blank (figure 6). In the case of the forming process of uniform Grade 5 blank it is seen that concentration of plastic strains is in a pole of the cup. The mechanical properties of the material in the weld zone were estimated based on the relationship between the hardness and strength of the material assuming that the material yield stress is in direct proportion to its hardness. The drawpiece shape obtained in numerical simulation of the stamping process of the welded blank a) blank-holder force 1000N.5 518.37 0. 4.3 368.37 0.2 [MPa] 236. Numerical simulation results of stamping process of spherical cup made of Grade 2 blank at the punch penetration of 10 mm: a) plastic strain distribution  [-].6 1002. RESULTS Figure 2 shows the shape of the drawpiece obtained during the numerical simulation of stamping process of TWB blank.8 798. Material/zone M1-GRADE 2 M2-GRADE 5 HAZ1 HAZ2 W COMPUTER METHODS IN MATERIALS SCIENCE Fig. weld materials. – 342 – .7 375. Table 2. 3.

). As a result of weld displacement. This might indicate that there is a possibility of drawpiece weakening and possible loss of material continuity in this area. 6.01. SUMMARY The main goal of the study was to develop a numerical model of stamping process of titanium TWB blanks.Innovative Economy (IE OP) financed from the European Regional Development Fund .02-00-015/08-00 is gratefully acknowledged. is much more difficult than stamping of the Acknowledgements. Numerical simulation results of stamping process of spherical cap made of welded blank Grade2||Grade5 for punch penetration of 10 mm: a) distribution of plastic strain  [-]. lubrication conditions etc. using rigid tools. Financial support of Structural Funds in the Operational Programme . 5. Comparison of strain distribution in the drawpiece made of a homogeneous material with those found in drawpiece made of TWB blanks shows that the presence of weld with different strength properties. – 343 – COMPUTER METHODS IN MATERIALS SCIENCE uniform blanks. The results provide important information about the process and may be useful for the design and optimization of the process run (selection of appropriate process parameters such as: blank-holder force.Project "Modern material technologies in aerospace industry". The simulation results show the efficiency of applying numerical calculations to studying stamping processes of TWB blanks. 5. Nr POIG.01. which will allow for further improvement. It should also be noted that in the area near the drawpiece top (on the border between more deformable material and the heat affected zone) there is a local increase in strains and significant thinning of the drawpiece material (figure 6a and 6b). . plastic strains increase in more deformable material and decrease in less deformable material (figure 6a). The calculations confirmed experimental results that stamping of titanium welded blanks that are characterized by different strength properties. In future these studies will be focused on a more accurate description of material mechanical characteristics in the heat affected zones and weld.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Fig. b) material thinning [mm]. It can be observed that there is limited formability in the zone corresponding to weld and that there is movement of this zone in the direction of less deformable material. Fig. Numerical simulation results of stamping process of spherical cup made of Grade5 blank as the punch penetrates the depth of 10 mm: a) distribution of plastic strains  [-]. introduces irregularity in the strain scheme in the deformed blank. b) material thinning [mm]. The performed simulations (FEM) of the process allow for analysis of deformation introduced into material during forming process and drawability assessment of the welded blanks.

.. 54/3... Meinders T. Piela A. 2008.L.. 56/11 (in Polish). Grosman F. 2010. Oceny właściwości mechanicznych i parametrów geometrycznych spoiny oraz obszarów jej przyległych dokonano na podstawie badań doświadczalnych.. 2000.6. Key Engineering Materials. Archives of Civil and Mechanical Engineering. bazującego na metodzie elementów skończonych (MES). Adamus J. Modelling and simulation of complex problems of nonlinear mechanics using the finite and discrete element methods. 99.. Archives of Civil and Mechanical Engineering. Piela A. Formability of laser welded blanks. Lisok J. 106. 17th International Scientific And Technical Conference “Design and technology of drawpieces and die stampings”... 2010.. 2003. Sinke J. Badano blachy tytanowe Grade2 i Grade 5 o grubości 0. Received: October 16. 2009 a.. Andreychuk P. Kumar Saravana G.L. 2012 Accepted: November 9. 6 (in Polish)... Narayanan Ganesh R..... Iacono C.. J. 8.. Int. Journal of Materials Processing Technology. Journal of Materials Processing Technology.. 2012 Received in a revised form: November 22. Zadpoor A.. Piela A. 2010.INFORMATYKA W TECHNOLOGII MATERIAŁÓW REFERENCES Adamus J. Rojek J. G. Journal of Materials Processing Technology. Deep drawing simulations of Tailored Blanks and experimental verification.A.. The study on numerical simulation of the laser tailor welded blanks stamping... Przeprowadzono analizę kształtowania czaszy kulistej z wsadu spawanego oraz materiałów jednorodnych. Zimniak Z. Hyrcza-Michalska M. 199–207. 2011. 103. Liu Z. 2010. 2007. Chena D. Tailor made blanks for the aerospace industry.. Model of welded joint in the metal charges used for testing pressformability. Ananda D.D. 2007. Rudy Metale. Monografie nr 174.. Journal of Materials Processing Technology. Prowadzone równolegle wstępne badania doświadczalne potwierdziły słuszność przyjętych założeń w zaprezentowanym modelu numerycznym procesu tłoczenia. Rojek J.. Politechniki Częstochowskiej (in Polish). Klassen M. Lisok J. 2006.G. Lacki P. 2000. Conf. Proc. Schubert E. A 420. Expert Systems with Applications. Prace IPPT IFTR REPORTS 4 (in Polish).. Huetink J. Cao J.. 2009 b. 4/3. 2004. A novel forming technology for tailor-welded blanks.net Adamus J.. Poznań:INOP (in Polish). Hyrcza-Michalska M. Kinsey B. Form. van den Berg A. 115. The analysis of forming titanium products by cold metalforming. Chen W.. Numerical simulation of the sheet-metal forming process of tailor-welded blanks (TWBs). Boudreau. Zerner C.scientific. Theoretical and experimental analysis of the sheet-titanium forming process. http://www. Model złącza spawanego we wsadach do tłoczenia blach „tailored blanks”. Archives of Metallurgy and Materials. Journal of Materials Processing Technology.8 mm.. 37. An expert system for predicting the deep drawing behavior of tailor welded blanks. Materials Science and Engineering. Fruitos O. Przestrzenny model procesu tłoczenia oraz obliczenia numeryczne wykonano przy użyciu programu ADINA v. Numerical simulation of car body elements pressing applying tailor welded blanks – practical verification of results. Stamping of the Titanium Sheets. 187–188. Fatigue behavior of tailor (laser)-welded blanks for automotive applications. 10/4. Walz C. Babu Veera K.. Light-weight structures produced by laser beam joining for future applications in automobile and aerospace industry. MODELOWANIE PROCESU TŁOCZENIA SPAWANYCH BLACH TYTANOWYCH TYPU TWB Streszczenie W artykule przedstawiono wyniki symulacji numerycznych procesu tłoczenia spawanych blach tytanowych typu TWB (Tailor-Welded Blanks).. Qiu X. Bhole S. 3/1.. 2001. Wyd.. Przegląd spawalnictwa. przez co mogą być przydatne na etapie projektowania i optymalizacji procesów tłoczenia. Uzyskane na drodze symulacji wyniki wskazują na trudności występujące podczas kształtowania blach spawanych oraz dostarczają istotnych informacji o przebiegu procesu.. Więckowski W. Adamus J.. Finite element analysis of a tailored blanks stamping process. Seplod G... Dokonano oceny tłoczności oraz możliwości kształtowania plastycznego badanych materiałów poprzez analizę porównawczą wyznaczonych rozkładów odkształceń plastycznych w materiale wytłoczek oraz zmiany grubości ścianek wytłoczek. Mater.. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 344 – . 2004.

calculated consolute temperatures are: 270 K for B4.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol.und Geowissenschaften. Germany *Corresponding author: andrzej. gamma ray detectors. or in the metastable B3-type structure either by molecularbeam epitaxy. 2009. groove rolling.44 eV for CdS.wozniakowski@us. Materials Science and Engineering Laboratory.. the CdSe (CdS) can be synthesized in the B4. Ab initio calculations of supercell formation energies were fit to cluster expansion Hamiltonians.. Telegrafenberg. 2000).. Madelung et al. Calculated consolute temperatures for all structures are in good quantitative agreement with experimental data. Depending on the growth conditions. 2013.. Maryland 20899-8520. and Monte Carlo simulations were used to calculate finite temperature phase relations. Tolbert & Alivisatos.. Gaithersburg. No. 13. 2003. 2000). USA 4 Present address: Technische Universität Darmstadt. OMAR ADJAOUD2. 40-007 Katowice. 1982).. Because of excellent properties Cd(S.. Fachbereich Material. Section 3. Hotje et al. Bankowa 12. photoconductors. Under high pressure. BURTON3 Institute of Materials Science. INTRODUCTION The cadmium chalcogenide CdSe1-xSx semiconducting alloy is characterized by a variable direct band gap which can be tuned by alloying. BENJAMIN P. 2000. 2 FIRST PRINCIPLES PHASE DIAGRAM CALCULATIONS FOR THE CdSe-CdS WURTZITE. The equilibrium zincblende structure is observed in CdS nanostructures (Banerjee et al. JÓZEF DENISZCZYK1. from 1. 300 K for B3. Excess vibrational contributions to the free energy were included. Fachgebiet Materialmodellierung. lasers and solar cells (Xu et al. 2006).pl Abstract The phase diagrams of CdSe1-xSx alloys were calculated for three different crystal structure types: wurtzite (B4). (Mujica et al. D-64287 Darmstadt. Petersenstr. and references cited therein). or by controlling the growth temperature (Wei & Zhang. It is known that CdS and CdSe occur at normal conditions both in the wurtzite and metastable zincblende structures.. Poland 2 GFZ German Research Centre for Geosciences. 2003.72 eV for CdSe to 2. and rocksalt (B1). 32. National Institute of Standards and Technology. The calculated phase diagrams have symmetric miscibility gaps for B3 and B4 structure types and a slightly asymmetric diagram for B1 structure. University of Silesia. CdSe1-xSx solid solutions have attracted great interest in recent years from both experimental and theoretical points of view (Xu et al. and with these..3.4. 1995. FEM 1 1. Banerjee et al. ZINCBLENDE AND ROCK SALT STRUCTURES ANDRZEJ WOŹNIAKOWSKI1. 14473 Potsdam. 2003. and 270 K for B1. Key words: clamping.edu. Mujica et al. 2009. zinc-blende (B3). Germany 3 Ceramics Division.Se) is used in optoelectronic devices. both B3 and B4 structures convert to the denser rocksalt- 345 – 350 ISSN 1641-8581 . Wei & Zhang. visible-light emitting diodes. 2000. Deligoz et al.

indicated that within experimental error Hf = 0 at 298 K. respectively. Breidi. Kresse & Furthmüller. The free ener- – 346 – . In Ref. or 2) the consolute temperature for CdSxSe1-x in B4 structure must be below room temperature. 1996a. In our investigations the well-converged cluster expansion system required calculation of the formation energy for 30-50 ordered superstructures. RCdS = 1. (2010) and Breidi (2011) originates from the different ab initio computational setup. Vmol.. B3 and B4 crystal structures. indicating that either: 1) the blocking temperature for Se/S diffusion is above TC. were performed using the Vienna ab initio Simulation Package VASP (Kresse & Hafner.727 cc/mol. Tolbert & Alivisatos.. Hotje et al. Lukas et al. despite differences in molar volume (Vmol. Electronic degrees of freedom were optimized with a conjugate gradient algorithm. embody the information regarding the energetics of an alloy. COMPUTER METHODS IN MATERIALS SCIENCE where α is a cluster defined as a set of lattice sites. Davies. B3 and B4. Recent measurements of formation enthalpies (Hf) for CdSxSe1-x B4-type solid solutions.CdS = 29. that one can have reasonable confidence that calculated phase diagrams faithfully reflect density functional theory (DFT) energetics. 0. 0.75 and 1. In both cases (B3 and B4). and different choices of supercells for which formation energies were calculated (this difference indicates that the sets of formation energies for at least one of these calculations. while excess vibrational free energy contributions were neglected. For the B4-type structure Breidi (2011) reports TC = 225 K < TC = 228 K for the B3 structure.INFORMATYKA W TECHNOLOGII MATERIAŁÓW structure phase (Mujica et al.. 2. 1984). 1994). Both cell parameters and ionic positions were fully relaxed for each superstructure of underlying B1. with the generalized gradient approximation for exchange and correlation potentials. Sufficiently large sets of formation energies are used. 1995. defined as E f  E CdS x Se1 x  xE CdS  1  x E CdSe . Jα. The difference between TC as calculated by Ouendadji et al. 2007). van de Walle & Asta. 2003. 0. (2009). summation is over all clusters α that are not equivalent by a symmetry operation and an average is taken over all clusters α′ that are equivalent to α by symmetry. 2010.. reported by Xu et al. COMPUTATIONAL DETAILS Calculations of formation energies. This may indicate that. Both studies predict miscibility gaps... 1996b) implementing the Blöchl’s projector augmented wave approach (Blöchl.0) were considered. 2006) (Xu et al. (2010) and Breidi (2011) are: TC = 315 K and 228 K. that CdS and CdSe form an ideal solution in the B4.934 cc/mol. and a plane-wave energy cutoff of 350 eV was used which yields  values that are converged to within a few meV per atom. Se = 4s24p4 and S = 3s23p4. 2011) phase diagrams for both B3 and B4 structures were determined. These measurements did not show the presence of a miscibility gap above 298 K. The predictive power of cluster expansion is controlled by cross2 1 n ˆ )2  validation score CVS    ( E i  E  . The VASP calculations were used to construct cluster expansion (CE) Hamiltonian in a form of polynomial in the occupation variables: E     m J    i  i ' (Sanchez et al. Jug & Tikhomirov..84 Å. (2009). 2010) only B3 structure was investigated. 2002a. Based on the VASP results. The Effective Cluster Interaction (ECI) coefficients. are based on sets of formation energies that are too small to yield converged effective Hamiltonians). Both predicted consolute temperatures differ from the critical temperature (TC = 298 K) reported by Xu et al.. mα denote the number of clusters that are equivalent by symmetry.e. 2011. 1994.25. while in (Breidi.structuretype.91 Å. 1993. The T-x phase diagram of the CdSe-CdS system was the subject of theoretical ab initio studies (Ouendadji et al. where (i )  n i 1  1 Ei is an ab inito calculated formation energy of su- ˆ represent the energy of perstructure i. and probably both. the First Principles Phase Diagram calculations were performed with the use of Alloy Theoretic Automated Toolkit (ATAT) software package (van de Walle & Ceder. For the B3-type structure the consolute temperatures (TC) reported by Ouendadji et al. (Ouendadji et al. Both configurational and excess vibrational contributions to the free energy are considered. van de Walle et al. only formation energies (at x = 0. i.5. All calculations were converged with respect to gamma centered k-point sampling. 2002). 2009). The aim of this study is to compare well converged calculations of CdSe-CdS phase diagrams in all three crystal structure types: B1. Valence electron configurations for the pseudopotentials are: Cd = 4d105s2.. 2003). at least above room temperature. 2002. 1981) and anion radii (RCdSe = 1. while E (i ) superstructure i obtained from CE with the use of the remaining (n – 1) structural energies.CdSe = 33.

 ) (3) The thermodynamic integration in (3). 2002): temperature at chemical potential stabilizing a given ground state of a system (here. RESULTS AND DISCUSSION Using the ab initio (VASP) method.5 meV/atom was reached for approximately 25 superstructures.  ) can be calculated through the thermodynamic integration described by equation (3) (van de Walle & Asta. B3-.  0 )  ( 1 . 2002).  )   1   ln  exp( N ( Ei  x i )  . are allowed to fluctuate. the internal energy (Ei) and concentration (xi) of constituents of an alloy with fixed number of atoms (N).1 meV/atom. A schematic diagram of the approach is: VASP calculations of formation enthalpies and vibrational free energies for a set of superstructures  fit a cluster expansion (CE = set of effective cluster interactions. calculations of the ground state energy were performed for the stoichiometric compounds CdSe and CdS and for the formation energies of many B1-. 0 )  ( E   x. 1. ECI) fit effective force constants to model excess vibrational contributions  Monte Carlo thermodynamic integration  phase diagram. The phase diagram calculations were performed with the use of the Monte Carlo thermodynamic integration within the semi-grand-ensemble (van de Walle & Asta. 0) and (1.INFORMATYKA W TECHNOLOGII MATERIAŁÓW gy contributed by lattice vibrations was introduced employing the coarse-graining formalism (van de Walle & Ceder. Increasing further the number of superstructures results in the fluctuations of CVS with standard deviation of order of 0. The thermodynamic potential  (per atom) associated with the semi-grand-canonical ensemble can be defined in terms of the partition function of the system in the form presented in equation (1) (van de Walle & Asta. 2002b). 2002b). For each superstructure the vibrational free energy was calculated within the quasi-harmonic approximation with the application of the bond-length-dependent transferable force constant approximation (van de Walle & Ceder. Figure 1 shows the dependence of the CVS on the number of calculated superstructures. was performed using the Monte Carlo method. Dependence of the CVS vs number of superstructures used in the fit of CE for B1. The starting point (0. In this ensemble. 36 – B3 and 34 – B4). (1) N  i  where the summation is over different atomic configurations (alloy states) and   1 /(k BT ) (kB is Boltzmann’s constant). 1) which does not encounter the phase transition. 1 )   0 ( 0 . – 347 – COMPUTER METHODS IN MATERIALS SCIENCE where E and x are the alloy’s average internal energy (calculated with the use of CE expansion) and concentration of constituents. 3. 1 ) (  0 . 2002): 1 (1 . for temperature (T) and chemical potentials () imposed externally. B3 and B4 structures. In the differential form (with variable T and ) equation (1) can be rewritten in a form given by equation (2): d (  )  ( E    x ) d   x d  . 0) is taken in the limit of low Fig. The convergence of CVS at values less than 1. The advantage of this approach is that it is based on the parameters-free ab initio calculations and leads to high quality effective Hamiltonians for multicomponent systems. and averaging was performed according to formula: A   Ai exp(  N ( Ei   xi )) /  exp(  N ( Ei   xi )) .   x ) d (  . The results presented in figure 1 strongly suggests convergence of the CE series. or B4-based superstructures (36 – B1. (2) i i Using the differential form given by equation (2) the thermodynamic potential  (  . The CE has the limitation that it only applies to a parent structure and its superstructures. The optimal number of superstructures were determined by minimizing the cross-validation score between ab initio computations and the cluster expansion prediction.  ( . along a continuous path connecting points (0. All formation energies were positive. the chemical potentials of end-members CdS and CdSe). which implies that no intermediate ground state structures were predicted.

Formation energies Ef calculated by VASP (cross) and fitted by cluster expansion (CE) for B4 (figure a). B3 and B4 crystal structures. thus the results in figures 3. This correlates with the higher predicted consolute temperature when only the configurational part of free energy is taken into account. detailed model studies of an effect of lattice COMPUTER METHODS IN MATERIALS SCIENCE – 348 – . Figures 4 are the calculated phase diagrams for the CdSe-CdS system in B1. The differences between VASP-calculated and CE-calculated are small. Temperature independent ECI were used to calculate the lower solvii in (a) and (b) and the upper curve in (c) (dash lines). except for the end-members compounds CdSe and CdS in B4 structure-type. The inter-atomic distance is expressed in units of the nearest neighbor distance (dnn). Note that formation energies for B1-based supercells are about twice as large as for the wurtzite and zinc-blende structures. note that these differences are an order of magnitude smaller than the values Ef. 2.. Fig. In most miscibility gap systems (Adjaoud. Temperature dependent ECI. et al. Low values of cross validation score and decreasing magnitudes of the ECI justify truncation of CE series and discarding the larger clusters. Fig. B3 (figure b) and B1 (figure c) structures. thus the results in figures 3a and 3b. It is evident that with increasing distance. Note the different scale used on vertical axis of figure c. The 3-body ECI. is very small. 2006) the inclusion of temperature-dependent ECI leads to a reduction in TC. However. 2009. Burton & van de Walle. are plotted as functions of inter-atomic separations in figures 2. yield the results which are plotted as the upper solvii in (a) and (b) and the lower solvus in (c) (solid lines). confirm the quality and predictive power of CE. B3 (figure b) and B1 (figure c) underlying crystal structures. Figures 3 are plots of the VASP-calculated supercell formation energies (Ef per cation) that were used to fit the ECI in figures 2. for B1 structure-type. Effective cluster interactions (ECI) as functions of interatomic distance (d/dnn) for the clusters taken into account in cluster expansion series for the B4 (figure a). 3. pair-ECI magnitudes decrease with oscillatory sign. which imply the inclusion of excess vibrational contributions to the free energy.INFORMATYKA W TECHNOLOGII MATERIAŁÓW The ECI. are atypical.

50. TC) = (0.50. 12. First-principles phase diagram calculations for the HfC-TiC. and temperature-dependent ECI.. that T-independent ECI predict lowest TC rather than higher. respectively.. Banerjee. 17953-17979.and rock salt structure-types.and zinc-blende structure-types. Rev. P. Burton. 4. B.and with excess vibrational free energy contributions (i.. Fig. 80. to TC = 270 K. without. zinc-blende. Geo.e. respectively. with vibrational effects included. van de Walle. Phys. Dash and solid solvii are the phase diagrams that were calculated with temperatureindependent-. Furthermore. 10647-10654. When only configurational free energy is taken into account the calculated consolute temperatures are: 220 K. that inclusion of vibrations into phase-diagram modeling of miscibility gap systems can increase the consolute temperature. 2000) have shown. Phys. 134112-134119. respectively. 270 K) when temperature dependent vibrational contribution is included. zinc-blende. First principles phase diagram calculations for the system NaCl-KCl: The role of excess vibrational entropy.INFORMATYKA W TECHNOLOGII MATERIAŁÓW cantly as compared to that of B4 structure. B. G. For the rock-salt structure inclusion of vibrations reduces the consolute temperature by 25%..P. Ayyub. R. Surprisingly. Miscibility gaps are predicted for all three systems. (2009). J.and B4-based solid solutions. Condens. 2006). 225. Rev. ZrC-TiC. and increases the consolute temperature by 30% and 23% for wurtzite. Chem. Projector augmented-wave method.. 1996). 230 K and 360 K for wurtzite-. Jayakrishnan. Effect of the size-induced structural transformation on the band gap in CdS nanoparticles.. For the B4 structure (figure 3a) the miscibility gap is symmetric with critical point (xC.. A.. 2006. 4.61) and reduce the critical temperature: from TC = 360 K. (2009) for the TiC-ZrC system.. Calculated consolute temperatures for B1. the inclusion of excess vibrational contributions to the free energy destabilizes the B3. van de Walle. 2000.. A.. TC) = (0. Inclusion of the vibrational contribution enhances asymmetry (xC = 0. when only formation energy is taken into account. The results in (a) and (b). B. For the B1 structure (figure 4c) the phase diagram obtained on the basis of temperature independent ECI is almost symmetric (xC = 0. R. B. 222229.. Burton. both without.P. et al. Phys. 220 K) when only configurational degrees of freedom are taken into account. investigations of the vibrational entropy change upon order-disorder transition in Pd3V system (van de Walle & Ceder.51). E. Slightly above room temperature complete solid solution is possible in the zincblende structure. 2009. respectively. and HfC-ZrC solid solutions. that the relaxation of bonds can change the sign of vibrational entropy differences as compared to the expectations based on the bond proportion model. respectively). and (xC. et al. B3 and B4 structure-types compare well with experimental critical temperature TC = 298 K reported by Xu. The main feature of the calculated phase diagrams is the consolute temperature (TC). P.and rock salt structure-types were calculated by the CE-method. contrary to similar studies (Burton et al.. O. For the B3 structure (figure 3b) the shape of the phase diagram does not change signifi- – 349 – COMPUTER METHODS IN MATERIALS SCIENCE . but for the B3 structure we obtained higher consolute temperatures: 230 K and 300 K. Calculated phase diagram of CdSxSe1-x alloy in B4 (figure a). is atypical. 50. vibrations on the phase stabilities of substitutional alloys have shown (Garbulsky & Ceder.and with T-dependent ECI. CONCLUSIONS Ab initio calculations of CdSe-CdS phase diagrams for wurtzite-. REFERENCES Adjaoud. B3 (figure b) and B1 (figure c). 1994. Steinle-Neumann. similarly as reported by Adjaoud. Matter. Blöchl.

Furthmüller. ściennie centrowaną strukturę soli kamiennej (B1). Hafner. S. Mujica. światłoczułych. Ceder.. Rev. OBLICZENIA Z PIERWSZYCH ZASAD DIAGRAMÓW FAZOWYCH DLA PSEUDOBINARNEGO SYSTEMU CdSe-CdS KRYSTALIZUJĄCEGO W SIECIACH WURCYTU. C. Diagramy fazowe zostały wyznaczone na podstawie potencjałów termodynamicznych obliczonych metodą całkowania termodynamicznego Monte Carlo. 1984. Weiss. J. Landolt-Bornstein Numerical Data and Functional Relationships in Science and Technology. Ma. B..P. Phys. Ceder. Kauzlarich. 100.. J. CdS. van de Walle. 2003. B.. Calphad.S) oraz ich stopy charakteryzują się szeroką bezpośrednią przerwą energetyczną i dlatego mogą być przydatne w urządzeniach optoelektronicznych. Mater. Ducastelle. Comput. Ceder. Sanchez.... Needs. Xu.. van de Walle.H. van de Walle. 54.. CdSe.. 2012 Accepted: December 8.53.. 26. Chem. Efficient iterative schemes for ab initio total-energy calculations using a planewave basis set. Cambridge Press. Kresse.A. and AlN-InN.. Temperature–pressure phase diagrams. G. Ab initio molecular dynamics for liquid metals. A. A. 113528-113534. Madelung. 6. 2011. 206-211. 2002. H. Alivisatos. R. The effect of lattice vibrations on substitutional alloy thermodynamics. The Calphad Method. Gratias. 1996a. 1259-1262. Mod. L. Received: October 29. Fries. 1088-1092. J. 1368-1374... Computational Thermodynamics. van de Walle. G. A.. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 350 – . Celem niniejszej pracy jest przeprowadzenie obliczeń diagramów fazowych oraz wyznaczenie krytycznej temperatury mieszalności dla stopów CdSe1-xSx krystalizujących w strukturach B4. 17b. 2002. Rev. 61.. 1995. Phys. G. 102. A.. 1996b. J. V. S. 1982. Berlin.M.. Sci. 1993. Lukas. 863-912. Navrotsky. Phys. Phys. and lattice dynamical properties of CdS. Munoz. Automating firstprinciples phase diagram calculations. Rezultaty uzyskane dla struktur sieci B4 oraz B3 charakteryzują się symetrycznymi lukami mieszalności. F. 2006. Mater. G. H. Hotje.. Związki Cd(Se.. Phase Equilib.. F. 74. 11169-11186. A. Phys. 1994. Kresse. M. S. Mater.P. S.... Phys. 2010. l’Université Pual Verlaine. 348-359. 23. 6944-6947. State Sci. 8993-9001. A. Eng. 1981. Generalized cluster description of multicomponent systems. 11-45. B3 i B1. 373. 62. Kattner. 27. 48. Mod. Physica A. Breidi.. K. 521-538. G. U. J.J.. Ceder... 558-561. 49. Comput. Tolbert. S. W przypadku struktur sieci B1 luka mieszalności wykazuje lekko asymetryczny charakter. Jug. A. ściennie centrowanej strukturze blendy cynkowej (B3). Asta. Anion substitution in zinc chalcogenides. Rev.. A.. Elastic. Rev. Ceder. 14251-14269.. M. 2003.. Rose. B. Wei. 24.. Density functional study of CdS1-xSex and CdS1-xTex alloys. ZnSe. D. Sci.K. detektorach promieniowania gamma. Asta. A..-H. M.S) w warunkach normalnych krystalizują w heksagonalnej strukturze wurcytu (B4) oraz metastabilnej. Phys.. K. Kresse. electronic.. Ab initio molecular simulation of the liquid-metal–amorphous-semiconductor transition in germanium. H.. Zhang. Comput. Arizona State University. F. 2000. Physica B. G. Cambridge.. 2012 Received in a revised form: December 3. Kresse.. Sci. Modelling Simul.. 124-130. Res. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set. Ciftci.. First-principles computation of the vibrational entropy of ordered and disordered Pd3V. 46424656. P. van de Walle. Structure stability and carier localization in CdX(X=S.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Burton. Springer-Verlag.. J. 270 K odpowiednio dla struktur sieci B4. 539-553. Hafner. 5972-5978. van de Walle... 1996.. J. B. W obliczeniach uwzględniono efekt drgań sieci. The wurtzite to rock salt transformation structural transformation in CdSe nanocrystals under high pressure. Se. Ghemid. 47. A. Mater. S. Contribution of the vibrational free energy to phase stability in substitutional alloys: Methods and trends. Arizona. Phys. Enthalpies of formation of CdSxSe1-x solid solutions. 2000. S. A.M. 75. Wyznaczone krytyczne temperatury mieszalności wynoszą 270 K. 2007. Sundman.. Garbulsky. Sol. Pod wpływem wysokiego ciśnienia struktury B4 oraz B3 zmieniają swoją formę krystaliczną i przekształcają się w gęstszą.. Appl.. B.. A. Davies. Tikhomirov. Colakoglu. Rev.. 5.. 334-350. M.. X. Deligoz. U. B. G. G... laserach oraz ogniwach słonecznych. E.. B. diodach elektroluminescencyjnych. Chem. G. Rubio. The alloy theoretic automated toolkit: A user guide. GaN-InN.. Binnewies.. Rev. Lattice constants and molar volume in the system ZnS. Ouendadji. 10... B3 oraz B1.. Rev. 2006. 2009.. 2002a... El Haj Hassan. PhD thesis. Ze względu na możliwość atrakcyjnych zastosowań półprzewodnikowe stopy CdSe1-xSx są w ostatnich latach przedmiotem rozważań teoretycznych oraz intensywnych badań doświadczalnych. Phys. Schultz. CdSe.. Zrealizowany proces obliczeń wskazuje na występowanie luk mieszalności w całym zakresie koncentracji CdSe1-xSx dla wszystkich rozpatrywanych sieci krystalicznych. II-V and II-VI compounds. Rev. and CdTe... First principles phase diagram calculations for the wurtzitestructure systems AlN-GaN. Thermodynamics of solid solution formation. Metz. PhD thesis. Self-driven lattice-model Monte Carlo simulations of alloy thermodynamic. 15-50. group III.B. structural and electronic properties of binary and pseudobinary semiconductors: an ab initio study. High pressure phases of group IV.. J.. a uzyskane rezultaty wykazują dobrą zgodność z dostępnymi w literaturze danymi eksperymentalnymi. 2002b. 128. Furthmüller. B. Journ.. Te) semiconductors. Y. O. BLENDY CYNKOWEJ ORAZ SOLI KAMIENNEJ Streszczenie Półprzewodnikowe związki Cd(Se. 2006. G. J. G. Meradji.D. 300 K. Phys.

and references cited therein). 40-007 Katowice. The phase diagram of the Zn1-xBexSe alloy was the subject of theoretical ab initio study with the use of the common tangent ISSN 1641-8581 351 – 356 . At elevated temperatures the two phases are more capable to be mixed over the wider concentration range on the Zn-rich side of phase diagram. Key words: clamping. point defects and their diffusion causes the degradation observed in the devices based on the ZnSe structures. University of Silesia. 2007. No. The calculated phase diagram shows the asymmetric miscibility gap with the upper critical solution temperature equal to 860 K (1020 K) when the lattice vibrations are included (excluded) in the free energy of the system. the alloying of ZnSe with BeSe was proposed to increase the resistance of the ZnSe structure to defect generation (Plazaola et al.. The knowledge of the T-x phase diagram might be very helpful in preparation of Zn1-xBexSe alloy.. Knowing that the BeSe compound is characterized by a strong covalent bonding (Vèrié.deniszczyk@us. 2003). 2010).. Presented results confirm the high quality of mapping of disordered alloy onto the lattice Hamiltonian. 2013. Due to mismatch of unit cell volumes and different elastic properties of ZnSe and BeSe constituents the preparation of the Zn1-xBexSe alloy may demand special conditions. Raman scattering study of lattice dynamics in Zn1-xBexSe alloy has proved that the atomic distribution in the alloy is not uniform but the ZnSe rich and BeSe rich regions form (Pagès et al. 1997). 2 PHASE DIAGRAM CALCULATIONS FOR THE ZnSe – BESE SYSTEM BY FIRST-PRINCIPLES BASED THERMODYNAMIC MONTE CARLO INTEGRATION ANDRZEJ WOŹNIAKOWSKI. Bankowa 12. what may indicate the presence of the immiscibility region in the (T-x) phase diagram of the alloy. Recent measurements have confirmed that the concentration of vacancies in Zn1-xBexSe alloy decreases with increasing concentration of Be atoms (Plazaola et al. However.edu.pl Abstract The T-x phase diagram of Zn1-xBexSe alloy is calculated by means of ab initio method supplemented with the lattice Ising-like model cluster expansion approach and the Monte Carlo thermodynamic computations.. 13. FEM 1.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. This findings indicate that the phase separation occurs in the samples investigated. The IIIV ZnSe compound is one of the promising materials for light emitting devices operating at short wavelengths (green and blue range). Poland *Corresponding author: jozef. 2003). However the dislocations. INTRODUCTION The alloying of semiconductors with the aim of tuning the band gap energy to achieve the values expected for device’s application in recent past was the subject of extensive experimental and theoretical research (Berghout et al. The investigations were focused especially on the wide-band gap II-IV semiconductors. We have proved that below the room temperature the miscibility of ZnSe and BeSe phases is possible only in the narrow range of concentration near the x = 0 and 1. groove rolling. JÓZEF DENISZCZYK* Institute of Materials Science.

while E (i ) structure i obtained from CE (equation (1)) using the (n – 1) other structural energies. The coefficients Jα of CE expansion (1) embody the information regarding the energetics of the alloy and are called the Effective Cluster Interaction (ECI). The present work aims to extend the study phase stability of Zn1-xBexSe system to cover both the configuration formation enthalpy and lattice vibrations free energy contributions to the total free energy of the system. For each superstructure the vibrational free energy was calculated within the quasi-harmonic approximation. 1984) is a method to parameterize the energy of a material as a function of its configuration. 2. With this aim the theoretical research was undertaken aimed to determine the T-x phase diagram of Zn1-xBexSe alloy by ab-initio method supplemented with the Monte Carlo (MC) phase diagram calculations within the semi-grandcanonical ensemble. which is recommended for difficult relaxation problems. in the phase diagram calculations the formation enthalpy for only few concentrations was considered. in our phase diagram calculations we assume the B3-type structure for Zn1-xBexSe alloy for the entire concentration range. Formation energies. Valence electron configurations for the pseudopotentials are Se = 4s24p4. supercell shape and atomic positions. the bond-lengthdependent transferable force constant approximation was used. 2002). To reduce the computational time needed for obtaining phonon densities of states for a set of superstructures involved in the cluster expansion procedure.. 2002. 1990) with the generalized gradient approximation (GGA) for exchange and correlation. 1996. Within the approximation the nearest- – 352 – . E  E Zn1 x Bex Se  xE BeSe  1  x E ZnSe  were calculated up to 20 atoms per unit cell. Garbulsky & Ceder. Furthermore. The predictive power of cluster expansion. 3. 1994. The cluster expansion (Sanchez et al. Based on the VASP results all remaining steps of the First Principles Phase Diagram (FPPD) calculations were performed with the use of Alloy Theo- retic Automated Toolkit (ATAT) software package (van de Walle & Ceder. The sum is taken over all clusters α that are not equivalent by a symmetry operation of the space group of the parent lattice.. Electronic degrees of freedom were optimized with a conjugate gradient algorithm. Zn = 3d104s2. 1996a. With this in mind. neglecting the lattice vibration contribution to the free energy of the system.. Luo et al. 2002a. Total energy calculations were done using the Vienna ab initio Simulation Package VASP (Kresse & Hafner.INFORMATYKA W TECHNOLOGII MATERIAŁÓW method of phase diagram construction (Berghout et al.. Each superstructure was relaxed with respect to volume. 1995. In the subsequent text the computational details are presented in Sec. ZnSe and BeSe. van de Walle et al. Kresse & Furthmüller.. while the average is taken over all clusters α′ that are equivalent to α by symmetry. defined by equation (1) is controlled by crossvalidation score defined by equation (2) 2 1 n ˆ )2  CV    ( Ei  E  (i )  n i 1  1 (2) COMPUTER METHODS IN MATERIALS SCIENCE where Ei is an ab inito calculated energy of superˆ represent the energy of superstructure i. Be = 2s2. All calculations were converged with respect to gamma centered k-point sampling. 1998). 1995). 1998. 1993. The multiplicities mα indicate the number of clusters that are equivalent by symmetry to a divided by the number of lattice sites. Ozolinš et al. COMPUTATIONAL DETAILS In the normal conditions the both end-member compounds. 2007 ). 2002.  i ' (1) where α is a cluster defined as a set of lattice sites. 2. The results are presented and discussed in Sec.. In the approach only formation enthalpy was taken into account. 1996b) using ultrasoft Vanderbilt-type pseudopotentials (Vanderbilt. crystallize in the zincblend (B3) type structure (Karzel et al. The energy.. and a plane-wave energy cutoff of 350 eV was used which yields  values that are converged to within a few meV per atom. In the first step VASP calculations were used to construct cluster expansion (CE) Hamiltonian. van de Walle & Asta. Typical wellconverged cluster expansion system contains of about 10-20 effective cluster interactions and requires the calculation of the energy of around 30-50 ordered structures (van de Walle et al. van der Ven et al. Section 4 is the conclusion. (per atom) is represented as a polynomial in the occupation variables given by equation (1): E     m J    i . The part of the free energy contributed by lattice vibrations was taken into account employing the coarse-graining formalism..

2002).1.006974 10 8. (CV = 0. 1) (-1. and the isotropic bending stiffness b. 0.965 -0.1) than the values of E itself. 1/2. 3/2) 7 6. 3/2) (-1/2.1.002262 base on the force constants deter(-1/2.012845 9 7..INFORMATYKA W TECHNOLOGII MATERIAŁÓW The ab initio calculations were performed for the end-member compounds ZnSe and BeSe and 33 reference superstructures containing up to 20 atb  oms.1) 6 5.1.1.1) (1/2. In the fitting procedure described in with only two independent terms: stretching stiffness Sec.001169 expansion. 1997) or the linear response duce the zero and point clusters. 1.910 -0. j   b ..001199 (0. 1. 1) (-1. 2002b). 1) sented in figure 2.1) 1 Zero cluster 0.000852 (0. 0. For all remaining superstructures the force constant matrices were predicted using the relaxed bond lengths and the chemical identities of bonds in each superstructure and employing the bond stiffness versus bond length functions evaluated for end-member compounds. 3/2) (-1. 0. the force constant matrix is approximated by the diagonal form described by equation (3) (Liu et al. With increasing distance between sites and low crystal symmetry of superstructures both of cluster the values of pair ECI falls down in an alternative approaches demand high computing oscillatory manner.009 -0.j . 1) (1/2. 1. 1) mined ab initio within the small (a) di.1.1.001999 14 2. This 2 Point cluster 0.0051) was obtained for 14 clusters in the An alternative way to determine the vibrational expansion (1). 0. RESULTS AND DISCUSSION COMPUTER METHODS IN MATERIALS SCIENCE . The bond stiff(1.1) 4 4.1) (1. 1991). All supercell energies are positive with re  spect to end-member compounds.003586 13 8. that the largest ECI intromethod (Parlinski at al. -1/2. Because of multiatomic composition cation sites.505 -0.. and are not applicaFigure 1 shows the formation energies E (per ble in the phase diagram calculations. no intermediate  i. Cluster Cluster Index Index eV/cation eV/cation besides the end-member comcoordinates coordinates Å Å pounds the residuals are at least (1. (3)   ground states was found.1. 2) one order of magnitude smaller (1.1) bond stiffness versus bond length 5 4. 2007) The phase diagram calculations were performed with the use of the Monte Carlo (MC) thermodynamic integration within the semi-grand-ensemble (van de Walle & Asta. individually for each clusters the largest ECI belongs to nearest neighbor superstructure. In this procedure. 2 the best cross-validation score s. 1) correct predictive power of CE (1.1) ness (b and s) constants calculated (1.000661 (-1/2.j is the distance of the longest pair within the cluster displacement approach the for – 353 – neighbor force constant matrix as function of bond lengths (volume) was calculated for the endmembers SeZn and SeBe of Zn1-xBexSe series.1.501 -0. The calculated and fitted energies (a) (a) di. power and are time consuming.1.019 -0. 1) (1. but ECI. do not coincide precisely. 1) relationships for Se-Zn and Se-Be (1.835 -0. cation atom) calculated for all reference superstructures by ab initio (VASP) method Table 1.000586 nearest neighbor bonds are pre(0.002446 12 8. 3/2) The bending and stretching (1. 1.j . di. It is evident. The cluster coordinates and correfree energy for each superstructure considered in the sponding effective cluster interactions are collected cluster expansion (1) is to apply the direct force in table 1.041703 8 6.1. Cluster coordinates and corresponding effective cluster interactions of clusters and using the cluster expansion taken into account in cluster expansion series defined by equation (1).003786 11 8. called transferable force constant approach (van de Walle & Ceder. 1/2. what indicates a miscibils  ity gap system.670 0. 3.1) (1.339 -0.505 0.001201 (1/2.1) computational methodology and 3 2. (1) using the ECI given in table 1. 1/2. 3/2) result confirms the quality of (1. ECI. 1.1.835 0. ½. (0. The Hamiltonian used in MC integration was of the cluster expansion form given by equation (1) with expansion parameters fitted to the formation enthalpy and vibrational free energy calculated by parallel computing. Among multisite theory (Giannozzi et al.1) (1.944 -0.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW Figure 3 shows the phase diagram of Zn1-xBexSe alloy calculated on the base of temperature independent ECI (fitted to formation enthalpies) and temperature dependent ECI (fitted to the sum of formation enthalpies and vibration free energies).65). a sequence of volumes of end-member compounds show the linear dependence on bond length. The arrows show the upper reduced solubility on the side of the diagram correcritical solution points. The contribution to free energy of vibrational degrees of freedom. Formation energies E calculated by VASP (cross) and obtained by Berghout et al.54) found by Berghout et al. can be related to the mismatch of ionic radii of cation ions (Zn – 74 pm and Be – 41 pm). indicating the enhanced miscibility of cation atoms on the Znrich side of phase diagram. The higher critical concentration resulted from our calculations can be attributed to an asymmetry of the phase diagram (figure 3). modifies considerably the phase diagram (figure 3). The asymmetric shape of calculated phase diagram might suggest the presence on the Zn-rich lefthand side of concentration range of some additional structure with the ground-state configuration energy lower than that of end-member compounds. It is well known. However our detailed search for the structures with negative ground-state formation enthalpy failed. An effect of viforce constants versus bond length for Se-Zn (left panel) and Be-Se (right panel) brations on the shape of phase diagram is bonds. The shape of our phase diagram differs noticeably from that calculated by Berghout et al. Calculated phase diagram based on the ECI fitted to the configuration formation enthalpy (dash line) and on the temperity gap between end--member compounds with ions ature-dependent ECI fitted to the sum of configuration and with very different ionic radius are asymmetric with vibration free energy (solid line). The critical concentration remains almost Fig. 2006). sponding to the smaller ion (Burton et al. The main effect of lattice vibrations is the shift of critical point to lower temperature (TC = 860 K). The critical temperature obtained within the restricted (to formation enthalpies) approach (TC = 1020 K) is significantly lower than the one Fig. The asymmetry of the phase diagram. 2.69. that the miscibilFig. With only the configurational degrees of freedom taken into account our modeling yields the critical point (xc. – 354 – COMPUTER METHODS IN MATERIALS SCIENCE . (2007) (TC = 1324 K). (2007) is much lower then our result. the phase diagram resulted from formation enthalpies solely is reduced slightly. 1. d) unchanged (xC = 0. 3.. (2007). 1020 K). TC) = (0. where no asymmetry can be observed. c) and bending b (figures b. Nearest neighbor stretching s (figures a. The main feature of the calculated phase diagram is the presence of asymmetric miscibility gap with upper critical solution temperature (TC). Crosses indicate an ab initio data points and lines represent linear fits negligible although the recess visible on used in the calculations of the vibrational free energy. fitted by cluster expansion Also the critical concentration (xc = 0. when taken into account.

Firszt.. Ghandehari. At elevated temperatures the two phases... Rev. Asta. Mater. 74.. Ceder. 78.. D.. Ceder. A.. Y.. M. W obliczeniach termodynamicznych uwzględniono również wkład do energii swobodnej pochodzący od drgań sieci. Ceder. A. Rev. Kattner. D.. Expected pronounced strengthening of II-VI lattices with beryllium chalcogenides.. G. Rev. 2010. 128. 1984. Schwarz. A. Phys. F. B. M. J. Hafner. B.. 4063-4066 Plazaola.. A.. 11169-11186... Saarinen. F. 6. B. Phys. 1995... 52. 7231-7242. 2002. K. Pagès. J. Dobrzynski. Rev. Tournie.. G. 1990. M. E.. Sanchez. Phys. A. Rev. Ducastelle. Obliczenia termodynamiczne oparte wyłącznie na części konfiguracyjnej energii swobodnej – 355 – COMPUTER METHODS IN MATERIALS SCIENCE . Ceder. Giannozzi. 51. Asta.V. Kresse. Garbulsky. H. 1996a. Self-driven lattice-model Monte Carlo simulations of alloy thermodynamic.. 57. 1998. van de Walle. Blaha. B. Rev. Köfferlein.M. 2007.. J. 73... B.. Mitchell. 11425-11438. Kresse... Journ. Ab initio molecular dynamics for liquid metals.L. 43. Schiessl.. Phys. Phys. P. 60-64. and AlN-InN. Diagram fazowy T-x wyznaczono na podstawie potencjału termodynamicznego wyliczonego metodą całkowania termodynamicznego Monte Carlo w ramach wielkiego rozkładu kanonicznego. Calphad. Hugel.65 of Be. T. van de Walle.. B. A. Phase transformation of BeSe and BeTe to the NiAs structure at high pressure. Aydinol. Phys. 94. Paszkowicz. Firszt.. Appl. G.. 75. M. 43. Phys.W. B 47. Zr.K. 67-72. M. S. 54.Z.. The alloy theoretic automated toolkit: A user guide. Phys. Bormann. are more capable to be mixed over the wider concentration range on the Zn-rich side of phase diagram. W obliczeniach termodynamicznych wykorzystano Hamiltonian sieciowy w postaci tzw. Rev. 16471653.. G. The upper critical solution temperature of 860 K is predicted at concentration x = 0. 100. Kresse. Automating firstprinciples phase diagram calculations.. Transferable force-constant modeling of vibrational thermodynamic properties in fcc-baased Al-TM (TM = Ti. Eng. Ceder. M. Parlinski. Phys. Kresse. Lattice dynamics and hyperfine interactions in ZnO and ZnSe at high external pressures.. 558-561. Firstprinciples study of the energy-gap composition dependence of Zn1-xBexSe ternary alloys. W.. Modelling Simul. Meczynska. Vanderbilt. Zaoui..Q.J. G. A. W.. Appl. S. Non-random Be-to-Zn substitution in ZnBeSe alloys: Raman scattering and ab initio calculations. de Gironcoli. Phys. J. B. Postnikov. J. 11-45. Uzyskany z obliczeń diagram fazowy T-x charakteryzuje się asymetryczną luką mieszalności.. Mater...Rev. 1997.. G.. 1991.. GaN-InN. 7058-7064. 23. van de Walle. Zunger. M. D. 461-469. Sci. Paszkowicz. 6427-64443.. Furthmüller. H. Rev. U..D. C. O.. Generalized cluster description of multicomponent systems.INFORMATYKA W TECHNOLOGII MATERIAŁÓW CONCLUSIONS Ab initio based Monte Carlo modeling of x-T phase diagram of Zn1-xBexSe alloy has shown that in the normal condition the solubility of Zn and Be cations is possible only on the Zn-rich side of phase diagram.. Flyktman. K. Eur. Cu-Ag and Ni-Au intermetallics: First-principles study of temperature-composition phase diagram and structures. W. Das. Defect characterization of ZnBeSe solid solutions by means of positron annihilation and photoluminescence techniques. Glas. 521-538.. H. Phys. G. Rev. K. J. P.. Phys. D. 1993. OBLICZENIA Z PIERWSZYCH ZASAD DIAGRAMU FAZOWEGO UKŁADU ZnSe–BESE METODĄ CAŁKOWANIA TERMODYNAMICZNEGO MONTE CARLO Streszczenie W pracy przedstawiono rezultaty badań teoretycznych stabilności fazowej roztworu stałego Zn1-xBexSe w zależności od temperatury i koncentracji składników.P. 15-50. van de Walle. Sci.. L. Ferhat. Hafner. 1998.. van de Walle.P.. DiSalvo.. J. Furthmüller.. A. B. Współczynniki energetyczne rozwinięcia klastrowego wyznaczono drogą dopasowania Hamiltonianu sieciowego do wartości entalpii tworzenia wyliczonych kwantowymi metodami z pierwszych zasad dla 33 nadstruktur Zn1-xBexSe w całym zakresie koncentracji. 14251-14269. 10.. Pasternak. Ag-Au. Burton.. Phys.. First-principles investigation of phase stability in LixCoO2. Phys. Karzel.Lett. który wyznaczono dla 33 nadstruktur w ramach przybliżenia kwazi-harmonicznego. van der Ven. Gosh. K.P.. B. Steiner. J. Rev. The effect of lattice vibrations on substitutional alloy thermodynamics.. G.. 1997. J. Potzel.. B.. 113528-113534. M.. Sci. Reniewicz... Asta.. Kawazoe. G.. 2975-2987. A. S. Pavone. F. ZnSe and BeSe.. Hiller. J.... Eng. Linear-programming method for obtaining effective cluster interactions in alloys from total-energy calculations: Application to the fcc Pd-V system. Comput. First principles phase diagram calculations for the wurtzitestructure systems AlN-GaN. Baroni.. 49.. Legowski.. Rev. R. F..G. Ruoff.. Liu. 2002. G. B.. A. P. B... Chafi.. 2006.. 334-350.. above which the solubility of cations is possible over the all concentration range. van de Walle. 1996b.. A. 75.. Ab initio calculation of phonon dispersions in semiconductors. 26. G. G. H. 53. First-principle determination of the soft mode in cubic ZrO_2. J. Gratias. B. 104117-15.. 2002a. Mod. 41. Phys. B..... 7892-7895. 539-553. Phase Equilib. Rev. Vèrié. 348-359. Trail.. G. Hf) alloys. Ab initio molecular simulation of the liquid-metal–amorphous-semiconductor transition in germanium. P.M. 2002b.. S. 1994. 58. Phys. F. Kalvius. A.. Efficient iterative schemes for ab initio total-energy calculations using a planewave basis set. 2003. Physica A.. F. rozwinięcia klastrowego (typu modelu Isinga).. Kresse. Ozolinš. 2007. A. Soft-selfconsistent pseudopotentials in a generalized eigenvalue formalism. Phys. 205112-205121. 1996. C. U. Greene. Hafner. Mater. Simon. Li Z. V. Cu-Au. W. Rev. Luo. J. Wolverton. REFERENCES Berghout. 1995. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set.S.

Prezentowane badania wykazały. W podwyższonej temperaturze. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 356 – . 2012 Accepted: November 30. 2012 Received in a revised form: November 19. Received: September 20. Uwzględnienie drgań sieci daje obniżoną temperaturę krytyczną TC = 860 K (xC = 0. mieszalność faz ZnSe i BeSe jest możliwa również w roztworach Zn1-xBexSe o wzbogaconej zawartości cynku.69 i TC = 1020 K. 400 K < T < TC. że poniżej temperatury pokojowej mieszalność faz ZnSe i BeSe jest możliwa wyłącznie w wąskich zakresach koncentracji (x  0 oraz x 1).INFORMATYKA W TECHNOLOGII MATERIAŁÓW dają diagram fazowy z górnym punktem krytycznym: xC = 0.65).

cellular automata finite element (CAFE) method was used to investigate microstructure evolution during four passes of ECAP and its incremental variant I-ECAP. ul. Poland 3 Warsaw University of Technology. many researchers have put an effort to obtain UFG magnesium alloys which. 30-059 Kraków.. ultrafine grained metals. which could be useful to estimate mechanical properties of the produced UFG metal. Mathematical description of interactions is introduced by transition functions (transition rules). Cellular automata finite element (CAFE) method Cellular automata (CA) technique is used in material science to provide digital representation of material microstructure and simulate its evolution during processing (Das. due to their low density. United Kingdom AGH University of Science and Technology. An internal state variable model describes total dislocation density evolution and transfers this information to the CA space. INTRODUCTION 1. Mickiewicza 30. 2 MODELLING MICROSTRUCTURE EVOLUTION DURING EQUAL CHANNEL ANGULAR PRESSING OF MAGNESIUM ALLOYS USING CELLULAR AUTOMATA FINITE ELEMENT METHOD MICHAL GZYL1. However. Cellular automata finite element (CAFE) approach is ISSN 1641-8581 357 – 363 .com Abstract 1 Equal channel angular pressing (ECAP) is one of the most popular methods of obtaining ultrafine grained (UFG) metals. The cellular automata space dynamics is determined by transition rules. 2013. The CAFE simulation results were verified for the temperature range 200-250 °C and strain rate 0. called the cellular automata space. LECH OLEJNIK3 2 University of Strathclyde. *. Glasgow G1 1XJ. Narbutta 85. it can be widely used in industrial practice. ANDRZEJ MILENIN2. so called I-ECAP.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol.gzyl@gmail. al. 2006. 75 Montrose Street. magnesium alloys. Poland *Corresponding author: michal. cellular automata finite element method 1. Since I-ECAP can deal with continuous billets.01-0. are very promising materials for weight and energy saving applications. 02-524 Warszawa. Madej et al. good agreement with experimental data was achieved. strain rate and temperature obtained from FE simulation. The current state of each cell is determined by the state of its neighbours and its own state in the previous step.5 s-1. only relatively short billets can be processed by ECAP due to force limitation. In this work. 2012). Key words: severe plastic deformation. ANDRZEJ ROSOCHOWSKI1. The developed CAFE model calculates the mean grain size and generates a digital microstructure prediction after processing. Recently. whose parameters are strain. No. It was reported that microstructure refinement during ECAP is controlled by dynamic recrystallization and the final mean grain size is dependent mainly on processing temperature. The interactions between cells describe dynamics of a simulated physical phenomenon. A solution to this problem could be recently developed incremental variant of the process. Material is represented as a lattice of finite cells. 2010. 13. equal channel angular pressing. Svyetlichnyy.1. Fitting and verification of the model was done using the experimental results obtained from I-ECAP of an AZ31B magnesium alloy and the data derived from literature.

2. The cellular automata (CA) space plays a role of digital material representation in meso scale. Equal channel angular pressing (ECAP) was developed by Segal (1995). the billet is fed by a small increment. Figueiredo and Langdon (2010) presented a model. which have the same cross section (circular or rectangular) and intersect at an angle that usually varies from 90° to 135°. Increase in dislocation density is caused by storage of dislocations while decrease in dislocation density results from annihilation of dislocations.INFORMATYKA W TECHNOLOGII MATERIAŁÓW the example of a multi scale modelling approach. Plastic strain is introduced into metal by simple shear. a billet is pushed through a die with two channels. Due to force limitation only relatively small amount of material can be processed at each stage. Since the billet dimensions remain unchanged. 2008.. This is the combination of micro scale modelling using cellular automata and finite element (FE) analysis.. incremental ECAP (IECAP) has been developed by Rosochowski and Olejnik (2007). 2007. the newly formed recrystallized grains are not able to fully consume the initial coarse grains. the process can be repeated in order to accumulate a desired high strain. Recently. the most popular method of metal forming simulations. Random pentagonal neighbourhood and absorbing boundary conditions are used to define CA space and its dynamics. which states that the mechanism of grain refinement during ECAP processing of AZ31 magnesium alloy is dynamic recrystallization (DRX). They also introduced a ‘critical grain size’ term in order to explain that homogenous grain size distribution is possible to achieve only if the initial mean grain size is small enough. 2. Microstructure evolution is described by transition functions. in which a very large strain is imposed to the material to refine its grain microstructure and improve its strength. the stages of material feeding and plastic deformation are separated. Lapovok et al. Overview In the present work. Therefore. The change of dislocation densi- – 358 – . 2009). Severe plastic deformation (SPD) processes. especially when ultra fine grained materials are considered. Then. Ding et al. the same as in classical ECAP. artificial grains are represented by CA cells. That is the reason why ECAP is not widely used in industrial practice. The mode of deformation is simple shear. whose parameters are the macro scale integration point variables obtained from FE simulation: strain. The internal state variable method (Pietrzyk.1. In the ECAP process. 2.3. is used to describe changes that occur in the micro scale during plastic deformation. The tool configuration consists of a punch working in a reciprocating manner and a die leading and feeding the material in consecutive steps. In I-ECAP. the punch approaches and plastically deforms it. Equal channel angular pressing and its incremental variant 1. The advantages of using multi scale simulation approach are evident. microstructure evolution during ECAP and its incremental variant is modelled using cellular automata finite element (CAFE) technique. When the punch moves away from the die. cellular automata transition rules usually used to describe DRX during hot forming of metals are applied in the presented model. Dislocation density evolution Evolution of dislocation density during hot metal forming is controlled by two competing processes: strain hardening and thermal softening caused by recovery or recrystallization. MODEL DESCRIPTION 2. 2002).2. This hypothesis is based on the occurrence of a bimodal microstructure after ECAP processing. strain rate and temperature (mapped from FE nodes). when feeding stops and the billet is in the fixed position. are very promising. This observation was made earlier by other researchers (Janeček et al. which treats dislocation density as a material variable. Otherwise.. By using the CAFE method not only the mean grain size but also microstructure homogeneity and grain size distribution can be calculated. which reduces the feeding force dramatically. 1. Microstructure evolution during ECAP of Mg alloys COMPUTER METHODS IN MATERIALS SCIENCE In order to improve mechanical properties of magnesium and other metals many thermomechanical processes have been proposed. which occurs at the channel intersection. Dislocation density in each cell controls the nucleation process.

2006). Cellular automata space evolution where: M – Taylor factor. T – temperature. grain grows until its virtual energy related to its grow potential is greater than zero. which describe how each CA cell state changes depending on its own and its neighbour state after the previous time increment. p2 – fitting parameters (table 2). privileged areas for the dislocation density rise are grain boundaries and their vicinity (Galiyev et al. Decrease in dislocation density is dependent on temperature as dislocation annihilation is characterized as thermally activated process (Madej et al. A1 3.32 .. A1. Parameters of dislocation density evolution equation (1).3. kJ/mol (Das. Q1. This is a simplified form of the formula introduced by Roberts and Ahlblom (1978): T p2   crit  p1   1/ 3 (2) where: γ – grain boundary energy.01 s-1. Fitting coefficients were determined using Hooke-Jeeves optimization method..01 s-1 based on these parameters. Orientation of each grain is described by three Euler angles: φ1. dislocation accumulation is more probable for grains with orientation closer to basal than prismatic or – 359 – COMPUTER METHODS IN MATERIALS SCIENCE 0. b – Burgers vector.38 b. A critical value of dislocation density in a CA cell must be reached to create a new grain nucleus in this cell. Parameters of equations (1) were derived from literature (table 1). Table 1. (2000). Mathis et al. φ2 (Bunge notation). 1.  cell changes its state to recrystallized when one of its neighbours is recrystallized. together with experimental results obtained for pure magnesium in similar conditions (Klimanek & Poetzsch. kJ/mol (Barnett. 2010) 21 M. 2006) 2. A2 – fitting parameters. 2000): d iCA  d iCA  d iCA  A1 M  Q2  CA Q   exp 1 d iCA  A2  iCA   d i 1 exp  b  RT   RT  (1) 2. Grain boundary energy is dependent on the misorientation angle between neighbouring grains and is evaluated from the equation derived by Read and Shockley (1950) for low angle grain boundaries. Since the slip on basal plane is the most favourable deformation mechanism for magnesium. R – gas constant. 2003). Q1 – effective activation  – efenergy. The nucleation process will be favoured where the level of stored energy is higher comparing to other areas.INFORMATYKA W TECHNOLOGII MATERIAŁÓW ty due to strain increment is described by the equation (1) (Sellars and Zhu. The CA space dynamics can be described by the following rules:  cell becomes the site of nucleation when its dislocation density exceeds a critical value. are illustrated in figure 1. 2002. 2006) Fig.. Q2 – apparent energy. (Chino. Φ. The presented approach is similar to the model developed by Mecking and Kocks (1981) where dislocation density evolution is also introduced as competition between dislocations storage and annihilation. nm (Chino. 2003) 147 Microstructure evolution during processing is modelled using transition rules. the equation (2) is used to calculate critical dislocation density.  iCA 1 – dislocation density at previous strain increment. for high angle grain boundaries it is kept constant.85e3 A2 30 Q2. Following these arguments. Wang and Huang (2003) showed the relation between crystallographic orientation and texture component in hcp metals.  fective strain rate. new grains cannot be consumed by other recrystallized grains. dε– effective strain increment. Dislocation density dependence on strain at 200°C and strain rate equal to 0. 2004). p1. Misorientation between neighbouring grains is calculated using a method presented by Zhu et al. Dislocation density evolution at 200 °C and strain rate 0.

The initial microstructure of I-ECAP processed material was heterogeneous. RESULTS FE simulations were run using Abaqus/Explicit commercial software.5 s-1. 225 and 250°C and strain rate within range 0. AZ31B billets and dies were heated up to 250°C using electric heaters. COMPUTER METHODS IN MATERIALS SCIENCE p1 4. it corresponded to 100 x 100 µm2 area of real material. The corresponding digital representation of asreceived material was obtained by uniform grain growth and simulation of DRX (figure 2b). CA microstructure evolution calculations and visualisation were performed using a self-developed software.  – misorientation angle.092 p7 1.3 µm (figure 2d).. Route BC was used. A double-billet variant of I-ECAP was realised using a 1 MN hydraulic servo press (Rosochowski et al. Since grain boundaries are prevailed areas for nucleation. At each time step. more new nuclei are expected to occur when there are more grains in the CA space. Table 2. Grain growth rate is associated with virtual energy assigned to each new nucleus. The billets were fed using a screw jack whose action was synchronised with the reciprocating movement of the punch. additional experimental results obtained for different process parameters were derived from literature (table 3). Strain rate values during both processes were significantly different. p5.6.1463 p8 0. In order to investigate temperature and strain rate effect on the final mean grain size and microstructure homogeneity. The grain growth energy is introduced using an empirical equation (4): E grain  p 7 exp( p8T )10 /   0. p8 – fitting parameters. Simulations were performed for temperatures: 200. The same conditions as during I-ECAP experiment were used in simulation.4176 p3 0. Processing at elevated temperatures is being simulated.8 mm. the total dislocation density increment is divided by a number of randomly chosen cells N. based on the readings obtained from a thermocouple located near deformation zone. coarse grains were surrounded by smaller ones (figure 2a). p6 – fitting parameters. p4. The parameter is dependent on temperature and strain rate obtained from FE simulation. The punch movement followed an externally generated sine rods. which extrinsic dislocation density will be increased.6677e18 p4 0. The process is stopped when the grain growth energy is equal to zero. In this condition. As the new grain grows its energy is being lowered. p7.094 (3) where: N – number of cells with increased extrinsic dislocation density.INFORMATYKA W TECHNOLOGII MATERIAŁÓW pyramidal one.15 µm (figure 2c) what is similar to the results obtained by Suwas et al.01-0. Since mean DRXed grain size is dependent mostly on temperature. twinning was not revealed in microscopic observations. which meant that after each pass the billet was rotated by 90°. the grain has no potential for further expansion. 4. – 360 – . The feeding stroke was equal to 0. grain growth energy is the function of temperature and misorientation angle.5 Hz and amplitude equal to 1. which resulted in the total strain of 4. what could be attributed to DRX during hot extrusion of supplied p5 1.7  gg (4) where: T – temperature.2 mm. nd – number of grains in CA space. Evolution of N is given by equation (3):  p4 nd 5 exp( p 6T ) N  p 3 p waveform with frequency 0. Temperature during processing was kept constant within ±2 °C.0756e70 p2 -31. The simulated microstructure was similar to the real one and the predicted grain size was 6. EXPERIMENTAL PROCEDURE Commercially extruded AZ31B magnesium rods with 17 mm diameter were machined using the EDM cutting technique in order to obtain bars with square cross-section 10 x 10 mm2 and length equal to 100 mm. p3. Parameters of CA space evolution equations (2-4). therefore twinning is not taken into account.06699 p6 -0. CA space dimensions were 400 x 400 cells. (2007). The mean grain size obtained after I-ECAP processing was equal to 6. 2008). which could indicate that processing temperature has a dominant effect on the final grain size. Moreover. A motor driven screw jack was controlled using National Instruments hardware and software (LabVIEW). A die with 90° angle between channels was used to conduct four passes of I-ECAP.004261 3.

it is – 361 – COMPUTER METHODS IN MATERIALS SCIENCE . Grain refinement is limited by a temperature. (2006) 250 225 200 225 Strain rate. 2. Microstructure images of as-received material (a) and after 4th pass of I-ECAP at 250 °C (c). c) d) Fig. (2009) conducted ECAP of AZ31 magnesium alloy at 200°C and 225°C using a die with 120° channel angle. (2009) Jin et al. Initial temperature.4 µm. The mean grain size after first pass was measured to be 4. Ding et al. (2009) Ding et al.7 µm. (2006) was equal to 15. the same as the average grain size of CA representation. A digital material representation was generated using non-uniform grain growth: 80% of grains grew slower than others and the mean grain size was 15. The initial mean grain size of the material processed by Jin et al. coarse grains were surrounded by colonies of very small a) b) c) d) Fig.3 0.5 0. respectively.8 µm. Simulations parameters. The initial microstructure and its digital material representation are shown in figure 3. 3. a) b) a) b) c) d) Fig. respectively. After 4 passes at 200°C the grain size was reduced to 1. When processing at 225°C. After first pass at 225°C. coarse grains and few smaller grains were observed (figure 4a).6 µm.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 3.1 µm while the model predicted 4. corresponding simulation results in (b) and (d). The microstructure obtained after 4 passes at 200°C and the corresponding results of modelling microstructure evolution are shown in figure 3 as well. (2009). (2006). 4. The mean grain size after first pass depends strongly on the initial microstructure. °C I-ECAP (this work) Ding et al. The initial mean grain size was 7.3 0. Microstructure images of as-received material (a) and after 4th pass of ECAP at 200 °C (c) obtained by Ding et al. corresponding simulation results in (b) and (d). Results obtained from the developed CAFE model are in good agreement with experimental data (figure 5). s-1 0. corresponding simulation results in (b) and (d).25 µm. Further deformation is needed to refine and homogenize a microstructure.8 µm while the model predicted 1. ° 90 120 120 90 grains (figure 4c). while experimental result was 2.2 µm. heterogeneous grain size distribution is obtained. respectively. A significant grain refinement is observed after first pass but it is not sufficient to refine the coarse grain dominated microstructure. Microstructure images of as-received material (a) and after 1st pass of ECAP at 225 °C (c) obtained by Jin et al. The interior of coarse grains was not consumed during DRX process. Only grains smaller than ~15 µm can be fully recrystallized during first pass of ECAP at 200°C. The mean grain size calculated using developed model was equal to 2.01 Channel angle.2 µm. As a result. the microstructure was less homogenous than at 200°C.

Numerical results were verified using experimental data from conventional and incremental ECAP. A.. Future work will be focused on modelling grain reorientation due to deformation by simple shear.D. REFERENCES Barnett.. FEM simulation of incremental shear. B.. Sakai. J. Axt. R.. I. Zeng. 145-150..R. 52. C. 2008.. Materials processing by simple shear. L.. A. Rosochowski. Rosochowski. 29. Fig. S. Nyilas. 53-62. 45.... T.X. Estrin.. 584-586. Microstructure evolution of AZ31 Mg alloy during equal channel angular extrusion. S. Olejnik. Pietrzyk. R. W.V. 1978.. S. W. 1-7. 419-422. Materials Science Forum.M. M. L.. Chino... W. T. eds. Chang. 754-760. a heterogeneous grain size distribution after first pass of ECAP for initially coarse grained microstructure was predicted as well as its further homogenization.. 503-508. Materials Science and Engineering A..P. Development of the Multi-scale Analysis Model to Simulate Strain Localization Occurring During Material Processing. Financial support from Carpenter Technology Corporation is kindly acknowledged. 1950. Cueto. Lapovok. Gottstein. Journal of Materials Science. Mathis.. R.. CONCLUSIONS COMPUTER METHODS IN MATERIALS SCIENCE A multi scale CAFE approach was developed in order to model microstructure evolution during equal channel angular pressing. 21. Jin. Read. 4827–4836. 5. Res.. Q. 2002. 287–318. Metallurgical and Materials Transactions A. Hodgson. T. Kinetics of flow and strainhardening. 1– 14.. Proc.B. Estrin. 2006. 2003. Conf. Langdon. Enhanced Superplasticity in a Magnesium Alloy Processed by Equal-Channel Angular Pressing with a BackPressure. Hoshika. U. Materials Science and Engineering A. 2000. M. Lukac.. Kaibyshev. 139-144. Pietrzyk.G. Kocks. Materials Science and Engineering A.. Advance Engineering Materials. Ding.. L.. 157-164. Mater. 2008.. latter was obtained from I-ECAP experiment. 116–120. Materials Science Forum. A nucleation criterion for dynamic recrystallization during hot working. In particular.. M. V. Dragomir-Cernatescu. Recrystallization during and following hot working of magnesium alloy AZ31... M.-S. Chen. T. Esaform 2007. D.. Y. M. 801-813. 1865-1875. Segal. Mechanical properties and microstructure of a Mg alloy AZ31 prepared by equal-channel angular pressing.. 275–289.. K.. Modeling mixed microstructures using a multilevel cellular automata finite element framework.. Kao. Microstructural modelling of aluminium alloys during thermomechanical processing. 2010. M. 47. A.INFORMATYKA W TECHNOLOGII MATERIAŁÓW shown that the mean grain size as small as 2 μm can be obtained at 200°C. F. 2007. K. 1995. Rev. P. L. Ungar. 2007. Lin. 2006. Acta Metallurgica. Effects of Processing Parameters on the Grain Refinement of Magnesium Alloy by Equal-Channel Angular Extrusion. 247-252.125-126. Roberts. Computations were performed for various temperatures and strain rates that are typical for processing of magnesium alloys... 705-711. 415-425. Poetzsch. Microstructure evolution under compressive plastic deformation of magnesium at different temperatures and strain rates. The model correctly predicted both the mean grain size after subsequent passes of ECAP/I-ECAP and microstructure homogeneity. Reorganization of cellular space during the modeling of the microstructure evolution by – 362 – ... Ding.. 419-422. 2007. M. M. Klimanek.. Mao.W. 509-514. Chinesta... 2004. 2889–2894. Grain refinement and mechanical behavior of a magnesium alloy processed by ECAP.. Mean grain size obtained from experiments and CAFE simulations.. 2012.F. T. Through-process modeling of microstructure evolution in hot forming of steels. Mecking. 471.. P.. Olejnik.G. Svyetlichnyy. Hellmig. Acta Materialia. Madej. Langdon. 2010. 280. Although smaller grain size cannot be obtained at given temperature. Spain. 462.. D.. B... The evolution of non-basal dislocations as a function of deformation temperature in pure magnesium determined by X-ray diffraction. 197. Continuous Dynamic Recrystallization in Magnesium Alloy.G. Zhu.. D. Popov.J. 2006. Journal of Materials Processing Technology. P. H. Acknowledgements.. Popov. X. Computational Materials Science.. 2009. Acta Metallurgica. Kumar.M. 2002. Mechanical properties of AZ31 Mg alloy recycled by severe deformation. Suwas. A. 78. R. A. 1981. 429-433. R. 10. Evolution of crystallographic texture during equal channel angular extrusion (ECAE) and its effects on secondary processing of magnesium. further processing leads to microstructure homogenization... T. Y. E. 40A. P. Materials Science and Engineering A. Janeček. 5.. Richert. Double-Billet Incremental ECAP. Ahlblom.. Sellars.. Phys. 26. Das. 653-658. J. Materials Science and Engineering A. C.. Zaragoza... Krieger. Y. S. Dislocation Models of Crystal Grain Boundaries. Lee. Galiyev. W.. G. Former were derived from literature. Figueiredo. 423. 2003. 324. 16. Materials Science and Engineering A. Arch Comput Methods Eng. Shockley. Materials Science Forum.

01-0. 2000. że mechanizm rozdrobnienia ziarna podczas ECAPu jest sterowany przez proces rekrystalizacji dynamicznej. Rozwiązaniem problemu może być opracowany inkrementalny wariant tego procesu. Huang. I-ECAP. a ostateczna średnia wielkość ziarna jest zależna głównie od temperatury procesu. Y. W niniejszej pracy sprzężona metoda automatów komórkowych i elementów skończonych (cellular automata finite element – CAFE) została wykorzystana do opisu rozwoju mikrostruktury podczas czterech przejść ECAPu i jego inkrementalnego wariantu.5 s-1. Ostatnie wyniki wskazują. 81. Model zmiennej wewnętrznej opisuje wzrost całkowitej gęstości dyslokacji i przekazuje tę informację do przestrzeni automatów komórkowych. 2003. Wyniki symulacji metodą CAFE zostały zweryfikowane dla zakresu temperatur 200-250°C oraz prędkości odkształcenia 0. J. Calculation of misorientation distribution between recrystallized grains and deformed matrix. 2012 Received in a revised form: November 4. 37–41. Materials Chemistry and Physics. W. – 363 – COMPUTER METHODS IN MATERIALS SCIENCE . Wang.. Jednak z powodu dużych sił potrzebnych do przeprowadzenia procesu. 153–162. co może być przydatne w wyznaczaniu własności mechanicznych otrzymanego materiału. Texture analysis in hexagonal materials. 11–26. że przy użyciu I-ECAPu mogą być przetwarzane nieskończenie długie elementy. których parametrami są odkształcenie. Computational Materials Science. 60. uzyskano bardzo dobrą zgodność z wynikami eksperymentalnymi. I-ECAPu. tylko relatywnie krótkie wstępniaki mogą być wyciskane. Mechanizm rozdrobnienia ziarna podczas przeróbki plastycznej stopów magnezu różni się znacząco od metali takich jak aluminium lub miedź i ich stopy. 2012 Accepted: November 21. G. Zhu. prędkość odkształcenia oraz temperatura – uzyskane z symulacji metodą elementów skończonych.. Yu. Ze względu na to. Y. 42. 2012 MODELOWANIE ROZWOJU MIKROSTRUKTURY PODCZAS RÓWNOKANAŁOWEGO WYCISKANIA KĄTOWEGO STOPÓW MAGNEZU PRZY UŻYCIU METODY CAFE Streszczenie Równokanałowe wyciskanie kątowe (equal channel angular pressing – ECAP) jest jedną z najpopularniejszych metod otrzymywania ultra drobnoziarnistych metali.. Dynamika zmian w przestrzeni automatów komórkowych jest determinowana przez reguły przejścia.C. N. Received: September 20. Mao..INFORMATYKA W TECHNOLOGII MATERIAŁÓW frontal cellular automata. Scripta materialia.. tzw. może on znaleźć szerokie zastosowanie w praktyce przemysłowej. Dopasowanie oraz weryfikacja modelu zostały wykonane przy wykorzystaniu wyników uzyskanych z przeprowadzonego procesu inkrementalnego ECAPu stopu magnezu AZ31B oraz danych literaturowych. Opracowany model CAFE oblicza średnią wielkość ziarna oraz generuje cyfrowy obraz mikrostruktury.

COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol. 2001). so that two (or more) "Kirkendall planes" can be expected? (van Dal at al. 2013. 1990). Gusak. thin-film electronic devices (Boettinger at al. both experimental characterization and computer simulations have revealed many new phenomena that are not yet fully explained by existing theories and models. FMSci&C. No. nonstability and discontinuity (van Loo at al. (ii) stability of individual Kirkendall plane. could the inert particles placed at the initial contact interface migrate differently in the diffusion zone. The bi-velocity method (Darken method) base on the postulate of the unique transport of the mass due to diffusion. Is the Kirkendall plane unique? In other words. interdiffusion. 2001. manifests itself in many phenomena. 13. the efforts directed to- wards its implementation into physics and thermodynamics are far from being accepted. AGH University of Science and Technology. The method deal with the 1) composition dependent diffusivities. 2007. The reason may be attributed to the inherent experimental difficulties involved in the measurement of material velocities.. A list of examples include: (i) multiple Kirkendall planes (bifurcations). generation of stress and in plastic deformation of the material. trifurcations. Poland *Corresponding author: bwierzba@agh. coatings. 30-059 Kraków. Al. 1947) always accompanies interdiffusion. INTRODUCTION In recent years.edu. the development of porosity. These diffusion-induced processes are of concern in a wide variety of structures including composite materials. However the existing methods dealing with Kirkendall trajectories do not quantify the bifurcations. momentum and energy conservations (Danielewski 364 – 367 ISSN 1641-8581 .pl Abstract In this study some aspects of Kirkendall and lattice plane migration in binary diffusion couples are studied by means of numerical simulations by bi-velocity method. It is shown that the method allows for calculation the trajectory of the Kirkendall plane in binary diffusion couples. While the Darken's treatment of diffusion has withstood the test of time. bi-velocity method. 2010). 2) different partial molar volumes of components. trajectory 1... Key words: kirkendall plane. and (iii) discontinuity of the Kirkendall velocity at moving interphase boundaries (van Dal at al. Mickiewicza 30. The Kirkendall effect (Smigelskas & Kirkendall. a number of fundamental questions still remains to be answered. weld junctions. the migration of inclusions inside the diffusion zone. 1990). The method based on the rigorous mathematical derivation of mass. Regardless of intensive work in this field. The rationalization and formal description of the Kirkendall effect are by no means trivial. 2 THE MIGRATION OF KIRKENDALL PLANE DURING DIFFUSION BARTEK WIERZBA Interdisciplinary Centre for Materials Modelling... Experiments confirm that the “fiducial markers” may have different trajectories. The bi-velocity method is a generalization of Darken method of interdiffusion. van Loo at al. 3) the stress field during the diffusion process and 4) entropy production.

(2)  s    i si . s. respectively. The bi-velocity allow also to calculate the energy and entropy conservations. J id   i id . the internal energy conservation law becomes:  i     J id  J idrift  . 2010). in a case when no external forces are considered is given by the Nernst-Planck equation (Nernst. i. RESULTS In this paper the three different methods of calculation of the position of Kirkendall plane in the binary A-B diffusion couples are shown and compared. i 1 2 (4)  drift    c11d  c2 2d  (7) – 365 – COMPUTER METHODS IN MATERIALS SCIENCE . 2010). Di the intrinsic diffusion coefficient and i and Mi are partial molar volume and molecular mass of the i-th component. the pressure evolution can be approximated from dilatation of ideal crystal: dp E   2    i i M i id  . The purpose of this paper is to use the bi-velocity method to calculate the Kirkendall trajectory (position of the Kirkendall plane). RT x i  1. Gusak. u. Tsi  ui  ich   i M i p and J id  ciid  ci Di  ch  i   i p  . Thus. the drift velocity can be rewritten in the following form:    M  i 1 i i i 2 d i   drift   0 . 2010) The core of the bi-velocity method is the mass balance equation: The pressure. 1889. when diffusion is not negligible can be calculated from partial GibbsDuhem relation.INFORMATYKA W TECHNOLOGII MATERIAŁÓW & Wierzba. The first two methods base on the drift velocity and its integral the last method assumes that the position of the Kirkendall plane is defined by the local maximas of the entropy distribution. ich is the chemical potential. The density i is related with concentration. i = Mici.. 2 . energy and entropy densities in multicomponent systems. THE BI-VELOCITY METHOD (DANIELEWSKI AT AL. The method can be used when the gradient of the mechano-chemical potential (from fairly well known thermodynamic properties) and diffusivities as a function of concentrations (from measured tracer diffusivities) are known. The Kirkendall velocity can be calculated from the drift velocity. The overall internal energy. 2004. respectively. the rotations of the system are neglected (rot υ = 0). 1) the "curve method" (Aloke. the entropy. 2) the "trajectory method" and 3) "entropy approximation".  drift .   dt 3 1  2v  x  i 1  (5) where E and v are the Young modulus and Poisson ratio. To calculate the drift flux the Volume Continuity Equation (VCE) is used. i = j i. generated during diffusion process is described by the Cauchy stress tensor. p. Planck. The differential form of VCE follow: 3. J d i  id   drift  ui d drift  ui i   i i     i ci p t x x (6) where ui denote the internal energy. p denote the pressure field acting on the components. and J drift i (1) denote the diffusion and drift flux. j. R and T are the gas constant and temperature. It allows to calculate the densities. Assuming the time independent external forcing (V ext t  0) . of the mixture can be defined:  drift     i i M iid . 1890):  u .e. 2. (3) The drift velocity. Assuming the diffusion process in the binary system and that the partial molar volumes are constant and equal. The method is limited to the axiatoric part of the stress tensor only. respectively. of the mixture can be calculated from components counterparts  u  i  1. Mainly. The "curve method". The diffusion flux. t x where i is the mass density. equation (4). i 1 i i 2 Finally. i 1 2 where id denote the diffusion velocity. 2 . drift velocity.

Substituting equac tion (2) the drift velocity in a binary system when pressure filed is neglected can be rewritten in the following form: where K is the positions of the Kirkendall plane at time t = t*. The position of the Kirkendall plane can also be calculated by following the marker during the diffusion process. c) trajectory and d) entropy bi-velocity method. equation (12). The data used to calculate the diffusion process are presented in table 1. – 366 – . N i  ci . "The entropy approximation". "The trajectory method".INFORMATYKA W TECHNOLOGII MATERIAŁÓW From Euler theorem the overall molar volume equal to the inverse of the overall concentration. 1. Figure 1 shows the comparison of presented above methods of calculating the position of Kirkendall plane. The position(s) of the Kirkendall plane(s) can be found at the point of intersection(s) between the drift velocity curve and the straight line calculated from equation (11). The comparisons of the position of Kirkendall plane (vertical line) by different calculation methods: b) velocity curve. the Kirkendall plane is a plane of initial contact moving at parabolic dependence. a) The concentration profile in binary A-B diffusion couple. COMPUTER METHODS IN MATERIALS SCIENCE Fig.e. i.  drift D  ch D  ch  N1 1 1  N 2 2 2 RT x RT x ch i (9) xK  t2     drift  xK  t1  . Thus:  K   drift  0. In this studies it is postulated that the local maxima's calculated on the entropy distribution curve shows the positions of Kirkendall planes (the most favored places) in the diffusion couples. t   dx xK  dt 2t (11)  drift  N   N  d 1 1 d 2 2  (8) where the molar ratio. t  dt t1 t2 (12) When the ideality sweeping statement is assumed.   1 c .   ln ci  RT thus the drift velocity x x has its final form:  drift   D1  D2  N1 x (10) In a diffusion controlled interaction. thus the position of the Kirkendall plane: In each time the new position of the Kirkendall plane is calculated.

A.. 2012 Accepted: : December 12.. t = 36000 s. Die elektromotorische Wirkamkeit der Ionen.. T = 1273K. 6717-6727. Algorytm obliczania trajektorii płaszczyzny Kirkendalla pozwala również na wyznaczenie: 1) stężeń składników. 4. Smigelskas. Interface migration and the Kirkendall effect in diffusion-driven phase transformations. Campbell.. 3347-3373.. 3352-3355. Soc. 4. J..02 cm A diffusion coefficient [cm2 s-1] NA·10-10 B diffusion coefficient [cm2 s-1] 10-10 REFERENCES Aloke. 58. F. 463. Figure 1 shows that the three presented methods give similar results when only one position of the Kirkendall plane is expected.. common for all components. decision number DEC-2011/03/B/ST8/ 05970. Thin Films. Guyer.. McFadden. Acknowledgments. Acta Metall. Mater. Wierzba. The data used in simulations of diffusion in binary A-B diffusion couple. (3) the stress field during the diffusion process and (4) entropy production. Chem. Gusak. P. and the drift velocity  drift . W. 171. This work has been supported by the National Science Centre (NCN) in Poland. 1890. Cserhati. M. Microstructural Stability of the Kirkendall Plane in Solid-State Diffusion. 1947.. 130-142. A. Rapp. A. 38. A. 1990. The software CADiff available from author. J. 2010. C. 129-140 (in German). 2001. 2012 Received in a revised form: November 21.. 17691779.. J. and Nanosystems. E. Phys. Pieraggi. The Kirkendall effect in solid state diffusion. 2010. Z. R.. E. Thermodynamically consistent bi-velocity mass transport phenomenology. H. 561-576 (in German). Eindhoven Boettinger. Kirkendall. The drift velocity. Danielewski. M. Wiley-Vch Verlag GmbH & Co. Metoda dwu-prędkości jest uogólnieniem metody Darkena. Pokazano. M. A. KRYTERIA EWOLUCJI PŁASZCZYZNY KIRKENDALLA PODCZAS PROCESU DYFUZJI Streszczenie W artykule zaprezentowana została metoda dwu-prędkości pozwalająca na wyznaczenie ewolucji płaszczyzny Kirkendalla podczas procesu dyfuzji. Proc. E. Zinc Diffusion in Alpha Brass. B. Acta Mat. J. thickness. R. 3) energii oraz 4) produkcji entropii w wieloskładnikowych fazach skondensowanych. d = 0.. (2) different partial molar volumes of components. W. Technische Universiteit Eindhoven. M.. J. Figure 1 shows a) the concentration profile and comparison of the calculation of Kirkendall plane where different methods were used: b) the velocity curve. Rev. The examples presented in this work show that the entropy curve can be used to approximate the position of the Kirkendall markers when the diffusion coefficients are composition dependant. Weinheim. C. The model was applied for the modeling of the trajectory of the Kirkendall planes in a binary diffusional couples. J. Diffusion-controlled Solid State Reactions In Alloys. Gusak. The method based on the postulate that each component’s velocity  i must be divided into two parts: id the unique diffusion velocity. B. J. Nernst. A. The method effectively deals with (1) composition dependent diffusivities. A. Opiera się ona na całkowych prawach zachowania masy. Annu Rev Phys Chem 40. Planck.D. 2) prędkości dryftu. Received: October 25. which depends on diffusion potential gradient and is independent of the choice of the reference frame. 2012 – 367 – COMPUTER METHODS IN MATERIALS SCIENCE ... CONCLUSIONS The bi-velocity method for interdiffusion allows quantitative and qualitative description of the mass transport process in binary systems. pędu oraz energii. Kodentsov. B. że metoda pozwala na poprawne wyznaczenie położenia płaszczyzny Kirkendalla podczas procesu dyfuzji w układach dwu-składnikowych. The presented simulation results were calculated using the bi-velocity method. allows to calculate the position and trajectory of the Kirkendall planes during diffusion process. van Dal. 2004. G. Computation of the Kirkendall velocity and displacement fields in a one-dimensional binary diffusion couple with a moving interface. Phys. 1889. 86. AIME. c) trajectory method and d) entropy bi-velocity method in a binary A-B diffusion couple. van Loo. M. van Loo F.. Lett. Ber die potentialdierenz zwischen zwei verdnnten lsungen binrer elektrolyte. 2007.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Table 1. Trans.

rwth-aachen. multi-phase field method. 2 MODELING OF STATIC RECRYSTALLIZATION KINETICS BY COUPLING CRYSTAL PLASTICITY FEM AND MULTIPHASE FIELD CALCULATIONS ONUR GÜVENC1. A simplified nucleation model at the grain level is used to initiate the recrystallization process. BERND BÖTTGER2. Intzestrasse 10. Intzestrasse 5. Key words: static recrystallization. considering the evolution of grain orientations and dislocation density. which occurs in interpass times. Calculation strategies based on Monte Carlo Potts (Raabe. 2008) and multi-phase ISSN 1641-8581 368 – 374 . MARKUS APEL2.. INTRODUCTION Microstructural changes play a major role in hot working processes. not only because the microstructure defines force requirements for forming through the flow stress but also since the microstructure defines final product properties. D-52056 Aachen. Static recrystallization is often simply modeled based on Johnson-Mehl-Avrami-Kolmogorov (JMAK) equations which are linked to the visco-plastic flow behavior of the material. whose material parameters have been calibrated using experimental flow curves of the considered 25MoCrS4 steel. they assume that the microstructure associated with a material point can be described by average values of grain size.. static recrystallization (SRX). influences the microstructure evolution. dislocation density or even strain. Operating on the macro-level. Under these assumptions. GERHARD HIRT1 1 Institute of Metal Forming. The grain microstructure is generated via a Voronoi tessellation. 2008). 13. misorientations and local accumulations of deformation. Germany 2 ACCESS e.COMPUTER METHODS IN MATERIALS SCIENCE Informatyka w Technologii Materia ł ów Publishing House AKAPIT Vol.. The deformation of the RVE. Such semi-empirical models are not able to predict the SRX grain microstructure. orientation) is transferred to the FDM grid used in the multiphase field approach by a dedicated interpolation scheme. Germany *Corresponding author: bambach@ibf. but the grain morphology predicted by the 2d model still differs from experimental findings. GOTTFRIED LASCHET2. it is possible to approximate the SRX kinetics obtained from the stress relaxation test. No. driving forces for static recrystallization are calculated based on the mean energy per grain and the curvature of the grain boundaries. RWTH Aachen University. crystal plasticity FEM. RWTH Aachen. the flow stress and the final product properties. Static recrystallization (SRX) is one of the most dominant mechanisms during inter-pass periods of hot rolling or forging processes and it is a common practice to model its kinetics using Johnson-Mehl-Avrami-Kolmogorov (JMAK) type equations. hot forming. vertex (Piekos et al. In this paper. They disregard effects of grain topology.V. THOMAS HENKE1. commonly used models such as those proposed by Sellars (1990) lack spatial resolution. 2013. 1999). In the phase field calculation. The microstructure is modeled by a representative volume element (RVE) of a homogeneous austenitic grain structure with periodic boundary conditions. D-52072 Aachen. is calculated using a crystal plasticity finite element (CP-FEM) formulation. cellular automata (Gawad et al.de Abstract In multi-step hot forming processes. an approach for the simulation of static recrystallization of austenitic grains is presented which is based on the coupling of a crystal plasticity method with a multiphase field approach. The deformed grain structure (dislocation density. MARKUS BAMBACH1*. However. periodic microstructure modeling 1.

Material The steel grade 25MoCrS4. 2009). In this paper.5. the phase field method offers a promising approach for modeling static recrystallization after plastic deformation due to its implicit definition of the grain boundaries as a diffusive interface. 1983).50 0. 2. A 2d microstructure is generated via a Voronoi algorithm. However. an Avrami exponent of n = 0.23 – 0.5 denotes the time required to reach 50% recrystallization..5     (1) in which n is the Avrami exponent and t0.40 – 0. material..29 0.1. avoiding the complexity of handling their topology one-by-one. – 369 – COMPUTER METHODS IN MATERIALS SCIENCE field methods (Takaki & Tomita. strain rate:  at T = 1100°C.2. its theoretical foundation on irreversible thermodynamics allows for the implementation of models based on the minimization of the free energy functional of the polycrystalline microstructure (Steinbach. predicted values are critically compared with the results of stress relaxation test. 2011). and at five different strain rates:  -1 10 100 s . 1 shows deformation response of the material under uniaxial compression at 1100°C. a case hardening steel for gearing applications for automotive and aerospace industry. crystal plasticity finite element (CP-FEM) simulations have gained momentum and have now reached a level of high predictive quality.90 0. After the predefined strain level was reached. In addition. In order to take advantage of the phase field approach in an SRX model effectively. For details of the sample preparation we refer to Xiong et al. (2011). Among those methods. Its chemical Grade C Mn Si Cr Mo composition is given in 25MoCrS4 0.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 2. In recent years. Chemical composition of 25MoCrS4 (1.1.3. 2.60 – 0. Fig. . Finally. Exact procedures of sample preparation and experimental methodology are described elsewhere (Henke et al. The possibility to implement grain-scale flow stress evolution models and to derive intra. the cross-head of the servo-hydraulic testing machine was kept at constant height and the force response of the specimen was measured over time. Compression tests The hardening response of the material was obtained through a set of compression tests at 1100°C  = 0. used to set up a CPFEM model with random grain orientations. respectively. the microstructural evolution of a commercial steel grade (25MoCrS4) during SRX after a hot uniaxial compression test is simulated by coupling CP-FEM calculations and phase field simulations.e.50 Table 1.and inter-grain crystalline interactions during deformation enables the generation of a representative deformed microstructure for a phase field simulation (Roters et al.01 s-1 the test case strain:  = 0. Light optical microscopy (LOM) was used before and after SRX in order to determine the grain size evolution and nucleation site preference. a representative initial (i. for all considered strain rates. For the given values of XRX(t) and t0. deformed) state of the microstructure is a necessary starting condition. Stress relaxation tests The SRX kinetics of the material were examined by stress relaxation tests.01. 0. Once the time evolution of XRX is known. 1.7326) according to DIN 17210 (Values are in wt. %).2. This simplifies the simulation of interface migration. and subjected to uniaxial compression.56 was determined by regression for  = 0.. was selected as application Table 1. JMAK kinetics of SRX can be evaluated by determining the unknown parameters of the modified Avrami equation: X RX n   t    1  exp   ln 2    .40 0. 2010) were proposed as an attempt to capture the heterogeneities at the micro level. EXPERIMENTAL ANALYSIS 2. The results of CP-FEM simulation are mapped onto the finite difference grid of the multi-phase field SRX simulation.   t0.40 – 0. 2010). LOM results show that the average grain sizes before and after the SRX are 36 µm and 7 µm. the results of intricate models with spatial resolution are often not compared to experimental results.15 – 0. The decrease of the reaction force and the respective stress values were then converted to recrystallized volume fractions (XRX) according to the procedure described by (Gronostajski et al. The compression test specimens (without lubrication pockets) were deformed to a pre-strain below the critical strain for dynamic recrystallization (DRX) at the different strain rates.

The parameter set can be calibrated with the compression test results. initial and saturation values of the slip resistance can be determined (0 = 8 MPa. which can be calculated using the equation h     h0 1  c  . Taylor factor M = 3). it is assumed that each grid point has the index and orientation value of its nearest neighboring node as shown in Fig. 1 shows the comparison of experimental and numerical responses of the material.1. 1. In addition. 3. 2012). MICROSTRUCTURE MODEL 3.  and c denote the stress state and the critical resolved shear stress on the active slip system. h0 . nucleation sites of the SRX are not found inside the grains.e. (2)  c   h   .. the transfer of data from a FEM mesh to a FDM grid requires a dedicated interpolation scheme. 2009). Ed   Gb 2 . Fig.  0 is any convenshear rate at the active slip system. 2. the well-known phenomenological deformation law by Hutchinson (1976) is used within the simulation software DAMASK (Roters et al.. If the material is assumed to be isotropic throughout the deformation process (i. 3.INFORMATYKA W TECHNOLOGII MATERIAŁÓW In addition. – 370 – . In order to transfer the grain index and the grain orientation data from nodes to the grid. but on the grain boundaries.2. a and  sat are material parameters (Kalidindi et al. 1992). a = 2).   sat  a (4) Three types of data have to be mapped from the CP-FEM output to the phase field simulation: The grain index.  c   0   Gb  . However. The law is defined by The macro-scale stress-strain curve can be converted to its micro-scale counterpart by the method proposed by Taylor (1938). m characterizes the strain rate sensitivity and finally h is the function defining the incremental value of c in terms of shear increments on a chosen slip system . the local orientations have to be averaged after the mapping to determine the mean grain orientations using circular statistics (Berens. (5) (6) COMPUTER METHODS IN MATERIALS SCIENCE Fig.  ient reference shear rate. Comparison of CPFEM and compression test results of 25MoCrS4 at 1100°C at various strain rates. sat = 16 MPa) and the other model parameters can be calculated numerically (h0 = 300 MPa. Coupling    0  c n 1 m sgn   . Note that each experiment has been repeated five times in order to take the experimental scatter into account. the mean grain orientation and the mean stored energy per grain. Deformation model In order to simulate the plastic hardening behavior. The stored energy of the deformation can be calculated from the flow stress increase using the equations In equation (4).  1 (3)  is the where  denotes the active slip system.

3.INFORMATYKA W TECHNOLOGII MATERIAŁÓW Ed      c 0 G 2 . Then. the local stored energies were converted to mean energies per grain using equation (7) with G = 32.54 and converted to the FDM grid. at the moment no dislocation After the deformation simulation.  is the interface thickness.2 was imposed in the stress relaxation experiments. In addition. c and 0 are the final and initial values of the shear resistance (Taylor. grain evolution can be modeled by minimization of the free energy functional – under the assumption of a double obstacle function – which leads to the popular formulation of multi-phase field method. The index and orientation of a node(circle) is assigned to all grid points (square) which are far away from the other nodes. i   mij[ ij (i 2 j   j  2i  j 2 (i   j ))  2 2      ij  i j Gij ]  (8) Fig. σij denotes the interfacial energy between adjacent grain boundaries. Recrystallization Fig. Therefore. 2006). the interpolated values at the gird points are obtained as the weighted sum of adjacent finite element mesh nodes. the nodal energies are mapped onto the grid points. 4. In equation (8).01 s-1) under hostrain rate (T = 1100oC and  mogenous boundary conditions was simulated with periodic digital microstructure generated by a planar Voronoi tessellation of 25 randomly oriented grains. Deformation Isothermal uniaxial compression at a constant  = 0. (6) and (7). heterogeneous system. Influence of each node is inversely proportional to its distance to an individual GP. 3. 1934). 3. Note that.1. In order to avoid the occurrence of DRX. the mean stored energy per grain is found by calculating their arithmetic mean on the grain area.2 GPa and α = 0. Recrystallization model In the equations (5). the nucleation of new grains was assumed to take place at interfaces and triple junctions with site saturation as initial condition. The weights are inversely proportional to the relative distance between grid point and finite element node. For an isothermal. Finally. expressed via equation (8) (Eiken et al. SIMULATION AND RESULTS 4. and the same pre-strain was used in the model. In the mapping.5 µm. G(T ) is the temperature dependent shear modulus. ij is related to the grain boundary curvature and ΔGij is the contribution of the stored energy (Ed). equation (8) models the combined effects of curvature and stored energy on the interface migration. In the SRX simulation  was taken to be 1.. 4. The energy Ed of each grid point (GP) is determined by interpolating that of the nodes. mij is the grain boundary mobility. a maximum strain of 0. Ed is the stored energy due to deformation. 4.  defines a proportionality constant. (7) density gradients inside the grains are taken into account. the surface – 371 – COMPUTER METHODS IN MATERIALS SCIENCE .2. 2. Evolution of slip resistance and misorientation are shown in Fig. as illustrated in Fig. 3.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW a) energy was set to 3e-7 Jcm-2 and the mobility was assumed to be 5e-3 cm4/Js. It was found that with the aforementioned parameter set. The evolution of the morphology of the deformed microstructure during the SRX was also predicted. – 372 – . Comparison of mean grain sizes calculated by LOM and CPFEM-PF after recrystallization Mean grain diameter / µm After recrystallization LOM 7 CPFEM-PF 8.5e-2 Jcm-3. JMAK model and the stress relaxation experiment. 4. When these values are chosen. the SRX kinetics and grain sizes of the phase field simulation are in good correlation with the experimental values as seen in Fig. 6. However. Table 2. 5 and Table 2. (a)  c and (b) misorientation at the end of deformation. Nucleation was restricted to occur at the grain boundary interfaces and at triple junctions and only at sites where the stored energy exceeds 2. the recrystallized grain front is directional which results in a cuboidal final microstructure as shown in Fig. COMPUTER METHODS IN MATERIALS SCIENCE Fig. XRX kinetics of CPFEM-PF-simulation.1 s (b) t = 1 s (c) t = 100 s Fig. 6. (a) t = 0. Progression of recrystallization through time: (a) Growth of nuclei on the interfaces.2 b) Fig. (b) their competitive growth towards nonrecrystallized grains and (c) the fully recrystallized microstructure. 5. the Avrami exponent calculated from the phase field model is observed to be larger than 1 in contrast to the exponent from the experiment.

1-21. W. Kuziak. Wierzbanowski. Overview of constitutive laws. Pulit. Eiken. J.g. 1999. P. Bacroix. DISCUSSION 5. I. Henke.. experiments. number of sites per interface) that is available.. Pietrzyk. Experimental Uncertainties affecting the Accuracy of Stress-Strain Equations by the Example of a Hensel-Spittel Approach. Ziemba. Madej. Acknowledgements. S. T. A MATLAB Toolbox for Circular Statistics. Hirt. 2009. J. Materials Science Forum. 1-9. 69-72. 2010. 348. Philosophical magazine A-Physics of Condensed Matter Structure Defects and Mechanical Properties. B. Eisenlohr. Mathematical and Physical Sciences. J. 348-352.e. 14th International ESAFORM Conference on Material Forming: ESAFORM 2011. Böttger.. Journal of the Mechanics and Physics of Solids.. K. homogenization and multiscale methods in crystal plasticity finite-element modeling: Theory.. 73.. REFERENCES Berens.1. Physical Review E. Bieler. 571-572. 2339‐2358. applications. 31.. F. a 3-D model with an improved nucleation model at the grain level is necessary to predict the SRX kinetics and the final grain shape more accurately. Material parameters are calibrated using the experimental results at the macro-scale. Anand.. J.... Metal Science Journal . D. The mapping algorithm averages out local gradients and should be improved in the future.. 231-236. 5. J. – 373 – COMPUTER METHODS IN MATERIALS SCIENCE The SRX kinetics and grain size calculated with the phase field method show a good correlation with that of the stress relaxation experiments... 1983. resulting in unrealistic cuboidal final grain geometry. 6. CONCLUSION In this work a coupled CP-FEM-phase field model based on the mean stored deformation energy and grain boundary curvature is applied to predict SRX kinetics. Gawad. the applied model comprises only physical quantities. Belfast.. due to the usage of the mean orientation and stored energy per grain in the phase field simulation.. Gronostajski. B.. Multiphase-field approach for multicomponent alloys with extrapolation scheme for numerical application.. Due to the low experimental value of the Avrami exponent (n < 1). Acta Materialia. This problem can be solved in the future by accounting for dislocation gradients and abnormal sub-grain growth in order to improve the simple nucleation model at the grain level. The morphological evolution of the microstructure is found to be directional. A. In spite of the good correlation between model and experiment. . The authors gratefully acknowledge the financial support of the Deutsche Forschungsgemeinschaft (DFG) for the support of the depicted research within the Cluster of Excellence “Integrative Production Technology for High Wage Countries”. 537‐569.. Multiscale model of dynamic recrystallization in hot rolling. 79. Hutchinson.. K.W. e. Bounds and Self-Consistent Estimates for Creep of Polycrystalline Materials. Moreover. Bambach. L. it was necessary in this model to fill all possible nucleation sites. surface energy and interfacial mobility which allow for physical interpretation. Software. This mean energy per grain serves as driving force in the recrystallization simulation with the multi-phase field approach. Even though the model is phenomenological.. Crystallographic texture evolution in bulk deformation processing of FCC metals. 1152-1211.. E.. T. Tjahjanto. being a spatially resolved method. 2008. 1976. Tarasiuk. 1992. G.INFORMATYKA W TECHNOLOGII MATERIAŁÓW 5. In contrast to conventional JMAK based statistical models. Proceedings of the Royal Society of London. CP-FEM and coupling The microstructural deformation of 25MoCrS4 is successfully modeled by CPFEM using a wellestablished phenomenological hardening law. An extension to 3D space would increase the number of possible nucleation sites per interface and make the predicted grain morphology more realistic. P.. 58. C.2. Use of stored energy distribution in stochastic vertex model.. Journal of Stat. Bronkhorst. the inter-/intra-grain scatter of the grain orientation and hardening are captured and mapped onto finite difference grid effectively. Piekos. 2008. the phase field method takes the heterogeneity of the stored energy and boundary curvature into account. Static recrystallization A dedicated mapping scheme was used to couple the multi-phase field model with a CPFEM deformation model whose deformation conditions correspond to experimental results. 1. M. 101‐127. R... kinematics. the scatter is unintentionally averaged during mapping which results in a loss of resolution. 17.. Raabe. 2006. D. Hantcherli. 2011. The nucleation mechanism of the 2-D model leads to a rather unrealistic grain shape when the model is adjusted to the experimentally obtained SRX kinetics. Raabe. M. 71-77. Kalidindi. International Journal of Material Forming. Recovery and recrystallization of Cu after hot deformation. However. Introduction of a scalable three-dimensional cellular automaton with a probabilistic switching rule for the discrete mesoscale simulation of recrystallization phenomena. Steinbach.. H. D. L. 40. The usage of a 2D model and the assumption of site saturation restrict the number of nucleation sites (i. Roters.

INFORMATYKA W TECHNOLOGII MATERIAŁÓW Roters. 2012 COMPUTER METHODS IN MATERIALS SCIENCE – 374 – . 307‐ 324. D. Przy tych założeniach możliwe było oszacowanie kinetyki SRX na podstawie badań relaksacji naprężeń. Taki pół-empiryczny model nie jest w stanie przewidzieć mikrostruktury ziaren dla SRX.. Taylor. 52. J. rekrystalizacja statyczna (ang.. Modelling Microstructural Development during Hot Rolling. orientacja) jest przekazywana do siatki różnic skończonych w modelu pola wielofazowego stosując metodę interpolacji. Tech. P. MODELOWANIE KINETYKI REKRYSTALIZACJI STATYCZNEJ POPRZEZ SPRZĘŻENIE PLASTYCZNOŚCI KRYSZTAŁÓW MES Z OBLICZENIAMI PÓL WIELOFAZOWYCH Streszczenie W wielostopniowych procesach obróbki plastycznej. Physical and Engineering Sciences.. Part I.6 in Consideration of Dynamic Recrystallization. Steel Research International. Eisenlohr. I. 1990. 17.. 2010. Met. Tjahjanto. Wietbrock. W. Mikrostruktura jest generowana za pomocą wieloboków Voronoi. 62. Kords. W niniejszym artykule przedstawiono podejście do symulacji statycznej rekrystalizacji austenitu wykorzystujące połączenie plastyczności kryształów z metodą pola wielofazowego.. Phase-field models in materials science. Parametry modelu materiału wyznaczono na podstawie doświadczalnych krzywych płynięcia dla stali 25MoCrS4. 2012 Accepted: November 3. które jest powiązane z lepkoplastycznym płynięciem materiału. Z drugiej strony przewidywana w modelu 2D morfologia ziaren wciąż odbiega od wyników doświadczalnych. Raabe. F. siły pędne dla statycznej rekrystalizacji są obliczane na podstawie średniej energii w ziarnie i krzywizny granic ziaren. Taylor. W celu zainicjowania rekrystalizacji stosowany jest uproszczony model zarodkowania na poziomie ziarna.M. G. DAMASK: the Düsseldorf Advanced MAterial Simulation Kit for studying crystal plasticity using an FE based or a spectral numerical solver. C.. Theoretical. Mikrostruktura jest modelowana za pomocą reprezentatywnych elementów objętości (ang: Representative Volume Element . A. 362-387.I.. Proceedings of the Royal Society A: Mathematical. International Journal of Mechanical Sciences. W obliczeniach pola faz. 1934. M. Saeed-Akbari.RVE) jednorodnej struktury ziaren austenitu z okresowymi warunkami brzegowymi.. 82. 6.SRX) występująca w czasach przerw między odkształceniami. D. Xiong. C. Statyczna rekrystalizacja jest często modelowana korzystając z równania Johnson-MehlAvrami-Kolmogorov (JMAK). M. Y... B. 320-328.. 1072-1081. Obliczenia odkształcenia RVE są prowadzone połączonymi metodami plastyczności kryształów i MES. Tomita. Static recrystallization simulations starting from predicted deformation microstructure by coupling multi-phase-field method and finite element method based on crystal plasticity. Plastic Strain in Metals. Takaki. 2012 Received in a revised form: October 29. G. T. Modeling the Flow Behavior of a HighManganese Steel Fe-Mn23-C0. 1-31. Sci.. Inst. 2009. Diehl. Odkształcona struktura ziaren (gęstość dyslokacji. 2011. Bambach.. Received: September 21. 145. static recrystallization . wpływa na rozwój mikrostruktury. Modelling and Simulation in Materials Science and Engineering. Mats.. 3-10. Procedia IUTAM. naprężenie uplastyczniające oraz właściwości gotowego produktu. Steinbach. 2012.I. z uwzględnieniem rozwoju orientacji ziaren oraz gęstości dyslokacji. 3.. Sellars. 127-136.. 1938. The Mechanism of Plastic Deformation of Crystals.