2011/10 Development of a multiscale approach for the characterization and modeling of heterogeneous materials : (Application to polymer nanocomposites) Majid BANIASSADI Université de Strasbourg-CNRS Institut de Mécanique des Fluides et des Solides UNIVERSITE DE STRASBOURG École Doctorale Mathématiques, Sciences de l'Information et de l'Ingénieur Institut de Mécanique des Fluides et des Solides THÈSE présentée pour obtenir le grade de: Docteur de l’Université de Strasbourg Discipline : Mécanique des matériaux Spécialité : Micromécanique par Majid BANIASSADI Development of a multiscale approach for the characterization and modeling of heterogeneous materials : Application to polymer nanocomposites Soutenue publiquement le 19 Décembre 2011 Membres du jury Directeur de thèse : Co-Directeur de thèse : Prof. Saïd AHZI, Université de Strasbourg Prof. René MULLER, Université de Strasbourg Rapporteur externe : Rapporteur externe : Prof. Moussa NAïT ABDELAZIZ, École Polytechnique de l'Université de Lille Prof. Sébastien MERCIER, Université Paul Verlaine-Metz Examinateur : Examinateur : Examinateur : Prof. Hamid GARMESTANI, Georgia Institute of Technology, Atlanta-USA Prof. Abdel-Mjid NOURREDDINE, Université de Strasbourg Dr. David RUCH, Centre de Recherche Public Henri Tudor, Luxembourg Invité : Invité : Invité : Prof. Yves REMOND, Université de Strasbourg Prof. Madjid FATHI, University of Siegen, Siegen-Germany Dr. Valérie TONIAZZO , Centre de Recherche Public Henri Tudor, Luxembourg Nom du Laboratoire: IMFS N° de l’Unité FRE 3240 This thesis is dedicated to my parents (Parvaneh Khadiv & Mahmoud Baniassadi) for their love, endless support and encouragement. ACKNOWLEDGMENT I would like to express my gratitude to all those who gave me the possibility to complete this thesis. I want to express my sincere gratitude to my advisor, Professor Said Ahzi, who throughout my doctoral studies has contributed with excellent scientific support and encouragement to commence and achieve this work. I have furthermore to thank my co-advisor, Prof. Muller, who supported me with excellent scientific help, particularly in the domain of polymers. I wish also to express my deepest gratitude to Prof. Garmestani for his valuable ideas and suggestions and fruitful discussions. His encouragements have been a major reason for me to start and advance this work. I am deeply indebted to Prof. Remond for being an inspiration for me and providing an all-out support during these years. I am also grateful to the Department of Advanced Materials and Structures from the Public Research Center - Henri Tudor for the excellent technical support and FNR-Luxembourg for the financial support, and also to my dear colleagues from AMS HT especially Dr. Ruch, director of AMS HT, Dr. Toniazzo, Dr. Laachachi, Dr. Addiego and Dr. Hassouna. Special thanks go to Prof. Fathi from the University of Siegen, Prof . Patlazhan from the University of Moscow and Prof. Gracio from the University of Aveiro for encouraging me to follow the academic research, and to Prof. Nait-Abdelaziz from the University of Lille and Prof. Mercier from the University of Metz for their hints and suggestions. I am heartily thankful to my greatest source of inspiration, my parents, who have always been there for me, understanding and unconditionally supportive of my endeavors as I pursued this goal. Special thanks are reserved for Mr. Ghazavizadeh and Mr. Mortazavi from IMFS, Mrs. Amani from Georgia Tech, Dr. Li from PNNL, and Mrs. Sheidaei from Michigan State University, Mr. Kaboli from the University of Strasbourg, Mr. Safdari from Virginia Tech, Mr. Wen, Mr. Barth, Mr. Nierenberger, Mr. Wang , Mrs. Lhadi, Dr. Joulaee, Mr. Essa and Dr. Mguil from IMFS, Mrs. Morais, Mrs. Vergnat, Dr. Angotti and Mr. Delgado-Rangel from AMS HT, Mr. Etesami and all my friends and colleagues from IMFS and AMS HT for helping me to follow my researches. Contents Résumé .......................................................................................................................................... 13 Abstract ......................................................................................................................................... 21 Introduction ................................................................................................................................... 25 References ............................................................................................................................. 32 Chapter I ...................................................................................................................................... 35 Literature Survey ....................................................................................................................... 37 I.1. Random heterogeneous material ..................................................................................... 39 I.2. Two-Point Probability Functions .................................................................................... 39 I.3. Two-Point Cluster Functions .......................................................................................... 41 I.5. Approximation of higher order correlation functions ..................................................... 44 I.6. Homogenization methods for effective properties .......................................................... 45 I.7. Assumption of Statistical Continuum Mechanics .......................................................... 46 I.8. Reconstruction................................................................................................................. 47 I. References .......................................................................................................................... 51 Chapter II .................................................................................................................................... 53 Using SAXS Approach to Calculate Two-Point Correlation Function..................................... 55 II.1. Introduction .................................................................................................................... 57 II.2. Correlation between SAXS data and two-point correlation functions........................... 58 II.3. Structural characterization ............................................................................................. 61 II.4. Conclusion ..................................................................................................................... 66 II. References ......................................................................................................................... 67 Chapter III................................................................................................................................... 69 New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials. 71 III.1. Introduction .................................................................................................................. 72 III. 2. Approximation of tree-point correlation functions ...................................................... 75 III. 3. Approximation of four-point correlation function....................................................... 80 III. 4. Approximation of N-point correlation function .......................................................... 85 III. 5. Results ......................................................................................................................... 86 III. 6. Conclusion ................................................................................................................... 95 ix III. References ....................................................................................................................... 96 Chapter IV ................................................................................................................................... 99 A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions .............................................................................................................. 101 IV. 1. Introduction ............................................................................................................... 103 IV. 2. Development of a Monte Carlo reconstruction methodology ................................... 106 IV. 3. Optimization of the statistical correlation functions ................................................. 117 IV. 4. Three-phase solid oxide fuel cell anode microstructure ............................................ 119 IV. 5. Reconstruction of multiphase heterogeneous materials ............................................ 120 IV. 6. Conclusion ................................................................................................................. 127 IV. References ..................................................................................................................... 128 Chapter V .................................................................................................................................. 131 Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast ................................................................................................................................... 133 V.1. Introduction................................................................................................................. 135 V.2. Computer generated model ......................................................................................... 137 V.3. Thermal conductivity .................................................................................................. 139 V.4. Mechanical model ....................................................................................................... 140 V.5. Experimental part........................................................................................................ 144 V.6. Results and discussion ................................................................................................. 146 V.7. Conclusion ................................................................................................................... 150 V. References....................................................................................................................... 151 Chapter VI ................................................................................................................................. 155 Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM ............................................................................. 157 VI.1. Introduction ................................................................................................................ 159 VI.2. Reconstruction of heterogeneous materials using two-point cluster function .......... 160 VI.3. Statistical characterization of microstructures ........................................................... 163 VI.4. FEM characterization of multiphase heterogeneous materials................................... 163 VI.5. Result and discussion ................................................................................................. 165 VI.6. Conclusion .................................................................................................................. 172 VI. References ..................................................................................................................... 173 x Conclusion and Future Work ...................................................................................................... 175 Appendix .................................................................................................................................... 179 Appendix A ............................................................................................................................. 181 Appendix B ............................................................................................................................. 184 xi Résumé 12 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Résumé 13 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé 14 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Les fonctions de corrélation à deux points sont une catégorie de descripteurs statistiques bien connus pour décrire théoriquement la morphologie et les relations entre morphologie et propriétés d’un matériau. Nous approfondissons dans ce travail les connaissances liées à l’application des fonctions de corrélation à deux points pour la reconstruction et l’homogénéisation de matériaux composites. Plus particulièrement, les fonctions de corrélation à deux points ont été déterminées à partir de données expérimentales provenant de différentes techniques comme la microscopie électronique à balayage (MEB) ou à transmission (TEM), la diffusion des rayons X aux petits angles (SAXS), et de la méthode de Monte-Carlo. Dans une première application, nous avons exploité des données SAXS provenant de la caractérisation d’un composite polymère à deux phases. Pour cela, une matrice polystyrène (PS) chargée de nanoparticules d’oxyde de zirconium (ZrO2) a été sélectionné. Par ailleurs, la morphologie de ce matériau a été observée par MEB au moyen du mode de détection transmission (STEM). L’évolution de l’intensité des rayons X diffusés I en fonction du vecteur d’onde h est représentée à la figure 1 dans le cas du PS chargé de 3 % et 5 % en poids de ZrO2. Les fonctions de corrélation à deux points pour les composites PS-ZrO2 (3 % et 5% de charges) sont montrées à la Figure 2. . Figure 1. Intensité des rayons X diffusés I en fonction du vecteur d’onde h dans le cas de la nanopoudre ZrO2 et des nanocomposites PS-ZrO2 (3 % et 5% de charges) (avec correction d’absorption et élimination du fond continu) 15 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Figure 2. Fonctions de corrélation à deux points des composites PS-ZrO2 (3 % et 5% de charges) Afin d’augmenter la précision de l’approche continuum statistique, des fonctions de corrélation de plus grand ordre doivent en principe être déterminées. Ainsi, une nouvelle méthodologie d’approximation a été développée pour obtenir des fonctions de corrélation à N-points dans le cas de microstructures hétérogènes de matériaux sans gradient fonctionnel (FGM). Des fonctions de probabilité conditionnelle ont été utilisées pour formuler l’approximation théorique proposée. Dans cette approximation, des fonctions de pondération ont été considérées pour connecter des sous-ensembles de fonctions de corrélation d’ordre N-1 et estimer la totalité des ensembles de fonctions de corrélation d’ordre N. Dans le cas de l’approximation des fonctions de corrélation d’ordre 3 et 4, de simples fonctions de pondération ont été utilisées. Les résultats provenant de cette nouvelle approximation, dans le cas des fonctions de corrélation à trois points, ont été comparés à la fonction de probabilité réelle déterminée à partir d’une microstructure tridimensionnelle à trois phases générée par ordinateur. Cette reconstruction tridimensionnelle a été obtenue à partir d’une microstructure bidimensionnelle (résultant d’images MEB) d’un matériau à trois phases. Cette comparaison a prouvé que notre nouvelle approximation est capable de décrire des fonctions statistiques de corrélation de plus grand ordre et ce, avec une grande précision. La comparaison ente les fonctions de corrélation à trois points simulées et approximés est montrée à la figure 3 dans le cas de phases (noire-noire-noire). 16 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Figure 3. Fonctions de corrélation à trois points pour un composite à trois phases. Les corrélation à trois points sont montées pour les phases (noire-noire-noire). Les fonctions à deux points provenant de différentes techniques ont été calculées et exploitées pour reconstruire la microstructure de systèmes hétérogènes. Une nouvelle méthodologie MonteCarlo a été développée comme moyen de reconstruction tridimensionnel (3D) de la microstructure de matériaux hétérogènes, sur la base de fonctions statistiques à deux points. L’aspect le plus pertinent de la méthodologie de reconstruction présentée est sa capacité de réalisée des reconstructions 3D à partir d’image MEB 2D pour un système à trois phases, extrapolable à un système à N phases. La reconstruction tridimensionnelle d’un système hétérogène a été exploitée pour prédire le seuil de percolation de matériaux hétérogènes. Des micrographies MEB d’une anode constituée de trois phases et utilisée dans les piles à combustible à oxyde solide (rouge : nickel, bleu : ZYS, noir : vides), et l’image de l’anode reconstruite selon trois directions sont respectivement montrées aux figures 4 et 5. 17 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Figure 4. Micrographie MEB de la microstructure d’une anode constituée de trois phases et utilisée dans les piles à combustible à oxyde solide (rouge : nickel, bleu : ZYS, noir : vides) (a) (b) Figure 5. a) Volume reconstruit d’une microstructure d’anode, b) Sections du volume selon l’épaisseur (rouge : nickel, bleu : ZYS, noir : vides). Enfin, la théorie continuum statistique a été utilisée pour prédire la conductivité thermique effective et le module élastique effectif d’un composite polymère. Pour cela, nous avons proposé l’utilisation de la théorie continuum statistique à fort contraste pour prédire les propriétés élastique et thermique effectives d’un nanocomposite. En particulier, des échantillons de nanocomposites isotropes contenant des monofeuillets d’argile orientés de manière aléatoire ont été générés et utilisés pour calculer les fonctions de corrélation statistique à partir de notre modèle. L’orientation, la forme et la distribution spatiale des nanoargiles ont été pris en compte à travers les fonctions statistiques de corrélation à deux et trois points. 18 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Ces fonctions de corrélation ont été exploitées pour calculer les propriétés thermiques et élastiques effectives du nanocomposite. Pour valider notre approche théorique, nous avons réalisé des mesures expérimentales de ces propriétés dans le cas de nanocomposites polyamide/nanoargile avec des concentration en nanoparticules d’argile de 1 %, 3 % et 5 %. Les résultats de la simulation ont montré que la rigidité effective de la matrice est significativement augmentée par l’ajout d’une faible quantité de feuillets d’argile exfoliés. La conductivité thermique effective et le module élastique effectif ont été comparés avec nos résultats théoriques. Une bonne corrélation entre expérience et simulation a été obtenue dans le cas de la conductivité thermique. L’effet de l’ajout de nanoargiles sur les propriétés thermiques et mécaniques effectives du nanocomposite polymère chargé d’argile a été étudié à l’aide des approches théoriques et expérimentales. Toutefois, dans ce travail de recherche, le module élastique prédit est supérieur au module élastique expérimental, ce qui peut être dû à la présence de morphologies intercalées pour des taux d’argile élevés et à l’anisotropie des propriétés des nanoargiles. Par rapport à la matrice vierge de polyamide, les résultats théoriques et expérimentaux montrent une augmentation de la conductivité thermique effective et du module élastique effectif du composite en fonction de la fraction volumique de nanoargile. L’évolution du module élastique simulé et expérimental avec la température est représentée à la figure 6 pour la matrice PA vierge et ses composites avec OMMT (1%, 3% et 5 %). La comparaison entre la conductivité thermique expérimentale et théorique du PA et de ses nanocomposites avec OMMT est quant à elle montrée à la Figure 7. Figure 6. Module élastique expérimental et théorique d’un composite à deux phases en fonction de la température T pour le PA vierge et ses composites avec OMMT (1%, 3% et 5 % en poids) 19 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Résumé Figure 7. Comparaison entre la conductivité thermique expérimentale et théorique du PA et de ses nanocomposites avec OMMT 20 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Abstract Abstract 21 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Abstract 22 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Abstract Microstructural two-point correlation functions are a well-known class of statistical descriptors that can be used to describe the morphology and the microstructure-properties relationship. A comprehensive study has been performed for the use of these correlation functions for the reconstruction and homogenization in nano-composite materials. Two-point correlation functions are measured from different techniques such as microscopy (SEM or TEM), small X-Ray scattering (SAXS) and Monte Carlo simulations. In our study, SAXS data is used to calculate Two-Point correlation function correlation for two phase polymer composite. The selected material is polystyrene (PS) filled with zirconium oxide nanoparticles (ZrO2). The nanocomposite morphology was first examined by scanning transmission electron microscopy (STEM) and SAXS. Higher order correlation functions must be calculated or measured to increase the precision of the statistical continuum approach. To achieve this aim, a new approximation methodology is utilized to obtain N-point correlation functions for multiphase heterogeneous materials. The twopoint functions measured by different techniques have been exploited to reconstruct the microstructure of heterogeneous media. A new Monte Carlo methodology is also developed as a mean for three-dimensional (3D) reconstruction of the microstructure of heterogeneous materials, based on two-point statistical functions. The salient feature of the presented reconstruction methodology is the ability to realize the 3D microstructure from its 2D SEM image for a three-phase medium extendable to n-phase media. Three dimensional reconstruction of heterogeneous media have been exploited to predict percolation of heterogamous materials. In this study, the reconstruction methodology is used to reconstruct 3D microstructures of a threephase anode structure in a solid oxide fuel cell (SOFC) from a 2D SEM micrograph. Finally, Statistical continuum theory is used to predict the effective thermal conductivity and elastic modulus of polymer composites. Two-point and three-point probability functions as statistical descriptor of inclusions have been exploited to solve strong contrast homogenization for effective thermal conductivity and elastic modulus properties of nanoclay based polymer composites and computer generated microstructure. To validate our modeling approach, we conducted several experimental measurements and FEM calculation. 23 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction 24 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction Introduction 25 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction 26 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction Development of advanced microstructure reconstruction methodologies is essential to access a variety of analytical information associated with complexities in the microstructure of multiphase materials. Several experimental and theoretical techniques such as X-ray computed tomography (CT), scanning and computer generated micrographs have been used to obtain a sequence of two-dimensional (2D) images that can be further reconstructed in a 3D space. However, due to cost of sample preparation processes, simulation methods are often more applicable in reconstruction of heterogeneous microstructures in different areas [1-7]. Using lower-order statistical correlation functions, Torquato [8] established the reconstruction of one- and two-dimensional microstructures with short-range order using stochastic optimization. However, he later showed that the lower-order correlation functions cannot solely represent a two-phase heterogeneous material and therefore more than one solution may exist for a specific low-order correlation function [8]. Sheehan and Torquato [9] later introduced more orientations in the correlation functions to effectively eliminate the effect of artificial anisotropy. In the case of multi-phase materials, Kröner [10] and Beran [11] have developed statistical mathematical formulations to link correlation functions to properties in multiphase materials. Using higherorder correlation functions, one can account for the contribution of shape and geometry effects [8]. Torquato [12] also developed a new hybrid stochastic reconstruction technique for reconstruction of three-dimensional (3D) random media by using the information from the lineal path function and the two-point correlation functions during the nucleation annealing process. Different optimization techniques such as simulated annealing and maximum entropy have been applied in order to improve the reconstruction procedure [13]. In addition to 3D reconstruction processes based on probability functions, these functions can be used to account for more details of microstructure heterogeneities and for the relationships between microstructure, local and effective properties of multi-phase materials. The effective properties can be obtained via perturbation expansions [14, 15]. One general approach for the prediction of the effective properties of a two-phase material with properties of each phase near the average ones is called “weak-contrast” expansion. However, in materials with a high degree of contrast between the properties of their phases, “strong-contrast” theory is applied. Brown [16] suggested an expansion for effective dielectric property of two-phase heterogeneous materials. This 27 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction expansion for perturbation homogenization was modified and extended for elasticity by Torquato [17] for two-phase materials and later the solution was extended to multi-phase materials by others [15, 18]. Several numerical methods can be used to obtain the effective thermal/electrical conductivity as well as effective elastic properties of multiphase composites of complex geometries containing arbitrary oriented inhomogeneities [19-21]. In this thesis, statistical correlation functions have been exploited to reconstruct microstructurs and to develop a multiscale homogenization approach. Two-point correlation functions are the lowest order of the correlation functions that can describe the morphology and the microstructure properties relationships. Two-point correlation functions can be measured using SAXS data or SEM/TEM images for different microstructures. Monte Carlo simulation is a numerical technique that is capable of predicting two-point or higher order correlation functions. Higher order correlation functions can be approximated using lower order of correlation functions. In this study, a new approximation has been developed to predict the higher order correlation functions based on the lower order ones which efficiently facilitate the characterization of the effective properties. In this research work, a new Monte Carlo methodology is developed and implemented as a mean for three-dimensional (3D) reconstruction of multi-phase microstructures, based on two-point statistical functions. Finally, Statistical continuum theory of strong contrast has been exploited to predict effective thermal and elastic properties of two phase heterogeneous materials using two-point and threepoint correlation functions. To validate our modeling approach, we also conducted experimental measurements and FEM simulations. The details of each of the 6 chapters are provided in the following. we should note that these chapters are reproduced from our published paper in international journals. Chapter 1 consist of literature survey where we briefly present what is statistical descriptor of heterogeneous materials and then we consider Monte Carlo simulation to predict the statistical correlation function of heterogeneous materials. We also briefly present homogenization methods for the effective properties. At the end of the chapter, we give the definitions for reconstruction of heterogeneous materials and we explain the annealing reconstruction technique. 28 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction In chapter 2, capability of the statistical continuum approach is directly linked to statistical information of microstructure. Two-point correlation functions are the lowest order of correlation functions that can describe the morphology and the microstructure-properties relationship. In this chapter, SAXS data is used to calculate two-point correlation function correlation for two phase polymer composite. The selected material is polystyrene (PS) filled with zirconium oxide nanoparticles (ZrO2). In chapter 3, higher order correlation functions must be calculated or measured to increase the precision of the statistical continuum approach. To achieve this aim a new approximation methodology is utilized to obtain N-point correlation functions for non-FGM (functional graded materials) heterogeneous microstructures. Conditional probability functions are used to formulate the proposed theoretical approximation. In this approximation, weight functions are used to connect subsets of (N-1)-Point correlation functions to estimate the full set of N-Point correlation function. For the approximation of three and four point correlation functions, simple weight functions have been introduced. The results from this new approximation, for three-point probability functions, are compared to the real probability functions calculated from a computer generated three-phase reconstructed microstructure in three-dimensional space. This threedimensional reconstruction was based on an experimental two-dimensional microstructure (SEM image) of a three-phase material. This comparison proves that our new comprehensive approximation is capable of describing higher order statistical correlation functions with the needed accuracy. In chapter 4, a new Monte Carlo methodology is developed as a mean for three-dimensional (3D) reconstruction of the microstructure, based on two-point statistical functions. The salient feature of the presented reconstruction methodology is the ability to realize the 3D microstructure from its 2D SEM image for a three-phase medium extendable to n-phase media. In the realization procedure, different phases of the heterogeneous medium are represented by different cells which are allowed to grow. The growth of cells, however, are controlled via several optimization parameters related to rotation, shrinkage, translation, distribution and growth rates of the cells. Indeed, the proposed realization algorithm can be categorized as a member of dynamic programming methods and is designed so comprehensive that can realize any desired microstructure. 29 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction To be more specific, at first the initial 2D image is successfully reconstructed and then the final optimization parameters are used as the initial values for the initiation of the 3D reconstruction algorithm. This work presents a novel hybrid stochastic methodology based on the colony and kinetic algorithm for the simulation of the virtual microstructure. The simulation procedure involves repeated realizations where each realization in turn consists of nucleation and growth of cells. For each of the subsequent realizations, the controlling parameters get updated by minimization of an objective function (OF) at the end of the preceding realization. Here, the OF is defined based on the two-point correlation functions from the simulated and real microstructures. The kinetic growth algorithm is established on the cellular automata approach which facilitates the simulation procedure. Comparison of the two-point correlation functions from different sections of the final 3D reconstructed microstructure with the initial real microstructure shows a satisfactory agreement which confirms the proposed methodology. In chapter 5, we propose the use of strong contrast statistical continuum theory to predict the effective elastic and thermal properties of nanocomposites. Three-dimensional isotropic nanocomposite samples with randomly oriented monolayer nanoclay s are computer generated and used to calculate the statistical correlation functions of the realized model. The nanoclay orientation, shape and spatial distribution are taken into account through two-point and threepoint probability functions. These correlation functions have been exploited to calculate effective thermal and elastic properties of the nanocomposite. To validate our modeling approach, we conducted experimental measurements of these properties for Nanoclay/Polyamide nanocomposites with concentrations of 1, 3 and 5 wt. % of nanoclay particles. The simulation results have shown that effective stiffness can be increased significantly with small amounts of particle concentration for the exfoliated clay monolayers. The predicted effective conductivity and elastic modulus have been compared to our experimental results. Effective thermal conductivity shows satisfactory agreement with experimental data. The effects of nanoclay additives on the effective mechanical and thermal properties of nanoclay based polymer composites have been investigated using experimental and simulation analyses. In this research however, the predicted results for elastic modulus overestimate the experimental data, which might be due to the increasing intercalated structure for high concentration of nanofiller and to anisotropic properties of nanoclay. Relative to the 30 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction pure polyamide matrix, both the modeling and the experiments show an increase of the effective thermal conductivity and effective elastic modulus of the composite as a function of the nanoclay volume fraction. In chapter 6, the previously developed reconstruction methodology (in chapter 4) is extended to three-dimensional reconstruction of a three-phase microstructure, based on two-point correlation functions and two-point cluster functions. The reconstruction process has been implemented based on hybrid stochastic methodology for simulating the virtual microstructure. While different phases of the heterogeneous medium are represented by different cells, growth of these cells is controlled by optimizing parameters such as rotation, shrinkage, translation, distribution and growth rates of the cells. Based on the reconstructed microstructure, finite element method (FEM) was used to compute the effective elastic modulus and effective thermal conductivity. In addition, the statistical approach based on two-point correlation functions and our proposed approximation of three point correlation functions (Derived in chapter 3 ) was also used to directly estimate the effective properties of the generated microstructures. Good agreement between the predicted results from FEM analysis and statistical methods was found which confirms the efficiency of the statistical methods for the prediction of thermo-mechanical properties of three-phase composites. Our results from statistical approach were also compared to the case of the previous(existing) three-point correlation approximation [22]. This comparison shows that our new approximation yields better results. Finally, to conclude this thesis, general conclusions and remarks are reported. Some perspectives and suggestions for the continuity for this research work are exposed. 31 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction References [1] Bochenek B, Pyrz R. Reconstruction of random microstructures--a stochastic optimization problem. Computational Materials Science. 2004;31(1-2):93-112. [2] Chung S-Y, Han T-S. Reconstruction of random two-phase polycrystalline solids using low-order probability functions and evaluation of mechanical behavior. Computational Materials Science. 2010;49(4):705-719. [3] Liang ZR, Fernandes CP, Magnani FS, Philippi PC. A reconstruction technique for three-dimensional porous media using image analysis and Fourier transforms. Journal of Petroleum Science and Engineering. 1998;21(3-4):273-283. [4] Pierret A, Capowiez Y, Belzunces L, Moran CJ. 3D reconstruction and quantification of macropores using X-ray computed tomography and image analysis. Geoderma. 2002;106(3-4):247-271. [5] Sundararaghavan V, Zabaras N. Classification and reconstruction of three-dimensional microstructures using support vector machines. Computational Materials Science. 2005;32(2):223-239. [6] Talukdar MS, Torsaeter O. Reconstruction of chalk pore networks from 2D backscatter electron micrographs using a simulated annealing technique. Journal of Petroleum Science and Engineering. 2002;33(4):265-282. [7] Tariel V, Jeulin D, Fanget A, Contesse G. 3D multiscale segmentation of granular materials. Image Anal Stereol. 2011;27(1):23-28. [8] Torquato S. Random heterogeneous materials : microstructure and macroscopic properties. New York: Springer; 2002. [9] Sheehan N, Torquato S. Generating microstructures with specified correlation functions. Journal of Applied Physics. 2001;89(1):53-60. [10] Kroner E. Bounds for effective elastic moduli of disordered materials,. J Mech Phys Solids 1977;25:137-155. [11] Beran MJ. Statistical continuum theories. New York: Interscience Publishers; 1968. [12] Yeong CLY, Torquato S. Reconstructing random media. PHYSICAL REVIEW E. 1998;57(1):495506. [13] Manwart C, Hilfer R. Reconstruction of random media using Monte Carlo methods. PHYSICAL REVIEW E. 1999;59(5):5596. [14] Fullwood DT, Adams BL, Kalidindi SR. A strong contrast homogenization formulation for multiphase anisotropic materials. Journal of the Mechanics and Physics of Solids. 2008;56(6):2287-2297. [15] Tewari A, Gokhale AM, Spowart JE, Miracle DB. Quantitative characterization of spatial clustering in three-dimensional microstructures using two-point correlation functions. Acta Materialia. 2004;52(2):307-319. [16] Brown JWF. Solid Mixture Permittivities. The Journal of Chemical Physics. 1955;23(8):1514-1517. [17] Torquato S. Effective stiffness tensor of composite media--I. Exact series expansions. Journal of the Mechanics and Physics of Solids. 1997;45(9):1421-1448. 32 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Introduction [18] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. Effective conductivity in isotropic heterogeneous media using a strong-contrast statistical continuum theory. Journal of the Mechanics and Physics of Solids. 2009;57(1):76-86. [19] Giraud A, Gruescu C, Do DP, Homand F, Kondo D. Effective thermal conductivity of transversely isotropic media with arbitrary oriented ellipsoïdal inhomogeneities. International Journal of Solids and Structures. 2007;44(9):2627-2647. [20] Spanos PD, Kontsos A. A multiscale Monte Carlo finite element method for determining mechanical properties of polymer nanocomposites. Probabilistic Engineering Mechanics. 2008;23(4):456-470. [21] Wang M, Pan N. Elastic property of multiphase composites with random microstructures. Journal of Computational Physics. 2009;228(16):5978-5988. [22] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. A new approximation for the three-point probability function. International Journal of Solids and Structures. 2009;46(21):3782-3787. 33 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey 34 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey Chapter I 35 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey 36 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey Literature Survey 37 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey 38 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey I.1. Random heterogeneous material A random heterogeneous material is a class of materials which is composed of different materials or states, such as a composite and a polycrystals. “Microscopic” length scale is much larger than the molecular scale but much smaller than the characteristic length of the macroscopic sample .The heterogeneous material can be supposed as a continuum on the microscopic scale, and therefore its effective properties can be defined [1]. Statistical methods, using correlation functions, are one of the most practical and powerful approaches to estimate properties of heterogeneous materials [1]. Properties of materials can be approximated by using different order of statistical correlation functions [1-3]. In multiphase materials, the first order correlation functions represent volume fractions of different phases and do not describe any information about the distribution and morphology of phases [1]. If M-number of random points are inserted within a given microstructure and the number of points in phase-i is counted as Mi, the one-point probability function ( P1i ) is defined as the volume fraction through the following relation, as M (the total number) is increased to infinity P1i Mi M vi (1) M of where Vi is the volume of phase i (Φi), Vtotal is the total volume and vi is the volume fraction of phase i. Clearly, for two phases microstructure: 2 ¦Vi i 1 2 Vtotal and ¦v i 1 (2) i 1 I.2. Two-Point Probability Functions Now assign a vector ݎԦstarting at each of the random points in a heterogeneous microstructure. Depending on whether the beginning and the end of these vectors fall within phase-1 or phase-2, § · § · § · § · there will be four different probabilities ( P212 ¨ r ¸ , P2 21 ¨ r ¸ , P211 ¨ r ¸ and P2 22 ¨ r ¸ ) defined as [1]: © ¹ © ¹ © ¹ © ¹ 39 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey ­ ®r M M of ¯ § · P2ij ¨ r ¸ © ¹ M ij § · § ·½ r 2 r1 , ¨ r1 Mi ¸ ¨ r 2 M j ¸ ¾ © ¹ © ¹¿ (3) where, Mij are the number of vectors with the beginning in phase-i ( I i ) and the end in phase-j ( I j ). Eq. (3) defines a joint probability distribution function for the occurrence of events constructed by two points (ݎԦଵ andݎԦଶ ) as the beginning and end of a vector ݎԦ when it is randomly inserted in a microstructure. The two-point function can be defined based on two other probability functions such that [1]: ­§ ·§ ·½ § · P Probability b bili ®¨ r1 Mi ¸ ¨ r 2 M j ¸ ¾ P Probability b bili b l ¨ r 2 M ¸ ¹© ¹¿ © ¹j ¯© § · P2ij ¨ r ¸ © ¹ (4) The first term on the right hand side is a conditional probability function. At very large distances, rÆf, the probability of occurrence of the beginning point does not affect the end point and the two points become uncorrelated or statistically independent and the conditional probability function reduces to a one-point correlation function: ­§ ·§ ·§ ·½ § · Probability ®¨ r o f ¸ ¨ r1 Mi ¸ ¨ r 2 M j ¸ ¾ P Probability b b bili l ¨ r1 Mi ¸ (5) ¹© ¹¿ © ¹ ¹© ¯© The two-point function will then reduce to [1]: § · P2ij ¨ r , r o f ¸ © ¹ § · Probability ¨ r1 Mi ¸ Probability(r 2 M j ) © ¹ (6) or, lim r of P2ij r Q iQ j (7) For the case of a two-point function in a two phase composite, we have symmetry for non FGM microstructure [1]: P2ij r P2ji r (8) 40 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey For a three-phase composite, the indices (i, j) in the probability functions representation extend to three and as a result we have nine probabilities ( P211 , P222 , P233 , P212 , P221 , P213 , P231 , P223 , P232 ). Due to normality conditions the following equations are satisfied: ¦ ¦ P r ij 2 1 (9) i 1,3 j 1,3 ¦ P r ij 2 vi (10) j 1,3 ¦ P r ij 2 vj (11) i 1,3 Satisfying all three conditions for a three-phase composite ( i , j{1,2,3}) and knowing that the probability functions are symmetric ( P2ij = P2ji ) results in the important conclusion that only three of the nine probabilities are independent variables. For instance, we can choose P211 or (P11), P212 or (P12), and P222 or (P22) as the three probability parameters. I.3. Two-Point Cluster Functions Two-point cluster function is the other microstructure descriptor of heterogeneous materials which can reflect more precise information for heterogeneous materials [4].The two-point cluster function (TPCCF) P2C ii (r ) is the probability of finding both points (starting and ending point of vector ( r )) in the same cluster of one of the phase (i). This quantity is a useful signature of the microstructure as it reflects clustering information. Incorporation of such information in addition to the lower-order two-point cluster functions have led to the formulation of rigorous bounds on transport and mechanical properties of two-phase media [1, 4]. I.4. Monte Carlo simulation of Correlation functions The one-point probability function of the phase p is defined by the probability of occurrence of random points in this phase [1]. Therefore, one-point correlation function for each phase indicates the volume fraction of this particular phase. Convergence to the real volume fraction by the soft core algorithm (allowing for penetrable inclusions) is one of the advantages of using Eq.1 for randomly distributed penetrable inclusions. 41 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey Two-point correlation functions are determined based on the probability of occurrence of the head and tail of each vector in a particular phase. For example for the nanoclay polymer composites, there exist exactly two states, phase-1 (polymer matrix) and phase-2 (nanoclay particles). Therefore, four different configurations of Two-point correlation functions are obtained. These should satisfy normality conditions which results in the important conclusion that only one of the four functions is independent (See Fig. 2) . the Monte Carlo estimation of Two-point correlation function are acquired by assigning large number of random vectors within the generated microstructure and examining the number fraction of the sets (of vectors) which satisfy the different types of correlation functions . Fig. 2. Two-point correlation functions for three composites with 3 wt% of nanoclay. Three-point correlation functions for phase P can be interpreted as the probability that three points at positions x1, x2, x3 are found in phase P. The vectors x2-x1, x3-x1 and x3-x2 are invariant by translation and just depend on the relative positions of the points [1]. Thus, the three-point correlation functions can also be interpreted as the probability of finding three points in a certain 42 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey triangular configuration as shown in Fig. 3, This interpretation can be generalized for N-point correlation functions [1]. Fig. 3.Vectors for Three-point correlation function Statistical homogenization techniques are limited by the use of explicit equations for calculating governing multiple integral solutions. Therefore, the direct Monte Carlo approach cannot be used to achieve a fast algorithm to estimate the effective properties of heterogeneous materials. Generally, N-point correlation functions are defined as probability of occurrence of N-points which are invariant relative to a fixed position in desired phases. The expression of these functions for a given phase q can be written as [1]: q,...,q Pn (x1, x 2 ,..., x n ) Pr obability(x1 Phase(q) x 2 Phase(q) ... x n Phase(q)) (12) Where, xi is the vector position of the points in the microstructure. 43 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey I.5. Approximation of higher order correlation functions More detailed morphological description of heterogeneous materials is obtained by using higher order correlation functions. Measuring higher order correlation functions is difficult because of the increase of the number of independent variables to define correlation functions. For instance, the approximation of three-point correlation functions using two-point correlation functions is one of the best possible approach for calculating three point correlation functions. Several simple analytical approximations were reported for three point correlation functions. Adams [5] proposed an approximation of three point correlation functions using two-point probability functions: 1 ii 1 ii iii P3 x1 , x2 , x3 # P2 x1 , x3 P2 x1 , x2 2 2 (13) Garmestani et al. [6] also proposed another approximation for three point correlation functions: § x1 x3 iii P3 x1 , x2 , x3 # ¨ ¨ x1 x3 x2 x3 © · § xx 2 3 ¸ P ii x , x ¨ 2 2 3 ¸ ¨ x1 x3 x2 x3 ¹ © · ¸ P ii x , x (14) ¸ 2 1 3 ¹ These two approximations do not satisfy all normalization relations. Mikdam et al. [7]. proposed a new approximation for two phase materials that satisfies the normalization relations. §§ x1 x3 P3 x1 , x2 , x3 # ¨ ¨ ¨¨ x x x x 2 3 ©© 1 3 iii · § x2 x3 ¸ P ii x , x ¨ ¸ 2 2 3 ¨ x1 x3 x2 x3 ¹ © · P ii x , x · ¸ P ii x , x ¸ 2 2 3 ¸ 2 1 3 ¸ P2ii 0 ¹ ¹ (15) 44 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey Fig. 4. Schematic representation of vectors for approximation of three point correlation functions I.6. Homogenization methods for effective properties The effective property Ke is defined by a relationship between an average of a generalized local flux F and an average of a generalized local intensity G [1]: F v Ke .G (16) Table 1 summarizes the average local flux F and the average local intensity G for some physical linear problems like conductivity, magnetic permeability, elastic moduli, viscosity and fluid permeability. Table 1 F, G and Ke for different physical problems [1] General effective property Average generalized flux Average generalized intensity G Ke F Thermal conductivity Heat flux Temperature gradient Electrical conductivity Electric current Electric field Magnetic permeability Magnetic induction Magnetic field To estimate the bulk properties of such heterogeneous materials, multiscale homogenization approaches are utilized. The multiscale homogenization techniques might be well categorized 45 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey into the following six classes: statistical methods such as strong-contrast [2, 3], inclusion-based methods such as self-consistent or Mori-Tanaka [8], numerical methods such as finite element analysis and asymptotic methods [9], variational/energy based methods such as HashinShtrikman bounds [10], and empirical/semi-empirical methods such as Halpin-Tsai and classical upper and lower bounds (Voigt–Reuss) [11]. Here, we specifically turn our attention to the statistical continuum mechanics of strong-contrast which, although difficult to implement, is applicable to any form of micro-structural inhomogeneity and relies heavily on the statistical information of the microstructure reflected in the correlation functions. In other words, to predict the effective properties of heterogeneous media with a high degree of contrast between the properties of phases and indistinguishable morphology of phases, strong-contrast approach is highly suitable [1]. As pointed out earlier, one of the well-known applications of n-point correlation functions can be found in properties characterization. For this, exact perturbation expansions are used to predict the effective stiffness/thermal properties of a macroscopically isotropic two phase composite media. Manipulating integral equations for the local “cavity” strain field and polarization leads to finding series’ expansions for the effective stiffness tensor or thermal tensor [1]. Unlike the classical homogenization methods the statistical approach accounts not only for the interactions between the phases but also for the distribution of the phases [1]. I.7. Assumption of Statistical Continuum Mechanics Statistical information of the microstructure can be used to predict the effective properties. There are some assumption for the samples and the domains as follows: A. All the random variables of the heterogeneous media such as stress, strain, stiffness,... have to obey the ergodic hypothes therefore the ensemble average of each variable can be defined as follows [1]: c c( x) 1 V ³ V c( x)dV ¦ c( x) (17) B. Distribution of the considered property over the particles of the media is assumed statistically homogenous. This assumption doesn’t prevent using the heterogeneous microstructures. 46 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey Since the microstructure can be heterogeneous in each section however to calculate the overall elastic properties the microstructure is assumed to be statistically homogenous. C. The considered bodies which are infinite in extent are assumed to be in equilibrium condition at each point. I.8. Reconstruction Experimental and numerical Reconstruction of heterogeneous materials to get an accurate structure can be used to characterize and optimized heterogeneous materials. there are different experimental techniques such as x-ray tomography or focused ion beam/scanning electron microscopy (FIB/SEM) which are used to reconstruct three dimensional microstructures. For numerical reconstruction, statistical information are extracted from the microstructure of the considered heterogeneous material and can be used to reconstruct three dimensional microstructures [1, 12-17]. I.8.1 X-Ray Computed Tomography X-Ray Computed Tomography is a non-destructive technique that can be utilized to reconstruct micro-heterogeneous materials such as metal matrix composites. In this technique, X-ray beams hits a rotating sample and two-dimensional projections are recorded using a detector in the other side of the sample (see Fig. 5) [15, 17]. Fig. 5. Principle of x-ray tomography [17] 47 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey In classical tomography (attenuation tomography), three dimensional reconstruction is performed by combining the two dimensional projections. This technique has some limitation such as [15]: x Resolution limited to about 1000-2000x the object cross-section diameter; x Blurring of material boundaries; x Weak attenuation contrasts for imaging ; x Complicated data acquisition and interpretation due to the image artifacts (beam hardening); x Large data volumes and difficulty of visualization and analysis However, this technique has several strengths such as [15]: x Non-destructive 3D imaging x Easy sample preparation required x Extraction of sub-voxel level details. I.8.2 FIB/SEM FEI's DualBeam™ (FIB/SEM) systems are used for 3D microscopy and reconstruction of micro-and-nano-composites. For this purpose, dual-beam FIB/SEM is utilized to obtain microscopic two-dimensional (2D) SEM images in x–y plane by sectioning the specimen from the surface in the vertical direction along z axis (see Fig. 6). Using Auto Slice and View software (FEI Co.) serial-sectioning, SEM slices are stitched together to perform reconstruction. The dualbeam FIB/SEM is composed of ion beam which allows milling of the surface while the imaging is conducted by the electron gun [12]. 48 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey Fig. 6. Principle of FIB/SEM [16] I.8.3 Reconstruction using statistical descriptor (computer realization) The reconstruction of random media using limited microstructural information (correlation functions) is one the intriguing inverse problem in engineering. Various reconstruction techniques have been developed to generate realizations with lower-order correlation functions [13, 14]. In what follows, we briefly explain one of the most popular reconstruction approaches which was developed using annealing optimization technique [1, 18]. Using a set of correlation functions, partial information of heterogeneous media can be provided. This information can be used to reconstruct and characterize random media. Generally, in a reconstruction procedure, we would like to generate a microstructure with specified set of two-point correlation functions. Numerical reconstruction of heterogeneous media can be utilized to solve an optimization problem for a random generated microstructure. Monte Carlo reconstruction, using annealing technique is an optimization technique that can be used to reconstruct heterogeneous materials [13, 14, 18]. In this method, at the first step, a random image are generated with the same volume fraction of target sample then annealing optimization technique is used to move pixel of each phase for minimizing error between correlation function of realized model and sample. An initial random configuration is generated until the one point function is similar to the target sample. Then, an initial “temperature” is selected considering periodic boundary conditions and a correlation function is calculated for this configuration. The result are then been compared to the 49 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey original target correlation function. Two pixels with different phases are chosen at random then swapped; ensuring the volume fraction of each phase is preserved. Then, the same correlation functions are calculated and the Mean Square Error (Error) is compared to the corresponding correlation functions. In this method, the Metropolis algorithm is chosen as the acceptance criterion for the pixel interchange and P is the acceptance probability for the pixel interchange as follows: ܲሺݎݎݎܧௗ ՜ ݎݎݎܧ௪ ሻ ൌ ቐ ͳǡ ݁ οಶೝೝೝ ି οா ் οா ் ൏Ͳ (18) Ͳ Where ΔError=Errornew−Errorold and function of T will be defined base on step of annealing solution. This process is repeated until the convergence to the target correlation functions. 50 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey I. References [1] Torquato S. Random heterogeneous materials : microstructure and macroscopic properties. New York ; London: Springer; 2002. [2] Torquato S. Effective stiffness tensor of composite media--I. Exact series expansions. Journal of the Mechanics and Physics of Solids. 1997;45(9):1421-1448. [3] Pham DC, Torquato S. Strong-contrast expansions and approximations for the effective conductivity of isotropic multiphase composites. Journal of Applied Physics. 2003;94(10):65916602. [4] Jiao Y, Stillinger FH, Torquato S. A superior descriptor of random textures and its predictive capacity. Proceedings of the National Academy of Sciences. 2009;106(42):17634-17639. [5] Adams BL, Canova GR, Molinari A. A Statistical Formulation of Viscoplastic Behavior in Heterogeneous Polycrystals. Textures and Microstructures. 1989;11(1):57-71. [6] Garmestani H, Lin S, Adams BL, Ahzi S. Statistical continuum theory for large plastic deformation of polycrystalline materials. Journal of the Mechanics and Physics of Solids. 2001;49(3):589-607. [7] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. A new approximation for the three-point probability function. International Journal of Solids and Structures. 2009;46(21):3782-3787. [8] Nemat-Nasser S, Hori M. Micromechanics : overall properties of heterogeneous materials. 2nd rev. ed. Amsterdam ; New York: Elsevier; 1999. [9] Dumont JP, Ladeveze P, Poss M, Remond Y. Damage mechanics for 3-D composites. Composite Structures. 1987;8(2):119-141. [10] Hori M, Munasighe S. Generalized Hashin-Shtrikman variational principle for boundaryvalue problem of linear and non-linear heterogeneous body. Mechanics of Materials. 1999;31(7):471-486. [11] Affdl JCH, Kardos JL. The Halpin-Tsai equations: A review. Polymer Engineering & Science. 1976;16(5):344-352. [12] Edward R, Principe L. How to Use FIB-SEM Data for 3-D Reconstruction. 2005. [13] Jiao Y, Stillinger FH, Torquato S. Modeling heterogeneous materials via two-point correlation functions: Basic principles. Physical Review E. 2007;76(3):031110. [14] Jiao Y, Stillinger FH, Torquato S. Modeling heterogeneous materials via two-point correlation functions. II. Algorithmic details and applications. Physical Review E. 2008;77(3):031135. [15] Ketcham R. X-ray Computed Tomography (CT). 2011. [16] Reuteler J. Introduction to FIB-SEM. 51 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter 1: Literature Survey [17] Merle P. X-Ray Computed Tomography on Metal Matrix Composites. Vienna University of Technolog: Insitute of Materials Science and Testing 2000. [18] Yeong CLY, Torquato S. Reconstructing random media. Physical Review E. 1998;57(1):495. 52 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function Chapter II 53 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function 54 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function Using SAXS Approach to Calculate Two-Point Correlation Function: (Application to Polystyrene/Zirconia Nanocomposite) 55 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function 56 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function II.1. Introduction Statistical continuum theory correlates the morphology of microstructures to physical properties of heterogeneous materials through correlation functions. In this framework, statistical n-point correlation functions provide a mathematical representation of heterogeneous materials morphology [1]. Particularly, one-point correlation functions gives information about the volume fraction of each constituent (phase) of the heterogeneous material [1]. The distribution, orientation and shape of the heterogeneous material phases are described by two-point or higher order correlation functions, which can be in general determined from appropriate microstructure measurements [2]. These measurements must be representative of the material morphology, i.e. the experimental information must reflect all the variation of phase distribution within the material. The heterogeneity, introduced through the polymer-based nanocomposites, can be represented by: i) the overall distribution of the nanoparticles within the polymer matrix and ii) the local heterogeneity of the nanoparticles which is called dispersion state [3]. What dictates the material properties is actually the dispersion state of the nanoparticles. We therefore consider that the dispersion of the nanoparticles within polymer matrix is the key distribution parameter to take into account in the statistical theory. To have information about nanoparticles dispersion, transmission electron microscopy (TEM) or x-ray scattering can be used [4]. However, in the case of TEM analysis, the TEM images are only relevant when the entire dispersion gradients of the nanoparticles are represented [5]. Particularly, uniform nanoparticles dispersion is not usually achieved. In this case, the microstructure is characterized by a mixture of single particles and aggregates containing more than one particle (aggregation). Note that, the nanoparticles aggregate size can reach several hundred nanometers depending on the nanoparticle size, processing method and the chemical interactions between the nanoparticle and the matrix. Therefore, the calculated correlation functions strongly depend on the magnification at which the TEM images are recorded. Using a high magnification, the correlation function will be dictated by the position within the heterogeneous material where the microscopy images are taken (e.g. whether the TEM images are chosen to include aggregates or not) [6]. In other words, the resolution can be high but the representative area (or volume) is much larger that the selected image [7]. On the contrary, using a low magnification, more representative information about the dispersion of the nanoparticles 57 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function will be obtained. In this case, the statistics are high but the resolution is low. As an alternative, dispersion state of nanoparticles in the polymer-based nanocomposites can be characterized by small-angle x-ray scattering (SAXS) measurement [8]. SAXS is an easy and fast method that is applied to a volume of the order of several cubic millimeters (high statistics) without compromising the resolution. The obtained scattering signal of the nanoparticles reflects the size distribution and shape of the nanoparticles (form factor) and their position with respect to each other (structure factor). For example, a high dispersion state of the nanoparticles within a polymer matrix will be characterized by an average particle size near that of a single particle and eventually a homogeneous interparticle distance. SAXS signal can be consequently exploited to calculate two-point correlation functions with a high accuracy since it produces a very accurate representation of the material morphology [9-12]. In this work, SAXS data is exploited to calculate two-point correlation function correlation for two phase polymer composite. The selected material is polystyrene (PS) filled with zirconium oxide nanoparticles (ZrO2). The nanocomposite morphology was first examined by scanning transmission electron microscopy (STEM) and SAXS. The two-point correlation functions were then calculated from SAXS measurements, while the three-point correlation functions can be approximated [13] from two-point correlation functions relation . II.2. Correlation between SAXS data and two-point correlation functions Small-angle x-ray scattering technique relies on electron density scattered from heterogeneities particles whose size typically ranges between 1 and 1000 nm, depending on the equipment configuration [14-16]. The scattered intensity depends on the difference between a local electronic density U from the scattered heterogeneities and its surrounding, which can be represented by an average density U . The local fluctuation K of the electron density can be defined as follows: K UU . (1) Assuming a statistically isotropic system with no long-range order, a correlation function that considers the amplitude of the density fluctuations can be defined as: 58 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function J r K2 KAKB (2) where A and B are two distinct points in the medium represented by the vectors r1 , r2 , r r2 r1 and J r is the characteristic or autocorrelation function depending on the position r. J r can be defined as follows: J r K(r1 )K(r2 ) (3) For random distribution of heterogeneities, the autocorrelation function J r satisfy the following conditions: J r 0 K and J r o f 0 . It is convenient to define the auto2 covariance of phase-1 for a statistically homogeneous media as [1] : J r K(r1 )K(r2 ) P211 (r ) I12 (4) where I1 is the volume fraction of phase 1 (fillers) and P211 (r ) is the two-point probability function. Recalling that U r is the number of electrons per unit volume, a volume element dV at position r will contain U r u dV electrons. The intensity of the x-ray scattering I as a function of the scattering vector h over the entire volume V is given by the following Fourier integral [17] : I (h h)) ³³³ ³³³ dV dV U(r )U(r )e 1 2 1 ihr 2 ³³ U(r )U(r )e 1 2 ihr dr1dr2 (5) V Summing all pairs with the same relative distance, then integrating over all relative distances, seems to be a logical course. An autocorrelation function can be defined as: U2 (r ) { ³³³ dV1U(r1 )U(r2 ) (6) which allows to rewrite I h as: I(h) h) ³³³ dVU (r)e 2 ihr (7) 59 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function implying that the intensity distribution in h or reciprocal space, is uniquely determined by the structure of the density field. Considering statistical isotropy, Debye [9, 10] proved that sin(hr ) hr eihr (8) As a result, the average scattering intensity reduces to: ³ 4Sr dr UU (r ) 2 I ( h) 2 sin(hr ) hr (9) Recalling the autocorrelation function, J , the above equation can be rewritten: I (h) Vn02 ³ 4Sr 2 dr J(r ) sin(hr ) hr (10) where n0 is the mean density of electrons. Or, J (r ) 1 2S Vn02 2 ³ f 0 I ( h) sin(hr ) 2 h dh hr (11) here, n0 is a constant. Using equation (4), the equation (11) can be rewritten as follow: J (r ) 11 where P2 1 2S Vn02 P211 (r ) I12 r represents 2 ³ f 0 I (h) sin(hr ) 2 h dh hr (12) the two-point probability correlation function which measures the 11 spatial distribution of the heterogeneities (phase-1) in the matrix (phase-2). P2 r should verify the following condition: P211 r I1 P211 r I1 when 2 r 0 when r o f . (13) The second condition in equation (13) is an indicator of the degree of homogeneity of the distribution of heterogeneities in the matrix (i.e. if the second condition is not verified then the distribution of the heterogeneities are not homogeneous in the matrix). 60 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function II.3. Structural characterization 3.1. Materials The polymer matrix of the studied nanocomposite, polystyrene (PS), was supplied by Scientific Polymer Products Inc. It has a molecular weight of about 120,000 g/mol. As for the zirconium oxide (ZrO2) nanofiller, it was provided by Sigma Aldrich under the reference # 544760 (average particle size < 100 nm according to the datasheet). The specific surface area (measured by the Brunauer, Emmett, and Teller method) and the density of ZrO2 were 25 m2/g and 5.89 g/cm3, respectively. II.3.2. Preparation of the nanocomposites All nanocomposites were prepared by melt mixing. The following material systems were extruded by means of a micro-compounder DSM (reference Xplore 15 mL): neat PS, PS + 1 wt. % ZrO2, PS + 3 wt. % ZrO2, and PS + 5 wt. % ZrO2. During this procedure, each system was compounded during 5 minutes at 230°C with a screw co-rotating speed of 200 rpm. To avoid oxidation phenomena, the extrusion was carried out under argon gas. The produced materials were extruded cylinders, 5 mm in diameter. Thermogravimetric analysis was performed after the processing step to measure the effective amount of nanoparticles within PS. The results indicated that the amount of nanoparticles used for the processing are preserved. II.3.3. Scanning transmission electron microscopy We performed a structural characterization of the nanocomposite by scanning transmission electron microscopy (STEM) to verify the presence of aggregates within the polymer matrix. STEM analyses of PS-ZrO2 nanocomposites were carried out using a scanning electron microscope FEI Quanta FEG 200 apparatus at 7 kV. The STEM samples were ultra-thin films (70 nm-thick) that were prepared with a Leica EM FC6 cryo-ultra-microtome at 25°C using a trimming diamond blade. Fig. 1 shows some tendency to aggregation whatever the amount of filler. The size of the aggregates is much less than 200 nm except for very few cases for which the size of the aggregates is in the micrometric range. The tendency to aggregation can be explained by the fact that the particles were not coated, which does not enable to increase the interaction between the 61 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function oxide particle and the polymer matrix. It is also thought that the used micro-compounder do not enable to reach an optimal dispersion state. This can be explained by the geometry of the extrusion screws that does not permit to obtain a high enough elongational flow to completely break up the aggregates into primary particles. However, the low beam energy of the scanning electron microscope does not enable to observe the local distribution state of the nanoparticles, i.e. the dispersion state, with the transmission mode. To characterize this local distribution of the particles, small-angle x-ray scattering technique was employed. Fig. 1. STEM micrographs at two magnifications, 5 000 (a, c and e) and 50 000 (b, d and f), of the composites PS +1 wt. % ZrO2 (a and b), PS + 3 wt. % ZrO2 (c and d), and PS + 5 wt. % ZrO2 (e and f) 62 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function II.3.4. Small-angle x-ray scattering Small-angle x-ray scattering tests were performed by means of a Panalytical X’Pert Pro MPD device to study the scattering signal of the nanoparticles within the polymer matrix, and hence to obtain physical and structural information. Particularly, the analysis of the scattering signal enables to characterize the size and shape of the particles (form factor) as well as their relative ordering (structure factor) [12]. The radiation used O =1.54 Å (Cu KD), at 45 kV and 40 mA was generated by an x-ray tube operating at 40 kV and 45 mA. A focused parallel mirror and a PIXcel detector were employed with specific slits in order to obtain the highest resolution at small-angle. Also, to attract strong signal from the nanoparticles, the background noise from the SAXS curves, i.e. the scattering curve of neat PS, was systematically subtracted. The last treatment of the curves consisted of assessing the intensity of primary beam that passes through the nanoparticles (absorption correction). The scattering intensity I is plotted as a function of scattering vector h = (4π/O) sin (θ) where θ is the scattering angle. Each SAXS test was repeated on three specimens. Fig. 2 shows representative scattered intensity I(h) of ZrO2 characterized alone (as-received powder), and characterized within PS matrix (investigated amounts: 3 and 5 wt. %). A high reproducibility of I(h) curves was found for each material. It is to be noted that no scattering signal was obtained for 1 wt. % of ZrO2 within PS, and hence the scattering curve of this system was not plotted in Fig. 2. This is certainly due to the resolution of the x-ray scattering equipment that does not enable to characterize such a low content of particles (1 wt. %) within a polymer matrix. For the other investigated ZrO2 amounts, no long-range Bragg peak is noted on the scattering curves, indicating no ordering of ZrO2 nanoparticles and hence no interaction strength between the particles as for example Van-der-Waals or hard-sphere interactions. Consequently, the scattering signal of the nanoparticles is only induced by the form factor. The initial parts of I(h) curves, below h = 0.07 nm-1, show a continuous decrease of the intensity with h that suggests some large aggregate. This observation is in line with the aggregation tendency noted on STEM images (Fig. 1). It is to be noted that in the case of a well-dispersed system, an initial "plateau" of scattering intensity at very low h would have been noted, followed by a gradual decrease of the intensity with h. Above h = 0.07 nm-1, the presence of some oscillation indicates the presence of scattering objects that have a relative uniform size. These objects are most 63 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function probably single particles and small aggregates constituted by few single particles. By means of Guinier plot (ln I as a function of h2), we found a radius of gyration Rg (deduced from the slope of ln(I)-h2 curve) of about 13.5 nm for as-received ZrO2, 5% of ZrO2 in PS and 3 % of ZrO2 in PS. Considering ZrO2 nanoparticles as having a spherical shape, the average particle size (diameter), deduced from the relationship 2×Rg×(5/3)0.5= 34.8 nm. Despite the presence of some big aggregates in the micrometric range (Fig. 1), we considered that the most representative information about the distribution of the particles is provided by SAXS. I(h) curves (Fig. 2) are hence used for the calculation of the two-point correlation functions. Fig. 2. The scattered intensity I as a function of scattering vector h for ZrO2 nanopowder and PS-ZrO2 composites (3 and 5 wt %) (background- and absorption-corrected curves) . The two-point probability functions representing the distribution of the ZrO2 nanoparticles within the PS matrix are calculated using equation (12) and reported in Fig. 3. Note that since 64 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function 11 SAXS diagram of 1 % ZrO2 in PS do not show any signal from the nanoparticles, P2 r curve presents negative values and does not verify the second condition in equation (13). Therefore P211 r for 1 wt. % ZrO2 cannot be exploited to calculate the physical properties of the nanocomposite. However, the two-point probability function P2 r , for 3 and 5 wt. % ZrO2 (see 1 Fig. 3) verifies the limits given in equation (13). Fig. 3. Two-point correlation functions TPCF for PS-ZrO2 composites (3 and 5 wt. % of ZrO2) 65 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function II.4. Conclusion Polystyrene (PS) nanocomposites were produced by melt mixing technique using zirconium oxide (ZrO2) as fillers. The spatial dispersion of nanoparticles within the polymer matrix was characterized by STEM and SAXS measurements. A non-uniform dispersion of the nanoparticles within the polymer matrix with a tendency to aggregation is obtained. The SAXS signals are used to calculate the correlation functions that represent the spatial dispersion of the nanoparticles considered as the key distribution parameter in such heterogeneous materials. The calculated correlation functions can be used in conjunction with the strong contrast version of the statistical continuum theory to predict the effective mechanical and thermal properties for both 3 and 5 wt. % ZrO2.. 66 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function II. References [1] Torquato S, Haslach HW. Random Heterogeneous Materials: Microstructure and Macroscopic Properties. Applied Mechanics Reviews. 2002;55(4):B62-B63. [2] Jiao Y, Stillinger FH, Torquato S. Modeling heterogeneous materials via two-point correlation functions: Basic principles. Physical Review E. 2007;76(3):031110. [3] Alexandre M, Dubois P. Polymer-layered silicate nanocomposites: preparation, properties and uses of a new class of materials. Materials Science and Engineering: R: Reports. 2000;28(12):1-63. [4] Kashiwagi T, Harris RH, Zhang X, Briber RM, Cipriano BH, Raghavan SR, et al. Flame retardant mechanism of polyamide 6-clay nanocomposites. Polymer. 2004;45(3):881-891. [5] Kashiwagi T, Fagan J, Douglas JF, Yamamoto K, Heckert AN, Leigh SD, et al. Relationship between dispersion metric and properties of PMMA/SWNT nanocomposites. Polymer. 2007;48(16):4855-4866. [6] Li DS, Baniassadi M, Garmestani H, Ahzi S, Reda Taha MM, Ruch D. 3D Reconstruction of Carbon Nanotube Composite Microstructure Using Correlation Functions. journal of computational and theoretical nanoscience. 2010;7(8):1462-1468. [7] Lingaiah S, Sadler R, Ibeh C, Shivakumar K. A method of visualization of inorganic nanoparticles dispersion in nanocomposites. Composites Part B: Engineering. 2008;39(1):196201. [8] Bandyopadhyay J, Sinha Ray S. The quantitative analysis of nano-clay dispersion in polymer nanocomposites by small angle X-ray scattering combined with electron microscopy. Polymer. 2010;51(6):1437-1449. [9] Debye P, Anderson HR. The correlations Function and Its Application. JOURNAL OF APPLIED PHYSICS. 1957;28(6):4. [10] Debye P, Anderson HR, Brumberger H. Scattering by an Inhomogeneous Solid 2. The Correlations Function and Its Application. JOURNAL OF APPLIED PHYSICS. 1957;28(6):679-683. [11] Frisch HL, Stillinger FH. Contribution to the Statistical Geometric Basis of Radiation Scattering. The Journal of Chemical Physics. 1963;38(9):2200-2207. [12] Gunier A, Fournet G. Small Angle Scattering of X-Rays., New York: John Wiley; 1955. [13] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. A new approximation for the three-point probability function. International Journal of Solids and Structures. 2009;46(21):3782-3787. [14] Brumberger H. Modern Aspects of Small-Angle Scattering. Boston: Kluwer Academic Publishers; 1995. 67 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter II: Using SAXS Approach to Calculate Two-Point Correlation Function [15] Feigin LA, Svergun DI. Structure Analysis by Small-Angle X-ray and Neutron Scattering. New York Plenum Press; 1987. [16] Cullity BD, Stock SR. Elements of X-ray Diffraction, . New Jersey: Prentice Hall; 2001. [17] Glatter O, Kratky O. Small Angle X-ray Scattering. New York: : Academic Press; 1982. 68 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Chapter III 69 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials 70 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials 71 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials 72 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials III.1. Introduction Description and characterization of heterogeneous systems have become of extreme importance to scientists during the past decades. Many techniques have been developed to realise threedimensional descriptions of heterogeneous systems [1]. Statistical continuum mechanics provides a robust alternative for the reconstruction and characterization techniques of heterogeneous systems. The reconstruction techniques have been empowered by the development of numerous simulation methodologies in recent years. Anisotropic features, orientation distribution, shape and geometrical features can be extracted from statistical correlation functions. Yeong and Torquato [1, 2] have initiated the study of microstructure reconstruction using correlation functions. Random heterogeneous materials were reconstructed from low order correlation functions via stochastic optimization annealing techniques. Different types of microstructures were investigated to examine the limitations of the reconstruction techniques to include short-range order. An exact mathematical formulation of the reconstruction algorithm was presented by Yeong and Toquarto [1, 2]. In a recent work, Garmestani and coworkers [3] have developed a new Monte Carlo (MC) methodology using Colony and kinetic growth algorithm. This approach have been developed to reconstruct the microstructure of twophase composites using statistical correlation functions [3]. This was recently extended by Baniassadi and co-workers [4] to three-dimensional multiphase composites, specifically applied to planar section solid oxide fuel cell materials, to develop three-dimensional microstructures. Li and co-workers[5] have presented a novel Monte Carlo technique by incorporating geometry, distribution and waviness of virtual nanotube fillers for the reconstruction of Carbon Nanotube (CNT) polymer composites. In this approach, the nanotubes were described as a chain of links and the reconstruction was performed by the optimization of the waviness, geometry and preferential distribution of CNTs. Characterization of mechanical, magnetic, electrical and thermal properties can be performed directly from descriptors such as N-point statistics. Different statistical continuum approaches (weak-contrast and strong-contrast) have been developed to account for the material heterogeneity through probability functions (Kröner [6]; Beran [7]; Phan-Thien and Milton [8]; Dederichs and Zeller [9] , Willis [10]; McCoy [11]; Torquato [1, 12, 13]; Sen and Torquato [14]). Weak contrast technique is based on perturbation from the average property and can be 73 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials utilized for heterogeneous materials with small variation in the multi-phase properties. A strongcontrast expansion for two-phase isotropic media was developed by Brown [15] for the effective conductivity that resulted in convergent integrals. Other scientists, such as Torquato [1] and Fullwood [16], developed this method for n-dimensional space and anisotropic multiphase heterogeneous materials. N-point correlation functions have a long history in important science and engineering applications going back to the invention of X-ray scattering and diffraction early last century. Statistical information in the form of pair-correlation functions can be extracted by using scattering data [17, 18]. Small angle X-ray scattering technique has been used to get information on the distribution of inclusions and dispersion of particles [19] . Corson [20] has developed methodologies linking properties of two-phase structures to the experimentally calculated two-, and three-point probability functions. In this approach the probability functions are assumed to be isotropic. In 1987, Adams et al. [21] introduced a set of two-point probability functions based on spherical harmonics. The spectral technique was used to account for orientation and point-to-point correlations in the microstructure. Garmestani and others [22-27] have later extended the statistical continuum approach to both composites and polycrystalline materials using two-point functions. Mikdam et al. [28] have developed an approximation for the 3-point correlation functions based on two-point functions. In other researches, Mikdam et al. [29] and Baniassadi et al. [17, 30] have applied the strong-contrast formulation to predict the effective electrical and thermal conductivity of a two-phase composite material where the distribution, shape and orientation of the two phases are taken into account using two-point and three-point correlation functions. In the present work, we propose to use the conditional probability to derive a comprehensive formulation of the N-point correlation functions for multiphase non FGM heterogeneous materials. The approximation of the used probabilities and the use of the boundary conditions allowed us to derive a new and broad approximation of the N-point probability functions. We show the capability of this new approach by comparing our predicted results to results from the computed real probability functions (for a computer generated microstructure) for three-point correlation functions. 74 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials III. 2. Approximation of tree-point correlation functions III. 2.1. Decomposition of Higher Order Statistics Higher order correlations can incorporate more details of the morphology of the secondary phases. Theoretically, a unique microstructure can be reconstructed by using an infinite order correlation function. In statistical mechanics formulations, it is necessary to exploit higher order correlation functions for a better identification of heterogeneous systems. In the current work, Npoint correlation functions have been approximated by use of (N-1)-point correlation functions. To obtain this approximation, N-point correlation function was partitioned into N subsets of (N1)-point correlation functions. For instance, the set of X of points or events (x1,x2, …xN), the subsets of X are given below: X=ሼݔଵ , ݔଶ ,…., ݔ ሽ (1) ܷܵܵܶܧܵܤሺܺሻ ൌ ሼሼݔ ሽǡ ሼݔଵ , ݔଶ ,…., ݔିଵ ሽ ǡ ሼݔଵ , ݔଶ ,…., ݔ ሽ ǡ ǥ ሽ(2) In this work, we denote by CN(x1, x2, ….xN) the N-point probability function for the occurrence of the point (x1, x2, ….xN) in a desired phase (occurrence of the event ሺݔଵ ݔ תଶ תǥ ݔ תே ሻ): CN (ݔଵ , ݔଶ , ǥݔே ) ൌ ܲሺݔଵ ݔ תଶ תǥ ݔ תே ሻ (3) Here, ܲሺݔଵ ݔ תଶ תǥ ݔ תே ሻ represents the probability of the eventሺݔଵ ݔ תଶ תǥ ݔ תே ሻ. For simplicity, the following properties of this correlation function are shown for the case of N=3 C3 (ݔଵ , ݔଶ , ݔଷ ) = C3 (ݔଶ , ݔଵ , ݔଷ ) = C3 (ݔଵ , ݔଷ , ݔଶ )ൌ ܲሺݔଵ ݔ תଶ ݔ תଷ ሻ (4) C3 (ݔଵ , ݔଶ , ݔଶ ) = ܲሺݔଵ ݔ תଶ ݔ תଶ ሻ= C2 (ݔଵ , ݔଶ ) (5a) C3 (ݔଶ ,ݔଷ ,ݔଶ ) = ܲሺݔଶ ݔ תଶ ݔ תଷ ሻ = C2 (ݔଶ , ݔଷ ) (5b) III. 2.2. Decomposition of two-point correlation functions Two-point correlation function is the probability of finding the beginning and ending points of a random vector with length r in a desired phase. According to the probability theory, two compatible events can be independent or dependent under favorable conditions. Dependency and independency of the two events in a heterogamous system depend on the length of the vector r. 75 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials This means that for very small r (very small in comparison to the RVE dimension), the probabilities of the occurrence of two events x1 and x2 are within the correlation limit (or dependant). However, for very large values of r these probabilities will be independent. For a general formulation, we introduce the dependency weight factor, ଵଶ , which will allow us to express the 2-point correlation function in terms of the 1-point probability functions in the following multiplicative decomposition: ܥ2ሺݔଵ ǡ ݔଶ ሻ ൌ ܲሺݔଵ ݔ תଶ ሻ ൌ ܹଵଶ ܥ כ1ሺݔଵ ) ܥ כ1ሺݔଶ ሻ ൌ ܹଵଶ ܲ כሺݔଵ )ܲ כሺݔଶ ሻ (6) The dependency factor is a function of the vector length: ܹଵଶ ൌ ݂ሺȁݎȁሻ (7) For very large r, the independence of events x1 and x2 yields the following: ݈݅݉ȁȁ՜ஶ ݂ሺȁݎȁሻ ൌ ͳ (8) In addition, for very small length of r (ȁȁ ՜ Ͳሻǡif the two events x1 and x2 are compatible we have ݈݅݉ȁȁ՜ ݂ሺȁݎȁሻ ൌ ଵ ሺ௫భ ሻ (9) However, if the two events are incompatible we have: ݈݅݉ȁȁ՜ ݂ሺȁݎȁሻ ൌ Ͳ (10) Note that the indices in the dependency weight factor ܹ represent the order of the correlation function (upper index b) and the number indicating each of factors needed (lower index a). III. 2. 3. Decomposition of three-point correlation functions A decomposition methodology is presented here to represent and estimate three-point correlation functions by use of two-point correlation functions. A full set of information for the two-point correlation functions must be available for the correct representation of the 3-point functions. First, the set of points (x1,x2,x3) is selected in a heterogeneous system and an analysis is performed according to conditional probability. Fig. 1 illustrates the three random points 76 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials representing the set (x1,x2,x3). Assume the occurrence of the event x1 (point x1 is found in specified phase). The probabilities of finding second and third points (x2 and x3) in the same phase are given by the following conditional probabilities: ܲሺݔଶ ȁݔଵ ሻ ൌ ሺ௫భ ת௫మ ሻ ܲሺݔଷ ȁݔଵ ሻ ൌ ሺ௫భ ת௫య ሻ ሺ௫భ ሻ ሺ௫భ ሻ (11) (12) Fig. 1.Three random points selected to calculate the three-point correlation function The three-point probability function for the occurrence of the event (x1,x2,x3) is equal to the sum of the probabilities of the following possible events. In this, we introduce the dependency factors (ܹଵଷ , ܹଶଷ and ܹଷଷ ) used to formulate our proposed approximation for the three-point correlation functions: The probability of occurrence of x1 followed by x2 and then x3 can be expressed as: ܹଵଷ ܲ כሺݔଵ ሻ ܲ כሺݔଶ ȁݔଵ ሻ ܲ כሺݔଷ ȁݔଵ ሻ (13) 77 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Similarly, the probability of occurrence of x2 followed by x1 and then x3 is : ܹଶଷ ܲ כሺݔଶ ሻ ܲ כሺݔଵ ȁݔଶ ሻ ܲ כሺݔଷ ȁݔଶ ሻ (14) Finally, the probability of occurrence of x3 followed by x1 and then x2 is given by : ܹଷଷ ܲ כሺݔଷ ሻ ܲ כሺݔଵ ȁݔଷ ሻ ܲ כሺݔଶ ȁݔଷ ሻ (15) Therefore, the three-point correlation function C3(x1,x2,x3) is given by the following approximation which adds the above probability approximations of the three possible events: ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܲሺݔଵ ݔ תଶ ݔ תଷ ሻ ൎ ܹଵଷ ܲ כሺݔଵ ሻ ܲ כሺݔଶ ȁݔଵ ሻ ܲ כሺݔଷ ȁݔଵ ሻ ܹଶଷ ܲ כሺݔଶ ሻ כ ܲሺݔଷ ȁݔଶ ሻ ܲ כሺݔଵ ȁݔଶ ሻ ܹଷଷ ܲ כሺݔଷ ሻ ܲ כሺݔଵ ȁݔଷ ሻ ܲ כሺݔଶ ȁݔଷ ሻ (16) We can then write: ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൎ ܹଵଷ כ మ ሺ௫భ ǡ௫మ ሻכమ ሺ௫య ǡ௫భ ሻ భ ሺ௫భ ሻ ܹଶଷ כ మ ሺ௫భ ǡ௫మ ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫మ ሻ ܹଷଷ כ మ ሺ௫య ǡ௫మ ሻכమ ሺ௫య ǡ௫భ ሻ భ ሺ௫య ሻ (17) The weight functions ܹ can now be calculated using the boundary conditions: The first boundary condition is: ݈݅݉௫భ ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ (18) where, x1՜ λ is meant to satisfy the following conditions: |r12ȁ ՜ λ and |r13ȁ ՜ λ. Therefore ݈݅݉௫భ ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܹଵଷ כ భ ሺ௫భ ሻכభ ሺ௫మ ሻכభ ሺ௫య ሻכభ ሺ௫భ ሻ భ ሺ௫భ ሻ భ ሺ௫భ ሻכభ ሺ௫య ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫య ሻ ܹଶଷ כ భ ሺ௫భ ሻכభ ሺ௫మ ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫మ ሻ ܹଷଷ כ (19) Applying this boundary condition we get: ݔଵ ՜ λܹଵଷ ൌ Ͳ (20) Similarly, for ଶ and ଷ we get: ݔଶ ՜ λܹଶଷ ൌ Ͳ ݔଷ ՜ λܹଷଷ ൌ Ͳ (21) (22) 78 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials The second boundary condition is: ݈݅݉ ௫భ ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ (23) ௫మ ՜ஶ ௫య ՜ஶ From equality condition for left side and right side of Eq. (17) we have: ݔଵ ՜ λǡ ݔଶ ՜ λǡ ݔଷ ՜ λܹଵଷ ܹଶଷ ܹଷଷ ൌ ͳ (24) Third boundary condition: ݈݅݉௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଶ ሺݔଶ ǡ ݔଷ ሻ (25) This yields: ݈݅݉௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܹଵଷ כ మ ሺ௫య ǡ௫మ ሻכమ ሺ௫య ǡ௫మ ሻ భ ሺ௫య ሻ మ ሺ௫మ ǡ௫మ ሻכమ ሺ௫య ǡ௫మ ሻ భ ሺ௫మ ሻ ܹଶଷ כ మ ሺ௫మ ǡ௫మ ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫మ ሻ ܹଷଷ כ (26) From this boundary conditions and using Eq. (17) we get: ݔଵ ՜ ݔଶ ܹଷଷ ൌ Ͳ (27) By applying similar methodology for ݔଶ and ݔଷ we obtain: ݔଵ ՜ ݔଷ ܹଶଷ ൌ Ͳ (28) ݔଶ ՜ ݔଷ ܹଵଷ ൌ Ͳ (29) Therefore, necessary conditions for weight function are referred as follows (see details in additional Appendix A): ݔ ՜ λܹଷ ൌ Ͳǡ ܹଷ ് Ͳ݂݅ ് ݆ݎ (30) ݔ ՜ λሺ݅ ൌ ͳǡ ǥ ͵ሻǡσ ܹଷ ൌ ͳ (31) ݔ ՜ ݔ ሺ݂݆ ് ݅ݎሻǡܹଷ ൌ Ͳǡ ݇ ് ݅ܽ݊݀݇ ് ݆ (32) 79 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials By assuming ܹଷ = fi (R1, R2, R3) where R1, R2 and R2 are the lengths of the radii between the three random points shown in Fig. 1. One choice for the weight functions that verifies all of the above boundary conditions is given by the proposed following radii ratios: ோଵ ܹଵଷ ൌ ோଵାோଶାோଷ ܹଶଷ ൌ ோଵାோଶାோଷ ܹଷଷ ൌ ோଵାோଶାோଷ (33) ோଶ (34) ோଷ (35) III. 3. Approximation of four-point correlation function We consider four random points arranged as a tetrahedron which encompasses a sphere of the radius Ri with the following four-point probability function (Fig. 2): Fig. 2. Four random points selected to calculate the four-point correlation function Similarly to the development of the three-point correlation functions in the previous section, the four-point correlation function has been approximated using three and two-point correlation functions. ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ܲሺݔଵ ݔ תଶ ݔ תଷ ݔ תସ ሻ ൎ 80 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials ܹଵସ ܲ כሺݔଵ ሻ ܲ כ൫ሺݔଶ ݔ תଷ ሻหሺݔଵ ݔ תଷ ሻ൯ ܲ כ൫ሺݔସ ݔ תଶ ሻหሺݔଵ ݔ תଶ ሻ൯ ܲ כ൫ሺݔସ ݔ תଷ ሻหሺݔଵ ݔ תସ ሻ൯ ܹଶସ ܲ כሺݔଶ ሻ ܲ כ൫ሺݔଷ ݔ תଵ ሻหሺݔଶ ݔ תଵ ሻ൯ ܲ כ൫ሺݔଵ ݔ תସ ሻหሺݔଶ ݔ תସ ሻ൯ ܲ כ൫ሺݔଷ ݔ תସ ሻหሺݔଶ ݔ תଷ ሻ൯ ܹଷସ ܲ כሺݔଷ ሻ ܲ כ൫ሺݔଵ ݔ תଶ ሻหሺݔଷ ݔ תଶ ሻ൯ ܲ כ൫ሺݔସ ݔ תଵ ሻหሺݔଷ ݔ תଵ ሻ൯ ܲ כ൫ሺݔସ ݔ תଶ ሻหሺݔଷ ݔ תସ ሻ൯ ܹସସ ܲ כሺݔସ ሻ ܲ כ൫ሺݔଶ ݔ תଵ ሻหሺݔସ ݔ תଵ ሻ൯ ܲ כ൫ሺݔଷ ݔ תଵ ሻหሺݔସ ݔ תଵ ሻ൯ כ ܲ൫ሺݔଶ ݔ תଷ ሻหሺݔସ ݔ תଶ ሻ൯ (36) Or, ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ଵସ ܥ כଵ ሺݔଵ ሻ כ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଵ ǡ ݔଷ ǡ ݔସ ሻ ܥ כଷ ሺݔଵ ǡ ݔଶ ǡ ݔସ ሻ ܥଶ ሺݔଵ ǡ ݔଶ ሻ ܥ כଶ ሺݔଵ ǡ ݔଷ ሻ ܥ כଶ ሺݔଵ ǡ ݔସ ሻ ଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଷ ሺݔଶ ǡ ݔଵ ǡ ݔସ ሻ ܥ כଷ ሺݔଶ ǡ ݔଵ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥଶ ሺݔଵ ǡ ݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔଵ ሻ ܥ כଷ ሺݔଷ ǡ ݔଵ ǡ ݔସ ሻ ܥଶ ሺݔଷ ǡ ݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ସସ ܥ כଵ ሺݔସ ሻ כ ܥଷ ሺݔସ ǡ ݔଵ ǡ ݔଶ ሻ ܥ כଷ ሺݔସ ǡ ݔଵ ǡ ݔଷ ሻ ܥ כଷ ሺݔସ ǡ ݔଶ ǡ ݔଷ ሻ ܥଶ ሺݔସ ǡ ݔଵ ሻ ܥ כଶ ሺݔସ ǡ ݔଶ ሻ ܥ כଶ ሺݔସ ǡ ݔଷ ሻ (37) The weight functions ܹ are calculated using boundary conditions: The first boundary condition is: ݈݅݉௫భ ՜ஶ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ܥଵ ሺݔଵ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ (38) This limit can be written as: 81 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials ݈݅݉ ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ௫భ ՜ஶ ൌ ܹଵସ ܥ כଵ ሺݔଵ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܹଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܹଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ܹସସ ܥ כଵ ሺݔସ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔସ ǡ ݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔସ ǡ ݔଷ ሻ ܥ כଷ ሺݔସ ǡ ݔଶ ǡ ݔଷ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଶ ሺݔସ ǡ ݔଶ ሻ ܥ כଶ ሺݔସ ǡ ݔଷ ሻ (39) Applying the boundary condition we get: ݔଵ ՜ λܹଵସ ൌ Ͳ (40) Similarly for ଶ , ଷ and ସ , we obtain ݔଶ ՜ λܹଶସ ൌ Ͳ (41) ݔଷ ՜ λܹଷସ ൌ Ͳ (42) ݔସ ՜ λܹସସ ൌ Ͳ (43) The second boundary condition is: ݈݅݉௫భ ՜ஶ ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ௫మ ՜ஶ ௫య ՜ஶ ௫ర ՜ஶ (44) We can also write: ݈݅݉௫భ ՜ஶ ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ௫మ ՜ஶ ௫య ՜ஶ ௫ర ՜ஶ 82 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials ܹଵସ ܥ כଵ ሺݔଵ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܹଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଵ ሺݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܹଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଷ ሻ ܥଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܹସସ ܥ כଵ ሺݔସ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ (45) Applying this boundary condition leads to: ݔଵ ՜ λǡ ݔଶ ՜ λǡ ݔଷ ՜ λݔସ ՜ λܹଵସ ܹଶସ ܹଷସ ܹସସ ൌ ͳ (46) Third boundary condition: ݈݅݉௫భ ՜௫మ ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ܥଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ (47) ݈݅݉ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ௫భ ՜௫మ ଵସ ܥ כଵ ሺݔଶ ሻ כ ܥଷ ሺݔଶ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥ כଷ ሺݔଶ ǡ ݔଶ ǡ ݔସ ሻ ܥଶ ሺݔଶ ǡ ݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଷ ሺݔଶ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଷ ሺݔଶ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥଶ ሺݔଶ ǡ ݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔଶ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ସସ ܥ כଵ ሺݔସ ሻ כ ܥଷ ሺݔସ ǡ ݔଶ ǡ ݔଶ ሻ ܥ כଷ ሺݔସ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔସ ǡ ݔଶ ǡ ݔଷ ሻ ܥଶ ሺݔସ ǡ ݔଶ ሻ ܥ כଶ ሺݔସ ǡ ݔଶ ሻ ܥ כଶ ሺݔସ ǡ ݔଷ ሻ 83 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials (48) The boundary conditions in Eq. (47) require the following conditions: ݔଵ ՜ ݔଶ ܹଷସ ൌ Ͳܹܽ݊݀ସସ ൌ Ͳ (49) Similarly, for ଶ and ଷ we get: ݔଵ ՜ ݔଷ ܹଶସ ൌ Ͳܹܽ݊݀ସସ ൌ Ͳ (50) ݔଶ ՜ ݔଷ ܹଵସ ൌ Ͳܹܽ݊݀ସସ ൌ Ͳ (51) And finally: ݔଷ ՜ ݔସ ܹଵସ ൌ Ͳܹܽ݊݀ଶସ ൌ Ͳ (52) Therefore, necessary conditions for weight functions are obtained as (see details in additional Appendix B): ݔ ՜ λܹସ ൌ Ͳǡ ܹସ ് Ͳ݂݅ ് ݆ݎ (53) ݔ ՜ λሺ݅ ൌ ͳǡ ǥ Ͷሻǡ σ ܹସ ൌ ͳ (54) ݔ ՜ ݔ ሺ݂݆ ് ݅ݎሻǡܹସ ൌ Ͳ݂݆ ് ݇݀݊ܽ݅ ് ݇ݎ (55) Assuming that ୧ସ are function of area fractions of the tetrahedron faces (areas) in Fig. 2, all boundary condition can be shown to be satisfied through the following ratios: ܹସ ൌ ܹଶସ ൌ ܹଷସ ൌ ܹସସ ൌ ሺ௫మ ǡ௫య ǡ௫ర ሻ ሻାሺ௫ ሺ௫భ ǡ௫మ ǡ௫య మ ǡ௫య ǡ௫ర ሻାሺ௫మ ǡ௫భ ǡ௫ర ሻାሺ௫భ ǡ௫య ǡ௫ర ሻ ሺ௫భ ǡ௫య ǡ௫ర ሻ ሺ௫భ ǡ௫మ ǡ௫య ሻାሺ௫మ ǡ௫య ǡ௫ర ሻାሺ௫మ ǡ௫భ ǡ௫ర ሻାሺ௫భ ǡ௫య ǡ௫ర ሻ ሺ௫మ ǡ௫భ ǡ௫ర ሻ ሺ௫భ ǡ௫మ ǡ௫య ሻାሺ௫మ ǡ௫య ǡ௫ర ሻାሺ௫మ ǡ௫భ ǡ௫ర ሻାሺ௫భ ǡ௫య ǡ௫ర ሻ ሺ௫భ ǡ௫మ ǡ௫య ሻ ሺ௫భ ǡ௫మ ǡ௫య ሻାሺ௫మ ǡ௫య ǡ௫ర ሻାሺ௫మ ǡ௫భ ǡ௫ర ሻାሺ௫భ ǡ௫య ǡ௫ర ሻ (56) (57) (58) (59) 84 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials where, ܣ൫ݔ୧ ǡ ݔ୨ ǡ ݔ୩ ൯ is the area of the side of the tetrahedron encompassing the three points ݔ୧ ǡ ݔ୨ ǡ ݔ୩ . III. 4. Approximation of N-point correlation function The methodology for deriving the approximations in previous sections can be extended to Npoint correlations which unfortunately yields a lengthy procedure for N>3. Thus, we limit our analysis to the following brief general description of the methodology: షభ ቀ ቁ ܥ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ǥ Ǥ ǡ ݔ ሻ ൌ σୀଵሺ୧୬ כ షమ ςసభ ሺషభሻ ൫௫ ǡǥǤǡ௫ሺషభሻ ൯ షభ ቀ ቁ షయ ςసభ ሺషమሻ ൫௫ ǡǥǤǡ௫ሺషమሻ ൯ షభ ቀ ቁ כ షర ςసభ ሺషయሻ ൫௫ ǡǥǤǡ௫ሺషయሻ ൯ షభ ቀ ቁ כǥሻ షఱ ςసభ ሺషరሻ ൫௫ ǡǥǤǡ௫ሺషరሻ ൯ (60) where ୧୬ are the dependency weight functions. In the formulation above, ൫ݔ୫ ǡ ǥ ୧ ǥ ǡ ݔ୮ ൯, is defined as the subset of (N-1)- points that include xi as a member of the subset. The weight functions must satisfy the following limiting boundary conditions: ݔ ՜ λܹ ൌ Ͳǡ ܹ ് Ͳ݂݅ ് ݆ݎ (61) ݔ ՜ λሺ݅ ൌ ͳǡ ǥ ݊ሻǡ σ ܹ ൌ ͳ (62) ݔ ՜ ݔ ሺ݂݆ ് ݅ݎሻǡܹ ൌ Ͳ݂݆ ് ݇݀݊ܽ݅ ് ݈݈݇ܽݎ (63) In the next section, we will present numerical results but only for three-point correlation functions. 85 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials III. 5. Results III. 5.1. Approximation of Three-Point correlation functions for a three-dimensional reconstructed microstructure In this section, the numerical verification of the above approximations is conducted to show the accuracy and the limitations of this methodology. Here, we chose to show results only for N=3. For this, a numerical Monte Carlo program was constructed and used to calculate different 3point statistical functions for a three-dimensional reconstructed microstructure. Monte Carlo methodology is used to reconstruct 3D microstructures of a three-phase anode structure in a solid oxide fuel cell from an experimental 2D SEM micrograph (see Fig. 3) [4]. The three phases shown on the SEM micrograph in Fig. 3 are nickel, yttria-stabilized zirconia (YSZ) and voids [4]. Fig. 3- SEM micrographs of a three-phase Anode microstructure of Solid Oxide Fuel Cell [4] (red: Nickel, blue: YSZ, Black: voids) 86 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials The methodology for the reconstruction is based on a two-point statistical function as a microstructure descriptor. Colony and Kinetic Growth algorithms are used to enable the realization process based on an optimization methodology described in the next chapter (see also [4]). The generated 3-D reconstruction of the microstructure is shown in Fig. 4. (a) (b) Fig. 4. a) Three-dimensional reconstructed image of the Anode microstructure b) several sections through the depth of the 3D microstructure (red: Nickel, blue: YSZ, Black: voids) . Phase 1 (Blue color) Phase 2 (Red color) Phase 3 (black color) [4] For a three-phase composite, we have nine two-point probabilities. Due to normality conditions and knowing that the probability functions are symmetric the number of independent two-point correlation functions reduce to three. For instance, we can choose C2(red-red), C2(black-black), and C2(red-black), as the three probability parameters. As an example, the diagrams of the three independent two-point correlation functions for the anode microstructure of Fig. 3 are shown in Fig. 5, 6 and 7. These results are finally used to approximate three-point correlation functions using Eq. (17). 87 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Fig. 5. Two-point correlation functions (TPCF) for the three-phase composite. Here we show the 2-point correlation function for the red-red phases Fig. 6.Two-point correlation functions for the three-phase composite. Here we show the 2-point correlation function for the black-black phases 88 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Fig. 7. Two-point correlation functions for the three-phase composite. Here we show the 2-point correlation function for the red- black phases Three-point correlation functions were estimated by the use of Monte Carlo theory. In this approach a large number of vectors have been generated randomly within the 3D microstructure and the probability of the occurrence of desired events were calculated. The results show very good agreement between numerical Monte Carlo simulation of the real sample and the approximation method. The three-point correlation functions based on the two selected vectors (R12=(9.62)i+(9.62)j+(9.62)k (constant length), R13=xi (varying length with ݔൌ ȁܴଵଷ ȁ)) originating from a random point are calculated while the length of one and the angle between the two remain unchanged. In Fig. 8 the simulations are performed for the probability of occurrence of the three points (X1, X2 and X3) in phase 1 (red phase). The result is plotted against the length of one vector (R13) while satisfying the conditions above. 89 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Fig. 8. Three-point correlation functions for the three-phase composite. The corresponding 3point correlations are shown for the (red- red-red) phases (average error = 0.046). In Fig. 9 the simulations has been carried out to approximate probability of occurrence of the three points (X1, X2 and X3) in phase 3 (black phase). 90 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Fig. 9. Three-point correlation functions for the three-phase composite. The corresponding 3point correlations are shown for the (black- black-black) phases (average error = 0.049). The corresponding three-point correlation functions plotted in Fig. 10 and Fig. 11 show that the approximation based on the methodologies described here match fairly well the simulated correlations calculated from the three dimensional reconstructed microstructure. 91 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Fig. 10. Three-point correlation functions for the three-phase composite. The corresponding 3point correlations are shown for the (red- black-red) phases( average error = 0.052). Fig. 11. Three-point correlation functions for the three-phase composite. The corresponding 3point correlations are shown for the (red- black-blue) phases (average error = 0.066). 92 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials Three- Point correlation functions have been simulated and approximated for a variety of vector lengths from 10 to 400 unit lengths of the representative volume element (2000 units) which is adapted to the convergence range of three point correlation functions for reconstructed RVE. The average errors are reported in Fig. 12 for a large amount of data (more than 50000 three-point correlation functions) and various types of three-point correlation functions. We note that the error was calculated using the following equation where THPCF represents the three point correlation function, ݎݎݎܧൌ ห்ுிሺሻೞೠೌ ି்ுிሺሻೌೝೣೌ ห ்ுிሺሻೞೠೌ (64) Fig. 12. Average error for various types of three-point correlation functions III. 5.2. Approximation of three-point correlation functions for computer generated of hard-sphere microstructure Three-dimensional isotropic virtual samples with randomly distributed hard spheres are generated and used to calculate the statistical two-point correlation functions of high density spheres. In this study, Three-point correlation functions have been approximated using two-point correlation functions which are calculated using Monte-Carlo simulations. The sphere geometry 93 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials is defined by a radius and center of spheres. The center of the spheres has been allocated randomly inside a cubic volume. In the next step, two-point correlation functions are determined based on the probability of occurrence of the head and tail of each vector in a particular phase (spheres). The size of representative volume element (RVE) has been verified using convergence of the two-point correlation function [31]. Three- point correlation functions – THPCP (sphere-sphere-sphere) have been simulated and approximated using Eq. (17) for a large amount of vectors and the errors are reported via average length of these vectors in Fig. 13. In this work, we have studied more than 40000 three-point correlation functions with different magnitude of vector lengths from 10 to 400 of unit lengths of cubic RVE, we note that the chosen RVE dimensions was 1000 in this simulation; the error has been calculated using Eq. (64). Although we see a large dispersion of the error in Fig. 13, we must note that the average error has been found to be equal to eight percent. Fig. 13. Error of Three-point correlation between the simulated and approximation for the (sphere, sphere, sphere) phases 94 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials III. 6. Conclusion In the present study, a new formulation is proposed to obtain a relation between the higher and lower order correlation functions for heterogeneous materials. The approximation was developed using the conditional probability theory and the formulation is valid for multiphase heterogeneous materials. Comparison between the three-point correlation functions computed from a 3D reconstructed microstructure and from the proposed approximation shows satisfactory agreement. The compared results confirm the capability of our proposed approximation scheme to estimate N-point correlation functions using the information from the lower order (N-1)-point correlation functions. In future work, the authors would like to incorporate two-point cluster functions as suitable descriptor of microstructures [32] to find more precise approximation. An investigation of different type of weight functions needs also to be conducted. 95 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials III. References [1] Torquato S. Random heterogeneous materials : microstructure and macroscopic properties. New York: Springer; 2002. [2] Yeong CLY, Torquato S. Reconstructing random media. Physical Review E. 1998;57(1):495. [3] Garmestani H, Baniassadi M, Li DS, Fathi M, Ahzi S. Semi-inverse Monte Carlo reconstruction of two-phase heterogeneous material using two-point functions. IJTAMM. 2009;1(2):6. [4] Baniassadi M, Garmestani H, Li DS, Ahzi S, Khaleel M, Sun X. Three-phase solid oxide fuel cell anode microstructure realization using two-point correlation functions. Acta Materialia. 2011;59(1):30-43. [5] Li DS, Baniassadi M, Garmestani H, Ahzi S, Reda Taha MM, Ruch D. 3D Reconstruction of Carbon Nanotube Composite Microstructure Using Correlation Functions. Journal of Computational and Theoretical Nanoscience. 2010;7:1462-1468. [6] Kröner E. Modeling Small Deformation in Polycrystals. Amsterdam: Elsevier; 1986. [7] Beran MJ. Statistical continuum theories. New York: Interscience Publishers; 1968. [8] Phan-Thien N, Milton GW. New Bounds on the Effective Thermal Conductivity of N-Phase Materials. Proceedings of the Royal Society of London Series A, Mathematical and Physical Sciences. 1982;380(1779):333-348. [9] Dederichs PH, Zeller R. Variational treatment of the elastic constants of disordered materials. Zeitschrift für Physik A Hadrons and Nuclei. 1973;259(2):103-116. [10] Willis JR. Variational and related methods for the overall properties of composites. Adv Appl Mech. 1981;21:78. [11] McCoy JJ. On the calculation of bulk properties of heterogeneous materials. Appl Math. 1979;37:13. [12] Torquato S. Effective electrical conductivity of two-phase disordered composite media. J Appl Phys. 1985;58(10):8. [13] Torquato S. Effective stiffness tensor of composite media--I. Exact series expansions. Journal of the Mechanics and Physics of Solids. 1997;45(9):1421-1448. [14] Sen AK, Torquato S. Effective conductivity of anisotropic two-phase composite media. Physical Review B. 1989;39(7):4504. [15] Brown JWF. Solid mixture permittivities. The Journal of Chemical Physics. 1955;23 (8):4. [16] Fullwood DT, Adams BL, Kalidindi SR. A strong contrast homogenization formulation for multi-phase anisotropic materials. Journal of the Mechanics and Physics of Solids. 2008;56(6):2287-2297. 96 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials [17] Baniassadi M, Addiego F, Laachachi A, Ahzi S, Garmestani H, Hassouna F, et al. Using SAXS approach to estimate thermal conductivity of polystyrene/zirconia nanocomposite by exploiting strong contrast technique. Acta Materialia. 2011;59(7):2742-2748. [18] Debye P, Anderson HR. The correlations Function and Its Application. Journal of Applied Physics. 1957;28(6):4. [19] Glatter O, Kratky O. Small angle X-ray scattering. London: Academic; 1982. [20] Corson PB. Correlation functions for predicting properties of heterogeneous materials. II. Empirical construction of spatial correlation functions for two phase solids. J Applied Physics. 1974;45(b). [21] Adams BL, Morris PR, Wang TT, Willden KS, Wright SI. Description of orientation coherence in polycrystalline materials. Acta Metallurgica. 1987;35(12):2935-2946. [22] Garmestani H, Lin S, Adams BL. Statistical continuum theory for inelastic behavior of a two-phase medium. International Journal of Plasticity. 1998;14(8):719-731. [23] Garmestani H, Lin S, Adams BL, Ahzi S. Statistical continuum theory for large plastic deformation of polycrystalline materials. Journal of the Mechanics and Physics of Solids. 2001;49(3):589-607. [24] Gokhale AM, Tewari A, Garmestani H. Constraints on microstructural two-point correlation functions. Scripta Materialia. 2005;53(8):989-993. [25] Li DS, Saheli G, Khaleel M, Garmestani H. Microstructure optimization in fuel cell electrodes using materials design. CMC-Computers Materials & Continua. 2006; 4(1)::11. [26] Lin S, Garmestani H, Adams B. The evolution of probability functions in an inelasticly deforming two-phase medium. International Journal of Solids and Structures. 2000;37(3):423434. [27] Saheli G, Garmestani H, Adams BL. Microstructure design of a two phase composite using two-point correlation functions. Journal of Computer-Aided Materials Design. 2004;11(2):103115. [28] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. A new approximation for the three-point probability function. International Journal of Solids and Structures. 2009;46(21):3782-3787. [29] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. Effective conductivity in isotropic heterogeneous media using a strong-contrast statistical continuum theory. Journal of the Mechanics and Physics of Solids. 2009;57(1):76-86. [30] Baniassadi M, Laachachi A, Makradi A, Belouettar S, Ruch D, Muller R, et al. Statistical continuum theory for the effective conductivity of carbon nanotubes filled polymer composites. Thermochimica Acta. 2011;520(1-2):33-37. 97 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter III: New Approximate Solution for N-Point Correlation Functions for Heterogeneous Materials [31] Kanit T, Forest S, Galliet I, Mounoury V, Jeulin D. Determination of the size of the representative volume element for random composites: statistical and numerical approach. International Journal of Solids and Structures. 2003;40(13-14):3647-3679. [32] Jiao Y, Stillinger FH, Torquato S. A superior descriptor of random textures and its predictive capacity. Proceedings of the National Academy of Sciences. 2009;106(42):1763417639. 98 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Chapter IV 99 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions 100 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions: (Application to Three-Phase Solid Oxide Fuel Cell Anode Microstructure) 101 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions 102 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions IV. 1. Introduction There is a growing need for a mathematical linkage between microstructure and some of the important properties in materials [1-3]. Such a linkage can provide the means to design microstructures with optimum properties [4, 5] . Representation of microstructures based on npoint correlation functions has a long history going back to the discovery of x-ray scattering and the understanding that the result of scattering provides statistical information in the form of paircorrelation functions [6]. The community of small angle scattering has a rich history of developing structure functions to get information about the microstructure in the form of particle size and distributions. More recently, reconstruction methodologies based on two-point functions have evolved as a challenging problem [7]. Yeong and Torquato [8] introduced a stochastic optimization technique that enables one to generate realizations of heterogeneous materials from a prescribed set of correlation functions. They have provided examples of realizable two-point correlation functions and introduced a set of analytical basis functions for their representations. They have presented an exact mathematical formulation of the reconstruction algorithm. Jiao and co-workers [9] has also shown that the two-point functions alone cannot completely specify a two-phase heterogeneous material. As a result they have developed an efficient and isotropypreserving lattice-point algorithm to generate realizations of materials. Kröner [10, 11] and Beran [12] have developed statistical mathematical formulations to link correlation functions to properties in multiphase materials. Analytical techniques based on onepoint probability have a significant drawback in that important characteristics such as shape and geometry are not considered. Thus, to determine the contribution of shape and distribution effects, higher order probability functions must be developed. Corson [13-15] was among the first to attempt to incorporate shape and geometry effects by using an experimental form of the two- and three-point probability functions. In this formulation, Corson assumes that the probability functions are independent of orientation. In 1987, Adams introduced a set of two-point probability functions based on spherical harmonics [16] . The harmonics were used to account for orientation and point-to-point correlation in the microstructure. Garmestani and co-workers later extended the statistical continuum approach to both composites and polycrystalline materials using two-point functions [4, 17-24]. 103 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Torquato and co-workers have developed a procedure for the realization for a two-phase media using stochastic optimization techniques [9]. A stochastic reconstruction technique was used to generate random heterogeneous media with specified correlation functions. An optimization technique was applied to the two orthogonal directions and the autocorrelation functions for the generated two orthogonal sets are then calculated between these two sets [25]. The comparison is then used as a means for the reconstruction methodology by examining autocorrelation functions that display no appreciable short-range order [25]. Elsewhere, Torquato tried to develop a new methodology to reconstruct 3D random media by using the information from 2D sections [8]. In this methodology, a hybrid stochastic reconstruction technique was developed for the optimization of the lineal-path function and the two-point correlation functions during nucleation annealing technique [8]. In most of the numerical setup reviewed above, the simulated annealing methodology was used to reconstruct the random media while in our proposed reconstruction algorithm, the realization procedure is implemented using several optimization parameters which controls the overall reconstruction of heterogeneity. Heterogeneity can be observed in a wide range of natural and artificial substances [26]. Heterogeneity can be recognized in a material system by the local measurements of particle orientation and size distribution. Two mechanisms of nucleation and grain growth are examples of processing controlling the development of heterogeneities. Heterogeneity can take place during casting (as a result of nucleation) and crystallographic grain orientation distribution during grain growth [26]. It is clear that by using the grain growth as a function of time and morphology a certain level of heterogeneity can be developed. Inspired by the two mechanisms of nucleation and grain growth, we founded our proposed algorithm of heterogeneity reconstruction on three steps: generation, distribution and growth of cells. For illustration, Table 1 lists the technical equivalent of the three steps for two metallurgical processes. 104 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Table 1- Different steps of heterogeneity generation in two metallurgical processes Sim. Steps Cell generation Cell distribution Growth of cells Casting Nucleation Nucleation rate Grain growth Powder metallurgy Powder Packing Sintering Process (particles) In the present study, a cellular automata approach [27] was utilized to implement the kinetic growth of cells. The cellular automaton model used for kinetic growth of cells is similar to the Eden fractal model employed as an efficient tool to simulate some natural spatiotemporal phenomena [27]. It has been noted that the grain boundaries (boundary morphology) in heterogeneous materials look highly like fractalian geometries [28] . In this study, the Monte Carlo simulation is the primary modeling tool for the development of the realization methodology. Our Monte Carlo approaches rely on the definition of important parameters that affect nucleation and grain growth as parts of a kinetic growth model. The microstructure is then evolved and optimized by manipulating the prescribed parameters of the model through an objective function (OF) minimization for the statistical correlation function. In a previous work [29], we have developed a two-dimensional reconstruction methodology for two-phase composite materials. Under this methodology, random realizations are generated using statistical correlation functions based on the Monte Carlo simulation. The microstructures are then explored and modified by mimicking the natural processes of materials synthesis to predict the final realization. A kinetic growth model [27] was combined with a colony algorithm based on the Monte Carlo methodology. The present work concentrates on the 3D realizations as compared to our previous 2D-based work [29]. A three-phase anode microstructure of a solid oxide fuel cell is considered, which increased the order of the statistical representation. 105 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions IV. 2. Development of a Monte Carlo reconstruction methodology A new algorithm is presented based on Monte Carlo methodology for the reconstruction of microstructures using two-point statistical functions [30]. The realization process includes three steps: 1) generation, 2) distribution, and 3) growth of cells. Here, cells (or alternately grains or particles) refer to initial geometries assigned to each phase before the growth step. During the initial microstructure generation, basic cells are created from the random nucleation points and then the growth occurs as in crystalline grain growth in real materials [31, 32] . After distribution of nucleation points and assignment of basic cell geometries, the growth of cells starts according to the cellular automaton approach. The three steps of realization algorithm are repeated continuously to satisfy the optimization parameters until an adequately realistic microstructure is developed as compared statistically to the true microstructure. It is worth noting that in various steps of algorithm execution, several controlling parameters are developed that facilitate minimization of the objective function (OF) which is an index of successful realization. Before the 3D realization process, the microstructure of interest is reconstructed in 2D using the planar basic cells, as depicted schematically in Fig. 3. First of all, a sufficiently fine 2D grid is produced. Then for each phase and based on their associated volume fractions, a number of basic cells of arbitrary geometries representing the rough initial shape of existing phases are placed at some random nucleation points. Then these entities are allowed to grow in the next step. Fig. 3 illustrates the growth of three typical cells after being generated in several evolutionary stages. Afterwards, the procedures of basic cells distributions, examining the volume fractions and growth continue until the cells meet each other and the grid is filled. 106 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Stage (1) Stage (2) Stage (3) Stage (4) Stage (5) Stage (6) Stage (7) Stage (8) Fig. 3. Step-by-step growth of three typical cells in a 2D grid During simulations, it was observed that simulation results are insensitive to the rough initial geometry of the basic cell. Additionally, the computer code was designed such that overlapping of dissimilar basic cells is avoided. Furthermore, the distribution form of basic cells, or, more precisely, the closeness or clustering of similar basic cells is controlled by colony algorithm detailed in subsection 2.2. At the end of a 2D reconstruction, the objective function (OF) which is defined based on the three independent two-point correlation functions as OF (P 11 2 real ) ( P211 ) sim ( P212 )real ( P212 ) sim ( P222 ) real ( P222 ) sim 2 2 2 (1) where the subscripts real and sim indicate, respectively, the relevant values from the real and simulated microstructures, is evaluated numerically. For the subsequent reconstructions in Fig. 4, the optimization parameters such as shrinkage of basic cells, growth factors in the X- and Ydirection, parameters of the colony algorithm, rotation angles of basic cells and so on are varied such that the objective function of Eq.(1) is minimized. The procedure of reconstruction and optimization is repeated until the objective function takes a sufficiently close to zero value and 107 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions meanwhile less than the Monte Carlo (M-C) repeat error. This repeat error depends on the microstructure. 108 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Start Reconstruction Generation of cells Distribution of cells Growth of cells Compute the two-point correlations (Simulation) Optimization NO Compute the two-point correlations (From Experimental SEM Images) OF < M-C repeat error YES Reconstruction is done. Fig. 4. Basic steps in the realization algorithm (OF = objective function; MC=Monte Carlo) 109 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions IV. 2.1. 3D cell generation After successful reconstruction of 2D microstructure, certain simulation parameters including optimum growth factors in the X- and Y-direction, colony parameters and shrinkage factor are inherited by the 3D realization algorithm. For 3D generation of basic cells, the 2D cell can be extruded to form a 3D one based on the extrusion shape function: ZM hM x, y (2) where M refers to the phase of interest. Some typical simple forms of the function h are listed in Table 2. In this work, however, different but constant extrusion functions leading to cubic basic cells were used. Table 2- Typical mathematical forms for extrusion shape function shape Equation Ellipsoid r c 2 ((k (M )) 2 ZM Torus r (k (M ))2 R(M ) x 2 y 2 ZM Cube ZM x2 y2 ) a, b, c and k(߮) are constants for each phase a 2 i b2 rak (M ) 2 k(߮) and R(߮) are constant for each phase a and k(߮) are constant for each phase The cells are then allowed to undergo sort of a local shrinkage through a shrinkage function, S, defined as: S ª f1 x, y, z, E , p1 º « » « f 2 x, y, z , E , p2 » «¬ f3 x, y, z , E , p3 »¼ (3) Where x , y and z are cartesian coordinate inside the extrusion shape define by ZM . The mathematical forms of fi can be, for example, based on simple polynomial functions. The dependency of the transformation matrix on local Cartesian coordinates can be used to develop a 110 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions methodology for the 3D simulation. In the matrix above 0 E 1 is a random variable and pi is the optimization parameter satisfying 0 d pi d 1 . Each of the three components of the S vector takes values from the interval > 0,1@ , hence the term shrinkage function. In this work, the simple forms of fi E pi were selected to represent the shrinkage function that only scales down the initial basic cell. Local rotation of basic cells is another operation that can be performed to achieve optimum reconstruction. The three local rotation matrices are represented by the following: Qx 0 0 ª1 º «0 cos(T ( E , p )) sin(T ( E , p )) » x 4 x 4 » « «¬0 sin(T x ( E , p4 )) cos(T x ( E , p4 )) »¼ (4) Qy ªcos(T y ( E , p5 )) 0 sin(T y ( E , p5 )) º « » 0 1 0 « » « sin(T y ( E , p5 )) 0 cos(T y ( E , p5 )) » ¬ ¼ (5) Qz ªcos(T z ( E , p6 )) sin(T z ( E , p6 )) 0 º « sin(T ( E , p )) cos(T ( E , p )) 0 » z 6 z 6 « » «¬ 0 0 1 »¼ (6) where rotation angles, T x ,T y ,T z , depend on the random parameter, E , and the optimization factor, pi . The mathematical form of the rotation angles may be represented by T x, y , z 2SE pi (7) with E and pi satisfying the same conditions that they have in Eq. (3). 111 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions IV. 2.2. Cell distribution The subsection of cell distribution consists of two parts: distribution of cells’ centers and the relative positioning of identical cells. For the first part a random generator function was defined to calculate the Xc ,Yc, and Zc coordinates for the initial position of the cells in the Monte Carlo simulation. XC F (E , p7 ) (8) YC F ( E , p8 ) (9) ZC F ( E , p9 ) (10) where E and pi have the same definitions as in Eq. (3) or (7) and F can assume different forms depending on the expertise of the user. One possible form of dependency, for example, can be represented as F LE pi (11) where L is the dimensional length of the 3D grid in the X-, Y-, or Z-direction depending on the coordinate under consideration. Here, we have used the simple linear form of Eq.(11) , F=Lβ. For the second part, the analysis can be performed according to the desired model whether the overlapping or penetration of identical phases is allowed or not. In other words, the models can allow for coalescence of the particles (cells) using the colony algorithm resulting in agglomeration or can allow for the model to remain devoid of any agglomeration of particles using the contactless procedure. The flow diagram provided in Fig. 5 helps to better understand the distribution procedure. By generating a cell, if the simulated volume fraction of the corresponding phase is lower than the input volume fraction, then the center of the particle is relocated using Eq. (8),(9) and Eq.(10). If coalescence is allowed and the new cell overlaps with another similar one, the new cell is placed at the generated coordinates otherwise the next condition is checked. This new condition, discussed in detail in the following paragraph, controls the state of bundling or clustering of homogenous cells. If this last condition is not satisfied it 112 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions means that the location of the new cell is far from the regions of space occupied by similar particles and there is no similar entity in its neighborhood. On the contrary, if the conditional term is satisfied it means that the new cell is going to be located in the neighborhood of some other similar particle(s) and has an adverse effect on the minimization of the objective function. Therefore it should be rejected and a newer center location (coordinates) be generated. Different alternate coordinates are selected until this criterion is met. 113 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Fig. 5. Algorithm for cell distribution. 114 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions In the colony algorithm, one possible form of conditional statement is C E ! : exp :Ii where C E En (12) where 0 E 1 is a random variable, Ii is the volume fraction of the phase of interest, and n, : are two optimization parameters. Indeed, by changing the colony parameter n (power of the bundling distribution function) and Ω (input probability criterion), the clustering rate of cells can be monitored. On the right hand side of the above inequality, the proposed exponential form guarantees the stability of the algorithm. IV. 2. 3. Cell growth For implementation of the final step of the realization process, i.e. the cell growth, the well known cellular automaton approach (CA) is utilized [33]. The model has the potential for being used in computability theory (mathematical logic), physics, theoretical biology and microstructural reconstruction. The concept is explored on a grid of sites with each site capable of assuming a finite number of states. By assigning an initial state to each site of the grid, the following process can be generated (or the growth of the grid) according to the states of the neighboring sites along with a few growth rules which are usually similar for all sites. Concisely, a cellular automaton consists of a site space with a neighborhood relation, a set of states and a local transition function. The neighborhood relation considered in this work is of Neumann type (Fig. 6). In Neumann neighborhood for a 3D lattice, six adjacent sites on top, bottom, right, left, front and back of a central site are regarded as its neighbors whose states contribute to the determination of the subsequent growth state of the grid. 115 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Fig. 6. Von Neumann neighborhood relation in a 3D grid of sites Indeed, the growth argument applies only to the sites on the exterior layer of each grain. The transition or update function exploited to predict the directional growth can be either deterministic or stochastic and can be applied either synchronously or asynchronously. Here stochastic transition functions are chosen whereby the model is updated synchronously. Given the Neumann neighborhood, six directional transition functions are suggested corresponding to six directions/neighbors around each site. For every site, six conditional statements are checked in the following way: \ i E , pi pi E ! 0 (13) Here, i=1,2, …6 , β ( 0 E 1 ) is a random variable and pi is an optimization parameter (0≤pi≤1). If the condition (25) is satisfied, the growth continues in that direction by one site provided not already occupied. Then the procedure continues to examine the other directions and other sites on the exterior layer. The adopted kinetic growth model can be regarded as an extended version of the Eden fractal algorithm [27] used in biology and chemistry for describing the growth of bacterial colonies and deposition of materials. The current proposed growth methodology not only does not suffer from the instability issues but also it is capable of allowing growth in any preferential orientation which is useful when simulating anisotropic materials. This is because of the introduction of 116 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions optimization parameters in the present algorithm that allow control over the growth of all cells individually. IV. 3. Optimization of the statistical correlation functions In this work, material’s heterogeneity is represented by statistical distribution functions. A hypothetical statistical function is optimized and compared to the experimental statistical distribution functions. Stochastic optimization methodologies incorporate probabilistic (random) elements, either in the input data (the objective function, the constraints, etc.), or in the algorithm itself (through random parameters, etc.) or both [34]. By applying different optimization parameters to the simulations, a minimum error is achieved through minimization of the objective function that is constructed from the comparison of the two-point correlation function of the experimental and simulated images. A direct simple search optimization technique [34] was used for finding the minimum objective function. The optimization technique was applied in two stages: first step is used to extract the optimization factor for a 2D image (rotation factor in Z axis, shrinkage factor in the XY plane, colony factors, grain growth factors in the XY plane. In the second step, the optimization and other parameters (rotation about the X and Y axis, grain growth in the XZ or YZ plane) are used as initial input parameters for the 3D reconstruction. One of the main advantages of this technique is the decreased time of optimization. IV. 3.1. Percolation Percolation analysis is one of the most complicated and time-consuming computational methodologies in engineering. Percolation algorithms are used to exploit the continuity of objects and morphologies that are affected by certain properties and processes. Many different types of algorithms are presented to solve percolation problem, but some of them are not efficient and others are only useful for specific tasks [27]. As one of the important applications of percolation analysis in the realization and reconstruction methodologies for a heterogeneous microstructure, it is usually necessary to check percolation of the different phases during cell generation. In every step of the percolation, the continuity of cells is checked and the number of cells are recalculated for the entire cluster. The knowledge of the percolation cluster numbers 117 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions [35], allows other higher order statistical correlation cluster functions to be recalculated. The percolated phase can then be shown as one color for graphical representation. In this procedure, the boundary of the percolated regions is calculated for the simulated microstructures. A new Monte Carlo methodology for percolation is used to examine the extent of the clustering process in the heterogeneous material (Fig. 7). In this model, every cell is assigned a number (cluster number) that evolves through the cell growth process. A random node is selected and for every node a minimum cluster value of neighboring nodes will be assigned as shown in Fig. 6. This process is repeated until percolation is completed. This algorithm is very simple and it converges very quickly. The simulation processes for the percolation in this approach occur simultaneously for all cells and phases. Select Random Site in Grid Phase Network Calculate minimum value of cluster number of phases Allocate minimum cluster numbre to all neighbor’s site No Percolation completed? Yes Save Cluster Number Fig. 7. Algorithm of percolation based on the Monte Carlo methodology. 118 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions IV. 4. Three-phase solid oxide fuel cell anode microstructure Performance and properties of solid oxide fuel cell are determined by microstructure of components, just like most other engineering materials. For example, our previous studies revealed that the degradation mechanism in fuel cell anode depends on anode support microstructure [36]. It is very important to understand the relationship between microstructure and properties. Verification of modeling performance requires the capability of microstructure reconstruction. In this study we developed a new microstructure reconstruction method and applied on fuel cell anode. The anode microstructure of a solid oxide fuel cell (SOFCs) is presented in Fig. 8. Due to its functionality and operational environment requirements, SOFC anodes must have high catalytic activity for hydrogen oxidation, high electronic conductivity, and sufficient open porosity for unimpeded transport of gaseous reactants and products. SOFCs must also be stable at SOFC operating temperatures in reducing environments. The material of choice for long-term stability, chemical and mechanical compatibility with the YSZ electrolyte and low cost is Ni-YSZ cermet [37, 38] (see Fig. 8). The nickel serves as an electrochemical catalyst and electronic conductor. The YSZ provides mechanical strength, inhibits coarsening of the nickel particles, provides porosity for gaseous transport to the electrolyte, and yields an anode material with a coefficient of thermal expansion (CTE) that is similar to that of the YSZ electrolyte [39]. Within the porous structure of the anode material, nickel particles typically protrude from the YSZ substrate into the pores. The line at which the three phases (nickel, YSZ, and porosity) meet is referred to as the as the triple-phase boundary (TPB). In the active part of the anode, near the electrolyte, the active species converge for the electrochemical reaction at the TPB. Pathways must be provided to transport the species to the TPB in order for it to be active. Electrons are conducted through the nickel, the oxide ions are conducted within the YSZ and hydrogen gas flows through the porosity to the TPB. Some investigators have observed degradation in electrochemical performance during testing with Ni-YSZ anodes. In 1996, Iwata [40] fabricated a roughly 3-mm-thick anode by mixing and cold pressing 8-YSZ (8 mol% yttrium doped zirconium) and nickel-oxide (NiO) powders. An 8YSZ electrolyte was then deposited to the anode substrate by plasma spray (to ~200 µm thickness). Iwata performed duration tests of 211 hr at 927 °C, and 1015 hr at 1008 °C with cells 119 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions made of these anode/electrolyte layers. Both tests exhibited anode performance degradation apparently proportional to the duration and experienced temperatures. Clearly, the features of the anode microstructure can have a significant influence on its long-term electrochemical performance. The goal of this study is to develop a methodology to reconstruct the three-phase microstructure of the SOFC anode to facilitate the subsequent performance and degradation studies. Fig. 8. SEM micrographs of a three-phase Anode microstructure of Solid Oxide Fuel Cell (red: Nickel, blue: YSZ, Black: voids) IV. 5. Reconstruction of multiphase heterogeneous materials A three-phase anode microstructure of solid oxide fuel cells (SOFC) is considered for the reconstruction methodologies introduced above. The three constituents of this anode are Nickel, YSZ and voids (see Fig. 8). The methodology uses the two-point correlation functions calculated from the 2D SEM micrographs as an input to produce different 2D and 3D realizations of the microstructure with special attention to the percolation of the porous media. For illustration of the proposed methodology, Fig. 9a shows the phase distribution for a computer-generated three-phase composite (red, green, and white) with a 20% volume fraction for red phase, 20% for the green phase and 60% for the white phase. This microstructure is simulated to examine the reproducibility of the details of the microstructure represented by the two-point correlation functions (Fig. 9b). This is accomplished by using the same first-order 120 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions statistics and input simulation parameters for both the red and green phases in this trial realization. The results for the two-point correlation functions (P11 for red-red and P22 for greengreen) in Fig 9b show that the realization of the red and green phases are statistically indistinguishable Thus, we can conclude that the proposed methodology well controlled by the input parameters of the Monte Carlo algorithm. 0.25 RED PHASE 0.2 GREEN PHASE TPCF 0.15 0.1 0.05 0 0 200 400 600 800 1000 r (a) (b) Fig. 9. a) 2D simulation image of a three-phase microstructure (red, green and white) with 20% for red and green and 60% white. b) the corresponding 2D probability statistics (TPCF = 2-point correlation function) After the above illustration based on a numerical 2D microstructure, now we consider the real microstructure of the SOFC anode. Fig. 10 shows the 2D SEM micrograph of the anode microstructure and the corresponding two separate 2D realizations. The two-point correlations calculated from the SEM micrograph are used as initial input for the realizations in Fig. 10b and c. The corresponding two-point correlation functions plotted in Fig. 10d-e show that the realizations based on the methodologies described here match fairly well the original correlations calculated from the SEM micrograph. 121 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions (a) (b) TPCF P11 TPCF P11 P 12 0.4 0.25 0.09 0.35 EXPRIMENTAL RESULT 0.2 SIMULATION1 0.3 0.15 0.1 0.05 EXPERIMENTAL 0.08 SIMULATION1 0.07 0.25 0.2 0.15 0.06 0.05 0.04 0.03 0.02 0.1 0.01 0.05 0 EXPERIMENTAL Simulation1 Simulation2 SIMULATION2 TPCF(P12) TPCF(PHASE2) SIMULATION2 TPCF(PHASE1) (c) 0 0 200 400 r (d) 600 800 0 0 100 200 300 400 r 500 (e) 600 700 800 0 200 400 r 600 800 (f) Fig. 10. 2D realizations for an experimental image and comparison of the two-point correlation functions (TPCF). a) the 2D SEM micrograph for the anode microstructure (from Fig 1). b) realization-1 c) realization-2 d) the 2-point correlation function ( P211 or P11) for the red-phase, e) the 2-point correlation function ( P222 or P22) black phase f) 2-point correlation (black-red) function ( P212 or P12) The 2D reconstruction requires simulation parameters for cell generation, nucleation, and growth that are calculated during the optimization process to arrive at a final microstructure. These parameters along with the input two-point statistical functions are now used as input parameters for the 3D realizations. Fig. 11 presents four 2D sections through the depth of the 3D realizations for the input three-phase anode microstructure. For this realization we have used the 2D microstructure in Fig. 10b. Table 3 and 4 show the final simulation parameters for the 3-D reconstruction. 122 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions Table 3- RVE properties Pixel RVE Type Cube X Dimension 205 Y Dimension 154 Z Dimension 116 Table 4- Reconstruction parameters Cell Shrinkage Cube ª0.5º «0.5» « » ¬«0.5¼» S Rotation Distribution Colony F=Lβ disabled Cell growth P4=1 P5=1 \ i E ,.001 .001 E ! 0 P6=1 123 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions (a) (b) (d) (e) Fig. 11. 2D sections in the z-direction of the 3D image for the reconstructed microstructure. a) Layer close to the bottom surface , b) Layer in the middle area, c) Layer middle and top d) Layer close to the top surface The boundaries of the percolated regions of the porous phase for the 3D realization are identified for one of the 2D sections (Fig. 12a) and is shown in Fig. 11b. The three independent two-point correlation functions are compared with the original experimental SEM micrograph in Fig. 13ac. The results show that the methodologies adopted here can produce microstructures with the same statistical information based on two-point statistics in a 3D microstructure. The 3D microstructure is then plotted from the 2D sections and shown in Fig. 14-a, b. 124 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions (a) (b) Fig. 12. a) A 2D section of the 3D image for the reconstructed microstructure (black=porosity); b) the corresponding percolation of voids (porosity) showing the percolation clusters by similar colors other than white 125 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions 0.25 EXPRIMENTAL RESULT SIMULATION TPCF(PHASE1) 0.2 0.15 0.1 0.05 0 0 200 400 r 600 800 (a) 0.4 EXPERIMENTAL RESULT TPCF(PHASE2) 0.35 SIMULATION RESULT 0.3 0.25 0.2 0.15 0.1 0.05 0 0 200 400 r 600 800 (b) 0.09 TPCF(P1-2) 0.08 0.07 0.06 0.05 0.04 EXPERIMENTAL RESULT 0.03 SIMULATION RESULT 0.02 0.01 0 0 400 r 200 600 800 (c) Fig. 13. Comparison of the two-point correlation functions from the experimental and the 3D realizations. a) 2-point correlation functions P211 or P11 (red-red, phase 1), b) P222 or P22 for the porous phase (black-black, phase 2), c) P212 or P12 (black-red) 126 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions (a) (b) Fig. 14: a) Three-dimensional reconstructed image of the Anode microstructure. b) several sections through the depth of the 3D microstructure (red: Nickel, blue: YSZ, Black: voids). IV. 6. Conclusion A Monte Carlo methodology is used to reconstruct 3D microstructures of a three-phase anode structure in a solid oxide fuel cell (SOFC) from a 2D SEM micrograph. The methodology is based on two-point statistical functions as microstructure descriptors. The realization uses a hybrid stochastic reconstruction technique for the optimization of the two-point correlation functions during different 3D realizations. Colony and kinetic growth algorithms (cellular automata) are used to enable the realization process based on an optimization methodology. The main challenge in the 3D reconstruction is the degree of complexity due to the increased number of microstructure parameters as compared to 2D realization. Another important aspect of the new methodology is the establishment of a simple numerical routine to examine the percolation of desired phases relevant to fuel cell technology [35]. Comparison of the two-point correlation functions from different sections of the final 3D reconstructed microstructure with the initial real microstructure shows good agreement. This supports the capability of our proposed methodology to reconstruct 3D microstructure from an experimental 2D SEM result. 127 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions IV. References [1] Torquato S, Stell G. Microstructure of two-phase random media. II. The Mayer-Montroll and Kirkwood-Salsburg hierarchies. J Chem Phys. 1983;78:3062-3072. [2] Torquato S, Stell G. Microstructure of two-phase random media. I. The n-point probability functions. J Chem Phys. 1982;77:2071-2077. [3] Torquato S. Random heterogeneous materials : microstructure and macroscopic properties. New York: Springer; 2002. [4] Saheli G, Garmestani H, Adams BL. Microstructure Design of a Two Phase Composite Using Two-point Correlation Functions. international Journal of Computer Aided Design. 2004;11(2-3):103 - 115. [5] Adams BL, Lyon M, Henrie B, Kalidindi SR, Garmestani H. Spectral integration of microstructure and design Textures Of Materials, ed. 2002;408-4:487-492. [6] Debye P, Anderson HR, Brumberger H. Scattering by an Inhomogeneous Solid 2. The Correlations Function and Its Application. JOURNAL OF APPLIED PHYSICS. 1957;28(6):679-683. [7] Liang ZR, Fernandes CP, Magnani FS, Philippi PC. A reconstruction technique for threedimensional porous media using image analysis and Fourier transforms. Journal of Petroleum Science and Engineering. 1998;21:273–283. [8] Yeong CLY, Torquato S. Reconstructing random media. PHYSICAL REVIEW E. 1998;57(1):495-506. [9] Jiao Y, Stillinger FH, Torquato S. Modeling heterogeneous materials via two-point correlation functions: Basic principles. PHYSICAL REVIEW E. 2007;76. [10] Kröner E. Statistical Continuum Mechanics. Wien: Springer Verlag; 1972. [11] Kroner E. Bounds for effective elastic moduli of disordered materials,. J Mech Phys Solids 1977;25:137-155. [12] Beran MJ. Statistical continuum theories. New York,: Interscience Publishers; 1968. [13] Corson PB. Correlation Functions for Predicting Properties of Heterogeneous Materials. I. Experimental Measurement of Spatial Correlation Functions in Multiphase Solids. J Applied Physics. 1976;45(a):3159-3164. [14] Corson PB. Correlation functions for predicting properties of heterogeneous materials. II. Empirical construction of spatial correlation functions for two phase solids. J Applied Physics. 1974;45(b):3165. 128 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions [15] Corson PB. Correlation functions for predicting properties of heterogeneous materials. I. Experimental measurement of spatial correlation functions in multiphase solids. J Applied Physics. 1974;45(a): 3159. [16] Adams BL, Morris PR, Wang TT. Description of Orientation Coherence in Polycrystalline Materials. Acta Metall. 1987:2935-2946. [17] Lin S, Garmestani H, Adams B. The Evolution of Probability Functions in an Inelastically Deforming Two-Phase Medium. International journal of solids and structures. 2000;37(2):423. [18] Lin S, Adams BL, Garmestani H. Statistical continuum theory for inelastic behavior of twophase medium. International Journal of Plasticity. 1998;17(8):719-731 [19] Li DS, Saheli G, Khaleel M, Garmestani H. Quantitative Prediction of Effective Conductivity in Anisotropic Heterogeneous Media Using Two–point Correlation Functions Computational Materials Science. Computational Materials Science. 2006;38(1):45-50. [20] Li DS, Saheli G, Khaleel M, Garmestani H. Microstructure optimization in fuel cell electrodes using materials design. CMC-COMPUTERS MATERIALS & CONTINUA. 2006;4(1):31-42. [21] Li D, Garmestani D. Microstructure Sensitive Design and Quantitative Prediction of Effective Conductivity in Fuel Cell Design. In: Khan A, editor. The 13th International Symposium on Plasticity and Its Current Applications, Alaska.2007. [22] Gokhale AM, Tewari A, Garmestani H. Constraints on Microstructural Two-Point Correlation Functions. SCRIPTA MATERIALIA. 2005;53(8):989-993 [23] Garmestani H, Lin S, Adams B, Ahzi S. Statistical Continuum Theory for Texture Evolution of Polycrystals. Journal of the Mechanics and physics of Solids. 2001;49:589-607. [24] Garmestani H, Lin S, Adams B, Ahzi S. Statistical Continuum Mechanics Analysis of an Elastic Two-Isotropic-Phase Composite Material. Composites: Part B. 2000;31:39-46. [25] Sheehan N, Torquato S. Generating microstructures with specified correlation functions. JOURNAL OF APPLIED PHYSICS. 2001;89(11):53-60. [26] Riosa PR, Siciliano F, Sandimc HRZ, Plautd RL, Padilhad AF. Nucleation and Growth During Recrystallization. Materials Research. 2005;8(3):14. [27] Gould H, Tobochnik J, Wolfgang C. An Introduction to Computer Simulation Methods: Applications to Physical Systems (3rd Edition): Addison-Wesley Longman Publishing Co., Inc. Boston, MA, USA 2005. [28] Cao Q-z, On P-z. Fractal Interfaces in Heterogeneous Eden-like Growth. PHYSICAL REVIEW LETTERS. 1991;67(1):4. 129 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter IV: A New Monte Carlo Solution for Reconstruction of Heterogeneous Materials Using Two-Point Correlation Functions [29] Garmestani H, Baniassadi M, Li DS, Fathi M, Ahzi S. Semi-inverse Monte Carlo reconstruction of two-phase heterogeneous material using two-point functions. International Journal of Theoretical and Applied Multiscale Mechanics. 2009;1:134-149. [30] Kalos MH, Whitlock PA. Monte-Carlo Methods: WIiley-VCH Verlag GmbH & Co. KGaA; 2004. [31] Blikstein P, Tschiptschin AP. Monte Carlo Simulation of Grain Growth. Materials Research. 1999;2(3):4. [32] El-Khozondar R, El-Khozondar H, Gottstein G, Rollet A. Microstructural Simulation of Grain Growth in Two-phase Polycrystalline Materials. Egypt J Solids,. 2006;29(1):35-47. [33] Ilachinski A. Cellular automata Texte imprimÂe a discrete universe. Singapore: World Scientific; 2001. [34] Spall JC. Introduction to stochastic search and optimization: Wiley- Interscience; 2003. [35] Asiaei S, Khatibi AA, Baniasadi M, Safdari M. Effects of Carbon Nanotubes Geometrical Distribution on Electrical Percolation of Nanocomposites: A Comprehensive Approach. Journal of Reinforced Plastics and Composites. 2009:0731684408100701. [36] Liu W, Sun X, Pederson LR, Marina OA, Khaleel MA. Effect of nickel-phosphorus interactions on structural integrity of anode-supported solid oxide fuel cells. Journal of Power Sources. 2010;195(21):7140-7145. [37] Zhu WZ, Deevi SC. A Review of the Status of Anode Materials for Solid Oxide Fuel Cells. Materials and Science Engineering. 2003;A362:228-239. [38] Fuel Cell Handbook. Morgantown, West Virginia: U.S. Department of Energy; 2004. [39] Holtappels P, Vogt U, Graule. T. Ceramic Materials for Advanced Solid Oxide Fuel Cells. Advanced Engineering Materials. 2005;75(5):292-302. [40] Iwata T. Characterization of Ni-YSZ Anode Degradation for Substrate-Type Solid Oxide Fuel Cells. J Electrochemical Society. 1996;143(5). 130 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast Chapter V 131 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast 132 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast: (Application to Nanoclay Based Polymer Nanocomposites) 133 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast 134 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast V.1. Introduction The improvement of mechanical, thermal, gas barrier and fire resistance properties of organic polymer materials is a major concern, particularly in the domains of transportation, building construction, and electrical engineering. Polymer nanocomposites often exhibit physical and chemical properties dramatically different from the corresponding pure polymers. Numerous and recent studies have shown the interest of the use of clay nanoparticles (above all modified montmorillonites) as nanofillers for several polymers [1, 2]. The usual volume fraction of clay that has been used is in the range of 5 to 10 wt% organo-modified montmorillonite. The reasons are the high aspect ratio (more than 1000), the high surface area (more than 750 m2/g) and the high modulus of these lamellar nanoparticles (170 GPa). Depending upon the processing conditions and characteristics of both the polymer matrix and organoclay, the in-situ dispersion of organoclay inside the host polymer by melt blending can be more or less achieved, leading to intercalated or exfoliated nanocomposites. Recently most of the researches about layered silicates are focused especially on montmorillonites (MMT), as the reinforcing phase due to availability and versatility of these types of nano fillers [3]. Depending on the process conditions and on the polymer/nanofiller affinity, The layered silicates dispersed into the polymer matrix can be observed in different states of intercalation and/or exfoliation [4]. The best performances are commonly achieved with the exfoliated structures [5]. Besides that, the insertion of clay materials into a polymer matrix led to a significant decrease of the diffusion coefficient of various gases into the composites [3, 6]. Over the last few years, development of computer engineering, and of numerical methods for molecular dynamics simulations, allowed a detailed study of the structure of nano-objects and their thermomechanical properties,which is in general difficult or even impossible to study by other methods. Among different modeling techniques, Molecular Dynamics (MD) are now becoming standard means for the simulation of matter at the molecular scale [7]. Now-a-days MD is considered as the most realistic simulation technique as well as an alternative to 135 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast experiment in atomic scale science [8]. Recent studies verified that material properties acquired from MD simulations could be efficiently utilized in order to perform homogenization for effective thermal and mechanical properties of the nanocomposite material [9]. In MD simulations the structure is usually considered to be free of any impurities and defects, which leads to an upper bound of the experimental results for the modulus and the thermal conductivity [8, 10]. Particularly, molecular dynamic (MD) methods have been actively used to study montmorillonite lamellar structures and intercalate in the interlamellars space[11-13]. Several homogeneization methods have been used in the literature to predict effective properties of nanocomposite properties. For instance, the effective mechanical properties of such nanocomposites have been investigated using inclusion-based theories which call for the Eshelby solution for ellipsoidal inclusions in a homogeneous medium [14-22]. For example, the generalized Mori-Tanka model has been exploited to predict the effective elastic modulus of the starch/clay nanobiocomposites [23]. Similarly, the effective thermal conductivity of composites with ellipsoidal inclusions have been widely considered using various micromechanical models in the literature[24, 25]. In this work, we used a strong contrast [26-29] multiscale statistical method to predict the overall modulus and thermal conductivity of montmorillonite polymer based nanocomposites. To take into account the geometrical information on inclusions and their distribution in the matrix, a statistical continuum approach has been developed based on statistical correlation functions [29]. In this study two-point and three-point correlation functions have been taken into account to describe the microstructure. Using Monte Carlo simulation, two-point correlation functions of the realized nanostructures have been extracted and in a following step three point correlation functions have been estimated based on the previously determined two-point correlation functions [30]. From the two-point and three-point correlation functions, the effective thermal conductivity of the nanocomposites was calculated using a strong contrast expansion. To validate our proposed statistical approach, we conducted experimental tests to measure both the elastic and thermal properties for polyamide/MMT nanocomposites with 1, 3 and 5 wt% of nanoparticles. We then compared our simulate results to the experimental one. 136 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast V.2. Computer generated model In this research, Three-dimensional isotropic virtual samples with randomly oriented disks as mono layer nanoclays are generated and used to calculate the statistical two-point correlation functions of the realized model. These statistical correlation functions have been utilized as nanostructure descriptor to approximate the strong contrast solution for thermal and mechanical properties of nanocomposites. In this solution two-point and three-point correlation functions have been exploited as input function to solve the strong contrast equations for the effective thermal and elastic properties. In this study, three point correlation functions have been approximated using two-point correlation functions which are calculated using computer generated sample for nanocomposite nanostructures(see Fig. 1). Fig. 1. Two-point correlation function 137 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast An exfoliated nanoclay is created as the set of two parallel random surfaces with a specified distance equal the thickness of the nanoclay particles. The disk geometry is defined by a normal vector to the nanoparticles surface. The center of the disk has been allocated randomly inside a cubic volume. Then the normal vector is specified by random homogeneous functions given below which surveys uniformly on the surface of a sphere [31]. ­ ® ¯M T cos 2S v 1 (1) 2u 1 Where T >0, 2S > and M > 0, S @ are spherical coordinates as shown in Fig..2 and where u, v are random variables belonging to @0,1> M In this simulation, the soft-core algorithm is used to generate nanoclay particles which allows for penetration [32]. Thus a new plate of the nanoclay is randomly placed somewhere in the unit cell regardless of the ones already present. In other words, regions of space may be occupied by more than one nanoclay. However, the reason for using the soft core approach is its simplicity and its reduced computational time. Besides, by using this algorithm, one can simulate nearly every volume fraction of nanoclay in the composites. 138 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast Fig. 2. spherical coordinate of normal vector V.3. Thermal conductivity To evaluate the effective conductivity at macroscopically anisotropic two-phase composites, the strong-contrast expansion approach has been further improved by establishing an integral equation for the cavity intensity field [29]. The nth order tensorial expansions are expressed in terms of integrals over products of certain tensorial fields and a determinant of N-point statistical correlation functions which make the integrals convergent for the infinite volume limit. Owing to the procedure of solving the integral equations which produces absolutely convergent integrals, no additional renormalization analysis is needed. Another salient aspect of this expansion is that when truncated, at finite order, they give reasonably accurate estimates at rather all concentrations even though the contrast between the conductivities is high. Assuming isotropic properties of the PA matrix and nanoclay particles, the effective conductivity tensor Oe of the nanocomposite is determined using the strong-contrast formulation of the statistical continuum theory [27]: ^Oe O I ` .^Oe 2O I ` -1 R R ª S S 1, 2 S S 1 S S 2 º R 1 1 3 I O « 2 » M 1, 2 d2 R³ S S S S1 1 S1 2 E SR S1 1 «¬ »¼ 1 S S ª S S (1, 2,3) S 2 (1, 2) S 2 (2,3) º R 2 2 R 3 OR d E SR ³³ « S S » M 1, 2 M 2,3 d2d3 .... S S S «¬ S1 1 S1 2 S1 1 S1 2 S1 3 »¼ (2) Here, we have adopted the shorthand notation consisting in representing x1, x2, x3 by 1 and 2, 3 respectively. In Eq. (2), I is the second-order identity tensor, O is the reference conductivity, R M (1, 2) is a second-order tensor defined below, and E SR is the polarizability: R ESR OS O R OS d 1 O R (3) 139 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast The subscript R stands for the reference phase, which is chosen here to be the nanoclay phase, and the subscript/superscript S stands for the PA matrix. R The second order tensor M (1, 2) is defined by: R M (1, 2) 1 3tt I : OR x1 x2 3 where : is the total solid angle contained in a 3-dimensional sphere and t (4) ( x1 x2 ) s . S 1 , x1 x2 1 S2 1, 2 and S3 1, 2,3 are the probability functions that contain the microstructure information. s s The one-point probability function, S (1) , is the volume fraction of the nanoparticles. The twoS 1 point probability function, S (1, 2) , is calculated from the Monte Carlo simulation. The three-point S 2 probability function, S3 1, 2,3 , is calculated from the following analytical approximation [30]: s ª º S p 2,3 x1 x2 x1 x3 p p « S3 1, 2,3 # S 2 1,3 S 2 1 1, 2 » 2 p 1 3 + « x1 x2 x1 x3 » S1 1 x1 x2 x1 x3 ¬ ¼ p (5) Fig. 4 defines the variables used in this approximation in local coordinates. V.4. Mechanical model Exact perturbation series (weak-contrast expansions) are valid for two phase media with small variation of effective conductivity and elastic moduli of composites [29]. In general, strongcontrast expansions take a larger radius of convergence than weak-contrast expansion for the same reference properties. The statistical theory of strong contrast has been used to determine the effective stiffness tensor of macroscopically isotropic two-phase composites. In this approach, an integral equation for the strain field leads to an exact series expansions for the effective stiffness tensor of two-phase composite media. In this method, N-point correlation functions show up in the final equations that characterize the microstructure. The general term of the expansion for a reference phase q is written as follows [28]: 140 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast M p2 L( q ) : ª¬ L(eq ) º¼ f 1 M p I ¦ Bn( p ) (6) n 2 Where MP is the volume fraction of phase P and I is the fourth-order identity tensor, I ijkl º 1ª G G «G G » ik il il jk ¼ 2¬ (7) In Eq. (6) the tensor coefficients (Bn ) are the following integrals over products of the U Tensors and the Sn represent the N-point correlation functions for phase P: B2( p) ³H d 2U ( q) (1, 2) ª¬ S2( p) (1, 2) M p2 º¼ (8) n2 § 1 · B d 2...³ dnU ( q ) (1, 2) : U ( q ) (2,3) (1) ¨ ¨ M p ¸¸ ³ © ¹ ( p) U (q)(n 1, n)' n (1,..., n), n t 3, ( p) n n r x1 x2 , t r r (9) (10) In Eq. (6) the effective tensor L(eq ) is given by : L(eq ) ^C e ^ ` C ( q ) ` I A( q ) : ª¬Ce C ( q ) º¼ 1 (11) Where Ce is the effective stiffness tensor, Cq is the stiffness tensor of the reference phase and A(q) is a forth order constant tensor [28]. Here '(np) (1,..., n) is a position-dependent determinant that is calculated using N-point correlation function for a given phase p by: 141 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast ( p) S2 (1, 2) ( P) S3 (1, 2,3) ( p) S1 (2) ( p) S2 (2,3) 0 0 0 0 ' (np ) (1,..., n) (12) ( p) Sn1 (1, 2,..., n 1) ( p) Sn2 (2,3,..., n 1) ( p) ( p) Sn (1, 2,..., n) Sn1 (2,3,..., n) ( p) S2 (n 2, n 1) ( p) S3 (n 2, n 1) ( p) S1 ( n 1) ( p) S2 (n 1, n) The tensor U is calculated based on the position-dependent fourth-order H(r) and the related tensor for phase q, L (q): (q) U ijkl (r ) q) (q) L(ijmn H mnkl (r ) ­° ª ºG (d 2)Gq (q) ª¬ dK q 2(d 1)Gq º¼ ® « K pq P pq » ij H mmkl (r ) d ( K 2 G ) d q q °¯ ¬« ¼» (d 2)Gq (q) P pq H ijkl (r )` d ( K q 2Gq ) (13) Where d is space dimension and the tensor H( r ) is the symmetrized double gradient tensor [28] which is given below: (q) (r ) H ijkl 1 1 d 2: ª¬ dK q 2(d 1)Gq º¼ r ªD qG ijG kl d ªG ik G il G ilG jk º dD q ªG ij tk tl G kl ti t j º ¬ ¼ ¬ ¼ ¬ d (d D q ) (14) 2 ª¬G ik t j tl G il t j tk G ik ti tl G ij ti tk º¼ d (d 2)D q ti t j tk tl º¼ The constant tensor for phase q is expressed as: L( q ) ª º (d 2)Gq ª¬ dKq 2(d 1)Gq ¼º « k pq /h P pq / s » d ( Kq 2Gq ) ¬« ¼» (15) 142 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast Where d is space dimension and where k pq and P pq are introduced as bulk and shear modulus polarizabilities, K q and Gq are respectively the bulk modulus and the shear modulus of the reference phase and Λh and Λs are the fourth-order hydrostatic and shear projection tensors [29]. k pq and P pq are given by the following relations: K p Kq 4 K p Gq 3 k pq P G p Gq pq Gp Gq ª¬3K q / 2 4Gq / d º¼ (16) (17) K q 2Gq For macroscopically isotropic media, Eq. (6) can be simplified as [29]: ª k pq M p2 « /h «¬ keq f P pq º / s » M p I ¦ Bn( p ) Peq »¼ n 2 (18) In this work, the calculations have been performed for the first and second terms of B and other terms have been neglected because of the complexity of the calculations: ªk º P M p2 « pq / h pq / s » M p I B2( p ) B3( p ) Peq »¼ ¬« keq B2( p) B3( p) ³H d 2U ( q) (1, 2) ª¬ S2( p) (1, 2) M p2 º¼ § 1 · ( q) ( q) ( p) ¨¨ ¸¸ ³ d 2...³ dnU (1, 2) : U (2,3)'3 (1,...,3) © Mp ¹ ' 3( p ) (1,..., 3) S 2( p ) (1, 2) S1( p ) (2) S3( P ) (1, 2,3) S 2( p ) (2,3) (19) (20) (21) (22) 143 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast We recall that tor three-point correlation functions, we are using the analytical approximation in Eq. (5)(see Fig. 3). Fig. 3. Representation of vectors in spherical coordinate V.5. Experimental part V.5.1. Materials The polyamide (PA) resin (viscosity 35p, at 240°C) was supplied by Scientific Polymer. The PA density was 0.99 g.cm-3 (at 23°C). The filler was a commercial organo-modified montmorillonite, Cloisite 30B (OMMT) and was purchased from Southern Clay Co. The modifier was methyl bis-2-hydroxyethyl tallow ammonium and its concentration was 90 meq per 100 g of clay. This treatment leads to a good dispersion in the polar polymer matrix and allows preparing intercalated or exfoliated nanocomposites. The density of organo-modified montmorillonite was 1.98 g.cm-3 (at 23°C). 144 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast V.5.2. Nanocomposites preparation PA and OMMT were first dried at 80°C during 4 hours. PA-OMMT nanocomposites were then prepared by melt-mixing, the molten PA pellets and the OMMT at different weight fractions of clay, using a co-rotating twin-screw extruder (DSM Xplore), at 180°C for 5 min, with a rotation speed of 150 rpm. The investigated weight fractions of OMMT in PA nanocomposites were 0, 1, 3 and 5 wt%. V.5.3. Transmission electron microscopy Transmission electron microscopy (TEM) analyses of PA-OMMT nanocomposites were carried out using a LEO 922 apparatus at 200 kV. The ultrathin films (70 nm thick) were prepared with a LEICA EM FC6 cryo-ultramicrotome at 25 °C. V.5.4. Mechanical properties The evaluation of the mechanical properties of PA and its nanocomposites was carried out using a Dynamic Mechanical Analyzer (DMA 242C-Netzsch). Storage (E’) and loss (E’’) modulus were measured as a function of temperature (-175 °C to +70 °C) with a dynamic temperature ramp sweep at 2 °K.min-1. Measurements were performed using the single cantilever bending mode at a frequency of 1 Hz. The storage modulus is the elastic response to deformation, whereas the loss modulus is the dissipative response corresponding to the energy lost during the cyclic deformation of the material. All DMA samples were pressed and cut in the form of 9.7010.40 mm-long, 1.15-1.47 mm-thick and 4.95-5.9 mm-wide specimens. To check the reproducibility of the experimental data and to ensure their consistency, 3 specimens were tested for each formulation. V.5.5. Laser flash Thermal diffusivity and thermal conductivity of studied materials were measured by the laser flash method. This technique entails heating the front side of a small, usually disk-shaped planeparallel sample by a short (≤ 1ms) laser pulse. The temperature rise on the rear surface is measured versus time using an infrared detector. All samples were coated on both faces with a very thin layer of colloidal graphite. The thermal diffusivity a(T) values can then be converted to 145 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast thermal conductivity λ(T) by using the specific heat Cp(T) and bulk density ρ(T) of studied material according to: λ(T) = ρ(T) • Cp(T) • a(T) (23) The samples in the shape of discs, 12 mm in diameter and 1 mm in thickness were prepared by compression molding. The measurements were carried out from room temperature to 100 °C under an argon flow. Three samples were tested for each system and the uncertainty for the determination of thermal diffusivity was evaluated to ±3%. V.6. Results and discussion V.6.1. Thermal conductivity Thermal conductivity of neat PA decreases from its room temperature value of 0.127 W.m -1.K-1 with increasing temperature (see Fig. 4). In our calculation, the thermal conductivity of nanoclay particles has been estimated using a semi-inverse strong contrast approach [33] for the compressed powder sample at about 0.55(W.m-1.K-1). We have neglected the effect of temperature on this property. We analyzed the thermal conductivity for PA/nanoclay with 1, 3 and 5 wt%. The corresponding volume fractions are obtained from the two-point correlation functions (see Fig. 1) as 0.55%, 1.6% and 2.5%, respectively. Our results show that the addition of nanoclay leads to an increase in thermal conductivity of PA. Moreover, the higher the amount of nanoclay, the higher the thermal conductivity becomes. As shown in Fig. 4, the thermal conductivity of the PA-OMMT composites predicted using the strong contrast approach fits quite well with the experimental results. The simulated curves are not smooth because we used non-smooth experimental data of conductivity for pure polymer as a function of temperature (see Fig. 4). 146 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast Fig. 4. Comparison between experimental and simulation thermal conductivity of PA and its nanocomposites with OMMT V.6.2. Thermo-mechanical properties Since montmorillonite can be used for improving thermal stability, it is important that it does not dramatically deteriorate the mechanical properties (stiffness). To predict the elastic modulus of the composite, values of the elastic modulus of nanoclay found in the literature [11, 18] were used, Enanoclay= 176 Gpa [11, 18]. The elastic modulus of the PA matrix is shown in Fig. 5 as function of temperature. Fig. 5 shows the effect of the nanoclay on the mechanical properties (storage modulus E’) obtained by DMA measurements as well as those obtained using statistical continuum theory. At room temperature, PA exhibits a significant storage modulus (E’25°C = 550 MPa. The addition of 147 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast 1-5 wt.% nanoparticles did not have any impact on E’25°C. Below 0°C, the values of E’ of the composites containing 1 or 3 wt. % nanoclay are similar. However E’ increases by ~20% when 5wt.% clay is added to PA. It is found that E’ of the composites predicted by our simulations fit well with the experimental data for 1 wt. % . However, simulated values of E’, for the composites containing more than 1 wt. % nanoclay, are unfortunately higher than the experimental ones for the same composition. In the next, we will attempt to explain these discrepancies. Fig. 5. Experimental and simulated elastic modulus of two phases composite as a function of temperature T for neat PA and its composites with OMMT (1, 3 and 5 wt. %). TEM analyses of the PA-OMMT nanocomposites were performed in order to investigate the distribution and the dispersion of OMMT into the PA matrix. Fig. 6 shows two TEM images for two different nanofiller contents, 3 and 5wt%. The images show decreasing exfoliation state of nanofillers with increasing volume fraction of the fillers. 148 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast The discrepancies between the experimental and the theoretical results or the elastic modulus can be explained by the dispersion of the nanoclays in PA matrix. Indeed, the statistical continuum theory calculations assume that nanoclays in PA are in exfoliated state (Fig. 7). On the other hand, the experimental results showed that the nanoclays are in exfoliated state in the composites PA - 1wt. % OMMT and in both exfoliated-intercalated state in the composites PA – 3 wt. % OMMT and PA – 5 wt. % OMMT. Fig. 6.TEM micrographs of PA-3%OMMT and PA-5%OMMT nanocomposite Fig. 7. Polymer/clay nanocomposite morphologies 149 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast V.7. Conclusion In the present study, the effects of nanoclay additives on the effective mechanical and thermal properties of nanoclay based polymer composites have been investigated using both experimental and simulation analysis. In the present study, statistical continuum theory is used to predict the effective thermal conductivity and elastic modules of nanoclay based polymer composites. In this research, Monte Carlo simulations have been performed to find two-point probability functions of each phase. Two-point and three-point probability functions, as statistical descriptors of inclusions(fillers) distribution have been used to solve strong contrast homogenization for the effective thermal and mechanical properties of nanoclay based polymer composites. The predicted thermal conductivity results have shown satisfactory agreement with experimental data. However, the predicted effective elastic modulus results for high concentration of nanoclay overestimate the experimental data. This discrepancy is probably due to increasing intercalated structure of nanoclay for high nanofiller concentrations. 150 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast V. References [1] Gilman JW. Flammability and thermal stability studies of polymer layered-silicate (clay) nanocomposites. Applied Clay Science 1999;15:31. [2] Matadi R, Gueguen,O., Ahzi, S.,Gracio, J. , Muller, R. , Ruch, D. . Investigation of the Stiffness and Yield Behaviour of Melt-Intercalated Poly(methyl methacrylate)/Organoclay Nanocomposites: Characterisation and Modelling. Journal of Nanoscience and Nanotechnology 2010;10:2956. [3] Sinha Ray S, Okamoto M. Polymer/layered silicate nanocomposites: a review from preparation to processing. Progress in Polymer Science 2003;28:1539. [4] Vaia RA, Giannelis EP. Lattice Model of Polymer Melt Intercalation in OrganicallyModified Layered Silicates. Macromolecules 1997;30:7990. [5] Pollet E, Delcourt C, Alexandre M, Dubois P. Organic-Inorganic Nanohybrids Obtained by Sequential Copolymerization of ε-Caprolactone and L,L-Lactide from Activated Clay Surface. Macromolecular Chemistry and Physics 2004;205:2235. [6] Yoshitsugu Kojima, Arimitsu Usuki, Masaya Kawasumi, Akane Okada, Yoshiaki Fukushima, Toshio Kurauchi, Kamigaito O. Mechanical properties of nylon 6-clay hybrid. J.Mater. Res 1993; 8 5. [7] Haile JM. Molecular dynamics simulation : elementary methods. New York: Wiley, 1992. [8] Mortazavi B, Khatibi, A. A., Politis, C. Molecular Dynamics Investigation of Loading Rate Effects on Mechanical-Failure Behaviour of FCC Metals. Journal of Computational and Theoretical Nanoscience 2009;6:644. [9] Afaghi Khatibi A, Mortazavi B. A Study on the Nanoindentation Behaviour of Single Crystal Silicon Using Hybrid MD-FE Method. Advanced Materials Research 2008;32:3. [10] Komanduri R, Chandrasekaran N, Raff LM. Molecular dynamics (MD) simulation of uniaxial tension of some single-crystal cubic metals at nanolevel. International Journal of Mechanical Sciences 2001;43:2237. [11] Chen B, Evans JRG. Elastic moduli of clay platelets. Scripta Materialia 2006;54:1581. [12] Hackett E, Manias E, Giannelis EP. Computer Simulation Studies of PEO/Layer Silicate Nanocomposites. Chemistry of Materials 2000;12:2161. [13] Suter JL, Boek ES, Sprik M. Adsorption of a Sodium Ion on a Smectite Clay from Constrained Ab Initio Molecular Dynamics Simulations. The Journal of Physical Chemistry C 2008;112:18832. [14] Benveniste Y. A new approach to the application of Mori-Tanaka's theory in composite materials. Mechanics of Materials 1987;6:147. 151 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast [15] Eshelby JD. The Determination of the Elastic Field of an Ellipsoidal Inclusion, and Related Problems. Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences 1957;241:376. [16] Hori M, Nemat-Nasser S. Double-inclusion model and overall moduli of multi-phase composites. Mechanics of Materials 1993;14:189. [17] Hu GK, Weng GJ. The connections between the double-inclusion model and the Ponte Castaneda-Willis, Mori-Tanaka, and Kuster-Toksoz models. Mechanics of Materials 2000;32:495. [18] Luo J-J, Daniel IM. Characterization and modeling of mechanical behavior of polymer/clay nanocomposites. Composites Science and Technology 2003;63:1607. [19] Mori T, Tanaka K. Average stress in matrix and average elastic energy of materials with misfitting inclusions. Acta Metallurgica 1973;21:571. [20] Mura T. Micromechanics of defects in solids. Lancaster: Nijhoff, 1987. [21] Nemat-Nasser S, Hori M. Micromechanics : overall properties of heterogeneous materials. Amsterdam ; New York: Elsevier, 1999. [22] Sheng N, Boyce MC, Parks DM, Rutledge GC, Abes JI, Cohen RE. Multiscale micromechanical modeling of polymer/clay nanocomposites and the effective clay particle. Polymer 2004;45:487. [23] Chivrac F, Gueguen O, Pollet E, Ahzi S, Makradi A, Averous L. Micromechanical modeling and characterization of the effective properties in starch-based nano-biocomposites. Acta Biomaterialia 2008;4:1707. [24] Mercier S, Molinari A, El Mouden M. Thermal conductivity of composite material with coated inclusions: Applications to tetragonal array of spheroids. Journal of Applied Physics 2000;87:3511. [25] Milhans J, Ahzi S, Garmestani H, Khaleel MA, Sun X, Koeppel BJ. Modeling of the effective elastic and thermal properties of glass-ceramic solid oxide fuel cell seal materials. Materials & Design 2009;30:1667. [26] Pham DC, Torquato S. Strong-contrast expansions and approximations for the effective conductivity of isotropic multiphase composites. Journal of Applied Physics 2003;94:6591. [27] Sen AK, Torquato S. Effective conductivity of anisotropic two-phase composite media. Physical Review B 1989;39:4504. [28] Torquato S. Effective stiffness tensor of composite media--I. Exact series expansions. Journal of the Mechanics and Physics of Solids 1997;45:1421. [29] Torquato S. Random heterogeneous materials : microstructure and macroscopic properties. New York ; London: Springer, 2002. 152 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter V: Homogenization of Mechanical and Thermal Behavior of Nanocomposites Using Strong Contrast [30] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. A new approximation for the three-point probability function. International Journal of Solids and Structures 2009;46:3782. [31] Weisstein EW. Sphere Point Picking., vol. 2010: From MathWorld--A Wolfram Web Resource. [32] Ghazavizadeh A, Baniassadi M, Safdari M, Ataei AA, Ahzi S, Grácio J, Patlazhan S, Ruch D. Evaluating the effect of mechanical loading on the electrical percolation threshold of carbon nanotube reinforced polymers: A 3D Monte-Carlo study. Journal of Computational and Theoretical Nanoscience. [33] Baniassadi M, Addiego F, Laachachi A, Ahzi S, Garmestani H, Hassouna F, Makradi A, Toniazzo V, Ruch D. Using SAXS approach to estimate thermal conductivity of polystyrene/zirconia nanocomposite by exploiting strong contrast technique. Acta Materialia 2011;59:2742. 153 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM 154 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM Chapter VI 155 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM 156 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM 157 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM 158 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM VI.1. Introduction Heterogeneous media are abundantly found in a wide range of synthetic materials such as composites or natural materials such as living tissues. As a microstructural descriptor of heterogeneous materials, statistical correlation functions are among the most efficient ones. Mechanical, thermal, electrical and in general physical properties characterization of heterogeneous materials can be realized directly by means of such descriptors which are further known under the general designation of N-point correlation functions [1-5]. TPCFs are the basic statistical functions required to evaluate the effective/homogenized properties of micro/nanostructures. Homogenization approaches developed based on statistical continuum mechanics such as weak-contrast or strong-contrast approach are able to evaluate the effective properties through n-point correlation functions. Multi-phase heterogeneous materials with slight variation of properties are closely simulated by applying weak-contrast expansions. For the case of large differences between the properties of phases, strong-contrast technique is the suitable one for physical characterization purposes [1, 5, 6]. Micro/nanostructural reconstruction is another equally valuable application area of TPCFs besides physical properties characterization. Statistical continuum mechanics can be exploited to provide a robust alternative to X-ray tomography for the reconstruction of heterogeneous materials. Statistical reconstruction of heterogeneous media has become an intriguing inverse problem which has found application in various fields of engineering and biology to obtain 3D realization from the lower order correlation functions. Reconstruction using TPCFs is much simpler and less expensive than the other rival methods such as X-ray tomography or stitching technique [5, 710]. In this chapter, we extend our previously developed reconstruction methodology (In chapter 4) to 3D microstructure reconstruction based on two-point correlation functions and two-point cluster functions. Using a hybrid stochastic methodology for simulating the virtual microstructure, growth of the phases represented by different cells is controlled by optimizing parameters such as rotation, shrinkage, translation, distribution and growth rates of the cells. We used the finite element method (FEM) to predict the effective thermo-mechanical properties such as the elastic modulus and thermal conductivity of the reconstructed microstructure. We also used the strong 159 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM contrast statistical method, based on two-point and three-point correlation functions. The twopoint correlation functions are calculated from the computer generated microstructure. For the three-point correlation functions, we used two approximations, the existing approximation of Mikdam et al. [11] and our proposed new approximation detailed in Chapter 3. Comparison of the results from both approaches and FEM simulations show that our new approximation, for the tree-point correlation functions, gives a better agreement with the FEM results. VI.2. Reconstruction of heterogeneous materials using two-point cluster function (TPCCF) The previously developed algorithm based on Monte Carlo methodology for the reconstruction of microstructures using two-point correlation functions is now extended by the use of an additional microstructure descriptor, the two-point cluster functions. In the next, we briefly summarize the reconstruction methodology. The realization process includes three steps: 1) generation, 2) distribution, and 3) growth of cells. Here, cells (or alternately grains or particles) refer to initial geometries assigned to each phase before the growth step. During the initial microstructure generation, basic cells are created from the random nucleation points and then the growth occurs as in crystalline grain growth in real materials. After distribution of nucleation points and assignment of basic cell geometries, the growth of cells starts according to the cellular automaton approach. The three steps of realization algorithm are repeated continuously to satisfy the optimization parameters until an adequately realistic microstructure is developed as compared statistically to the true microstructure. It is worth noting that in various steps of algorithm execution; several controlling parameters are developed that facilitate minimization of the objective function (OF) which is an index of successful realization. This objective function is defined based on the three independent two-point correlation functions ( P2ij ) and two-point cluster functions ( P2c ii ) as follows: OF (P ) ij 2 real 2 3 ( P2ij ) sim ¦ ( P2ii )real ( P2ii ) sim ¦ ( P2c ii ) real ( P2cii ) sim 2 i 1 2 i 1 2 (1) where the subscripts real and sim indicate, respectively, the values from the real and reconstructed microstructures. The procedure of reconstruction and optimization is repeated until 160 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM the objective function takes a value that is of the same order as the Monte Carlo (M-C) repeat error. The material heterogeneity is represented by statistical two-point correlation functions and twopoint cluster functions. Hypothetical statistical functions are optimized and compared to the intial statistical functions of the sample microstructure. Stochastic optimization methodologies incorporate probabilistic (random) elements, either in the input data (the object function, the constraints, etc.), or in the algorithm itself (through random parameters, etc.) or both . By applying different optimization parameters to the simulations, a minimum error is achieved through minimization of the objective function (Eq. 1) that is constructed from the comparison of the two-point correlation function and two-point cluster functions of the sample and simulated (realization) microstructures. A direct simple search optimization technique was used for finding the minimum objective function. Fig. 1 depicts a schematic of the extended reconstruction algorithm. We recall that two-point cluster function is the probability of finding both beginning and ending points of a random vector in the same phase and same cluster. 161 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM Fig. 1. Basic steps in the realization algorithm (OF = objective function; MC=Monte Carlo) 162 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM VI.3.Statistical characterization of microstructures Exact perturbation expansions were used to predict the effective elastic modulus and thermal conductivity of two phase heterogeneous materials [5] . In this chapter , we have compared the effective properties of heterogeneous materials for two different approximation of three-point correlation functions. The first approximation (see Eq. 2 below) has been developed by Mikdam [11] and the second approximation has been proposed in the third chapter of this dissertation (see Eq. 3 below). ª º S p x , x x1 x2 x1 x3 p p « S x , x + S 2 x1 , x2 » 2 p 2 3 S3 x1 , x2 , x3 # « x1 x2 x1 x3 2 1 3 » S1 x1 x1 x2 x1 x3 ¬ ¼ p § x2 x3 p S3 x1 , x2 , x3 # ¨ ¨ x1 x2 x1 x3 x2 x3 © § x1 x3 ¨ ¨ x1 x2 x1 x3 x2 x3 © · § x1 x2 ¸S p x , x S p x , x + ¨ ¸ 2 1 3 2 1 2 ¨ x1 x2 x1 x3 x2 x3 ¹ © (2) · ¸S p x , x S p x , x ¸ 2 2 3 2 1 2 ¹ · ¸S p x , x S p x , x ¸ 2 1 3 2 2 3 ¹ (3) Here, S 2 x1 , x2 and S3 x1 , x2 , x3 are the two and three point correlation functions, p p respectively. The effective conductivity and elastic modulus of the composite material can be determined using the strong-contrast formulation of the statistical continuum theory considering the isotropic properties of the phases. VI.4. FEM characterization of multiphase heterogeneous materials The computer generated sample and the 3D reconstructed microstructure based on two-point correlation functions and two-point cluster functions are used for our FEM characterization. Finite Element simulations were carried out using ABAQUS/Standard (Version 6.10). Due to the extensive computationally time, only ten layers of the real specimens were included in the modeling. For the purpose of thermal modeling, the specimen was meshed using eight-node linear heat transfer brick (DC3D8-type) elements. For the mechanical modeling, the eight-node 163 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM linear brick, 3D stress with reduced integration (C3D8R-type) elements, were used. Each mesh element was assigned to the corresponding phase. Fig. 2. Finite element illustration and boundary condition of computer generated and reconstructed microstructure (Left: computer generated and right: reconstructed microstructures), for thermal and mechanical loading. In order to obtain the thermal conductivity of the specimen, constant heating surface heat flux was applied to a plane in the X direction while cooling surface heat flux equal to the cooling heating flux was applied in the opposite surface. In this way, steady state heat transfer criteria will be fully observed and by averaging the temperatures in each surface, the created temperature gradient as a function of distance in the specimen can be evaluated. The loading condition for the thermal and mechanical models are illustrated in Fig. 2. Using a one-dimensional form of the Fourier law, the thermal conductivity of the specimens was obtained. In order to obtain the elastic modulus of the specimen, a small strain was applied to the loading surface in its normal direction while the opposite surface was fixed only in its normal direction. By summation of the reaction forces in the fixed surface, the applied stress was calculated. Then, the elastic modulus of the specimen was obtained using Hook's law. 164 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM VI.5. Result and discussion The computer-generated three-phase sample is assumed to contain 10% volume fraction of red phase, 30% of the green phase and 60% of the black phase (see Table 1 and Fig. 3). This computer-generated three-phase sample is reconstructed and imported to the ABAQUS package for the FEM characterization. Fig. 3 shows 2D sections of three arbitrary layers taken from through-the-depth of the corresponding reconstructed microstructure in the ABAQUS package. These sections are arbitrarily chosen from the top, middle and bottom parts of the 3D reconstructed domain. The corresponding two-point correlation functions ( P211 or (P11) for red-red and P222 or (P22) for black-black and P212 or (P12) for red-black ) are calculated for both computer-generated and reconstructed microstructures shown in Fig. 4. As shown in this Fig., there is a good agreement between the two-point correlation functions of the reconstructed and computer-generated microstructures. The reconstruction process is performed based on the two-point correlation and the two-point cluster functions which had been extracted from computer generated microstructure. To check the validity of the reconstruction process, two-point cluster function for non-percolated phase (red-red) is calculated and shown in Fig. 5. Good agreement between the calculated two-point cluster functions for the two microstructures is obtained which strongly confirms the validity of the reconstruction process. Table 1.Phases properties Phase Number Phase 1 Phase 2 Phase 3 Volume Percent 60% 10% 30% Phase color black red blue 165 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM (a) Layer 25 (b) Layer 75 (c) Layer 125 Fig. 3. 2D arbitrary sections in the z-direction of the 3D reconstructed microstructure. a) Layer close to the bottom surface , b) Layer in the middle area, c) Layer close to the top surface 166 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM 0.12 0.1 TPCF (Microstructure 1) TPCF 0.08 TPCF (Microstructure 2) 0.06 0.04 0.02 0 0 100 200 300 r 400 500 (a) 0.7 0.6 TPCF (Microstructure 1) TPCF 0.5 TPCF (Microstructure 2) 0.4 0.3 0.2 0.1 0 0 100 r 200 300 400 500 (b) 0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 TPCF TPCF (Microstructure 1) TPCF (Microstructure 2) 0 100 200 r 300 400 500 (c) Fig. 4. a) Two-point correlation function (P11) for the red-phase, b) two-point correlation function (P22) for the black-phase c) two-point correlation function (P12) for the black-red phases for the computer generated and reconstructed microstructures. 167 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM 0.12 TPCCF (Microstructure 1) TPCCF (Microstructure 2) 0.1 TPCCF 0.08 0.06 0.04 0.02 0 0 50 100 150 200 250 r 300 350 400 450 500 Fig. 5. Two-point cluster function P2c 11 (TPCCF) for the red-phase, The boundaries of the percolated regions of different phases are identified for one of the phases (red-phase) in 2D section shown in (Fig. 6a) in which the phase percolation is less than the percolation threshold. The percolated aggregates have been recognized using different colors in Fig. 6b. In Fig. 6b and c, wide percolated clusters have been observed in the cut section images. As the other two phases are intrinsically percolated and their corresponding two-point correlation functions and two-point cluster functions are identical, there was no need to analyze the percolation in these phases. 168 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM (a) (b) (c) (d) Fig. 6. a) An arbitrary 2D section of the 3D reconstructed microstructure (black=porosity); b ,c ,d) the corresponding percolation of voids (porosity) showing the percolation clusters by similar colors other than white 169 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM In Fig. 7, the temperatures and Von Mises stress contours are represented for different cases for the local properties. The normalized properties (both thermal and elastic) for the three phases are taken (1,10,1) for case 1, (1,1,10) for case 2 ,(1,10,10) for case 3. Table 2 summarizes values of the phase properties assumed for different target samples. As it can be clearly seen from Fig. 7, there is fine agreement between these three cases with respect of the obtained fields of temperature and stress values. As a result, the differences in the obtained thermal conductivity and elastic modulus of the two microstructures (sample and reconstructed) are less than 1% error. Elastic properties and thermal conductivity of these microstructures have been compared using strong contrast (with existing and new approximation) and FEM analysis of 3-D reconstructed microstructures (cases). Fig. 7. Temperatures and Mises stress contours (Left : computer generated and right: reconstructed microstructure) The elastic properties for the three target samples are shown in Fig. 8 (left). The FEM results show very good agreement with strong contrast results which were obtained using statistical correlation functions of the microstructure along with both existing and new approximation for 170 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM the three-point correlation functions. Similarly, thermal conductivity for the samples has been calculated using FEM analysis of 3-D reconstructed microstructure and strong contrast technique. A small gap has been observed between the results obtained from the FEM and strong contrast method (Fig. 8 (right)). We note that the results with our new approximation are much closer to the FEM simulations. This shows the validity of the proposed statistical homogenization technique for three-phase heterogeneous materials and our approximation. Table 2. Phase’s properties Sample Sample 1 Sample 2 Sample 3 (Thermal conductivity and elastic modules) (Thermal conductivity and elastic modules) (Thermal conductivity and elastic modules) Phase 1 1 1 1 Phase 2 10 1 10 Phase 3 10 10 1 Fig 8. Elastic module of reconstructed microstructure using FEM and Strong contrast technique (first approximation and second approximation)(left), Thermal conductivity of reconstructed microstructure using FEM and strong contrast technique (first approximation and second approximation).(right) 171 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM VI.6. Conclusion A Monte Carlo methodology is developed to reconstruct 3D microstructures of a three-phase microstructure. Two-point correlation functions and two-point cluster functions are used as microstructure descriptors in the reconstruction procedure. Using a hybrid stochastic reconstruction technique, optimization of the function during different 3D realizations is performed repeatedly. The main challenge in the 3D reconstruction is incorporating two-point cluster function as complimentary statistical descriptor to perform reconstruction technique. Comparison of the two-point correlation functions from different sections of the final 3D reconstructed microstructure with the initial computer generated microstructure (sample microstructure) shows good agreement. In addition, we have shown that the thermo-mechanical properties of the generated and reconstructed microstructures are close by means of FEM simulations. This supports the capability of our proposed methodology to reconstruct 3D microstructure. We have also used the statistical homogenization technique to compute the effective elastic and thermal properties. The comparison of the results with those of the FEM simulations shows a fairly good agreement. This agreement between the two approaches suggest that the statistical approach is a reliable approach, particularly when the new approximation for the tree-point correlation functions is used. 172 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Chapter VI: Three-dimensional Reconstruction and Homogenization of Heterogeneous Materials Using Statistical Correlation Functions and FEM VI. References [1] Fullwood DT, Adams BL, Kalidindi SR. A strong contrast homogenization formulation for multi-phase anisotropic materials. Journal of the Mechanics and Physics of Solids. 2008;56(6):2287-2297. [2] Kröner E. Statistical Continuum Mechanics. Wien: Springer-Verlag, ; 1977. [3] Pham DC, Torquato S. Strong-contrast expansions and approximations for the effective conductivity of isotropic multiphase composites. Journal of Applied Physics. 2003;94(10):65916602. [4] Torquato S. Effective stiffness tensor of composite media--I. Exact series expansions. Journal of the Mechanics and Physics of Solids. 1997;45(9):1421-1448. [5] Torquato S. Random heterogeneous materials : microstructure and macroscopic properties. New York: Springer; 2002. [6] Wang M, Pan N. Elastic property of multiphase composites with random microstructures. Journal of Computational Physics. 2009;228(16):5978-5988. [7] Bochenek B, Pyrz R. Reconstruction of random microstructures--a stochastic optimization problem. Computational Materials Science. 2004;31(1-2):93-112. [8] Liang ZR, Fernandes CP, Magnani FS, Philippi PC. A reconstruction technique for threedimensional porous media using image analysis and Fourier transforms. Journal of Petroleum Science and Engineering. 1998;21(3-4):273-283. [9] Manwart C, Hilfer R. Reconstruction of random media using Monte Carlo methods. Physical Review E. 1999;59(5):5596. [10] Talukdar MS, Torsaeter O. Reconstruction of chalk pore networks from 2D backscatter electron micrographs using a simulated annealing technique. Journal of Petroleum Science and Engineering. 2002;33(4):265-282. [11] Mikdam A, Makradi A, Ahzi S, Garmestani H, Li DS, Remond Y. A new approximation for the three-point probability function. International Journal of Solids and Structures. 2009;46(21):3782-3787. 173 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Conclusion and Future Work 174 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Conclusion and Future Work Conclusion and Future Work 175 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Conclusion and Future Work 176 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Conclusion and Future Work In this study , statistical two point correlation functions as microstructure descriptors of heterogeneous media has been utilized to reconstruct the microstructures and homogenization thermal conductivity and elastic modulus of nanocomposites. different techniques such as Monte Carlo, SAXS data analysis and image processing of TEM/SEM images were exploited to calculated two point correlation functions. in future work , we are looking to extract statistical correlation functions using SAXS data analysis of anisotropic multiphase heterogeneous materials. Due to the complexity of calculating higher order correlation functions, in this research a new novel formulation has been proposed to obtain a relation between the higher and lower order correlation functions for heterogeneous materials using the conditional probability theory. This approximation is valid for N-Point correlation functions of multiphase heterogeneous materials . Comparison between the three-point correlation functions from the final 3D reconstructed microstructure and the approximate correlation functions shows satisfactory agreement. In future work, we would like to extend the weight functions of approximation to achieve optimum solution for N-Point correlation functions. Statistical two point correlation functions can be exploited to realize two or three dimensional microstructure of heterogamous materials. In this research work, a new Monte Carlo methodology is developed to reconstruct 3D microstructures of a N-phase microstructure. Two-point statistical functions are used as microstructure descriptors in the reconstruction procedure. Using a hybrid stochastic reconstruction technique, optimization of the two-point correlation functions during different 3D realizations is performed repeatedly. The main challenge in the 3D reconstruction is the possibility to incorporate other statistical descriptors similar two-point cluster function and lineal path functions as complimentary statistical descriptor to perform reconstruction technique. In the final step of this research project, statistical two-point correlation functions were used to homogenize thermal conductivity and elastic modulus of isotropic nanocomposite. For this, strong contrast homogenization approach was used. One advantage of this approach is to take into account the details of the microstructure which plays a very important role on the physical 177 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Conclusion and Future Work properties of materials. For this purpose, two-point probability functions was calculated using Monte Carlo technique to represent the distribution, shape and orientation of nanofillers (inclusions). In this approach , three point correlation functions have been estimated using the two point correlation functions then the effective thermal conductivity and elastic modulus of nanocomposite was calculated using strong contrast approach. It will be interesting to calculate four-point and five-point correlation function using the new developed approximation in future study for seeing the influence of the higher order correlation functions on the effective properties of heterogeneous materials. in future work , we would like to extend the numerical solution for calculating stiffness tensor and thermal conductivity tensor of multiphase anisotropic heterogonous media. 178 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix Appendix 179 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix 180 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix Appendix A: Verification of the Boundary Conditions for the Approximated Three-Point Probability Function: In this section, different limiting conditions (ݔଵ ՜ λǡ ǥ ሻare examined. A.1 First, we consider the case: ଵ ՜ λǣ ݈݅݉ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ௫భ ՜ஶ ሺܹଵଷ ൌ Ͳሻ כ ሺܹଷଷ ൌ Ǥͷሻ כ ܥଶ ሺݔଶ ǡ ݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔଵ ሻ ܥଶ ሺݔଶ ǡ ݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ሺܹଶଷ ൌ Ǥͷሻ כ ܥଵ ሺݔଵ ሻ ܥଵ ሺݔଶ ሻ మ ሺ௫య ǡ௫మ ሻכమ ሺ௫య ǡ௫భ ሻ భ ሺ௫య ሻ (A.1) ݈݅݉ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ௫భ ՜ஶ ሺǤ ͷሻ כ భ ሺ௫భ ሻכభ ሺ௫మ ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫మ ሻ ሺǤͷሻ כ మ ሺ௫మ ǡ௫య ሻכభ ሺ௫భ ሻכభ ሺ௫య ሻ భ ሺ௫య ሻ ݈݅݉௫భ ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ (A.2) (A.3) Similarly for ݔଶ ՜ λ: ݈݅݉௫మ ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଵ ሺݔଶ ሻ ܥ כଶ ሺݔଵ ǡ ݔଷ ሻ (A.4) And when ଷ ՜ λ ݈݅݉௫య ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଵ ሺݔଷ ሻ ܥ כଶ ሺݔଵ ǡ ݔଶ ሻ A.2 (A.5) Considering the case: ଵ ՜ ଶ 181 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix ݈݅݉௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ሺܹଵଷ ൌ ͲǤͷሻ כ ሺܹଷଷ ൌ Ͳሻ כ మ ሺ௫భ ǡ௫భ ሻכమ ሺ௫య ǡ௫మ ሻ భ ሺ௫భ ሻ ሺܹଶଷ ൌ Ǥͷሻ כ మ ሺ௫భ ǡ௫భ ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫భ ሻ మ ሺ௫య ǡ௫భ ሻכమ ሺ௫య ǡ௫భ ሻ భ ሺ௫య ሻ (A.6) ݈݅݉௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଶ ሺݔଶ ǡ ݔଷ ሻ (A.7) Similarly, we have: ݈݅݉௫మ ՜௫య ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଶ ሺݔଷ ǡ ݔଵ ሻ (A.8) ݈݅݉௫య ՜௫భ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଶ ሺݔଵ ǡ ݔଶ ሻ (A.9) A.3 Now, consider the case: ୧ ՜ λሺ ൌ ͳǡʹǡ͵ሻǣ ݈݅݉ ௫భ ՜ஶ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܹଵଷ כ ௫మ ՜ஶ ௫య ՜ஶ ܹଷଷ כ A.4 భ ሺ௫భ ሻכభ ሺ௫మ ሻכభ ሺ௫మ ሻכభ ሺ௫య ሻ ൌ భ ሺ௫య ሻ భ ሺ௫భ ሻכభ ሺ௫మ ሻכభ ሺ௫య ሻכభ ሺ௫భ ሻ భ ሺ௫భ ሻ ܹଶଷ כ భ ሺ௫భ ሻכభ ሺ௫మ ሻכభ ሺ௫మ ሻכభ ሺ௫య ሻ భ ሺ௫మ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ (A.10) Finally, let’s consider the case: ୧ ՜ ୨ ሺ ൌ ͳǡʹǡ͵ሻሺ ൌ ͳǡʹǡ͵ሻǣ ݈݅݉ோ՜ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܹଵଷ כ మ ሺ௫భ ǡ௫భ ሻכమ ሺ௫భ ǡ௫భ ሻ భ ሺ௫భ ሻ ܹଶଷ כ మ ሺ௫భ ǡ௫భ ሻכమ ሺ௫భ ǡ௫భ ሻ భ ሺ௫భ ሻ మ ሺ௫భ ǡ௫భ ሻכమ ሺ௫భ ǡ௫భ ሻ భ ሺ௫భ ሻ ܹଷଷ כ (A.11) And therefore, we have: ݈݅݉ ௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܥଵ ሺݔଵ ሻ (A.12) ௫య ՜௫మ This approximation is also valid for incompatible events (e.g. when x1 and x2 fall in two different phases) because in the limit, the terms containing correlation functions vanish to zero. For example in the case of incompatible event for x1 and x2 we have: ݈݅݉௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ܹଵଷ כ మ ሺ௫య ǡ௫మ ሻכమ ሺ௫య ǡ௫భ ሻ భ ሺ௫య ሻ మ ሺ௫భ ǡ௫మ ሻכమ ሺ௫య ǡ௫భ ሻ భ ሺ௫భ ሻ ܹଶଷ כ ൌͲ మ ሺ௫భ ǡ௫మ ሻכమ ሺ௫మ ǡ௫య ሻ భ ሺ௫మ ሻ ܹଷଷ כ (A.13) 182 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix Therefore we have: ݈݅݉௫భ ՜௫మ ܥଶ ሺݔଵ ǡ ݔଶ ሻ ൌ Ͳ (A.14) and finally: ݈݅݉௫భ ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ Ͳ (A.15) 183 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix Appendix B: Verification of the Boundary Conditions for the Approximated Four-Point Probability Function: In this section, different limiting conditions (ଵ ՜ λǡ ǥ ሻare probed. B.1 First, we consider the case: ݔଵ ՜ λǣ ݈݅݉ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ௫భ ՜ஶ ሺܹଵସ ൌ Ͳሻ ܥ כଵ ሺݔଵ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܹଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܹଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଶ ሺݔଷ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ ସସ ܥ כଵ ሺݔସ ሻ כ భ ሺ௫భ ሻכమ ሺ௫ర ǡ௫మ ሻכభ ሺ௫భ ሻכమ ሺ௫ర ǡ௫య ሻכయ ሺ௫ర ǡ௫మ ǡ௫య ሻ భ ሺ௫భ ሻכభ ሺ௫ర ሻכమ ሺ௫ర ǡ௫మ ሻכమ ሺ௫ర ǡ௫య ሻ (B.1) Using boundary conditions of Eq. (54 and 55), We have: ܹଵସ ൌ Ͳ (B.2) ܹଶସ ܹଷସ ܹସସ ൌ ͳ (B.3) By substituting weight functions and simplifying Eq. (B.1), we get: ݈݅݉௫భ ՜ஶ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ܥଵ ሺݔଵ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ (B.4) Similarly, we have: ݈݅݉௫మ ՜ஶ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ܥଵ ሺݔଶ ሻ ܥ כଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ (B.5) ݈݅݉௫య ՜ஶ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ܥଵ ሺݔଷ ሻ ܥ כଷ ሺݔଵ ǡ ݔଷ ǡ ݔସ ሻ (B.6) ݈݅݉௫ర ՜ஶ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ܥଵ ሺݔସ ሻ ܥ כଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ (B.7) 184 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix B.2 Considering the case: ଵ ՜ ଶ ݈݅݉ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ௫భ ՜௫మ ܹସସ ܥ כଵ ሺݔସ ሻ כ ܹଵସ ܥ כଵ ሺݔଶ ሻ כ ܥଷ ሺݔଶ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥ כଷ ሺݔଶ ǡ ݔଶ ǡ ݔସ ሻ ܥଶ ሺݔଶ ǡ ݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܹଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଷ ሺݔଶ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଷ ሺݔଶ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥଶ ሺݔଶ ǡ ݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ܹଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔଶ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ య ሺ௫ర ǡ௫మ ǡ௫మ ሻכయ ሺ௫ర ǡ௫మ ǡ௫య ሻכయ ሺ௫ర ǡ௫మ ǡ௫య ሻ మ ሺ௫ర ǡ௫మ ሻכమ ሺ௫ర ǡ௫మ ሻכమ ሺ௫ర ǡ௫య ሻ ݈݅݉௫భ ՜௫మ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ=ܥଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ (B.8) (B.9) Similarly, we have: ݈݅݉௫మ ՜௫య ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ=ܥଷ ሺݔଵ ǡ ݔଷ ǡ ݔସ ሻ (B.10) ݈݅݉௫య ՜௫ర ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ=ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔସ ሻ (B.11) ݈݅݉௫ర ՜௫భ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ=ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ (B.12) This approximation is also valid for incompatible events (e.g. when x1 and x2 fall in two different phases) because in the limit, the terms containing correlation functions vanish to zero. For example in the case of incompatible event for x1 and x2 we have: ݈݅݉ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ௫భ ՜௫మ ଵସ ܥ כଵ ሺݔଵ ሻ כ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ܥ כଷ ሺݔଵ ǡ ݔଷ ǡ ݔସ ሻ ܥ כଷ ሺݔଵ ǡ ݔଶ ǡ ݔସ ሻ ܥଶ ሺݔଵ ǡ ݔଶ ሻ ܥ כଶ ሺݔଵ ǡ ݔଷ ሻ ܥ כଶ ሺݔଵ ǡ ݔସ ሻ ଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଷ ሺݔଶ ǡ ݔଵ ǡ ݔସ ሻ ܥ כଷ ሺݔଶ ǡ ݔଵ ǡ ݔଷ ሻ ܥ כଷ ሺݔଶ ǡ ݔଷ ǡ ݔସ ሻ ܥଶ ሺݔଵ ǡ ݔଶ ሻ ܥ כଶ ሺݔଶ ǡ ݔଷ ሻ ܥ כଶ ሺݔଶ ǡ ݔସ ሻ ଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଷ ሺݔଷ ǡ ݔଶ ǡ ݔସ ሻ ܥ כଷ ሺݔଷ ǡ ݔଶ ǡ ݔଵ ሻ ܥ כଷ ሺݔଷ ǡ ݔଵ ǡ ݔସ ሻ ܥଶ ሺݔଷ ǡ ݔଵ ሻ ܥ כଶ ሺݔଷ ǡ ݔଶ ሻ ܥ כଶ ሺݔଷ ǡ ݔସ ሻ 185 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Appendix ܹସସ ܥ כଵ ሺݔସ ሻ כ య ሺ௫ర ǡ௫భ ǡ௫మ ሻכయ ሺ௫ర ǡ௫భ ǡ௫య ሻכయ ሺ௫ర ǡ௫మ ǡ௫య ሻ మ ሺ௫ర ǡ௫భ ሻכమ ሺ௫ర ǡ௫మ ሻכమ ሺ௫ర ǡ௫య ሻ (B.13) By substituting three point correlations function using Eq. (18) in Eq. (39) and calculating limit, we have: ݈݅݉௫భ ՜௫మ ܥସ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ǡ ݔସ ሻ ൌ ܥଷ ሺݔସ ǡ ݔଵ ǡ ݔଶ ሻ B.3 (B.14) Now, consider the case: ݔ ՜ λሺ݅ ൌ ͳǡʹǡ͵ǡͶሻǣ ݈݅݉௫భ ՜ஶ ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ௫మ ՜ஶ ௫య ՜ஶ ௫ర ՜ஶ ܹଵସ ܥ כଵ ሺݔଵ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଵ ሻ ܹଶସ ܥ כଵ ሺݔଶ ሻ כ ܥଵ ሺݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܹଷସ ܥ כଵ ሺݔଷ ሻ כ ܥଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଷ ሻ ܥଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܹସସ ܥ כଵ ሺݔସ ሻ כ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔସ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻ ܥ כଵ ሺݔସ ሻ ൌ ܥଵ ሺݔଵ ሻ ܥ כଵ ሺݔଶ ሻ ܥ כଵ ሺݔଷ ሻܥଵ ሺݔସ ሻ (B.15) B.4 Finally, let’s consider the case: ݔ ՜ ݔ ሺ݅ ൌ ͳǡʹǡ͵ǡͶሻܽ݊݀ሺ݆ ൌ ͳǡʹǡ͵ǡͶሻǣ ݈݅݉ோ՜ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ݈݅݉௫భ ՜௫మ ܥସ ൫ݔଵ ǡ ݔଶ ǡ ݔଷǡ ݔସ ൯ ൌ ݈݅݉௫ర ՜௫మ ܥଷ ሺݔଵ ǡ ݔଶ ǡ ݔଷ ሻ ൌ ௫య ՜௫మ ௫ర ՜௫మ ݈݅݉௫ర ՜௫మ ܥଶ ሺݔଶ ǡ ݔସ ሻ ൌ ܥଵ ሺݔଶ ሻ ௫య ՜௫మ (B.16) 186 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 187 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 188 Majid BANIASSADI, University of Strasbourg, Strasbourg, 2011 Development of a multiscale approach for the characterization and modelling of heterogeneous materials : Application to polymer nanocomposites In this research, a comprehensive study has been performed in the use of two-point correlation functions for reconstruction and homogenization in nano-composite materials. Two-point correlation functions are measured from different techniques such as microscopy (SEM or TEM), scattering and Monte Carlo simulations. Higher order correlation functions must be calculated or measured to increase the precision of the statistical continuum approach. To achieve this aim a new approximation methodology is utilized to obtain N-point correlation functions for multiphase heterogeneous materials. The two-point functions from different techniques have been measured and exploited to reconstruct microstructure of heterogeneous media. A new Monte Carlo methodology is developed as a mean for three-dimensional (3D) reconstruction, of the microstructure of heterogeneous materials, based on two-point statistical functions. The salient feature of the presented reconstruction methodology is the ability to realize the 3D microstructure from its 2D SEM image for a three-phase medium extendable to n-phase media. Three dimensional reconstruction of heterogeneous media have been exploited to predict percolation of heterogamous materials. Finally, Statistical continuum theory is used to predict the effective thermal conductivity and elastic modulus of polymer composites. Two-point and three-point probability functions as statistical descriptor of inclusions have been exploited to solve strong contrast homogenization for effective thermal and mechanical properties of nanoclay based polymer composites. To validate our modeling approach, we conducted several experimental measurements for nanoclay/polymer of composite. Comparison of our predictions with the experimental results led to a good agreement. this allows us to conclude that the proposed methodlogy is accurate. Développement d'une approche multi-échelle pour la caractérisation et la modélisation des matériaux hétérogènes: Application aux polymères nanocomposites Dans ce projet de recherche, une étude approfondie a été effectuée en utilisant des fonctions de corrélation à deux points pour la reconstruction et l'homogénéisation de nano-matériaux composites. Ces fonctions de corrélation à deux points sont mesurées à l’aide de différentes techniques telles que la microscopie (MEB ou TEM), la diffraction des rayons X et les simulations de type Monte Carlo. Des fonctions de corrélation d'ordre supérieur doivent être calculées ou mesurées si l’on souhaite augmenter la précision de l'approche de la méthode statistique. Pour atteindre cet objectif, une nouvelle méthodologie d’approximation est utilisée pour obtenir des fonctions de corrélation à N-points pour les matériaux hétérogènes multiphasiques. Les fonctions à deux points ont été mesurées à partir de techniques différentes et exploitées pour reconstituer la microstructure des milieux hétérogènes. Dans la suite de ce travail, une nouvelle méthodologie Monte Carlo est développée comme outil pour la reconstruction en trois dimensions (3D) de la microstructure des matériaux hétérogènes, fondée sur les fonctions statistiques à deux points (TPFC). La caractéristique principale de la méthodologie de reconstruction présentée ici est la capacité de réaliser la microstructure 3D à partir de son image SEM 2D pour un milieu à trois phases extensible à n-phases. Trois reconstructions tridimensionnelles des milieux hétérogènes ont été exploitées pour prédire la percolation des matériaux hétérogames. Enfin, la théorie de la statistique des milieux continus est utilisée pour prédire la conductivité thermique effective ainsi que le module d'élasticité des composites polymères. Des fonctions de probabilité à deux points et à trois points, utilisées comme descripteurs statistiques des inclusions (renfort) ont été exploitées pour résoudre le problème de l’homogénéisation à fort contraste des propriétés thermiques et mécaniques effectives des matériaux composites à base de polymère/ nano-argile. Pour valider notre approche de modélisation, nous avons mené plusieurs mesures expérimentales pour les composites polymère/nanoargile. La comparaison de nos prédictions avec les résultats expérimentaux ont conduit à un bon accord ce qui confirme la qualité et la précision de la méthodologie proposée.