[pt] Um dos principais objetivos da engenharia de petróleo é desenvolver e aplicar técnicas capazes de aumentar a produtividade de poços de petróleo, incluindo a estimulação do poço por operações de fraturamento hidráulico. Estudos sobre a propagação de fraturas podem ser feitos analiticamente para algumas situações simplificadas envolvendo homogeneidade, isotropia e condições de contorno simples do meio geológico, ou pela aplicação de métodos numéricos, como o método dos elementos finitos, para casos mais complexos. A presente pesquisa apresenta análise numérica de fraturamento hidráulico utilizando o método estendido dos elementos finitos (XFEM), em conjunto com o modelo constitutivo de dano da Zona Coesiva (MZC). No método estendido dos elementos finitos a geometria da fratura se torna independente da malha, permitindo a propagação da fratura através do domínio, dispensando sucessivas gerações de malha necessárias na aplicação do método convencional dos elementos finitos. Os resultados numéricos obtidos foram comparados com soluções analíticas assintóticas no caso limite em que o regime da propagação é dominado pela rigidez da rocha, obtendo uma boa concordância. Adicionalmente, foram investigados os efeitos de diferentes parâmetros do fluido de injeção e as características de propagação da fratura quando a interface entre diferentes camadas geológicas é inclinada, mostrando dependência do ângulo de inclinação, das propriedades do material e das tensões in-situ. / [en] One of the main objectives of petroleum engineering is to develop and apply techniques capable of increasing the productivity of oil wells, including the stimulation of well by hydraulic fracturing operations. Studies on the propagation of fractures can be done analytically for some simplified situations involving homogeneity, isotropy and simple boundary conditions of the geological medium, or by the application of numerical methods, such as the finite element method, for more complex cases. The present research presents a numerical analysis of hydraulic fracturing using the extended finite element method (XFEM), in conjunction with the damage constitutive model of Cohesive Zone (MZC). In the extended finite element method the fracture geometry becomes independent of the mesh, allowing the propagation of the fracture through the domain without successive mesh generations as necessary in the conventional finite element method. The computed numerical results were compared with asymptotic analytical solutions in the limit case in which the propagation regime is dominated by the rigidity of the rock with good compatibility. In addition, this study investigates the effects of different parameters of the injection fluid and the fracture propagation characteristics when the interface between different geological layers is inclined, shows dependency between the angle of inclination with the properties of the material and the in-situ stresses.

Tato disertační práce vychází z výzkumu v rámci francouzsko-českého programu doktorátu pod dvojím vedením na pracovišti Institut français de mécanique avancée v Clermont-Ferrand a na Ústavu fyziky materiálu AV v Brně. Úvodní výzkumný úkol na brněnském pracovišti se zabýval numerickou analýzou pole napětí v okolí čela trhliny v tenké kovové fólii. Zvláštní pozornost byla zaměřena na vliv speciálního typu singularity v průsečíku čela trhliny s volným povrchem. Těžiště disertační práce spočívá v numerickém modelování a stochastické analýze problémů šíření trhlin se složitou geometrií v dvojrozměrném prostoru. Při analýze těchto problémů se dříve zřídka používaly numerické metody, a to z důvodu vysoké náročnosti na výpočtový čas. V této disertaci je ukázáno, že aplikací moderních metod numerické mechaniky a vhodných technik v analýze spolehlivosti lze tyto problémy řešit s pomocí numerických metod i na PC. Ve spolehlivostní analýze byla využita lineární aproximační metoda FORM. Pro rychlost šíření trhlin se vycházelo z Parisova-Erdoganova vztahu. Pro parametry tohoto vztahu byl použit dvourozměrný statistický model, který postihuje vysokou citlivost na korelaci obou parametrů. Mechanická odezva byla počítána rozšířenou metodou konečných prvků (XFEM), která eliminuje výpočetní náročnost a numerický šum související se změnou sítě v klasické metodě konečných prvků. Prostřednictvím přímé diferenciace bylo odvozeno několik vztahů pro derivace funkce odezvy, čímž se dosáhlo lepší numerické stability a konvergence spolehlivostní analýzy a výrazného zkrácení doby výpočtu. Problém zatížení s proměnou amplitudou byl řešen aplikací transformace zatížení metodou PREFFAS. Využití distribuce výpočtů v síti PC umožnilo další zrychlení analýzy.

La prédiction de l’état de fissuration est un enjeu crucial pour l’analyse des structures en béton armé, qui nécessite le recours à la modélisation et à la simulation numérique. Le calcul par éléments finis des structures en béton armé pose au moins deux problèmes majeurs :d’une part il existe peu de modèles permettant de traiter à la fois l’initiation, la propagation et l’ouverture des fissures, d’autre part le diamètre généralement faible des armatures métalliques par rapport aux dimensions des structures étudiées nécessite des maillages particulièrement fins. On propose donc des solutions à ces deux problématiques basées sur l’utilisation de fonctions de niveau (level set). L’endommagement et la fissuration du béton sont modélisés à l’aide de l’approche TLS (Thick Level Set). Cette méthode,développée en tant que méthode de régularisation des modèles d’endommagement locaux, utilise une level set afin d’introduire une longueur caractéristique. Cela permet de rendre aisée la localisation de la position des fissures, et donc d’enrichir le champ de déplacement parla méthode des éléments finis étendus (X-FEM) afin de modéliser l’ouverture des macro-fissures. Concernant la modélisation des armatures, une nouvelle approche multidimensionnelle est proposée. Une représentation volumique des armatures par la méthode X-FEM est utilisée dans les zones d’intérêt afin d’obtenir des résultats précis tout en simplifiant la procédure de maillage, tandis qu’une représentation linéique est utilisée dans le reste de la structure afin de réduire le nombre de degrés de liberté du calcul. La méthode de transition développée ici permet d’assurer la cohérence des résultats obtenus / Prediction of cracking is a key point for the analysis ofreinforced concrete structures, which requires the use of Modeling and numerical simulation. The analysis of reinforced concrete structures using the finite element method raises two issues: on one hand, few models areable to deal with the initiation, the propagation and the opening of cracks, on the other hand the diameter of thereinforcements which is usually small compared to the dimensions of the structures necessitates very fine meshes. Some solutions to these two problematics areproposed, based on the use of level set functions.Damage and cracking of concrete are modeled using theThick Level Set (TLS) approach. This method,developped as a mean to regularize local damagemodels, uses a level set to introduce a characteristic length. It makes the location of the cracks easy, whichallows to enrich the displacement field with the eXtendedFinite Element Method (X-FEM) in order to model the macro-cracks opening. Concerning the modeling of thereinforcements, a new multidimensionnal approach isproposed. A volumic representation of the reinforcements with the X-FEM method is used in the zones of interest to get accurate results while simplifying the meshing process, whereas a lineic representation isused elsewhere to decrease the number of degrees of freedom. The developed transition method insures the consistency of the results.

Composite laminates are being more employed as fundamental structures due to its low weight and high stiffness. An example of this innovation is the primary structures of modern aircraft, which are lighter than the material formerly used. To predict the material response as load gradually increases can be quite demanding due to composite\'s complex failure mechanism. Hence superior computational models should be further investigated to precisely predict the mechanical behavior of composite media. This dissertation addresses an extended finite element procedure based on the layerwise displacement theory to simulate purely mode I delamination failure in composite laminates. The present model has the potential to perform structural analyzes in a pre-delaminated structure and also considering progressive failure. The type of element to be employed at the discretion of the model is the rectangular 4-node iso-parametric homogeneous element whose displacement field is approximated based in the layerwise theory. There are four types of degrees of freedom, one displacement in each direction, and one degree of freedom associated to the strong discontinuity. Numerical examples already solved in the bibliography are suggested in this dissertation, which demonstrate the potential of the model developed to accurately simulate pure mode I delamination in case of the investigation here is further elaborated. In addition, one possibility of future development of this dissertation is the modeling of fracture mode I without the need to discretize the cohesive planes as realized in traditional Cohesive Zone Methods. / Compósitos laminados estão sendo mais empregados como estruturas fundamentais devido ao seu baixo peso e alta rigidez. Um exemplo dessa inovação são as estruturas primárias das aeronaves modernas, que são mais leves do que as os materiais empregados antigamente. Prever a resposta do material à medida que a carga aumenta gradualmente pode ser difícil devido ao complexo mecanismo de falha dos compósitos. Portanto, modelos computacionais mais refinados devem ser investigados a fim de se prever um comportamento mecânico mais preciso. Esta dissertação aborda um procedimento de elementos finitos estendido baseado na teoria de deslocamento layerwise para simular falhas de delaminação modo I em laminados compósitos. O modelo abordado tem potencial para realizar análises em uma estrutura prédelaminada além de falha progressiva. O tipo de elemento a ser empregado na discrição do modelo é o isoparamétrico, homogêneo de 4 nós, retangular, e o campo de deslocamento é baseado na teoria layerwise. Existem quatro tipos de graus de liberdade, um deslocamento em cada direção, e um grau de liberdade associado à forte. Sugere-se nesse trabalho, exemplos, que são comparados com a bibliografia, e que apontam que o modelo desenvolvido nesta dissertação tem o potencial de simular o fenômeno de delaminação em modo I com acurácia, caso o estudo nessa dissertação seja estendido. Além disso, uma possibilidade de desenvolvimento futuro desse trabalho é a modelagem da fratura modo I sem a necessidade de discretizar os planos coesivos entre as lâminas, como realizado em métodos coesivos tradicionais.

Motivados pelas atuais tendências e suportados pelo grande interesse de indústrias do segmento aeronáutico, estudos e desenvolvimentos vêm sendo conduzidos na área de análise estrutural de materiais compósitos. Todavia, mesmo havendo várias contribuições científicas e tecnológicas nesta área, este assunto continua sendo um campo aberto e bastante promissor para novas pesquisas, devido a sua extensa complexidade e imediata aplicação. A ausência de um modelo capaz de projetar com elevada precisão uma estrutura aeronáutica com presença de escalonamento fabricada em material compósito, que pode sofrer modo de falha translaminar, motivou o presente trabalho. É sabido que o método de elementos finitos estendidos (XFEM - eXtendend Finite Element Method, do Inglês) vem sendo usado de maneira robusta para análise de propagação de trincas em elementos estruturais tridimensionais isotrópicos durante os últimos anos, mas não em compósitos. De forma a contribuir com a pequena quantidade de trabalhos científicos referentes a métodos XFEM 3D para análise de estruturas fabricadas em materiais compósitos não convencionais, como estruturas com escalonamento de camadas e laminados espessos, é apresentada uma nova metodologia implementada como uma ferramenta computacional para analisar quase estaticamente este tipo de estrutura. O modelo é baseado no aprimoramento do \"Método da Seção de Ouro\" que é aplicado em conjunto com uma versão aprimorada do critério de falha de Puck, permitindo assim definir com precisão e baixo custo computacional a iniciação e direção de uma trinca. Esta informação é utilizada para iniciar uma rotina baseada em XFEM, que é usada para o enriquecimento dos elementos finitos que vão falhando progressivamente durante a análise. A nova metodologia (implementada computacionalmente) apresenta convergência uma ordem de grandeza maior quando comparada com o algoritmo tradicional, sendo aproximadamente 20 vezes mais eficiente em termos computacionais. O modelo é ainda avaliado quanto a seus resultados em comparação com dados provenientes de ensaios experimentais, demonstrando uma boa convergência entre as previsões computacionais e os resultados obtidos em laboratório. / Supported by current trends and by the great interest of aeronautic industries, studies and developments have been made in the field of high performance composite materials. Nonetheless, even with the scientific and technological contributions, the matter is still a field wide open and promising for new research due to its high complexity and immediate application. The absence of a model capable of universally reproducing mechanical behavior of composite materials tapered structures, which can suffer translaminar failure mode, motivated the present work. It is well known that the eXtended Finite Elements Method (XFEM) has been used robustly for analysis of crack propagation in isotropic tri-dimensional structural elements lately but not for composites. In order to contribute with the scares amount of available works on 3D XFEM application on non-conventional composite material structures, such as tapered structures and thick laminates, a new methodology is presented as a computational tool for quasi-static analysis of this type of component. The model derives from \"Golden Section Method\" that is applied along with an enhanced version of Puck\'s failure criterion, which allows a low computational cost and high precision estimation of crack initiation and direction of propagation. This information is used to trigger an XFEM based routine that is applied for enriching the elements progressively during analysis. The new methodology (computationally implemented) has a convergence rate one order of magnitude greater than traditional implementation, roughly 20 times more efficient in terms of computational processing. Finally, to assure robustness, the model is validated against standardized and specifically developed experiments, showing good convergence between numerical predictions and results obtained in the laboratory.

[pt] Um elevado número de reservatórios de hidrocarbonetos é naturalmente fraturado. Quando sujeitos a estimulação hidráulica, as fraturas naturais podem influenciar a propagação da fratura hidráulica, que pode tomar uma forma geométrica complexa, criando redes de fraturas no reservatório. De forma a melhor entender e simular tais fenômenos, um elemento baseado no Método dos Elementos Finitos Estendidos (XFEM) é proposto. A formulação do elemento inclui interseção e cruzamento entre fraturas, atrito entre as faces das fraturas, comportamento acoplado entre deslocamentos, poro-pressões e pressões do fluido da fratura, absorção de fluído da fratura para o meio poroso (leak-off) e a eventual perda de pressão nas faces da fratura (filter cake). Os fundamentos teóricos e os aspectos relevantes da implementação são apresentados. Um conjunto de análises é realizado de forma a validar em separado as diferentes funcionalidades do elemento implementado. Finalmente, os resultados de quatro aplicações práticas são analisados e discutidos: dois conjuntos de ensaios de laboratório de interseção de fratura, propagação de fratura hidráulica num modelo sintético multi-fraturado e percolação na fundação fraturada de uma barragem. Conclui-se que o código implementado fornece previsões muito boas do comportamento acoplado do meio fraturado e tem capacidade de simular corretamente a interação entre fraturas hidráulicas e naturais. Pode também verificar-se que o comportamento hidráulico dos modelos e a propagação e interseção de fraturas são muito influenciados por parâmetros tais como o diferencial de tensões in-situ, ângulo entre fraturas, a abertura hidráulica das fraturas e a condutividade transversal das faces da fratura. / [en] A large number of hydrocarbon reservoirs are naturally fractured. When subjected to hydraulic fracturing treatments, the natural fractures may influence the propagation of the hydraulic fracture, which can grow in a complicated manner creating complex fracture networks in the reservoir. In order to better understand and simulate such phenomena an element based on the eXtended Finite Element Method is proposed. The element formulation comprises fracture intersection and crossing, fracture frictional behaviour, fully coupled behaviour between displacements, pore and fracture fluid pressure, leak-off from the fracture to the surrounding medium and the eventual loss of pressure due to filter cake. The theoretical background and implementation aspects are presented. A set of analyses is performed in order to validate different features of the implemented element. Finally, the results of four practical applications are analysed and discussed: two laboratory hydraulic fracture tests, hydraulic fracture propagation in a multifractured synthetic model and percolation through a dam fractured foundation. It is concluded that the implemented code provides very good predictions of the coupled fluid-rock fracture behaviour and is capable of correctly simulating the interaction between hydraulic and natural fractures. Moreover, it is shown that the hydraulic behaviour of the models and the intersection between fractures are very sensible to parameters such as differential in-situ stresses, angle between fractures, initial hydraulic aperture and fracture face transversal conductivity.

Ce mémoire de thèse à été réalisée dans le cadre d'une collaboration scientifique avec "La Manufacture Française des Pneumatiques Michelin". Il porte sur l'analyse mathématique et numérique de la convergence et de la stabilité de formulations mixtes ou hybrides de problèmes d'optimisation sous contrainte avec la méthode des multiplicateurs de Lagrange et dans le cadre de la méthode éléments finis étendus (XFEM). Tout d'abord, nous essayons de démontrer la stabilité de la discrétisation X-FEM pour le problème d'élasticité linéaire incompressible en statique. Le deuxième axe, qui représente le contenu principal de la thèse est dédié à l'étude de certaines méthodes de multiplicateur de Lagrange stabilisées. La particularité de ces méthodes est que la stabilité du multiplicateur est assurée par l'ajout de termes supplémentaires dans la formulation faible. Dans ce contexte, nous commençons par l'étude de la méthode de stabilisation de Barbosa-Hughes appliquée au problème de contact unilatéral sans frottement avec XFEM cut-off. Ensuite, nous construisons une nouvelle méthode basée sur des techniques de projections locales pour stabiliser un problème de Dirichlet dans le cadre de X-FEM et une approche de type domaine fictif. Nous faisons aussi une étude comparative entre la stabilisation avec la technique de projection locale et la stabilisation de Barbosa-Hughes. Enfin, nous appliquons cette nouvelle méthode de stabilisation aux problèmes de contact unilatéral en élastostatique avec frottement de Tresca dans le cadre de X-FEM. / This Ph.D. thesis was done in collaboration with "La Manufacture Française des Pneumatiques Michelin". It concerns the mathematical and numerical analysis of convergence and stability of mixed or hybrid formulation of constrained optimization problem with Lagrange multiplier method in the framework of the eXtended Finite Element Method (XFEM). First we try to prove the stability of the X-FEM discretization for incompressible elastostatic problem by ensured a LBB condition. The second axis, which present the main content of the thesis, is dedicated to the use of some stabilized Lagrange multiplier methods. The particularity of these stabilized methods is that the stability of the multiplier is provided by adding supplementary terms in the weak formulation. In this context, we study the Barbosa-Hughes stabilization technique applied to the frictionless unilateral contact problem with XFEM-cut-off. Then we present a new consistent method based on local projections for the stabilization of a Dirichlet condition in the framework of extended finite element method with a fictitious domain approach. Moreover we make comparative study between the local projection stabilization and the Barbosa-Hughes stabilization. Finally we use the local projection stabilization to approximate the two-dimensional linear elastostatics unilateral contact problem with Tresca frictional in the framework of the eXtended Finite Element Method X-FEM.

Composite laminates are being more employed as fundamental structures due to its low weight and high stiffness. An example of this innovation is the primary structures of modern aircraft, which are lighter than the material formerly used. To predict the material response as load gradually increases can be quite demanding due to composite\'s complex failure mechanism. Hence superior computational models should be further investigated to precisely predict the mechanical behavior of composite media. This dissertation addresses an extended finite element procedure based on the layerwise displacement theory to simulate purely mode I delamination failure in composite laminates. The present model has the potential to perform structural analyzes in a pre-delaminated structure and also considering progressive failure. The type of element to be employed at the discretion of the model is the rectangular 4-node iso-parametric homogeneous element whose displacement field is approximated based in the layerwise theory. There are four types of degrees of freedom, one displacement in each direction, and one degree of freedom associated to the strong discontinuity. Numerical examples already solved in the bibliography are suggested in this dissertation, which demonstrate the potential of the model developed to accurately simulate pure mode I delamination in case of the investigation here is further elaborated. In addition, one possibility of future development of this dissertation is the modeling of fracture mode I without the need to discretize the cohesive planes as realized in traditional Cohesive Zone Methods. / Compósitos laminados estão sendo mais empregados como estruturas fundamentais devido ao seu baixo peso e alta rigidez. Um exemplo dessa inovação são as estruturas primárias das aeronaves modernas, que são mais leves do que as os materiais empregados antigamente. Prever a resposta do material à medida que a carga aumenta gradualmente pode ser difícil devido ao complexo mecanismo de falha dos compósitos. Portanto, modelos computacionais mais refinados devem ser investigados a fim de se prever um comportamento mecânico mais preciso. Esta dissertação aborda um procedimento de elementos finitos estendido baseado na teoria de deslocamento layerwise para simular falhas de delaminação modo I em laminados compósitos. O modelo abordado tem potencial para realizar análises em uma estrutura prédelaminada além de falha progressiva. O tipo de elemento a ser empregado na discrição do modelo é o isoparamétrico, homogêneo de 4 nós, retangular, e o campo de deslocamento é baseado na teoria layerwise. Existem quatro tipos de graus de liberdade, um deslocamento em cada direção, e um grau de liberdade associado à forte. Sugere-se nesse trabalho, exemplos, que são comparados com a bibliografia, e que apontam que o modelo desenvolvido nesta dissertação tem o potencial de simular o fenômeno de delaminação em modo I com acurácia, caso o estudo nessa dissertação seja estendido. Além disso, uma possibilidade de desenvolvimento futuro desse trabalho é a modelagem da fratura modo I sem a necessidade de discretizar os planos coesivos entre as lâminas, como realizado em métodos coesivos tradicionais.

Dans un contexte de vulnérabilité militaire des sous-marins, les ingénieurs et chercheurs doivent être capables de prédire le comportement des structures fissurées. Ainsi, la modélisation de la transition des changements de modes de propagation de fissure (cisaillement-traction et inversement) des métaux sous sollicitations extrêmes devient un outil incontournable ou essentiel. Des critères tridimensionnels de direction de propagation de fissure développés pour une rupture par cisaillement ou par ouverture sont exposés. Des formules de direction de propagation semi-analytiques et analytiques, fonctions des facteurs d’intensité des contraintes et du coefficient de Poisson, sont ainsi proposées. L’interprétation de ces formules laisse envisager la prise en compte des effets tridimensionnels dans de futures simulations 3D de propagation de fissure. Une étude du problème en deux dimensions est également développée, proposant une formule analytique du critère en cisaillement. De plus un algorithme automatique de transition cisaillement-traction a été implémenté dans le code de calcul de dynamique explicite Europlexus, développé par le CEA. Une méthodologie d’identification des paramètres du modèle pour un matériau donné et pour un cas quasi-statique a été proposée. Confronté à l’interprétation de deux expériences connues de propagation dynamique (expériences de Kalthoff et de Ravichandran), le modèle proposé a montré sa pertinence. De plus, afin de mieux connaître le comportement à rupture de l’acier à Haute Limite Élastique Soudable, deux études expérimentales dédiées au suivi de la propagation dynamique d’un front de fissure ont été développées et validées sur des essais de rupture sous chargement quasi-statique et dynamique de type choc. Cette étude expérimentale a permis d’observer que les branchements de fissures, relevés sur les essais sous chargement quasi-statique, n’apparaissent plus sous chargement dynamique et pour des sollicitations en mode I pur. Les méthodes théoriques et numériques développées dans ces travaux de thèse permettent donc de simuler, automatiquement et avec un unique modèle, les changements de modes de rupture au cours d’une propagation dynamique de fissure. De plus, les protocoles expérimentaux exposés dans ce manuscrit permettent d’appréhender les phénomènes de transition cisaillement-traction en soulevant l’importance de la vitesse de sollicitation et du mode de sollicitation de l’essai. / We propose an approach to the simulation of the shear-tensile transition in dynamic crack growth based on two points: a new crack propagation criterion which is suitable for shear, and an algorithm which is capable of handling the transition from shear mode to tensile mode and back in the same simulation. The new crack propagation criterion for brittle crack growth is based on the maximum shear stress rather than the maximum hoop stress. The shear stress direction becomes the new crack’s direction in which propagation is initiated for shear-type failure. The stress state at the crack’s tip is obtained through a local approach which can be used even in the case of extensive plasticity. Additionally, we propose to control the transition from shear mode to tensile mode during the simulation of crack propagation using an equivalent strain estimated at the crack’s tip. Depending on a threshold strain, the propagation direction is predicted using the maximum shear stress (in the shear case) or the maximum hoop stress (in the tensile case).

Sous l’effet des sollicitations mécaniques répétées induites par les passages des trains, on observe l’apparition de fissures de fatigue de contact dans les rails. Une fois amorcées, celles-ci peuvent se propager et mener à la rupture du rail. Dans un contexte d’intensification du trafic et d’augmentation globale des vitesses de circulation, il devient stratégique pour SNCF d’optimiser sa politique de maintenance. Afin de définir des pas de surveillance adaptés et une planification optimisée des opérations de maintenance, une meilleure connaissance des mécanismes d’endommagement par fatigue du rail s’avère nécessaire. Tendre vers cette stratégie de maintenance prédictive passe par la mise en place d’outils de simulation numérique adaptés. Dans ce contexte, une chaîne d’outils a été développée : détermination des sollicitations transmises au rail, des champs de contraintes et de déformations résiduelles, localisation des zones critiques vis-à-vis du risque de fissuration. L’étape suivante consiste à estimer le risque lié à la présence de fissures et à étudier leurs propagations. Elle constitue une partie des objectifs de ces travaux de thèse. La résolution du problème tri-dimensionnel d’une structure fissurée, avec contact et frottement entre les lèvres, est effectuée grâce à un modèle tri-dimensionnel éléments finis étendus multi-échelles. Ce modèle fait appel à une formulation mixte stabilisée où chaque champ est écrit à l’aide d’enrichissement. La fissure est représentée grâce à une stratégie implicite-explicite. Le problème est résolu à l’aide du solveur non-linéaire LATIN. Une étude empirico-numérique a permis de proposer des formules a priori assurant à la méthode de résolution un taux de convergence proche de l’optimal. La simulation de la propagation des fissures de fatigue est réalisée à l’aide de critères spécifiques, adaptés à un chargement multi-axial et non-proportionnel, et d’une loi de propagation dédiée en mode mixte. La confrontation des résultats de simulation avec des essais réalisés sur une configuration cylindre-plan a validé la stratégie X-FEM/LATIN à deux échelles. Tous ces développements ont été implémentés dans le code de calcul éléments finis CAST3M. Des contraintes résiduelles réalistes, provenant d’un logiciel externe, ont été introduites. Cette étape a requis la mise en place d’une procédure de transfert des champs entre les deux maillages (celui utilisé pour le calcul des contraintes résiduelles et celui utilisé pour la résolution du problème de mécanique élastique linéaire de la rupture). L’étude de la flexion du rail a révelé l’influence de ce phénomène uniquement lors du passage du chargement sur la fissure.Enfin, une étude numérique a montré la très forte influence de l’orientation du chargement tangentiel, des contraintes résiduelles et de la présence de plusieurs fissures sur la direction et les vitesses de propagation des fissures de fatigue. / To optimize the rail grinding strategy, the prediction of crack growth rates has a vital role. Contact, with friction between the crack faces, notably occurs in rolling contact fatigue (RCF) problems. These time-dependent, multi-axial, non proportional loadings may lead to a crack initiation and propagation, and sometimes to the development of very complex 3D crack network. Numerical simulations of frictional fatigue crack are efficiently performed using the eXtended Finite Element Method (X-FEM). Within this method, the mesh does not need to conform to the crack geometry. Most difficulties associated to complex mesh generation around the crack and the re-meshing steps during the propagation are hence avoided. A 3D two-scale frictional contact fa-tigue crack model developed within the X-FEM framework is presented. It allows the use of a refined discretization of the crack interface independent from the underlying finite element mesh and adapted to the frictional contact crack scale. The model is used here to analyze the crack propagation, rate and direction, under rolling contact fatigue. The wheel-rail contact loading is modeled as a traveling hertzian load. The stress intensity factors are computed at the crack tips during the wheel passage. Criteria for determining crack growth direction under multiaxial non proportional conditions and mixed mode Paris’ law are used. Actual residual stresses are accounted for in the simulation. They are determined thanks to a dedicated model used at SNCF in which the asymptotic mechanical state of the rail is computed when submitted to cyclic loads. A non-uniform elastic-plastic stabilized state is calculated and introduced, by projection of the mechanical fields onto the finite element mesh, in the crack propagation simulation. All this strategy has been implemented in CAST3M and is now used to model 3D frictional crack growth under RCF.