Geometric Multigrid (GMG) methods are widely used in numerical analysis to accelerate the convergence of partial differential equations solvers using a hierarchy of grid discretizations. These solvers find plenty of applications in various fields in engineering and scientific domains, where solving PDEs is of fundamental importance. Using multigrid methods, the pace at which the solvers arrive at the solution can be improved at an algorithmic level. With the advance in modern computer architecture, solving problems with higher complexity and sizes is feasible - this is also the case with multigrid methods. However, since hardware support alone cannot achieve high performance in execution time, there is a need for good software that help programmers in doing so. Multiple grid sizes and recursive expression of multigrid cycles make the task of manual program optimization tedious and error-prone. A high-level language that aids domain experts to quickly express complex algorithms in a compact way using dedicated constructs for multigrid methods and with good optimization support is thus valuable. Typical computation patterns in a GMG algorithm includes stencils, point-wise accesses, restriction and interpolation of a grid. These computations can be optimized for performance on modern architectures using standard parallelization and locality enhancement techniques. Several past works have addressed the problem of automatic optimizations of computations in various scientific domains using a domain-specific language (DSL) approach. A DSL is a language with features to express domain-specific computations and compiler support to enable optimizations specific to these computations. Halide and PolyMage are two of the recent works in this direction, that aim to optimize image processing pipelines. Many computations like upsampling and downsampling an image are similar to interpolation and restriction in geometric multigrid methods. In this thesis, we demonstrate how high performance can be achieved on GMG algorithms written in the PolyMage domain-specific language with new optimizations we added to the compiler. We also discuss the implementation of non-trivial optimizations, on PolyMage compiler, necessary to achieve high parallel performance for multigrid methods on modern architectures. We realize these goals by: • introducing multigrid domain-specific constructs to minimize the verbosity of the algorithm specification; • storage remapping to reduce the memory footprint of the program and improve cache locality exploitation; • mitigating execution time spent in data handling operations like memory allocation and freeing, using a pool of memory, across multiple multigrid cycles; and • incorporating other well-known techniques to leverage performance, like exploiting multi-dimensional parallelism and minimizing the lifetime of storage buffers. We evaluate our optimizations on a modern multicore system using five different benchmarks varying in multigrid cycle structure, complexity and size, for two-and three-dimensional data grids. Experimental results show that our optimizations: • improve performance of existing PolyMage optimizer by 1.31x; • are better than straight-forward parallel and vector implementations by 3.2x; • are better than hand-optimized versions in conjunction with optimizations by Pluto, a state-of-the-art polyhedral source-to-source optimizer, by 1.23x; and • achieve up to 1.5$\times$ speedup over NAS MG benchmark from the NAS Parallel Benchmarks. (The speedup numbers are Geometric means over all benchmarks)

No description available.

Le but de cette étude est d'améliorer les performances, en termes d'espace mémoire et de temps de calcul, des simulations actuelles de l'Interaction mécanique Pastille-Gaine (IPG), phénomène complexe pouvant avoir lieu lors de fortes montées en puissance dans les réacteurs à eau sous pression. Parmi les méthodes de raffinement de maillage, méthodes permettant de simuler efficacement des singularités locales, une approche multi-grille locale a été choisie car elle présente l'intérêt de pouvoir utiliser le solveur en boîte noire tout en ayant un faible nombre de degrés de liberté à traiter par niveau. La méthode Local Defect Correction (LDC), adaptée à une discrétisation de type éléments finis, a tout d'abord été analysée et vérifiée en élasticité linéaire, sur des configurations issues de l'IPG, car son utilisation en mécanique des solides est peu répandue. Différentes stratégies concernant la mise en oeuvre pratique de l'algorithme multi-niveaux ont également été comparées. La combinaison de la méthode LDC et de l'estimateur d'erreur a posteriori de Zienkiewicz-Zhu, permettant d'automatiser la détection des zones à raffiner, a ensuite été testée. Les performances obtenues sur des cas bidimensionnels et tridimensionnels sont très satisfaisantes, l'algorithme proposé se montrant plus performant que des méthodes de raffinement h-adaptatives. Enfin, l'algorithme a été étendu à des problèmes mécaniques non linéaires. Les questions d'un raffinement espace/temps mais aussi de la transmission des conditions initiales lors du remaillage ont entre autres été abordées. Les premiers résultats obtenus sont encourageants et démontrent l'intérêt de la méthode LDC pour des calculs d'IPG. / The aim of this study is to improve the performances, in terms of memory space and computational time, of the current modelling of the Pellet-Cladding mechanical Interaction (PCI),complex phenomenon which may occurs during high power rises in pressurised water reactors. Among the mesh refinement methods - methods dedicated to efficiently treat local singularities - a local multi-grid approach was selected because it enables the use of a black-box solver while dealing few degrees of freedom at each level. The Local Defect Correction (LDC) method, well suited to a finite element discretisation, was first analysed and checked in linear elasticity, on configurations resulting from the PCI, since its use in solid mechanics is little widespread. Various strategies concerning the implementation of the multilevel algorithm were also compared. Coupling the LDC method with the Zienkiewicz-Zhu a posteriori error estimator in orderto automatically detect the zones to be refined, was then tested. Performances obtained on two-dimensional and three-dimensional cases are very satisfactory, since the algorithm proposed is more efficient than h-adaptive refinement methods. Lastly, the LDC algorithm was extended to nonlinear mechanics. Space/time refinement as well as transmission of the initial conditions during the remeshing step were looked at. The first results obtained are encouraging and show the interest of using the LDC method for PCI modelling.

Este trabalho propõe uma nova abordagem, baseada em wavelets, para o método Multigrid Algébrico (WAMG). Nesta nova abordagem, a Transformada Discreta Wavelet é aplicada na matriz de coeficientes do sistema linear gerando uma aproximação dessa matriz em cada nível do processo de multiresolução. As vantagens da nova abordagem, que incluem maior facilidade de paralelização e menor tempo de montagem, são apresentadas com detalhes e uma análise quantitativa de convergência do método WAMG é realizada a partir da sua aplicação em problemas testes. O WAMG também é testado como pré- condicionador para métodos iterativos no subespaço de Krylov na análise magnetostática e magnetodinâmica (regime permanente senoidal) pelo Método dos Elementos Finitos, e em matrizes esparsas extraidas das coleções Matrix Market e da Universidade da Flórida. São apresentados resultados numéricos comparando o WAMG com o Multigrid Algébrico tradicional e com os pré-condicionadores baseados em decomposições incompletas de Cholesky e LU. / In this work we propose a wavelet-based algebraic multigrid method (WAMG) as a linear system solver as well as a prediconditioner for Krylov subspace methods. It is a new approach for the Algebraic Multigrid method (AMG), which considers the use of Discrete Wavelet Transform (DWT) in the construction of a hierarchy of matrices. The two-dimensional DWT is applied to produce an approximation of the matrix in each level of the wavelets multiresolution decomposition process. The main advantages of this new approach are presented and a quantitative analysis of its convergence is shown after its application in some test problems. The WAMG also is tested as a preconditioner for Krylov subspace methods in problems with sparse matrices, in nonlinear magnetic field problems and in 3D time-harmonic Electromagnetic Edge-based Finite Element Analysis. Numerical results are presented comparing the WAMG with the standard Algebraic Multigrid method and with the preconditioners based on the incomplete Cholesky and LU decompositions.

A necessidade de obter solução de grandes sistemas lineares resultantes de processos de discretização de equações diferenciais parciais provenientes da modelagem de diferentes fenômenos físicos conduz à busca de técnicas numéricas escaláveis. Métodos multigrid são classificados como algoritmos escaláveis.Um estimador de erros deve estar associado à solução numérica do problema discreto de modo a propiciar a adequada avaliação da solução obtida pelo processo de aproximação. Nesse contexto, a presente tese caracteriza-se pela proposta de reutilização das estruturas matriciais hierárquicas de operadores de transferência e restrição dos métodos multigrid algébricos para acelerar o tempo de solução dos sistemas lineares associados à equação do transporte de contaminantes em meio poroso saturado. Adicionalmente, caracteriza-se pela implementação das estimativas residuais para os problemas que envolvem dados constantes ou não constantes, os regimes de pequena ou grande advecção e pela proposta de utilização das estimativas residuais associadas ao termo de fonte e à condição inicial para construir procedimentos adaptativos para os dados do problema. O desenvolvimento dos códigos do método de elementos finitos, do estimador residual e dos procedimentos adaptativos foram baseados no projeto FEniCS, utilizando a linguagem de programação PYTHONR e desenvolvidos na plataforma Eclipse. A implementação dos métodos multigrid algébricos com reutilização considera a biblioteca PyAMG. Baseado na reutilização das estruturas hierárquicas, os métodos multigrid com reutilização com parâmetro fixo e automática são propostos, e esses conceitos são estendidos para os métodos iterativos não-estacionários tais como GMRES e BICGSTAB. Os resultados numéricos mostraram que o estimador residual captura o comportamento do erro real da solução numérica, e fornece algoritmos adaptativos para os dados cuja malha retornada produz uma solução numérica similar à uma malha uniforme com mais elementos. Adicionalmente, os métodos com reutilização são mais rápidos que os métodos que não empregam o processo de reutilização de estruturas. Além disso, a eficiência dos métodos com reutilização também pode ser observada na solução do problema auxiliar, o qual é necessário para obtenção das estimativas residuais para o regime de grande advecção. Esses resultados englobam tanto os métodos multigrid algébricos do tipo SA quanto os métodos pré-condicionados por métodos multigrid algébrico SA, e envolvem o transporte de contaminantes em regime de pequena e grande advecção, malhas estruturadas e não estruturadas, problemas bidimensionais, problemas tridimensionais e domínios com diferentes escalas. / The need for solving large linear systems arising from the discretization of partial differential equations modelling physical phenomena motivates the search for scalable numerical techniques. Multigrid algorithms are instances of such techniques.In order to provide a suitable assessment of the solution obtained by such algorithms, an error estimator must be associated to the numerical solution of the discretized problem. In this context, this thesis proposes the reutilization of the hierarchical matrix structures of transfer operators and the restriction to algebraic multigrid methods to speed up the process of solving the linear systems associated with the contaminant transport equation in saturated porous media. In addition, it features the implementation of residual estimates for problems involving constant or non-constant data, the regimes of small- or large-scale advection and the proposal of employing the residual estimates associated to the source term and to the initial condition to build adaptive procedures for the problem data. The development of the computer codes of the finite element method, residual estimator and adaptive procedures were based on the FEniCS project, using the programming language PYTHONR and developed on the Eclipse platform. The implementation of the algebraic methods with reutilization relied upon the libray PyAMG. Grounding on the idea of reutilizing the hierarchical structures, fixed and automatic parameters multigrid methods were proposed and extended to non-stationary iterative methods such as GMRES and BICGSTAB. The numerical results demonstrate that the residual estimator captures the behavior of the real error of the numerical solution, and provide adaptive algorithms for the data whose output mesh yields a numerical solution alike to that obtained from a uniform mesh with more elements. Moreover, the methods with reutilization are faster than those that do not reuse the structures. Besides, the efficiency of such methods can also be observed in the solution of an auxiliary problem, which is necessary for deriving the residual estimates in the regime of large-scale advection. These results encompass both the type SA algebraic multigrid method and those pre-conditioned by them. Moreover, they involve the transport of contaminants in regime of small- and large-scale advection, structured and non-structured meshes, bi- and tridimensional problems and domains with different scales.

A necessidade de obter solução de grandes sistemas lineares resultantes de processos de discretização de equações diferenciais parciais provenientes da modelagem de diferentes fenômenos físicos conduz à busca de técnicas numéricas escaláveis. Métodos multigrid são classificados como algoritmos escaláveis.Um estimador de erros deve estar associado à solução numérica do problema discreto de modo a propiciar a adequada avaliação da solução obtida pelo processo de aproximação. Nesse contexto, a presente tese caracteriza-se pela proposta de reutilização das estruturas matriciais hierárquicas de operadores de transferência e restrição dos métodos multigrid algébricos para acelerar o tempo de solução dos sistemas lineares associados à equação do transporte de contaminantes em meio poroso saturado. Adicionalmente, caracteriza-se pela implementação das estimativas residuais para os problemas que envolvem dados constantes ou não constantes, os regimes de pequena ou grande advecção e pela proposta de utilização das estimativas residuais associadas ao termo de fonte e à condição inicial para construir procedimentos adaptativos para os dados do problema. O desenvolvimento dos códigos do método de elementos finitos, do estimador residual e dos procedimentos adaptativos foram baseados no projeto FEniCS, utilizando a linguagem de programação PYTHONR e desenvolvidos na plataforma Eclipse. A implementação dos métodos multigrid algébricos com reutilização considera a biblioteca PyAMG. Baseado na reutilização das estruturas hierárquicas, os métodos multigrid com reutilização com parâmetro fixo e automática são propostos, e esses conceitos são estendidos para os métodos iterativos não-estacionários tais como GMRES e BICGSTAB. Os resultados numéricos mostraram que o estimador residual captura o comportamento do erro real da solução numérica, e fornece algoritmos adaptativos para os dados cuja malha retornada produz uma solução numérica similar à uma malha uniforme com mais elementos. Adicionalmente, os métodos com reutilização são mais rápidos que os métodos que não empregam o processo de reutilização de estruturas. Além disso, a eficiência dos métodos com reutilização também pode ser observada na solução do problema auxiliar, o qual é necessário para obtenção das estimativas residuais para o regime de grande advecção. Esses resultados englobam tanto os métodos multigrid algébricos do tipo SA quanto os métodos pré-condicionados por métodos multigrid algébrico SA, e envolvem o transporte de contaminantes em regime de pequena e grande advecção, malhas estruturadas e não estruturadas, problemas bidimensionais, problemas tridimensionais e domínios com diferentes escalas. / The need for solving large linear systems arising from the discretization of partial differential equations modelling physical phenomena motivates the search for scalable numerical techniques. Multigrid algorithms are instances of such techniques.In order to provide a suitable assessment of the solution obtained by such algorithms, an error estimator must be associated to the numerical solution of the discretized problem. In this context, this thesis proposes the reutilization of the hierarchical matrix structures of transfer operators and the restriction to algebraic multigrid methods to speed up the process of solving the linear systems associated with the contaminant transport equation in saturated porous media. In addition, it features the implementation of residual estimates for problems involving constant or non-constant data, the regimes of small- or large-scale advection and the proposal of employing the residual estimates associated to the source term and to the initial condition to build adaptive procedures for the problem data. The development of the computer codes of the finite element method, residual estimator and adaptive procedures were based on the FEniCS project, using the programming language PYTHONR and developed on the Eclipse platform. The implementation of the algebraic methods with reutilization relied upon the libray PyAMG. Grounding on the idea of reutilizing the hierarchical structures, fixed and automatic parameters multigrid methods were proposed and extended to non-stationary iterative methods such as GMRES and BICGSTAB. The numerical results demonstrate that the residual estimator captures the behavior of the real error of the numerical solution, and provide adaptive algorithms for the data whose output mesh yields a numerical solution alike to that obtained from a uniform mesh with more elements. Moreover, the methods with reutilization are faster than those that do not reuse the structures. Besides, the efficiency of such methods can also be observed in the solution of an auxiliary problem, which is necessary for deriving the residual estimates in the regime of large-scale advection. These results encompass both the type SA algebraic multigrid method and those pre-conditioned by them. Moreover, they involve the transport of contaminants in regime of small- and large-scale advection, structured and non-structured meshes, bi- and tridimensional problems and domains with different scales.

Este trabalho propõe uma nova abordagem, baseada em wavelets, para o método Multigrid Algébrico (WAMG). Nesta nova abordagem, a Transformada Discreta Wavelet é aplicada na matriz de coeficientes do sistema linear gerando uma aproximação dessa matriz em cada nível do processo de multiresolução. As vantagens da nova abordagem, que incluem maior facilidade de paralelização e menor tempo de montagem, são apresentadas com detalhes e uma análise quantitativa de convergência do método WAMG é realizada a partir da sua aplicação em problemas testes. O WAMG também é testado como pré- condicionador para métodos iterativos no subespaço de Krylov na análise magnetostática e magnetodinâmica (regime permanente senoidal) pelo Método dos Elementos Finitos, e em matrizes esparsas extraidas das coleções Matrix Market e da Universidade da Flórida. São apresentados resultados numéricos comparando o WAMG com o Multigrid Algébrico tradicional e com os pré-condicionadores baseados em decomposições incompletas de Cholesky e LU. / In this work we propose a wavelet-based algebraic multigrid method (WAMG) as a linear system solver as well as a prediconditioner for Krylov subspace methods. It is a new approach for the Algebraic Multigrid method (AMG), which considers the use of Discrete Wavelet Transform (DWT) in the construction of a hierarchy of matrices. The two-dimensional DWT is applied to produce an approximation of the matrix in each level of the wavelets multiresolution decomposition process. The main advantages of this new approach are presented and a quantitative analysis of its convergence is shown after its application in some test problems. The WAMG also is tested as a preconditioner for Krylov subspace methods in problems with sparse matrices, in nonlinear magnetic field problems and in 3D time-harmonic Electromagnetic Edge-based Finite Element Analysis. Numerical results are presented comparing the WAMG with the standard Algebraic Multigrid method and with the preconditioners based on the incomplete Cholesky and LU decompositions.

L'imagerie photoacoustique (IPA) des tissus biologiques permet de combiner les avantages des imageries optique et ultrasonore. Le principal contraste endogène pour l’IPA provient des vaisseaux sanguins en raison de la forte absorption de l'hémoglobine par rapport aux tissus environnants. De plus, les vaisseaux sanguins sont à peu près cylindriques et la concentration d'hémoglobine peut être supposée uniforme à l'intérieur des veines. Comme première contribution, nous avons développé dans cette thèse un modèle analytique de fluence optique pour plusieurs inhomogénéités cylindriques parallèles incorporées dans un milieu turbide. Les modèles analytiques n’existent que pour les cas simples. Pour traiter des situations plus complexes, comme les tissus biologiques, les méthodes numériques sont nécessaires. La deuxième contribution de cette thèse consiste à développer un solveur multigrilles de l'équation de diffusion optique et donc de proposer une méthode numérique efficace pour résoudre la fluence optique. Enfin, notre troisième contribution concerne la reconstruction de la tomographie quantitative photoacoustique (TQPA). Basée sur les modèles efficaces présentées dans les première et seconde contributions, nous avons proposé une méthode de reconstruction basée sur le modèle direct analytique pour les cas simples et une méthode d'inversion basée sur multigrille pour les cas plus réalistes. Les avantages de la méthode d'inversion basée sur multigrille sont présentés à la fois en terme de temps de calcul et de vitesse de convergence. Une validation expérimentale est présentée dans le dernier chapitre de cette thèse, prouvant la validité et l'analyse des performances des méthodes développées. / Photoacoustic imaging (PAI) of biological tissues tries to combine the advantages of optical and acoustical imaging. The main endogenous contrast for PAI is derived from blood vessels due to the strong absorption of hemoglobin compared to the background tissues. Furthermore, blood vessels are roughly cylindrical and hemoglobin concentration can be assumed to be uniform inside the vessel. Therefore, the blood vessels can be considered as “cylindrical inhomogeneities”. As a first contribution, we have developed in this thesis an analytical model of optical fluence for multiple parallel cylindrical inhomogeneities embedded in an otherwise homogeneous turbid medium. Analytical models only exist for simple cases. To deal with more complex situations like biological tissues, numerical methods are required. The second contribution of this thesis is to develop a multigrid solver of optical diffusion equation and therefore to propose an efficient numerical method to resolve the optical fluence. Finally, our third contribution is concerned with quantitative PA tomography (QPAT) reconstruction. Based on the efficient models presented in the first and second contributions, we have proposed an analytic-based reconstruction method for simple cases and a multigrid-based inversion scheme for more realistic cases. The advantages of multigrid-based inversion scheme are shown in both computation and convergence speed. An experimental validation is presented in the last chapter of this thesis, proving the validity and analyzing the performances of the developed methods.

Les problèmes de mécanique du contact sont des problèmes multi-échelles mettant en jeux de nombreux phénomènes physiques. Les premières études concernant ce domaine datent de la fin du XIXème siècle et les développements majeurs ont été réalisés au cours du XXème siècle en parallèle du besoin croissant des ingénieurs de prévoir le comportement des matériaux sous sollicitations tribologiques. L'évolution des besoins industriels et les avancées technologiques réalisées dans le domaine du numérique conduisent à réaliser des simulations tridimensionnelles ayant pour objectif la prédiction du comportement de pièces sous sollicitations thermo-mécaniques transitoires, pour, soit alléger les structures, augmenter le niveau de sollicitations, étendre la durée de vie... Ces simulations se révèlent très souvent coûteuses en termes de temps de calcul et d'espace mémoire et nécessitent par conséquent l'utilisation de super calculateurs. Dans ce contexte, cette thèse propose un modèle innovant basé sur les techniques multigrilles avec raffinement local afin de réaliser ces simulations pour des coûts numériques faibles. Ce modèle est basé sur les équations de Lamé généralisées et l'équation de la chaleur de Fourier discrétisée à l'aide des différences finies. Le système linéaire obtenu est résolu à l'aide de la méthode itérative de Gauss-Seidel couplée avec les techniques multigrilles. Ces techniques permettent d'accélérer la convergence d'un problème en utilisant plusieurs grilles et des opérateurs de transfert. Afin de garantir une convergence optimale et de minimiser la taille mémoire dans le cas de variations de propriétés importantes, des techniques numériques de localisation et d'optimisation ont été mises en place. Les applications visées ici sont centrées sur l'utilisation de revêtements ou de matériaux innovants pour permettre les gains attendus. Des validations du modèle ont été effectuées en comparant nos résultats avec ceux issus de la littérature. Des études paramétriques ont permis d'étudier l'influence de l'épaisseur du revêtement, de la valeur du module de Young mais aussi d'une couche à gradient de propriété sur le champ de contrainte et la tenue du système revêtement/substrat sous sollicitation de contact. Des études similaires ont été conduites sous sollicitations thermiques. L'intérêt porté aux variations de propriétés des matériaux selon toutes les directions de l'espace a conduit à étudier l'effet de la microstructure, qui est constituée de grains ayant chacun leurs propriétés propres, sur les champs de contraintes. La mise en évidence de cet effet est explicitement montrée au travers de calculs de durée de vie utilisant des descriptions statistiques de type Weibull. La dispersion observée sur les résultats est conforme aux observations expérimentales. / Contact mechanic problems are multi-scale and involve numerous physical phenomena. These problems have been studied since the end of the XIXth century and major developments have been made during the XXth century due to the necessity for engineers to predict material behavior under tribological loads. Currently, industrial demands and technological breakthroughs drive people to consider three-dimensional simulations to study this behavior under thermo-mechanical loads. The objectives are multiple: reduce of the size of structures, increase of material resistance, improvement of fatigue life... These simulations, which often require high numerical costs in terms of memory size and CPU time, have to be performed on super computers. In this context, this work proposes an innovative model based on multigrid methods using a local refinement strategy in order to perform these simulations at a low numerical cost. The model is based on the Lamé elasticity equations and the Fourier heat equation which have been discretized using a finite difference framework. The obtained linear system is solved using the Gauss-Seidel iterative method coupled with multigrid techniques. These methods allow an acceleration of the convergence speed, using different grids and transfer operators. In order to obtain an optimum convergence speed and decrease the required memory size, local refinement strategies and optimization techniques have been used. Several calculations required hundred millions of points, can be solved on a personal computer within a few hours. Applications focus on the use of a coating or innovative materials which allow improvements in terms of fatigue life. The model has been validated against results found in the literature. Parametric studies allowed to analyse the influence of the coating thickness, the Young's modulus ratio or the use of a graded layer on the stress field and on the coating/substrate system behaviour under contact loads. Similar studies have been performed under thermal loads. Special attention has been paid to the material property variations along all space directions. It has lead us to consider a material microstructure which is composed of grains with their individual properties. The influence of the microstructure on the fatigue life phenomenon is clearly highlighted using statistical Weibull charts. The dispersion observed in the numerical results tends to be similar to experiments found in the literature.

Ce travail de thèse s’inscrit dans le cadre de l’étude de l’Interaction mécanique Pastille-Gaine (IPG) se produisant dans les crayons combustibles des réacteurs à eau pressurisée. Ce mémoire porte sur le développement de méthodes de raffinement de maillage permettant de simuler plus précisément le phénomène d’IPG tout en conservant des temps de calcul et un espace mémoire acceptables pour des études industrielles. Une stratégie de raffinement automatique basée sur la combinaison de la méthode multi-grilles Local Defect Correction (LDC) et l’estimateur d’erreur a posteriori de type Zienkiewicz et Zhu est proposée. Cette stratégie s’appuie sur l’erreur fournie par l’estimateur pour détecter les zones à raffiner constituant alors les sous-grilles locales de la méthode LDC. Plusieurs critères d’arrêt sont étudiés afin de permettre de stopper le raffinement quand la solution est suffisamment précise ou lorsque le raffinement n’apporte plus d’amélioration à la solution globale.Les résultats numériques obtenus sur des cas tests 2D élastiques avec discontinuité de chargement permettent d’apprécier l’efficacité de la stratégie proposée.Le raffinement automatique de maillage dans le cas de problèmes de contact unilatéral est ensuite abordé. La stratégie proposée dans ce travail s’étend aisément au raffinement multi-corps à condition d’appliquer l’estimateur d’erreur sur chacun des corps séparément. Un post-traitement est cependant souvent nécessaire pour garantir la conformité des zones de raffinement vis-à-vis des frontières de contact. Une variété de tests numériques de contact entre solides élastiques confirme l’efficacité et la généricité de la stratégie proposée. / This Ph.D. work takes place within the framework of studies on Pellet-Cladding mechanical Interaction (PCI) which occurs in the fuel rods of pressurized water reactor. This manuscript focuses on automatic mesh refinement to simulate more accurately this phenomena while maintaining acceptable computational time and memory space for industrial calculations. An automatic mesh refinement strategy based on the combination of the Local Defect Correction multigrid method (LDC) with the Zienkiewicz and Zhu a posteriori error estimator is proposed. The estimated error is used to detect the zones to be refined, where the local subgrids of the LDC method are generated. Several stopping criteria are studied to end the refinement process when the solution is accurate enough or when the refinement does not improve the global solution accuracy anymore.Numerical results for elastic 2D test cases with pressure discontinuity shows the efficiency of the proposed strategy.The automatic mesh refinement in case of unilateral contact problems is then considered. The strategy previously introduced can be easily adapted to the multibody refinement by estimating solution error on each body separately. Post-processing is often necessary to ensure the conformity of the refined areas regarding the contact boundaries. A variety of numerical experiments with elastic contact (with or without friction, with or without an initial gap) confirms the efficiency and adaptability of the proposed strategy.