Les simulations numériques à l'échelle du pore sont fréquemment utilisées pour étudier le comportement des écoulements multiphasiques largement rencontrées dans phénomènes naturels et applications industrielles. Dans ce travail, la morphologie de structures liquides et l'action capillaire sont examinées à l'échelle des pores par la méthode de Boltzmann sur réseau (LBM) à plusieurs composants selon le modèle de Shan-Chen. Les résultats numériques obtenus sont en bon accord avec les solutions théoriques. Les simulations numériques sont étendues à microstructures complexes au-delà du régime pendulaire.La LBM a été utilisée pour modéliser l'écoulement multiphasique à travers un milieu poreux idéalisé dans des conditions de drainage primaire quasi-statique. Les simulations LBM ont fourni une excellente description du déplacement de l'interface fluide-fluide à travers les grains. Pendant le drainage, les simulations LBM sont capables de reproduire la déconnexion d'une phase dans le milieu granulaire sous la forme de ponts pendulaires ou structures liquides complexes. Malheureusement, le temps de calcul nécessaire pour ce type de simulations est assez élevé. Afin d’optimiser les ressources de calcul, nous présentons un modèle 2D (modèle Throat-Network) basé sur des solutions analytiques pour décrire l'écoulement biphasique à travers un ensemble de disques dans un temps de calcul très réduit, donc le modèle 2D est susceptible de remplacer les simulations LBM lorsque les ressources de calcul sont limitées. L'approche souligne l'importance de simuler le problème a l'échelle de la gorge du pore pour obtenir les relations volume - pression capillaire locales. Le modèle Throat-Network est un point de départ pour le modèle hybride proposé pour résoudre les problèmes en 3D. Le modèle hybride combine l’efficacité de l’approche réseau de pores et la précision du LBM à l’échelle des pores. Le modèle hybride est basé sur la décomposition de l’échantillon en petits sous-domaines, dans lesquels des simulations LBM sont effectuées pour déterminer les propriétés hydrostatiques principales (pression capillaire d'entrée, courbe de drainage primaire et morphologie du liquide pour chaque gorge du pore). Malgré la réduction significative des temps de calcul obtenus avec le modèle hybride, le temps n’est pas négligeable et les modélisations numériques d'échantillons de grandes tailles ne sont pas réalistes. Les approximations données par les méthodes Incircle et MS-P, qui prédisent les propriétés hydrostatiques, sont comparées à celles de LBM et du modèle hybride. / Numerical simulations at the pore scale are a way to study the behavior of multiphase flows encountered in many natural processes and industrial applications. In this work, liquid morphology and capillary action are examined at the pore-scale by means of the multicomponent Shan-Chen lattice Boltzmann method (LBM). The accuracy of the numerical model is first contrasted with theoretical solutions. The numerical results are extended to complex microstructures beyond the pendular regime.The LBM has been employed to simulate multiphase flow through idealized granular porous media under quasi-static primary drainage conditions. LBM simulations provide an excellent description of the fluid-fluid interface displacement through the grains. Additionally, the receding phase trapped in the granular media in form of pendular bridges or liquid clusters is well captured. Unfortunately, such simulations require a significant computation time. A 2D model (Throat-Network model) based on analytical solutions is proposed to mimic the multiphase flow with very reduced computation cost, therefore, suitable to replace LBM simulations when the computation resources are limited. The approach emphasizes the importance of simulating at the throat scale rather than the pore body scale in order to obtain the local capillary pressure - liquid content relationships. The Throat-Network model is a starting point for the a hybrid model proposed to solve 3D problems. The hybrid model combines the efficiency of the pore-network approach and the accuracy of the LBM at the pore scale to optimize the computational resources. The hybrid model is based on the decomposition of the granular assembly into small subsets, in which LBM simulations are performed to determine the main hydrostatic properties (entry capillary pressure, capillary pressure - liquid content relationship and liquid morphology for each pore throat). Despite the reduction of computation time, it is still not negligible and not affordable for large granular packings. Approximations by the Incircle and the MS-P method, which predict hydrostatic properties, are contrasted with the results provided by LBM and the hybrid model. Relatively accurate predictions are given by the approximations.

Cette thèse étudie les instabilités dynamiques des milieux granulaires modèles saturés à l’aide d’un appareil triaxial classique. Les instabilités englobent la liquéfaction et les effondrements en compression isotrope drainée, les frottements saccadés en compression triaxiale drainée. Ces instabilités apparaissent spontanément à des contraintes effectives de confinement imprévisibles. Elles sont accompagnées de très rapides et très fortes surpressions interstitielles, malgré un drainage approprié; ce que ne présentent pas les milieux granulaires naturels. En compression isotrope drainée (consolidation), des effondrements locaux naissent instantanément. En compression triaxiale drainée, on observe de larges frottements saccadés quasi-périodiques caractérisés par des déformations volumiques et axiales contractantes. De temps en temps, ces effondrements et frottements saccadés locaux peuvent se développer en liquéfaction menant à une destruction complète de la structure granulaire. Les données à haute résolution temporelle issues de ce travail ont permis la découverte d’une nouvelle famille de liquéfaction dynamique et statique. L’étude des émissions acoustiques passives a permis l’identification de signature spectrale caractéristique. Pour les frottements saccadés, la phase de glissement peut être interprétée comme une consolidation dynamique, limitée par l’unique surface en dessous de la ligne critique de rupture dans le plan des contraintes effectives. La séquence temporelle précise des événements exclut que la pression interstitielle soit la cause principale des instabilités. Cependant, le rôle important de la surpression interstitielle est démontré dans des relations quantitatives entre les incréments de contraintes, et de déformations et l’éphémère surpression interstitielle stabilisée développée pendant la phase de glissement. Cela montre finalement la nature quasi-déterministique de ces instabilités dynamiques. Ces relations empiriques sont basées uniquement sur l’amplitude maximale de l’accélération verticale de très courte durée et sont gouvernées indépendamment par la pression de confinement et par l’indice des vides. La similarité de la surpression interstitielle entre différentes instabilités suggère fortement quelques mécanismes similaires de déclenchement, probablement à partir de ré-arrangements de la micro-structure granulaire. / This thesis reports a laboratory study on the dynamic instabilities of model saturated granular material using a triaxial apparatus. The term instability consists of isotropic collapse and liquefaction under isotropic compression and of stick-slip under triaxial compression in drained condition. The instabilities spontaneously occur at unpredictable effective stress with unexpected buildup of excess pore pressure irrespective of fully drained condition, contrasting with the instability-free behaviour of natural granular materials. In isotropic compression, instantaneous local collapse happens and in triaxial compression, very large and quasi-periodic stick-slip occurs with sudden volumetric compaction and axial contraction. Sometimes, these local failures (collapse and stick-slip) can develop into total liquefaction failure, destroying completely the granular structure. High time-resolved data permit the discovery of a new family of dynamic and static liquefaction. Passive acoustic measurements allow the identification of typical spectral signature. For stick-slip phenomenon, the slip phase with constant duration of stress drop can be interpreted as dynamic consolidation at constant deviatoric stress, limited by a unique boundary inside the critical state line in the effective stress plane. The precise temporal sequence of mechanical measurements excludes the generated pore pressure as the main cause of the instabilities. However, the role of pore pressure is emphasised by consistent quantitative relations between the amplitude of incremental stresses, incremental strains and the ephemeral stabilised excess pore pressure developed during the dynamic event, leading to the quasi-deterministic nature of granular instabilities. These empirical relations are based only on the short-lived maximum vertical acceleration and governed separately by the confining pressure and the initial void ratio. The similarity of pore pressure evolution for different kinds of instability strongly suggests some common speculative triggering mechanisms, probably originated from different rearrangements of the granular micro-structure.

A granular material is a collection of discrete, solid particles of macroscopic size dispersed in an interstitial fluid, in which the fluid has an insignificant effect on the particle dynamics. Because they exhibit fascinating properties because of dissipative interactions, due to their importance in geophysical and industrial processes, flows of granular materials have been the focus of large amount of research involving physicists and engineers. A good understanding of the physics of granular materials is desired in order to design efficient processing and handling systems. Granular materials can be heaped like a solid, and can flow like a fluid. Though the two distinct regimes of granular flows are well described by kinetic theory (rapid flows) and plasticity theories (quasi-static), the intermediate dense flow regime, where collisional and frictional interactions are important, is not yet described successfully. In this thesis, we examine the applicability of kinetic theory for dense granular flows, the structure and dynamics in sheared inelastic hard disks systems and dynamics of sheared non-spherical particles. Two complementary simulation techniques, the discrete element (DE) technique for soft particles and the event driven (ED) simulation technique for hard particles, are used to examine the extent to which the dynamics of an unconfined dense granular flow can be well described by a hard particle model when the particle stiffness becomes large. First, we examine the average co-ordination number for the particles in the flow down an inclined plane using the DE technique using both linear and Hertzian contact models. The simulations show that the average co-ordination number decreases below 1 for values of the spring stiffness corresponding to real materials such as sand and glass, even when the angle of inclination is only 1olarger than the angle of repose. The results of the two simulation techniques for the Bagnold coefficients (ratio of stress and square of the strain rate) and the granular temperature (mean square of the fluctuating velocity) are found to be in quantitative agreement. In addition, we also conduct the comparison of the pre-collisional relative velocities of particles in contact. Since momentum is transported primarily by particle contacts in a dense flow, the relative velocity distribution is a sensitive comparison of the dynamics in the two simulation techniques. It is found that the relative velocity distribution in both simulation techniques are well approximated by an exponential distribution for small coefficients of restitution, indicating that the dynamics of a dense granular flow can be adequately described by a hard particle model. The structure and dynamics of the two-dimensional linear shear flow of inelastic disks at high area fractions are analysed. The event-driven simulation technique is used in the hard-particle limit, where the particles interact through instantaneous collisions. The structure (relative arrangement of particles) is analysed using the bond-orientational order parameter. It is found that the shear flow reduces the order in the system, and the order parameter in a shear flow is lower than that in a collection of elastic hard disks at equilibrium. The distribution of relative velocities between colliding particles is analysed. The relative velocity distribution undergoes a transition from a Gaussian distribution for nearly elastic particles, to an exponential distribution at low coefficients of restitution. However, the single-particle distribution function is close to a Gaussian in the dense limit, indicating that correlations between colliding particles have a strong influence on the relative velocity distribution. This results in a much lower dissipation rate than that predicted using the molecular chaos assumption, where the velocities of colliding particles are considered to be uncorrelated. The orientational ordering and dynamical properties of the shear flow of inelastic dumbbells in two dimensions are studied, as a first step towards examining the effect of shape on the properties of flowing granular materials. The dumbbells are smooth fused disks characterised by the ratio of the distance between centers (L) and the disk diameter (D), and the ratio (L/D)varies between 0 and 1 in our simulations. Area fractions studied are in the range 0.1 to 0.7, while coefficients of normal restitution from 0.99 to 0.6 are considered. The simulations are similar to the event driven simulations for circular disks, but the procedure for predicting collisions is much more complicated due to the non-circular shape of the particles and due to particle rotation. The average orientation is measured using an orientational order parameter S, which varies between 0 (for a perfectly disordered fluid) and 1 (for a fluid with the axis of all dumbbells in the same direction). It is found that there is a gradual increase in ordering as the area fraction is increased, as the aspect ratio is increased or as the coefficient of restitution is decreased, and the order parameter has a maximum value of about 0.5 for the highest area fraction and lowest coefficient of restitution considered here. However, there is no discontinuous nematic transition for all the parameters studied here. The axis of the dumbbells are preferentially oriented along the extensional axis (at an angle of 45ofrom the flow direction) at low area fraction, but the orientation is closer to the flow direction as the area fraction is increased. The orientation distribution is calculated, and it is found that the orientation distribution is well described by a function of the form P(θ) =(1/π)+ (2S/π)cos(2(θ−θp)), where θis the angle from the flow direction and θpis the principal orientation direction. The mean energy of the velocity fluctuations in the flow direction is found to be higher than that in the gradient direction and the rotational energy, though the difference decreases as the area fraction increases, due to the efficient collisional transfer of energy between the three directions. The distributions of the translational and rotational velocity are found to be Gaussian distributions to a very good approximation. The equation of state for the pressure is calculated, and it is found to be remarkably independent of the coefficient of restitution. The pressure and dissipation rate show relatively little variation when scaled by the collision frequency for all the area fractions studied here, indicating that the collision frequency determines the momentum transport and energy dissipation even at the lowest area fractions studied here. The mean angular velocity of the particles is examined in some detail. It is found that the mean angular velocity is equal to half the vorticity at low area fractions, but the magnitude of the mean angular velocity systematically decreases to less than half the vorticity as the area fraction is increased, even though the stress tensor is symmetric.

Under loading conditions found in many geotechnical structures, it is common to observe failure in zones of high localized strain called shear bands. Existing models predict these localizations, but provide little insight into the micromechanics within the shear bands. This research captures the variation in microstructure inside and outside of shear bands that were formed in laboratory plane strain and two-dimensional discrete element method (DEM) biaxial compression experiments. Plane strain compression tests were conducted on dry specimens of Ottawa 20-30 sand to calibrate the device, assess global response repeatability, and develop a procedure to quantitatively define the onset of localization. A new methodology was employed to quantify and correct for the additional stresses imparted by the confining membrane in the vicinity of the shear band. Unsheared and sheared specimens of varying dilatancy were solidified using a two-stage resin impregnation procedure. DEM tests were performed using an innovative servo-controlled flexible lateral confinement algorithm to provide additional insights into laboratory results. The solidified specimens were sectioned and the resulting surfaces prepared for microstructure observation using bright field microscopy and morphological analysis. Local void ratio distributions and their statistical properties were determined and compared. Microstructural parameters for subregions in a grid pattern and along predefined inclined zones were also calculated. Virtual surfaces parallel to the shear band were identified and their roughnesses assessed. Similar calculations were performed on the DEM simulations at varying strain levels to characterize the evolution of microstructure with increasing strain. The various observations showed that the mean, standard deviation, and entropy of the local void ratio distributions all increased with increasing strain levels, particularly within regions of high local strains. These results indicate that disorder increases within a shear band and that the soil within the shear band does not adhere to the classical concept of critical state, but reaches a terminal void ratio that is largely a function of initial void ratio. Furthermore, there appears to be a transition zone between the far field and the fully formed shear block, as opposed to an abrupt delineation as traditionally inferred.

Integrated theoretical and experimental analysis is carried out in this work to investigate the fundamental failure mechanisms and flow patterns involved in the process of fluid injection into dense granular media. The experimental work is conducted with aqueous glycerin solutions, utilizing a novel setup based on a Hele-Shaw cell filled with dense dry sand. The two dimensional nature of the setup allows direct visualization and imaging analysis of the real-time fluid and grain kinematics. The experimental results reveal that the fluid flow patterns show a transition from simple radial flow to a ramified morphology while the granular media behaviors change from that of rigid porous media to localized failure that lead to development of fluid channels. Based on the failure/flow patterns, four distinct failure/flow regimes can be identified, namely, (i) a simple radial flow regime, (ii) an infiltration-dominated regime, (iii) a grain displacement-dominated regime, and (iv) a viscous fingering-dominated regime. These distinct failure/flow regimes emerge as a result of competition among various energy dissipation mechanisms, namely, viscous dissipation through infiltration, dissipation due to grain displacements, and viscous dissipation through flow in thin channels and can be classified based on the characteristic times associated with fluid injection, hydromechanical coupling and viscoelastoplasticity. The injection process is also analyzed numerically using the discrete element method (DEM) coupled with two fluid flow scheme, a fixed coarse grid scheme based on computational fluid dynamics (CFD) and a pore network modeling scheme. The numerical results from the two complementary methods reproduce phenomena consistent with the experimental observations and justify the concept of associating the displacement regimes with the partition among energy dissipation mechanisms. The research in this work, though fundamental in nature, will have direct impacts on many engineering problems in civil, environmental and petroleum engineering such as ground improvement, environmental remediation and reservoir stimulation.

The present work is concerned with the mathematical modelling of a bed of granular material in a gravitational field vertically fluidised by a vibrating surface. The particles are in rapid motion, and lose energy by inelastic collisions. The steady state is maintained by a balance of the rate of dissipation of energy in inelastic particle collisions and the rate of transfer of energy due to particle collisions with the vibrating surface. The limit where the energy dissipation due to inelastic collisions is small compared to the mean kinetic energy of the particles is considered. This non-equilibrium steady state is similar to a dilute gas at equilibrium with a uniform temperature and an exponentially decaying density, obtained from the ideal gas equation of state. From the analysis of this state, four non-dimensional numbers are derived which uniquely specify the state of the system. A perturbative analysis about the uniform temperature state is carried out and analytical solutions to the macroscopic variables of the system are obtained using two types of approximations. The first is a hydrodynamic model using constitutive relations from the general kinetic theory of granular media, and the second is a kinetic theory formulation derived exclusively for the vibro-fluidised bed. The latter permits an anisotropy between the horizontal and vertical directions due to the anisotropic nature of the source of energy at the bottom wall. The kinetic theory is extended to incorporate the corrections due to the high density effects, which is similar to the Enskog correction to dense gases. An event driven (ED), or hard sphere molecular dynamic (MD), simulation of the vibrated bed is carried out. The quantitative predictions of the theories are validated by the simulation. A systematic probing of the parameter space within the ED simulations revealed two new phenomena in a vibro-fluidised bed which are inhomogeneous in the horizontal direction. These are convection rolls similar to the Rayleigh-Benard instability in fluids, and a clustering instability leading to a phase separation. The instabilities are characterised using a phase diagram. The homogeneous states close to these new states are adequately described by the models developed here. An analysis of the stability of this state could have implications in understanding the instabilities in driven granular materials (such as in sheared media and fluidised beds) in general, and pattern formation in vibrated beds in particular.

A new simple design approach will be described that utilizes test results from the Repeated Load Triaxial Apparatus to establish the risk level of permanent deformations in the unbound granular layers (UGL) in pavement constructions under consideration of the seasonal effects. From this data a serviceability limit line (plastic shakedown limit) stress boundary for the unbound granular materials (UGM) was defined for different moisture contents. Below this line the material will have stable behavior. The serviceability limit line was applied in a finite-element (FE)-program FENLAP to predict whether or not stable behavior occurs in the UGM. To calculate the stress in the UGL, a nonlinear elastic model (Dresden Model), which is described in the paper, was implemented into the FE-program. The effects of changing moisture content during Spring-thaw period and asphalt temperature on pavement structural response were investigated. Additionally, permanent deformation calculations for the UGL were performed taking the stress history into consideration. The results clearly demonstrate that, for pavement constructions with thick asphalt layers, there is no risk of rutting in the granular base, even at high number of load repetitions. The study showed that the proposed design approach is a very satisfactory simple method to assess the risk against rutting in the UGL, even without the calculation of the exact permanent deformation of the pavement construction.

Cette thèse concerne la modélisation du comportement mécanique de poudres ductiles au cours de leur mise en forme par le procédé de compression à froid. L’approche utilisée consiste à modéliser la poudre par un échantillon numérique de 50 particules sphériques distribuées aléatoirement. Chaque particule est maillée en éléments finis avec un comportement élasto-plastique et interagit par contact avec ses voisines. Sous l’effet du chargement, les particules se déforment et subissent de grandes transformations.La méthode développée consiste à imposer aux frontières de l'échantillon numérique des conditions aux limites en déplacement ou en force afin de simuler une réponse continue équivalente en termes de contrainte-déformation pour plusieurs chemins de chargement depuis les faibles valeurs de densité relative jusqu'aux fortes valeurs. Outre le fait que la méthode permet d’appliquer par la simulation des chemins de chargement inaccessibles aux techniques expérimentales, elle fournit de nombreuses indications sur la phénoménologie du comportement aux frontières de l'échantillon en relation avec la morphologie des particules déformées plastiquement et l’évolution des surfaces de contact inter-particules.Cette réponse obtenue sera étudiée au moyen des notions de l'élasto-plasticité classique, à savoir la surface de charge et la loi d'écoulement. L'accent est mis sur l’analyse de l’écoulement plastique et l'existence du potentiel plastique. Les résultats montrent que la direction d’écoulement est approximativement unique pour les états de contrainte éloignés du point de chargement. En revanche, la direction d’écoulement cesse d’être unique à proximité de point de chargement, ce qui révèle le caractère anisotrope de l’écrouissage. Ce comportement, propre aux poudres ductiles, est intimement lié à la déformation des grains. La dégradation anisotrope des propriétés mécaniques résultant du mécanisme de dilatance est également étudiée. / This thesis is concerned with the modelling of the mechanical behaviour of ductile powders during cold compaction process. A numerical method was implemented, in which a powder sample was assimilated to an assembly of 50 meshed particles in a cubic box which was submitted to compressive loadings simulated by means of the finite element method. Each particle was meshed and assigned an elastic-plastic constitutive behaviour. Particles interacted through mechanical, frictional contact and underwent large transformation during loading.The method developped in the thesis consisted in imposing boundary conditions as displacements or forces on the bounding walls of the numerical sample. The mechanical response of the sample was averaged to obtain the equivalent response of a continuum in terms of stress and strain. Various loading paths, including loading paths that are technically not attainable by experimental means, were applied up to different values of relative densities, from low (60%) to high values (98%). Much information on the phenomenological behaviour of the numerical sample could be obtained, in relation with the morphology of deformed particles and the evolution of contact surfaces.The results were interpreted based on the concepts of classical elasto-plasticity, i.e., yield surface and flow rule. The main focus of the study was the analysis of incremental plastic flow and the existence of a plastic potential. Results showed that the flow rule postulate, based in the plastic potential, could be considered as valid for stress states relatively far from the loading point. In the vicinity of the loading point, the direction of the plastic strain increment vector ceases to be unique. This behaviour, which is an original feature of ductile powders mechanical behaviour, was attributed to the anisotropic strain-hardening processes at stake, related to the anisotropic formation of contact surfaces between particles. The drop in mechanical properties related to dilatancy was also studied.

Ce travail de thèse est consacré à l'étude des écoulements biphasiques grains-gaz dans un régime de blocage (jamming). Ce régime caractérisé par un écoulement interrompu en temps fini et fréquemment rencontré dans la nature ou les applications industrielles présente un caractère stochastique. Sa compréhension nécessite la mesure de son débit, de ses fluctuations et de la probabilité d'arrêt liée à la géométrie de confinement, à la microstructure granulaire et au gaz interstitiel. Une approche numérique discrète couplant la méthode de la dynamique des contacts non-régulière (Non Smooth Contact Dynamics) pour la dynamique des particules et une méthode mésoscopique de type homogène équivalente pour le gaz est développée. La statistique d'écoulement obtenue est en bon accord avec des résultats expérimentaux de la littérature : le débit vérifie une loi de puissance croissante en fonction de l'ouverture avec un débit moyen affecté par la présence du gaz. Ces résultats statistiques quantitatifs sont appliqués à l'estimation du taux de relocalisation du combustible nucléaire fragmenté et confiné dans son tube-gaine présentant un ballonnement local suite à une situation accidentelle de type Perte de Réfrigérant Primaire. / In this work, we investigate particle-gas two-phase flows in the jamming regime where the flow stops in finite time. In this regime, which occurs quite often in nature and industrial applications, the flow is stochastic and needs therefore to be characterized by the jamming probability as well as the flow rate and its fluctuations that depend on the confining geometry, granular microstructure and gas properties. We developped a numerical approach based on the coupling of the Non Smooth Contact Dynamics for the solid phase and a mesoscopic method for the gas phase. We find that the flow rate as a function of the opening is well fit by a power law in agreement with reported experimental data. The presence of a gas affects only the mean flow rate, the flow statistics being sensibly the same as in the absence of the gas. We apply our quantitative statistical results in order to estimate the relocation rate of fragmented nuclear fuel inside its cladding tube as a result of a local balloon caused by an accident (loss-of-refrigerent accident).

La dégradation géométrique de la voie ferrée sur les Lignes à Grande Vitesse (LGV) est un phénomène qui entraîne des coûts de maintenance importants pour assurer une qualité de circulation et de sécurité. Cette dégradation géométrique est imputable, en partie au tassement de la couche de ballast qui constitue l'une des parties de la voie ferrée.Le tassement du ballast est un phénomène difficile à estimer et à prédire car sa nature granulaire et ses caractéristiques particulières entraînent des variabilités de ses propriétés mécaniques. Ce travail de thèse est consacré à l'étude du comportement mécanique du ballast, pour le développement d'un modèle prédictif du tassement sur voie ferrée à partir de la caractérisation mécanique initiale du matériau (grâce à l'utilisation du pénétromètre léger Panda) et des sollicitations auxquelles la voie est soumise. Grâce à l'étude paramétrique développée sur un banc d'essais à échelle réelle, nous avons constitué et validé un modèle de prédiction qui est basé sur une loi de relaxation logarithmique.La modélisation numérique par éléments discrets en utilisant la méthode de la Dynamique des Contacts permet d'étudier les propriétés mécaniques du matériau ballast à l'échelle des grains. L'analyse des déformations transitoires montre une dépendance claire de la déformation moyenne par rapport à la contrainte appliquée et au rapport d'aspect en raison de l'action du frottement aux frontières. Les fluctuations de ces déformations, sont très importantes et semblent évoluer avec la déformation moyenne. Finalement, la pertinence de la caractérisation du ballast à partir de l'essai Panda a été vérifiée à partir d'une étude paramétrique sur les mécanismes d'enfoncement grâce à l'utilisation de cette approche numérique discrète. / Railway track degradation on high-speed lines is a phenomenon which causes high maintenance costs to ensure quality traffic and safety.This geometric degradation is due in part to the vertical strain (settlement) of the ballast layer, which compose the track.The ballast settlement is a difficult phenomenon to estimate and predict because its granular nature and characteristics of this layer induces variability of its mechanical properties.This work is devoted to the study of the mechanical behavior of ballast, to develop a predictive model of track settlement from the initial mechanical characterization of the material (by means a light penetrometer Panda) and the loading on track. By means several test on a full-scale model of railway track, we established and validated a predictive model based on a logarithmic relaxation law. Numerical modeling by discrete element method using the Dynamic Contacts allows to study the mechanical properties of ballast at grain-scale.Transient deformation analysis shows a clear dependence of the average deformation with respect to the stress and aspect ratio as a result of the frictional feedback at the boundaries. Fluctuations of these deformations are significant and seem to evolve with the average deformation. Finally, the relevance of the ballast characterization using the Panda test has been verified by means a parametric study on mechanisms during the penetration process by discrete element method.