Masonry structures are one of the most common housing type built all around the world. Nonetheless computational methodologies implemented in classical engineering soft-ware seem not to be adapted to fine analysis of masonry structures. The Non-Smooth Contact Dynamics method developed by JJ Moreau and M Jean (the Dis-crete Element Method implemented in the software LMGC90) can perform the modeling of divided media and thus seem particularly well adapted to this type of computation. This work aims to better understand the behavior of the algorithm used in the LMGC90 soft-ware in the resolution of this kind of problems. The impact on the local and global behaviors of structures are considered and discussed. Both the notion of reliability of model and the no-tion of quality of the simulation are focused on. This work is particularly based on the comparison between experimental and numerical re-sults obtained with the LMGC90 software. The test consists in the modeling of a simple dry masonry structure placed on a tilting table. Three interaction laws are used to model the Sig-norini-Coulomb condition. The stability of the wall modeled with rigid bodies is studied re-garding with the influence of numerical parameters calibrating the Gauss-Seidel algorithm used to resolve such a multi-contact problem. Local behaviors are also considered as well as CPU effectiveness.

The contact dynamics is a complex problem which brings many difficulties when it comes to simulation. A particular way to express the contact problem has been developed by Moreau and Jean. The LMGC90 software aims at performing computation using this formalism. Being initially written in Fortran, the software suffers some limitations due to the language. Furthermore its architecture allowed robustness but the scientific requirements are becoming so demanding that this structure reaches its limit. Here is presented an original architecture were the software is split in several parts, each referring to an operator or a service needed by the formulation of the physical problem. This architecture is made possible by using some programming features as Fortran modules, or a fake inheritance mechanism. The bulk and contact modelling part are separated as much as possible and each part generates algebraic systems and interactions in an anonymous way to feed the contact solver. The storing of data is also put apart from the other modules so that all data relevant to the computation of a body are gathered at only one place. This structure makes extensions easier to add and still allows to have complex strategies management. The current software already covers a wide range of application fields like granular material, masonry structures, fracture of heterogenous media and multi-physics couplings. The design work presented here would aims at rend the code more flexible and simplify the couplings with other software or libraries. The ultimate goal being to be able to have a dynamic model representation switch at run-time simulation.

We use contact dynamic simulations to perform a systematic investigation of the effects of particles shape angularity on mechanicals response in sheared granular ma-terials. The particles are irregular polyhedra with varying numbers of face from spheres to "double pyramid" shape with a constant aspect ratio. We study the quasi-static behav-ior, structural and force anisotropies of several packings subjected to triaxial compression. An interesting finding is that the shear strength first increases with angularity up to a maximum value and then saturates as the particles become more angular. Analyzing the anisotropies induced by the angular distributions of contacts and forces orientations, we show that the saturation of the shear strength at higher angularities is a consequence of fall-off of the texture anisotropies compensated by an increase of the tangential force anisotropy. This is attributed to the fact that at higher angularity, particles are bet-ter connected (or surrounded) leading to an increase of friction mobilization in order to achieve the deformation. Moreover, the most angular particles also have very few sides so that, this effect is enhanced by the increase of the proportion of face-side and side-side contacts with angularity.

As soon as we are interested in the simulation of interacting objects raises the issue of the choice of a relevant numerical model and of technical approaches suitable to the solving strategy. It concerns the choice of objects bulk behavior (which reduces to kinematic parametrization if they are considered as rigid), object shape description, contact behavior, contact detection technics and solving strategy (e.g. time evolution, time integration, contact solver, etc.) The principal difficulty comes from the choice of a model and numerical technics and strategies. However models and strategies are usually related. Various models and strategies are already presented in this book which explains why this chapter only focuses on specificities and sophistications necessary to the modeling of collection of objects with complex shapes described by polyhedra.