Cracking in reinforced concrete structures is a major concern, as it may impact the structural durability and sometimes also its functionality. The simulation represents an appropriate tool to answer the question of the control of the cracking. But it supposes to validate the numerical methods and especially the choice for constitutive laws. But softening laws for concrete (quasi-brittle material) imposes the use of regularization techniques, if the well-known mesh dependency effect wants to be avoided. But questions on their applicability on industrial structures (numerical cost as refined meshes are needed) and their domain of validity (calibration of the parameters) are still asked. For this PhD thesis, we propose to start from a computation on a reinforced concrete containment building and to define a methodology which includes a multi-scale approach to go from the full structure to Representative Structural Volumes. At this scale, a fine simulation will be investigated including an improvement in the modeling of the heterogeneities (prestressed tendons especially) and the development of appropriate regularization techniques. This methodology (multi-scale approach) will be general enough to be applied on any softening structures.