Abstract
Modern approaches to the modeling of composites are no longer limited to the use of a single approach for the whole structure or for all the degradation mechanisms. On the contrary, modern advances enable the definition of truly multiscale models in order to describe the degradation. Thus, homogenized models can be rigorously deduced from the underlying micromechanics. In the past few years, LMT-Cachan has made a number of contributions to the three key points of these multiscale approaches: (1) the improvement of the reference model on the fine scale, (2) the definition of a controlled correspondence between the scales, and (3) the definition of the associated homogenized model. Here, the complete approach is formalized as a modeling pyramid. Each mechanism of degradation is described on the more relevant scale within an ‘hybrid micromechanical model’. Based on the reference modeling, constitutive laws can be transfered within the unique framework of damage mechanics for being applied within commercial softwares. As an illustration, we focus more specifically on the homogenized law obtained for transverse cracking. The constitutive law and the material parameters issued from the homogenization, which define the model on the higher scale, are reviewed. Their identification is studied in detail. An important key point of the pyramidal approach appears here. Since it allows the interpretation of every quantity on different scales (both at the micromechanical and at the mesomechanical scales), the most relevant scale can be used for the identification of a chosen property. We limit ourselves to a ‘classical’ identification. We mean by classical identification a procedure based on straight specimens. This process, to a certain extent, uses a parametric simulation of the nonlinear model based on a finite element representation of the test samples. The complete model is then used for the simulation of an industrial sample with hole. That example emphasizes the interest of underlying micromechanial variables for experimental validation.