Computational aspects of a damage mesomodelling for laminates Bernard Douchin, Pierre Ladevèze L.M.T. - Cachan, E.N.S. Cachan / C.N.R.S. / Université Paris 6 61, avenue du Président Wilson, 94235 Cachan Cedex, France Abstract The damage of composites in general -and of laminates in particular- involves several complex mechanisms. Various damage mechanisms co-exist and the simulation of these structures requires special precautions. One of the key points is the choice of a scale able to represent all the involved phenomena. A specific approach -initiated in the L.M.T. - Cachan and presented in several previous papers- bases the laminates modelling on the meso-scale [1]. This meso-modelling for laminates represents any stacking sequence as successive homogeneous layers throughout the thickness separated by interlaminar interfaces. The two basic constituents behaviours are described by internal variables which allows to represent all the involved damage mechanisms : fibre breaking, matrix micro-cracking parallel to the fibre direction, adjacent layers debonding [2], [3]. The damage mesomodel is a semi-discrete modelling approach for which the damage state is locally uniform within the mesoconstituents. The damage variables evolution is described by damage models with delay effect, so as to achieve the independence of the numerical results vis-à-vis the F.E. mesh [4]. The same models are notably used in dynamics too [5]. The implementation of the here-above modelling have lead to the realisation of different prototype softwares [6], [7]. Although the results show a good correlation between models and experiences, some numerical difficulties remain, specifically related to the treatment of the strains and damages localisation. We thus here focus on the laminates computation aspects. The classical incremental algorithms being inadequate for the computation of high localisations, a more robust strategy is developed. An error estimator for the elasto- viscoplastic with softening behaviours is introduced, the efficiency of this computations being coupled with a fine control of the computational parameters. In order to increase the numerical robustness the LArge Time INcrement (LATIN) method is selected as computation strategy [8]. This method -introduced in 1985- have shown good robustness and efficiency for several problems, in particular in the (visco)plastic cases ; we present here an extension to the softening behaviours, applied on a one dimension model problem. The LATIN method uses a drastically different scheme from the classical step by step approaches. It's an iterative procedure which defines at each iteration an approach solution on the whole time-space domain. We start the iterative scheme with the solution of the elasticity problem. The basic idea is to segregate the difficulties : the equations are split into two groups, the non-linear equations group and the global in space variable equations group. Each iteration will consist in two stages : we find alternatively a solution of the first, then of the second group, by a local stage followed by a linear global stage. The selection of