The present study is devoted to the generalization of the Nonuniform Transformation Field Analysis (NTFA), a modelreduction approach introduced by the authors. First, the local fields of internal variables are decomposed on a reduced basis of modes. Second, the effective (average) dissipation potential of the phases is replaced by accurate approximations. The reduced evolution equations of the models, in other words the homogenized constitutive relations, can be entirely expressed explicitly in terms of quantities which are pre-computed "off-line". The example of creep of polycrystalline ice is used to assess the accuracy of the models. Their predictions, both the overall response and the local response, are shown to be in good agreement with full-field simulations with a significant speed-up.
Reduced-order modelsA common engineering practice in the analysis of composite (or polycrystalline) structures is to use effective or homogenized material properties instead of taking into account all details of the individual phase properties. Unfortunately when the individual constituents are nonlinear, the exact description of the effective constitutive relations requires the determination of the local fields (at the microscopic scale). For structural computations, the consequence of this theoretical result is that the two levels of computation, the level of the structure and the level of the unit-cell, remain intimately coupled. The nested resolution of these coupled problems (known as FE 2 analysis) is so far limited by their formidable size. It is therefore quite natural to resort to model-reduction techniques achieving a compromise between analytical approaches, which are costless but often very limited by nonlinearity, and full-field simulations which resolve all complex details of the exact solutions, but come at a very high cost. This is the aim of reduced-order modelling (ROM) in general ( [1, 2]). However, ROM is often understood as a way to reduce the cost of a numerical simulation. In homogenization problems, the objective is not only to reduce the computational cost, but also to arrive at an explicit constitutive model, where internal variables are identified and evolution equation for these internal variables are explicitly derived. The model which is reduced is not only the computational model, but also and primarily, the constitutive model.General procedures to achieve the model reduction do exist, at least for the computational model. However, without a proper physical insight in the problem to be reduced, it is likely that these general procedures will not deliver satisfactory answers. The model initially proposed by the authors ( [3]) and further developed in [7][8][9] and by other authors ( [4, 5]) is limited in scope to materials with a microstructure, comprised of constituents with a certain type of constitutive relations involving internal variables. In turn, the approximations on which the model is based are physically sound in this context. The three main ingredients of the model are as follows....