Multiscale parameter identification

In this work a multiscale approach is introduced which allows for the identification of small scale mechanical properties by means of large scale test data. The proposed scheme is based on the computational homogenization method in which a small scale representative volume element is related to each large scale material point and the large scale material response is directly obtained via homogenization of the small scale field variables. Application of this computational homogenization method usually requires that the microstructure of the material be well characterized, i.e., that the constitutive behavior of all constituents of the heterogeneous material is known. This condition is circumvented here by the solution of an inverse optimization problem, which provides the fine scale material properties as a result. Therefore the objective function compares large scale experimental results to field values, simulated with the computational homogenization method. Discrete analytical expressions for the sensitivities are derived, and the performance of different gradient-based optimization algorithms is compared for linear elastic problems with various microstructures