Data-Driven multiscale modeling in mechanics

We present a Data-Driven framework for multiscale mechanical analysis of materials. The proposed relies on the formulation in mechanics (Kirchdoerfer and Ortiz 2016), with material data being directly extracted from lower-scale computations. Particular emphasis is placed two key elements: parametrization history, optimal sampling state space. demonstrate an application prediction behavior sand, prototypical complex history-dependent material. In particular, model able to predict response under nonmonotonic loading paths, compares well against plane strain triaxial compression shear banding experiments.