CEE Seminar - Deep-Learning Enabled Multiscale Poromechanics: From Brittle Fracture to Ductile Flow
Many geomechanical applications, such as geological disposal of nuclear waste and CO2, require reliable predictions of the multiscale thermo-hydro-mechanical responses of fluid-infiltrating porous media exposed to extreme environments. While hierarchical multiscale method such as DEM-FEM and FEM2 are proven to be effective to link simulations between two length scales, the multi-porosity nature of fractured porous media often leads to complex geometrical attributes that spans multiple length scales [1,2,3,4]. This multiscale coupling is particularly difficult to model for anisotropic materials where the principle direction of material responses is sensitive to the size of representation elementary volume [5]. In this seminar, we will present an alternative homogenization approach that takes advantage of deep reinforcement learning....