MiMoDiM project aims at developing a new class of constitutive models able to account for microstructure complexity of granular materials across the scales. A physics informed use of complex network analysis and machine-learning techniques will enable to identify new relevant statistical microstructures descriptors from discrete element simulations and micro-tomography images. By focusing on the description of mesoscale mechanics, this project will offer a unique possibility to feed constitutive models withmicroscale data. By accounting for all failure modes of granular materials and for several microstructure reinforcement techniques, this project will contribute to improve both the safety and the sustainability of engineering structures such as earth dams, wind turbines, nuclear plants, pipelines or any other infrastructure built on (or buried in) geomaterials.