Multiscale Modeling of Granular Materials

de Yang Liu

Granular materials have a “discrete” nature whose global mechanical behaviors are originated from the grain scale micromechanical mechanisms. The intriguing properties and non-trivial behaviors of a granular material pose formidable challenges to the multiscale modeling of these materials. Some of the key challenges include upscaling of coarse-scale continuum equation form fine-scale governing equations, calibrating material parameters at different scales, alleviating pathological mesh dependency in continuum models, and generating unit cells with versatile morphological details. This dissertation aims to addressing the aforementioned challenges and to investigate the mechanical behavior of granular materials through multiscale modeling. Firstly, a three-dimensional nonlocal multiscale discrete-continuum model is presented for modeling the mechanical behavior of granular materials. We establish an information-passing coupling scheme between DEM that explicitly replicates granular motion of individual particles and a finite element continuum model, which captures nonlocal overall response of the granular assemblies. Secondly, a new staggered multilevel material identification procedure is developed for phenomenological critical state plasticity models. The emphasis is placed on cases in which available experimental data and constraints are insufficient for calibration. The key idea is to create a secondary virtual experimental database from high-fidelity models, such as discrete element simulations, then merge both the actual experimental data and secondary database as an extended digital database to determine material parameters for the phenomenological macroscopic critical state plasticity model. This expansion of database provides additional constraints necessary for calibration of the phenomenological critical state plasticity models. Thirdly, a regularized phenomenological multiscale model is investigated, in which elastic properties are computed using direct homogenization and subsequently evolved using a simple three-parameter orthotropic continuum damage model. The salient feature of the model is a unified regularization framework based on the concept of effective softening strain. The unified regularization scheme is employed in the context of constitutive law rescaling and the staggered nonlocal approach to alleviate pathological mesh dependency. Lastly, a robust parametric model is presented for generating unit cells with randomly distributed inclusions. The proposed model is computationally efficient using a hierarchy of algorithms with increasing computational complexity, and is able to generate unit cells with different inclusion shapes.