Doctor of Philosophy (PhD)


Chemical Engineering

Document Type



Increasingly, imaging technology allows porous media problems to be modeled at microscopic and sub-microscopic levels with finer resolution. However, the physical domain size required to be representative of the media prohibits comprehensive micro-scale simulation. A hybrid or multiscale approach is necessary to overcome this challenge. In this work, a technique was developed for determining the characteristic scales of porous materials, and a multiscale modeling methodology was developed to better understand the interaction/dependence of phenomena occurring at different microscopic scales. The multiscale method couples microscopic simulations at the pore and sub-pore scales. Network modeling is a common pore-scale technique which employs severe assumptions, making it more computationally efficient than direct numerical simulation, enabling simulation over larger length scales. However, microscopic features of the medium are lost in the discretization of a material into a network of interconnected pores and throats. In contrast, detailed microstructure and flow patterns can be captured by modern meshing and direct numerical simulation techniques, but these models are computationally expensive. In this study, a data-driven multiscale technique has been developed that couples the two types of models, taking advantage of the benefits of each. Specifically, an image-based physically-representative pore network model is coupled to an FEM (finite element method) solver that operates on unstructured meshes capable of resolving details orders of magnitude smaller than the pore size. In addition to allowing simulation at multiple scales, the current implementation couples the models using a machine learning approach, where results from the FEM model are used to learn network model parameters. Examples of the model operating on real materials are given that demonstrate improvements in network modeling enabled by the multiscale framework. The framework enables more advanced multiscale and multiphysics modeling – an application to particle straining problems is shown. More realistic network filtration simulations are possible by incorporating information from the sub-pore-scale. New insights into the size exclusion mechanism of particulate filtration were gained in the process of generating data for machine learning of conductivity reduction due to particle trapping. Additional tests are required to validate the multiscale network filtration model, and compare with experimental findings in literature.



Committee Chair

Thompson, Karsten

Available for download on Friday, January 18, 2019