In this research, gas transport, diffusion, and adsorption through micro-structured porous materials with the help of data-driven methods are investigated. Different types of the geological porous media from organic-rich shale to tight carbonate and clays have been digitally analysed and simulated to characterize various fluid-solid interactions. Considering the heterogeneous structure of the many nature-made porous materials, a multi-scale pore network modeling approach has been presented which couples the effects of micro-pores, meso-pores, and fractures at the same time. Considering the high computational cost of the multi-scale multi-physical systems, machine learning (ML) is employed to make statistical surrogate models with minimal accuracy losses. Several physical features of the porous material have been predicted using deep convolutional neural networks based on the segmented images as input. Properties like absolute permeability, gas permeability, gas storage capacity and capillary pressure have been successfully predicted by the proposed machine learning model with averaged r-squared of around 0.9. In some of the cases like permeability, ML predicted values have been compared to the micro-scale laboratory experiments and relative error of 13 % has are reported which is reasonable considering 3 to 4 orders of magnitude lower computational cost. The outcome of this study is to equip researchers with a series of ML-assisted tools to accelerate numerical simulations of several fluid-solid interactions in porous materials. As an example, the proposed methodology can be used in screening of the suitable CO_2 subsurface storage sites based on analysis of the pore-scale images of shale-deposits.
- pore network modeling
- gas diffusion
- porous media
- machine learning
Multiscale modeling of gas transport in partially saturated heterogeneous media assisted by machine learning
Rabbani, A. (Author). 31 Dec 2021
Student thesis: Phd