A new way to parameterize hydraulic conductances of pore elements: A step towards creating pore-networks without pore shape simplifications. (July 2017)

• Record Type: Journal Article
• Authors: Miao, Xiuxiu; Gerke, Kirill M.; Sizonenko, Timofey O.
• Is Part Of: Advances in water resources. Volume 105(2017)
• Page Start: 162
• Page End: 172
• Publication Date: 2017-07
• Subjects: Pore-network -- Dimensionless hydraulic conductance -- X-ray microtomography -- Porous rocks -- Permeability; Hydrology -- Periodicals; Hydrodynamics -- Periodicals; Hydraulic engineering -- Periodicals; 551.48
• DOI: 10.1016/j.advwatres.2017.04.021
• Languages: English
• ISSNs: 0309-1708

Abstract

Highlights: 3292 pore throat cross-section elements were extracted from 3D images of rocks. Hydraulic conductances were computed numerically on all 3292 cross-sections. Circularity and convexity were the best predictors of hydraulic conductances. A novel way to parameterize conductances using neural network was proposed. Novel approach resulted in 90% of predictions lying within the 20% error bounds. Abstract: Pore-network models were found useful in describing important flow and transport mechanisms and in predicting flow properties of different porous media relevant to numerous fundamental and industrial applications. Pore-networks provide very fast computational framework and permit simulations on large volumes of pores. This is possible due to significant pore space simplifications and linear/exponential relationships between effective properties and geometrical characteristics of the pore elements. To make such relationships work, pore-network elements are usually simplified by circular, triangular, square and other basic shapes. However, such assumptions result in inaccurate prediction of transport properties. In this paper, we propose that pore-networks can be constructed without pore shape simplifications. To test this hypothesize we extracted 3292 2D pore element cross-sections from 3D X-ray microtomography images of sandstone and carbonate rock samples. Based on the circularity, convexity and elongation of each pore element we trained neural networks to predict the dimensionless hydraulic conductance. The optimal neural network provides 90% of predictions lying within the 20% error bounds compared against direct numerical simulation results. Our novel approach opens a new way to parameterize pore-networks and we outlined future improvements to create a new class of pore-network models without pore shape simplifications. Graphical abstract: