Yanlong Wang, Jinhua Liu, and Ting Zhang, and Jin Zhang


Multiple-point geostatistics, reconstruction, porous media, data template, training image, pore network model


Multiple-point geostatistics (MPS) is a new branch of geoscience, allowing extracting the characteristics from training images and copying them to simulated images. To reconstruct 3-D porous media is quite important to the science of mechanisms of fluids flow and industrial fields. However, accurate reconstruction of porous media is not easy using traditional interpolation methods. Therefore, a real 2-D micro-CT image and MPS are used to reconstruct the 3-D structures of porous media. A cross-section of porous media, with the resolution of 10 microns, obtained by micro-CT scanning is used as an original training image. By using MPS and extracted sample points from each former regenerated training image, each newly reconstructed image is taken as a new training image to predict its next layer. Then, the stochastic 3-D images of porous media are generated by stacking all the reconstructed layers successively. The permeability of the target image composed of real sandstone volume data and the reconstructed images was compared using lattice Boltzmann method (LBM). A pore network model was extracted by using maximal ball algorithm, by which some important parameters of reconstructed images were compared with those of the real volume data. The experimental results show that although the structures of porous media reconstructed through this method are stochastic, all the reconstructed results have similar structures, which are close to those of real volume data obtained by micro-CT scanning, demonstrating better reconstructed results than what two-point geostatistics does.

Important Links:

Go Back