Permeability prediction of soft clay based on digital models reconstructed by an improved Markov chain Monte Carlo method

3D digital models can facilitate the understanding of micro- and nanoscale pore structures and enhance the permeability prediction accuracy of geological materials. However, the reconstruction of digital models using the conventional Markov chain Monte Carlo (MCMC) algorithm cannot well reflect the microstructure of the heterogenous media and needs to be improved. In this study, referring to the box-counting method (BCM) of fractal theory, the conventional probability distribution function was extended to different scale probability functions, and the concept of value paths was introduced to control the reconstruction quality. Subsequently, a comparison to the conventional MCMC algorithm was conducted based on the statistical function and the pore fractal dimension to verify the effectiveness of the improved algorithm. Finally, the reconstruction model was applied in finite element simulation to predict the permeability of soft soil. The verification results showed that the improved MCMC algorithm can well realize the reconstruction of heterogeneous porous media since the squared correlation coefficients of statistical functions of the 3D reconstruction and the original image are above 0.98, and the pore fractal dimension of reconstructions using the improved MCMC was closer to the original image than that using the conventional MCMC. Furthermore, in permeability prediction, permeability simulations using images reconstructed with an improved MCMC algorithm are closer to the experimental values, with a maximum prediction error of 1.87 times (that of MCMC was 4.21 times) the corresponding experimental value.