Utilizing physics-informed neural network and geotechnical distance field for solving three-dimensional nonlinear consolidation

Abstract

Solving three-dimensional (3D) nonlinear consolidation is complex and computationally expensive. This study proposes a framework for solving 3D nonlinear consolidation by utilizing an improved physics-informed neural network with hard constraints coupling with machine learning based geotechnical distance functions for threedimensional spatial interpolation. The performance of the developed framework was assessed by comparing pore water pressure data between the developed framework and those obtained from COMSOL Multiphysics. In addition, the impact of vertical hydraulic conductibility heterogeneity, compression index, and void ratio on long-term settlement was also evaluated and discussed. It was found that the proposed framework showed a reliable estimation of the 3D distribution of pore water pressure across the 3D domain, achieving results that are comparable to data obtained from COMSOL. In addition, the heterogeneity of hydraulic conductivity can be successfully considered using the developed framework, which enables assessing the long-term settlement of a clay deposit with high uncertainty of hydraulic conductivity. Overall, the developed framework shown in this study can be applied to complex consolidation problems with low computational costs and high accuracy.