In this work, we design a multiscale simulation method based on the Generalized Multiscale Finite Element Method (GMsFEM) for numerical modeling of fluid seepage under permafrost condition in heterogeneous soils. The complex multiphysical model consists of the coupled Richards equation and the Stefan problem. These problems often contain heterogeneities due to variations of soil properties. For this reason, we design coarse-grid spaces for the multiphysical problem and design special algorithms for solving the overall problem. A numerical method has been tested on two- and three-dimensional model problems. A a quasi-real geometry with a complex surface is considered for the three-dimensional case. We demonstrate the efficiency and accuracy of the proposed method using several representative numerical results.
During construction works, it is advisable to prevent strong thawing and an increase in the moisture content of the foundations of engineering structures in the summer. Since the density of water and ice differ, due to the difference bulging of the foundation sections can occur when it freezes back in winter. In this work, the effect of fiber-reinforced piles on the thermal field of the surrounding soil is investigated numerically; that is, the study of the influence of aggregates with high and low thermal-physical properties on the temperature of frozen soils is conducted. Basalt and steel fiber reinforcement are compared. The difficulty of this work is that the inclusions inside piles are too small compared to the pile itself. Therefore, to solve the Stefan problem, a generalized multiscale finite element method (GMsFEM) was used. In the GMsFEM, the usual conforming partition of the domain into a coarse grid was used. It allowed reducing problem size and, consequently, accelerating the calculations. Results of the multiscale solution were compared with fine-scale solution, the accuracy of GMsFEM was investigated, and the optimal solution parameters were defined. Therefore, GMsFEM was shown to be well suited for the designated task. Collation of basalt and steel fiber reinforcement showed a beneficial effect of high thermal conductive material inclusion on freezing of piles in winter.
We consider the multispecies model described by a coupled system of diffusion–reaction equations, where the coupling and nonlinearity are given in the reaction part. We construct a semi-discrete form using a finite volume approximation by space. The fully implicit scheme is used for approximation by time, which leads to solving the coupled nonlinear system of equations at each time step. This paper presents two uncoupling techniques based on the explicit–implicit scheme and the operator-splitting method. In the explicit–implicit scheme, we take the concentration of one species in coupling term from the previous time layer to obtain a linear uncoupled system of equations. The second approach is based on the operator-splitting technique, where we first solve uncoupled equations with the diffusion operator and then solve the equations with the local reaction operator. The stability estimates are derived for both proposed uncoupling schemes. We present a numerical investigation for the uncoupling techniques with varying time step sizes and different scales of the diffusion coefficient.
10 December 2022 in Journal of Computational and Applied Mathematics
In this paper, we consider the dual continuum fluid filtration problem. The mathematical model is described by a coupled system of partial differential equations for two pressure fields. We propose a novel PINN-based partial learning approach to solve this problem. We employ a partially explicit time scheme, where the implicit scheme is applied to a highly permeable continuum, and the explicit scheme is used for the low permeable continuum. We utilize discrete-time physics-informed neural networks (PINNs) to compute the implicit part of the problem that is challenging to solve directly. By leveraging the slow variations in the low permeable continuum and the robustness of PINNs to data noise, we decouple the equations and solve them sequentially. At each time layer, we first solve the implicit part and then treat it as a known function to solve the explicit part. The explicit part is spatially discretized using the finite element method with standard linear basis functions. We utilize transfer learning to accelerate the computation of the implicit part using the parameters from the previous time layer as the initial parameters for training the current time layer. To test the proposed method, we consider two two-dimensional model problems. For each problem, the method achieved good accuracy, with transfer learning significantly speeding up the computation.
8 May 2024 in Journal of Computational and Applied Mathematics