Accelerating fluid flow simulations through doubly porous media using a FEM-assisted machine learning approach

Abstract

This paper introduces an innovative methodology for simulating fluid flow through doubly porous media. This technique combines finite element (FE) simulations and machine learning (ML) methods. The approach involves training ML models with a dataset of FE simulation outcomes, enabling predictions of macroscopic permeability under varied conditions. This strategy addresses the limitations of conventional simulation methods, which suffer from high computational demands and lack of systematic insight. The approach's precision and efficiency are confirmed through numerical experiments, achieving a coefficient of determination of 0.998. The outcomes underline the potential of the FEM-assisted ML strategy to considerably enhance fluid flow simulations in porous media. The technique can find applications across diverse domains within civil engineering. Furthermore, developing an intuitive graphical user interface (GUI) streamlines the application of the proposed approach.