Machine Learning-Based Prediction of Throat Clogging by Fine Migration in Porous Media

Summary Clogging in porous media, driven by the migration and deposition of fine particles, leads to reduced permeability and impacts fluid flow, affecting applications from groundwater management to energy extraction. This study presents a Machine Learning Model (MLM) designed to predict throat clogging and permeability impairment within porous media under varying hydro-physical conditions. Using a Computational Fluid Dynamics-Discrete Element Method (CFD-DEM) framework, we generated extensive data on pore structures, particle sizes, flow velocities, and surface roughness. The MLM, comprised of three sub-models, predicts clogging dynamics and associated permeability reductions by analyzing fine particle behavior at the pore level. Each sub-model was trained using data from CFD-DEM simulations and validated to ensure high predictive accuracy. Our approach achieved a throat classification accuracy of 95.2% and an R-squared value of 0.9981 for permeability impairment predictions. Unlike traditional CFD-DEM simulations, which require days to complete, the MLM provides reliable results in minutes, demonstrating its efficiency and practicality. This study underscores the potential of machine learning in advancing the predictive modeling of fine particle clogging, providing timely insights for managing clogging effects in environmental, industrial, and energy applications.