Machine Learning Predictive Model for Permeability Alteration Induced by Microplastics Migration in Porous Media

Summary Microplastics (MPs), are particles below 5 mm from larger plastic degradation that have a wide distribution and ecological impact. To understand their impact on porous media and groundwater quality, a novel machine learning-based approach was introduced. The approach uses a combination of Computational Fluid Dynamics and the Discrete Element Method (CFD-DEM) to generate a dataset that trains a Machine Learning Model (MLM) to predict the dynamics of throat clogging and permeability impairment resulting from the migration of microplastic particles. Using this approach, a dataset of over 190,000 data points was generated, covering factors such as porous media geometry, fine particle size, flow velocity, and fine particle concentration. The developed MLM accurately forecasts the temporal progression of clogged throats and permeability impairment with 95% accuracy in predicting clogged throats and a R-squared value of 0.99 in predicting permeability impairment. Constructed by interconnecting three Machine Learning Sub-Models (MLSMs), the initial MLSM classifies throats based on their size and location to identify clogged throats. Subsequently, a pore volume regression MLSM determines the pore volume at which each clogged throat becomes obstructed. Lastly, the permeability impairment regression MLSM predicts the reduction in permeability based on clogged throat information and associated pore volumes.