Machine Learning-Assisted Exploration of a Two-Dimensional Nanoslit for Blast Furnace Gas Separation

Abstract

Diffusion-induced gas separation is crucial for industrial applications, while the determination of specific conditions is still challenging. Here, molecular dynamics simulation data were used to train machine learning models to identify the effective separation conditions for blast furnace gas confined in nanosilts with different absorption strengths (graphene and graphene oxide). The diffusion coefficients and exponents of the blast furnace gas were obtained as a database by molecular dynamics (MD) simulations. Several environmental and structural controlling factors (such as temperature, layer distance, atomic number, ionization potential, etc.) were extracted through importance analysis. And the relationships between these factors and diffusion properties were further established by the Kernel Ridge Regression algorithm. Based on the differences in diffusion coefficients, specific binary gas mixtures in the blast furnace gas with competitively high separation potential have been screened out by a trained machine learning model and verified by MD simulations. The simulation strategy can provide theoretical guidance for the structural design of membranes for diffusion-controlled gas separation.