Machine Learning-assisted Study Of The Gas Adsorption And Separation Performance Of Amorphous Carbon

Porous materials as adsorbent for gas separation has a wide and important application in industrial production.The development of advanced adsorbent with large gas uptake and high selectivity is required for efficient gas separation.Amorphous carbon is a commonly used adsorption material in industry because of its large specific surface area,stable physical and chemical properties and low production cost.However,the microscopic disorder of carbon materials makes it difficult to study the relationship between properties and structure at the atomic level.The investigation on amorphous carbon materials to meet specific needs still relies on trial-and-error process.Combining high-throughput computing with machine learning techniques can efficiently explore the potential phase space,analyze a large amount of data to reveal conformational relationships and significantly improve efficiency of material discovery.In this thesis,a hybrid reverse Monte Carlo method is combined with molecular dynamics simulations to achieve rapid construction of nitrogendoped amorphous carbon models.Taking an important carbon capture system(CO2/N2/CH4 mixture)in energy and environment fields as a typical research system,the adsorption separation performance of small molecules of carbon dioxide,nitrogen and methane in a large number of amorphous carbon structures is simulated based on the grand canonical Monte Carlo simulation and ideal adsorption theory.The relationship between various structural features and the adsorption capacity and selectivity of gas mixtures was analyzed using machine learning.It was found that the average pore size and specific surface area of amorphous carbon are the main factors affecting the adsorption separation performance,whereas the doping amount of nitrogen is a minor factor.High adsorption capacity and high selectivity are difficult to be achieved simultaneously.The amorphous carbon materials with improved separation performance are expected to large specific surface area and proper pore size which is closed to the diameter of gas molecular.