Rechnergestütztes Design flexibler MOFs für die Trennung von Kohlenwasserstoffen.
Physikalische Chemie von Festkörpern und Oberflächen, Materialcharakterisierung
Zusammenfassung der Projektergebnisse
The efficient separation of hydrocarbon mixtures such as ethylene and ethane, propylene and propane, and xylene isomers is of a high interest of the petrochemical industry, but also highly challenging and energy consuming. New advanced porous materials - metal organic frameworks (MOFs) have great potential for the separation purposes, and simulation plays a guiding role in searching for the new materials with desirable properties. In this work, for the first time thousands of fully flexible models of MOFs from CoRE MOF database are thoroughly screened to reveal their separation performance, employing classical molecular dynamics simulations for calculating diffusion coefficient. Structure-property relationships are revealed during the screening, which shows the most preferable characteristics for the fastest diffusion through the MOFs. The simulations performed showed that high throughput computational screening of thousands of fully flexible MOFs is almost computationally prohibitively expensive. The computationally obtained results revealed 5 the most promising MOFs from CoRE MOF database for each pair m-xylene/p-xylene, o-xylene/p-xylene, ethylene/ethane, and propane/propylene. More importantly, the training database built with these structural characteristics. Machine learning approaches are employed for determining diffusion properties in MOFs, for this purpose a complete set of generalized classical force-field inspired descriptors (CFID) is employed. GradientBoostingRegressor was used to build a regression model for the data of diffusion coefficients of hydrocarbons studied and Classical Force-field Inspired Descriptors.
