Accelerated Screening of Covalent Organic Frameworks for Ethylene Glycol/1,2-Butanediol Separation by Interpretable Machine Learning.

The separation of ethylene glycol (EG) and 1,2-butanediol (1,2-BDO) azeotrope in the synthesis process of EG via coal and biomass is becoming increasingly commercial and of environmental importance. Selective adsorption is deemed as the most promising method because of energy savings and environment favorability. In this study, we developed an interpretable decision tree (DT) model to facilitate high-throughput screening of covalent organic frameworks (COFs) as adsorbents for the separation of EG/1,2-BDO mixtures, achieving an R² value of 0.96. The interpretable decision tree analysis has shown that using the difference in isosteric heat (ΔQ_st) between EG (Q_st-EG) and 1,2-BDO (Q_st-BDO), combined with the largest cavity diameter (LCD), is effective for selecting the optimal COFs for EG/1,2-BDO separation. COFs with ΔQ_st greater than 6.5 kcal/mol and an LCD ranging from 3.6 to 4.8 Å typically exhibit superior performance and can serve as preselection criteria to accelerate the screening process. Six COFs with high EG working capacity and exceptional adsorption selectivity for EG/1,2-BDO were selected using the selection principle. All the selected COFs containing strong electronegative groups. The electronegative groups can significantly amplify the disparity in adsorption strength between EG and 1,2-BDO, thereby boosting separation efficiency. The principles proposed in this work can be used to guide the design of COFs for effective separation of EG and 1,2-BDO.