Abstract Membrane separation has been extensively studied as a cost-effective CO 2 separation method, and polysilsesquioxane (PSQ)-based membranes are expected to be robust membranes with high thermal and mechanical stability and processability. In this study, a prediction model for CO 2 permeance and CO 2 /N 2 permselectivity as target variables was generated by applying machine learning to experimental data collected in our previous studies as explanatory variables. On the basis of this model, two new urea-containing PSQ-based membranes were prepared, and their CO 2 separation performance was evaluated. Among them, a membrane synthesized through the 1:1 copolymerization of (3,6-dioxaoctane-1,8-diyl)bis- N - N ’-(triethoxysilylpropyl)urea and bis(triethoxysilyl)ethane demonstrated high performance, achieving a CO 2 permeance of 1.3 × 10 –6 mol m –2 s –1 Pa –1 (4.0 × 10 3 GPU) and a CO 2 /N 2 permselectivity of 13. A membrane was also prepared using (triethylamine-2,2’,2”-triyl)tris- N - N ’-(triethoxysillylpropyl)urea as a monomer, which resulted in inferior CO 2 separation performance. However, increasing the calcination temperature significantly increased the CO 2 permeance, whereas the CO 2 /N 2 permselectivity slightly decreased, likely because of the thermal degradation of the urea units, resulting in the formation of void spaces.
Kanematsu et al. (Fri,) studied this question.