Detecting Reactive Products in Carbon Capture Polymers with Chemical Shift Anisotropy and Machine Learning

Aminopolymers are attractive sorbents for CO2 direct air capture applications as their amines readily react with atmospheric levels of CO2 to form chemisorbed species. The identity of the chemisorbed species varies upon experimental conditions like amine chemistry, support material, CO2 loading, and humidity, forming a variety of carbonyl-type sites. 13C solid-state nuclear magnetic resonance (NMR) is often used to help elucidate the identity of the chemisorbed species however the chemical shift range for carbonyl sites is small and comparable to observed chemisorbed 13C peak widths. Herein, application of a 2D chemical shift anisotropy (CSA) recoupling pulse sequence (ROCSA) is used to obtain CSA tensor values at each isotropic chemical shift, overcoming the isotropic peak resolution limitation. CSA tensor values describe the local chemical environment and can readily differentiate between chemisorbed products. To aid this experimental technique, we also developed a k-nearest-neighbor (KNN) classification model to distinguish chemisorbed compounds via their CSA tensor parameters. The combination of 2D CSA measurements coupled with a KNN classification model enhances the ability to accurately identify chemisorbed products especially in the case of mixtures. This methodology is demonstrated on poly(ethylenimine) in a solid-support γ-Al2O3 exposed to CO2 followed by incomplete regeneration at 100 °C and shows a mixture of strongly bound chemisorbed products, ammonium carbamate and urea.