(18b) Machine Learning-Aided Discovery and Design of Metal-Organic Frameworks for Sour Gas Sweetening
Hydrocarbon sources, such as crude petroleum, natural gas, and biogas, can contain acidic hydrogen sulfide (H2S). Unacceptable amounts of the acid gas should be removed, and the process is referred to as sour gas sweetening. Adsorption-based separation of H2S using novel porous materials may provide tremendous opportunities in this critically important process. Specifically, the diverse chemical and geometrical properties of metal-organic frameworks (MOFs) hold great promises. While a handful of MOFs have been investigated experimentally and/or computationally for H2S removal to date, the vast materials space of MOFs resulting from the myriad combinations of metal clusters and organic linkers remain unexplored. Furthermore, the key chemical and geometrical features of MOFs for effective and efficient removal of H2S have not yet been fully unraveled to facilitate rational design of MOFs. To address these missing components, we employ machine learning with molecular simulations. As is known, to establish a successful machine learning model, it is important to include descriptors representative of the performance. In this study, our model considers several well-established structural descriptors such as the largest cavity diameter. Moreover, we develop new geometric and chemical descriptors to specifically capture the adsorption environments of MOFs that may contribute significantly to the adsorption strengths of MOFs to gases. Our results show that, by incorporating the aforementioned new descriptors, our machine learning model can be highly predictive of the sour gas separation. We anticipate that our study can pave the way for future discovery and synthesis of MOFs for sour gas sweetening.