AI-powered Tool for the Automated Estimation of Gas Diffusivities in Nanoporous Materials
Machine learning has recently seen an increased adoption in Natural Sciences and Technology, including the domain of Material Science, where the expected impact to the environment and human energy footprint is very significant.
The scope of the current project is the development of an automated process (and related tool) for the calculation of the properties (namely diffusivities) in novel nanoporous materials, and more specifically metal-organic frameworks (MOFs). These materials could make membranes for the separation of gas mixtures, a process which accounts for an impressive 10%-15% of the global energy consumption. Such a tool will facilitate the collection of massive diffusion data to train Machine Learning models for the prediction of diffusivity in MOFs.
Although Machine Learning has been incorporated in the study of nanoporous solids, such as metal-organic frameworks (MOFs) and Covalent-organic frameworks (COFs), the research has been focused on gas sorption properties,1,2 while gas diffusion is neglected. This has to do mainly with the implications involved with the calculation of diffusivity in high-throughput computations. The absence of automated processes that incorporate non-conventional molecular dynamics approaches, such as transition-state theory (TST), hampers the development of AI tools that could advance the field of materials for separation membranes.
Indicative tasks that the project implies:
(a) Identify a MOF structure database and perform some initial calculations with a transferable force field
(b) Development of tool that identifies the topological characteristics, such as the reaction axis that connects the center of the MOFs cages
(c) Development of automated process that uses knowledge from (b) (reaction axis) and performs the TST simulations
(d) Employ the tool to execute diffusion simulations in an extended MOF database3 and gather data
(e) Identify and gather features/descriptors to build the dataset for the ML training.
(f) Build and evaluate a predictive model based on the above feature set.
(1) Demir, H.; Keskin, S. Computational Insights into Efficient CO2 and H2S Capture through Zirconium MOFs. J. CO2 Util. 2022, 55, 101811.
(2) Krokidas, P.; Karozis, S.; Moncho, S.; Giannakopoulos, G.; Brothers, E. N.; Kainourgiakis, M. E.; Economou, I. G.; Steriotis, T. A. Data Mining for Predicting Gas Diffusivity in Zeolitic-Imidazolate Frameworks (ZIFs). J. Mater. Chem. A 2022, 10, 13697–13703.
(3) Chung, Y. G.; Haldoupis, E.; Bucior, B. J.; Haranczyk, M.; Lee, S.; Zhang, H.; Vogiatzis, K. D.; Milisavljevic, M.; Ling, S.; Camp, J. S.; Slater, B.; Siepmann, J. I.; Sholl, D. S.; Snurr, R. Q. Advances, Updates, and Analytics for the Computation-Ready, Experimental Metal-Organic Framework Database: CoRE MOF 2019. J. Chem. Eng. Data 2019, 64, 5985–5998.