Automated AI-driven screening of nanoporous materials for enhanced carbon dioxide adsorption

Abstract

Millions of crystalline nanoporous materials have been identified for carbon capture, making the task of measuring the adsorption performance of each individual nanopore using computer simulations wholly unfeasible. Furthermore, experimentally synthesising and calculating the adsorption properties of each sample is even more impractical. A screening framework is thus required to identify a smaller number of promising candidates for further investigation. In this presentation, we introduce our work which deploys several distinct mathematical techniques to efficiently characterise nanopore structures, leading to a rapid high throughput nanopore screening mechanism. Our automated cloud-based materials screening tool is composed of two distinct parts. Firstly, several computational geometric and topological descriptors are determined for each individual nanopore, allowing us to immediately discount samples with unfavourable structural properties. Next, Grand Canonical Monte Carlo (GCMC) simulations are deployed to calculate the target adsorption figures of merit. The computed results can be ingested into machine learning algorithms to produce a surrogate model to the original simulations, thus accelerating the estimation of desired adsorption properties of the candidate nanoporous materials.