Inverse design of porous materials for natural gas storage using machine learning and molecular simulation

Abstract

Adsorbed Natural Gas (ANG) storage is an important technology for energy and environment applications because it can operate at room temperature and lower pressure compared to conventional Compressed Natural Gas (CNG) and Liquefied Natural Gas (LNG). Although several attempts have been made to utilize the adsorbents such as zeolite and metal-organic frameworks (MOFs), no structure has been reached the target methane deliverable capacity value (315 cm$^3$(STP)cm$^{-3}$) proposed by the U.S. Department of Energy. Moreover, millions of nanoporous materials have been developed using computer-aided design, but they did not even reach the experimental record. It is our opinion that this absence of high-performance materials is caused by the insufficient size of material space and the difficulty of implementing additional factors such as flexibility of porous materials. To overcome these problems, the first and second studies in this thesis focus on the development of inverse design system which has a much larger pool of materials. (1) A neural network model has been developed for the inverse design of porous materials, and (2) the available porous material space has been expanded to 247 trillion using topology-guided design. And the last chapter focuses on (3) the storage of natural gas using the flexibility of metal-organic frameworks. Flexible wine-rack MOFs were designed by computer-aided method, and their deliverable capacities were evaluated using molecular simulations. As a result, about 100 record-breaking candidates were found, including both rigid and flexible materials.