Simultaneous Design of Gas Separation Membranes and Schemes through Combined Process and Materials Informatics

Abstract

The performance of membrane separation technology is dependent on the membrane material and process structureaspects that have not typically been scrutinized simultaneously. Therefore, this study was aimed at concurrently designing membrane-separation-related materials and process structures. Machine learning was adopted to predict the gas permeability of new membrane materials and to design high-permeability materials, with systems exhibiting noteworthy separation performance identified from ∼10 million new material candidates. Furthermore, to simultaneously optimize the separation-enabling membrane materials and process structures, a method was developed to implement a machine-learning-based superstructure approach, and an efficient Bayesian-optimization-derived structure search was performed. Consequently, combinations of process structures and novel membrane materials that satisfied specific gas-separation requirements were discovered, suggesting the integration of process and material informatics. This study underlines the viability of simultaneously optimizing the design of membrane materials and process structures using experimental data, virtual molecule candidate materials, process simulations, machine learning, and Bayesian optimization.