(235c) Machine Learning Model Based Guidelines for Material Selection in Membrane-Based Gas Separation Processes

Membrane processes have received much attention due to low cost for production and ease of operation in various gas separation processes [1]. To enhance gas separation performance, there has been active research in process design and optimization, as well as membrane material development. The selection of an appropriate membrane material, from numerous materials that have been developed, has the greatest impact on the performance of the membrane separation process [2]. Previous studies have mainly focused on physical properties of materials such as permeability and selectivity to select membranes efficiently [3]. However, various factors such as feed flow rate, pressure, and module length are also closely related to process performance in membrane processes. In particular, the trade-off relationship between product purity and recovery which are determined by these operating and design parameters is an important feature that must be considered in the membrane material selection procedure [4].

In this study, guidelines for membrane material selection from a process-scale perspective are proposed, and the optimal membrane is selected by applying this method to an actual material database The study consists of two parts: an analysis and optimization at the process level, and an application and validation using an open-source material database. First, a machine learning (ML)-based membrane process model is developed to derive feasible ranges of permeance and selectivity that can achieve a product purity of more than 99%. Next, an optimization model for product production cost is derived using the ML-based model. The optimization model derives operating and design parameters (feed flow rate, pressure, and module length) that minimize the production cost by given membrane properties, such as permeance and selectivity. The proposed methodology is applied to the “Clean, Uniform, Refined with Automatic Tracking from Experimental Database” (CURATED) [5] of covalent organic frame work (COF) to select the optimal membrane for green ammonia based H₂/N₂ separation process. The permeance and selectivity of each COF are derived from and molecular dynamics and Grand Canonical Monte Carlo simulations.

This study has the following scientific contributions. With the development of a membrane process model based on ML, a feasible operating range based on membrane performance can be derived according to a wide range of various variables. The model thus developed can be used as a guide line for new material development in the future.

Literature cited:

[1] C. Altintas, S. Keskin, Molecular Simulations of MOF Membranes and Performance Predictions of MOF/Polymer Mixed Matrix Membranes for CO 2 /CH 4 Separations, ACS Sustainable Chemistry and Engineering. 7 (2019) 2739–2750. https://doi.org/10.1021/acssuschemeng.8b05832.

[2] A.N.V. Azar, S. Velioglu, S. Keskin, Large-Scale Computational Screening of Metal Organic Framework (MOF) Membranes and MOF-Based Polymer Membranes for H2/N2 Separations, ACS Sustainable Chemistry and Engineering. 7 (2019) 9525–9536. https://doi.org/10.1021/acssuschemeng.9b01020.

[3] H.C. Gulbalkan, Z.P. Haslak, C. Altintas, A. Uzun, S. Keskin, Assessing CH4/N2 separation potential of MOFs, COFs, IL/MOF, MOF/Polymer, and COF/Polymer composites, Chemical Engineering Journal. 428 (2022) 131239. https://doi.org/10.1016/j.cej.2021.131239.

[4] B. Gu, Mathematical Modelling and Simulation of CO2 Removal from Natural Gas Using Hollow Fibre Membrane Modules, Korean Chemical Engineering Research. 60 (2022) 51–61. https://doi.org/10.9713/kcer.2022.60.1.51.

[5] D. Ongari, A. V. Yakutovich, L. Talirz, B. Smit, Building a Consistent and Reproducible Database for Adsorption Evaluation in Covalent-Organic Frameworks, ACS Central Science. 5 (2019) 1663–1675. https://doi.org/10.1021/acscentsci.9b00619.