Unifying Mixed Gas Adsorption in Molecular Sieve Membranes and MOFs using Machine Learning

TL;DR

This work tackles the challenge of predicting gas adsorption isotherms across two distinct classes of porous materials—carbon molecular sieve membranes (CMSMs) and metal-organic frameworks (MOFs)—including binary mixtures. It introduces a data-fusion approach that trains both neural networks and gradient-boosted trees on a combined dataset built from CMSM Grand Canonical Monte Carlo simulations and the CoRE MOF database, using a compact fingerprint of physical descriptors to predict pure and mixed gas uptakes. The study shows that joint training across material classes is essential for transferability, with neural networks (and to a comparable extent XGBoost) accurately capturing adsorption trends and smoothing isotherms, even for unseen materials like CALF-20 at 313K. The results offer a data-efficient pathway to screen gas-separation materials, potentially reducing the need for exhaustive simulations across the vast material landscape while enabling reliable predictions for mixtures. All mathematical notation is presented with proper delimiters, e.g., $q_A$, $q_B$, $d_A$, $d_B$, $T$, $P$, $GSA$, $\xi$, $\rho^{-1}$, $D_i$, and $d_{max}$.

Abstract

Recent machine learning models to accurately obtain gas adsorption isotherms focus on polymers or metal-organic frameworks (MOFs) separately. The difficulty in creating a unified model that can predict the adsorption trends in both types of adsorbents is challenging, owing to the diversity in their chemical structures. Moreover, models trained only on single gas adsorption data are incapable of predicting adsorption isotherms for binary gas mixtures. In this work, we address these problems using feature vectors comprising only the physical properties of the gas mixtures and adsorbents. Our model is trained on adsorption isotherms of both single and binary mixed gases inside carbon molecular sieving membrane (CMSM), together with data available from CoRE MOF database. The trained models are capable of accurately predicting the adsorption trends in both classes of materials, for both pure and binary components. ML architecture designed for one class of material, is not suitable for predicting the other class, even after proper training, signifying that the model must be trained jointly for proper predictions and transferability. The model is used to predict with good accuracy the CO2 uptake inside CALF-20 framework. This work opens up a new avenue for predicting complex adsorption processes for gas mixtures in a wide range of materials.

Figures (7)

• Figure 1: Monomers (a) pyrrole and (b) pyridine used to build different $6$F-CMSM polymers. Equilibrated morphologies of three representative membranes. The inverse density $(1/ \rho)$ of these membranes are (c) highest $(0.86\cubic cm\per g)$, (d) average $(0.79\cubic cm\per g)$, and (e) lowest $(0.69\cubic cm\per g)$. The simulation box lengths are in $\AA$ units. The bottom color legends show the constituent elements in the polymer membrane.
• Figure 3: Correlation coefficients of input and output features.
• Figure 4: Distribution of physical properties of the combined $6$F-CMSM and MOF dataset.
• Figure 5: Training and validation loss vs. epochs for a neural network having $(n, h) = (160, 7)$ using (a) MSE and (b) MAE metrics respectively.
• Figure 7: Comparison of gas loading (q) predictions using (a) NN with $(n, h) = (2^{14}, 14)$ trained on the combined dataset and (b) NN with $(n, h) = (2^{10}, 10)$ trained only using $6\text{F-CMSM}$ data.