Artificial neural networks to correlate structure and CO₂ separation performance of mixed matrix membranes containing 2D fillers

This study presents an innovative method to accurately predict CO₂ permeability and the selectivity of CO₂/N₂, CO₂/CH₄, and CO₂/H₂ in mixed matrix membranes (MMMs) containing polymers and two-dimensional (2D) nanoparticles. A number of neural network models were used to examine the connection between six input variables (feed pressure, polymer type, filler content, 2D filler, additive type, and modification process) and two output variables (permeability and selectivity). The proposed method was tested on different neural network architectures using measurements like Mean Absolute Error (MAE) and Correlation Coefficient (R²). The neural network models were constructed with one, two, and three hidden layers, each containing a variation of neurons. These findings indicate the existence of a workable model that effectively mitigates bothunderfitting and overfitting occurrences. Another test on the suggested neural network model showed that the type of polymers, the amount of fillers, and the feed pressure had the most significant impact on gas permeability and selectivity. The proposed approach holds significant promise for predicting gas transport properties while minimizing the need for substantial time and financial resources.