Predicting the solubility of gases, vapors, and supercritical fluids in amorphous polymers from electron density using convolutional neural networks

Abstract

A twin convolutional neural network is proposed to predict the pressure and temperature-dependent sorption of gases, vapors, and supercritical fluids in amorphous polymers, using spatial electron density distribution. These distributions are obtained as 3D tensors (images) from DFT calculations. The dataset, compiled from over 250 literature sources, comprises nearly 15 000 experimental measurements of 79 gases’ uptakes (0.01–50 wt%) in 102 different polymers. These measurements, spanning pressures from 1 × 10⁻³ to 7 × 10² bar and temperatures from 233 to 508 K, include nearly 500 solvent–polymer systems, ranging from low-pressure sorption in membrane glassy polymers to high-pressure solubility of supercritical fluids in molten polymers. The irreducible mean absolute percentage error (MAPE) is estimated to be around 20%, with a brief discussion on the sources of data variability. In 150 epochs, the model achieved a 32% MAPE on a test set of 1600 measurements concerning 22 polymers not previously encountered by the model.