Combining High-Throughput Molecular Simulations and Machine Learning to Optimize Adsorption Processes

High-throughput molecular simulations utilizing massively parallel high-performance computers allow now for the generation of an unprecedented amount of adsorption data. However, given the high dimensionality of state space (temperature, pressure, and composition) and/or the large number of porous materials governing the adsorption behavior, molecular simulation alone is not effective in finding optimal materials and conditions for adsorption processes. This talk will highlight the development of surrogate machine learning models that are trained on molecular simulation data and enable finding optimal material/condition combinations for gas storage or chemical separation applications: (a) desorptive drying for hydrogen-bonding solute/solvent mixtures in all-silica zeolites, (b) adsorption of BTEX mixtures onto zeolite nanosheets at membrane reactor conditions, and (c) gas storage in diverse porous materials.