Use and improvement of machine learning in adsorption research and processes

Abstract

This work investigates multiple methods of machine learning for an application of surrogate modelling and optimization of an experimentally validated Pressure Swing Adsorption model, evaluation their accuracy and precision compared to the more widely use method of Artificial Neural Networks. In addition, some means of improving the machine learning models accuracy with at-hand process knowledge and parameters were explored, which was followed by the optimization of the Purity and Recovery parameters of the system, finishing with a quantification of the total computational time employed. All steps described were developed and finished successfully using the open source Python programming language and the expected and unexpected results were discussed and the optimization was finalized and expanded.