Two-step evolutionary multi-objective optimisation of pressure swing adsorption processes with monolith columns

Abstract

Monolithic adsorbents offer an opportunity to intensity cyclic adsorption processes, but uniformity of channel size and flow distribution have a detrimental effect on separation performance. Mathematical modelling and optimisation techniques require repeated process simulations up to cyclic steady state but the real monolith model representing the response of a distribution of channels is computationally expensive. This study explores the possibility of employing the ideal single channel monolith model to do an initial search, followed by a secondary search with the computationally more complex and more accurate real monolith model. Two case studies have been considered here to cover the different nature of the product of interest (i.e., heavy or light), and whether the optimisation is constrained or unconstrained. For unconstrained optimisation, the optimum decision variable values found with the ideal monolith model are similar to those obtained when only the real monolith model is used for all the functional evaluations (i.e., the real optimum). However, the corresponding objective function values were not the same due to the ideal and real monolith model predictions differing for certain combinations of decision variables. In this case, a quick secondary refinement search with the real monolith model yielded the real optimum objectives. In the case of constrained optimisation, the optimum objective and decision variable values predicted from the initial search differed substantially from the real optimum. Optimum values close to the real optimum could still be obtained with the two-step search strategy. The two-step search strategy required approximately half the computational effort, compared to the approach where only the real monolith model was used for all the evaluations.