Predictive Control of Pressure Swing Adsorption

Pressure swing adsorption requires a repeated cycle of four steps. The periods of these steps, or other defined terminal conditions, determine the rate and quality of the product, and its cost. In transient situations such as upsets or grade changes, it is not intuitively obvious how the steps should be progressively altered to bring the plant to the desired operating point in an optimal fashion. The present work considers the problem of real-time maximisation of the production of a single adsorber, and maintaining a setpoint concentration in its product receiving vessel. In a modelling exercise, these objectives have been met using predictive control based on completion of the present step, plus two full future cycles to reduce the end-effect. The approach sought to be fast and robust by suitable linearisation of the system. This allowed MILP solution in the mixed logical dynamical (MLD) framework as a mixed integer dynamic optimisation (MIDO). However, this problem was ultimately solved faster and more reliably by testing all combinations for constraint violations and the objective value.