The VSA Process for Oxygen Enrichment: Process Description and Dynamic Modeling Using Neural Networks

Air separation for oxygen enrichment via the pressure/vacuum swing adsorption (PSA/VSA) process is an important unit operation in many large industrial processes. However, complexities associated with it being a bulk separation process operating in a cyclic, non-equilibrium, unsteady state manner presents significant difficulties in understanding the physics involved. Also, previous attempts in modeling the system numerically have resulted in complications associated with the solution of several, coupled, non-linear, partial differential equations with varying boundary conditions. Empirical methods, such as neural networks, can provide an alternative modeling strategy for highly non-linear processes, such as PSA/VSA. This paper describes the implementation and performance of a dynamic neural network model on a pilot scale, single bed, oxygen vacuum swing adsorption plant. Furthermore, the ability of neural networks to capture the dynamics of this process lends itself to future study in model predictive control of PSA/VSA systems.