Premium
Piecewise linear model predictive control of a rapid pressure swing adsorption system
Author(s) -
Urich Matthew,
Rama Rao Vemula,
Kothare Mayuresh V.
Publication year - 2020
Publication title -
aiche journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.958
H-Index - 167
eISSN - 1547-5905
pISSN - 0001-1541
DOI - 10.1002/aic.16998
Subject(s) - pressure swing adsorption , model predictive control , control theory (sociology) , controller (irrigation) , nonlinear system , computer science , system identification , process (computing) , control engineering , engineering , chemistry , adsorption , control (management) , artificial intelligence , physics , organic chemistry , quantum mechanics , database , agronomy , biology , measure (data warehouse) , operating system
Rapid pressure swing adsorption (RPSA) is a gas separation technology used in the small‐scale oxygen concentrator devices. These devices are commonly used to produce high purity (~90%) oxygen from air for oxygen rehabilitation therapy, but can also produce a much wider range of oxygen purities for other applications. RPSA is a complex, cyclic, nonlinear switched logic process resulting from the coupling of gas adsorption, heat transfer, flow reversal effects, and process logic switches. For RPSA devices to operate satisfactorily, feedback control is critical but challenging due to their inherent complexity. In this article, we present a piecewise linear model predictive control framework for operation and control of a single‐bed RPSA system. A set of coupled, nonlinear partial differential equations with flow switching conditions is used as a plant model for the RPSA process. Subspace system identification with pseudo‐random binary sequence signals applied to this plant model at multiple operating points is used to generate a family of piecewise linear models for use in the model predictive controller algorithm. Detailed descriptions of the RPSA plant model, the multiple linear model identification procedure, the controller formulation and model switching logic are presented. The closed‐loop system is evaluated in simulation using several realistic set point tracking and disturbance rejection cases.