Optimization of pressure swing adsorption via a trust-region filter algorithm and equilibrium theory. (August 2021)

• Record Type: Journal Article
• Authors: Uebbing, Jennifer; Biegler, Lorenz T.; Rihko-Struckmann, Liisa; Sager, Sebastian; Sundmacher, Kai
• Is Part Of: Computers & chemical engineering. Volume 151(2021)
• Publication Date: 2021-08
• Subjects: Pressure Swing Adsorption -- Nonlinear Programming -- Trust-Region Filter Method -- Equilibrium Theory -- Reduced Modeling; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2021.107340
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: Significant reduction of the computational costs for the trust-region step of the applied trustregion filter algorithm. Reduced modeling via equilibrium theory applicable to various PSA cycle operations. Optimization of a 4-column 9-step PSA process for gas separation including multiple internal recycle streams. Comparison of different simplifications of the applied trust-region filter method. Abstract: Optimization of pressure swing adsorption (PSA) remains a challenging task, as these are periodic dynamic systems governed by nonlinear PDE systems. This study develops an optimization strategy that incorporates reduced PSA models and also samples information from a high-fidelity 'truth' model. The reduced model is based on equilibrium theory and can be applied to optimize a large variety of cyclic adsorption processes. The optimization is performed via a trust-region filter (TRF) method. The TRF method uses reduced models to minimize the information required from the high-fidelity adsorption model during mathematical optimization. Moreover, this method guarantees convergence to the optimum of the high-fidelity model. This approach is applied to optimize a 4 column pressure swing adsorption process, modeled by partial differential algebraic equations, for the separation of a CO 2 /CH 4 mixture. The results show that the reduced model significantly reduces the computational time of the method's trust-region step compared to previous studies.