Koopman-based data-driven control for continuous fluidized bed spray granulation with screen-mill-cycle. (July 2021)

• Record Type: Journal Article
• Authors: Maksakov, Anton; Palis, Stefan
• Is Part Of: Journal of process control. Volume 103(2021)
• Page Start: 48
• Page End: 54
• Publication Date: 2021-07
• Subjects: Granulation -- Fluidized bed -- Neural networks -- Koopman embedding; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2021.05.006
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: Continuous fluidized bed spray granulation processes with external screen-mill cycle are known to be susceptible to nonlinear oscillations. As has been shown recently, mass control may induce periodical oscillation of the particle size distribution in fine grinding regimes, which requires the design of additional stabilizing control. In this contribution a data-driven control approach is investigated. It is based on the identification of the Koopman-based linear representation of the system and does not require any manual system modeling as it is based purely on input–output measurements. The resulting representation makes possible a linear–quadratic control design, that does not depend on a specific operational point. The proposed approach was successfully validated in a numerical study. Highlights: Koopman-based linearization of continuous fluidized bed spray granulation with screen-mill-cycle. Discovery of linear Koopman subspace using deep neural networks. Stabilization control design for nonlinear oscillations of particle size distribution.