Greybox model for multistage spray drying plants constrained to small datasets

Robert Miklos*, Lars Norbert Petersen, Niels Kjølstad Poulsen, Christer Utzen, John Bagterp Jørgensen, Hans Henrik Niemann

*Corresponding author for this work

Research output: Contribution to journalJournal articleResearchpeer-review

Abstract

Dynamic models of spray drying plants are required for many multivariable control strategies for spray dryers, for example, for model predictive control. Often, the model and its parameters are determined by fitting the model to experimental data. When the experimental data is generated, the experiments disturb normal production and introduce production losses. These losses increase the expenses for determining the model. Therefore, the experimentation time must be limited, which then limits the amount of data available for the model fitting. The limited amount of data restricts the complexity of the spray dryer model. In this article, we propose a greybox model with less complexity than the available models in the literature. The greybox model is derived by combining the process structure, physical relations and data from experimental data. The model is validated with experimental data from an MSD20 pilot plant. Further, the greybox model is compared to a model estimated by subspace identification. The comparison shows that the greybox model has a higher fit to the validation data than the subspace method.
Original languageEnglish
JournalAdvanced Control for Applications: Engineering and Industrial Systems
Number of pages22
ISSN2578-0727
DOIs
Publication statusAccepted/In press - 2021

Keywords

  • Greybox
  • Learning‐based model predictive control
  • Model validation
  • Modeling
  • Process control
  • Spray dryers
  • Subspace identification
  • MPC

Fingerprint Dive into the research topics of 'Greybox model for multistage spray drying plants constrained to small datasets'. Together they form a unique fingerprint.

Cite this