Improving operation in an industrial MDF flash dryer through physics-based NMPC

Nonlinear model predictive control (NMPC) has increased popularity thanks to the availability of black-box models, systematically obtained from plant data via machine-learning procedures. Although this may be a good approach in the medium term, pure data-driven models quickly become outdated with ch...

Full description

Saved in:
Bibliographic Details
Published in:Control engineering practice 2020-01, Vol.94, p.104213, Article 104213
Main Authors: Santos, Pedro, Pitarch, José Luis, Vicente, Alberto, de Prada, César, García, Ángel
Format: Article
Language:English
Subjects:
Grey-box model
Humidity control
MDF dryer
NMPC
Stiction
Wood industry
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Nonlinear model predictive control (NMPC) has increased popularity thanks to the availability of black-box models, systematically obtained from plant data via machine-learning procedures. Although this may be a good approach in the medium term, pure data-driven models quickly become outdated with changes in the process operation, so re-identification (training) routines are periodically required. Moreover, if the NMPC includes any economic objective to drive the process to a more efficient operation, these models often go beyond their validity range, which translates in a degraded control performance (high plant-model mismatch). In this work, the above issues are considered to model the drying section of an industrial medium density fibreboard (MDF) plant. The main contribution of this work is proposing a lumped-parameter grey-box model built upon first principles and completed with experimental equations, obtained from constrained regression with plant data. Upon this model, a moving-horizon estimator (MHE) is designed to estimate unmeasured inputs and states. Then, a mixed economic-tracking NMPC, which explicitly considers actuator-stiction compensation, is proposed to drive the dryer to the humidity setpoint by the most efficient path. Both optimisations are solved within the system sampling interval thanks to an efficient discretisation of the system dynamics by orthogonal collocation, and to the high computational performance achieved by modern software tools for numerical optimal control. The controller was firstly tested in simulation with a rigorous model of a generic MDF dryer, and it is currently implemented in the actual plant achieving significant performance improvements.
ISSN:0967-0661
1873-6939
DOI:10.1016/j.conengprac.2019.104213