Jump label

Service navigation

Main navigation

You are here:

Main content


Model-based optimizing control of reactive distillation


Reactive distillation (RD) combines chemical reaction and separation by distillation in one piece of equipment. It is the most often used integrated process because by applying reactive distillation, the number of pieces of equipment and the energy demand can be reduced significantly, in particular for equilibrium-constrained reactions. The dyn group has cooperated with the fluid separations group of Prof. A. Gorak (FVT) for many years in the dynamic mathematical modelling and optimal operation and control of reactive distillation processes. In the context of the MOBOCON project, a very challenging reactive distillation process was chosen for the demonstration of online optimizing control at a pilot plant. 


The process, which has been developed by the FVT group, can be dynamically switched between two different target products. The reaction mechanism consists of two steps:


Both products are of interest industrially, but their individual production is difficult. By a proper choice of the operating conditions in a reactive distillation column, either product can be produced with a high selectivity in a single apparatus.

Dynamic mathematical modelling
From several mathematical models of different complexity, an equilibrium-stage process model was chosen and implemented in MATLAB® and Python/CasADi for model-based control purposes. The resulting DAE model has 955 variables, among them 305 dynamic states.

Multi-rate state estimation
For model-based control, the internal states of the optimization model must be initialized using plant measurements. Possible estimators are particle filters or the DAE version of the EKF. A new formulation of the particle filter for DAE systems has been developed in which different sampling intervals of different measurements can be considered.

diagram_startupModel-based optimizing control
For nonlinear model-based control, the dyn group has developed a software-platform (called DO-MPC) that employs state-of-the-art methods for dynamic nonlinear optimization, such as automatic differentiation and multiple shooting, and supports robust multi-stage NMPC formulations.

The goal in MOBOCON is to apply direct optimizing control with an economic cost function to the real reactive distillation process in a 50 mm diameter pilot plant. Key challenges are the size and the nonlinearity of the process model and the inevitable plant-model mismatch as the complex physico-chemical phenomena are only approximately represented even by highorder models. The online optimizing controller should also handle the startup of the plant and the transition between the two modes of operation that lead to the two different target products.


Völker, M. C., Sonntag, C., Engell, S.: Control of integrated processes: A case study on reactive distillation in a medium-scale pilot plant. Cont. Eng. Pract., 2007, 15(7), 863–881.

Idris, E.N.A.; Engell, S.: Economics-based NMPC strategies for the operation and control of a continuous catalytic distillation process. J. of Proc. Control, 2012, 22(10), 1832–1843.

Lucia, S., Tatulea-Codrean, A., Schoppmeyer, C., Engell, S.: An Environment for the Efficient Testing and Implementation of Robust NMPC. In: Proc. IEEE Multi-conference on Systems and Control, Antibes, 2014.

Model parameter estimation of the SMB process

Simulated Moving Bed (SMB) chromatography is a preparative chromatographic process which establishes a countercurrent movement of the solid bed by switching the feed and the product ports periodically in a ring of chromatographic columns. Compared to batch chromatography, the SMB process is more efficient in terms of solvent consumption and utilization of the adsorbent, but it is more difficult to design and to operate. Model-based online optimization techniques can improve the economic performance of the process. Due to the high sensitivity of the purities of the products to the adsorption behavior of the system, it is crucial to employ accurate models in the optimization.
The accuracy of the model parameters can be improved by optimal dynamic experiment design which is investigated in our research. By means of this technique, it has been possible to design start-up plant experiments which lead to a drastic reduction of the expected confidence intervals of the parameters of the isotherms of the individual columns, while respecting purity constraints for one of the product ports during the time horizon of the experiment. A critical aspect is the evaluation of the objective function of the experimental design problem, for which the adjoint-based derivative evaluation framework implemented in the optimal experiment design package VPLAN (IWR Heidelberg) is used together with an efficient implementation of the process model of the chromatographic plant.
We have also developed a technique for the estimation of individual column parameters based upon concentration measurements of the product ports when they are connected to the column under consideration.
In 2015, the dyn group will install a new SMB chromatography system which is capable of running the novel multi-column solvent gradient process (MCSGP), and we will study the online optimization of this mode of operation.


Lemoine-Nava, R., Engell, S.: Individual Column State and Parameter Estimation in the Simulated Moving Bed Process: An Optimization-based Method. In: Proc. of the 19th IFAC World Congress, Cape Town, 2014.

Interaction between operators and advanced control solutions

desktop_pictureA challenge in advanced optimization-based control is the interaction with the operators. Professional plant operators have a good experience-based understanding of the behavior of the plant and can learn the effect of manual changes and disturbances whereas they cannot handle the interactions and constraints as well as the optimization-based solution. However, the operators are responsible for the monitoring of advanced controllers and have the authority to accept or to discard the proposed control moves and to switch the optimizing controller off if its behavior seems wrong to them. 
Acceptance by the operators is therefore crucial to the long-term success of advanced control solutions; they must be viewed as a useful and reliable support in the daily work of the operators. Our goal is to provide information about the results of the optimization in order to convince and motivate the operators to rely on them. This requires that understandable explanations and reliable diagnosis of the results are provided in order to build trust.
Moreover, the abilities of humans to detect and handle critical situations are important and will be needed for a long time to maintain an efficient operation of processing plants. Rather than asking the operators to make a yes/no decision on the acceptance of the proposed control moves, an improved performance could result from a two-way interaction between the model-based optimization and the operators so that their process experience can be used to influence the optimization, e.g. by anticipating changing conditions that are not known to the optimizer in advance.

Optimizing control of intensified processes

Intensified continuous processes have a large potential for the sustainable production of high-quality, high-value and customer-specific products at competitive prices in a sustainable fashion. To realize the potential of this technology, tightly controlled, fully automated and optimized production is a must. Starting with the F3 project, the dyn group has investigated the optimizing control of intensified processes, in particular of tubular reactors.

Optimizing control and state estimation of a continuous polymerization reactor
moboconThe goal is to design and to implement optimizing control for a tubular polymerization reactor with side feeds of monomer and initiator to maximize the product throughput while meeting tight quality constraints. The molecular weight and the residual monomer of the produced polymer are considered as the quality indicators.


The tubular reactor system is described by a set of nonlinear PDEs and exhibits a very complex and dynamic behavior due to the superposition of the effects of the side feeds. Our approach combines state estimation by particle filtering and dynamic model-based optimization. A key challenge is the computation time due to the high order of the discretized process model. The weighted essentially non-oscillatory scheme (WENO) is used to simulate the system without the need for a very fine discretization grid.


Bouaswaig, A.E., Engell, S.: WENO scheme with static grid adaptation for tracking steep moving fronts. J. of Chem. Eng. Sci., 2009, 64(14), 3214–3226.

Hashemi, R., Engell, S.: Optimizing control and state estimation in a tubular polymerization reactor. In: Proc. of the 19th IFAC World Congress, Cape Town, 2014.

Integrated control and sensing for sustainable operation of flexible intensified processes

consensThe goal of the HORIZON 2020 SPIRE project CONSENS is to advance the continuous production of high-value products in flexible intensified continuous plants by introducing novel online sensing equipment and closed-loop control of the key product parameters. CONSENS focuses on flexible continuous plants but the results will be transferable also to large-scale continuous processes. The research and development is driven by industrial case studies from three different areas, spanning the value chain of chemical production: complex organic synthesis, specialty polymers, and formulation of complex liquids. The project results will be validated in pilot plants, including production containers that have been developed in the F3 Factory project.
In the dyn group, PAT-based control concepts for two CONSENS processes will be developed, an organic synthesis in a tubular reactor, and the production of high-viscosity polymers in two pieces of equipment, a tubular and a kneader reactor.


The challenge for both processes is to enhance the process performance applying the novel sensor data. In order to cope with the mismatch between plant models and reality, online model parameter estimation and iterative optimization using gradient modifiers will be applied. The model will be used for optimizing control. In addition, the dyn group will develop strategies for plant-wide control for modular continuous plants, integrating local control strategies into a plant-wide process control strategy.


Finkler, T., Kawohl, M., Piechottka, U., Engell, S.: Realization of online optimizing control in an industrial semi-batch polymerization. J. of Proc. Control, 2014, 24(2), 399–414.