Ralf Gesthuisen

Model based operation of emulsion polymerization reactors with evaporative cooling: Application to vinyl acetate homopolymerization ☆

Abstract This work aims at maximizing the productivity of emulsion homopolymerization processes. A dynamic model of emulsion polymerization processes is extended by the inclusion of vaporization from the liquid phases in the reactor to the gaseous phase. The multi-component gas–liquid mass transfer phenomenon is described by the Maxwell–Stefan diffusion equations, which are solved by a special algorithm. A novel operation strategy is developed for running a reactor optimally with respect to batch time. This strategy is applied first to an industrial scale reactor, which is run without using evaporative cooling. Then, based on the extended model, controlled vaporization is included by which additional heat is removed from the reaction system. This makes it possible to extend the restrictions imposed by the limited heat removal of the cooling jacket considerably. Simulation results are presented for the homopolymerization of vinyl acetate in an industrial scale reactor operated in semi-batch mode. The results show that a significant amount of heat can be removed by evaporative cooling thus leading to higher productivity.

Fixed structure multirate state estimation

In most chemical processes only some measurements are available online while other measurements are available infrequently and often with long delays. Multirate state estimation can optimally combine these different classes of measurements to improve the estimation quality compared to the fast measurements alone. The following new aspects of multirate state estimation are discussed in this paper: Firstly, we discuss the behaviour of variable and fixed structure estimation schemes and conclude that a fixed structure is preferable. Secondly, the moving horizon state estimator is extended to the multirate case in a fashion that corresponds to the fixed structure state estimation scheme. Finally, a performance gauge is proposed which measures the maximal improvements that can be obtained by using additional noisy slow measurements. The theoretical findings are supported by a simulation example.

MULTIRATE STATE ESTIMATION USING MOVING HORIZON ESTIMATION

Abstract In most chemical processes only some measurements are available online while other measurements are available infrequently and often with long delays. Multirate state estimation can optimally combine these different classes of measurements to improve the estimation quality compared to the fast measurements alone. The nature of measurements at different sampling intervals which are subject to delays makes the application of traditional one step state estimators cumbersome. There is but one state estimation scheme which naturally suggests the inclusion of these different classes of measurements, the Moving Horizon Estimator (MHE). In this paper, we extend the MHE concept to the multirate case (MMHE). We present two forms, a variable structure and a fixed structure MMHE and present the relevant equations. We recommend the fixed structure estimator as it has superior noise reductions qualities. The proposed scheme is supported by a simulation example.

State estimation for semicontinuous emulsion copolymerization using calorimetric data

Semicontinuous emulsion copolymerization is one of the major processes in the production of polymer lattices. For optimization or control purposes it is necessary to measure important product qualities such as the composition of the copolymer or the average chain length of the macromolecules. Measurements of such properties for a multiphase process are expensive and usually not available online. State estimation is besides reaction calorimetry a possibility to overcome this problem. Starting from a complex model of the considered process we develop a simplified model which enables us to estimate the concentrations of the monomers in the particle phase and thus the composition of the copolymer using a constrained extended Kalman filter (CEKF). We show that this approach is robust to modeling errors. The CEKF is also compared to an EKF. From the results it follows that the CEKF has a better robustness to errors of the initial values.

CONTROL STRUCTURE SELECTION FOR A REACTIVE DISTILLATION COLUMN

Abstract In recent years integrated processes have attracted considerable attention in both academic research and industrial applications. The integration, for instance of reaction and separation, reduces the degrees of freedom and requires tight control to operate the processes at optimal conditions. From the available number of measurements and input variables, a control structure has to be identified which guarantees the best performance. In this paper, we discuss the application of a systematic control structure selection consisting of pre-screening of structures and calculation of the attainable performance for the heterogenous catalyzed esterification of methyl acetate in a batch reactive distillation column. The approach is applied to a linearized model of 85 th order of the nonlinear system. To verify the predictions of the control structure selection method, linear controllers are designed for the favorable structures by using a frequency response approximation method. The resulting linear controller is applied to the nonlinear model and shows good performance.

Simultaneous estimation of heat transfer coefficient and reaction heat in semi-batch processes

In this paper, we propose an extended Kalman filter (EKF) to estimate the heat of reaction and heat transfer coefficient. The EKF is chosen as it is well known in the process industries and the most frequently used state estimator. It can be used to estimate the heat of reaction and heat transfer coefficient.

Estimating the Prediction Uncertainty of Dynamic Neural Network Process Models

Abstract In this paper, the explicit calculation of prediction intervals for multistep-ahead predictions from dynamic neural network models is described. Usually, asymptotic methods that are based on linearized descriptions of neural networks are applied with the potential problem of large coverage errors and too optimistic prediction intervals. The potential Sources of these problems are the negligence of the network parameter uncertainties and the non-normality of error distributions. To overcome these restrictions, the bootstrap method is used here. The bootstrap is a tool from computational statistics. New formulations are introduced to apply the bootstrap to nonlinear time series models with exogenous inputs. The proposed method is illustrated by an analysis of a neural network model of a bioreactor benchmark problem.

State Estimation for Tubular Reactors with Measurements at the Outlet Only

Abstract In this paper we address the design of finite dimensional observers for systems with distributed parameters and measurements only at the outlet. Necessary properties of the models for observability are discussed. If the partial differential equation systems (PDE-systems) fulfill these restrictions they are lumped by polynomial approximation (orthogonal collocation). The observer design is based on the resulting lumped parameter model. Using graph theory it is shown that the resulting model is observable. The design method is applied to the production of polyamid-6 in an isothermal tubular reactor with recycle stream (Gupta, 1987) with viscosity measurement at the reactor outlet.

Time Optimal Control of the Molecular Weight in a Semi Batch Emulsion Polymerization

Abstract In this contribution a hierarchical approach to the control of the molecular weight in a semi-batch emulsion polymerization is derived, which ensures minimum batch time. Rather than performing a numerical optimization to determine an optimal trajectory for the process at any point in time, the existing boundary conditions are examined and used to determine the time optimal operation at each time step. This procedure provides the set points for the molar holdups of monomer and chain transfer agent (CTA) in the system. Either a decentralized PI controller or a nonlinear model predictive controller (NMPC) can be applied to track the calculated trajectory. As the set point calculation is a solution of a system of algebraic equations it can easily be performed on-line. It is shown that this concept can be used to produce multi modal molecular weight distributions and can also handle disturbances in the process, such as an unexpected nucleation of particles.

Zustandsschätzung für Rohrreaktoren mit Messungen am Reaktoraustritt (State Estimation for Tubular Reactors with Measurements at the Outlet only)

In diesem Beitrag wird der Entwurf von Beobachtern mit konzentrierten Parametern für Systeme mit verteilten Parametern mit Messungen nur am Austritt des Systems diskutiert. Die notwendigen Voraussetzungen für den Entwurf eines Beobachters für das durch polynomiale Approximation und orthogonale Kollokation diskretisierte System werden hergeleitet. Die Beobachtbarkeit wird mit der Graphentheorie nachgewiesen. Als Beispiel für die Anwendbarkeit des Verfahrens dient der Prozess der kontinuierlichen Produktion von Polyamid-6 in einem isothermen Rohrreaktor mit Recyclestrom, wobei am Reaktoraustritt die Viskosität, sowie zwei Konzentrationen in der Polymerschmelze gemessen werden.