Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Optimizing model predictive control of multi-column chromatographic processes


C. Grossmann

ETH Zurich, no. DISS. ETH NO. 18499, pp. 253

The manufacturing and downstream processing of single enantiomer drugs and therapeutic biopharmaceuticals requires advanced separation and purification techniques in order to meet the stringent requirements imposed by the regulatory agencies such as the Food and Drug Administration (FDA) or the European Medicines Agency (EMEA). Multi-column chromatographic processes, like simulated moving bed (SMB) and multi-column solvent gradient purification (MCSGP) processes, have attracted interest in the fields of fine chemicals, pharmaceuticals and biotechnology for this task. Multi-column chromatographic processes can be rapidly and reliably scaled up from drug development to industrial production, a rather important feature in industries where time-to-market is crucial. However, optimal operation of multi-column chromatographic processes is still an open and challenging issue. This is because of the uncertainty involved in determining the physical data that is used in the process models. The models describing the dynamics of such processes are rather complex, due to their cyclic and hybrid nature, with inlet/outlet port switching, strong nonlinearities and delays. The full economic potential of the multi-column chromatographic processes can be realized by using a feedback control scheme that will guarantee the fulfilment of product constraints while respecting process limitations in the face of uncertainty. The current work deals with different aspects of the optimization and control of multi-column chromatographic processes.
The main contributions made by this thesis can be put into three categories:

(1) Systematic methods to build reduced order linear models: This thesis proposes a "cycle to cycle" optimizing control scheme based on linear model predictive control (MPC). A fundamental part of this approach is the linear dynamical model that describes the evolution of the process on a cycle to cycle basis. Two different approaches to build the linear models are presented. The first is a modeling framework that describes multi-column chromatographic processes with nonlinear and hybrid first principle models that can be cast in a general formulation. This formulation is then simplified in a number of steps that yield a reduced order linear model that captures the most important features of the process. These reduced order linear models are then used in the MPC formulation effectively.
The second approach is a system identification method developed in the course of this thesis that can be used to build linear dynamical models of low order from input/output data. The system identification method, called NucID, is shown to be superior to common system identification techniques when applied to sets of data with missing entries, which is often the case in the process industry. The NucID method allows one to build reduced order models from data sets with missing entries or identification experiments, where the outputs do not have to be sampled as frequently as the inputs can be changed. This will lower the cost and waste of the products during the identification experiment.

(2) Application of optimizing model predictive control to multi-column chromatographic processes: The cycle to cycle controller was implemented for the MCSGP process and its performance was assessed through simulations on a virtual MCSGP plant. Two different mixtures to be separated were considered: an antibody variant separation case taken from the literature and a peptide mixture separation case as an industrial example. The controller was able to fulfill the purity requirements while increasing the productivity of the process with respect to the current practice reported in literature. This represents the first automatic control scheme developed for the MCSGP process. The performance of the cycle to cycle controller for the SMB process was assessed both through simulation studies and experiments. The simulations evaluate the performance of the controller under a wide range of nonlinear chromatographic conditions described by the generalized Langmuir isotherms. The results showed that even without the knowledge of the underlying nonlinear adsorption behavior, it is possible to operate an SMB at maximum productivity while fulfilling purity constraints by using the cycle to cycle control schemes. This is a major breakthrough since the time consuming task of determining the complete adsorption isotherm of a new mixture to be separated becomes redundant. The simulation studies also showed that the cycle to cycle controller can effectively use the five degrees of freedom of the SMB unit, i.e the four sectional flow rates and the switch time, as manipulated variables in order to further increase the productivity of the process. In the case of chiral SMB separations it can be advantageous to combine different monitoring techniques to get a better control performance. This thesis proposes also a multi-rate control scheme that extends the capability of the cycle to cycle controller to make use of additional measurements that may work at widely varying time scales, e.g. HPLC analysis and UV detectors. Its performance was assessed through simulations on the chiral separation of the guaifenesin enantiomers. The biggest advantage of this approach is demonstrated by the results: it is possible to detect regeneration problems of the liquid and solid phase in the SMB process, which cannot be done with average concentration measurements alone.

(3) Development and implementation of online monitoring techniques for chiral SMB separations and experimental work to validate the control concepts for the SMB process: Experimental implementation of the cycle to cycle controller has been carried out for achiral (mixture of uridine and guanosine) and chiral (a mixture of Guaifenesin enantiomers) separations in an eight-column laboratory scale SMB unit. The experimental runs were designed to challenge the performance of the controller under linear and nonlinear chromatographic conditions, with uncertainties in the system and major disturbances to be rejected. The controller was able to fulfill the purity constraints while maximizing the productivity in the wide variety of experiments and scenarios that were investigated. In the frame of the chiral SMB separations two different monitoring techniques were developed and implemented. The first technique combines a UV detector and a polarimeter that measure the absorbance of UV light and the rotation angle of polarized light, respectively. Even though the controller fulfilled the purity constraints within the experimental accuracy relying on this monitoring technique, the performance of the controller was hindered by the low accuracy of the polarimeter. Therefore, a further analytical technique was developed and implemented. The second monitoring technique was an HPLC online monitoring system to analyze the average concentration of the product streams from cycle to cycle. This monitoring system overcomes the limitations imposed by optical detectors to monitor the concentrations of chiral species. Its accuracy and reliability are superior to the previous monitoring technique and are entirely satisfactory. The HPLC monitoring system represents a valuable contribution to monitor and control SMB separations, especially in the frame of the cycle to cycle optimizing controller. The experimental and simulation results presented in this thesis clearly validate the most valuable asset of the SMB cycle to cycle controller: the controller can deliver the specified purities and maximize the productivity even if the isotherm governing the separation is unknown and despite disturbances in the SMB unit. The time consuming task of determining the complete adsorption isotherm of a new mixture to be separated becomes redundant. The controller and the approach presented in this thesis offer a fast and reliable way to set up chiral SMB separations in a shorter time.


Type of Publication:

(03)Ph.D. Thesis

M. Morari

File Download:

Request a copy of this publication.
(Uses JavaScript)
% Autogenerated BibTeX entry
@PhDThesis { Xxx:2009:IFA_3494,
    author={C. Grossmann},
    title={{Optimizing model predictive control of multi-column
	  chromatographic processes}},
Permanent link