DE19842550B4 - Method for on-line parameter estimation and process observation for SMB chromatography processes - Google Patents

Abstract

Method for observing an SMB chromatography process for separating two components with a plurality of columns and two product tapping points by on-line estimation of column parameters, characterized in that continuous concentration measurements are made at the product tapping points, from which each column for each of the components to be separated the apparent dispersion coefficient D _ap and the parameter γ characterizing the adsorption behavior are determined by minimizing an error functional.

Description

In the chemical industry The so-called life science products are considered the most promising Area of the next Years. However, since pharmaceutical preparations and fine chemicals increasingly complex requirements and stricter legal requirements have to do justice have to here efficient processes for gentle product separation, such as B. chromatographic separation processes are used. The continuous Operation of chromatographic separation processes with apparent solid countercurrent as an SMB process (simulated moving bed, Simulated countercurrent) wins to separate a mixture of substances increasingly important in two groups. Because of the high product costs the economical operation of these systems places high demands to litigation and Process monitoring.

The SMB process is accomplished by cascading any number of individual liquid chromatography columns or segments so that there is an endless loop. Arrangements with 8 to 24 columns or segments are common. The feed and solvent supply as well as the removal of the two products are arranged between the columns and define four different zones of the process, each between the inlets and outlets. The apparent counterflow is approximately achieved by periodically switching the inflows and outflows in the direction of the fluid flow, ie after a specified switching period, all inflows and outflows are switched one column at a time. Zone I lies between the desorbent supply and the extract removal, Zone II between the extract removal and the feed supply, Zone III between the feed supply and raffinate discharge and Zone IV between raffinate discharge and desorbent supply. The zones are assigned based on their functionality and their relative position to the inflows and outflows. The zones are not assigned to specific chromatography columns, the function of the individual columns changes with each switching period. A schematic sketch of the process can be found in 1 , Adsorptive separation processes based on this principle are described in (Broughton et al. 1961), (Rasche, 1992), (Negawa and Shojii, 1995), (Priegnitz, 1995). Detailed descriptions of the possible technical versions of this process can be found in the references above. They also serve to document the state of the art. A technical implementation of the process in a single column is described in (Schoenrock, 1992), in (Hotier, 1998) an operating mode with variable switching periods and constant circulating current is described. The SMB process is characterized by complex, mixed continuous / discrete system dynamics that are difficult to control from a control point of view. The main problem here is the start-up process and the stable operation of the systems in the vicinity of the economic optimum. The currently known methods for controlling SMB processes are based predominantly on a pure specification of the relevant liquid flows. In (Masao and Tanimura, 1 986) a method is described to regulate these liquid flows within the columns. The aim of this method is to reverberate the liquid flows at a constant setpoint using pressure measurements. A similar procedure is described in (Cohen et al., 1997). Describes a method with a setpoint that is variable over time but fixed (Kearney and Hieb, 1992). All of these methods are not based on the measurement of the actual system states and product concentrations, but only use the volume flows calculated in advance, so that due to the high sensitivity of the process to disturbances and parameter fluctuations in the vicinity of the optimal operating point, stable operation of the systems in the vicinity of the economic optimum is not always possible.

Recent studies have shown that not only the product concentrations are essential for the control of the SMB process, but also that the location and the temporal development of all adsorption and desorption fronts are of crucial importance for the process control. However, these quantities cannot be measured directly, since appropriate concentration measurements within the columns are not technically possible and measuring points after each individual column often mean an unacceptable financial outlay. To make matters worse, the process dynamics are determined by variable, depending on the operating parameters, dead times and - depending on the position of the concentration plateaus in the columns - the sensitivity of the quality variables with regard to the influencing variables varies greatly. Among other things, this also means that malfunctions often only have visible effects after a very long time (several hours to days). To date, there are only a few approaches to improve the conventional driving style by means of advanced control, including concentration measurements and observation of the internal concentration profiles, in order to operate the processes in the vicinity of the economic optimum. In (Holt, 1995a) a control and measurement concept for an SMB process is described, which in (Holt, 1995b) was once again specially developed for a para-xylene separation. In this method, the liquid flows and switching periods are determined from on-line measurements of the concentrations at two points in the circuit flow with the aid of an algorithm in such a way that a product property, such as. B. the purity, a setpoint. In contrast to the invention described here, this method requires more Measuring points for product monitoring and also does not allow observation of the concentration profiles in the columns.

In (Marteau et al., 1994) a Process monitoring concept aromatics separation using SMB chromatography. The measuring point arrangement proposed there contains four measuring points, where two are arranged in the recycle stream and one each Concentration of the extract and raffinate stream monitored. This method needed four measuring points for surveillance of the process and the product composition. However, for cost reasons it would be advantageous, fewer measuring points to use.

The invention described here is intended the disadvantages of the above approaches avoid. In particular, it should enable with only two measuring points both the product quality monitor online as well as the course of the concentration profile within a column do. From the knowledge of the concentration profile in the columns then the main positions relevant to litigation of the adsorption and desorption fronts. On that basis can e.g. B. further developed a control algorithm with which the required product properties also in the event of malfunctions and Fluctuations in parameters are observed.

The invention shown here forms for one thing the basis, but also represents a method with the immeasurable Concentration profile of the plant can be monitored at any time as well the relevant process parameters can be determined online.

The invention relates to a method for parameter estimation and process monitoring for simulated countercurrent chromatography processes. The invention allows the continuous determination of all columns existing concentration profiles from the continuous measurement the outflow concentration at two points in the process.

The invention is particularly suitable for process monitoring of simulated countercurrent chromatography processes in the area of Sugar separation (Rearick et al., 1997) e.g. B. Breakdown of fructose-glucose Solutions, Separation of sucrose from Molasse, (Broughton, 1983).

The measuring devices become direct in the lines for installed and deliver the two products extract and raffinate on-line the concentrations of the two components. One for sugar separation suitable measuring equipment is z. B. described in (Altenhöner et al., 1997).

When arranging the measuring points had to considered be that the Properties of the chromatography columns on the geometry and Pack of columns depend, which due to production-related reasons slight fluctuations subject. In extreme cases, each individual column has different ones Characteristics. The measuring points were therefore arranged so that the System parameters column-specific can determine. For cost reasons it is another concern, if possible few measuring points to use. The one chosen here Arrangement of the measuring points has additional the advantage that the Measurements carried out as part of the profile and parameter determination at the same time for product monitoring can be used.

The invention is based on the fact that for a large number of separations the physical behavior of the adsorption column can be described with sufficient accuracy by the following equation:

Here, c _i (x, t) is the concentration of component i at location x at time t and u is the flow velocity in the columns. A description of this model, its physical basis and its solution can be found in (Dünnebier et al., 1998). γ _i and D _{ap, i} (i = A, B) are four parameters that are characteristic of the system and must be determined on-line from the measured values.

Parameter- and profile determination

Provided that substances A and B do not interact with each other during adsorption, the parameter pairs γ _A and D _{ap, A} and γ _B and D _{ap, B} can be determined separately.

The parameter determination described below is column-specific. That pillar whose parameters are determined is referred to as the "relevant pillar".

über einen noch zu spezifizierenden Zeithorizont (t₁, t₂).The outflow concentration c _{i, mess} (x = L, t) measured in extract or raffinate flow with unknown parameters γ _i and D _{ap, i} (i = A, B) is compared with a numerically determined outflow concentration c _i (x = L, t; γ _i , D _{ap, i} ) balanced. The parameters γ _i and D _{ap, i are} known in the numerical simulation. The comparison is carried out via these two parameters by optimizing the quadratic quality function

over a time horizon to be specified (t ₁ , t ₂ ).

c _i (x = L, t; γ _i , D _{ap, i} ) is the solution of the partial differential equation (1) with boundary and initial conditions c i (x = 0, t; γ i , D ap, i ) = C egg (t) (3) c i (x, t = 0; γ i , D ap, i ) = C 0, i (X). (4)

With the (2) optimal parameters γ _i = γ _{i, opt} and D _{ap, i} = D _{ap, i, opt} , c _i (x, t = 0; γ _{i, opt} , D _{ap, i, opt} ) the concentration profile at the beginning of the bar inside the relevant column again.

The boundary and initial conditions (3) and (4) are not known and are approximated. For investigation The boundary and initial conditions become the measured concentrations in extract and raffinate deduction in two successive cycles needed. The parameters can be determined after the second bar. This is why this bar referred to as the "current clock". to A distinction is made between the first cycle from which measurements are used and the "previous cycle".

L denotes the column length and te the cycle time. The "direction" is the flow direction desorbent chosen.

Determination of γ _A and D _{ap, A}

The operating parameters of the system must be selected so that (after a start-up time) the desorption front of substance A, which denotes the weakly adsorbable component, passes the raffinate removal point EX in every cycle. After the start-up phase, in in the area of the extract deduction, the in 2 shown concentration distribution.

To determine the parameters γ _A and D _{ap, A} and the concentration profile, the outlet concentrations in the extract deduction from the previous and the current cycle, C _{A, mess, alt} (t) and c _{A, mess} (t), are used.

approximation the initial condition

In the previous bar, the relevant pillar is the first pillar of Section II (cf. 2a ). The inlet concentration in this column corresponds to the concentration in the extract hood and is therefore measurable.

In the second zone, a plateau of approximately constant concentration forms in the concentration profile of substance A (Dünnebier et al., 1998). For this reason, the initial _{profile is} chosen as approximately constant equal to C _{A, mess, alt} (t = 0).

To approximate the concentration profile in the relevant column at the end of the previous bar, c _{A, old} (x, t = te; γ _A , D _{aP, A} ), the partial DGL (1) with boundary and initial values is used
c _{e, A, alt} (t) c _{A, mess, alt} (t)
c _{0, A, old} (t) c _{A, mess, old} (0)
solved.

After relocking, the relevant pillar is the last pillar of Section I (cf. 2 B ). The initial value in the current measure thus corresponds c 0, A (x) = c A old (x, t = te; γ A , D ap, A (5)

approximation the boundary condition

linear approximert. γ_A ist dabei der aktuelle Wert der Optimierung. (6) entspricht einer Modellierung des Konzentrationsprofils durch lineare Approximation und anschließender Approximation des Ausflusses durch ein Modell ohne Diffusion.From rough knowledge of the approximate profile it is clear that the concentration of substance A in zone I is very low at the end of each cycle. Noteworthy concentrations can only be found in the end area of the last column. After relocking, the formerly last pillar of Section I is the penultimate pillar of Section I (cf. 2 B and c ). For this reason, the non-measurable inlet concentration in the relevant column is assumed to be essentially zero in the current cycle. The initial range is determined by the empirically determined formula

linear approximated. γ _A is the current value of the optimization. (6) corresponds to modeling the concentration profile using linear approximation and then approximating the outflow using a model without diffusion.

The solution of (1) with boundary and initial values (6), (5) approximates c _A (x, t; γ _A , D _{ap, A} ) in the relevant column.

Establish of the time horizon

The time horizon of the quality functional (2) is chosen so that the Desorption front of substance A the extract deduction within this Period happened.

Determination of γ _B and D _{ap, B}

The operating parameters of the plant must be selected so that (after a start-up time) the adsorption front of substance B passes the raffinate withdrawal point RA in every cycle (cf. 3 ).

To determine the parameters γ _B and D _{ap, B} and the concentration profile, the outlet concentrations in the extract deduction from the previous and the current cycle, c _{B, mess, alt} (t) and c _{B, mess} (t), are used.

approximation the initial condition

The initial condition C _{0, B} (x) is approximated using the entry concentration of substance B into the relevant column in the previous cycle, C _{B, mess, alt} (t). In the previous cycle, the relevant column is the first column of section IV. The inlet concentration in this column is measurable, since the raffinate deduction is in front of the column under consideration in the previous cycle (cf. 3a ). c B, mess, old (t) = c B, measured (t) (7)

linear approximert.From rough knowledge of the approximate profile it is clear that the concentration of substance B in zone IV is very low at the beginning of each cycle (cf. 3a and c ). Noteworthy concentrations can only be found in the initial area of the first column. For this reason, the solution of the partial differential equation (1) in the relevant column (previous cycle) is approximated by essentially assuming the zero line as the edge distribution at time t = 0. The initial range is determined by the empirically determined formula

linear approximated.

The solution of (1) with boundary and initial values (7), (8) is c _{B, old} (x, t; γ _B , D _{ap, B} ). c _{B, old} (x, t = te; γ _B , D _{ap, B} ) approximates the concentration curve of substance B in section IV at the end of the previous bar. After relocating, the relevant pillar is the last pillar of section III (cf. 3b ). In the current measure, the initial profile is approximately through c 0, B (x) = c B, old (x, t = te; γ B , D ap, B ) (9) certainly.

approximation the boundary condition

It is known from practical studies and extensive numerical simulations (Dünnebier et al., 1998) that a plateau of approximately constant concentration forms in the concentration profile of substance B in the third zone (cf. 3b and d ). For this reason, the inlet concentration in the relevant pillar, which is the last pillar of section III after relocations, constant as c e, B (x) = c B, old (0, t = te; γ B , D ap, B ) (10) selected.

To approximate c _B (x = L, t; γ _B , D _{ap, B} ) the partial DG (1) is solved with initial and boundary values (9) and (10).

Establish of the time horizon

The time horizon of the quality functional (2) is chosen so that the Adsorption front of substance B the raffinate withdrawal within this Period happened.

determination the concentrations in the other columns

Is independent of the parameter estimate now additionally an estimate of the inner states of all pillars possible. Based on the known initial state of an unloaded system one can, using the process model, the impressed input variables and the system parameters determined on-line, the internal concentrations calculate in the entire system.

Summary

• 1. Die aktuellen, säulenspezifischen Parameter eines zur Automatisierung des Prozesses geeigneten Systemmodells.
• 2. Die aktuellen Konzentrationen der Produktströme.
• 3. Das nicht direkt meßbare örtliche Konzentrationsprofil zu Beginn des Taktes im Inneren der Säule, die sich in Fließrichtung des Desorbens vor der jeweiligen Meßstelle befindet (Desorptionsfront A und Adsorptionsfront B).
• 4. Aus der Kenntnis der säulenspezifischen Parameter und der Stellgrößenhistorie kann das vollständige Konzentrationsprofil im Inneren aller Säulen berechnet werden.

As a result, the method delivers on-line using only two measuring points:

• 1. The current, column-specific parameters of a system model suitable for automating the process.
• 2. The current concentrations of the product flows.
• 3. The local concentration profile, which cannot be measured directly, at the beginning of the cycle inside the column, which is located in the flow direction of the desorbent in front of the respective measuring point (desorption front A and adsorption front B).
• 4. From the knowledge of the column-specific parameters and the manipulated variable history, the complete concentration profile inside all columns can be calculated.

bibliography

U. Altenhöner, M. Meurer, J. Strube, and H. Schmidt-Traub. Parameter estimation for the simulation of liquid chromatography. Journal of Chromatography A, 769: 59-69, 1997.

D. B. Broughton et al. US Pat. 2985589, 1,961th

D. B. Broughton. Sucrose extraction from aqueous solutions featuring simulated moving bed. US Pat. 4404037, 1,983th

C. Cohen, R. Jacob, G. Bureau du Columbier, and G. Hotier. Process for regulating at east one fluid flow circulating in a simulated moving bed chromatography separation 100p. U.S. Pat. 5,685,992, 1997.

G. Dünnebier, S. Engell, K.-U. Klatt, H. Schmidt-Traub, and J. Strube and I. Weirich. Modeling of simulated moving bed chromatography processes with regard to process control design. Comp. chem. Engng., 22: S855-S858, 1998.

R. E. Holt. Control process for simulated moving adsorbent bed separations. U.S. Pat. 5457260, 1995.

R. E. Holt. Control process for simulated moving bed para-xylene separation. U.S. Pat. 5,470,482, 1995.

G. Hotier. Process for simulated moving bed separation with a constant recycle rate. U.S. Pat. 5762806, 1998.

M. M. Kearney and K. L. Hieb. Time variable simulated moving bed process. U.S. Pat. 5102553, 1992.

P. Marteau, G. Hotier, N. Zanierszydlowski, A. Aoufi, and F. Cansell. Advanced control of C-8 aromatics separation with real time multipoint online raman-spectroscopy. Process control and Quality, 6: 133-140, 1994th

A. Masao and M. Tanimura. method for controlling a simulated moving bed system. US Pat. 4599115, 1986.

M. Negawa and F. Shojii. simulated moving bed separation system. US Pat. 5456825, 1995. J. W. Priegnitz. Small scale simulated moving bed separation system. US Pat. 5470464, 1995th

J. F. Rasche. Simulated moving bed chromatography separation. U.S. Pat. 5122275, 1992. D.E. Rearick, M. Kearney, and D. D. Costesso. Simulated moving bed technology in the sweetener industry. Chemtech, pages 36-40, 1997.

KWR Schoenrock. Simulated moving bed apparatus using a single sorbent bed for separating com ponents from a fluid stream. US Pat. 5156736, 1992.

Claims (4)

Method for observing an SMB chromatography process for separating two components with a plurality of columns and two product tapping points by on-line estimation of column parameters, characterized in that continuous concentration measurements are made at the product tapping points, from which each column for each of the components to be separated the apparent dispersion coefficient D _ap and the parameter γ characterizing the adsorption behavior are determined by minimizing an error functional. A method according to claim 1, characterized in that the determined parameters D _ap and γ are used in a simulation model based on the dispersion model with linear isotherm (DLI model). Method according to 1, characterized in that the determined parameters D _ap and γ are used for the early detection of changes in individual column packs. A method according to claim 2, characterized in that the simulation model to determine the position of the fronts of concentration profiles is used, so that a process control is feasible by manipulating flow rates and cycle times.