1. Optimisation and control of chromatography
Sebastian Engell
Abdelaziz Toumi Laboratory of Process Control Biochemical and Chemical Engineering Department
Universität Dortmund

2. Contents Introduction Batch chromatography SMB chromatography
Preparative chromatography
Simulated Moving Bed technology
Reactive chromatography
Batch chromatography
Motivation, problem formulation, modelling
Parameter estimation
Feedback control
SMB chromatography
Optimisation of the operation regime
Control strategies
Optimisation-based control of a reactive SMB-process
Conclusions and future challenges

3. Preparative chromatography
= Chromatography for production, not analytical chemistry
Batch Process: F e d ( A + B )
Eluent (E)
(A+E)
(B+E) C ,
flexible, standard process in analytical and development labs
multi-components separation
intensification by gradient elution
expensive in large scale
highly diluted products

4. Simulated Moving Bed technology
Process intensification:
True Moving Bed (TMB)
Practical implementation as a
simulated moving bed process:
Adsorbent is fixed in several chromatographic columns.
Periodic switching of the inlet/outlets => moving bed is simulated.
Complex mixed discrete and continuous dynamics

5. SMB chromatography: process dynamics
Continuous flows and discrete switchings
Axial profile builds up during start-up
Same profile in different columns in cyclic steady state
Periodic output concentrations

6. The VARICOL process Variable length column process (NovaSEP 2000)
Periodic but asynchronous switching of the ports

7. Industrial applications of SMB I
Petro-chemicals
Universal Oil Products (USA), US Patent (Brougthon und Gerhold 1961), 120 units sold (Sarex, Molex , Parex etc..)
Institut Francais du Pétrole (France), largest SMB-Plant in the world implemented in South Korea (Eluxyl) ….
Sugar industry
Amalgamated Sugar Co. (USA) operates SMB-plants with a total capacity of tonn HFCS (2001)
Cultor Corporation (Finland) patented new operating modes which includes ,,Sequential-’’ and ,,Multistage’’ SMB (FAST)
Appelxion has installed more than 90 ,,Improved’’ SMB-Plants, 3 of them in Europe (in Spain for the production of Pinitol)

8. Industrial applications of SMB II
Pharmaceutical substance development
Considerable amount of pure chiral drugs is required for the clinical phases.
Binary separations of enantiomers
Drugs purified using SMB-processes
Prozac (Elli Lilly & Co, USA)
Citalopram (Lundbeck, Denmark)
...
SMB-Plants of large scale
Aerojet Fine Chemicals (Sacramento, USA)
Bayer (Leverkusen, Germany)
Daicel (Japan)
Novasep (Nancy, France)
800 Millimeters SMB-Plant Aerojet Fine Chemicals (Sacramento, USA)

9. Prediction of application areas
Fraction of installed units
International Strategic Directions (Los Angeles, USA)

10. Reactive chromatography
Integration reduces equipment costs.
In-situ adsorption drives the reaction beyond the equilibrium.
Conversion of badly separable components
Loss of degrees of freedom and flexibility
Complex dynamics, narrow range of operation A
B+C
Injection A B A C
Chromatographic bed
+ catalyst
Mazzotti/Morbidelli et al. (ETH-Zürich)
Ray et al. (Singapore National University)
Schmidt-Traub et al. (Universität Dortmund)
DFG-Research Cluster Integrated Reaction and Separation Processes at Universität Dortmund since 1999
fractionation
tanks A B C

11. RSMB for glucose isomerisation (Fricke and Schmidt-Traub)
Cyclic Steady State PurEx=70 %
extract
feed
eluent
6 columns interconnected in a closed loop arrangement
ion exchange resin (Amberlite CR-13Na)
immobilized enzyme Sweetzyme T (Novo Nordisk Bioindustrial)
switching
eluent (water)
extract
feed
Zone II
Zone I
Zone III

12. Contents Introduction Batch chromatography SMB chromatography
Preparative chromatography
Simulated Moving Bed technology
Reactive chromatography
Batch chromatography
Motivation, problem formulation, modelling
Parameter estimation
Feedback control
SMB chromatography
Optimisation of the operation regime
Control strategies
Optimisation-based control of a reactive SMB-process
Conclusions and future challenges

13. Batch chromatography: challenge
Separation of 2-component mixtures in isocratic elution mode
Goals:
Maximize productivity for given column setup!
Meet product specifications at all times!
Adjust for
plant/model mismatch or
changes in separation characteristics!
Extension of this concept to multi-component mixtures

14. Batch chromatography: optimisation
Mathematical formulation of the optimisation problem:
maximise the productivity
purity requirements
recovery requirements
flow rate limitation due to maximum pressure drop
Online optimisation: nested approach (Dünnebier & Klatt)

15. Orthogonal collocation Normalised formulation
General Rate Model
Fluid phase:
Solid phase:
Isotherm:
Orthogonal collocation
Finite elements
Galerkin
Numerical Scheme by Gu
Stiff ODE system
ODE solver
Integration
Solid phase
Parabolic pde system
Normalised formulation
Solution ci(x,t)
Fluid phase
Simulation is 2-5 orders of magnitude faster than real time.
Universal model, can include reaction etc..

16. Batch chromatography: Parameter estimation - results
Enantiomer separation
EMD by MERCK, Darmstadt
R = fast eluting
Initial set of model parameters from offline experiments
Model adaptation by online estimation of
1 mass transfer coefficient
1 adsorption parameter per component
good fit of measured and simulated elution profiles

17. Batch chromatography: Control scheme

18. Batch chromatography: Control results for sugar separation
Task:
Reach steady state after initial disturbance!
Realise set-point change!
Specifications of the experiment:
System: Fructose (A) Glucose (B)
Feed concentration: 30 mg/ml each
Specified purities: 80 % each New Setpoints: 85 % each

19. Dealing with model mismatch
Unfeasible set-point
Constraints are violated.
The process is operated inefficiently.
Model mismatch
Additional feedback control layer to establish the constraints

20. Feedback control Hanisch 2002 Initial condition:
Adjust switching times to keep
the purity constraints
Adjust operating parameters
to minimize the waste part
Initial condition:

21. Online optimisation Disadvantage of the purity control scheme:
Optimality is lost!
Solution:
Measurement-based online optimisation
Redesigned ISOPE algorithm
Combines the measurement information and the model to construct a modified optimisation problem.
Iteratively converging to the real optimum although model mismatch exists.
Can handle constraints with model mismatch.
Gradient-modification
optimisation algorithm
Batch chromatography
Measurements
Set-point
Gao & Engell: Measurement-based online optimisation with model-mismatch, ESCAPE 14.

22. Production rate surfaces:
Simulation study: enantiomer separation
Elution profiles:
“real plant”
Purity specification: 98%
Recovery limit: 80%
Flow rate: ≤ 0.42 cm/s
Production rate surfaces:
“Real plant”
Optimisation model

23. Result of iterative optimisation

24. Contents Introduction Batch chromatography SMB chromatography
Preparative chromatography
Simulated Moving Bed technology
Industrial applications of SMB
Reactive chromatography
Batch chromatography
Motivation, problem formulation, modelling
Parameter estimation
Feedback control
SMB chromatography
Optimisation of the operation regime
Control strategies
Optimisation-based control of a reactive SMB-process
Conclusions and future challenges

25. Reminder: SMB dynamics

26. Choice of the (nominal) operating regime
Triangle theory (Morbidelli and Mazzotti)
Based on the True Moving Bed process model
Wave theory (Ma & Wang 1997)
HELPCHROM (Novasep)
Based on a plate model, propriatory software
Approaches based on rigorous modelling
Heuristics, simulation-based-methods (Schmidt-Traub et al., Biressi et al.)
Genetic algorithms (Zhang et al. 2003)
Iterative approach (Lim and Joergensen, 2004)
SQP-based approach (Klatt and Dünnebier, Toumi)

27. Mathematical modeling: Full model
Hybrid Dynamics
Node Model (change in flow rates and concentration inputs)
Synchronuous switching (new initialization of the state)
Continuous chromatographic model (General Rate Model)
Numerical approach (Gu, 1995, Toumi)
Finite Element Discretization of the fluid phase
Orthogonal Collocation for the solid phase
stiff ordinary differential equations solved by lsodi (Hindmarsh et al.)
Efficient and accurate process model (672 state variables for nelemb=10, nc=1,Ncol=8)

28. Model-based Optimisation I
Sequential approach
simulation until cyclic steady state is reached
Simultaneous/multiple shooting
cyclic steady state is included as an additional constraint
MUSCOD-II (Bock et. al.) DFG project (EN 152/34-1)
Process dynamic
cyclic steady state
Purities
Pressure drop
SMBOpt (Toumi et. al.)

29. SMB vs. VARICOL (single shooting)
Verzögerer
VARICOL is more efficient than SMB
VARICOL result gives clue for the choice of the distribution of the columns over the zones.

30. SMB vs. PowerFeed (multiple shooting)
26.0 % higher Productivity