e Pergamon

Computers and Chemical Engineering Supplement (1999) S715-S7l8

© 1999 Elsevier Science Ltd. All rights reserved
PII: S0098·1354199/00188·X

Simulation of Chromatographic Processes

using MATLAB
Bernt Nilsson*, Per Borgqvist, Anders Axelsson and Guido Zacchi
NUTEK Competence Centre for BioSeparation and Dept. of Chemical Engineering I,
Lund Institute of Technology, P.O.Box 124, SE-221 00 Lund, Sweden
email: bernt.nilsson@kat.lth.se http://www.kat.lth.se
Abstract
This paper deals with the dynamic simulation of chromatographic processes using MATLAB. Three different
simulation tools are presented,; BIOPR02, CPS toolbox and an application using of FEMLAB. Our main
application is separation of biomolecules, like proteins, using different kinds of stationary phases. BIOPR02 has
a user-friendly graphical interface and is mainly based on simplified models for fast simulation. The CPS toolbox
can facilitate the development of a simulation of different classes of models based on a number of simulation
methods. FEMLAB is a general modelling and simulation system based on finite element method.

Keywords: Chromatography, Computer tool, Modelling, Numerical methods, Process model, Simulation.

INTRODUCTION
Chromatography is one of the most important
separation methods in pharmaceutical and
biotechnology industry. It is often operated in a
stepwise manner creating a production cycle. This
paper deals with the dynamic simulation of
chromatographic processes using MATLAB. Three
different simulation tools are presented; BIOPR02,
CPS toolbox and an application ofFEMLAB.
BIOPR02 is a program implemented in MATLAB
with a user-friendly graphical interface. It is mainly
based on simplified models for fast simulation to make
simply parameter studies. The CPS toolbox can
simulate different classes of models based on a
number of simulation methods. It can be used for
simulations of whole batch cycles. Finally the paper
briefly presents the use of FEMLAB in
chromatographic applications. FEMLAB is a general
modelling and simulation system based on finite
element method. see Comsol (1998). It is
commercially available as an add-on to MATLAB.
PREPARATIVE CHROMATOGRAPHY
Chromatography is used both in analytical and
preparative applications in pharmaceutical and
biotechnology industry. It is based on the interaction
between a mobile phase and a stationary phase. The
interaction can be based on different chemical and
physical mechanisms that creates the actual separation,
Ion exchange chromatography uses the charge of 'the
molecules as the separating mechanism, affinity
chromatography utilises the affinity of the molecules
to adsorb to active sites on a stationary carrier and gel
filtration separates the molecules due to their diffusive
properties in a stationary gel phase. Chromatographic
processes are often performed in series creating an
important part of the downstream process. A general
reference to the subject is Sofer and Hagel (1997).
A preparative chromatography process is often
performed in packed bed columns. The packing
material is a porous gel, i.e. small beads created by a
matrix of a polymer structure. There are other kinds of
column configurations, like expanded beds and
moving beds, but this work is focused on the
traditional packed bed column. A separation is
achieved when one or many of the substances in the
fluid, the mobile phase, is adsorbed on the packing
material, the stationary phase. The separation is
complete when the packing material is saturated. Then
the column is washed for impurities. Finally the
adsorbed substance is eluted by using an eluant at a
certain pH or ion strength. Before the separation can
be restarted the column must be regenerated by an
aditional washing and cleaning. This creates a batch-
wise process with many important design issues to be
considered, like operating conditions for each phase,
recovery of product, cycle time, column efficiency,
etc. All these parameters have to be considered in
order to get optimal conditions.
CHROMATOGRAPHY MO})ELS
Chromatographic processes can be described with
different types of models which all are dependent in
time and space, i.e.vinstationary distributed parameter
systems. These are expressed as a set of partial
differential equations, PDEs. The different models can
be classified as homogeneous or heterogeneous
models. Homogenous models describe the adsorption
in the column dimension. Heterogeneous models
describe it in a microscopic bead dimension which is
coupled with the macroscopic column dimension.
5716 Computers and Chemical Engineering Supplement (1999) S715-S718
Homogeneous Chromatography Models
All quantities are expressed in the same space
dimensions in homogeneous models. One first
example is a simple chromatography model based on
convection and adsorption. The concentration in the
void liquid is modelled by a plug-flow convection and
an adsorption term as
ae v Fc I-£caq
"Jt=- £c J;---e:-Jt
where e is the void liquid concentration, t is time and z
is the column axial dimension, v is the liquid velocity,
E.. is the void fraction and q is the concentration of the
adsorbed substance. The adsorption can be modelled
by a kinetic expression or by an equilibrium isotherm.
The adsorption equilibrium is often expressed as a
Langmuir isotherm like
qm".e
q= KJ +e
where qmax is the maximum adsorption capacity and Ktl
is the dissociation coefficient. This model with
adsorption equilibrium becomes a nonlinear first-order
PDE in time and one in space dimension. For a
detailed discussion see chap. 6 in Varma and
Morbidelli (1997). The second example of a simple
model utilizes Langmuir kinetics instead of the
isotherm. Langmuir kinetics is expressed as
aq
Jt = kad,e(qm;u - q) - kJcsq
where kadf is the adsorption coefficient and kd", the
desorption coefficient. This model thus consists of a
set of two equations and it has a famous analytical
solution by Thomas, see for instance Chase (1984).
An extension of the two homogeneous models
presented above is to add dispersion in the void liquid
as
de Daxaze vae I-£caq
"Jt=e: aZ2 - E
c
az ---e:-Jt
where Dux is the axial dispersion coefficient.
Heterogeneous Chromatography Models
A model of a porous gel bead is described with
diffusion and adsorption as
Dc(a Zp 2a p ) I aq
Jt=e;
a p
a r Z +;J; - £p Jt
where p is the pore liquid concentration, r is the
radius, De is the effective diffusion and Ep is the bead
porosity. The last term can be described by a kinetic or
an equilibrium expression. The model describes the
change of the concentration profile in the bead. The
pore concentration at the boundary, PII, is used in the
mass transfer term in the coupled void liquid equation
ae Dax azc v de I-Ec
-=-- ------ka(c-p)
at e, a ZZ Ec a Z Ec R facilitate the development of a simulation setup of
and here k is the mass transfer coefficient and a is the
mass transfer area. PII is the pore liquid concentration
at the bead surface, i.e, at r=R. This example of a
heterogeneous model becomes a set of two parabolic
PDEs in two space dimension. The model is called
heterogeneous as the two space dimensions are
described separately. They only interact at a boundary.
This model with kinetics is presented in Carlsson et al.
(1994) and with equilibrium in Bautista et al. (1998).
BIOPR02
Some of the models presented above have analytical
solutions for special cases and many of them can be
found in the literature. One set of solutions has been
implemented in a MATLAB program called
BIOPR02, see Tauson and Levander (1997) and
Berggren (1998). The program can simulate three
different classes of models:
i) Homogeneous kinetic model describes plug-flow
convection and Langmuir kinetics and it can be
solved for pulse and step responses.
ii) Diffusive model contains descriptions of film mass
transfer, pore diffusion without accumulation and
adsorption equilibrium, see Arnold et al. (1985).
iii) Gel filtration model capture dispersion, film mass
transfer, pore diffusion without any adsorption.
The interface helps the user to calculate a number of
physical parameters based on the selected protein. If
protein data is lacking in the program a user-defined
protein can be created. The user can select between the
calculated parameters or define new ones.
I i" .......
,ro;-
-",,-
~
~
rTi'i1
rm-
r-;;--
j t"~~ !!'* _ t .(li<' ~
1 1:'.;.1)4 . ,... ~ .. ~
....---- l ~
.•:
......
.....
'lO"I ~
~
A. . . . .
~
~
~
Figure I: Parameter settings for Example I
Example 1: Breakthrough Responses
The conditions for a breakthrough response simulation
for the protein BSA (Bovine Serum Albumine) is
defined in Figure 1. Two simulations of the
breakthrough response are shown in Figure 2. The
steepest response is a result of the diffusive model
with both film mass transfer and pore diffusion
resistance. The second response is the Thomas
solution, i.e, the kinetic model.
CPS TOOLBOX
The CPS toolbox is a set of MATLAB M-files that
chromatography models. Parabolic partial differential
equations are often solved numerically by the method
of lines, MOL. This means that the space dimension is
 Computers and Chemical Engineering Supplement (/999) S715-S718
Figure 2: Breakthrough responses for Example I based
on diffusion (steepest response) and kinetic model.
discretized and the resulting system becomes a large
set of ordinary differential equations in time, ODEs. In
CPS toolbox the space derivatives can be discretized
with three different methods, finite difference method,
FDM, finite element method, FEM and orthogonal
collocation method, OCM. One example of MOL
applied to one of the homogeneous models described
above is the following.
2e
de D""d vde I-t:'·(k c{ ) k )
Jt= 7: d Z2 - t: d Z ---;;:- ad, qm>x -q - Jcsq
c (1989).
The model describes dispersion, convection and
Langmuir adsorption kinetics. The general form of the
discretized model becomes
d a D(U ( K
M-=;-=- K/lC') - -
V (
Kia + K1Be)
o t e:. 2a+ 2 . t;.
_ I ~ t: (klkl,qma,Koa -kad,KI~qa - kJ<JKI~q)
c
c.c ~ The polynomials are weighted by a coefficient which
where a is the value of the concentration at certain
points in the domain. The discretization methods
calculates the M and K matrices. Note that all terms
except one are linear. The resulting ODE system can
be solved using solvers in MATLAB. The most
frequently used solver is ode15s, an implicit,
multistep NDF method for stiff and sparse problems,
see Shampine and Reichelt (1997).
Finite Difference Method
In FDM the space dimension is discretized into a set
of grid points. Space derivatives of a quantity are
approximated with a relation between the values of the
quantity at the grid points. The relation is found by
using Taylor expansions. The accuracy of the
approximation defines the number of values and the
relation coefficients that are used in the
approximation. A second order accurate central-
difference approximation of the first and second
derivatives looks as follows
dC cj +1 -ci-I d 2C CH I - 2cj +cj _ 1
dZ'" 2h ; d Z2 '" h2
In the CPS toolbox there are first, second and fourth
order FD-approximations. In column models FDM
needs upwind correction and there is a set of backward
difference approximations of first, second and fourth
5717
order. A good introduction is found in Davis (1984)
and high order approximations are presented in
Schiesser (1991).
Finite Element Method
In FEM the solution is approximated with a set of
finite elements. The elements are piece-wise
polynomials, the so-called base functions. The method
of weighted residuals, MWR, is used to find the
element coefficients using Galerkin. This means that
for every element i a weighted integral can be defined
as
f Rj(t,z)W(z) = 0
where R is the residual form of the equation and W is
the weighting function. This is the same as the base
function in the Galerkin MWR. The integral results in
one equation for each base function, i.e. each element,
see Davis (1984) for details. These equations can be
expressed in matrix form and the M and K matrices.
Each matrix element in M and K is calculated by an
integral. In the case of column models dominated by
convection the FEM must be modified with
streamline-diffusion stabilization, see Szepessy
Orthogonal Collocation Method
Parabolic PDEs are successfully solved by the global
orthogonal collocation method, see Villadsen and
Michelsen (1978). The solution of the problem is
approximated by a set of orthogonal polynomials.
These polynomials are defined in the whole domain.
is found by the method of weighted residuals. MWR.
The weights are Dirac functions at the collocation
points and the MWR-integral becomes a set of
equations. These can be expressed in matrix form and
the M and K matrices are found. For a good
introduction see chap. 8 and 12 in Rice and Do (1995).
Example 2: Intraparticle behaviour
The dynamic response of a heterogeneous model with
Langmuir kinetics is shown in Figure 3. The problem
is solved with a hybrid discretization method using 50
....
:;[* 4[j})
I
-0 5
o 50 100 ' 50 200 250 300 350
norm. tirr..
Old oons .n 10 N r1iclee
100 150 200 250 300 3 50
Figure 3: Dynamic responses of 10 grid points in the
void liquid with corresponding particle grid points.
 57t8 Computers and Chemical Engineering Suppl ement (/999) S715-S718
internal grid points -in the void and 4 points in the
particles, in total 450 states . In the upper plot the
responses of 10 (of 50) internal grid points in the void
is shown. The lower plot shows the responses of the
corresponding 10 pore liquid concentrations in the
particle. The void equation is discretizcd with a fourth
order FDM with upstream correction. The particles are
discretized using Galerkin-FEM with 4 linear base
function elements.
SIMULATION WITH FEMLAB
FEMLAB is a new general problem solving tool based
on finite element method, see Comsol (1998), and it is
an add-on to MATLAB. It has an user-friendly
graphical interface for problems defined in two space
dimensions.
t
Sr--- f--
.,t.s as
Figure 4: The geometry definition in Example 3.
Example 3: Column Profiles
A modification of the heterogeneous model with
Langmuir kinetics has been implemented in FEMLAB.
The void liquid is described in one domain, the upper
domain in Figure 4, and the gel beads are defined in
the lower d<1main. The domains interact at the phase
boundary. The void liquid flows horisontal from -I to
1 in the upper domain and the pore liquid diffuse
vertical from 0.4 to 0 in the lower domain. The result
from a simulation using streamline-diffusion
stabilization is shown in Figure 5. The figure shows
the concentration surface profile in the column just
before breakthrough. It is clearly seen that there is a
quite steep concentration gradient in the pores.
os
Figure 5: Concentration surface in the column. It
correspond to the geometry definition in Figure 4.
CONCLUSIONS
This paper has presented three different MATLAB
tools for simulation of chromatographic processes.
The BIOPR02 program uses simplified models in
order to make fast simulations for parameter studies.
The CPS toolbox facilitates the development of
simulations based on the state-of-the art in numerical
solution technique. This toolbox is well suited for
whole batch cycle simulations. The use of FEMLAB
in chromatography applications is briefly discussed.
ACKNOWLEDGEMENTS
The authors thank D. Levander, P.-A. Tauson and J.
Berggren for their work on versions of BIOPRO.
REFERENCES
Arnold, F.H., Blanch, H.W., and Wilke, C.R., (1985),
Chem Eng J., 30, B9-B23.
Bautista, L.F., Martinez, M., and Aracil, J., (1998),
"Phenomenological Modelling of Chromatographic
Processes in Biotechnology", In Proc. of SPICA 98,
Tondeur (ed).
Berggren, J., (1998), "Modellering av kromatografisk
separation av proteiner", (in Swedish), MSc Thesis,
LUTKDHI(TKKA-5005)11-47/(f998), Dept. of
Chemical Engineering I, Lund University, Lund,
Sweden.
Carlsson, F., Axelsson, A., Zacchi, G., (1994),
"Mathematical modelling and parametric studies of
affinity chromatography", Computers Chem.Engng.,
Vol 18, Suppl, p.657-661.
Chase, H.A., (1984), "Prediction of the Performance
of Preparative Affinity Chromatography", J. of
Chrom., 297, pp 179-202.
COMSOL, (1998), FEMLAB- Reference Manual,
COMSOL AB, Stockholm, Sweden.
Davis, M.E, (1984), Numerical Methods and
Modeling for Chemical Engineers, chap 2-4.
Rice, R.G :.. and Do, D.D., (1995), Applied
Mathematics and Modeling for Chemical Engineers,
Wiley, chap. 8 and 12.
Schiesser, W.E., (1991), The Numerical Method of
Lines, Academic Press.
Shampine, L.F., and Reichelt, M.W., (1997), "The
MATLAB ODE Suite", SIAM J. SCi. Comput., Vol.
18, No.1, pp. 1-22.
Sofer, G., and Hagel, L., (1997), Handbook ofProcess
Chromatography, Academic Press.
Szepessy, A., (1989), "Convergence of the Streamline
Diffusion Finite Element Method for Conservation
Laws", PhD thesis, Dept. of Mathematics., Chalmers
Institute of Technology, Goteborg, Sweden.
Tauson, P.-A., and Levander, D., (1997), "Simulering
av kromatografi- och adsorptionskolonner", (in
Swedish), MSc Thesis, LUTKDHI(TKKA-5020)/I-
68/( I997), Dept. of Chemical Engineering I, Lund
University, Lund, Sweden.
Varma, A., and Morbidelli, M., (1997), Mathematical
Methods in Chemical Engineering, Oxford University
Press, chap 6.
Villadsen, J., and Michelsen, M.L., (1978), Solution of
Differential Equation Models by Polynomial
Approximation, Prentice-Hall.