Separation Science and Technology

ISSN: 0149-6395 (Print) 1520-5754 (Online) Journal homepage: http://www.tandfonline.com/loi/lsst20

Modelling of packed bed adsorption columns for

the separation of green tea catechins

Miguel Monsanto, Adithya Thota Radhakrishnan & Edwin Zondervan

To cite this article: Miguel Monsanto, Adithya Thota Radhakrishnan & Edwin Zondervan
(2016) Modelling of packed bed adsorption columns for the separation of green tea catechins,
Separation Science and Technology, 51:14, 2339-2347, DOI: 10.1080/01496395.2016.1211146

To link to this article: http://dx.doi.org/10.1080/01496395.2016.1211146

Accepted author version posted online: 22

Jul 2016.
Published online: 22 Jul 2016.

Submit your article to this journal

Article views: 26

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=lsst20

Download by: [Eindhoven University of Technology] Date: 19 September 2016, At: 23:36
SEPARATION SCIENCE AND TECHNOLOGY
2016, VOL. 51, NO. 14, 2339–2347
http://dx.doi.org/10.1080/01496395.2016.1211146

Modelling of packed bed adsorption columns for the separation of green tea
catechins
Miguel Monsantoa, Adithya Thota Radhakrishnana, and Edwin Zondervanb
a
Department of Chemical Engineering and Chemistry, Eindhoven University of Technology, Eindhoven, The Netherlands; bDepartment of
Production Engineering, Bremen University, Bremen, Germany

ABSTRACT ARTICLE HISTORY

Green tea is a rich source of catechins, which when purified have a high economic value as they can be Received 16 December 2015
used as a supplement in several products, to increase their health benefits. Catechins are regarded as Accepted 6 July 2016
desired components with several applications in a variety of areas such as foods, cosmetics and KEYWORDS
pharmaceuticals. A multicomponent sorption model has been developed for the separation of Adsorption model;
catechins from liquid tea streams, with macroporous resins in a packed bed column. Two commer- breakthrough curves;
cially available food grade resins were considered: Amberlite XADHP and Diaion HP20. For the macroporous resin; process
desorption step, two food grade solvents are used: water and ethanol. The adsorption and desorption design; tea polyphenols
behaviour is subsequently mathematically described with one-dimensional axial dispersed plug flow
model that can accurately simulate the dynamics of the solvent swing sorption columns. The model
parameters were regressed from experimental data. Five components are modelled in the competi-
tive sorption: the main four catechins present in green tea and caffeine. The model was used for the
process design and optimization for the recovery of catechins from green tea.

Introduction
The growing market of functional foods is related to
the increasing consumer demand for more natural and
healthier food additives, like catechins, which are the
main polyphenols present in green tea. Catechins have
been studied widely for their potential health aspects,
and upon isolation and purification from green tea
could be used to meet these consumer demands.[1–3]
The structure of the four main catechins from green tea
(epicatechin (EC), epigallocatechin (EGC), epicatechin gal-
late (ECg) and epigallocatechin gallate (EGCg)) is presented
in Fig. 1.
At an analytical and preparative level, chromatogra-
phy has been commonly used for the separation of
green tea catechins,[4] while adsorption is commonly
used for the separation of polyphenols from tea. The
use of macroporous resins as an adsorbent has shown
good potential for the separation and purification of
green tea catechins. Adsorption has the advantage of
allowing the use of food grade materials and solvents.
However, since the reported purities are not very high,
additional purification steps are usually required,[5,6]
because catechins and caffeine adsorb competitively
on the hydrophobic resin. It is, therefore, important
to have a process that separates catechins with high
yield and low amounts of caffeine.
For this purpose, the present study employs a sol-
vent swing adsorption technique where two different
commercially available macroporous resins (Amberlite
XAD7HP and Diaion HP20) are used. The resin selec-
tion is based on the work from Sevillano et al.,[7] where
the resin XAD7HP is found to be preferred for the
adsorption of catechins, and the HP20 is the best
resin for caffeine adsorption.
By using a solvent swing operation there is a change
in the sorption equilibrium of the components, due to
the change in the composition of solvent during the
sorption process.[8]
In a packed bed adsorption column, nearly all
solutes can be adsorbed for a limited amount of time.
The adsorption process starts with an unsaturated bed
of adsorbent and in an ideal fixed bed adsorption with
plug flow conditions (very small internal and external
mass-transfer resistances and negligible axial disper-
sion), equilibrium is reached almost instantaneously.
The objective of this work is to develop a mathematical
model that can describe the sorption of green tea poly-
phenols on a packed bed column with a macroporous

CONTACT Edwin Zondervan edwin.zondervan@uni-bremen.de Department of Production Engineering, Bremen University, Leobener strasse,
Bremen, Germany.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/lsst.
© 2016 Taylor & Francis
2340 M. MONSANTO ET AL.
Figure 1. Green tea catechins: (a) epicatechin (EC), (b) epigalloca-
techin (EGC), (c) epicatechin gallate (ECg) and (d) epigallocatechin
gallate (EGCg).
resin. The input data for the model consist of kinetics,
adsorption equilibrium and adsorbent characteristics. As
an output, the model simulates breakthrough and elution
curves for catechins and caffeine (for the adsorption
cycles), which can be used for process optimization, thus
avoiding time consuming and costly experiments.
The adsorption kinetics for the mass transfer can be
described by adsorption on local sites and mass transport
in series, with two types of models: either based on a local
adsorption assumption between the solid and bulk phase
or based on the mass-transfer resistance between adsor-
bent particle and fluid phase.[9]
The first class of models (local adsorption) are used
for high mass-transfer rates, where the effect of mass-
transfer resistance to the particle can be neglected and
include local equilibrium or local kinetic theories.
The second class (mass-transfer resistance) may
include a rigorous or approximated approach. The rig-
orous approaches are time consuming as they account
for the mass-transfer inside the particle along its radius,
while approximated models are less time consuming.
The most widely used approximated method is the linear
driving force (LDF) concept.[10]
To reduce the calculation time and describe the
adsorption kinetics, the LDF concept is used with a
lumped parameter for the overall mass-transfer resis-
tance. The thermodynamic equilibrium is modelled
with the multicomponent Langmuir isotherm.[9,11,12]
The resulting model can be used for process design and
optimization (column dimensions and operational para-
meters) to separate the catechins from green tea. The
dynamic column process model predicts the output of
adsorption column(s) with a solvent swing adsorption,
where the feed is green tea extract and the eluent is a
water–ethanol solution.
Experimental
Adsorbents
Two non-functionalized food grade polymeric adsorbents
are used: Amberlite XAD7HP and Diaion HP-20,
obtained from Sigma-Aldrich (Zwijndrecht, The
Netherlands). The XAD7HP is a non-ionic aliphatic
acrylic polymer and the HP20 has a polystyrenic matrix
cross-linked with divinylbenzene. The adsorbents are pre-
treated in three steps to remove the salts: first washed with
water, then with ethanol and afterwards again with water.
Reagents and equipment
For the high-performance liquid chromatography
(HPLC): glacial acetic acid (from Merck KGaA,
Darmstadt, Germany), analytical grade acetonitrile
(from Sigma-Aldrich) and Milli-Q water were used.
Catechins standards and freeze-dried green tea powder
were supplied by Unilever R&D (Vlaardingen, The
Netherlands).
The HPLC equipment is an Agilent 1220 Infinity LC
gradient system with a variable wavelength detector and
the column is a phenyl-hexyl (Luna 5 µm Phenyl-Hexyl),
from Phenomenex, Inc (Utrecht, The Netherlands).
The HPLC pump used in the column experiments was
from Knauer (Smartline Pump 1000) (Berlin, Germany),
and the glass column (1.5 cm diameter and variable bed
height) was from Omnifit (Distrilab BV, The Netherlands).
The ultracentrifuge is a Optima L-90K from Beckman
Coulter (Utrecht, The Netherlands).
Sample preparation
Green tea was prepared by extraction of freeze-dried tea
powder with water at 85°C in an Erlenmeyer. The tea
extract was afterwards centrifuged for 20 min at 20,000
rpm and the two resulting phases were separated by
decanting.
Dynamic sorption experiments
The glass column (Omnifit) was packed with the wet
resin up to the target bed height (see Table 1). The
column was initially flushed with water, after which the
clear tea extract was pumped via the bottom of the
column with an HPLC pump. Samples were then

Table 1. Packed bed column parameters for the XAD7HP and
HP20 experiments.
Experiment 1 Experiment 2
Resin XAD7HP HP20
Column diameter (m) 0.015 0.015
Bed length (m) 0.030 0.028
Adsorption flow rate (cm3/min) 0.89 1.0
Desorption flow rate (cm3/min) 1.0 1.0
taken at defined intervals at the outlet of the column
and analysed via HPLC (Fig. 1). In the next step, the
column was again washed with water to remove non-
adsorbed components. In the last step, a 70% (m/m)
ethanol–water eluent was used for the desorption. All
the experiments were performed at room temperature
(20°C).
HPLC analysis for the quantification of catechins
and caffeine
The four main green tea catechins (EC, EGC, ECg and
EGCg) and caffeine (Caff) concentrations were deter-
mined by gradient elution on a C18 column (Luna 5
μm Phenyl-Hexyl, from Phenomenex, Inc.).
Two mobile phases are used for the gradient elution
with a flow rate of 1 cm3/min: mobile phase A is acetic
acid in water (2% v/v) and mobile phase B is acetic acid
in acetonitrile (2% v/v). The following gradient cycle
was used: 5% of B for 40 min, 18% of B for 10 min,
18%–50% of B over 0.1 min, 50% of B for 5 min, 50%–
5% of B over 0.1 min and 5% of B for 20 min.
The catechins concentration is calculated by adding
the concentration of the four individual catechins.
Model development
The use of mathematical models which describe the
column dynamics of adsorption process allows the pre-
diction of breakthrough and elution curves.[13] For the
process design it is possible to simulate the adsorption/
desorption cycles, which can be used for operational
process design optimization.
Multi-component Langmuir isotherm
Due to the multicomponent nature of the tea solution, the
multicomponent Langmuir adsorption isotherm was used
to characterize the amount of adsorbate adsorbed on the
adsorbent. The multicomponent Langmuir equation (Eq.
(1)) is applied for modelling the adsorption behaviour of
the studied species based in a monolayer adsorption and
considering no interactions with the adsorbates.[14]
SEPARATION SCIENCE AND TECHNOLOGY 2341
Qm;i bi Ci
qi ¼ P (1)
1 þ nj¼1 bj Cj
where qi is the amount of component i adsorbed andQm;i is
the maximum adsorption capacity of component i. b is the
Langmuir constant and C is the adsorbate concentration of
each components i,j, in a n-component solution.
Column adsorption model
To model the dynamic behaviour of an adsorbent particle
on the column, a mass balance is set up across a slice of
the packed bed column. The flow is represented by using
Ruthven’s axially dispersed plug flow,[15] which includes
both the convection and axial dispersion effects. The
differential fluid phase mass balance is given by Eq. (2).
The following assumptions are taken into consid-
eration for this model: (i) instantaneous and isother-
mic adsorption, (ii) negligible bed pressure drop, (iii)
spherical adsorbent particles and (iv) perfect radial
mixing.
The LDF approximation (see Eq. (3)) was used for
the adsorption kinetics.[16] The LDF parameters can be
easily regressed to fit the experimental data.
@Ci @Ci @ 2 Ci ð1  εb Þ @qi
¼ v þ Daz;i 2  ρp (2)
@t @z @z εb @t
@qi  
¼ kov;i qs;i  qi (3)
@t
where εb is the bed porosity, Ci is the concentration of
the component i, Daz;i is the axial dispersion coefficient,
ρp is the density of the adsorbent, qi is the concentra-
tion of component i on the adsorbent surface and kov;i
is the overall mass-transfer resistance.
For the model calculations, Eqs. (2) and (3) are set
up for the individual components (four catechins and
caffeine) and solved simultaneously. The coupled par-
tial differential equations (PDEs) are discretized using
upwind finite differences and solved in MATLAB.
To calculate the overall mass-transfer coefficient kov;i
a lumped parameter is used, representing the intrapar-
ticle diffusion and the external mass-transfer resistances
(see Eq. (4)).
1 1 1
¼ þ (4)
kov;i kint;i kext;i
wherekext;i is the external mass-transfer resistance and
kint;i is the internal mass-transfer resistance.
The initial condition for starting the column opera-
tion (first step) is a packed column filled with water. A
step response at the inlet of the column is then applied,
2342 M. MONSANTO ET AL.

where the initial conditions of one step match the final
conditions of the previous step.
The Danckwerts boundary conditions are used at the
inlet (z = 0) and at the outlet (z = L) of the column (Eq.
(5)) and the initial conditions are given by Eq. (6).[17]
@Ci v   @cCi
jz¼0 ¼ Ci  Cifeed ; j ¼0 (5)
@z Daz;i @z z¼L
Ci ðz; t Þ ¼ 0 (6)
where v is the superficial fluid velocity.
To model the breakthrough curves, the coupled
PDEs with initial and boundary conditions are discre-
tized into ordinary differential equation (ODEs), using
upwind finite differences (Eqs. (7) and (8)), to repre-
sent the flow of an adsorbate through a column. The
system of ODEs was solved in MATLAB using an
implicit solver (ode15s) for stiff differential equations,
based on numerical differentiation formulas (NDFs).
dCi Ci ðzn Þ  Ci ðzn1 Þ
¼ (7)
dz Δz
The sensitivity analysis of the axial dispersion coeffi-
cient Daz;i shows that an increase of 200% has no influ-
ence on the model output (Fig. 2). This result shows that
the mass transfer and not the axial dispersion is the main
responsible factor for the apparent dispersion.
The results in Fig. 2 show that when the overall
mass-transfer coefficient (kov;i ) is reduced by 10%, the
model results are strongly influenced. This result was
already expected since the mass transfer is the limiting
factor for the resin saturation rate.
For the isotherm parameters ðQm;i and bi Þ, an
increase of, respectively, 10% and 100% show a signifi-
cant influence on the shape of the breakthrough curves
and on the calculated results (see Fig. 2).
The sensitivity analysis shows that three ðkov;i , Qm;i
and bi Þ of the four tested parameters strongly influence
the column model and should be regressed from the
experimental data.
Parameter regression

d2 Ci Ci ðzn1 Þ  2Ci ðzn Þ þ Ci ðznþ1 Þ
¼ (8)
dz2 Δz2
Results and discussion
Sensitivity analysis
A sensitivity analysis was performed to evaluate which
parameters influence the calculated results, when using
the column adsorption model. Four parameters ðDaz;i ,
kov;i , Qm;i and bi Þ from the model were tested (Fig. 2).
Figure 2. Experimental setup for the dynamic sorption
experiments.
The inbuilt MATLAB function fsolve was used as an
optimization routine to regress the parameters kov;i ,
Qm;i and bi . This function uses the Levenberg–
Marquardt (LM) algorithm to minimize the residual
between the numerical and experimental data.[18] The
LM algorithm is an iterative technique for solving non-
linear least-square problems. It finds the minimum in a
multivariate function that can be expressed as the sum
of squares of nonlinear functions.
Two packed bed column experiments were performed
at similar conditions using two resins: XAD7HP and
HP20 (Table 1).
The breakthrough and elution curves comparing the
model and experimental results are shown in Figs. 3
and 4 for Experiments 1 and 2, respectively.
The results in Fig. 3 clearly demonstrate that the
model based on the regressed parameters for the
XAD7HP resin has a good fitting. The EGC is the
fastest component to reach saturated bonding, while
the ECg still did not reach saturation at 150 min.
Upon elution the different adsorption capacities of the
catechins can be observed. The ratio of the compounds
in the eluent is different from the ratio of the com-
pounds in the feed. While EGC and caffeine have
similar amounts in the feed, in the eluent only half
the amount of caffeine to EGC is found. In the specific
case of the elution curves for ECg and caffeine, the
model over predicts the maximum concentration
value. Nevertheless, the model follows the trend for all
the components.

Figure 3. Sensitivity analysis results for the adsorbed components: EC, EGC, ECg, EGCg and Caff. (black line, original; red line,
modified) for the selected parameters: Daz,i, kQV,I, Qm,i and bi.
The results in Fig. 4 show that the model is able to
predict the behaviour of all the components, in both the
breakthrough curves and the elution curves, with a high
accuracy, for the HP20 resin.
The breakthrough curves in Figs. 3 and 4 also show
some overshoot for the smaller components (EGC and
caffeine), which is related to the components’ different
rates of diffusion and affinities for the resin. In this
case, the smaller components are more adsorbed in an
initial stage and afterwards partially desorbed, when the
bulkier components with more affinity arrive to the
same adsorption sites.
The regressed parameter values are presented in
Tables 2 and 3 for Experiment 1 and 2, respectively.
There is a clear difference between the values of the
Langmuir equilibrium constants ðQm;i , bi Þ for the two
resins. The XAD7HP resin has a somewhat higher Qm
for the gallated catechins (ECg and EGCg), while the
HP20 resin has a much higher Qm for the gallated
catechins (ECg and EGCg) and caffeine.
SEPARATION SCIENCE AND TECHNOLOGY 2343
The XAD7HP resin has a higher affinity for all the
components, as reflected in their Langmuir constant bi,
when compared with the HP20 resin. This can be
explained by the fact that the HP20 resin has a much
higher affinity for the caffeine than for the catechins,
indicating that it is a good option to use HP20 first to
separate caffeine from the other components.
Process design
The model that represents the column dynamics can be
used to estimate operational parameters of a packed bed
column(s) process, for the recovery of catechins from
green tea. Two operating designs are tested: in Design 1
a two-column process is used with the objective of max-
imizing the amount of catechins and minimize the con-
tamination of the catechins with the caffeine; in Design 2
a single column is used with the purpose of maximizing
the amount of catechins recovered. All the columns have
2344 M. MONSANTO ET AL.

sold as a dried powder, a drying step is necessary. This

can be done with a Spry drying process (you can refer to
the paper with the modelling of the SpryDry), where the
ethanol is completely removed and the final product is a
green tea catechins concentrated powder. The final objec-
tive is to achieve an efficient process design for the
separation of green tea catechins with high yield and
purity. In a preliminary process design, the regressed
parameters are optimized for the operational time of the
columns, in an adsorption and desorption cycle.

Figure 4. Model (–) and experimental (×) breakthrough and
elution curves for Experiment 1 (XAD7HP resin); for experimen-
tal conditions, see Table 1.
the same dimensions: length of 2.2 m and a diameter of
0.2 m. For the desorption, a 70% (m/m) ethanol–water
mixture was used. Food grade ethanol is used not only to
pre-treat the resins but also for the desorption of the
target components. The target components will be present
in a water–ethanol solution. Since green tea catechins are
Design 1
Design 1 (Fig. 5) consists of two adsorption columns in
series connected by a buffer tank and operated in a batch
mode, where the green tea feed is set for a flow rate of 14 L/
min. Tea extract is fed to the first column via the bottom
and the output stream at the top of the column is after-
wards collected in the buffer tank. The output of the buffer
tank is the input of the second column. The output of the
second column is the final stream of the process.
Based on the resin screening results from Sevillano
et al., the first column will be packed with HP20 resin
(favourable for caffeine adsorption) and the second will
be packed with the resin XAD7HP (preferred for the
adsorption of catechins).
In total four operating modes are considered, based on
the possibility of using the output from the adsorption or
from the desorption in each column (Fig. 5). The feed for
the buffer tank, can be taken from the output of either the
adsorption or desorption of the HP20 column. The out-
put of the whole process can either be the adsorption or
the desorption of the second column (XAD7 column).

Table 2. Regressed parameters values for Experiment 1.

Experiment 1 (XAD7HP resin)
Adsorption Desorption
Qm;i (g/g) bi ðdm3 =gÞ kov;i (10−3 s−1) Qm;i (g/g) bi ðdm3 =gÞ kov;i (10−3 s −1
)
EC 0.25 2.59 0.87 0.25 0.063 43.86
EGC 0.25 1.63 1.05 0.25 0.057 33.05
ECg 0.30 16.95 0.12 0.30 0.041 37.37
EGCg 0.31 3.12 0.67 0.31 0.058 37.64
Caffeine 0.22 4.13 0.59 0.22 0.049 07.79

Table 3. Regressed parameters values for Experiment 2.

Experiment 2 (HP20 resin)
Adsorption Desorption
Qm;i (g/g) bi ðdm3 =gÞ kov;i (10−3 s−1) Qm;i (g/g) bi ðdm3 =gÞ kov;i (10−3 s−1)
EC 0.10 0.047 17.3 0.10 0.011 11.5
EGC 0.02 0.026 169.1 0.02 0.011 41.6
ECg 0.96 0.069 1.2 0.96 0.012 19.6
EGCg 0.52 0.041 3.4 0.52 0.011 17.8
Caffeine 0.33 0.119 2.3 0.33 0.021 10.9
 SEPARATION SCIENCE AND TECHNOLOGY 2345

desorption (elution) output of column 2 (XAD7HP

resin). In this mode, column 1 is used to preferably adsorb
the caffeine, while column 2 mostly adsorbs the catechins.
In Fig. 6, the concentration profiles for a complete
operational cycle are presented. For the two-column
cycle, the model optimized operating time is 40 min
for the adsorption step and 45 min for the desorption
step. A washing step of 2 bed volumes (BV) is per-
formed between the adsorption and desorption steps.
One complete operational cycle achieves a 52% yield of
catechins, while reducing the yield of caffeine to 19%.
To better evaluate the relative amounts of catechins
and caffeine after the process a relative purity of cate-
chins (Pcat,caff) is defined (Eq. (9)):
concentration of catechins ðg=LÞ
Pcat;caff ð%Þ ¼  100
concentration of catechins þ caffeine ðg=LÞ
(9)
In the green tea feed (input for the first column) the
relative purity of catechins was 78%. After one cycle,
the catechins’ relative purity increased to 91%.
Figure 5. Model (–) and experimental (×) breakthrough and
elution curves for Experiment 2 (HP20 resin); for experimental
conditions, see Table 1. Design 2
In Design 2, a single column packed with XAD7HP
resin was used and the operational time is optimized
The objective of Design 1 is to maximize the yield of with the single objective of maximizing the yield of
catechins (Ycat) and minimize the yield of caffeine catechins (Ycat).
(Ycaff), see Eqs. (7) and (8). The process flow rate is set at 9.3 L/min and a
mass of catechins in output ðgÞ washing step of 2 bed volumes (BV) is performed
Ycat ð%Þ ¼  100 (7) between the adsorption and desorption steps. The con-
mass of catechins in feed ðgÞ
centration profiles of the components (four catechins
mass of caffeine in output ðgÞ
Ycaff ð%Þ ¼  100 (8) and caffeine) in one operational cycle are presented in
mass of caffeine in feed ðgÞ
Fig. 7. For this design, the optimal operational time is
The operational mode which achieves the best result is the 20 min for the adsorption step and 65 min for the
one that uses the adsorption output of column 1 (HP20 desorption step (Fig. 7), achieving a yield of catechins
resin), as the input for the buffer, followed by the of 89% and a yield of caffeine of 88%.

Figure 6. Two-column process design diagram (left) and the four possibe operational modes (right).
2346 M. MONSANTO ET AL.
Figure 7. Column operation cycle (adsorption + washing ±
description) for Design 1.
Figure 8. Column operation cycle (adsorption + washing ±
description) for Design 2.
Conclusions
The dynamics of multicomponent solvent swing
adsorption packed bed columns have been simulated
with an axial dispersed column model. Three sensitive
model parameters, overall mass-transfer coefficient
 
kov;i ; maximum adsorption capacity ðQm;i Þ and the
Langmuir constant ðbi Þ; were regressed from the
experimental data.
Five green tea components (four catechins present in
green tea and caffeine) were modelled in the competitive
sorption and show a good fitting to the experimental data.
The modelled breakthrough and elution curves were able to
represent the competitive adsorption between all the
components.
The values of the regressed parameters for the XAD7HP
and HP20 resins demonstrate the different affinities
towards the macroporous resins. The XAD7HP resin has
a higher affinity for all the components, when compared
with the HP20 resin. On the other hand, the HP20 resin has
a much higher affinity for the caffeine than for the cate-
chins, resulting in a good option to preferentially adsorb
caffeine.
For the process design, two operating designs are
simulated and optimized for the operational time of
the packed bed columns.
Design 1 uses a two-column process with the multi-
objective of maximizing the amount of catechins and
minimizing the amount of caffeine. After one opera-
tional cycle, the yield of catechins was 52% and the
yield of caffeine was 19%, for an operation time per
column of 95 min. The relative purity of catechins to
caffeine increased from 78% to 91%, when compared
with the feed green tea solution.
If instead of the multi-objective optimization, the
only objective is to maximize the amount of catechins,
a single column with an operation time of 100 min can
be used (Design 2), achieving a yield for the catechins
of 89%.
Overall, Design 1 proves to be effective for separat-
ing the caffeine and still allows the recovery of more
than half of the catechins, while Design 2 is very effi-
cient for the recovery of catechins.
The current model can also be extended to include
cleaning and regeneration steps, as well as to determine
the optimum number of cycles until regeneration is
needed.
Acknowledgements
Special thanks to David Méndez Sevillano for his contribution.
Funding
This work was supported by the Institute for Sustainable
Process Technology (ISPT), The Netherlands.
References
[1] Balentine, D.A.; Wiseman, S.A.; Bouwens, L.C.M.
(1997) The chemistry of tea flavonoids. Critical
Reviews in Food Science and Nutrition, 37: 693–704.
[2] Khan, N.; Mukhtar, H. (2007) Tea polyphenols for
health promotion. Life Sciences, 81: 519–533.
[3] Velayutham, P.; Babu, A.; Liu, D.M. (2008) Green tea
catechins and cardiovascular health: An update.
Current Medicinal Chemistry, 15: 1840–1850.
[4] Amarowicz, R.; Shahidi, F.; Wiczkowski, W. (2003)
Separation of individual catechins from green tea
using silica gel column chromatography and HPLC.
Journal of Food Lipids, 10: 165–177.

[5] Vuong, Q.V.; Golding, J.B.; Nguyen, M.; Roach, P.D.
(2010) Extraction and isolation of catechins from tea.
Journal of Separation Science, 33: 3415–3428.
[6] Zhao, R.Y.; Yan, Y.; Li, M.X.; Yan, H.S. (2008) Selective
adsorption of tea polyphenols from aqueous solution of
the mixture with caffeine on macroporous crosslinked
poly(N-vinyl-2-pyrrolidinone). Reactive & Functional
Polymers, 68: 768–774.
[7] Sevillano, D.M.; van der Wielen, L.A.M.; Hooshyar, N.;
Ottens, M. (2014) Resin selection for the separation of
caffeine from green tea catechins. Food and
Bioproducts Processing, 92: 192–198.
[8] Wegmann, C.; Garcia, E.S.; Kerkhof, P.J.A.M. (2011)
Kinetics of acrylonitrile adsorption from an aqueous
solution using Dowex Optipore L-493. Separation and
Purification Technology, 81: 429–434.
[9] Siahpoosh, M.; Fatemi, S.; Vatani, A. (2009)
Mathematical modeling of single and multi-component
adsorption fixed beds to rigorously predict the mass
transfer zone and breakthrough curves. Iranian Journal
of Chemistry & Chemical Engineering-International
English Edition, 28: 25–44.
[10] Glueckauf, E.; Coates, J.I. (1947) Theory of chromato-
graphy .4. the influence of incomplete equilibrium on
the front boundary of chromatograms and on the
effectiveness of separation. Journal of the Chemical
Society, 241, 1315–1321.
SEPARATION SCIENCE AND TECHNOLOGY 2347
[11] Fournel, L.; Mocho, P.; Brown, R.; Le Cloirec, P. (2010)
Modeling breakthrough curves of volatile organic com-
pounds on activated carbon fibers. Adsorption-Journal
of the International Adsorption Society, 16: 147–153.
[12] Gogoi, P.; Saikia, M.D.; Dutta, N.N.; Rao, P.G. (2010)
Adsorption affinity of tea catechins onto polymeric
resins Interpretation from molecular orbital theory.
Biochemical Engineering Journal, 52: 144–150.
[13] Xu, Z.; Cai, J.G.; Pan, B.C. (2013) Mathematically
modeling fixed-bed adsorption in aqueous systems.
Journal of Zhejiang University-Science A, 14: 155–176.
[14] Markham, E.C.; Benton, A.F. (1931) The adsorption of
gas mixtures by silica. Journal of the American
Chemical Society, 53: 497–506.
[15] Ruthven, D. (1984) Principles of Adsorption and
Adsorption Processes; John Wiley & Sons, Inc: New York.
[16] Glueckauf, E. (1955) Theory of chromatography .10.
Formulae for diffusion into spheres and their applica-
tion to chromatography. Transactions of the Faraday
Society, 51: 1540–1551.
[17] Wehner, J.F.; Wilhelm, R.H. (1995) Boundary condi-
tions of flow reactor (Reprinted from Chem Engng, vol
6, pg 89–93, 1956). Chemical Engineering Science, 50:
3885–3888.
[18] Fan, J.Y. (2003) A modified Levenberg-Marquardt
algorithm for singular system of nonlinear equations.
Journal of Computational Mathematics, 21: 625–636.