Computers and Chemical Engineering 29 (2005) 1167–1183

Modelling and simulation in chemical engineering:

Tools for process innovation
Alirio E. Rodrigues ∗ , Mirjana Minceva
Laboratory of Separation and Reaction Engineering (LSRE), Department of Chemical Engineering,
Faculty of Engineering, University of Porto, Rua Dr Roberto Frias, 4200-465 Porto, Portugal

Available online 19 March 2005

Abstract

The Chemical Engineering Science movement has served well in solving problems from micro to macro scales. Ultimately, as Professor
R. Sargent pointed out, it would be better if translated in “Scientific Engineering”.
Examples of innovation as a combination of science (concept), technology and process/product will be presented.
The first example is perfusion chromatography based on the concept of “diffusivity augmented by convection” and on the technology
of fabrication of large-pore packings. The second example, Simulated Moving Bed (SMB), is based on a concept developed to overcome
the difficulties in implementing True Moving Bed processes. Technological contributions come from adsorbent materials and rotary valve
to simulate the solid movement. SMB is now a key technology for chiral separations. Modelling/Simulation tools provide sound basis for
design/operation by using the concept of “separation volume”.
The third example is from the area of multifunctional reactors. The SMB technology is extended to the simultaneous reaction/separation.
© 2005 Elsevier Ltd. All rights reserved.

Keywords: Chemical Engineering Science; Process innovation; Technology; Perfusion chromatography; Simulated Moving Bed; Multifunctional reactors

1. Introduction—the Chemical Engineering Science I remember the statement of Professor Le Goff (1970)
movement when I was a student in Nancy “Le Génie Chimique c’est
pas de la plomberie” and the philosophy of modelling that a
The Chemical Engineering Science movement illustrated chemical engineer uses when solving a new problem involv-
in the book of Bird, Stewart, and Lighfoot “Transport ing the writing of:
Phenomena” (2002) follows the approach initiated by
• conservation equations (mass, energy, momentum and
Kramers at Delft in his lecture notes “Physische Transportver-
electric charge);
schijnselen” (Kramers, 1958). It contains the basic material
• equilibrium laws at the interface(s);
to help in developing a mathematical model defined by Aris
• constitutive laws (e.g., ideal gas law);
(1994) as “a complete set of mathematical equations which
• kinetic laws of transport and reaction;
are supposed to correspond to some entity – its prototype –
• initial and boundary conditions;
which can be a physical, biological, social . . . entity although
• optimization criteria.
here we deal with physicochemical systems”.
A process model is a relation between “outputs” and “in- This methodology has been useful to analyze problems at
puts” (feed conditions, design parameters and process ad- various scales involved in chemical engineering:
justable parameters) in view of (i) scale-up from lab to in-
dustrial scale, (ii) prediction of process dynamics and (iii) • pore scale (catalyst and adsorbent): 1–1000 nm;
optimization of operating conditions (Shinnar, 1978). • particle scale: 10 ␮m–1 cm;
• reactor/separator scale: 1–10 m;
∗ Corresponding author. Tel.: +351 22 5081671; fax: +351 22 5081674. Models can be guided by objectives, which can seem con-
E-mail address: [email protected] (A.E. Rodrigues). tradictory: simplification of the reality/idealization and de-

0098-1354/$ – see front matter © 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2005.02.029
1168 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

tailed models to “better” know the reality. On one side they 2. The “art” of modelling
aim at simplification of reality or idealization; this is con-
densed in the words of Denbigh (1951) “In science it is al- Chemical engineers use some techniques in modelling
ways necessary to abstract from the complexity of the real such as: adimensionalization and scaling, averaging, appro-
world, and in its place to substitute a more or less ideal- priate choice of independent variables. However, physical
ized situation that is more amenable to analysis” and led insight is gained as a consequence of such manipulations. A
to the creation of new, simplified models which are a “dig- further step is transforming model results into useful tools
ital impression” of our profession: boundary layer theory, for real life situations.
film model for heat/mass transfer, residence time distribu-
2.1. Scaling and dimensionless groups
tion (RTD) theory (Levenspiel, 2002). On the other hand,
detailed models are necessary when simple models fail to Chemical engineers have some habits as normalization
capture essential behaviour of real systems; this is illustrated of variables; as a consequence of that mathematical opera-
by the Maxwell–Stefan treatment of multicomponent diffu- tion dimensionless groups appear with a physical meaning.
sion (Maxwell, 1952): Consider the problem of diffusion/reaction in an isothermal
n

porous catalyst with slab geometry. The mass balance in
xi Nj − x j Ni
di = ∇xi = (1) steady-state for irreversible reaction of order n is:
j=1
ct DijMS
d2 ci
De − kcin = 0 (3)
The basic laws of momentum transfer (Newton’s law: dz2
τyx = µ dνx
dy ), heat transfer by conduction (Fourier’s law: with boundary conditions (symmetry condition in the centre
qy = −k dT and surface condition)
dy ), mass transfer by diffusion (Fick’s law: jy =
−DAB dρ dci
dy at constant ρ) assume infinite velocity of propa-
A
z = 0, =0
gation of the signal; this problem is eliminated following the dz (4)
proposal of Maxwell (1952) for momentum transfer z = , ci = ciS
∂σyx ∂ux The normalization of space variable x = z/ and concentra-
σyx + τ = −µ (2) tion variable fi = ci /ciS leads to:
∂t ∂y
n−1
d2 fi 2 kciS
where τ is the time constant and the shear stress σ yx −  fn = 0 (5)
in a fluid or solid body σ yx ; similarly for mass trans- dx2 De i
fer (Westerterp, Kronberg, Benneker, & Dil’man, 1996) The dimensionless group φ (Thiele modulus) governing the
and heat transfer (Liu, Chen, & Xu, 1999) one should reaction/diffusion problem is:
∂q
write: j + τ ∂j ∂c ∂T
∂t = −D ∂x and q + τ ∂t = −k ∂x (Vernott and
n−1
kciS
Cattaneo—VC equation). In homogeneous substances, the 2 = φ2 = DaII (6)
De
relaxation time is 10−8 to 10−14 s and Fourier’s law works
for normal heating processes; but in biological systems τ is The physical meaning of the dimensionless group is: φ2 =
of the order of 10–30 s and VC equation applies. DaII = reaction rate/diffusion rate = diffusion time con-
A philosophy of modelling can be based in four points: stant/reaction time constant. Two extreme cases are (Fig. 1):

• start with simple models; obtain from such models in- (a) reaction rate  diffusion rate—concentration profile in-
formation which remains valid for more complex models side the catalyst is almost equal to the surface concentra-
(US$ 10 approach of Levenspiel (2002): “Always start by tion; the catalyst works on “chemical regime”;
trying the simplest model and then only add complexity to (b) reaction rate  diffusion rate—the catalyst works in
the extent needed”); “diffusional regime”.
• the validity of a model is not just a result of a “good fit”;
more important is the capability to predict the system be-
haviour under operating conditions different from those
used to get model parameters;
• good results can only be obtained if the model “well” rep-
resents the system;
• use models to obtain useful design parameters and their
dependence on operating conditions; use independent ex-
periments if possible to get model parameters.
In short, model development is a task to be carefully done
to avoid waste of energy in the next simulation step. “Keep
things as simple as possible, but not simpler” (Einstein). Fig. 1. Chemical regime and diffusional regime in an isothermal catalyst.
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1169

Fig. 2. LDF model.

Fig. 3. Breakthrough curves for unfavourable isotherms.

2.2. Averaging

The linear driving force (LDF) model of Glueckauf (1955) Kĉi

q̂i = (13)
sketched in Fig. 2 is a good example. For a spherical “homo- 1 + (K − 1)ĉi
geneous” adsorbent particle the mass conservation equation
is: A first dimensionless parameter appears: the “capacity pa-
qi0
  rameter” ξm = 1−ε ε ci0 of the adsorption column. A combi-
∂qi 1 ∂ ∂qi
= Dh 2 R2 (7) nation of the independent variables z and t in only one vari-
∂t R ∂R ∂R able T (throughput parameter of Vermeulen (1958)) defined
with boundary conditions (symmetry at the centre and equi- as the ratio of moles of solute passed through the bed sec-
librium with the fluid concentration at the surface through the tion located at ν = Az and the number of moles retained in
adsorption equilibrium isotherm f(ci ): 
the adsorbent 
contained in the volume ν. The new variable
T = ξ1m uzi t − 1 allows us to write the mass balance as
∂qi
R = 0, =0 dq̂i
= T and taking into account the adsorption equilibrium
∂R (8) dĉi
R = Rp , qiS = f (ci ) K
isotherm dq̂
dĉi = [1+ĉi (K−1)]2 we get the breakthrough curve
i

shown in Fig. 3:
Multiplying both members by R2 dR, integrating over the par-
ticle volume (between 0 and Rp ) and introducing the average 
concentration qi leads to: 1− K T
 ĉi = , K ≤ T ≤ 1/K (14)
∂ qi 3Dh ∂qi  3Dh qiS − qi 1−K
= =
∂t Rp ∂R Rp Rp αRp
2.4. From model results to real life
15Dh
= (qiS − qi ) = kh (qiS − qi ) (9) Back to the reaction/diffusion problem for first order re-
R2p
action in isothermal slab catalyst. The concentration profile
is fi = cosh(φx)
cosh φ and the effectiveness factor of the catalyst
φ
2.3. Choice of variables is η = tanh
φ . It is important to know the effectiveness fac-
tor to calculate the amount of catalyst in the reactor needed
Let us consider the equilibrium model of an isothermal to get a given reactant conversion. But the Thiele modu-
adsorption column with plug fluid flow of a diluted stream lus requires that the kinetic constant k must be known (and
(trace system). Model equations are the mass balance of the many times it is not). Hopefully, there are always bright peo-
solute in a bed volume element and the equilibrium law at ple around to transform theoretical results in practical tools.
the interface fluid/solid: Weisz and Prater (1954) changed η = f(φ) in a more useful plot
∂ci ∂ci ∂q∗ η = g(ηφ2 ), where ηφ2 does not require the knowledge of k;
u0 +ε + (1 − ε) i = 0 (10) but only measurable quantities since ηφ2 = rcobs 2
∂z ∂t ∂t iS De
(Fig. 4).
qi∗ = f (ci ) (11)
2.5. Obtaining useful relations between dependent
For an adsorption isotherm of “constant separation factor” variables
type and normalizing the dependent variables, ĉi = cci0i , q̂i =
qi
qi0 we get: Consider the diffusion/reaction/conduction problem in a
non-isothermal slab catalyst. For slab geometry and first order
∂ĉi ∂ĉi 1 − ε qi0 ∂q̂i irreversible reaction conservation equations of mass/energy
ui + + =0 (12)
∂z ∂t ε ci0 ∂t are:
1170 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

3. Models of US$ 10, 100 and 1000 (Levenspiel, 2002)

3.1. US$ 10 models

3.1.1. Equilibrium theory for adsorption columns.

Physical concepts from simple models
The simplest model of an adsorption column model as-
sumes isothermal operation, plug fluid flow, infinitely fast
mass transfer between fluid and solid phases (instantaneous
equilibrium at the interface) and trace system (Rodrigues &
Tondeur, 1981). Model equations are the mass balance in a
bed volume element and the equilibrium law at fluid/solid
interface:
∂ci ∂ci ∂ qi
u0 +ε + (1 − ε) =0 (20)
∂z ∂t ∂t
qi = qi∗ = f (ci ) (21)

where qi is the average concentration in the adsorbent and

qi∗ = f (ci ) is the concentration at the particle surface in equi-
librium with the fluid concentration ci . Using the cyclic re-
lation between partial derivatives we get De Vault equation
(1943):
Fig. 4. (a) Effectiveness factor vs. Thiele modulus and (b) effectiveness
factor vs. ηφ2 .  
∂z ui
uci = = (22)
∂t ci 1 + ε f (ci )
1−ε
d2 ci
De − k(T )ci = 0 (15)
dx2 Those interested in understanding adsorptive and chromato-
graphic processes will recognize this is the most important
d T
2
λe + (−$H)k(T )ci = 0 (16) result to retain. It shows that adsorption in fixed beds is a phe-
dz2 nomenon of propagation of concentration waves. The sim-
with boundary conditions: plest model shows that the nature of the equilibrium isotherm
is the main factor influencing the shape of the breakthrough
dci dT
z = 0, = =0 (17) curve (Fig. 6). The physical concepts to be retained are: dis-
dz dz persive waves are formed when isotherms are unfavourable;
z = , ci = ciS ; T = Ts (18) each concentration propagates with a velocity given by De
Vault equation. Compressive waves are formed for favourable
Multiplying the first equation by the heat of reaction (−$H) isotherms and the physical limit is a shock, which propagates
and adding the second we get, after integrating twice: ui
with a velocity ush = 1−ε $qi , where the slope of the chord
1+ ε $ci
De (−$H)
T − Ts = (ciS − ci ) (19) linking the feed state and the bed initial state appear instead
λe of the local slope of the equilibrium isotherm.
This equation was first derived by Damkholer (1943) and
provides a relation between concentration and temperature 3.1.2. Fluid flow in chemical reactors: residence time
in a point inside the catalyst (Fig. 5). distribution and tracer technology
Danckwerts (1953) approached the study of fluid flow in
reactors in a brilliant and simple way: “introduce a pulse
of tracer into the fluid entering the reactor and see when it
leaves”. The normalized outlet concentration versus time is
related with the residence time distribution.
The study of RTD of flowing fluids and its consequences
can be put under the umbrella of tracer technology. This is
important for chemical engineers, researchers in the medical
field, environment, etc., to diagnose the reactor behaviour,
drug distribution in the body, etc. Danckwerts built a theory
based on the characterization of fluid elements of a popula-
Fig. 5. Concentration and temperature profiles in a non-isothermal catalyst. tion inside the reactor (age and life expectation) and leaving
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1171

3.2. US$ 100 models: diffusion, convection and reaction

in isothermal catalysts—intuition is not enough

The importance of intraparticle convection in the catalyst

effectiveness was analyzed by Nir and Pismen (1997) for
first order irreversible reaction in isothermal catalysts. The
problem was first dealt with by Wheeler (1951); he concluded
that intraparticle convection would be important only for gas
phase systems at high pressure in catalysts with very large
pores. For the reaction A → B in slab catalysts, the mass
balance is:

d2 f df
2
− 2λm − 4φS2 f = 0 (24)
dx dx

with boundary conditions: f = 1 at x = 0and x = 1. Model pa-

k
rameters are: Thiele modulus φS =  De and intraparticle
ν0 
Peclet number λm = (ratio between time constants for
De
pore diffusion and convection). The concentration profile in-
side the catalyst is:

shα2 eα1 (2x−1) − shα1 eα2 (2x−1)

f = (25)
Fig. 6. (a) Unfavorable isotherms and dispersive fronts and (b) favorable sh(α2 − α1 )
isotherm and compressive front.

λm ± λ2m +4φS2
the reactor (residence time). Then, he introduced the “distri- where α1,2 = 2 . Asymmetric profiles (Fig. 7) can
bution” relative to each character; the residence time distri- be viewed as a result of mass transfer by convection inside
bution E(t) is then defined as E(t) dt being the fraction of fluid pores. The effectiveness factor is:
elements leaving the reactor with residence time between t
and t + dt. The next question is how to experimentally have 1/α1 − 1/α2
ηdc = (26)
access to E(t). This brings the tracer technology to the center coth α1 − coth α2
of the arena. The normalized response to an impulse of tracer
C(t) is directly related with the RTD, i.e., C(t) = τE(t); or the When convection is not important, i.e., λm = 0, the effective-
φS
normalized response to a step input of tracer F(t) curve of ness factor becomes ηd = tanh φS . The effect of convection
Danckwerts is E(t) = dFdt(t) . This is a characteristic of lin- can be seen in Fig. 8 where ηdc /ηd is plotted versus λm and
ear systems: the response to an impulse is the derivative of φS . In the intermediate region of Thiele modulus (similar re-
the response to a step input. How this linearity appears in action and diffusion rates), the effectiveness of the catalyst is
this macroscopic vision of fluid flow where Navier–Stokes improved by convection. The pore convection will apparently
ρ Dv
Dt = ρg − ∇P + µ∇ v applies in a detailed description
2
increase diffusivity and move the working regime of the cat-
is a matter to think about. alyst from diffusion to “chemical” controlled. The message
It is also interesting to note that the RTD is the in- is: intuition is not enough!
verse Laplace transform of the transfer function G(s), i.e.,
E(t) = L−1 G(s). This relation allows the calculation of the
moments of E(t) from G(s) and its derivatives at s = 0 (Van
der Laan theorem). Finally, the chemical engineer uses the
hydrodynamic characterization to connect with the reaction
kinetics obtained in a batch reactor, cbatch (t) and predict the
average outlet concentration in a real reactor:
 ∞
cS = E(t)cbatch (t) dt (23)
0

This result is valid for first order reactions. For other reaction
kinetics, it gives the limit when the flow is completely segre-
gated; in the limit of maximum micromixing the Zwietering
equation (1959) holds. Fig. 7. Asymmetric concentration profiles in large-pore catalysts.
1172 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

Fig. 8. ηdc /ηd vs. λm and φS .

3.3. US$ 1000 models: Computational Fluid Dynamics

(CFD)

Levenspiel (2002) summarizes the progress on the study

of fluid flow: “In the 19th century there were two approaches
to study fluid flow: hydrodynamics (dealt with ideal friction-
less fluid; highly mathematical stuff) and hydraulics devel-
oped by civil engineers” who amassed mountains of tables
of pressure drop and head loss of fluids in open and closed
channels of all sort . . .”. At the beginning of the 20th cen-
tury, Prandtl said “Hydrodynamics has little significance for
the engineer because of the great mathematical knowledge
required for an understanding of it and the negligible possi-
bility of applying its results. Therefore, engineers put their
Fig. 9. Two-dimensional flow in a reservoir (a) and RTD (b).
trust in the mass of empirical data collectively known as the
“science of hydraulics”. As Levenspiel says, “Prandtl was the
genius who patched together these different disciplines with sults are available at the following web address
his simple boundary layer theory. The result is modern fluid http://www.fe.up.pt/∼mmalves/cfd/reactor/index.htm.
mechanics”. The following pictures (Fig. 10) illustrate the concen-
Numerical methods for the solution of PDE’s are avail- tration evolution of a tracer inside the reactor. Initially, the
able and the combination of two solid disciplines appears reactor is full of water (blue), and a step profile in the con-
with a new name: “Computational Fluid Dynamics”. Twenty
years ago, we published in ISCRE8 “Residence time distri-
bution in laminar flow through reservoirs from momentum
and mass transport equations” (Brunier, Zoulalian, Antonini,
& Rodrigues, 1984). It is a problem of 2D flow in a reser-
voir of length L and height H, where a stationary laminar
flow exists between inlet and outlet (Fig. 9). The formulation
is made in terms of vorticity and stream function; the flow
field is obtained and the RTD is obtained by solving the mass
conservation equation:
 2 
∂(uΩ) ∂(νΩ) ∂ Ω ∂2 Ω
+ =v +
∂x ∂y ∂x2 ∂y2
∂ψ ∂ψ
u= ; v=−
∂y ∂x
 2  (27)
∂ ψ ∂2 ψ
+ = −Ω
∂x2 ∂y2
 2 
∂C ∂(uC) ∂(vC) ∂ C ∂2 C
+ + =D +
∂t ∂x ∂y ∂x2 ∂y2
This problem was recently solved with modern tools
(Fluent) (Madeira, Alves, & Rodrigues, 2004). Re- Fig. 10. Time evolution of tracer concentration in a 2D reservoir.
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1173

centration of a tracer (red) is imposed at the inlet boundary.

The reduced time is defined as θ = t/τ, where τ represents
the space-time. The streamlines are shown in black.
These pictures clearly illustrate that approximately at
θ = 0.22 one starts to ‘see’ tracer at the reservoir outlet. In
addition, even for a very long time of operation (about five
times the residence time), the reservoir is not completely full
of tracer, due to the large stagnant zone. The movie of con-
centration evolution is available at the following web address
http://paginas.fe.up.pt/∼mmalves/cfd/reactor/movies.htm.

4. Process simulation

The processes to be discussed below illustrate how science

and technology ingredients contribute to process innovation.
In the area of proteins separation the concept of “diffusiv-
ity augmented by convection” combined with the manufac-
ture of large-pore adsorbents lead to the development of a
process—perfusion chromatography (Afeyan et al., 1990).
Interesting enough the development was made by a spin-off
company of MIT and not by the traditional resin manufac-
turers. Also the development of SMB processes by UOP was
based on the concept of simulating the solid movement by
keeping it fixed combined with a technology development
of rotary valve to allow switching of the position of fluid
streams entering/leaving the system. Again the recent intro-
duction of this technology in pharmaceutical industry was
lead by a small company and not by the traditional owners of
SMB technology.

4.1. Perfusion chromatography

The first chromatographic experiment (1903) was reported

in 1905 by M. Tswett (Fig. 11a) to the Warsaw Society of Nat-
ural Sciences: “On a category of adsorption phenomena and
their application to biochemical analysis” (Rondest, 1972).
He coined the term “chromatography” inspired in the exper-
Fig. 11. Experiments of (a) M. Tswett and (b) D.T. Day.
iment: elution of a sample of green leaves extract through
a column of calcium carbonate which was separated in a 4.1.1. The concept behind perfusion chromatography:
yellow fraction (carotenes) and green fraction (chlorophyll). augmented diffusivity by convection in large-pore
This study was rediscovered in 1931 by the Nobel Prize R. materials
Kuhn working on natural pigments. The theory of adsorp- In chemical engineering, there are materials (catalysts,
tion chromatography was developed in 1940 by Tiselius and adsorbents and membranes) with large pores (>1000 Å) for
partition chromatography in 1941 by Martin and Synge (all
Nobel) (1941). Another vision of history shows David Tal-
bot Day (Heines, 1971), geologist and engineer at the Min-
eral Resources of the US Geological Survey, who presented
at the First International Petroleum Congress in Paris (1900)
one experiment where “crude oil forced upward through a
column packed with limestone changed in color and com-
position” (Fig. 11b). This is the basis of PONA analysis es-
tablished in 1914 and still used in petroleum industry. The
factors influencing the behaviour of a fixed bed column can
be classified in two categories: equilibrium and kinetic factors
(hydrodynamics, heat/mass transfer) (Fig. 12). Fig. 12. Factors governing the behaviour of an adsorptive process.
1174 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

transport and smaller pores to provide adsorption capacity

and catalytic sites. My interest in this area started with a
problem of measurement of effective diffusivity in large-pore
catalysts using a chromatographic method and tracer technol-
ogy. The analysis of results obtained by Ahn (1980) with a
conventional model led to the conclusion that effective diffu-
sivity was changing with flow rate. Results were reanalyzed
by assuming transport not only by diffusion, De but also by
convection (pore velocity v0 ) and the equivalence with the
conventional model where both mechanisms were lumped in
an apparent D̃e allowed us to show that (Rodrigues, Ahn, & Fig. 13. Enhancement factor for diffusivity due to convection, 1/f(λ).
Zoulalian, 1982):
1 is: v0 = au, where a is the ratio of particle and bed permeabil-
D̃e = De (28) ities. Van Deemter, Zuiderweg, & Klinkenberg’s equation
f (λ)
(1956) for conventional packings for the Height Equivalent
where the intraparticle Peclet number is λ = v0 /De = to a Theoretical Plate (HETP) is:
τd /τc .
The apparent diffusivity is augmented by convection and B 2 εp (1 − εb )b2
HETP = A + + τd u (34)
the enhancement factor is 1/f(λ). This result explains the func- u 3 [εb + εp (1 − εb )b]2
tioning of perfusion chromatography developed in 1990 for
the separation of proteins. where εp is the particle porosity,
 εb the interparticle porosity,
Based on the work of Nir and Pismen (1977) on diffusion, b = 1 + (1 − εp )m/εp and m is the slope of the equilibrium
convection and reaction in large-pore catalysts (5000 Å) data isotherm; or
from Ahn (1980) were analyzed. For a non-adsorbable tracer B
the “lumped” diffusion/convection model for transient state HETP = A + + Cu (35)
u
is:
For large-pore particles (Rodrigues, 1993) an extension of
∂2 c ∂c
D̃e = εp (29) the Van Deemter equation was presented:
∂x 2 ∂t
The particle transfer function is: B
HETP = A + + Cf (λ)u (36)
√ u
c̄ tanh τ̃d S
g̃p (s) = = √ (30) Equation (36) shows that at low velocities f(λ) ≈ 1 it is
c̄s τ̃d S
reduced to the classic Van Deemer equation (diffusion-
with an apparent diffusion time constant τ̃d = εp 2 /D̃e . controlled limit); at high velocities f(λ) ≈ 3/λ and the last term
The detailed diffusion/convection model is: of Rodrigues equation (Eq. (36)) becomes constant since the
velocity inside pores v0 is proportional to the bed superficial
∂2 c ∂c ∂c
De − v0 = εp (31) velocity u. The HETP reaches a plateau, which does not de-
∂x 2 ∂x ∂t pend on the solute diffusivity but only on the particle perme-
and ability and pressure gradient (convection-controlled limit).
 In large-pore supports the column performance is improved
λ 2
(e2r2 − 1)(e2r1
− 1) 2 + τd s since HETP is lower than with conventional supports (the C
gp (s) = (32)
(e 2 − e 1 )
2r 2r τd s term of Van Deemter equation is reduced) and the speed of
 separation is increased without loosing efficiency (Fig. 15).
with r1,2 = λ2 ± λ 2
2 + τd s, τd = εp 2 /De and λ =
v0 /De = τd /τc (intraparticle Peclet number).
Model equivalence (Fig. 14) leads to Eq. (28) where the
enhancement factor for pore diffusivity due to convection is
1/f(λ) shown in Fig. 13, with:
3 1 1
f (λ) = − (33)
λ tanh λ λ
The practical application of this concept was developed in
1990 for the separation of proteins (perfusion chromatogra-
phy). The pore velocity can be estimated from the equality
between bed pressure drop relative to the bed length and parti- Fig. 14. Model equivalence: (a) diffusion and convection and (b) “lumped”
cle pressure drop assuming that Darcy’s law is valid; the result diffusion.
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1175

Fig. 15. HETP vs. u (Van Deemter equation and Rodrigues equation).

The objective achieved with the use of large-pore packing

materials as shown in Fig. 16 is to reduce intraparticle mass
transfer resistances. Examples of packing materials are in
Fig. 17.

4.2. Simulated Moving Bed (SMB) processes

Fig. 16. How to decrease intraparticle mass transfer resistance?

4.2.1. The concept of SMB

Simulated Moving Bed is a powerful technique for prepar- biotechnology, pharmaceuticals and fine chemistry; the first
ative scale chromatography known since 1961 (first patent industrial unit was installed at UCB Pharma (Belgium) in
by UOP, Broughton & Gerhold, 1961). This technology 1999 by Novasep. SMB is now considered a key technology
was originally developed in the areas of petroleum refining for chiral separations. It is interesting to compare “old” and
and petrochemicals, known as the Sorbex process. Recently, “new” applications of SMB technology. In the Parex process
SMB technology has found new applications in the areas of (Fig. 18), for p-xylene recovery from a mixture of xylene

Fig. 17. Porous materials: (a) POROS packing, (b) Silica Monolith (Merck) and (c) POROSHELL (Agilent Technologies).
1176 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

Fig. 18. Parex process with rotary valve (AC: adsorbent chamber; RV: rotary valve; EC: extract column; RC: raffinate column; lines: 2—desorbent; 5—extract;
9—feed; 12—raffinate).

isomers column diameter is D = 10 m (maximum) and bed sorbed and carried with the solid phase. Section 2 is the zone
height H = 1 m and particles are of 600 ␮m diameter; in chi- of desorption of the less retained species B, while section
ral separations (Fig. 19), D = 1 m (maximum) and H = 0.1 m 3 is the zone of adsorption of the more retained component
(similar D/H ratio) with particles of 20 ␮m diameter. The A. The role of section 4 is to clean the eluent, which is then
adsorbent capacity is 200 kg/m3 in the Parex process com- recycled to the section 1 where the adsorbent is regenerated.
pared with 10 kg/m3 in chiral separations; the productivity is The operation of this ideal TMB has several drawbacks as-
120 kg/m3 h in the Parex and 1–10 kg/m3 h in chiral separa- sociated with the movement of the solid phase. A Simulated
tions. Moving Bed technique was developed in order to retain the
The principle of SMB operation can be best understood advantages of continuous and countercurrent flow without
by analogy with the equivalent True Moving Bed (TMB) pro- introducing the problems associated with the actual move-
cess. The TMB unit (Fig. 20) is divided into four sections: ment of the solid phase. In the SMB system, the adsorbent is
section 1, between the eluent and extract ports; section 2, fixed and the positions of the inlet and outlet streams move
between the extract and feed ports; section 3, between feed periodically.
and raffinate points; section 4, between the raffinate and the This shift, carried out in the same direction of the liquid
eluent inlet. In the ideal TMB operation, liquid and solid flow phase, simulates the movement of the solid phase in the op-
in opposite directions, and are continuously recycled: the liq- posite direction. The four liquid access lines between each
uid flowing out of section 4 is recycled to section 1, while column can be used to perform a discrete movement of the
the solid coming out of section 1 is recycled to section 4. In inlet and outlet streams in the same direction of the liquid
the TMB operation, the solid flow rate is constant all over phase.
the unit; however, the liquid flow rates differ from section In the Sorbex SMB technology developed by UOP
to section. Let us consider a feed mixture containing species (Fig. 18), a rotary valve is used to periodically change
A, the more retained component and recovered in the extract the position of the eluent, extract, feed and raffinate lines
(Fig. 19a) and species B, the less adsorbed species and pref- along the adsorbent bed. At any particular moment, only
erentially recovered in the raffinate. In sections 2 and 3, the four lines between the rotary valve and the adsorbent bed
two components must move in opposite directions. The less are active. However, there are alternative techniques to per-
retained component B must be desorbed and carried with the form the port switching, like the one developed by NovaSep
liquid phase, while the more retained species A must be ad- (France), which uses a set of individual on-off valves connect-
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1177

Fig. 19. (a) SMB technology for chiral separations and (b) SMB unit Licosep 12-26 (Novasep) at the LSRE.

ing the inlet and outlet streams to each node between columns other hand, the SMB approach examines each column indi-
(Fig. 19). vidually. Due to the switching of inlet and outlet ports, each
column plays a different role during whole cycle, depending
on its location. The model of the SMB unit is constituted of k
4.2.2. Modelling of SMB (k = number of the columns) identical models of the column,
The SMB can be modelled as an equivalent TMB where
the solid movement is taken into account and equivalence re-
lations are used to relate the results with a real SMB (Table 1) Table 1
Equivalence relations between a SMB and a TMB
or as a real SMB; each bed is analyzed individually and the
periodic change in boundary conditions is taken into account. SMB TMB
With reference to Fig. 20, the TMB unit is divided by inlet Solid phase velocity 0 us = Lc /t*
and outlet ports into four zones. The TMB steady-state in- Liquid phase velocity vSMB vTMB = vSMB − us
ternal concentration profiles are presented in Fig. 21. On the t* is the switching time and Lc is the column length.
1178 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

connected with each other by simple material balances on

the connecting nodes. With each switching of the inlet and
outlet ports each column should be updated in terms of flow
rate and inlet concentration. The flow rate in each column,
according to its location (section), can be calculated by mass
balance around the inlet and outlet nodes. The inlet concen-
tration of each column is equal to the outlet concentration of
the previous column, except for the feed and desorbent nodes.
The main difference between TMB and SMB approach is
that the stationary regime of the unit. The time dependence of
the boundary conditions in the SMB leads to cyclic steady-
state instead of a real steady-state present in the TMB model.
The cyclic steady-state is reached after a certain number of
cycles, but the system states are still varying over the time
because of the periodic movement of the inlet and outlet ports
along the columns (Fig. 22).
Whatever modelling approach is considered, the degree
of complexity of the model may vary significantly accord-
ing to the description of the unit operation (staged or dis-
tributed plug flow system), the mass transfer resistance (equi-
librium stage or mass transfer resistance within the fluid
and/or solid phase) and adsorption equilibrium (linear, Lang-
muir, bi-Langmuir or modified Langmuir).
The SMB and TMB model equations when axial dis-
persion plug flow for the liquid phase, plug flow for the
solid phase, homogeneous LDF for internal mass transfer
and any multicomponent adsorption isotherm are assumed is
presented in Table 2.

4.2.3. Design of SMB

Due to the relative complexity of its physical implemen-
tation, the definition of operating conditions (flow rates and
switching time) of a SMB adsorber is not a straightforward
Fig. 20. True Moving Bed (TMB). task. Some constraints have to be met to recover the more
strong adsorbed species (A) in the extract and the less ad-
sorbed species (B) in the raffinate. These constraints are ex-
pressed in terms of net fluxes of components in each section
considering an equivalent TMB. In section 1, the species A
must move upward to the extract port, in sections 2 and 3, the
species A must move downward to the extract port and the
species B must move to the raffinate port and in section 4,
the net flux of the species B has to be downwards (Fig. 20).
Q1 cA1
>1
QS qA1
Q2 cB2 Q2 cA2
>1 and <1
QS QB2 QS qA2
(37)
Q3 cB3 Q3 cA3
>1 and <1
QS QB3 QS qA3
Q4 cB4
<1
QS QB4
The simplest case may be formulated for systems with lin-
Fig. 21. Concentration profiles in TMB (A: more retained species; B: less ear uncoupled adsorption isotherms. The region of complete
retained species). separation (100% extract and raffinate purity) predicted by
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1179

Fig. 22. Cyclic steady-state concentration profiles at the beginning, half and end of the switching time period: (a) species A, (b) species B, (c) extract
concentration history and (d) raffinate concentration history.

the equilibrium theory is a triangle shown in Fig. 23a. The model and explore the influence of the flow rates in zones 1
section constraints are explicit inequality relations between and 4 (desorbent flow rate). The “separation volume” method-
solid and liquid flow rates in the four TMB sections (Ruthven ology offers two possibilities: if the flow rate in zone 1 (m1 )
& Ching, 1989; Storti, Mazzotti, Morbidelli, & Carrà, 1993): is fixed, the design leads to a (m2 × m3 × m4 ) volume for a
given separation requirement; if the flow rate in zone 4 (m4 )
m1 > KA KB < m2 , m3 < KA m 4 < KB (38) is fixed, the design will result in a (m2 × m3 × m1 ) volume
For non-linear isotherms the “triangle” is distorted as shown for a given separation requirement (see Fig. 25). In previous
in Fig. 23b. works (Azevedo & Rodrigues, 1999; Minceva & Rodrigues,
Higher mass transfer resistances lead to a decrease of the 2002; Rodrigues & Pais, 2004) it was found that the flow rate
separation region as shown in Fig. 24 (Pais, 1999). in zone 4 has less influence on the separation performance.
For the systems where mass transfer resistance inside par-
ticles is important, the “triangle theory” can only give initial 4.2.4. Optimization of SMB
guesses for a feasible operating point of the process, since it is The optimization of SMB unit considers selection of the
based on the assumption of the equilibrium model. Very often, operating conditions and/or geometric parameters that min-
100% purity in both extract and raffinate is either unnecessary imize/maximize a given objective function(s), in the frame
or would require an extremely large adsorbent inventory to be of previously defined constraints. Most commonly the ob-
accomplished. Furthermore, the approach based on equilib- jective is to maximize the SMB productivity and mini-
rium model does not allow explicit prediction of the product mize the eluent (desorbent) consumption, for a given con-
purities that are generally the main constraints for a feasible straints defined in terms of minimum required product purity
operating point. and/or recovery. Maximum productivity implies maximum
These are the reasons why the concept of separation vol- feed flow rate, max(QF ) = max(γ 3 − γ 2 ) and minimum elu-
ume was introduced (Azevedo & Rodrigues, 1999). The “sep- ent (desorbent) consumption implies minimum eluent flow
aration volume” methodology uses realistic mathematical rate min(QE ) = min(γ 1 − γ 4 ), where γj = (1−ε)
ε mj , j = 1, 2,
1180 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

Table 2
Transient SMB and TMB model equations
Simulated Moving Bed model equations
Mass balance over a volume element of the bed k
 
∂Cik 1 ∂2 Cik ∂Cik (1 − ε)
= γk∗ − = ∗
αk (qik − qik )
∂θ Pek ∂x 2 ∂x ε

Mass balance in the particle

∂qik ∗
= αk (qik − qik )
∂θ

Initial conditions
θ = 0 : Cik = qik = 0

Boundary conditions for column k

1 dCik
x = 0 : Cik = = Cik,0
Pek dx
where Cik,0 is the inlet concentration of species i in column k
x = 1:
For a column inside a section and for extract and raffinate nodes:
Cik = Cik + 1,0
ν∗
For the eluent node: Cik = 1∗ Cik+1,0
ν4
ν∗ νF
For the feed node: Cik = 3∗ Cik+1,0 − ∗ CiF
ν2 ν2

Global balances
Eluent node : ν1∗ = ν4∗ + νE ; extract node : ν2∗ = ν1∗ − νX
Feed node : ν3∗ = ν2∗ + νF ; raffinate node : ν4∗ = ν3∗ − νR

Multicomponent adsorption equilibrium isotherm

∗ = f (C , C )
qAk A Ak Bk
∗ = f (C , C )
qBk B Ak Bk

True Moving Bed model equations

Mass balanceover a volume element
 of the section j
∂Cij 1 ∂2 Cij ∂Cij (1 − ε)
= γj − = αj (qij∗ − qij )
∂θ Pej ∂x2 ∂x ε

Mass balance in the particle

∂qij ∂qij
= + αj (qij∗ − qij )
∂θ ∂x

Initial conditions
θ = 0 : Cij = qij = 0 Fig. 23. Separation region for: (a) linear isotherms and (b) non-linear
isotherm.
Boundary conditions for section j
1 dCij
x = 0 : Cij − = Cij,0
Pej dx
where Cij,0 is the inlet concentration of species i in section j 3, 4. The influence of γ 1 and γ 4 on the SMB productivity and
x = 1: eluent consumption is presented in Fig. 26. The separation
ν1
For the eluent node: Ci4 = Ci1,0 ; for the extract node: region increases by increasing γ 1 up to some value and the
ν4
Ci1 = Ci2,0
ν3 νF
vertex moves from lower to higher values of (γ 2 , γ 3 ), higher
For the feed node: Ci2 = Ci3,0 − CiF ; for the raffinate productivity could be reached. When the separation region
ν2 ν2
node: Ci3 = Ci4,0 size does not increase more, further increment in γ 1 will lead
and qi4 = qi1,0 , qi1 = qi2,0 , qi2 = qi3,0 , qi3 = qi4,0 just to increase of the eluent consumption, since (γ 1 − γ 4 )
increases. When the value of γ 1 is fixed (Fig. 26b) the sep-
Global balances
Eluent node: ν1 = ν4 + νE ; extract node: ν2 = ν1 −νX aration regions are similar until some value of γ 4 and then
Feed node: ν3 = ν2 + νF ; raffinate node: ν4 = ν3 −νR they start to decrease if the value of γ 4 further increases. The
value of the flow rate in zone 4 (γ 4 ) does not influence the
Multicomponent adsorption equilibrium isotherm feed flow rate and SMB unit productivity, but it affects the
∗ = f (C , C )
qAj A Aj Bj
∗ = f (C , C ) desorbent consumption. We proposed a two-level optimiza-
qBj B Aj Bj
tion procedure based on the concept of “separation volume”
and equivalent TMB design (Minceva, 2004).
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1181

Fig. 24. Influence of mass transfer resistance on the separation region for
non-linear isotherms.

4.3. Future directions in SMB: multicomponent, SMBR

Although the SMB technology offers many advantages

over preparative chromatography (leading to cleaner, smaller,
safer and faster processes) (Nicoud, 1997) the main disad-
vantage of this process is the limitation to the separation of
binary mixtures or of one component from a multicomponent
mixture.
Fig. 26. Influence of flow rates in: (a) section 1 and (b) section 4 on SMB
The pseudo-Simulated Moving Bed process—JO process productivity and eluent consumption.
of Japan Organo Co. (Ando, Tanimura, & Tamura, 1990;
Masuda, Sonobe, Matsuda, & Horie, 1993) has been suc-
cessfully applied in separation of a ternary mixture. The pro- species is collected in the raffinate while the more retained
cess cycle is divided into two steps (Fig. 27). In step 1, feed species is collected in the extract (Mata & Rodrigues, 2001).
and eluent streams are introduced into the system, equivalent The combination of the chemical or biochemical reaction
to a series of preparative chromatographic columns, and the with Simulated Moving Bed chromatographic separator has
intermediate component is produced. In step 2, similar to a been subject of considerable attention in the scientific re-
Simulated Moving Bed, there is no feed and the less adsorbed search the last 10 years. This integrated reaction–separation
technology adopts the name Simulated Moving Bed Reactor
(SMBR) technology. The first application of SMBR in zeo-
lite catalyzed alkylation reaction was patented by Zabransky
and Anderson (1977). Integration of reaction and separation
steps in one single unit has the obvious economical advantage
of reducing the cost of unit operations for downstream pu-
rification steps. In the case of reversible reaction, where the
conversion is limited by the chemical equilibrium removal
of products as they are formed allows achieving conversions
well beyond equilibrium values. For reactions in series or in
parallel, it may be possible the selective separation of desired
intermediate species. When a reaction product has an inhibit-
ing or poisoning effect, its removal from the reaction medium
also promotes enhanced yield.
The inversion of sucrose (A → B + C) was studied in a
SMBR (Azevedo & Rodrigues, 2001). Sucrose is introduced
in the middle of the unit with the feed stream. The reaction
is catalyzed by the enzyme invertase introduced in the unit
Fig. 25. Concept of separation volume. with the eluent stream. The sucrose reacts near to the feed
1182 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183

Fig. 27. Comparison between the two different techniques: (a) TMB and (b) pseudo-SMB JO process.

Simulation?); in the meantime we keep the reflection of As-

tarita: “the amount of information available grows continu-
ously but the amount of information that any one of us can
usefully digest does not grow”.

References

Afeyan, N., Fulton, S., Gordon, N., Mazsaroff, I., Varady, L., & Reg-
nier, F. (1990). Perfusion chromatography—An approach to purifying
Fig. 28. SMBR cyclic steady-state internal concentration profiles (Azevedo biomolecules. Bio-technology, 8, 203.
& Rodrigues, 2001). Ahn, B. (1980). Études des caracteristiques diffusionelles de transfer de
matière dans un reacteur catalytique à lit fixe d’oxidation menagée.
Ph.D. Thesis. Université de Technologie de Compiègne.
port; fructose and glucose are formed and separated in the Ando, M., Tanimura M., & Tamura, M. (1990). Method of chromato-
extract and raffinate, respectively. The internal concentration graphic separation. U.S. Patent No. 4,970,002.
Aris, R. (1994). Mathematical modelling techniques. New York: Dover
profiles are presented in Fig. 28. Pubs.
Asynchronous shifting of the inlet/outlet ports has been Azevedo, D. C. S., & Rodrigues, A. E. (1999). Design of a simulated
proposed in the Varicol process (Ludemann-Hombourger, moving bed in the presence of mass-transfer resistances. AIChE Jour-
Nicoud, & Bailly, 2000) as well as manipulation of feed con- nal, 45, 956.
centration and flow rate. Azevedo, D. C. S., & Rodrigues, A. E. (2001). Design methodology
and operation of a simulated moving bed reactor for the inversion of
sucrose and glucose–fructose. Chemical Engineering Journal, 82, 95.
Bird, B., Stewart, W., & Lightfoot, E. (2002). Transport phenomena. New
5. Conclusions York: John Wiley.
Broughton, D. B., & Gerhold, C. G. (1961). Design of a simulated moving
bed in the presence of mass-transfer resistances. U.S. Patent No.
The area of Modelling/Simulation characterizes a gen- 2,985,589.
eration of chemical engineers associated with the Second Brunier, E., Zoulalian, A., Antonini, G., & Rodrigues, A. (1984). Resi-
Paradigm of Chemical Engineering. Today, simulation tasks dence time distribution in laminar flow through reservoirs from mo-
can be simplified with the availability of friendly user soft- mentum and mass transport equations. ISCRE 8, EFCE publications
ware; the question of validation of results remains and in series no. 37. The Institution of Chemical Engineers, p. 439.
Damkholer, G. (1943). Ubertemperatur in Kontaktkornern. Zeitschrift fur
principle more time will be available to analyze results. Physikalische Chemie, 193, 16.
The Third Paradigm of Chemical Engineer should Danckwerts, P. V. (1953). Continuous flow systems—Distribution of res-
come out soon (is it Molecular Transformations/Modelling/ idence times. Chemical Engineering Science, 1, 1.
 A.E. Rodrigues, M. Minceva / Computers and Chemical Engineering 29 (2005) 1167–1183 1183

Denbigh, K. (1951). The thermodynamics of the steady state. London: Pais, L. (1999). Chiral separations by SMB chromatography. Ph.D. The-
Methuen. sis. FEUP.
De Vault, D. (1943). The theory of chromatography. Journal of American Rondest, J. (1972). Michel Tswett, botanist who was providence of
Chemical Society, 65, 532. chemists. La Recherche, 24, 580.
Glueckauf, E. (1955). Theory of chromatography. Part 10—Formulae Rodrigues, A. E., & Tondeur, D. (Eds.). (1981). Percolation processes:
for diffusion into spheres and their applicability to chromatography. Theory and applications. NATO ASI series E33. Dordrecht: Kluwer
Transactions of Faraday Society, 51, 1540. Academic Publication.
Heines, S. (1971). Chromatography—A history of parallel development. Rodrigues, A. E., Ahn, B., & Zoulalian, A. (1982). Intraparticle forced
Chemtech, 1, 280. convection effect in catalyst diffusivity measurements and reactor de-
Kramers, H. (1958). Physische Transportverschijnselen. Delft: TU Delft. sign. AIChE Journal, 28, 925.
Le Goff, P. (1970). Cours de Cinétique Physique et Chimique. ENSIC. Rodrigues, A. E. (1993). An extended Van Deemter equation (Rodrigues
Levenspiel, O. (2002). Modeling in chemical engineering. Chemical En- equation) for performing chromatographic processes using large-pore,
gineering Science, 57, 4691. permeable packings. LC–GC, 6, 20.
Liu, J., Chen, X., & Xu, L. (1999). New thermal wave aspects on burn Rodrigues, A. E., & Pais, L. S. (2004). Design of SMB chiral separations
evaluation on skin subjected to instantaneous heating. IEEE Transac- units using concept of separation volume. Separation Science and
tions on Biomedical Engineering, 46, 420. Technology, 39, 245.
Ludemann-Hombourger, O., Nicoud, R., & Bailly, M. (2000). Design of Ruthven, D. M., & Ching, C. B. (1989). Counter-current and simulated
a simulated moving bed: Optimal particle size of the stationary phase. counter-current adsorption separation processes. Chemical Engineer-
Separation Science and Technology, 35, 1829. ing Science, 44, 1011.
Madeira, L. M., Alves, M., & Rodrigues, A. E. (2004). Teaching non- Shinnar, R. (1978). Chemical reactor modeling—The desirable and the
ideal reactors with CFD tools. Chemical Engineering Education, 38, achievable. In D. Luss & V. Weekman Jr. (Eds.), Chemical reaction
245. engineering reviews, ACS symposium series: Vol. 72 (p. 1).
Martin, A. J. P., & Synge, A. L. M. (1941). Modeling in chemical engi- Storti, G., Mazzotti, M., Morbidelli, M., & Carrà, S. (1993). Robust de-
neering. Biochemical Journal, 35, 1358. sign of binary countercurrent adsorption separation processes. AIChE
Masuda, T., Sonobe, T., Matsuda F., & Horie, M. (1993). Process for Journal, 39, 471.
applications of preparative chromatography. Patent No. 5,198,120. Van Deemter, J., Zuiderweg, F., & Klinkenberg, A. (1956). Counter-
Mata, V. G., & Rodrigues, A. E. (2001). Separation of ternary mixtures current and simulated counter-current adsorption separation processes.
by pseudo-simulated moving bed chromatography. Journal of Chro- Chemical Engineering Science, 5, 271.
matography A, 939, 23. Vermeulen, T. (1958). Separation by adsorption methods. Advanced
Maxwell, J. C. (1952). The scientific papers of James Clerk Maxwell. Chemical Engineering, 2, 147.
New York: Dover. Weisz, P., & Prater, D. (1954). Interpretation of measurements in exper-
Minceva, M., & Rodrigues, A. E. (2002). Modeling and simulation of a imental catalysis. Advance in catalysis: Vol. 6. Academic Press.
simulated moving bed for the separation of p-xylene. Industrial and Westerterp, K., Kronberg, A., Benneker, A., & Dil’man, V. (1996). Wave
Engineering Chemistry Research, 41, 3454. Concept in the theory of hydrodimanic dispersion—A maxwellian
Minceva, M. (2004). Separation/isomerization of xylenes by simulated type approach. Trans. Instn. Chem. Engrs. Part A, 74, 944.
moving bed technology. Ph.D. Thesis. FEUP. Wheeler, A. (1951). Reaction rates and selectivity in catalyst pores. Ad-
Nicoud, R.-M. (1997). In R.-M. Nicoud (Ed.), Recent advances in indus- vances in Catalysis, 3, 249.
trial chromatographic processes (p. 4). France: NOVASEP, Nancy. Zabransky, R. F., & Anderson, R. F. (1977). Simulated moving bed alky-
Nir, A., & Pismen, L. (1977). Simultaneous intraparticle forced convec- lation process. U.S. Patent No. 4,049,739.
tion, diffusion and reaction in a porous catalyst. Chemical Engineering Zwietering, T. (1959). The degree of mixing in continuous flow systems.
Science, 32, 35. Chemical Engineering Science, 11, 1.