Chemical Engineering Journal 378 (2019) 121985

Contents lists available at ScienceDirect

Chemical Engineering Journal

journal homepage: www.elsevier.com/locate/cej

Modeling of fixed-bed columns for gas physical adsorption T

a,⁎ a b
Francisco Javier Gutiérrez Ortiz , Manuel Barragán Rodríguez , Ralph T. Yang
a
University of Seville, Escuela Técnica Superior de Ingeniería, Department of Chemical and Environmental Engineering, Camino de los Descubrimientos s/n, 41092 Sevilla,
Spain
b
University of Michigan, College of Engineering, Department of Chemical Engineering, 3074 H.H. Dow, 2300 Hayward Street, Ann Arbor, MI 48109-2136, USA

HIGHLIGHTS

• Modeling of the fixed-bed physical adsorption was simulated by Comsol.

• The modeling does not require one to obtain key parameters from the experiment and only a few data are needed.
• Boundary conditions, thermal effects and the homogeneous/heterogeneous modelings are discussed.
• The effects of mass transfer resistance, axial mass and heat dispersion on the breakthrough curves were simulated.
• The modeling was tested against five sets of experimental data, and good agreement was found.

ARTICLE INFO ABSTRACT

Keywords: Adsorption processes may be classified as purification or bulk separation, depending on the feed concentration of
Breakthrough curve the compounds to be adsorbed (adsorbates). The adsorption dynamics of a fixed-bed is crucial for a well-de-
Fixed-bed column signed adsorption process. This work focuses on physical adsorption and provides a modeling to predict the
Adsorption breakthrough curves using a minimum set of experimental data. The modeling does not require one to obtain key
Gas separation
parameters from the experiment unlike other modeling found in the literature. In addition, boundary conditions,
Modeling
Simulation
thermal effects, correlations to be used, and the homogeneous/heterogeneous modelings are discussed for 1-D
modeling, after verifying that it provides the same output as 2-D modeling does. Moreover, it is demonstrated
that pseudo-homogeneous modeling is realistic, so a more complex heterogeneous modeling is not necessary.
The modeling has been tested against five sets of experimental data: three cases of bulk separation and two cases
for purification. The simulation was carried out by Comsol Multiphysics software, and a good match between the
experimental data and the simulation output was achieved, which demonstrates the applicability of the mod-
eling, so it may be used with confidence. Purification can be modeled as an isothermal process, and no energy
balance equation is needed. However, for bulk separation, noticeable thermal effects may take place due to the
relatively high adsorbed gas that consequently releases a higher amount of heat; at the same time, the superficial
fluid velocity is reduced due to the decrease in the flow-rate, and the gas properties change, affecting the
breakthrough curves.

1. Introduction radial dispersion might be also important, depending on the extent of

In a real adsorption fixed-bed, the outlet concentration and tem-
perature profiles do not replicate the input because of the dispersion
and mass transfer resistance that needs to be considered in addition to
adsorption equilibrium. The latter one involves the time delay kinetics
due to the hold-up in the column, while the former one is related to
hydrodynamics, especially regarding the extent of mixing of the fluid
elements occurring due to turbulence as well as molecular diffusion and
conduction. Usually, axial dispersion is considered, but sometimes
adsorption and the heat released in the medium and exchanged with
the surroundings. In relation to the former, adsorption processes may be
classified as purification or bulk separation, depending on the feed
concentration of the components to be adsorbed. So, the radial effects
should not be disregarded a priori for bulk separation processes, while
the radial dispersion is almost negligible for purification processes.
Likewise, the separation behavior changes in an adsorption bed from
isothermal to adiabatic systems, and of course for non-adiabatic and
non-isothermal fixed-beds. In addition, the overall dynamics is affected

⁎
Corresponding author.
E-mail address: frajagutor@us.es (F.J. Gutiérrez Ortiz).

https://doi.org/10.1016/j.cej.2019.121985
Received 1 May 2019; Received in revised form 13 June 2019; Accepted 15 June 2019
Available online 17 June 2019
1385-8947/ © 2019 Elsevier B.V. All rights reserved.
F.J. Gutiérrez Ortiz, et al.
Nomenclature
as, esp particle specific superficial area (m−1)
C adsorbate concentration in gas phase (mol m−3)
C0 adsorbate feed concentration (mol m−3)
dp mean particle diameter (m)
D column internal diameter (m)
De effective diffusivity (m2 s−1)
Dm molecular diffusivity (m2 s−1)
Dz axial dispersion coefficient (m2 s−1)
Hads heat of adsorption (J mol−1)
F fluid flow-rate (m3 s−1)
h heat transfer coefficient (W m−2 K−1)
kext external mass transfer coefficient (m s−1)
k conductivity (W m−1 K−1)
KG overall mass transfer coefficient (s−1)
KL equilibrium constant of Langmuir (m3 mol−1)
L total bed length (m)
M molecular weight (mol g−1)
Nu Nusselt number (−)
P pressure
Pe Peclet number (−)
Pr Prandtl number (−)
q adsorbate concentration (or adsorbate loading) in solid
phase (mol kg−1)
by the presence of other solutes that may be absorbable, i.e., multi-
component systems.
For a well-designed system, a modeling to predict breakthrough
curves from basic kinetic and equilibrium data is very important, thus
reducing the number of experimental tests, which may be very labor-
ious and time-consuming. Thus, a comprehensive adsorption modeling
for both purification and bulk separation processes would be very
useful for design engineers.
Process simulation is a tool more and more used for process design
because it reduces the cost relative to tests carried out in the lab and
pilot plant before scaling up. Although many phenomena may occur
simultaneously in an adsorption process, there is a challenge to achieve
a modeling and simulation methodology that allows a reasonable pre-
diction of the system behavior even if a number of simplifications are
used.
In a previous paper, a specific adsorption model for hydrogen sul-
fide removal from landfill biogas on activated carbon from treated
sewage-sludge was developed, as a purification case-study [1]. Simu-
lation results were compared with experimental data obtained in an
experimental fixed bed, showing good agreement between them. In the
present work, the model is extended to be applied to both purification
and bulk separation adsorption processes. A discussion about the use of
one-dimensional (1-D) and two-dimensional (2-D) pseudo-homo-
geneous and heterogeneous modelings, as well as thermal effects
caused by the heat of adsorption, and the right selection of boundary
conditions and correlations are carried out. Thus, by solving a set of
partial differential equations (PDE) using Comsol Multiphysics soft-
ware, the breakthrough curves, temperature and concentration profiles
for gas and solid can be obtained.
Comsol is focused on the numerical solution of systems of partial
differential equations (PDE) by the Finite Element Method. This soft-
ware has a series of modules that can be coupled to model different
processes involving different branches of physics and chemistry.
However, after testing all of them, the module termed Coefficient Form
2
Chemical Engineering Journal 378 (2019) 121985
q* value of q in equilibrium with C (mol kg−1)
qm saturation value of q in the Langmuir isotherm (mol kg−1)
Ra Rayleigh number (−)
Re Reynolds number (−)
Sc Schmidt number (−)
Sh Sherwood number (−)
t time (s)
T temperature (K)
u interstitial velocity (m s−1)
w mass fraction
y molar fraction
z distance from the bed entrance (m)
Greek
Ε. bed void fraction (−)
ρb adsorbent bulk density (kg m−3)
μ dynamic viscosity (kg m−1 s−1)
Subscripts
f fluid
s solid
w column wall
out external/outside
PDE, which is a mathematical module without any specific physical
function that allows the user to enter a set of PDE, is the most appro-
priate to code the modeling in Comsol in a dimensional form because it
requires less computing resources; this is another novelty with respect
to the previous model.
When the equations vary with time (transitory regime), which is a
characteristic of any adsorption process, the solution method is divided
into four steps:
(1st) Spatial discretization by the Finite Element Method. After this
step, the time derivatives are the only ones present in the equations.
(2nd) Time-iterated integration using only the BDF (Backward
Differentiation Formulas) Method, whereby the differential equations
are solved at each time interval with data coming from the previous
time interval. After this step, a system of nonlinear algebraic equations
remains to be solved.
(3rd) Iterative linearization of the system of nonlinear equations by
the Damped Newton Method, which uses a damping factor to ensure
convergence. The smaller the factor, the slower the algorithm is, but the
convergence is better addressed. The result of applying this method to
nonlinear equations is a system of linear equations.
(4th) Direct or iterative solution with algorithms to solve the matrix
equations of the system of linear equations, which gives the final si-
mulation outputs.
2. Aims and scope
The main aim of the paper is to provide a predictive modeling of
fixed-bed adsorber dynamics both for purification and for bulk se-
paration processes based on physical adsorption applied to gas se-
paration process with favorable Type I isotherms, including multi-
component systems. The main novelty is the applicability of the
modeling, which needs only a few number of parameters and does not
require information from experimental breakthrough data to obtain
preceding mass transfer or axial dispersion coefficients as usually done
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

by other researchers. other hand, there may be a radial gradient of velocity or void fraction,
The paper describes and discusses the equations and boundary so both concentrations and temperature might change radially.
conditions, and how to deal with them in order to reach a suitable Moreover, a high thermal gradient due to cooling or heating through
numerical solution, using correlations and a minimum number of ex- the adsorber wall can lead to a radial concentration distribution. Thus,
perimental data, mainly related to the equilibrium and sorbent char- a 2-D heterogeneous axially dispersed plug flow model should be con-
acteristics. In order to avoid lengthening the document excessively, the sidered. Generally, it is accepted that if the diameter of the adsorber-to-
paper is focused on physical adsorption, so chemisorption is out of the particle sorbent diameter is greater than 10, the radial concentration
scope of the study, which will be developed in a future communication. gradient is negligible. Furthermore, in physical adsorption processes,
As commented on below, two-dimensional (2-D) axially dispersed plug the heat source is not high enough to involve radial variations. A
flow modeling will be further developed in a future work because no number of simulations have allowed verifying this for two case-studies,
noticeable difference between 1-D and 2-D modeling was verified. and therefore, no radial variations in velocity, concentration and tem-
The model predictions are compared with the results obtained from perature are considered. Thus, 2-D modeling is to be only taken into
five experimental studies published in the literature regarding both account in chemical adsorption, in another paper, as above mentioned.
purification and bulk separation processes in order to test the predictive The pressure drop along the bed may be calculated by Eq. (1):
power of the proposed model using minimum data, and to demonstrate 2u
us us P
its applicability.
s
+ us + = µf us
f
t f
z z z2 (1)
3. Mathematical modeling where represents an inertia term given by Ergun correlation (Eq. (2)):
(1 )2 (1 )
The mathematical modeling used to describe the dynamics of ad- = 150µf 3d 2
+ 1.75 f 3d
|us|
sorption considers a nonlinear adsorption isotherm with a plug flow p p (2)
contact model. It also includes transport phenomena due to dispersion For purification processes, the pressure drop is negligible because
and mass transfer resistance regarding the adsorption rate on the par- the fluid velocity does not change. For bulk separation, since these
ticles, as well as because of thermal effects relative to the heat exchange systems usually operate at a constant fixed-bed pressure by using a
throughout the bed. The overall mass transfer rate is represented by the pressure controller and a control valve set downstream from the bed,
Linear Driving Force (LDF) approximation, which considers the ad- Eq. (1) is not strictly necessary to obtain the breakthrough curves, and a
sorbent particles as a homogeneous phase in which the diffusion occurs constant pressure value can be considered.
with a constant effective diffusivity. For physical adsorption, other
surface phenomena are assumed to be fast enough, so the adsorption
3.2. Balance equations
rate is proportional to the deviation of the equilibrium through an
overall mass transfer coefficient. According to the LDF model, the
Mass and energy balance equations along with a kinetic rate ex-
overall mass transfer coefficient (KG) determines the rate at which the
pression for the adsorption are the set of equations to be considered in
adsorption occurs, so a high value of KG increases the rate of adsorp-
the modeling.
tion, due to the decrease in the mass transfer resistance; the system
Since the radial gradient of concentration may be neglected (1-D
performance improves, and the slope of the breakthrough curve rises.
modeling), the differential mass balance for the i-th adsorbate (for a
However, the fact that many simulated breakthrough curves are steeper
multicomponent system) in the fixed-bed is Eq. (3)
than the real ones may be because of the overall mass transfer coeffi-
cient is underestimated, so axial dispersion must be included to achieve Ci 2C 1 qi
+ (uCi ) = Dz, i 2i
a simulation to be closer to reality. The dispersion involves a back t z z s
t (3)
mixing for the adsorbate that enlarges the width of the mass transfer
where
zone in the bed. Thermal effects may be also important on the ad-
sorption performance. In the next sub-section, a set of equations and qi
= K G, i (qi* qi )
boundary conditions are described and discussed; variables, parameters t (4)
and properties symbols are detailed in the Nomenclature section.
The equilibrium adsorbed amount qi* in Eq. (4) can be calculated
from the equilibrium isotherm, which is associated with the extended
3.1. Assumptions
Langmuir isotherm (Eq. (5)) as a representative isotherm for Type I
isotherms:
In a previous study [1], isothermal conditions, negligible radial
dispersion and constant fluid velocity were assumed. In the present qm, i KL, i Ci
qi* =
paper, these assumptions are revisited, but the sorbent particles are 1+ KL, i Ci (5)
assumed to be spherical and homogeneous in size and density.
Isothermal conditions might be valid for a purification process, If the temperature changes, and the equilibrium constant (KL ) is
because the adsorbate amount on the solid surface is small and the heat known at a temperature (e.g., 298 K), the KL value can be obtained at
released in the adsorption is low, so the changes in temperature may be any other temperature by the van't Hoff equation (Eq. (6)), using the
neglected. However, this assumption is no longer valid for bulk se- isosteric heat ( Hads ), which may be assumed to be invariant with the
paration processes a priori, due to the higher adsorbate concentration surface loading:
in the feed. Therefore, a non-isothermal process or an adiabatic system dLnKL Hads
(if a good thermal isolation is used) must be considered. If the tem- =
dTf RT f2 (6)
perature changes, the equilibrium isotherm and the breakthrough curve
will change. If there are compounds that are not adsorbed, i.e., inert compounds,
The axial flow velocity will also change over time due to the loss of another differential mass balance is required for these latter or for the
adsorbate in a bulk separation process, but not for purification pro- total gas (overall mass balance) in the fixed-bed, given by Eqs. (7a) and
cesses, due to the very low feed concentration of adsorbate. On the (7b), respectively:

3
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

Cinert 2C
inert us modeling can be used with confidence. Similarly, several models in-
+ (uCinert ) = Dz, inert ;u=
t z z2 (7a) clude the contribution of the heat transfer through the wall (Eq. (10)).
In this work, the wall contribution is also checked, concluding that it
Ct 2C 1 qi u
+ (uCt ) = Dz, t 2t s ;u= s may be disregarded, but distinguishing among isothermal, adiabatic
t z z i
t (7b) and real (non-isothermal and non-adiabatic) systems. Thus, the heat
The energy balance may be solved for the two phases if a hetero- transfer and the existence of thermal isolation or external cooling of the
geneous model is used (Eqs. (8a) and (8b)), i.e., there is an equation for adsorber is very important to achieve a realistic simulation output.
the solid phase (subscript 's') and another one for the fluid phase
(subscript 'f'). Similarly, if a pseudo-homogeneous model is applied, 3.3. Boundary conditions
there is only one equation for the gas phase because the temperature is
the same in both phases (Eq. (9)). Usually, boundary conditions used to solve the adsorption equations
(second order PDE) are Dirichlet-type or Neumann-type for purification
Ts ks, e 2T qi
s Cp, s = s
+ ( Hads, i ) hsf as, esp (Ts Tf ) processes, as exposed in a previous work [1], although Robin-type
t 1 z2 t s
(8a) should be also considered as a flux-source type condition, as shown
below.
( f Cp, f Tf ) + (u f Cp, f Tf )
t z Initial condition (t = 0):
kf ,e 2T
f 1 h us = 0; (0 z L)
= + hsf as, esp (Ts Tf ) 4 in (Tf Tw )
z2 Din (8b) Ci = 0; qi = 0 (0 z L)
where Tf = Ts = Tinitial (0 z L)

1 =
wi
; Cp, f = Cp, i wi For inert compounds or any other at first present in the bed,
f
i Ci = Pi /(RTo ). Initially, there is thermal equilibrium between the bed
and the surroundings, and the total molar fraction must be 1.0; pressure
Tf (uTf )
( f Cp, f + (1 ) s Cp, s ) + f Cp, f is assumed as constant. If the adsorbent is not fresh but regenerated
t z from a previous adsorption cycle, the initial solute charge is not zero,
2T
f q hin but a residual one (qi = qres ); equally, if some compound which can be
= kf ,e + (1 ) ( Hads, i ) i 4 (Tf Tw )
z2 t s
Din (9) adsorbed is initially present in the bed, qi = qi*.
Boundary conditions (z = 0 and z = L)
It is assumed that the heat of adsorption is generated completely in
the solid phase, so there is a heat transfer from solid to the fluid phase. At z = 0:
For the solid phase, the terms are the accumulation of heat in the solid, us = us,0 ; (t > 0) Dirichlet type
the axial thermal conduction, the heat generated by adsorption and the Ci = Ci,0 =
Pyi,0
(t > 0) Dirichlet type
heat transfer to the fluid. For the fluid phase, the terms are the heat RTf ,0

accumulation in the gas, the convective term due to fluid flow, the axial
thermal conduction and the heat transfer from the solid, as well as the
heat transfer to the environment through the adsorber wall that takes
place only from the fluid phase. This latter term would be null in an
adiabatic system.
The axial thermal conduction, which may be considered as playing a
similar role to the axial dispersion in the mass balance, is justified due
to the existence of a number of heat transfer mechanisms through both
solid particles and fluid, as well as from particles towards particles and
fluid, involving conduction, convection and even radiation; some of
them are present in a quiescent fixed-bed and others depend on the
velocity fluid. Thus, normally heat transfer is expressed by a lumped
parameter (effective axial thermal conductivity). In pseudo-homo-
geneous models, no distinction between solid and fluid is considered,
and the heat is transferred axially only through the fluid, but hetero-
geneous models require two lumped parameters affecting the heat
transfer, one for the solid and the other for the fluid.
With respect to the energy balance in the column wall, Eq. (10) is
applied, taking into account the relatively high value of the thermal
conductivity of steel.
Tw
w Cp, w Aw = Din hin (Tf Tw )
t (10)
From simulation results, the heat accumulation in the wall is neg-
ligible under non-adiabatic conditions.
Usually, in the models found in the literature, it is assumed that the
gas phase and the solid phase reach thermal equilibrium in-
stantaneously, but no reason is given for it. In this work, this assump-
tion is tested, verifying that it takes place, so 1-D pseudo-homogeneous
Ts = Ts,0 = T0 (t > 0) Dirichlet type
Tf = Tf ,0 = T0 (t > 0) Dirichlet type
At z = L:
Ci
= 0 (t > 0) Neumann type
z
Tf Ts
z
= z
= 0 (t > 0) Neumann type
At the outlet, there is no change in concentrations as there are no
adsorbent particles and, hence, no adsorption occurs. The same is ap-
plied to the temperatures of solid and fluid.
If there is axial dispersion, the boundary conditions for Ci need to be
revised, because conservation of mass requires Robin-type conditions at
the ends. However, it leads to an underdetermined system due to an
unknown exit concentration or temperature. This issue is overcome by
assuming a continuous concentration at the outflow boundary, and a
well-posed problem is achieved by forcing a homogeneous Neumann
exit condition. In principle, this boundary condition would be valid if
there were no mass (or thermal) dispersion outside the adsorber.
Danckwerts proposed such a boundary condition at the outlet, but he
did not provide a proof of it [2]. However, considering as continuous
both the fluid-phase concentration and the flux at the outlet, Coppola
Dout hout (Tw To) and Devan [3] showed that the Danckwerts condition is appropriate for
the outlet of a fixed-bed adsorber with axial dispersion, even if the
dispersion coefficient downstream from the fixed-bed is not zero, thus
relaxing the previous constraint.
Regarding the boundary condition at the entry of the adsorber
(z = 0), an alternative Robin type boundary condition may be used
instead of the Dirichlet boundary condition. This other boundary con-
dition is called Danckwerts entry condition, which can be written as

4
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

follows: Tf
kf ,e = f Cp, f (Tf ,0 Tf )(t > 0)
z
Ci
Dz , i = u0 Ci,0 uCi (t > 0)
z

where Ci,0 is the concentration just before entering the adsorber and Ci
is the concentration just after entering the adsorber. In this way, the 3.4. Correlations to be used along with the balance equations
continuity of flux rather than that of concentration at the inlet
boundary would be claimed, because owing to diffusion the con- The use of correlations is key to solve the equations and to predict
centration just within the entrance of the reactor is less than the con- the dynamics of a fixed-bed adsorber in a more general way. However,
centration of reactant in the entering stream. In all the simulations there are a lot of correlations that might be applied. In this way, few
performed, Danckwerts boundary conditions work better than Dirichlet correlations have been selected to range all the operating conditions
conditions, especially for bulk separation. expressed by dimensionless numbers such as Peclet, Reynolds, Schmidt,
Danckwerts entry condition (z = 0) should be also applied to the Nusselt or Rayleigh.
temperature:
dp u dp u µ udp
Pez = ; Re = ; Sc = ; Pem =
Ts Dz µ Dm Dm
ks, e = s Cp, s (Ts,0 Ts )(t > 0)
z

Table 1
Correlations for the estimation of the model parameters.
Axial dispersion coefficient (Dz)
Wakao-Funazkri correlation, valid for 3 < Re < 104 [4]
Dz =
Dm
(20 + 1
2
ScRe )
4
Rastegar-Gu correlation for 0.001 < Re < 10 [5] 1 0 . 7Dm dp
= +
Pez Lu L 0.18 + 0.008Re 0.59
Wen-Fan correlation for 0.008 < Re < 400 and 0.28 < Sc < 2.2 [6]
Dm 1 ScRe
Dz = 0.3 +
2 1 + 3.8
ScRe

External mass transfer coefficient (kext)

Ranz-Marshall correlation modified by Wakao and Funazkri for 3 < Re < 104 [7,4] Sh = 2 + 1. 1Sc1/3Re 0.6
Dwivedi-Upadhyay correlation for 0.25 < < 0.97 and 1 < Re < 104 [8] Sh =
0.455
Re 0.59Sc 0.33

Overall mass transfer coefficient (KG)

Approximation by Glueckauf [9] dp dp 2
q* q0* b
1 2 0 b 2
= +
KG 3kext C0 15De C0
The effective diffusivity, De, is calculated by the Bosanquet correlation [10]; a typical value of 3 can be adopted for p 1 1
1
De = +
tortuosity ( p ) p Dm DK
The molecular diffusivity for binary mixtures, Dm , is calculated by the Fuller-Schettler-Gridding correlation [11] 1 1
0.5
10 7T1.75 +
Mj Mi
Dm = Di, j = 2
1 1
P v j3 + vi3

In case of multiple compounds, diffusion of each component is considered [12] Dm, i =

1 yi
yj
n
j i D
i, j

Solid-gas heat transfer coefficient (hsf)

Gunn correlation modified by Sun-Tenneti-Subramaniam for fixed-bed with high void fraction values [13], valid for 0.46 + 1.77 + 0.69 2
Nu = 3 + (1.37 2.4 + 1.2 2) Re 0.7Pr 1/3
0.35 < < 1 and 1 < Re < 105

Wall heat transfer coefficient (hint)

Li and Finlayson correlation [14] for 1-D model (0.05 < dp/D < 0.3, 20 < Re < 7600)
Nuint =
hint D
kf
= 2.03 Re 0.8 exp ( ) 6dp
D

External heat transfer coefficient between the walls of the bed and the ambient air (hout)
Morgan correlation valid for external free convection through long vertical cylinders [15] NuD = C RaDn
W
A value of 5 may be used for h out , which is typical for natural convection with external air.
m2K

Gas thermal conductivity (kg)

This parameter does not have a high influence on the breakthrough curves, so it can be approximated to the conductivity of the main compound. If such a compound does not exist,
Wilke approach to the Chapman-Enskog model, valid for gas mixtures [16], can be also used
n yi ki 1 1 2 1 1
kf = i=1 n y , where 2
j = 1 j ij µi 2 Mi 4 Mi
ij = 1+ 4 2 1+
µj Mj Mj

Effective axial thermal conductivities (ke) ks

0.28 0.757log10
Yagi, Kunii and Wakao correlation [17] modified by Krupiczka [18], disregarding radiation effects because of kf , e = kf (ks /kf ) kf
+ 0.75RePr
temperatures are relatively low
In this way, for fluid phase two terms are considered for effective thermal conductivity: one due to molecular
ks , e = k s
conduction in the fluid (quiescent bed) and another due to fluid convection, while for the solid phase there is only
one term related to molecular conduction in the solid.

5
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

hext D g air (Tf 298) D3 1 To solve Eq. (4), the General Form PDE can be selected with
NuD = ; RaD = with =
kf µair T P
uq = [ q1 q2 q3 ] (Eq. (12)):

The correlations used to estimate the model parameters [4–18] are d 2uq duq
ea + da + · =f
given in Table 1. dt 2 dt (12)

0 0 0 1 0 0 0
3.5. Some remarks on simulation
ea = 0 0 0 da = 0 1 0 = 0
0 0 0 0 0 1 0
Regarding the bulk separation, velocity, density and specific heat
K G,1 (q1* q1 ) 0 0
change with time, which involves the use of more computing resources
associated with the quality of space-mesh and time-mesh. In order to f= 0 K G,2 (q2* q2 ) 0
reduce these resources, density and specific heat are outside the time- 0 0 K G,3 (q3* q3)
derivative and space-derivative, but they are updated in each integra-
tion interval as well as the volumetric flow-rate and, hence, the su-
perficial velocity. In fact, the flow-rate is reduced due to the mass loss of 4. Results and discussion
different compounds that are adsorbed, and, simultaneously, increased
due to the increase in temperature and the pressure drop throughout Experimental data have been taken from references [19–25] to va-
the fixed-bed. The velocity changes with the flow-rate, and mass and lidate the proposed modeling, which is based on a minimum set of
energy balances are coupled through this variable. The update of the experimental data and the use of a reduced catalogue of correlations.
volumetric flow-rate after any integration step is: Most references are based on activated carbon (AC) due to its extremely
large, accessible internal surface, and large pore volume, especially,
Tnew Pprev Ct , out micropore volume, which allows multiple applications. In addition, AC
Fnew = Fprev
Tprev Pnew Ct , in with Fprev does not require prior stringent moisture removal, and the heat of ad-
sorption, or bond strength, on activated carbon is generally lower than
The pressure effect is disregarded in this work, since a negligible
that on other sorbents, so lower energy is needed for the sorbent re-
axial pressure gradient is assumed.
generation [26].
The implementation of equations in Comsol was carried out using
In the proposed modeling, simulated using Comsol, parameters are
the vector variable u (coefficient form PDE) for both the mass (uM) and
calculated by equations and correlations, but few experimental data are
energy (uE) balance equations (Eq. (11)):
needed, which is one of the key points of this paper. These data are the
d 2ui du following:
ea + da i + ( c ui ui + ) + ui + aui = f ,where i=M, E
dt 2 dt
Adsorbent data ρs: density of the adsorbent particles
(11)
εp: porosity of the adsorbent particles
Thus, for a three-component system, the matrixesassociated to dp: diameter of the adsorbent particles
uM = [C1 C2 C3 ] are: dpore: pore diameter of the adsorbent
Bed data ε: porosity of the bed
0 0 0 1 0 0 Dz ,1 0 0 0 0 0 D: bed diameter
ea = 0 0 0 da = 0 1 0 c = 0 Dz ,2 0 = 0 0 0 L: length of the bed, although this only influences the
time in which the breakthrough point is reached
0 0 0 0 0 1 0 0 Dz ,3 0 0 0
Operation data P: operating pressure
0 T0: fluid temperature at the bed inlet
= 0 yi,0: molar fractions of the feed compounds at the bed
0 inlet
F0: fluid flow-rate at the bed inlet.
q1 Isotherm data (with Lang- KL: equilibrium constant of the Langmuir isotherm.
0 0
u 0 0 0 0 0 t muir isotherm) qsat: saturation charge of the adsorbent.
1 q2 ΔHads: enthalpy of adsorption.
= 0 u 0 a= 0 0 0 f= s 0 0
t
0 0 u 0 0 0
Any isotherm can be adapted to a Langmuir isotherm. Likewise, phy-
q3
0 0
t
sicochemical properties of the compounds can be obtained from open
Likewise, for uE = [Tf Tw ]: database.
Most of these parameters will be normally known, since the ad-
Cp, f u 0 kf ,e 0
ea = 0 0 a = 0 0 = f
c= = 0 0 sorbent is characterized, the sizing of the column is available, and the
0 0 0 0 0 0 0 0 0 0
operating conditions must be set; isotherms should have been de-
0 termined but if not, some experiments should be carried out before
=
0 applying the modeling and simulation to predict the behavior of the
fixed-bed. In addition, some relationships among the 14 parameters
Cp, f + (1 ) s Cp, s 0
da =
f
allow the reduction of this number. Thus, except for very low column
0 w Cp, w diameters, bed porosity ranges from 0.3 to 0.5. Likewise, from the
column diameter and the flow-rate, the gas superficial velocity is ob-
qi h
(1 ) ( Hads, i ) t s
4 Din
in
0 tained, which should be kept constant when scaling-up, so dimension-
less Reynolds number is constant as present in almost all the correla-
f= (Tf Tw )
tions used. Similarly, the pore characteristics influence the effective
(Dhin (Tf Tw ) Dout hout (Tw To ))
0 4 diffusion coefficient, but usual values of micropore diameters (about
(Dout 2 D 2)
1–2 nm) and particle porosities (around 0.2–0.3) might be used if these

6
F.J. Gutiérrez Ortiz, et al.
parameters were unknown, as they are not too relevant.
4.1. Modeling validation
Two case-studies of purification and three cases of bulk separation
have been simulated by Comsol in order to verify the goodness of the
modeling proposed in this paper. Along with these cases, the reference
from which the experimental data were taken is cited. Tables S1–S5
show the data withdrawn from experimental papers to implement the
code in Comsol (see Supplementary Material) for the five case-studies
included in this work.
All the case-studies have been performed using both heterogeneous
and pseudo-homogeneous modeling. In all the cases, the results are
exactly the same. The reason for such a behavior is in the heat transfer
from sorbent to fluid, which is very high. In fact, as the heat transfer
coefficient ranges from 50 to 100 W·m−2·K−1, and particle diameter is
about 1–2 mm, the heat transfer resistance is very low, and the heat
released during the adsorption simultaneously increases the tempera-
ture in both the solid and the gas.
In addition, most of references used in the case-studies include a
section of modeling apart from the experimental data, whose results are
very close to those obtained in the modeling presented in this work,
with minor differences. In fact, the current output is oftentimes even
better than that used by others, and even sometimes no differences can
be found between the previous and current outputs. Figs. S1 and S2
(Supplementary Material) show a comparison between the modeling
results previously provided and the output of the current modeling for
the case-studies 2 (purification) and 3 (bulk separation) selected to il-
lustrate the discrepancies. These are not substantial, and among the
most important results regarding the modeling now presented is that no
fitting parameter is used in the current modeling.
4.1.1. Purification case-studies
No article providing experimental data of temperature for a
C0: 1980 ppm
C0 : 1065 ppm
C0 : 570 ppm
1600
C0 : 162 ppm
1200
C (ppm)
800
400
0 5000
Time (s)
Fig. 1. Breakthrough curves for the case-study 1 (H2S removal) at four H2S feed concentrations at 1 atm and room temperature (source of experimental data: [19]).
7
Chemical Engineering Journal 378 (2019) 121985
purification adsorption has been found, probably because the thermal
effects observed have been negligible. From all the simulation outputs,
the increase in temperature predicted by the model is always < 1 K.
The increase in temperature due to the release of adsorption heat in
purification processes is insignificant because the amount of adsorbate
adsorbed on the sorbent is very small. Therefore, purification processes
can be modeled as isothermal processes, and no energy balance equa-
tion is needed. In addition, simulation was performed using both
Danckwerts and Dirichlet boundary conditions at the bed inlet, ver-
ifying that the model output is the same.
- Case-study 1. Hydrogen sulfide removal [19]
Experiments were carried out in a fixed-bed cylindrical column (30-
mm ID, 100 mm height) and a thermally treated sewage–sludge was
used as an adsorbent. A gas composed of a mixture of CH4 (60 vol%),
H2S (variable, but < 2000 ppm) and CO2 (balance) was entered the
bed. A detailed description of the experiment and results can be found
elsewhere [19]. The H2S adsorption isotherms and breakthrough curves
were obtained at H2S concentrations of 1980, 1065, 570 and 162 ppm.
Fig. 1 illustrates the comparison between the results from the ex-
periment (points) and the model (solid lines), at different H2S inlet
concentrations. The breakthrough curves are steeper when the H2S feed
concentration is higher. The simulation outputs are the same as those
previously reported [1] but using less input parameters and direct
equations without converting them into dimensionless equations, thus
optimizing the simulation performance in Comsol. Note that the overall
mass transfer and the axial dispersion coefficient were not fitted from
experiments. It is observed that the predictions match well with the
experimental data, although the final slight curvature exhibited in the
experimental breakthrough curves deviates from the prediction; the
small slope in the final adsorption stages means that adsorption rate
becomes very small as the adsorbent is close to saturation. It might be
because of the overall mass transfer coefficient decreases over time,
10000 15000 20000
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

1.2 which cannot be represented by the constant value used in simulation.

(a) This effect of KG on the shape of the breakthrough curve is corrected in
1.0 the case studies relative to bulk separation, where the value of KG is
continuously updated. The other calculated parameter, the axial dis-
0.8 persion coefficient, also influences the adsorption performance and is
updated in bulk separation cases as well.
C/C0

0.6
- Case -study 2. Thiophene removal [20]

0.4 Sorption experiments were performed in a laboratory setup for the

removal of heterocyclic sulfur compounds by a commercial activated
200 ºC
0.2 150 ºC
carbon. Breakthrough curves were experimentally obtained at 100, 150
100 ºC and 200 °C for C4H4S inlet concentrations of 8.5, 15 and 30 ppm, and
flow-rates of 30, 42.5 and 55 NmL min−1, at temperature 100 and
0.0
0 20 40 60 80 100 120 140
150 °C. The thiophene was mixed in Ar as the carrier while the ad-
sorption of Ar was assumed negligible relative to thiophene.
Time (min) Fig. 2 shows the experimental results (points) as well as the curves
simulated (solid lines). The agreement between the modeling output
1.2 and the experimental results is very good in most tests.
(b) Like in the case-study 1, a higher concentration of the adsorbate at
1.0 the inlet causes the breakthrough time to be reached earlier, since the
amount of sorbent is the same but is spent earlier because a greater
amount of adsorbate is adsorbed.
0.8
No thermal effects were detected in any of the purification case-
studies.
C/C0

0.6
4.1.2. Bulk separation case-studies
0.4 The gas adsorbed in a purification process is, usually, < 1% of the
total gas entering the column, so the effect on the fluid flow-rate and
55 NL/min any thermal effect are negligible. To observe an appreciable increase in
0.2 42.5 NL/min the temperature, the concentration of the adsorbate should be about or
30 NL/min greater than 10 wt% of the total feed that corresponds to bulk separa-
0.0 tion processes, in which the amount adsorbed may be of the same
0 10 20 30 40 magnitude order as that non-adsorbed. In this case, the effect on the
Time (min) fluid flow-rate can become significant because the superficial fluid ve-
locity is reduced as the flow-rate decreases, and the gas density, the
1.2
fluid specific heat and other properties change due to the variation in
8.5 ppm the mass fraction of compounds. In addition, noticeable thermal effects
15 ppm
(c) may be expected due to the relatively high amount of adsorbed gas that
1.0 30 ppm consequently releases a higher amount of heat. These thermal effects
also affect the velocity and properties aforementioned.
0.8
- Case study 3. Methane and carbon monoxide adsorption [21,22]
C/C0

0.6
The simulated system is the multicomponent adsorption of the ef-
fluent coming from a steam methane reforming to purify the hydrogen
0.4
obtained. The gas is composed of H2, CH4 and CO, and all of them are
adsorbed, although hydrogen is barely adsorbed. A ternary mixture
0.2 (H2/CH4/CO of 60.4/28.1/11.5 vol%) was fed into the fixed-bed con-
taining a commercial activated carbon. The pressure was kept constant
0.0 in the fixed-bed by a back-pressure regulator. The adsorption bed was
0 20 40 60 80 100 120 140 160 180 initially saturated with H2 at the adsorption pressure.
Fig. 3 shows the breakthrough curves of the gas mixture, which
Time (min)
match reasonably well with the experimental data (points). The evo-
Fig. 2. Breakthrough curves for the case-study 2 (thiophene removal) at lution of H2 is determined by the evolution of CO and CH4. CO is less
1.5 bar: (a) at 100, 150 and 200 °C (15 ppm C4H4S, 30 NmL min−1); (b) for adsorbed than CH4 and starts to increase near 500 s, with a slope that
flow-rates of 30, 42.5 and 55 NmL min−1 (15 ppm C4H4S, 150 °C); (c) for 8.5, superposes on the experimental data. The maximum value is somewhat
15 and 30 ppm C4H4S (100 °C, 30 NmL min−1) (source of experimental data: lower than that experimental, and a small deviation is exhibited be-
[20]). tween 550 and 650 s; during this period, a plateau of the molar fraction
of the CO and H2 can be seen. After that, the agreement is pretty good.

8
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

1.0
H2 model
CH4 exper.
CO exper.
0.8

Molar fraction
0.6

0.4

0.2

0.0

0 200 400 600 800 1000 1200 1400 1600

Time (s)
Fig. 3. Breakthrough curves for the case-study 3 (CH4 and CO adsorption) at 9.8 bar and 289.1 K (source of experimental data: [21,22]).

315

313 T_0.1L
T_0.3L
311
T_0.5L
309
T_0.75L
307 T_L
Temperature (ºC)

305

303

301

299

297

295

293

291

289
0 200 400 600 800 1000 1200 1400 1600
Time (s)
Fig. 4. Simulated temperature evolution over time for the case-study 3 (CH4 and CO adsorption) at different locations (9.8 bar, 289.1 K).

When the sorbent becomes saturated of CH4, the corresponding
breakthrough curve appears showing a slight deviation regarding the
experimental data, probably because of the overall mass transfer coef-
ficient is not high enough or the axial dispersion coefficient should be
somewhat lower, although the values required for that are not very
different from those provided by correlations.
Fig. 4 shows the trend of temperature with time at 0.1L, 0.3L, 0.5L,
0.75L and L from the fixed-bed entry, all together to illustrate how the
temperature profile changes over time depending on the section of
fixed-bed. Fig. 5 compares the profiles obtained by simulation (solid
lines) with experimental data (points) at four different sections. It can
be seen that those match well with each other from the initial time up to
reach the maximum or peak value, corresponding to the maxima of
qi / t for CO and CH4 that take place at different moments; after that,

9
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

315 315

(a) (b)
310 310

305 305
Temperature (K)

Temperature (K)
300 300

295 295

290 290

0 200 400 600 800 1000 1200 1400 1600 0 200 400 600 800 1000 1200 1400 1600
Time (s) Time (s)

315 315

(c) (d)
310 310

305 305
Temperature (K)

Temperature (K)
300 300

295 295

290 290

0 200 400 600 800 1000 1200 1400 1600 0 200 400 600 800 1000 1200 1400 1600
Time (s) Time (s)

Fig. 5. Temperature evolution for the case-study 3 (CH4 and CO adsorption) at three locations: (a) 10 cm, ((b) 30 cm, (c) 50 cm and (d) 75 cm, at 9.8 bar and 289.1 K
(source of experimental data: [21,22]).

the simulated temperatures decrease because the bed (note that solid
and gas temperatures are equal) is eventually cooled by the heat ex-
change with the environment and by the feed gas entering the fixed-bed
at a lower temperature. However, the cooling is faster than that ob-
tained experimentally. There may be some reasons for this: (1) the
values of the heat transfer coefficients (especially, hin ) are computed
from correlations, but a better fitting can be obtained if those values are
manually adjusted; (2) the effective axial thermal conductivity (kf , e ) has
also a significant effect. Thus, for example, if hin is reduced and/or kf , e is
increased, the tail beyond the maxima has a lower slope, i.e., tem-
peratures diminish slowly over time. However, a too low value of hin
implies an approximation to an adiabatic system, and hence, higher
temperatures, depending also on the value of hout . Similarly, a too high
value of kf , e leads to a lower maximum for temperature. Moreover, in
these cases, the maxima might increase over the length, contrary to the
evolution shown in Fig. 5. Nevertheless, the first stage of the evolution
of temperatures due to the active adsorption is the key to know when
the maxima are reached and the rate for it. The slower rate of cooling
after that is less important in these predictions.
- Case study 4. Carbon dioxide adsorption [23]
The behavior of an adsorbent based on commercial granular acti-
vated carbon (AC) and activated carbon modified by ammonia
derivatives (N-doped AC) adsorbents for carbon dioxide capture was
tested. The modified adsorbent enhances the performance, because of
the additional chemical adsorption taking place. In this work, no che-
misorption is considered, so only pure (unmodified) sorbent (AC) is
used to validate the proposed modeling. Two experiments were carried
out for this latter sorbent at two inlet volumetric flow-rates of 50 mL/
min and 100 mL/min, and temperatures of 30, 45 and 60 °C. The inlet
composition of the feed gas is 15/85 CO2/N2 vol% at 1 bar.
From the isotherm proposed in the study, only the term relative to
physical adsorption is used, which was fitted following the Tóth iso-
therm. In the modeling, in addition to this latter, Langmuir isotherm is
applied by adjusting the isotherm to 30 °C and varying the parameters
with temperature by the van’t Hoff Equation. Since Reynolds number is
about 1, the correlations to be used are that by Rastegar-Gu for the axial
dispersion coefficient and the Dwivedi-Upadhyay correlation to obtain
the external mass transfer coefficient.
In Fig. 6, the breakthrough curves are drawn in different colors at 50
and 100 mL/min at the three mentioned temperatures. In addition, the
increase in temperature is lower than 1 °C, and no thermal effects are
exhibited. As expected, the breakthrough curves move towards the
right, i.e., to higher times, at lower flow-rate and temperature values,
showing a good agreement between the model and the experiment in all
the tests.

10
F.J. Gutiérrez Ortiz, et al.
Fig. 6. Breakthrough curves for the case-study 4 (CO2 adsorption) at 1.0 atm
and (a) 50 mL·min−1 and (b) 100 mL·min−1 (source of experimental data:
[23]).
- Case study 5. Carbon dioxide and nitrogen separation [24,25]
The study deals with CO2 capture by adsorption in post-combustion.
The stream of gas is composed of 80% N2 and 20% CO2. The adsorbent
used was activated carbon. The breakthrough curves were obtained at
different temperatures (25, 50, 100 and 150 °C), at 1.0 bar. Fig. 7 shows
the experimental breakthrough curves obtained by the researchers
(points), as well as the simulation output of the proposed modeling
(solid lines). As the temperature increases, the simulations match
equally well with the experimental results; indeed, the agreement is
very good for all the tests.
No experimental curves for temperature were presented in the
11
Chemical Engineering Journal 378 (2019) 121985
original study. However, Fig. 8 shows the evolution of temperature in
the four experiments performed under adiabatic conditions at the exit
of the bed, which does not exceed 6 K at 373 and 423 K, so similar
results should have to be obtained by simulating under isothermal
conditions. However, at 301 K and 323 K, the increase in temperature is
22 and 15 K, respectively, so there is a thermal effect on the fixed-bed
performance. Note that the effect described in the case-study 3, relative
to low values of hin , is clearly shown here; in fact, some simulations
have been performed using a non-adiabatic system, and the break-
through curves widen over time, because the heat released to en-
vironment leads to temperature profiles with lower peaks, which take
place sooner.
4.2. Thermal effects on breakthrough curves
When the temperature changes, the equilibrium constant of the
adsorption also varies, so that the breakthrough curves will change. The
modification of the equilibrium constant can be approximated by in-
tegrating the van't Hoff equation (Eq. (6)) by assuming that the ad-
sorption heat does not change with temperature; in this way, the new
equilibrium constant can be obtained from that obtained experimen-
tally at a reference temperature (Tref ):
Hads Tref
KL, T = KL, ref exp 1
RTref Tf
As adsorption is an exothermic process, the equilibrium constant
decreases when the gas temperature increases. Therefore, the perfor-
mance of the process enhances at a lower temperature, as can be seen in
some of the previous Figures.
In addition, the axial dispersion coefficient increases with tem-
perature, because it is directly proportional to the diffusivity that is
proportional to the 1.75th power of the temperature. If the axial dis-
persion coefficient rises, the slope of the breakthrough curve will de-
crease, and the breakthrough time will be reached sooner. It is because
a high axial dispersion coefficient enlarges the width of mass transfer
zone that, hence, arrives earlier at the end of the bed and a non-zero
exit concentration is detected earlier.
The temperature and concentration profiles have been shown for
five case-studies, although those relative to bulk separation are more
significant than those for purification processes. The heat released in
the adsorption is directly related to the breakthrough curve, since it is
proportional to the adsorption rate. The heat generated over time
reaches a maximum at any bed section, which corresponds to the
maximum slope of the breakthrough curve. From that point, the slope
changes from being increasing to decreasing (Figs. 5 and 8), because
when the slope begins to rise from zero (at a certain bed zone), heat of
adsorption starts to be released and reaches a maximum value corre-
sponding to maximum value of qi / t . Beyond this point, the slope
begins to decrease because of the heat released by adsorption decreases
to zero and the cooling through the environment. In this way, the re-
leased heat leads to an increase in temperature of this part of the bed;
likewise, heat is transmitted from the bed (solid) to the gas. As the
product of the heat transfer coefficient and the interface area is very
high, the heat transfer is fast, and the temperatures of both phases are
similar, so the curves appear overlapped; indeed, the simulation output
of heterogeneous and pseudo-homogeneous modeling are the same. The
gas heats up the solid disposed in the sections of sorbent downstream
from the mass transfer zone, which has not participated yet in the
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

1.4 1.4

(a) (b)
1.2 1.2

1.0 1.0

0.8 0.8

C/Co
C/Co

0.6 0.6
N2 N2
CO2 CO2
0.4 0.4

0.2 0.2

0.0 0.0
0 50 100 150 200 250 0 50 100 150 200 250
Time (s) Time (s)

1.4 1.4

(c) (d)
1.2 1.2

1.0 1.0

0.8 0.8
C/Co
C/Co

0.6 0.6
N2 N2
CO2 CO2
0.4 0.4

0.2 0.2

0.0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
Time (s) Time (s)

Fig. 7. Breakthrough curves for the case-study 5 (CO2 and N2 adsorption at (a) 301 K, (b) 323 K, (c) 373 K, (d) 423 K, at 1.02 bar and 30 mL min−1 (source of
experimental data: [24,25]).

adsorption process and, therefore, it is still at room/initial temperature.
Temperature peaks will be achieved just in the end of adsorption pro-
cess itself.
The shape of temperature profiles depends on the axial thermal
conductivity. Furthermore, if no axial thermal conductivity was con-
sidered, the maximum temperature achieved would be higher, because
the width of mass transfer zone would decrease, and the adsorption
would take place in a narrower band of fixed-bed where the heat is
released (Fig. 9). Similarly, a higher inlet concentration would lead to a
higher temperature, since the amount adsorbed per unit time at any
fixed- bed section would be greater, so the heat accumulated in that
area would be also higher as more heat is released.
The change of concentration in the radial axis (2-D model) is not
perceptible because the increase in the temperature that causes the
radial dispersion is not enough to show significant differences. These
might be observed if heat released was much higher, which will be
addressed in a future work.
5. Conclusions
So far, models of fixed-bed adsorption for gas separation has been
carried out ad hoc, by obtaining a few preceding key parameters to

12
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

324 338
322
336
320
318 334
Temperature (K)

Temperature (K)
316
332
314
312 330
310
328
308
306 326
304
324
302 (a) (b)
300 322
0 20 40 60 80 100 120 140 160 180 200 220 240 260 0 20 40 60 80 100 120 140 160 180 200 220 240 260
Time (s) Time (s)

380 427

379
426
378
Temperature (K)

Temperature (K)
377 425
376

375 424

374
423
373
(c) (d)
372 422
0 20 40 60 80 100 120 140 160 180 200 220 240 260 0 20 40 60 80 100 120 140 160 180 200 220 240 260
Time (s) Time (s)

Fig. 8. Temperature evolution for the case-study 5 (CO2 and N2 adsorption) at the outlet (z = L), 1.02 bar and 30 mL min−1, and at: (a) 301 K, (b) 323 K, (c) 373 K,
(d) 423 K (source of experimental data: [24,25]).

achieve the best fit with the experimental breakthrough curves for a
given adsorbent. In this work, a modeling is proposed to predict a priori
the breakthrough curves using a minimum number of data, mainly re-
lated to the sorbents and isotherms. Boundary conditions have been
discussed and a complete, but not too large, list of correlations have
been included. Five case-studies have been used to demonstrate the
goodness of the one-dimensional (1-D) pseudo-homogeneous modeling,
which could be applied to any other case of physical adsorption. In the
near future, the extension to chemisorption will be dealt with, con-
sidering two-dimensional (2-D) heterogeneous modeling. The modeling
and methods provided in this paper can be used with confidence to
assess existing fixed-beds and to design and evaluate a fixed-bed ad-
sorber on a larger scale.

13
F.J. Gutiérrez Ortiz, et al. Chemical Engineering Journal 378 (2019) 121985

doi.org/10.1016/j.cej.2019.121985.

References

[1] P.G. Aguilera, F.J. Gutiérrez Ortiz, Prediction of fixed-bed breakthrough curves for
H2S adsorption from biogas: Importance of axial dispersion for design, Chem. Eng.
J. 289 (2016) 93–98.
[2] P.V. Danckwerts, Continuous flow systems, Chem. Eng. Sci. 2 (1953) 1–13.
[3] A.P. Coppola, M.D. Levan, Adsorption with axial diffusion in deep beds, Chem. Eng.
Sci. 36 (1981) 967–971.
[4] N. Wakao, T. Funazkri, Effect of fluid dispersion coefficients on particle-to-fluid
mass transfer coefficients in packed beds: correlation of Sherwood numbers, Chem.
Eng. Sci. 33 (1978) 1375–1384.
[5] S.O. Rastegar, T. Gu, Empirical correlations for axial dispersion coefficient and
Peclet number in fixed-bed columns, J. Chromatogr. A 1490 (2017) 133–137.
[6] C.Y. Wen, L.T. Fan, Models for Flow Systems and Chemical Reactors, Marcel
Dekker, New York, Vol, 1975.
[7] W.E. Ranz, W.R. Marshall, Evaporation from drops. Part II, Chem. Eng. Prog. 48
(1952) 173–180.
[8] P. Dwivedi, S.N. Upadhyay, Particle-fluid mass transfer in fixed and fluidized beds,
Ind. Eng. Chem. Prod. Rd 16 (1977) 157–165.
[9] R.T. Yang, Gas Separation by Adsorption Processes. Series on Chemical
Engineering, Imperial College Press, London, 1997.
[10] R. Krishna, J.M. van Baten, Investigating the validity of the Bosanquet formula for
estimation of diffusivities in mesopores, Chem. Eng. Sci. 69 (2012) 684–688.
[11] E.N. Fuller, P.D. Schettler, J.C. Giddings, A new method for prediction coefficients,
Ind. Eng. Chem. 58 (1966) 18–27.
[12] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, 2nd Ed, John Wiley
& Sons, New York, 2002.
[13] B. Sun, S. Tenneti, S. Subramaniam, Modeling average gas-solid heat transfer using
particle-resolved direct numerical simulation, Int. J. Heat Mass Tran 86 (2015)
898–913.
[14] C. Li, B. Finlayson, Heat transfer in packed beds−a re-evaluation, Chem. Eng. Sci.
32 (1977) 1055–1066.
[15] F.P. Incropera, D.P. Dewitt, T.L. Bergman, A.S. Lavine, Fundamentals of Heat and
Mass Transfer, sixth Ed., John Wiley & Sons, New York, 2007.
[16] C.R. Wilke, A viscosity equation for gas mixtures, J. Chem. Phys 18 (1950)
517–519.
[17] N. Wakao, S. Kaguei, T. Funazkri, Effect of fluid dispersion coefficients on particle-
to-fluid heat transfer coefficients in packed beds”, Chem. Eng. Sci. 34 (1979)
325–336.
[18] R. Krupiczka, Analysis of thermal conductivity in granular materials, Int. Chem.
Eng. 7 (1967) 122–144.
[19] F.J. Gutiérrez Ortiz, P.G. Aguilera, P. Ollero, Biogas desulfurization by adsorption
on thermally treated sewage-sludge, Sep. Purif. Technol. 123 (2014) 200–213.
[20] P. Edinger, D. Grimekis, K. Panopoulos, S. Karellas, C. Ludwig, Adsorption of
thiophene by activated carbon: a global sensitivity analysis, J. Environ. Chem. Eng.
5 (2017) 4173–4184.
[21] H. Ahn, C. Chun, M. Park, I.-S. Ahn, C.-H. Lee, Thermal effects on the breakthrough
curve of a hydrogen ternary system at a fixed bed, Sep. Sci. Technol. 36 (2001)
2121–2145.
[22] S. Ahn, Y.W. You, K.H. Kim, C.H. Lee, Layered two- and four-bed PSA process for H2
recovery from coal gas, Chem. Eng. Sci. 68 (2012) 413–423.
[23] M.S. Shafeeyan, W.M.A.W. Daud, A. Shamiri, N. Aghamohammadi, Modeling of
carbon dioxide adsorption onto ammonia-modified activated carbon: kinetic ana-
lysis and breakthrough behavior, Energ. Fuel 29 (2015) 6565–6577.
[24] T.L.P. Dantas, F.M.T. Luna, I.J. Silva Jr., D.C.S. de Azevedo, C.A. Grande,
A.E. Rodrigues, R.F.P.M. Moreira, Carbon dioxide–nitrogen separation through
Fig. 9. Simulated temperature evolution over time for the case-study 3 (CH4 adsorption on activated carbon in a fixed bed, Chem. Eng. J. 169 (2011) 11–19.
[25] T.L.P. Dantas, S. Amorim, F.M.T. Luna, I.J. Silva Jr., D.C.S. de Azevedo,
and CO adsorption) at different locations under adiabatic conditions: (a) k f , e (b)
A.E. Rodrigues, R.F.P.M. Moreira, Adsorption of carbon dioxide onto activated
k f , e × 5 (c)k f , e = 0 . carbon and nitrogen-enriched activated carbon: surface changes, equilibrium, and
modeling of fixed-bed adsorption, Sep. Sci. Technol. 45 (2010) 73–84.
[26] R.T. Yang, Adsorbents: Fundamentals and Applications, John Wiley & Sons Inc,
Appendix A. Supplementary data Hoboken, New Jersey, 2003.

Supplementary data to this article can be found online at https://

14