You are on page 1of 11

Chemical Engineering Journal 253 (2014) 305315

Contents lists available at ScienceDirect

Chemical Engineering Journal


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

Modeling and simulation of the adsorption of biogas hydrogen sulde on


treated sewagesludge
F.J. Gutirrez Ortiz , P.G. Aguilera, P. Ollero
Departamento de Ingeniera Qumica y Ambiental, Universidad de Sevilla, Camino de los Descubrimientos, s/n, 41092 Sevilla, Spain

h i g h l i g h t s

 The dynamic behavior of a xed bed with treated sewagesludge for H2S removal is modeled.
 The Bohart and Adams model does not predict suitably the complete breakthrough curves.
 The Klinkenberg solution implies a linear isotherm, and the prediction is not good enough.
 The LDF model with a non-linear isotherm provides a good approach of the adsorption process.
 As an application, a scale-up of the lab scale to a larger one was performed and assessed.

a r t i c l e i n f o a b s t r a c t

Article history: For the design of a xed-bed adsorber, it is essential to know the prediction of the column breakthrough.
Received 4 April 2014 This paper compares two modeling approaches of the dynamic behavior of a xed bed that contains trea-
Accepted 30 April 2014 ted sewagesludge for biogas desulfurization. The rst approach is based on the simple model by Bohart
Available online 21 May 2014
and Adams, which has been widely used to describe the adsorption dynamics when chemical reaction
takes place, and the second one uses the linear driving force model (LDF), solving it analytically by Klin-
Keywords: kenberg equation and numerically with the aid of Aspen Adsorption. While the BohartAdams model
Biogas
uses rectangular isotherm to describe the equilibrium, the LDF model is solved with a more realistic
Hydrogen sulde
Sewagesludge
and suitable isotherm. The Klinkenberg solution implies a linear isotherm, and the prediction is not good
Simulation enough, although better than that of the approach of BohartAdams. The trouble is that the linear iso-
Adsorption therm used in the analytical solution is not realistic and a nonlinear isotherm is required. This paper pre-
Modeling sents a methodology to estimate the overall mass transfer coefcient dynamically as well as the isotherm
that best ts. Thus, the LDF model with the Freundlich isotherm provide a breakthrough curve that match
the experimental results much better than the BohartAdams approach. In order to apply the prediction
obtained, a scaling up of the lab scale was performed, discussed and assessed.
2014 Elsevier B.V. All rights reserved.

1. Introduction different processes for removal of hydrogen sulde from landll


biogas, but activated carbon adsorption process has been mainly
Landll biogas is a product of anaerobic degradation of organic studied by researchers in recent years, because their large surface
substrates, and conversion of the chemical energy contained in area and surface functional groups [2].
biogas, which is rich in CH4, into electricity is possible through Although sewagesludge surface is smaller than that of acti-
combustion in internal combustion engines [1]. These units can vated carbons, they can remove hydrogen sulde, once suitably
be seriously damaged by the hydrogen sulde (H2S), because it treated, and they are much cheaper than activated carbons. Thus,
can cause corrosion. Therefore, the removal of H2S from the biogas a xed-bed of treated sewagesludge may be an interesting option
is particularly crucial in order to use it as fuel for engine. for the biogas purication process.
Desulfurization of gaseous fuel can be done on various adsor- An experimental characterization of several treated sewage
bents depending on the temperature of the feed gas. There are sludge was previously carried out, and as a relevant result, the
adsorbent obtained by calcination at 700 C of one of the three
kinds of sludge (coded as LG700PA) showed an adsorption capacity
Corresponding author. Tel.: +34 95 448 72 68. twice of that of a commercial activated carbon without impregna-
E-mail address: frajagutor@etsi.us.es (F.J. Gutirrez Ortiz). tion [3].

http://dx.doi.org/10.1016/j.cej.2014.04.114
1385-8947/ 2014 Elsevier B.V. All rights reserved.
306 F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315

Nomenclature

C H2S concentration in gas phase KF coefcient of the Freundlich isotherm


C0 H2S feed concentration L total bed depth
Cb H2S breakthrough concentration q concentration of H2S (or load) in solid phase
dp mean particle diameter qe value of q in equilibrium with C
dpore mean pore diameter qs saturation value of q
De effective diffusivity t time
Di internal diameter of the bed tb breakthrough time
DK Knudsen diffusivity u gas interstitial velocity
Dm molecular diffusivity z distance from the bed entrance
ftor tortuosity factor
kBA BohartAdams rate constant Greek
kext external mass transfer coefcient e bed void fraction
Ka overall mass transfer coefcient ep particle porosity
Ke equilibrium constant of a linear adsorption isotherm q gas density
KL equilibrium constant of Langmuir l gas viscosity

A continuous xed-bed adsorber does not run under equilib- In this paper, different adsorption models are proposed to
rium conditions, so the ow conditions and mass-transfer aspects describe the breakthrough curves that were experimentally
throughout the column affect the performance [4]. The dynamic obtained in a previous work [3]. Thus, the pioneer model by Bohart
behavior of a xed bed column is described in terms of the efuent and Adams was selected to compare it to the linear-driving-force
concentrationtime prole, i.e., the breakthrough curve, which is (LDF) approximation, solved analytically by the Klinkenberg equa-
essential in the evaluation of the efcacy of an adsorber. The shape tion and numerically with the aid of Aspen Adsorption, depending
of the equilibrium isotherm (graphical representation of the gas on the adsorption isotherm. Likewise, the results of these
uptake quantity as a function of pressure at a constant tempera- approaches were compared to experimental data. For the valida-
ture) determines the shape of the breakthrough curve, which is tion of the models, some of the experimental points of each break-
also affected by both inter- and intra-particle transfer processes through test were reserved to compare their results with those
in the column and adsorbent [5]. Therefore, accurate measurement obtained using the model proposed. Additionally, in some cases,
of the isotherm and mass transfer phenomena are fundamental. replicates could be carried out and they were also compared to
This paper compares two modeling approaches of the dynamic the models. Finally, it was assumed that only the hydrogen sulde
behavior of a xed bed that contains thermally treated sewage was adsorbed although both the carbon dioxide and the methane
sludge for biogas desulfurization. Experimental results obtained could compete for the active adsorbent centers, being thus also
in a previous work [3] were use in the comparison of the two mod- adsorbed along with the H2S. Therefore, a very high selectivity of
eling approaches. The data and methodology given in this paper the adsorbent to the hydrogen sulde was assumed. From the com-
provide a basis for a dynamic adsorption model that could be used parison of modeling results and the experiment, this assumption
to design and evaluate an adsorber on a larger scale. As an applica- was quite right.
tion of that, a scale-up of the lab scale to a larger one with a biogas
ow-rate about 150 times higher was performed, discussed and 2.1. Breakthrough curve
assessed.
By assuming isothermal conditions and plug ow of gas through
2. Modeling the xed bed (zero axial dispersion), and taking into account that
changes in uid velocity are negligible because this is a trace sys-
Mathematical models are useful to understand the behavior of tem, i.e., the adsorbable component (H2S) concentration is low
xed-bed columns as well as to design and to optimize this process (<2000 ppm), the Eqs. (1) and (2) describe the dynamics of adsorp-
unit. To predict the dynamics of the hydrogen sulde adsorption on tion [6,8]:
carbonaceous material, such as activated carbon or sewagesludge, Mass differential balance equation for xed-bed:
the mathematical model must include the adsorption isotherm, the
@C @C 1  e @q
massenergy balance inside the adsorbent particle and the mass u 0 1
energy balance of the gaseous phase in the bed [6], as well as mass @t @z e @t
transfer phenomena (an equation for the mass ux). @q
In the xed-bed columns, the concentrations in gas phase f q; C 2
@t
change through the time and space. As the process is unsteady,
the development of a model that describes the concentration pro- where C is the H2S concentration in gas phase, u is the gas intersti-
le in the bed with the time is quite complex. Comprehensive tial velocity, e is the bed void fraction, q is the concentration of H2S
xed-bed models that take into account the nonlinear equilibrium (or load) in solid phase, t is the time and z is the distance from the
behavior as well as axial dispersion, resistance to mass transfer, bed entrance.
and sorption kinetics are described in terms of partial differential The initial and boundary conditions are the following:
equations, which generally involve a very difcult numerical solu- t 0 : Cz 0; qz 0 z  0
tion [7]. Consequently, simplied or shortcut methods for correlat-
z 0 : Ct; z C 0 t  0
ing the breakthrough curves of adsorbents columns are extensively
used for the initial design and analysis of xed-bed columns [8], where C0 is the concentration at the inlet of the bed.
and frequently, they provide a straightforward and helpful For Eq. (2), the idealized equilibrium model (instantaneous
approach. equilibrium is assumed at all points in the column and there are
F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315 307

no mass transfer resistances) is seldom accurate in predicting the 2.4. Linear-driving-force approximation
breakthrough behavior [6]. Thus, a nonequilibrium model is
needed to obtain a realistic dynamic response of the bed. The BohartAdams model and the Thomas model are based on
the assumption that the rate of adsorption is controlled by the sur-
2.2. The approach of Bohart and Adams face reaction. However, external and internal diffusion limitations
are not absent in the overall process. Moreover, adsorption may
Bohart and Adams [9] proposed a model for the adsorption of not be limited by kinetics of chemical reaction, but controlled by
one component that has been widely used to describe adsorption external and/or internal mass transfer. Thus, an alternative
dynamics when chemical reaction takes place, because they approach is to consider both inter- and intra-particle mass transfer
described the adsorption by a quasi-chemical rate equation. In to control the adsorption process.
this approach, the diffusion steps (in both lm and pore) are The linear-driving-force (LDF) approximation takes this
assumed to be very fast, and the surface step is rate-controlling. approach in such a way that a lumped overall resistance describes
The adsorption or surface rate (quasi-chemical rate law) is the mass transfer, and where the adsorbent particle is treated as a
following: homogeneous phase in which diffusion takes place with a constant
effective diffusivity; the reaction kinetics is assumed to be much
@q faster than the mass transport. Thus, the mass transfer rate is rep-
kBA Cqe  q 3
@t resented as proportional to the deviation from equilibrium:
where qe is the value of q in equilibrium with C, and kBA is the @q
BohartAdams rate constant. Now, q is the volume-average H2S K a qe  q 7
@t
concentration in solid surface. The model incorporates a rectangular
where Ka is the overall effective mass transfer coefcient.
adsorption isotherm, i.e., irreversible (the H2S adsorbed onto the
adsorbent surface cannot be desorbed), such that qe = qs (saturation
2.4.1. Analytical solution
value) for C > 0 and qe = 0 for C = 0.
If the equilibrium isotherm is linear (qe = KeC), an analytic
Thus, Eq. (1) can be analytically solved and written in a simpli-
expression can be derived, such as Klinkenberg did [13] using
ed linear form (Eq. (4)) [8]:
dimensionless equations as follows:
 
C0 kBA NL 1e @/ @ u
ln 1  kBA C 0 t b ; with N qs 4 Mass balance in the bed : 0 8
Cb u e @n @ s
where L is the total bed depth, Cb is the breakthrough concen-
tration and tb is the breakthrough time. The parameters N and @u
Adsorption velocity : /u 9
kBA can be obtained by linear regression from the slope and inter- @s
cept, respectively, of a linear plot of ln (C0/Cb  1) versus tb, if C0, L where:
and u are known.   
C q KeKaz 1  e z
Eq. (4) allows determining the breakthrough time under differ- / ; u ; n ; s Ka t 
ent experimental conditions as well as predicting the behavior at C0 qs u e u
different bed heights and hydrogen sulde concentrations. There- Thus, using the above mentioned initial and boundary condi-
fore, it could be a useful tool for design and scaling. tions, the analytical solution is:
  
C 1 p p 1 1
2.3. Adsorption isotherms  1 erf s  n p p 10
C0 2 8 s 8 n
The main difference between the BohartAdams model and where erf(x) is the error function of Gauss, dened by Eq. (11).
others similar models lies in the form of the isotherm assumed. Z x
2 2
Thus, for instance, the Thomas model [10] assumes a Langmuir iso- erf x p et dt 11
therm. The nonlinear isotherm of Langmuir has the following form: p 0

The coefcient of overall mass transfer (which includes both the


qe KLC
5 external and internal mass transfer) was determined as follows:
qs 1 K L C
2
1 dp dp
where KL is the equilibrium constant of Langmuir. When the 12
isotherm is highly favorable for adsorption, both models will give K e K a 6kext 60De
similar results [5]. Nevertheless, other possibilities for a better The external mass transfer coefcient (kext) was calculated by
description of the adsorption dynamics of H2S on thermally treated the RanzMarshall correlation (Eq. (13)) [14]:
sewagesludge consist of using other isotherms. This way, some
experiments were carried out to t Langmuir and Freundlich Sh 2:0 1:8Re1=2 Sc1=3 ; Sc < 400; 1 < Re < 70; 000 13
isotherms. where:
The experimental Freundlich isotherm is displayed as follows:
kext dp ue dp q l
qe K F C n 6 Sh ; Re ; Sc
Dm l qDm
where KF is a coefcient that mainly depends on the tempera- Dm, dp, q and l are the molecular diffusivity, particle diameter,
ture and the specic surface area of adsorbent. The exponent n also gas density and gas viscosity, respectively. Dm was calculated by
depends on the temperature. the FullerSchettlerGridding correlation (Eq. (14)) [15]:
Freundlich equation is more used in chemisorption than the  0:5
Langmuir equation due to its better t to the experimental results 1 1
103 T 1:75 Mgas MH2 S
[11]. This is because the Freundlich isotherm is more realistic con- Dm  2 14
P 1=3
sidering an energetically nonuniform surface instead of a uniform V 1=3
gas V H2 S
surface as in the Langmuir isotherm [12].
308 F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315

with Dm in cm2/s T in K, P in atm, M is molecular weight, and V is the was disposed. A simulated biogas composed by a mixture of CH4
diffusion volume. (60 vol.%), H2S (2000 ppm) and CO2 (balance) entered the bed, with
De is effective diffusivity, which may be calculated by the a supercial velocity at room temperature (25 C) of 2.6 cm/s.
Bosanquet equation (Eq. (15)): Three types of thermally treated sewagesludge were used as
 1 adsorbents. The thermal treatment was performed in a tubular fur-
ep 1 1 nace with different nal temperatures (500, 700 or 900 C), using a
De 15
ftor Dm DK holding time of 30 min and a heating rate of 5 C/min. In addition,
ep is the particle porosity and ftor is the tortuosity factor. two activated carbons, one fresh and the other chemically impreg-
Finally, DK is the Knudsen diffusivity, which is calculated by Eq. nated with NaOH, were used to establish a suitable comparison
(16): framework.
Physical and chemical characterization of the adsorbents was
 1=2
T carried out by inductively coupled plasma atomic emission spec-
DK 4850dpore 16 troscopy (ICP-OES), X-ray Diffraction (XRD), X-ray Photoelectron
M H2 S
Spectroscopy (XPS), Scanning Electronic Microscopy (SEM), physi-
where dpore is the mean pore diameter. sorption determination and mercury porosimetry. A detailed
The model was solved with the aid of Matlab R2010b description of the experimental apparatus, methods of investiga-
(Mathworks). tion and results can be found elsewhere [3].
In addition, the H2S adsorption isotherms of one of the adsor-
2.4.2. Numerical solution bents (LG700PA) was obtained at room temperature (25 C), using
The dynamic model (Eq. (1)) with the LDF approach (Eq. (7)) a bed height of 100 mm. H2S concentrations tested were 1980,
and a non-linear isotherm can be solved only by using a numerical 1065, 570 and 162 ppm, using nitrogen for diluting the mixture
method. Thus, they were implemented in and solved using the coming from the bottle with 2200 ppm of H2S. The ow rate of
commercial software package Aspen Adsorption V7.1 [16], in the gas mixture was always 1.1 L/min. Likewise, the amount of
which a bed of spherical particles with constant diameter equal adsorbent disposed inside the bed was 34 g in each test. To
to 2 mm (average size of the experimental particle size range) enhance the ow distribution of gases at the adsorbent bed inlet,
was specied. an inert bed of 50 mm height with glass spheres of 6 mm diameter
The methodology requires performing two steps. First, the iso- was located at the bottom of the adsorbent bed.
therm parameters of the H2S adsorption onto the adsorbent must
be tted using the adsorption capacity of the adsorbent at different 4. Results and discussion
H2S feed concentrations. The second stage consists of estimating
the overall mass transfer coefcient from the experimental break- Some of the tests performed in the previous experimental study
through curve obtained in the laboratory. This can be carried out at [3] were used to t the parameters of the models and then a com-
a given H2S feed concentration or using different values of concen- parison between experimental and simulation results for both the
trations. The isotherm was tted by the nonlinear least-squares BohartAdams (BA) model and the LDF model was made.
method called NL2SOL [17], which minimizes the nonlinear sum
of squares using analytic Jacobian. To estimate the overall mass 4.1. Results of the BohartAdams approach
transfer coefcient, the Upwind Differencing Scheme (UDS1) was
selected as discretization method (nite volume scheme) for solv- Fig. 1 shows ln (C0/Cb  1) were plotted versus time using exper-
ing the Eq. (1) with Eq. (7), which leads to an hyperbolic partial dif- imental data with one of the adsorbents selected (LL900A), at a bed
ferential equation, because UDS1 is an unconditional stable height of 70 mm and an H2S inlet concentration of 2036 ppm. By
method that reaches the convergence very quickly against other regression analysis, the slope and the origin ordinate were equal
schemes with the same number of nodes. to 0.207 and 9.874, respectively. The simple determination coef-
cient (r2) was 0.95. The model parameters tted were 67.10 m3/
3. Experimental kg-min and 3.27 kg/m3 for the BohartAdams (BA) rate constant
(kBA) and the adsorption capacity of the sorbent per volume unit
A high number of adsorption tests were carried out using a of the bed (N), respectively. With these parameters, the break-
xed-bed cylindrical column (30-mm ID) where the adsorbent through curves were simulated for different cases. Fig. 2 depicts

5
Ln( (C o/C )-1) = -0,2074t + 9,8741
4
Ln( (Co/C)-1)

0
20 25 30 35 40 45 50
Time (min)

Fig. 1. Experimental tting to the BA model for the adsorbent LL900A at a bed height of 70 mm and H2S inlet concentration of 2036 ppm.
F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315 309

500 L:7 cm; experimental for


fitting (Co=2036 ppm)

H2S outlet concentration (ppm)


L:7 cm; simulated by B-A
400 model

L:7 cm; experimental for


validation (Co=2036 ppm)
300
L:11 cm; simulated by B-A
model

200 L: 11cm; experimental


(Co=2036 ppm)

L:13cm; simulated by B-A


model
100
L:13cm; experimental
(Co=2274 ppm)

0
0 20 40 60 80 100 120
Time (min)

Fig. 2. Experimental and simulated breakthrough curves for the adsorbent LL900A, using the BH model at three bed-heights and two H2S feed concentrations.

Table 1
The same procedure was applied to other adsorbent (LF900A),
Experimental and simulated breakthrough time for the adsorbent LL900A. rst computing the parameters at a 70 mm bed height, and then
obtaining the simulated breakthrough curve at a bed height of
Breakthrough L = 70 mm L = 110 mm L = 130 mm
C0 = 2036 ppm C0 = 2036 ppm C0 = 2274 ppm
110 mm, using an H2S concentration of 2274 ppm. In this case,
the calculated N and kBA values were equal to 3.27 kg/m3 and
t200 Experimental (min) 35.8 64.0 69.4
t200 Simulated (min) 37.0 64.7 69.1
79.96 m3/kg min, respectively. Fig. 3 shows all the experimental
and model-based curves tested. The experimental and the simula-
tion results match quite well, especially in the steepest part of the
concentration breakthrough wavefront. However, the model does
the experimental breakthrough curves for the adsorbent LL900A not work well in the initial zone of breakthrough curve, at concen-
and those obtained using the BA model for different bed heights trations lower than 50 ppm, where the values of ln ((C0/Cb)  1)
(70, 110 and 130 mm) and two inlet concentrations (2036 ppm does not t to a straight line.
for the two rst bed heights and 2274 ppm when the bed height Additionally, Fig. 4 depicts the breakthrough data for a third
used was of 130 mm). The model was only tted from the experi- adsorbent (LG700 PA), at a bed height of 100 mm and an initial
mental data obtained at a bed height of 70 mm. Afterwards, the H2S concentration of 1980 ppm. In this case, the calculated N and
model was used for the other two bed heights, which matched well kBA values were equal to 5.58 kg/m3 and 10.18 m3/kg min, respec-
the experimental data. Furthermore, replicates were performed at tively. However, the simple determination coefcient (r2) was only
a bed height of 70 mm, and it can be observed that the experimen- 0.66, so the tting is worse regarding the other adsorbents. Now,
tal points are close to the model results. Therefore, the BA model the tting was extended beyond 500 ppm (upper limit in Figs. 2
suitably ts the breakthrough curves, providing breakthrough and 3), covering thus the complete range of concentrations up to
times similar to those experimentally obtained. Table 1 shows 1980 ppm (it must be taken into account that performing the t-
the experimental and model-based breakthrough times, using a ting to 500 ppm, r2 increased to 0.92). In addition, with the esti-
breakthrough concentration of 200 ppm. mated parameters, the breakthrough curve was simulated by

500
L:7 cm; experimental for
fitting
H2S outlet concentration (ppm)

400 L: 7 cm; simulated by B-A


model

L:7 cm; experimental for


validation - 1
300
L:7 cm; experimental for
validation - 2

200 L:11 cm; simulated by B-


A model

L:11 cm; experimental


100

0
0 10 20 30 40 50 60 70 80 90 100
Time (min)

Fig. 3. Comparison between experimental and simulated breakthrough curves for the adsorbent LF900A at two bed-heights (H2S feed concentration of 2274 ppm).
310 F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315

2000

Co=1980 ppm; for ng


1800
Co=1980 ppm; simulated by B-A model
Co=1980 ppm; for validaon
1600

H2S outlet concentration (ppm)


Co=1065 ppm; simulated by B-A model
Co=1065 ppm; experimental
1400
Co=570 ppm; simulated by B-A model
Co=570 ppm; experimental
1200

1000

800

600

400

200

0
0 50 100 150 200 250 300 350 400 450 500
Time (min)

Fig. 4. Experimental and simulated breakthrough curves for the adsorbent LG700PA with the BH model at a bed height of 100 mm and three H2S feed concentrations.

changing the H2S feed concentration (1065 ppm and 570 ppm), Table 2
and simulated curves did not match well the experimental curves. Parameters used in the Klinkenberg analytical solution.
Thus, e.g., at 570 ppm of inlet H2S concentration, for a break-
Parameter Value Unit
through concentration of 200 ppm, the BohartAdams equation
Particle density 2505.4 kg/m3
predicts a breakthrough time of 374 min. From the breakthrough
Bulk density 481.0 kg/m3
curve it is seen that the corresponding experimental breakthrough Bed void fraction 0.808
time is approximately 151 min. The predicted breakthrough time is Particle porosity 0.21
thus much greater than the observed breakthrough time. In this Bed length 0.10 m
case, in which the H2S feed concentration is changed, the model Bed diameter 0.03 m
Feed gas ow-rate 1.83E05 m3/s
still ts worse, as explained next. Indeed, the breakthrough curve
Particle diameter 0.002 m
is smoother at low concentrations than at high concentrations. This Pore diameter 6.52E09 m
is coherent with the assumption of a rectangular isotherm, which Tortuosity factor 3
may be a satisfactory approach for relatively large concentrations Gas density 1.1012 kg/m3
Gas viscosity 1.32E05 kg/m s
of strongly adsorbed H2S, such as was found in the previous exper-
imental study [3].

4.2. Results of the Klinkenberg analytical solution includes the parameters used. The value of tortuosity factor is
based on [18], and the rest of parameters were obtained from the
Fig. 5 shows the experimental and the theoretical breakthrough previous study [3] and from [19]. The equilibrium constants (Ke)
curves, using the analytical solution by Klinkenberg. Table 2 were computed by tting the experimental results of adsorption

2000
Co = 1980 ppm; Simulated
1800 by Klinkenberg Eq.

Co = 1980 ppm;
1600
H2S outlet concentration (ppm)

experimental

1400 Co = 1065 ppm; Simulated


by Klinkenberg Eq.

1200 Co = 1065 ppm;


experimental

1000
Co = 570 ppm; Simulated
by Klinkenberg Eq.
800
Co = 570 ppm;
experimental
600
Co = 162 ppm; Simulated
400 by Klinkenberg Eq.

Co = 162 ppm;
200 experimental

0
0 50 100 150 200 250 300 350
Time (min)

Fig. 5. Experimental and simulated breakthrough curves for the adsorbent LG700PA with the Klinkenberg solution at a bed height of 100 mm and four H2S feed
concentrations.
F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315 311

1,2
Log (q e ) = 0,6036 Log( C ) - 0,9834
2
R = 0,9983
1

0,8

Log qe
0,6

0,4

0,2

0
1 1,5 2 2,5 3 3,5
Log C

Fig. 6. Fit of the experimental data to a Freundlich isotherm for the adsorbent LG700PA.

isotherm to the linear isotherm: 8262.4, 11245.9, 13726.2 and Additionally, it was carried out an experimental data tting
22721.4 m3 gas/m3 ads corresponding to a H2S feed concentration using the Langmuir isotherm (Fig. 7), but it was worse than that
of 1980, 1065, 570 and 162 ppm, respectively, at which the overall for Freundlich. The parameters entered the simulator were
mass transfer coefcient was 5.55  103, 4.08  103, 4.41E04 kmol/kg and 20825.09 m3/kmol.
3.34  103, 2.02  103 s1. The external mass transfer coefcient By selecting the Freundlich isotherm, the parameters above
was 4.81  102 m/s. mentioned were further tted using the experimental data. The
In Fig. 5, it can be observed that the shapes of the H2S concen- nal values of the two parameters were 0.0688 m3/kg and
tration breakthrough front at different H2S feed concentration are 0.5841, respectively, with a tolerance of 106. Fig. 8 shows the
not very similar with those of experimental breakthrough curves. experimental isotherm and the isotherm obtained with the param-
The larger slope in the beginning of the real breakthrough curves eters calculated by regression and computed by simulation.
suggests a higher mass-transfer rate than that in the end, where
the small slope indicates that adsorption rate becomes very small 4.4. Dynamic estimation of the overall mass transfer coefcient
when adsorbent approaching to equilibrium. In fact, the break-
through curve prediction from Klinkenberg model is symmetrical, The overall mass transfer coefcient was estimated with the
and this approach cannot give a good t at different H2S feed con- experimental conditions for the adsorbent LG700PA, i.e., a biogas
centrations, especially at high concentrations. ow-rate of 1.1 L/min, a xed bed of 30 mm diameter and
100 mm height, a bed void fraction of 0.808 and an apparent den-
4.3. Fitting of the adsorption isotherms sity of 481 kg/m3. The H2S feed concentration was equal to the
experimental values used for obtaining the breakthrough curve.
The aim of measuring the adsorption isotherm was to simulate Results from the different estimates of the mass transfer coef-
the system tested in the laboratory more accurately in a nonequi- cient (Ka) from experimental breakthrough fronts at several con-
librium model that includes mass transfer resistances. centrations are shown in Table 4. These coefcients were
To t the Freundlich equation, the natural logarithm of the included in the LDF model (Eq. (7)) that, along with Eq. (1) and
adsorption capacity was graphically represented versus the natural the tted Freundlich isotherm, allowed calculating the travelling
logarithm of concentration, and by linear regression analysis, the front or wavefront.
isotherm parameters were obtained (Fig. 6). The adsorption capac- It should be noted that the overall mass transfer coefcient (Ka)
ity obtained from the experimental breakthrough curves for the increases as the H2S feed concentration increases, which at rst
adsorbent LG700PA was 9.91, 7.26, 4.74 and 2.23 mg H2S/g adsor- may seem strange. However, previous researchers [20,21] pro-
bent at 1980, 1065, 570 and 162 ppm of H2S feed concentration, posed a correlation for the prediction of the overall LDF rate con-
respectively. The isotherm equation for the adsorbent LG700PA is stant considering the value of q at equilibrium (qs) with feed
the following: concentration (C0), so Ka would be multiplied by the ratio C0/qs,
approximately. Using the Freundlich isotherm, this implies an
qe 0:1039 C 0:6036 17
increase in Ka when increasing the H2S feed concentration, as
As a rst step to apply and solve the Eqs. (1) and (7) with the aforementioned.
adsorption isotherm, these parameters were used as inputs in The predicted curves are close to those experimentally
Aspen Adsorption, in suitable units (Table 3). The two parameters obtained, except at 1065 ppm of H2S feed concentration, which
to be entered were 0.0835 m3/kg and 0.6036. may be due to experimental errors, since a value of Ka higher than
the obtained was expected, following the same trend as for other
Table 3
concentrations.
Adsorption capacities for the adsorbent LG700PA at different H2S feed concentrations.
Additionally, Fig. 9 illustrates the breakthrough curves in rela-
Concentration (mol H2S/m3) Adsorption capacity (mol/kg) tive terms (outlet concentration-to-inlet concentration ratio versus
8.8393  102 0.2915 the treated biogas volume) to better observe that the adsorbent
4.7545  102 0.2134 can treat a smaller volume of biogas as the feed concentration
2.5446  102 0.1394
increases. This is because the overall driving-force increases as
7.2321  103 0.0656
the H2S feed concentration increases. A higher pore and/or external
312 F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315

250
C /q e = 0,0668 C + 71,817
2
R = 0,9715
200

150

C/qe
100

50

0
0 500 1000 1500 2000 2500
C

Fig. 7. Fit of the experimental data to a Langmuir isotherm for the adsorbent LG700PA.

3,5E-04

3,0E-04
Adsorption capacity (kmol/kg)

2,5E-04

2,0E-04
Experimental

1,5E-04 Calculated

Simulated
1,0E-04

5,0E-05

0,0E+00
0,E+00 2,E-05 4,E-05 6,E-05 8,E-05 1,E-04
Concentration (kmol/m3)

Fig. 8. Experimental isotherm and Freundlich isotherm (computed and simulated) for the adsorbent LG700PA.

Table 4 overall mass transfer coefcient for all the curves (that obtained
Mass transfer coefcients estimates for the adsorbent LG700PA at different H2S feed at 570 ppm and coded by adding r1 in Fig. 10), i.e., a value of
concentrations by Aspen Adsorption. the mass transfer coefcient was tted from a breakthrough curve,
Concentration (ppm) Mass transfer coefcient (s1) and the simulated wavefronts barely changed. Probably, this is
1980 1.92  103 because the internal mass-transfer is the controlling step (using
1065 1.17  103 the Klinkenberg equation, the ratio between the internal and the
570 1.58  103 external mass-transfer resistances is about 3). It is observed that
162 1.15  103 the predictions matches well the experimental data, although the
nal slight curvature exhibited in the experimental breakthrough
deviates from the prediction, which may suggest that a more rigor-
lm diffusion rate will result in a steeper concentration wavefront, ous model than the LDF approximation should be taken into
which can be observed in the breakthrough curves. account to fully describe the dynamics of adsorption. However, it
can be concluded that the LDF model with a single lumped mass
transfer coefcient accounting for the non-linearity of the isotherm
4.5. Results of the LDF model with a non-linear isotherm provides an adequate approximation of the adsorption process.
These results give lower values of Ka as compared to those obtained
Fig. 10 illustrates the simulated breakthrough curves and those using the Klinkenberg solution and, however, the curves analyti-
experimentally obtained, at four inlet concentrations. Only 20 cally predicted (Fig. 5) seems to require a still higher value for Ka
nodes were used in the discretization, since more nodes gave the in order to make steeper them. This is not realistic, and it is prob-
same result. Additionally, the t was performed using the same ably due to the use of the linear isotherm.
F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315 313

0,9

0,8

0,7

0,6

C/Co
0,5

0,4
Co= 1980 ppm
0,3
Co=1065 ppm
0,2 Co=570 ppm

0,1 Co=162 ppm

0
0 50 100 150 200 250 300 350 400

Treated volume (L)

Fig. 9. Experimental breakthrough curves in relative terms for the adsorbent LG700PA at four H2S feed concentrations.

2000

1800
Co=1980 ppm;
Aspen simulated

1600 Co=1980 ppm;


H2S outlet concentration (ppm)

Aspen simulated_r1
Co=1980 ppm; for
validation
1400
Co=1065 ppm;
Aspen simulated
1200 Co=1065 ppm;
Aspen simulated_r1
Co=1065 ppm; for
1000 validation
Co=570 ppm; Aspen
simulated_r1
800 Co=570 ppm; for
validation
Co=162 ppm; Aspen
600 simulated
Co=162 ppm; Aspen
simulated_r1
400 Co=162 ppm; for
validation

200

0
0 50 100 150 200 250 300 350 400
Time (min)

Fig. 10. Experimental and Aspen-Adsorption-simulated breakthrough curves for the adsorbent LG700PA at four H2S feed concentrations.

Against the BohartAdams model that uses a rectangular iso- are the residence time (L/u) and particle size (dp). This latter is the
therm and assumes the surface-rate as limiting step, simulations same as in the lab scale, i.e., 2 mm on average. A higher residence
were carried out using the linear driving force (LDF) model which time may be achieved by increasing the height of the column,
take into account the experimental adsorption isotherm (Freund- while keeping the supercial velocity through the column. This
lich equation, in this case) and mass transfer process to control way, the overall mass transfer coefcient should be the same in
the adsorption process. It can be concluded that the tting was both scales, although the supercial velocity is not too relevant
much better using the LDF model, with the aid of Aspen Adsorp- since the main resistance to mass transfer is due to the internal
tion, as shown in Figs. 4 and 10, where the experimental break- solid diffusion. Thus, the supercial velocity through the bed depth
through curves are compared to the simulations performed using was chosen according to the lab results (0.026 m/s). Moreover,
both the BohartAdams method and the LDF model, respectively, the Reynolds number is low (Re < 40) so the viscous term of
at different H2S feed concentrations. Therefore, the LDF model that the Erguns law is dominant to compute the pressure drop in the
includes the effects of mass transfer rate with a nonlinear isotherm bed. Besides, to have near plug ow in the lab bed, the particle
results in a theoretical representation that more closely approxi- Peclet number (related to the ratio L/dp) should be high enough
mates the adsorption process studied in this paper. to cause a high bed Peclet number (related to the ratio L/Di),
whereas in a larger scale unit the particle Peclet number is of min-
4.6. Scale-up assessment imal importance, since dp is the same but both L and Di are much
greater, and thus the bed Peclet number is expected to be high
Finally, in order to apply the prediction of the breakthrough enough. However, since a scaling up is being performing, disper-
curve by the LDF model with the Freundlich isotherm, a scale-up sion effects might have to be considered in the lab scale modeling,
was performed and assessed. In xed beds, the critical parameters and this is a point for a future work.
314 F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315

Co:1065 ppm; D:0.37m; H:3m; v:0.026m/s Co:1065 ppm; D:0.117m; H:3m; v:0.26m/s
Co:1980 ppm; D:0.37m; H:3m; v:0.026m/s Co:1980 ppm; D:0.117m; H:3m; v:0.26m/s
2000

1800

H2S outlet concentration (ppm)


1600

1400

1200

1000

800

600

400

200

0
0 10 20 30 40 50 60 70 80 90
Time (h)

Fig. 11. Simulated breakthrough curves for scaling-up assessment for the adsorbent LG700PA at four H2S feed concentrations.

A scale up to 10 m3/h was established, which involves an inter- The BohartAdams model could be used as an acceptable
nal diameter (Di) of 0.37 m. Additionally, a bed depth (H) of 3.0 m approach to determine the breakthrough time, if the breakthrough
was chosen. Fig. 11 illustrates the breakthrough curve at two H2S concentration is within the sharper zone of the breakthrough
feed concentrations for this case using the lab scale supercial curve. However, at the initial and nal zones of this curve, the
velocity (0.026 m/s). Moreover, it was also scaled up using a higher curve is less steep and the extrapolation to other heights does
supercial velocity (0.26 m/s), keeping in mind to reduce the unit not work. Additionally, when varying the H2S concentration at
size and, hence, the space needed. Although, this higher velocity the xed-bed inlet, this method no longer give a good estimate
implies an internal diameter of 0.117 m, with this option, the mass of the breakthrough curves, due to the simplication of considering
is also reduced to the tenth part (using the same bed depth), so it is a rectangular isotherm (irreversible). Therefore, the use of this
expectable that the lifetime of the bed is equally reduced, as shown model must be taken with care.
in Fig. 11. The number of beds and the space required would have Beyond the relatively simple modeling, such as the Bohart
to be larger, and the pressure drop would be quite higher (about Adams approach (or the Thomas model), it is necessary to obtain
120 Pa), so the use of a high velocity is not recommendable. better solutions for the breakthrough curves by the rigorous model
By specifying a breakthrough concentration of 200 ppm, the bed equations that take into account an isotherm and mass transfer
lifetime (breakthrough time) would be 45 and 60 h at 1980 and resistances. The LDF model is perhaps the simplest concept to
1065 ppm, respectively. The lifetime of the adsorbent increases describe mass transfer while maintaining a sensible physical basis.
as the H2S feed concentration decreases. The pressure drop is lower The analytical solution requires a linear isotherm that is not realis-
than 10 Pa, computed by using the Ergun equation, so the power tic and then the breakthrough curves obtained deviate from the
consumption would not be signicant. However, the adsorbent experimental, especially at higher H2S feed concentrations. Thus,
load is 155 kg. If the estimated cost due to dehydration and ther- dynamic simulation of the adsorption process was carried out with
mal drying of the sewagesludge was assumed to be of the same the aid of Aspen Adsorption V7.1, which makes it possible to esti-
magnitude order of that computed for the thermal treatment from mate the overall mass transfer coefcient numerically and solve
the data obtained in the previous work [3], which was 0.27 /kg of the LDF model when using a nonlinear adsorption isotherm. The
fresh sludge, using an electricity cost of 0.12 /kWh [22], the total Freundlich isotherm tted quite well to the experimental data. This
cost of the adsorbent would be somewhat lower than 0.60 /kg, way, this modeling can provide a reasonable prediction of the
ve times cheaper than activated carbon on average. However, breakthrough curves, and hence, be used for scaling up. A scale-
due to its short lifetime predicted in the scale up, and in order to up assessment from lab scale to a larger scale with a ow rate
apply it in a full scale, the need of adsorbent would be huge so a about 150 times higher was performed, and it was veried that a
regeneration method should be studied to enlarge the use of the regeneration of the adsorbent is required.
adsorbent. This is currently under study and it will be included
in a future paper.
References
5. Conclusions [1] N. Abatzoglou, S. Boivin, A review of biogas purication processes, Biofuels
Bioprod. Bioren. 3 (2009) 4271.
For the design of a xed-bed adsorber, a dynamic model of [2] T.J. Bandosz, K. Block, Effect of pyrolysis temperature and time on catalytic
performance of sewage sludge/industrial sludge-based composite adsorbents,
adsorption helps to calculate the height required, operating life- Appl. Catal. B: Environ. 67 (2006) 7785.
time and adsorption capacity of the bed. This paper analyses the [3] F.J. Gutirrez Ortiz, P.G. Aguilera, P. Ollero, Biogas desulfurization by
dynamic response of different adsorbents, but especially one of adsorption on thermally treated sewagesludge, Sep. Purif. Technol. 123
(2014) 200213.
them (coded as LG700PA), with a relatively high adsorbent capac-
[4] V. Srivastava, B. Prasad, I. Mishra, I. Mall, M. Swamy, Prediction of
ity based on a previous experimental study [3]. The breakthrough breakthrough curves for sorption removal of phenol by bagasse y ash
curve was steeper at higher H2S feed concentrations, but its shape packed bed, Ind. Eng. Chem. Res. 47 (2008) 16031613.
barely changed with bed height and it was moved almost parallel, [5] K.H. Chu, Fixed bed sorption: setting the record straight on the BohartAdams
and Thomas models, J. Hazard. Mater. 177 (2010) 10061012.
which is reasonable since the mass of adsorbent varied proportion- [6] R.T. Yang, Gas Separation by Adsorption Processes (Series on Chemical
ally, and hence, the residence time. Engineering: vol. 1), Imperial College Press, London, 1987.
F.J. Gutirrez Ortiz et al. / Chemical Engineering Journal 253 (2014) 305315 315

[7] K.H. Chu, X. Feng, E.Y. Kim, Y.-T. Hung, Biosorption parameter estimation with [16] Aspen Technology, Inc. <http://www.aspentech.com/products/aspen-
genetic algorithm, Water 3 (2011) 177195. adsim.aspx>.
[8] D.O. Cooney, Adsorption Design for Wastewater Treatment, CRC Press, Boca [17] J. Dennis, D. Gay, R. Welsch, An adaptive nonlinear least-squares algorithm,
Raton, 1999. ACM T. Math, Software 7 (1981) 367383.
[9] G.S. Bohart, E.Q. Adams, Some aspects of the behavior of charcoal with respect [18] J.M. Zalc, S.C. Reyes, E. Iglesia, The effects of diffusion mechanism and void
to chlorine, J. Am. Chem. Soc. 42 (1920) 523544. structure on transport rates and tortuosity factors in complex porous
[10] H.C. Thomas, Heterogeneous ion exchange in a owing system, J. Am. Chem. structures, Chem. Eng. Sci. 59 (2004) 29472960.
Soc. 66 (1944) 16641666. [19] R.H. Perry, D.W. Green, J.O. Maloney, Perrys Chemical Engineers Handbook,
[11] J.H. Tsai, F.T. Jeng, H.L. Giang, Removal of H2S from exhaust gas by use of seventh ed., Mc-Graw Hill, 1997.
alkaline activated carbon, Adsorption 7 (2001) 357366. [20] S. Farooq, D.M. Ruthven, Heat effects in adsorption column dynamics. 2.
[12] Y.B. Zeldovich, On the theory of Freundlich adsorption isotherm, Acta Experimental validation of the one-dimensional model, Ind. Eng. Chem. Res. 29
Physicochom. (1935) 961974. (1990) 10841090.
[13] A. Klinkenberg, Numerical evaluation of equations describing transient heat [21] A. Malek, S. Farooq, Comparison of isotherm models for hydrocarbon
and mass transfer in packed solids, Ind. Eng. Chem. 40 (1948) 19921994. adsorption on activated carbon, AIChE J. 42 (1996) 31913201.
[14] W.E. Ranz, W.R. Marshall, Evaporation from drops. Part II, Chem. Eng. Prog. 48 [22] <http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/Electricity_
(1952) 173180. and_natural_gas_price_statistics>.
[15] E.N. Fuller, P.D. Schettler, J.C. Giddings, A new method for prediction
coefcients, Ind. Eng. Chem. 58 (1966) 1827.

You might also like