Chemical Engineering Science 54 (1999) 4031}4040

Estimation of overall e!ective coe$cient of heat transfer for

nonisothermal "xed-bed adsorption
Wencai Lin, S. Farooq*, Chi Tien
Department of Chemical & Environmental Engineering, National University of Singapore, Singapore 119260, Singapore
Received 20 October 1998; received in revised form 5 February 1999; accepted 10 February 1999

Abstract

One- and two-dimensional models for nonisothermal "xed-bed adsorption have been numerically solved. Concentration and
temperature pro"les obtained from the two-dimensional model are presented. By matching the exit concentration and temperature
pro"les from the two models, an empirical correlation has been proposed which allows estimation of the overall e!ective coe$cient of
heat transfer for the one-dimensional model.  1999 Elsevier Science Ltd. All rights reserved.

Keywords: Fixed-bed; Adsorption; Nonisothermal; Coe$cient

1. Introduction model may be made to match the column dynamics

Fixed-bed adsorption is widely used in the separation
of gaseous mixtures and its dynamics plays a key role in
the design and control of the processes. A number of
models have been developed to study the dynamics of
a nonisothermal adsorption bed (Ruthven et al., 1975;
Kaguei et al., 1989; Farooq and Ruthven, 1990a,b; Sun
and Meunier, 1991; Silva and Rodrigues, 1997). Most of
these models are one-dimensional in which the radial
temperature and concentration distributions are neglect-
ed (Ruthven et al., 1975; Sun and Meunier, 1991; Silva
and Rodrigues, 1997). Kaguei et al. (1989) proposed
a two-dimensional model by taking into account the
radial thermal conduction and showed that the heat of
adsorption obtained by one-dimensional analysis of
column breakthrough results is signi"cantly di!erent
from that obtained by two-dimensional analysis. A cor-
relation was proposed to relate the two heats of adsorp-
tion obtained from the two models. Farooq and Ruthven
(1990a) developed a two-dimensional nonisothermal
model for the idealized conditions of a linear equilibrium
isotherm and negligible "lm resistance to heat transfer at
the column wall. Even with these extreme approxima-
tions (which maximized the e!ect of radial temperature
gradient) it was shown that by adjusting the overall
e!ective coe$cient of heat transfer the one-dimensional
*Corresponding author. E-mail: chest@nus.edu.sg
predicted by the two-dimensional model under condi-
tions leading to the formation of a combined temperature
and concentration wave. Although such a match was not
possible under the conditions of pure thermal wave
formation, it was argued that in practical systems, favor-
able equilibrium and signi"cant "lm resistance to heat
transfer will render the one-dimensional model adequate,
regardless of the relative velocities of the tempera-
ture and concentration fronts. This conclusion was
experimentally veri"ed in a later study (Farooq and
Ruthven, 1990b).
It is clear from the above studies that the most crucial
parameter for the success of one-dimensional analysis is
the overall e!ective heat transfer coe$cient which is
normally obtained by "tting the experimentally mea-
sured temperature pro"le at the column exit with the
model solution (Kaguei et al., 1985; Malek and Farooq,
1996). Available correlations for estimating the para-
meter (that includes the e!ect of radial temperature
gradient) were mostly developed under steady-state con-
ditions and do not work well for transient thermal wave
propagation in an adsorption bed. In this study, a general
two-dimensional model for nonisothermal "xed-bed ad-
sorption has been numerically solved covering a wide
spectrum of possible operating conditions and physical
properties of the adsorption system. It is recon"rmed
that the concentration and temperature pro"les at the
column exit calculated from the more appropriate two-
dimensional analysis may be very closely matched with

0009-2509/99/$ - see front matter  1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 9 9 ) 0 0 1 0 9 - 8
4032 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040

a one-dimensional model using the overall e!ective heat where the dependence of equilibrium constant b on
transfer coe$cient as a "tting parameter. A correlation temperature is described by the Arrhenius law
has been proposed that represents well the observed


dependence of the overall e!ective heat transfer coe$c- !DH
b"b exp . (3)
ient on all the variables investigated and may now be  R

used for its a priori estimation.
10. The temperature dependence of gas and solid proper-
ties is assumed negligible.
2. Theory

The system considered is a "xed adsorption bed, which 3. Two-dimensional model

undergoes a step change in adsorbate concentration at
the inlet from a steady inert carrier gas #ow. After a peri- Subject to the above assumptions, the following equa-
od of time, the concentration breakthrough is attained tions constitute the two-dimensional model:
and the outlet concentration becomes equal to that of the Adsorbate mass balance:
inlet when the adsorbent is completely saturated with



*C * 1 * *C * *C
respect to the feed. The transient material mass and I# (;C )" x D I # D I
*t *x I x *x  0 *x *x  *x
energy equations are necessary to simulate this process.      
The following assumptions are made: (1!e ) *q
! o . (4)
1. Ideal gas law holds. e Q *t
2. E!ective thermal conductivities of the commercial
In order to consider the total gas concentration change
adsorbent particles are relatively high, and therefore
due to a change in temperature it is necessary to include
intraparticle temperature gradients can be neglected.
inert gas mass balance equation
Thermal equilibrium is assumed between the #uid and



the adsorbent particles, which is also a very common *C * 1 * *C * *C
assumption in adsorber calculations (Yang, 1987). A# (;C )" x D A # D A .
*t *x A x *x  0 *x *x  *x
3. A constant temperature is maintained at the column      
wall. (5)
4. Velocity distribution in the radial direction is
neglected. Accumulation of inert gas in the macropore voids is
5. The frictional pressure drop is assumed to be neglected in above balance.
negligible. Thermal-energy balance:
6. Both axial and radial dispersion are taken into ac-
 
* *
count in the component material balance equations of e o c #(1!e )o c #e (;o c )
*t E NE Q NQ *x E NE
the two-dimensional model. Only axial dispersion is 
considered in the one-dimensional model.



* * * *
7. Bulk #ow of heat and thermal dispersion in the axial " x K # K
x *x  0 *x *x  *x
direction is common in both the models. In addition,     
the two-dimensional model explicitly takes into ac- *q
count the radial thermal dispersion and a heat transfer #(1!e )o (!DH ) I. (6)
Q  *t
coe$cient con"ned at the wall. In the one-dimen-
sional model, heat transfer in the radial direction is Since the radial velocity distribution is neglected, the
lumped into a single overall e!ective coe$cient. velocity change in the axial direction due to adsorption
8. A linear driving force (LDF) model is used to describe can be obtained by the continuity equation
the adsorption rate
*o *(o ;) *q
*q e E#e E "!(1!e )o M . (7)
(1) *t *x Q *t
*t
"k(q*!q) 
In the axial direction, the Danckwerts boundary con-
and the mass transfer rate constant is assumed to be ditions are applicable for the mass and heat balance
independent of temperature. equations
9. The equilibrium isotherm is represented by the Lan-

 
gmuir isotherm equation *C *C
D I "; (C " !C " ), I "0
 *x D I V> I V\ *x
q* bP y  V  V*
" 2 , (2)
q* 1#bP y at t'0, (8a)
Q 2
 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040 4033

 

*C *C *CM *(;M CM ) 1 * *CM 1 *CM
D A "; (C " !C " ), A "0 I# I" xN I # I
 *x D A V> A V\ *x *h *xN Pe *xN  *xN Pe *xN 
 V 
  V *
  0    
at t'0, (8b) !
ab(qN *!qN ), (4a)


K "e ; o c (" !" ), *CM *(;M CM ) 1 * *CM 1 *CM
 *x D E NE V> V\ A# A" xN A # A (5a)
 V *h *xN Pe *xN  *xN Pe *xN ,
 0    

*
"0 at t'0. (8c) * *
*x [(oN #j)M ]# (oN ;M M )
 V* E E
*h *xN
For the velocity equation 


1 * *M 1 *M

*; " xN # #b
uu(qN *!qN ),
;"
V
"; ,
D *x
"0 at t'0. (9) Pe *xN  *xN Pe *xN 
 V* 0F   F 
(6a)
In the radial direction
*oN *(oN ;M )
 
*C *C E# E "!oN q*M
b(q*!qN ).
I " I (10a) Q Q
(7a)
*x *x
"0, *h *xN
 V 
  V 0
 

 
*C *C The dimensionless boundary conditions are:
A " A "0, (10b)
*x *x
 V  V0
 
*CM *C
I "Pe (CM " N M "N I
 
* * a >!C \), "0
"! U (! ). (10c) *xN N  I V   I V 
 *xN N
*x
"0,
*x K 5  V  V
 V  V0 0
at h'0, (8a)
The following initial conditions apply for the adsorber
prior to introducing the concentration step change at the
 
*CM *CM
inlet: A "Pe (CM " N !CM " N ), A "0
*x N  A V> A V\ *xN N
 V  V
C (t, x , x )"0 at t)0 for 0)x )R and
I    at h'0, (8b)
0)x ), (11a)


 
C (t, x , x )"C at t)0 for 0)x )R and *M *M
A   A  "Pe (" >!" \), "0
*xN N F V 
 V 
 *xN N
0)x ), (11b)  V  V

at h'0, (8c)
(t, x , x )" at t)0 for 0)x )R and
   


0)x ), (11c) *;M
 ;M " N "1, "0 at h'0, (9a)
V *xN N
q (t, x , x )"0 at t)0 for 0)x )R and  V
I   

 
0)x ), (11d) *CM *CM
 I " I "0, (10a)
*xN N *xN N
;(t, x )"; at t)0 for 0)x ). (11e)  V  V
 D 

 
The following dimensionless variables were used: *CM *CM
A " A "0, (10b)
*xN N *xN N
C C q ;  V  V
CM " I , CM " A , qN " , M " , ;M " , q" ,
I C A C
 
ID AD
q*
Q


;
D
;
D *M *M
"0, "! tPe (M !1). (10c)
o *xN N *xN N R 0F
t x x
h" , xN "  , xN " , oN " .  V 
  V
q  R  o
E The dimensionless initial conditions are:
The model equations, together with the initial and
CM (h, xN , xN )"0 at h)0 for 0)xN )1
boundary conditions [Eqs. (1)}(11)], take the following I   
dimensionless forms:
and 0)xN )1, (11a)

*qN
"b(qN *!qN ), (1a) C
*h CM (h, xN , xN )" A at h)0 for 0)xN )1
A   C 
hC exp(u) AD
qN *" I , (2a)
1#hC exp(u) and 0)xN )1, (11b)
I 
4034 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040

M (h, xN , xN )"1 at h)0 for 0)xN )1 Table 1

   Heat transfer correlations for #uid and sound phases
and 0)xN )1, (11c)



d 
qN (h, xN , xN )"0 at h)0 for 0)xN )1 Bi
Q
Bi "2.41#0.156
Q
!1
I    d
N
k 2
and 0)xN )1, (11d) k /k PQ"(1!e
 PQ D k
D
1!(k /k )B
D Q
;M (h, xN )"1 at h)0 for 0)xN )1.
 
(11e) (1!(k /k ))B k B#1 B!1
  ; D Q ln Q ! !
(1!(k /k )B) k B 2 1!(k /k )B
D Q D D Q


1!e 
where B"1.25
e
4. One-dimensional model
Nu Nu "2.0#1.1Pr  Re 
DQ DQ
As for the one-dimensional model, the material and


d
Nu Nu "0.523 1! N Pr  Re 
thermal-energy balance equations take the following UD UD d
forms: Pe
D
1 0.73e 0.5
" #
Pe Re Pr (1#(9.7e /Re Pr))
*CM *(;M CM ) 1 *CM D
I# I" I!
ab(qN *!qN ), (12) 1 0.67e 1
*h *xN Pe *xN  Pe " #
   0D Pe
0D
Re Pr 7(2!(1!2d /d ))
N
*CM *(;M CM ) 1 *CM
A# A" A, (13)
*h *xN Pe *xN 
  
* * where
[(oN #j)M ]# (oN ;M M )
*h E *xN E (Re Pr)/Pe
 b " 0D (17b)
D (8/N )#((Bi #4)/Bi )
1 *M D D D
" #b
uu(qN *!qN )
Pe *xN  and
F 
2 h 1.5(1!e )(d /d )
! 5 (M !1) (14) N " N . (17c)
R e o c ; D
 
E NE D 1 0.1
Re Pr/Pe #
with the same boundary conditions in the axial direction 0D Nu k /k
DQ Q D
and initial conditions as the two-dimensional model.
Bi , k /k , Nu and Pe appearing in Eqs. (17a)}(17c)
Q PQ D DQ 0D
were calculated from yet another set of available correla-
tions (Derkx and Dixon, 1997; Dixon, 1985) and these
5. Estimation of parameters are listed in Table 1. The dimensionless parameter
Bi ("(a R /k )) requires a and k which were ob-
Axial and radial dispersion coe$cients in the adsorber D UD 0D UD 0D
tained from the correlations for Nu and Pe . The
were determined from the following correlations pro- UD 0D
correlation for Nu is also given in Table 1.
posed in the literature (Tien, 1994; Gunn, 1987): UD
The following correlations also given in the same liter-
Axial dispersion coe$cient: ature (Derkx and Dixon, 1997; Dixon and Cresswell,
1 0.3 0.5 1979) were used to determine the e!ective axial and
" # . (15) radial thermal conductivities.
Pe Re Sc 1#3.8/(Re Sc)

1 1 (k /k )
Radial dispersion coe$cient: " # PQ D , (18a)
Pe Pe Re Pr+1#(1/N ) (8Bi /(Bi #4)),
1 1 e F D Q Q Q
" # (16) where
Pe Pe qRe Sc
0 D
where Pe "40!29 exp(!7/Re) and q"1.2 for 1.5(1!e )(d /d )
D N" N , (18b)
spheres. Q k /k [(1/Nu )#(0.1/k /k )]
PQ D DQ Q D
The "lm heat transfer coe$cient at the adsorber wall, K k Bi #4
0" PQ#b Q . (19)
a , was estimated from the following correlation given by k k D Bi
U D D Q
Dixon and Cresswell (1979) and Derkx and Dixon (1997):
K appears in the group Pe . As discussed earlier, Pe
0 0F D


a d 8b k d b k /k , Bi and Nu appearing in Eqs. (18a), (18b) and (19)
Nu " U N" D #2Bi PQ N 1# D , (17a) PQ D Q DQ
U k d /d Qk d k /k were calculated from the correlations listed in Table 1.
D N D PQ D
 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040 4035

Dixon (1985) also proposed the following relation be-

tween the overall e!ective heat transfer coe$cient, h , of
U
the one-dimensional model and the "lm heat transfer
coe$cient, a used in the two-dimensional model:
U
1 1 R
" # , (20)
h a lK
U U 0
where the wall and the bed resistances have been linearly
added. An adjustable parameter, l, has been introduced
Fig. 1. E!ect of convergence criteria on the numerical results: (a)
with the linearized bed resistance term to account for the
temperature pro"le; (b) concentration breakthrough. (Relative residual
e!ect of radial temperature distribution. l varies from () 0.00001; () 0.00005; () 0.0005).
3}6 depending on the operating conditions in a nonad-
sorbing "xed bed.
The dimensionless form of Eq. (20) is

1 1 Pe
" # 0F. (20a)
g t R l

Substituting Eq. (20a) in Eq. (14), gives

* *
[(oN #j )M ]# (oN ;M M )
*h E Q *xN E
 
1 *M 2g Fig. 2. E!ect of grid size on the numerical results: (a) temperature
" #b
uu(qN *!qN )! (M !1). (14a)
Pe *xN  R pro"le; (b) concentration breakthrough. (1) 10;100; (2) 15;200; (3)
F  20;250.

6. Numerical implementation
Several methods for the reduction of two-dimensional
packed-bed model to one-dimensional model have been
reviewed by Dixon (1996). Finlayson (1971) simpli"ed
a two-dimensional steady-state model for heat transfer in
a packed bed to a one-dimensional model by using one-
point collocation and concluded that only for the case of
very small tube Biot number ("a R/K ) the one-term
U 0
approximation is reasonable and the one-dimensional
model is applicable. Similar model reduction technique
was not applied in our study due to model nonlinearity
coming from adsorption equilibrium isotherm.
Eqs. (1a), (2a), (4a), and (5a)}(7a) constitute the two-
dimensional model and Eqs. (1a), (2a), (7a), and (12), (13)
and (14a) constitute the one-dimensional model. These
dimensionless equations were discretized in space using
"nite di!erence (control-volume) technique and the
power-law scheme (PLS) was adopted for discretization.
The resulting ordinary di!erential equations were integ-
rated in time using an implicit scheme (Patankar, 1980).
The values of the dependent variables, concentration,
temperature and velocity obtained from the one-dimen-
sional model are the average values at a given location
and the values at the exit as a function of time give the
so-called breakthrough curves. In the case of two-dimen-
sional model, radial variation of velocity was neglected
but concentration and temperature were allowed to vary
radially. Hence the concentration and temperature
breakthrough from the two-dimensional analysis repre-
sented the mixing cup values of these variables calculated
according to the following integral:
0 rf (r) dr
 ,
0 r dr

where f (r) represents the radial concentration or temper-
ature pro"le.
It can be seen from the computed results in Fig. 1 that
the computation is converged when the relative residuals
are less than 0.5;10\. Hence a relative residual of
0.5;10\ or less between two iterations was used as the
convergence criterion. Three grid sizes, 10;100,
15;200, 20;250, were chosen to study the e!ect of grid
size on the numerical results. Based on the results shown
in Fig. 2, a grid number of 20;250 were chosen in most
of the simulations. Finer grids were used for those para-
meter values, which gave very sharp wave fronts.
7. Results and discussion
In order to verify that the correlations chosen in this
study for estimating the heat and mass transfer para-
meters of the two-dimensional model are realistic, we
compared the simulation results with the experimental
results for methane breakthrough results in a "xed bed of
activated carbon. The experimental, equilibrium, kinetic,
4036 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040

Table 2 Table 3
Experimental conditions, equilibrium, and heat and mass transfer para- Equilibrium, kinetic and other physical parameters for the base case
meters for methane adsorption on activated carbon*
; "0.0199 m/s, P "2.4828 bar y "0.225
D 2 D
"0.4 m R "0.0175 m e "0.3994 c "1729.5 J/(kg K) k "0.079 J/(m s K) k"0.016/s
NI D
dp"2.58 mm c "950 J/(kg K) o "870 kg/m (!DH )"26440 J/mol q*"3.7843 mol/kg b "4.11;10\/bar
NQ Q  Q 
; "0.0156 m/s, "299.15 K, y "0.05 D "1.18 cm/s D "1.31 cm/s K "0.31 J/(m s K)
D  D  0 
P "2.4848 bar c "2257 J/(kg K) c "5193 J/(kg K) K "0.33 J/(m s K) a "536 W/(m K)
2 NI NA 0 U
k "0.95 J/(m s K) k "0.137 J/(m s K)
(!DH )"15956 J/mol
Q D 
q*"7.33 mol/kg b "1.995;10\/bar  n"0.9169
Q 
k"0.20/s D "2.3 cm/s D "2.6 cm/s
 0
K "0.39 J/(m s K) K "0.41 J/(m s K) a "634 W/(m K)
 0 U
Geankoplis (1993). For the rest of the numerical study, the adsorption of

Perry and Green (1997). propane on activated carbon was used as the base case.
Calculated from Eq. (15). The equilibrium, kinetic and physical parameter values
Calculated from Eq. (16). taken from Malek and Farooq (1996) are listed in
Calculated from Eq. (18).
Table 3. The other parameters are same as those in
Calculated from Eq. (19).
Calculated from Eq. (17). Table 2.
*All other parameters are from Malek and Farooq (1996). Representative two-dimensional concentration and
temperature pro"les in the bed at di!erent times are
shown in Fig. 4. It can be seen that the distribution of
temperature in the radial direction is more pronounced
than the adsorbate concentration in either adsorbed or
gas phase. The radial gradient of gas concentration exists
only in the early part of adsorption. Radial di!usion in
the gas phase seems to eliminate the concentration gradi-
ent in this phase. However, the adsorbed concentration
continues to remain high near the wall, diminishing to-
wards the center, since lower temperature close to the
wall favors adsorption.

8. Comparison between the one- and two-dimensional

Fig. 3. Comparison of experimental and computed temperature pro-
"les and concentration breakthrough for the adsorption of methane in
a "xed bed of activated carbon. The symbols are experimental and the
lines are results from the two-dimensional model () concentration
breakthrough, () temperature breakthrough, () temperature mea-
sured at the mid-point, run3, Malek and Farooq, 1996). The equilib-
rium, kinetic and heat transfer parameters used in this simulation are
given in Table 3.
and heat and mass transfer parameters are given in
Table 2. Langmuir}Freundlich isotherm was used to en-
sure more accurate material balance as suggested in the
original work from which the experimental results were
taken (Malek and Farooq, 1996). The results are shown
in Fig. 3. Concentration and temperature breakthrough
measured at the column exit were the average (mixing-
cup) values. On the other hand, the temperature probe
placed at the mid-point of the column gave the local
value. The very good agreement between the simulated
and measured pro"les suggests that the two-dimensional
nonisothermal model and the chosen correlations for the
estimation of heat and mass transfer parameters do in-
deed provide reliable prediction of real system behavior.
results
It has already been mentioned that for the same oper-
ating conditions the one-dimensional solution may be
made to approximate the corresponding two-dimen-
sional solution by using the overall e!ective coe$cient of
heat transfer, h [present in Eq. (14)], as the "tted para-
U
meter. Eq. (20), proposed by Dixon (1985) for estimating
h , relates the e!ective overall heat transfer parameter to
U
the "lm heat transfer coe$cient, a , radial thermal con-
U
ductivity, K , column radius, R , and a constant l. In
0
a non-adsorbing packed-bed system, l varies between
3 and 6. In this study, the breakthrough curves from the
one-dimensional model were "tted with the two-dimen-
sional model by adjusting l. There are 11 and 13 dimen-
sionless groups in the one- and two-dimensional models,
respectively, which are all listed in Table 4. It is clear that
among the dimensionless parameters present in the one-
dimensional model, besides g, which is related to the "tting
parameter, l, all the other 10 parameters are also found
in the two-dimensional model. This fact is particularly
useful since the correspondence between the two models
can be established by using the same set of parameter
values for the two cases. The two-dimensional model has
 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040 4037

Fig. 4. Calculated two-dimensional concentration and temperature pro"les in the bed at di!erent times (a) 40 s; (b) 3000 s; (c) 6000 s, mole fraction in
feed y "0.113; other parameters are same as in Table 3.
D

two additional dimensionless parameters resulting from
more appropriate description of mass and heat transfer
mechanism in the radial direction. E!ects of all the 13
dimensionless parameters of the two-dimensional model
on l were investigated. The range of investigation for the
various parameters, given in Table 4, were judiciously
at several values of the other parameters. Representative
"t between the two models and sensitivity of one-dimen-
sional model solution to l are shown in Fig. 5. The
root-mean-squares (RMS) error between the two models
was calculated according to the following equation:


chosen to ensure that the conditions for combined con- , D
RMS" G G , (21)
centration and thermal wave formation as well as pure N
thermal wave formation were well covered. The two
models were matched at 183 di!erent conditions. These where N is the number of time points at which error
conditions included variation of one parameter at a time calculation were made and D is normalized di!erence in
G
4038 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040

Table 4
Dimensionless groups of the two models

Dimensionless groups Base values Range of E!ect on l

investigation
one- two-
dimensional dimensional

Pe 0.116 0.027}0.48 No

0
Pe Pe 67.6 15.7}236 No
 
a a 145 48}3600 Yes
b b 0.322 0.03}3.2 No
Pe 0.107 0.026}0.43 Yes
0F
Pe Pe 58.6 14.3}190 No
F F
u u 4.7 0.6}46 Yes
u u 10.6 4.1}18.2 Yes
h h 23 1.1}230 Yes
j j 217 72}2116 Yes

1.5 1}3 No

23 7.4}34 No
R R
t 11.8 2}90 Yes
g Fitting parameter

Fig. 6. E!ects of dimensionless groups on l (line) correlation; (symbol)

results obtained matching one- and two-dimensional models.

As may be seen from the summary given in Table 4,

Fig. 5. Typical examples of "tting: (a) temperature; (b) concentration
breakthrough pro"les (1) two-dimensional model; 2, 3 and 4 are from
the one-dimensional model for l"10, 5 and 20, respectively); (c) RMS
error between the two models as function of l.
Pe\
t\
h
j

the temperature or concentration from the two models at
a given time. The normalization was made with respect
to the inlet value. In case of temperature breakthrough,
inlet value in 3C was used. Fig. 5c shows that RMS error
between the two models as a function of l goes through
a minimum. The best-"t value was chosen to be the
l value at which RMS error for temperature break-
through was minimum, which is much sharper and falls
within the relatively broader minimum obtained for con-
centration breakthrough.
seven out of the 13 parameters have e!ect on l. The
e!ects of these seven parameters on l when the other
parameters were "xed at the base values are shown in
Fig. 6. The following correlation is proposed for the
observed variations:
l"11.43#0.000556a
0F
;(u!17.8u#98.1)!1.98 ln u. (22)
Plot of the correlation is also included in Fig. 6.
A comparison between all 183 l values obtained from
simulation and the values given by the above correlation
is presented in Fig. 7. The correlation coe$cient of the
453 line is 0.999, which supports that the observed de-
pendence of l on the seven dimensionless parameters is
indeed well represented by the proposed correlation. The
usefulness of the proposed correlation and hence this
study becomes evident when CPU time required in
 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040 4039


a R
Bi solid-wall Biot number " UQ
Q k
PQ
C gas-phase concentration, mol/m
c heat capacity, J/(K kg)
N
d diameter, m
D dispersion coe$cient, m/s
D molecular di!usivity of gas, m/s
K
h overall heat transfer coe$cient, W/(m K)
U
!DH heat of adsorption, J/mol
K e!ective thermal conductivity J/(m s K)
K axial conductivity of the #uid, J/(m s K)
D
k thermal conductivity of gas, J/(m s K)
D
K radial conductivity of the #uid, J/(m s K)
0D
k stagnant bed radial ("axial) thermal con-
PQ
ductivity, J/(m s K)
Fig. 7. Comparison of the simulated l values with those given by the
k thermal conductivity of adsorbent, J/(m s K)
correlation. Q
k lump mass transfer coe$cient, s\
solving the two models are compared. Depending on the adsorbent bed length, m
parameter values, the two-dimensional model solution M molar weight of adsorbate gas
required 6}12 CPU hours on a Cray Supercomputer N the number of time points


(J90), whereas the one-dimensional model required only a d
Nu #uid}solid Nusselt number " DQ N
10}15 CPU min on the same machine. Similar computa- DQ k
tion in a Pentium (200 MHz) PC required on an average D


a d
16 h and 18 min, respectively. Nu wall Nusselt number " U N
U k
D


a d
Nu wall-#uid Nusselt number " UD N
8. Conclusion UD k
D
d;
General one- and two-dimensional models for adsorp- Pe N D
 D
tion in nonisothermal "xed beds have been proposed and 
;
numerically solved. It is shown that the e!ective overall Pe D
coe$cient of heat transfer is a very crucial parameter for  D

one-dimensional analysis of nonisothermal adsorber d ;e o c
Pe N D E NE
dynamics, as the agreement between the two models D k
depends on the chosen value for this parameter. It is D
d ;e o c
proposed that an appropriate value for this parameter Pe N D E NE
F K
may be estimated a priori from Eq. (20), with a and 
U ; e o c
K calculated according to Eqs. (17) and (19), respective- D E NE
0 Pe
ly, and l determined from the correlation, Eq. (22), de- F K

veloped in this study. Although Eq. (22) is empirical in d;
nature, all the dimensionless groups used in this equation Pe N D
0 D
appear in the two-dimensional model. These dimen- 0
R;
sionless groups cover all the physical and operating Pe D
parameters involved in nonisothermal adsorption calcu- 0 D
0
lations. The proposed correlation is the result of an d ;e o c
Pe N D E NE
exhaustive search on the e!ect of all the dimensionless 0D k
groups and may be considered to be representative of 0D
d ;e o c
a wide range of adsorber conditions. Pe N D E NE
0F K
0
R; e o c
Pe D E NE
Notation 0F K
0
c k
b constant of Langmuir equation, bar\ Pr NE
 k
D


a R P total pressure, bar
Bi wall-#uid Biot number " UD 2
D k q adsorbed phase concentration, mol/kg
0D
4040 W. Lin et al./Chemical Engineering Science 54 (1999) 4031}4040

q* equilibrium adsorbed phase concentration, B bed

mol/kg c inert gas
q* saturated equilibrium adsorbed phase con- f feed
Q
centration, mol/kg g gas
R universal gas constant ("8.314 J/(mol K)) h heat
R bed radius, m k adsorbate
d ;e o one one-dimensional
Re N D E p adsorbent particle
k
R radial
k two two-dimensional
Sc
oD w wall
K
temperature, K
t time, s
; velocity, m/s References
x radial distance

x axial distance Derkx, O. R., & Dixon, A. G. (1997). E!ect of the wall Nusselt number
 on the simulation of catalytic "xed bed reactors. Catalysis oday, 35,
y fraction of adsorbate
435.
Dixon, A. G. (1985). Thermal resistance models of packed-bed e!ective
Greek letters heat transfer parameters. A.I.Ch.E. Journal, 31(5), 826.
Dixon, A. G. (1996). An improved equation for the overall heat transfer
o q* coe$cient in packed beds. Chemical Engineering and Processing, 35,
a Q Q 323.
C
ID Dixon, A. G., & Cresswell, D. L. (1979). Theoretical prediction of
a #uid-solid heat transfer coe$cient, W/(m K) e!ective heat transfer parameters in packed beds. A.I.Ch.E. Journal,
DQ
a wall "lm heat transfer coe$cient, W/(m K) 25(4), 663.
U
a wall-#uid heat transfer coe$cient, W/(m K) Farooq, S., & Ruthven, D. M. (1990a). Heat e!ects in adsorption
UD column dynamics. 1. comparison of one- and two-dimensional mod-
a wall-solid heat transfer coe$cient, W/(m K)
UQ els. Industrial Engineering and Chemical Research, 29(6), 1076.
k Farooq, S., & Ruthven, D. M. (1990b). Heat e!ects in adsorption
b column dynamics. 2. experimental validation of the one-dimensional
;
D model. Industrial Engineering and Chemical Research, 29(6), 1084.
e bed void Finlayson, B. A. (1971). Packed bed reactor analysis by orthogonal
collocation. Chemical Engineering Science, 26, 1081.
h
g U Geankoplis, C. J. (1993). ransport processes and unit operations (3rd
e o c ; ed.), Englewood Cli!s, NJ: Prentice-Hall. Int.
E NE D
h b R C Gunn, D. J. (1987). Axial and radial dispersion in "xed beds. Chemical
   ID Engineering Science, 42, 363.
o q*R Kaguei, S., Shemilt, L. W., & Wakao, N. (1989). Models and experi-
u Q Q
o c ments on adsorption columns with constant wall temperature-
E NE radially varying and radially lumped models. Chemical Engineering
(1!e )o c Science, 44, 483.
j Q NQ Kaguei, S., Yu, Q., & Wakao, N. (1985). Thermal waves in an adsorp-
e c o
NE E tion column parameter estimation. Chemical Engineering Science, 40,
o density, kg/m 1069.
Malek, A., & Farooq, S. 1996. Kinetics of hydrocarbon adsorption on
(1!e )

activated carbon and silica gel. A.I.Ch.E. Journal, 43(3), 761.

e Patankar, S. V. (1980). Numerical heat transfer and -uid -ow. Washing-
ton: Hemisphere.
a
t U Perry, R. H., & Green, D. W. (1997). Perry's chemical engineering
e o c ; handbook (7th ed.), New York: McGraw-Hill.
E NE D Ruthven, D. M., Garg, D. R., & Crawford, R. M. (1975). The perfor-
(!DH )
u  mance of molecular sieve adsorption column: non-isothermal system.
R Chemical Engineering Science, 30, 803.
 Silva J. A. C., & Rodrigues, A. E. (1997). Fixed-bed adsorption of
D U represents the average adsorbate concen- n-pentane/isopentane mixture in pellets of 5A zeolite. Industrial En-
G


U !U gineering and Chemical Research, 36, 3769.
tration or temperature "   Sun, L. M., & Meunier, F. (1991). An improved "nite di!erence method
U
 for "xed-bed multicomponent sorption. A.I.Ch.E. Journal, 37(2),
Superscripts and subscripts 244.
Tien, C. (1994). Adsorption calculations and modeling, Boston: Butter-
worth-Heinemann.
0 initial Yang, R. T. (1987). Gas separation by adsorption processes. MA: Butter-
A axial worths, Stoneham.