llwptaa ad f2hcmicd

Bn&?kh& vol.4, pp.85-92

@IPapmooPressLtd.,1960. Printed
inGreatBritain

NUMERICAL TECHNIQUE FOR SOLVING PARTIAL

DIFFERENTIAL EQUATIONS WITH APPLICATIONS
TO ADSORPTION PROCESS

MAW-LINGWANG,*CHING-TIENLIout and RONG-YEUCHANG

Department of Industrial Chemistry & Chemical Engineering, National Tsing Hua University, Hsinchu, Taiwan,
Republic of China

(Received 19 June 1979)

Abstract-A new computational scheme of partial differential equations‘which describe the charac-
teristics of an adsorption process in a packed bed is developed. The solution for the concentration
and temperature profiles is obtained by first transforming the partial differential equations into
ordinary differential equations along characteristic paths and then solving them with a combination
of finite difference methods. The computed temperature and tzoncentration distributions are com-
pared with those,available from the literature. The computational algorithm is found more efficient
and more precise than previously documented ones.

ScowThe objective of this work is to develop a powerful numerical method to solve a system of
first and second order, nonlinear partial differential equations. The equations describing the general
adsorption process in a packed bed are chosen. The purpose of this study is to simulate concen-
tration and temperature profiles for adsorption processes. The objective has been achieved by
transforming the partial differential equations into ordinary differential equations along characteristic
paths and using finite difference methods-either the implicit Euler technique or the Runge-Kutta
Gill method-in combination with the Crank-Nicholson Six-point implicit formula. The adsorption of
methane from a helium stream in a fixed activated carbon bed is computed and the results are
compared with the experimental data given in the literature.

Conclusion and Significance-The agreement between experimental and computational results shows
that the numerical technique employed in this study is a powerful tool in solving partial differential
equations of the non-isothermal adsorption process. The calculational errors are minimized by
preventing the complicated calculation from moving out of bounds. The computational results are
very consistent with the experimental data reported in the literature and the number of points to be
calculated has been reduced to only about l/IO of those applied by Meyer & Weber[ 121.Therefore,
in the present study, the computation time is greatly reduced and the solution is more accurate.
These results are achieved by a more balanced number of calculations in the grid of the time and
axial length independent variables. In addition to the main significant results of the present Study for
saving much computation time, it is shown that the present model is a better one and the present
computational technique is more effident.

INI’RODUCTION model. The numerical computation used the method of

Before 1950, solutions of the partial differential equa- characteristics, associated with a modified Euler method.
tions which describe the adsorption process in a packed For most industrial applications, the model must in-
bed were limited to those based on simple mathematical clude concepts of micropore and macropore diffusion,
models[l]. Exact solutions were obtained only for sys- and nonline& equilibrium adsorption, coupled with heat
tems with linear adsorption isotherms, dilute concen- effects; these make the problem complex and difficult to
tration of the adsorbate and negligible heat effects. solve. Colwell & Dranoff [4] studied, with a digital com-
After the development of high speed computers and puter, the effects of nonlinear equilibrium and axial
the establishment of comparatively more complex mixing on sorption processes in a packed ion exchange
mathematical models, Rosen[Z] was the first to report an resin bed. Antonson & Dranoff [5], in 1970, also studied
analytical solution of a general adsorption model for packed bed adsorption columns; breakthrough curves
systems with a linear equilibrium adsorption isotherm. were generated for intraparticle-diffusion-controlled
Acrivos[3] used I the method of characteristics-an operation with nonlinear Langmuir equilibrium isotherm
extremely useful tool for many chemical engineering and cylindrical adsorbent particles. And in 1973, Garg &
problems, especially in adsorption. A system for drying a Ruthven[6] used an implicit finite difference scheme
diluted air-vapor mixture by passing it through a bed of (Crank-Nicholson method) to predict the asymptotic
silica gel particles was described by a one-dimensional forms of the breakthrough curves for an adsorption
column packed with molecular sieves. Other mathemati-
*To whom the correspondence should be sent. cal techniques which involve Green functions and
t&sent address: National Taiwan Institute of Technology, Laplace transformation [7,8] were also used to calculate
Taipei, Taiwan, R. 0. C. the breakthrough curves either for surface reaction or
85
86 et al.
M. L. WANG

for pore diffusion control. The above analyses, however, aT aT

were restricted to the operation of packed adsorption ;ii-+uz= ~(~~~c~*-c~,XT-T,)(y-*,)
beds under isothermal conditions.
For nonisothermal situations, a survey of adsorption
-%T- T,,)-3~- e)) (4)
literature indicates that most of the work was done under
adiabatic conditions, over a small temperature rangel%
111.The most complete discussion of nonisothermal ad-
%=%T-T,)tz$T,-T,) (5)
sorption in fixed beds is found in Meyer L Weber[l2]. w w
They established a mathematical model in which heat
and
and mass transfer resistance around the adsorption par-
ticles, wall effects and moderate heat loss to the sur- pu( 1 - y) = constant (6)
roundings are taken into consideration. However, their
numerical method results in long computation times, where
because it involves a very complicated procedure of
interpolation. Moreover, temperature and concentration
distributions in the bed, and the breakthrough curves,
deviate from the experimental values, especially in the
early stage of operation.
In desiining an adsorption process, the size of the bed
and the cycle time should be determined. It is essential to
have an effective and accurate numerical method which
takes short computation time. In this paper a powerful
numerical technique is presented to solve a mathematical
model proposed by Meyer & Weber[ 121 with some
modifications.

THEMODEL. OF AlSORFrrON
In industry, most adsorption processes are carried out
under unsteady state and nonisothermal conditions. The following initial and boundary conditions apply: at
Therefore, the physical properties and transfer t = 0, for all z and r
coefficients change with the temperature and com-
position changes in the packed bed. In adsorption bed T(t, z) = Tw(t,z) = Tp(f, r, z) = To (7)
(Fig. l), the adsorbate is removed from the carrier gas
when the feed stream is passed over the adsorbent in the y(t,z)=x(t,r,z) = Yo (8)
packed tower. The mathematical model to describe the
adsorption process uses the following equations: atz=O,forallt>O

T(t, z) = TF (9)
(1)
y(t, 2) = YF (10)

(2) dt, Z) = G&PF (11)

at r=O. for all r>O

(3)
&=aT,=O (12)
r------Fluid outlet, T, y ar ar

-Cooling fluid inlei

at r=r,,fort>O

(13)

&%=h(T-T,). (14)

Equations (3) and (4) which are used to describe the

concentration of adsorbate and the temperature of bulk
stream within the packed bed are different from that of
COOI Meyer & Weber’silZ]. In principle, Eqs. (1) and (2)
outle which describe the concentration of adsorbate and the
temperature within the particles are the same as that of
Bed void volume (bulk fluid)
Meyer & Weber’s. However, Eqs. (1) and (2) are affected
T, Y, P,l
Feed fluid inlet by Eqs. (3) and (4). Therefore, this will make the solu-
r,,v,,v, P tions
__ of___..
Eqs. (1) and (2) different from that obtained by
Fig. 1. Schematic diagram of adsorption process in a packed bed. Meyer iyr Weber.
 Technique for solving partial differential equations 81

NUMERICAL TECHNIQUE
(N=EO)
As indicated in the beginning, today’s mathematics z-const.
does not offer the exact solution to these nonlinear
partial differential equations. In this paper, a useful I3
computational technique is presented to evaluate the
meaning of the heat and material balance equations. The I
II
j+l
numerical procedures for calculation are obtained by
at 2
dividing the related independent variables into several
grids. In Fig. 2(a), a three-dimensional grid and two
natural characteristic paths of the system are identified. j

Definitive Eq. talc. Variable talc.

Character- Eq. for along the along the char. j-l

istic path char. path char. path path

/z
I v = dzldt (3), (4) Y,T i+l
i-l i i+2 LV=312)
II 2 = constant (l), (2), (5) X, T,, T,. /
J
r (K=El
Equations (3) and (4) can be solved along the in-
Fig. 2(a). Grid for numerical solution in the present study.
dividual characteristic curve I by employing the Implicit
Euler Method or the Runge-Kutta-Gill Method. Equa- t
tions (1) and (2) belong to the parabolic type of partial (N=~5.000)z_const,
differential equations, therefore, by using the Crank- I
Nicholson Six-point Implicit Formula, they can be
transformed into finite difference equations. The solu-
tions of Eqs. (1) and (2) are obtained by calculating all 5
jtl t=const.
the points following the characteristic curve II. Equation
(5) is solved by the Implicit Euler Formula along the I 4 2 3
j,
characteristic path II. A flow chart of the computational
algorithm is shown in Fig. 3. An example demonstrating jtl -
the calculating procedures follows.
As shown in Fig. 2(a), following the individual charac-
teristic curve, the values at point 3 can be obtained from
the known values of points 1 and 2 by assuming that the
values on points t = j. At are all known. Therefore, the
finite difference equations can be represented by the
following:

Yi.j+t = Yi-l,j+l -y1 -Y)(Y

_ _ -&;) m

' (L=8)

Ti,j+l = r-l,j+l +Bz’ Fig. 2(b). Grid for numerical solution by Meyer & Weber.
PC,
Equations (15) and (16) can be written as,
badC,z-C,dU -Y)(Y -zs)U- To)
l
yi,j+l = Yi-l,j+l -Arfl(yi,j+1, xsi.j+l>Ti,i+l, Gsi.j+lvTwi,j+,)
-h&U’- la,)-+@- Tw)) (18)
(16)
Ti.j+l= I;.-l,j+l -A~f2(Yi.j+1,xsi,j+l,Ti.j+l, Tpsij+l,Twi.j+J
T..w,,+I =T..+=,~..-T..) W.J
WJ mC 1.1 (19)
i=12 , ,..., Mtl
+GtT -7 W.l
..) 7 (17) j=12 , ,..., Nt 1.
mC, o
where By solving Eqs. (1) and (2), we get
y = (Yi.j+l + Yi-l,j+1)/2
$ = (E.,A - EdI)/(E, E., - E&) (20)

L = Vwi.,, + L_,,+,)/2 ~=(EIB-EIA)/(~E-E~E,) (21)

where
&S= txSi,j+l + xSi_f j+,)/2

Gs = (Tpsij+l+ T,.+,j+,)D

i=12 , I..., Mtl

j=12 9 Y..., Nt 1.
 M. L. WANGet 01.

Input Data & ITEL

x. y, T, Tp - Init. Cod. - TIME,
DIMENSIONS (Incl. Coeff.) cm,x,L M, N

No convergence

Fig. 3. Computational flow chart of the algorithm in present study.

Define
fk = (k.j,k + &.j+l,k)/2
’ (29)
FPk
=( &ij,k+ TPij+
,,)I2 (23) where
ji.j = (Yij+l + Yi.i)D
i= 1,2,...,Mtl
j=1,2,...,Ntl Fi,j = (Ti,j+lt T,j)/2
k=1,2 ,..., Ltl.
Then, A and B can be expressed in finite difference for l+W;.j = tTWf,j+l + TWi,j)/2

k= l,(r=O)
and the variable? in El, &. Es and E4 can also be
expressed by .?, Tp
60,
A =-+fZ-f,) Equations (20) and (21) are written in finite difference
(24)
form,
xi.j+l,k = Xi.j,k+‘At(&A- EzB)I(EI&- &E3) (30)
B2ke(F -F
57P2 PI
) (25)
T,i.j+l.k = TPii,kt At(E,B - E,A)/(E,Eq - E&3).
for k = 2,. . . , L (31)

1
In general, the above two equations can be written in a
&@k+, -lk-1) more general form,

(26) xi.j+ I& = ki.j., t Atfj(x.~.~+l.k+1~~i.j+l.k~~i,j+l.k-l,

TPi.j+i.t+l*
Tpij+l
. t, Tpij+l.Y-1) (32)
(27) TpiJ+,.a
= T,,,, ~A~f4(Xi.j+~,k+~,Xi.j+l.k~Xi.j+l.k-l,

fork=Ltl (r=rr) TPdJ+l,t+l,

Gi.j+~,~*
T,i.j+l.r-l) (33)

A=~[(~) i=1,2,...,Mtl
P
(jij - -fL+I) - 2(fL,+I - 9)
1
(28) j=1,2,..., Ntl
 Technique for solving partial differential equations 89

In the above statements, the slope, f,, (or f~), is the temperature and concentration profiles in the adsorption
average value between the values at points 1 and 3 which processes. By doing so, the calculation errors and com-
are shown in Fig. 2(a) (i.e. along the characteristic curve putation time are minimized.
I), and the slope, f3, (or f.,), is the average value between The advantages of the numerical method used in this
the values at points 2 and 3 (i.e. along the characteristic paper to solve the partial differential equations can be
curve II). The values at point 3 can not be calculated easily seen by comparing it with the one suggested by
directly because of the highly nonlinear equations. Meyer & Weber. Their numerical calculation procedure
Therefore, an iteration method is employed. The cal- follows the pattern shown in Fig. 2(b). In order to obtain
culation is terminated when the normalized difference of the values at point 5, the data for point 4 have to be
calculated values between two successive iterations is supplied. However, a very complicated method df inter-
within the tolerance (e.g. 0.00001). By following the same polation is needed to obtain the values at point 4 from
approach, the values of all points at t = (j t 1) At can be the data of points 1,2 and 3. It is difficult to get accurate
obtained. For example, the values at point 4 can be values by interpolation.
calculated from the known information at points 3 and 6. In addition, the method used by Meyer & Weber needs
In general, if the stability criteria of the system are much computation time. This can be explained by two
satisfied, the convergence of the solution is ensured. factors. First, from the grid for numerical solution as
However, the unique convergence limit is not necessarily indicated in Fig. 2(b), the location of point 5 is searched
reached. along the characteristic curve I, u = dddt. However, the
The interval used for calculating the values along the velocity, V, in general, is a function of the temperature
characteristic path should be restricted[l3]. The limita- and concentration within the packed bed. Therefore, the
tion on At’ for solving hyperbolic equations is given by, location of point 4 will always change during the iterative
calculation if the slope at point 4 is considered the
average value between the values at points 4 and 5. Thus,
the iteration will consume much more computation time
to search for the values at point 5. Second (as shown in
and the corresponding limitation for AZ is given by, Fig. 4) the changes of the concentration and temperature
within the packed bed with respect to z-position are very
sharp in a transient state.
$1. In carrying out the computation results in the present
study, 9 points are chosen for r(L = 8), 313 points are
chosen for z(M = 312). Time intervals are about 8&100
If AZ and At’ are chosen to satisfy these limits, the points depending upon the completeness of the ad-
solution will be convergent. sorption processes desired. In Meyer & Weber’s paper, 9
As indicated by Meyer & Weber, there is no stability points are chosen for r, 17 points are chosen for z and
criterion for the radial increments because the Crank- 15,000 points are chosen for t. As the intervals of z
Nicholson finite forms of Eqs. (1) and (2) are un- become smaller, the errors of slope to be estimated will
conditionally stable. become less and the results will become more accurate.
The choice of At in calculating Eq. (5) should satisfy In general it takes only 3-6 iterations to achieve the
the following stability criterion, convergence within the adsorption zone which occupies
about 20% of the whole length of the packed bed. But it
At Il&w I llCu < 1
only takes l-3 iterations to achieve the convergence
2 ( mC, mC,> ’ beyond the adsorption zone. Therefore, the number of
iterations needed to achieve the convergence in both
At the start of the calculation, a small time interval, At, cases is reasonable. However, the procedure of iterations
is chosen on account of the effect of high gradient in proposed by Meyer & Weber will consume much more

3121 I 0.12
\ ! I

t=0.015 hr

g 308-
t.
F!
3
6

Fbints as indicated corresponding

p lo Fig. Z(a) and 2(b)
Y 304-
1

Jo
3001 I I
0 0.25 0.50 0.75 I.0
- .7- oxis
Fig. 4. Characteristics of concentration or temperature in hulk stream vs z-axis at transient state.
90 td. L. WANG
et al.

computation time. Putting the time required for each in Figs. 5-8. Comparisons are also made with previous
iteration in both cases aside, a comparison of com- work[l2].
putation time needed between the present study and that Figures 5 and 6 show the concentrations of adsorbate
of Meyer & Weber can be estimated from the number of for the bulk stream within the packed bed. A comparison
points to be calculated. of the calculated results by Meyer & Weber, with the
The number of time intervals in the present study is one produced by the present study, based on their model
much less than the number of intervals used by Meyer & but with a different computation method, and the
Weber. Since the number of points to be calculated by experimental data is shown in Fig. 5. It can be easily
Meyer & Weber is about 10 times of the points cal- seen that the present method is a better one on account
culated in this research, only about l/10 of the com- of the balanced grid proposed by the present study.
putation time is needed in the present study. Figure 6 shows the calculated values based on the
present computational technique but with two different
models. The present model fits better with the experi-
mental data.
SWI’Ehf EXAMPLE AND TRANSFER The temperature distributions of solution for the bulk
COEFFICIENTS stream within the packed bed are shown in Figs. 7 and 8.
As an example, we use the removal of methane gas Comparisons of calculated results by Meyer & Weber
from a helium stream by adsorption on activated carbon, and the present study with the experimental data based
studied experimentally by Meyer & Weber. We use their on different kinds of computational techniques are
values for the physical properties of the substance and shown in Fig. 7. The calculated results presented in this
the characteristics of the adsorption packed bed. The study are much closer to the experimental data. The
heat and mass transfer coefficients are obtained from the corresponding temperature distributions are shown in
literature[l2,14]. The computational results are plotted Fig. 8.

0.12

.-s
t
e

%
E 0.00 -
4_
s
L
‘iz

time (hr.)
Fii. 5. Effluent concentration history, using Meyer & Weber model.

0.12
Computer results
- Present model
- - --- Meyer and Webr’s

time (hr.1
Fig. 6. Etlluent concentration history, using present computational method.
92 tvf.L. WANG
d cd.

x,mole fraction of adsorbate on the surface of adsorbent 2. J. B. Rosen, Ind. Engng Chem., 46, 15% (1954).
ymole fraction of adsorbate in bulk stream 3. A. Acrivos, Ind. Engng Chem., 4t?(4),703 (1956).
yFmole fraction of adsorbate in bulk stream at entrance 4. C. J. Cqlwell & J. S. Dranoff, I & EC Fundamentals, 8(2),
y0mole fraction of adsorbate within packed bed at t = 0 193 (1%9).
z axial coordinate of packed bed, cm 5. C. R. Antonson & J. S. Dranoff, Chem. Engng Progr. Symp.,
No. %, Vol. 65,27 (1970).
Greek svmbols 6. D. R. Gara & D. M. Ruthven, Chem. Enana -- Sci., 28, 791
a ‘volume fraction of voidage within adsorbent, cm’/cm3 of (1973). -
bed 7. D. Gidaspow, D. Dharia & L. Leung, Chem. Engng Sci., 31,
u volume fraction of solid adsorbent, cm3/cm3of bed 337 (1976).
voidane of packed bed 8. K. Kawazoe & Y. Takeuchi, /. of Chem. Engng of Japan,
; a/(a iu, _ 7(6), 431 (1974).
P density of fluid in bulk stream, g-mole/cm’ 9. P. J. Banks, Chem. Engng Sci. 27, 1143(1972).
PF densitv of fluid in bulkstreamat entrance, a-mole/cm-’ 10. D. J. Close & P. J. Banks, Chem. Engng Sci., 27, 1157(1972).
PfP density of fluid within adsorbent, g-mole/cm3 II. C.-y. Pan & D. Basmadjian, Chem. Engng Sci., 22, 285
Pa density of adsorbent particles, g/cm3 of bed (1%7).
P: J(1 - 6) = PA + 0) 12. 0. A. Meyer & T. W. Weber, AIChEJ. 13(3),457 (1%7).
0 quantity of adsorption in equilibrium, g-molelg-adsor- 13. R. Camahan, H. A. Luther & J. 0. Wilkes, Applied Numeti-
bent cal Methods, Wiley, New York, 449 (1%9).
14. R. H. Wilhelm, Chemical Reactor Theory A Review, ~349,
RERZRENCES Prentice-Hall, Englewood Cliffs, New Jersey, U.S.A. (1977).
1. 0. A. Hougen & W. R. Marshall, Jr., Chem. Eng. Progr..,
Trans. Se&on 43(4), 197(1974).