Beruflich Dokumente
Kultur Dokumente
00
Printed in Great Britain. 0 1986. Pergamon Preos Ltd.
and
N. S. RAGHAVAN
Alcan InternationalLimited, Kingston Laboratories,Kingston, Ontario, Canada K7L 424
Abstract-A simplitied dynamic model for a PSA air separationprocessis developedbased on linearized
mass transfer rate expressionsand binary Langmuir equilibrium. Constant pressure is assumed during
adsorption and desorption steps but the variation in flow rate through the column due to adsorption is
aeeounted for. The model predictions,using independentlymeasured kinetic and equilibriumdata are
comparedwith experimentalresultsobtainedin a simpletwo-bed air separationPSA systempackedwith a
carbon molecular sieve adsorbent. The model is shown to provide a good representation of the
experimentally observed behaviour over a wide range of conditions.
Pressure swing adsorption (PSA) processes, first de- The alternative approach which requires developing
veloped by Skarstrom (1959,1960,1972), in the form of a suitable dynamic model to account for the mass
the heatless drier have found widespread application transfer kinetics has been adopted by Mitchell and
in hydrogen purification and air separation, as well as Shendalman (1973), Chihara and Suzuki (1983),
in air drying. The rational design and optimization of Raghavan et al. (1985) and Carter and Wyszynski
such processes requires a detailed mathematical model (1983). However, in all these models the adsorbable
since the effects of the process variables (flow rates of species was assumed to be present only at low concen-
purge and feed, cycle time, bed length and pressure tration in an inert carrier so that the gas velocity could
ratio) are strongly coupled, making it difficult to arrive be regarded as constant along the length of the column.
at an optimal combination purely by empiricism Such an assumption is obviously not valid for air
and intuition. The earlier mathematical models separation since, even on a carbon sieve where oxygen
(Shendalman and Mitchell, 1972; Chan et al. 1981; is the preferentially adsorbed species, the variation in
Knaebel and Hill, 1982) were based on equilibrium fluid velocity due to adsorption will be significant. In
theory and the effects of mass transfer resistance were the present paper we report the development of a
ignored. The validity of such an approximation has simple dynamic model for a bulk separation involving
been confirmed experimentally for air separation on a the competitive adsorption of two components and
5A zeolite adsorbent by Flores-Femandez and allowing for the variation in fluid velocity through the
Kenney, 1983. However, for adsorption of CO2 on column. Experimental data are presented for PSA air
silica gel (Mitchell and Shendahnan, 1973) and C2HL separation on a carbon molecular sieve and it is shown
on both 4A and 5A zeolites (Hassan et al., 1985) that the model, with independently measured kinetic
experimental PSA studies reveal rather large devi- and equilibrium parameters, provides a good predic-
ations from the equilibrium theory predictions, sug- tion of the behaviour of the system.
gesting that, at least for these systems, kinetic effects
are probably important. Although equilibrium theory
can provide valuable guidance as to the range of THEORETICAL MODEL
acceptable operating conditions and the limits of The simple two-bed PSA cycle involves four steps as
performance, the quantitative value of such models is shown in Fig. 1. During step 1, a high pressure feed is
therefore limited. Furthermore, the equilibrium theory supplied continuously to bed 2, in which preferential
approach is obviously restricted to systems in which adsorption of the faster diffusing oxygen occurs. The
the adsorptive selectivity depends on a difference in relatively slowly diffusing nitrogen is removed as pure
equilibrium and it is clearly not applicable to systems, raffinate product. A small fraction of this stream is
such as air separation on a carbon sieve. The adsorp- expanded to low pressure and used to purge bed 1. In
tion equilibrium isotherms for oxygen and nitrogen on step 2, bed 1 is pressurized with feed while bed 2 is
carbon molecular sieves are almost identical and the blown down in the reverse flow direction. The same
preferential sorption of oxygen in these adsorbents is cycle is repeated in steps 3 and 4 with high pressure
of kinetic origin, resulting from the higher micropore flow and adsorption occuring in bed 1, and purging
diffusivity of that species. occuring in bed 2. In order to develop a mathematical
1333
1334 M. M. HASSANet al.
Adding eqs (1) and (2) and considering eq. (3),
I--Product
Da
Solid phase in bed 2.
k2
COI. 2 -
at
= k,(qz2 - 4a2) (3
0
Purge Roprossurbotlon ah2
at = k, Mu-qm) (6)
F**d Blowdown -
I I
0
I
0.4
11
.l
I
0.5 ml
I
a9-?
r
I I
I
I
I,
1
where
= 1 + b,C,, + b,C,
+
(--
l-& aqB2E(-J
>at
&
(2)
Reference Gas
4
TCD TCD
Ertr. Roff.
PSA PSA
COi. Cal.
1 2
350mm 350mm
FM1 X X FM2
35mm 35mm
v V
-
0
PC2
I
IT,
Feed RV4
Fig. 2. Schematic diagram of small pressure swing adsorber. FC, Flow controller; FM, flow meter; PC,
pressure gauge; PT, pressure transducer; T, thermocouple; PRV, pressure relief valve; RV. regulating valve;
SV, solenoid valve; V, shut-off valve; TCD, thermal conductivity detector.
and cycle time were investigated and the results are pressurization times were taken as 25 % of the adsorp-
summarized in Figs 3-10 and in Tables 1 and 2. tion or desorption times. In Run 13 the adsorp-
For all runs except Run 13, the blowdown or tion/desorption time was 90 s while the blow-
Air separation by pressure swing adsorption
10
cl x
c
ooooooo
3 00 000000
0000000
s
.I
.
5
I= Fig. 6. Effect of varying length/velocity ratio on approach to
m cyclic steady state (from initially clean beds). x , Run 8, L/o
z =30;0.Run7.L/v=25;O.Run6,L/v=2O.Forahrunsa
= 1.0, cycle time = 150s.
lo+1
0
*
5
.
10
15
.
20
*
25
J
Holf Cycle Number-
10-l
t <b>
. . I
10-l0 . 1
5
. 1
10
. 1
15
. 1
20
. 1
25
J
Half Cycle tkmbm-
Fig. 7. Effect of varying cycle time on approach to cyclic
Fig. 4. Approach of residual oxygen wncentration in product steady state (from initially clean beds). x , Run 9, Steps 1 or 3
to cyclic steady state value (beds initially clean). (a) Feed =!3O~Step2or4= 22_5s;O,RunlO,Stepslor3=40s,
=air(Runn(b) feed= 15%01inHecarrier(Run152For StepsZor4= 10s;O.Ruu3,Steps1or3=6Os,Steps2or4
both runs L/v = 25, purge/feed ratio = 1-0. cycle time E 15s;a,Run13,S~lor3=90qSteps2or4= lOs.For
= 150 s. Theoretical curves are shown as wminuous lines. all runs a = 1.65, L/v = 25.
M. M. Hw.er al.
Constant
Run Feed Purge Purge to W Time for Time for 02 N2 Product cont. at
No. rate rate feed lIlti0 Step 1 or 3 Step 2 or 4 steady state % 0, #
(SLPM) (SLPM) ratio (s) 9
(Q n, L.k f), flak, Expt. Theory
v * v
j
2 0.90 0.28 0.93 21 60 15 2.25 =.
0
3 0.96 0.551 1.65 25 60 15 12 1.1178 65 0.1904 1.50 1.48 u
5 1.21 0.644 1.60 20 60 15 12 0.8946 65 0.1523 2.55 2.91 yb
6 1.20 0.405 1.0 20 60 15 12 0.8940 65 0.1523 4.20 4.49 s
8
7 0.81
0.96 0.27
0.32 1.0
1.0 30
25 60 15
15 12
12 1.341
1.1178 65
65 0.2284
0.1904 2.45
2.95 2.01
2.70 1
9 0.96 0.551 1.65 25 !!I 22.5 11 1.0247 49.5 a1449 2.15 230
10 0.96 0.551 1.65 25 40 10 14 1.3048 73 0.2138 2.09 ::Z a
11 0.96 0.16 0.50 25 60 12 1.1178 65 0.1904 6.80 k!
12
13 0.96
0.96 0.551
0.551 1.65
1.65 25
25 90
90 ::.5
10 11
11 1.0247
1.0247 49.5
49.5 0.1449
0.1449 2.60
2.53 2.30
2.30 i
14+ 0.96 0.32 1.0 25 60 15 12 1.1178 65 0.1904 3.00 295 a
154 0.96 0.32 1.0 25 60 15 12 1.1178 6.28 5.21 f
16 0.96 0.32 1.0 25 90 22.5 11 1.0247 49.5 0.1449 4.70 3.90
Theoretical 0.96 0.551 1.65 25 25 6.25 16 1.4912 92.5 0.2709 - 221
bed in which low pressure flow desorp- on the front boundary of chromatograms and on the
tion is carried out, cm/s effectiveness of separation. J. &em. Sot. 1315.
Hassan, M. M., Raghavan, N. S., Ruthven, D. M. and
xAi C,,/C,, or CAi/C,, depending on Boniface. H. A., 1985, Pressure swing adsorption-II.
whether high pressure flow adsorption or A.I.Ch.E. J. 31, 2008-2016.
low pressure flow desorption occurs in the Knaebel, K. S. and Hill. F. B., 1982, Analysis of gas
ith bed, (-) purification by heatless adsorption. Paper No. 91d, Annual
A.1.Ch.E. meeting, Los Angeles, California.
XBi GJCHP or C,,/C,, depending on
Mitchell, J. E. and Shendalman. L. H., 1973, Study of heatless
whether high pressure flow adsorption or adsorption in the model system CO2 in He: part II.
low pressure flow description occurs in A.1.Ch.E. Symp. Ser. 69, No. 134, 25.
the ith bed, (-). Also equal to (I- XAi) Nakao, S. and Suzuki, M., 1983, Mass transfer coefficient in
X dimensionless distance, Z/L cyclic adsorption and desorption. J. &em. Engng Japan 16,
114.
yAi qAi/qhs. adsorbate concentration of com-
Raghavan. N. S., Hassan, M. M. and Ruthven. D. M., 1985,
ponent A in solid phase in bed i, (-) Numerical simulation of a PSA system-I. Isothermal
YBi q&q%, adsorbate concentration of com- trace component system with linear equilibrium and finite
ponent B in solid phase in bed i, (-) mass transfer resistance. A.1.Ch.E. J. 31, 385392.
Ruthven, D. M., Raghaven, N. S. and Hassan, M. M., 1986,
Z axial distance coordinate, cm
Absorption and diffusion of nitrogen and oxygen in a
t-1 dimensionless quantity carbon molecular sieve. Chem. Engng Sci. 41, 1325-1332.
Shendalman. L. H. and Mitchell, J. E., 1972, A study of
Greek letters heatless adsorption in the model system CO1 in He: part 1.
Chem. Engng Sci. 27, 1449.
= -
Lk Ai dimensionless parameter for Skarstrom, C. W., 1959, Use of adsorption phenomena in
QAi
V OH automatic plant-type gas analyzers. Ann. N.Y. Acad. Sci. 72,
component in bed i, (-) 751.
Skarstrom, C. W., 1960, U.S. patent 2,444,627 (to ESSO
Lk, Research and Engng Co.).
= - dimensionless parameter for Skarstrom, C. W., 1972, Heatless fractionation of gases over
V OH
solid adsorbents, in Recent Developments in Separation
component B in bed i, (-) Science, Vol. 2. CRC Press, Cleveland, Ohio.
dimensionless parameter for Villadsen, J. V.. 1970. Selected Approximation Methods for
BAl = b,G,,, Chemical Engineering Problems. Denmarks tekniske hoj-
component A in bed 1, (-) skile. Lyngsby.
B61 = b,C,,, dimensionless parameter for
component B in bed 1, (-) APPENDIX
BA2 = b,C,,, dimensionless parameter for When we consider the high pressure flow in bed 2 and the
component A in bed 2, {-) low pressure flow in bed 1 as the operations during step 1, the
equations to be considered for bed 2 are, eqs (1). (4)-(9), and
B B2 = b&HP, dimensionless parameter for
the equations which describe the operation in bed 1 are, eqs
component B in bed 2, (-)
(1% (13~(19).
E bed porosity, (-) Since steps 1 and 3 are identical except for the direction
Y = qBS/qAS, (-1 of flow, it is sufficient to consider the equations relevant to
step 1.
l--E qAS
*I =-*-, dimensionless parameter Equations (l), (4k(7) combined and written in dimension-
& CL, less form become:
for bed 1, (-) 1 a2xA2 - axA2
ax,2
-=-. - - v,*---
ar PeH dx ax
f--e qAS
dimensionless parameter
=-C
for ;d 2, (-;
-yA2 +YXA2aB2
>
>I (21)
REFERENCES
flB2 (I - xA2)
Carter, J. W. and Wyszynski, M. L., 1983, The pressure swing -yB2 .
x 1 +fiA2XA2+flB2 c1 -xA2)
adsorption drying of compressed air. C&m. Engng Sci. 38,
1093. Equations (5), (6) and (7) when combined give,
Chan, Y. N., Hill, F. B. and Wong, Y. W., 1981, Equilibrium
theory of a pressure swing adsorption process. Chem. ayAZ flAzXA2
dr = (IA2 - yA2 (22)
Engng Sci. 36,243. 1 + flA2xA2 + bB2 t1 - XA2) 3
Chihara, K. and Suzuki, M., 1983. Simulation of non-
ayB2 bB2 t1 - xA2)
isothermal pressure swing adsorption. J. &em. Engng -YB2 . c3)
Japan 16, 53. ~ a7 = 0LB2 1 + bA2XA2 + flB2 (I - xA2) >
Fiores-Fernandez, G. and Kenney, C. N., 1983, Modelling of The relevant boundary conditions, eqs (8) and (9) in dimen-
the pressure swing air separation process. Chem. Engng Sci. sionless form lead to
38, 827.
Forsim, 1976, A Fortran package for the automated solution ax,
of coupled partial and/or ordinary differential equation
systems. Atomic Energy of Canada Limited.
-I
ax r=O= *=0-
- xA2 (24)
- vz (j)*Ax
Ix-o-
(j, 1))
-(P-Bx (j, 1)
+yaB2 1 +BA*X&+&zff
fiBZ t1 - xA2)
-X,)
-yB2
>I. (26) -A,+PeHXAZ
- p2 (/).Ax
I.X-0-
(j. M+2))+F1
-(P*Bx(j,M+2)
(j, (35)
Similarly, for bed 1 the following dimensionless equations
are arrived at. based on (lo), (13)-(19): where
Fz 0) = kz
(x, (i) - lb,,
x
1 +BalXal +BBI (~--AI)
+YXA1aB1
i 1+/3~lX~l +#~l (1 - XA,)
- yBl
(27)
ayAl BAL~AI
-yAl j=)
-= az A1
{ 1 +BAIXAI+BBI (1 --AI) I
ayBl ,%l (I- XAI)
7 = aBl - yBl (2% Equations (22) and (23) now become:
1+ BA~XAI + Be1 (I- XAI)
w-0
(36)
(31)
ayB2 ti, 8B2 l1 -x, th)
aa,
-=
ax
- +l [aAl (
f +BAlx~l
BA~XAI
+fia2 (1~-XAI)
TsaB2
I 1+bB2XA2(/)+f?B2-(1-XA2(/))
-yB2(i) .
*I
&,l c1 - xAl)
+ YaBl (32) The expression for velocity is given by a set of algebraic
1 +BA~XAI+BB~ (~-XAI)
equation as follows:
1V2
(0
Equation (20) which is a valid condition for the startup of Ax (M + 2, i), Ax (j, M + 2) -
the cyclic operation with two clean beds, leads to the Ti2[-4x(X
i)-
Ax (M+2,M+2)
following two sets of dimensionless equations:
XA2 (x, r = 0) = 0; x,, (x, * = 0) = 0; = Fs(j)- Ax (j, 1)
YA2 (x, T = 0) = 0; Y, (x, r = 0) = 0 (33)
- Ax(M+2,1),Ax(j,M+2) _
Vz (1) (38)
x,, (x, r = 0) = 0: x,, (x, r = 0) = 0;
Ax(M+2,M+2) >
YA1 (X, z = 0) = 0, Y,, (x, z = 0) = 0. (34)
Note: ~~(l)=l.Oaud~(M+2)=0;/=2,M+l.
When eqs (21), (24) and (25) are combined and written in
collocation form based on a Legendre type polynomial to
represent the trial function, the following set of ordinary In eqs (35~(37)
differential equation results for the high pressure flow step in l/(Ax(l,M+2)-((Al-Ax(M+2,M+2)))
AI -
bed 2 (step 1):
M+l A,=Ax(M+~,M+~)/(Ax(I,M+~)-(As.Ax(M
~ = ,zz (P.Bx(j, i)- vz (j)Ax(j. i)) +2*M+2)))
A, = (Ax (1.1) - Pe&Ax (M + 2,l)
-(A~.(P.BxO. 1)
hl+1 A., = l/Ax (M + 2,1)
-F-~(/)*Ax(j,l))- c Ax(M+2,i))
As =Al*A4
i-2
-(A,-(P-Bx(j, 1) P = l/Pe,
M-F1 M = number of collocation points.
-Vz((/)-Ax&I))* C (A,~Ax(M+2,i)
i-2
1VI
(i)
-Ax (LO)) Ax(M+2,i)*Ax(j,M+2) -
Ax(M+2,M+2)
+(B,-(QmBx(j,M+2)-VI(jj.Ax(j,M
Al+1
+2)). 1 (Bs-Ax (M+2,1) Ax (M+2,1)Ax (j,M+2)
i=1
Ax(M+2,M+2) >
--Ax (1.0)) -X,, 60 PI (1) J = 2, M + 1. (42)
>
+B,*PeL.G*XM _
j_M+2(Q.Will dPI (M+2) o
Note: PI (1) - G and
dX =
-VI (/).Ax(i,
1))
In eqs (39)-(42)
-B,.PeL*G-Xu (Q*Bx(j,M++)
I j-M+2 B, - l/(Ax(l,M+2)-((BB-Ax(M+2.M+2)))
where +2.M+2)))
B, = (Ax (1.1) - Pe,)/Ax (M + 2, 1)
F, Ci)= tii CX,,, (i) - l)a,,
C BA== l/Ax(M+2,1)
BAIXAI ti) Bs = B,*B,
- -YA, ti1
x { 1 +BAl+x*I (i)+Bm (1 -X**(i)) > Q = l/PeL.
+~XAI (/)-aBl
-31
The dimensionless partial differential equations for the
-yBl(si) . high pressure tlow in bed 1 (Step 3) and the low pressure flow
in bed 2 (Step 3), were similarly written and reduced to sets of
Equations (28) and (29) can be written as: ODEs and algebraic equations by the orthogonal collocation
method. The resulting algebraic equations were solved by
~YAI ti) BAI. x,1 (3 using standard subroutine for solving system of equations,
a+ = KAI
{ 1 +8Al*XAl (i)+hm(l -xAl (i)) while the resulting ODEs were solved using Adams variable
step integration algorithm as provided in the Forsim package
- YAI ti) W)
of the Atomic Energy of .Canada Ltd. (Forsim. 1976).
> simultaneously.