Hydrogen Purification from Refinery Fuel Gas

by Pressure Swing Adsorption

A. Malek and S. Farooq
Dept. of Chemical and Environmental Engineering, National University of Singapore, Singapore 119260

Hydrogen purification and recovery from various process streams constitutes the
largest commercial use of pressure swing adsorption (PSA) technology. This study inves-
tigates the pet$ormance of a six-bed, dual-sorbent PSA operation for hydrogen purifica-
tion from the refinery fuel gas. Major impurities are methane, ethane, propane and
butane, and comprise 30 - 40% of the feed. The dual-sorbent PSA bed consists of an
initial layer of silica gel adsorbent for trapping heavier hydrocarbons and a subsequent
layer of activated carbon for removing lighter hydrocarbons. A numerical simulation
model of the H,-PSA process developed with all the essential features of the actual
operation shows that butane is more strongly adsorbed on activated carbon than silica
gel, and, hence, is less easily desorbed from the former using simple pressure reduction
in the PSA cycle. Therefore, the initial layer serves to prevent butane from degrading the
adsorptive capacity provided by activated carbon for other lighter hydrocarbons. The
simulation model agrees well with the experimental results from a laboratory unit as
well as with available H2-PSA plant data from a refinery. The results also indicate the
importance of heat effects in this process. Extensive parametric studies, which show
effects of feed velocity and cycle time on the variation of product recovery and purity
obtainable from the industrial unit, provide a valuable guide for its proper operation.

Introduction
In an earlier article (Malek and Farooq, 1997b), we dis- trace amounts of higher molecular weight hydrocarbons, such
cussed the development of a simulation model for a six-bed, as butane and aromatics, in addition to the light hydrocar-
ten-step pressure swing adsorption (PSA) process. The per- bons considered in our previous study. These trace impurities
formance of the numerical model was verified with experi- can seriously degrade the capacity of the activated carbon
mental results from a laboratory-scale PSA unit. Although adsorbent and, hence, the PSA performance, because they
the chosen configuration resembled that of an industrial op- are more strongly adsorbed than the lighter hydrocarbons.
eration for the purification of hydrogen from refinery fuel Typically a silica gel layer is added to remove the heavier
gas, a model system consisting of activated carbon adsorbent hydrocarbons prior to reaching the activated carbon layer,
and a feed mixture containing three adsorbable hydrocarbon which is more suitable for the lighter hydrocarbons. These
impurities, namely, methane, ethane, and propane in (inert) deficiencies in the model developed in our previous publica-
helium carrier, was considered as a precursor to modeling tion (Malek and Farooq, 1997b) have been addressed in this
the actual industrial operation. study.
The use of helium in place of hydrogen was a measure of The effect of the presence of butane in the feed stream on
safety precaution in high pressure laboratory experiments. the overall PSA performance and the role of the silica gel
The simulation of an actual commercial operation must in- layer have been examined in detail. The previous simulation
clude hydrogen as carrier, which is weakly adsorbed on acti- model is first extended to include butane in the feed and an
vated carbon. Moreover, refinery fuel gas normally contains additional layer of silica gel adsorbent. The extended model
is then verified with experimental results from the laboratory
Correspondencc concerning this article should he addressed to S Farooq unit with helium as carrier. Finally, the model is further ex-
Current addre% of A Malek Kvaerner R J Brown Pte Ltd , 73 Science Park Dr ,
Singapore 118254 tended to simulate the actual industrial operation and its

AIChE Journal September 1998 Vol. 44, No. 9 1985

performance is compared with available plant data. It is Table 1. PSA Experimental Information
shown that a mathematical model of an actual process, which Column Jacketed stainless steel
is experimentally verified at various stages of development, Internal diameter 4 cm
can be efficiently used for the purpose of evaluating the ef- Length 40 cm
fects of the design variables on the process performance. Adsorbent Activated carbon
Apparent density* 0.87 gecm-'
Particle porosity* 0.58
Mean pore radius* 18X lo-' cm (18 A)
Simulation of Dual-Sorbent PSA Process Specific surface area** 970 m2.g- '
Particle heat capacity** 0.95 J/g*K
The dual-sorbent PSA process was operated with the same Particle size 2.36-2.80 mm
cycle sequence adopted in the previous single-sorbent study (Tyler equiv. 8-mesh and 7-mesh)
(Malek and Farooq, 1997b). Therefore, the assumptions of Bed void fraction 0.4
the extended model are broadly similar to those made in the Adsorbent Silica gel
previous study. Only the modifications are elaborated here. Apparent density* 1.15 g e m - '
The main modification in the present dual-sorbent model Particle porosity* 0.44
Mean pore radius* 20 X lo-' cm (20 A)
is that the adsorbent physical properties, as well as the equi-
Specific surface area* 666 m 2 - g - '
librium parameters, mass-transfer, and heat-transfer kinetics Particle heat capacity** 0.92 J/g.K
for the adsorption of the hydrocarbons in the bed are now Particle size 2.36-2.80 mm
dependent on the location in the bed. In the dual-sorbent (Tyler equiv. 8-mesh and 7-mesh)
PSA studies conducted here, the beds contained an initial Bed void fraction 0.4
section of silica gel adsorbent, which amounted to 24.5% of Adsorbate Mirture of methane, ethane, propane and
butane
the bed length, followed by activated carbon adsorbent that
Carrier Helium
made up the rest of the bed (Figure 1). Hence, in the numeri-
cal solution of the PSA simulation model, the physical prop- Experimental range
Interstitial gas velocity 0.427-0.715 cm/s
erties of the adsorbent, as well as the equilibrium and kinetic Cycle time 600-840 s
parameters, changed at a dimensionless length of 0.245 from Pressure PH = 9.8 bar; PL = 2.8 bar
the bed inlet. It is worth mentioning here that the equilib- Temperature 299.15 K
rium capacity of activated carbon for all the hydrocarbons is
*Experimentally measured in this laboratory. Mean pore radius and spe-
generally greater than that offered by silica gel (Malek, 1996). cific surface area were determined using mercury porosimetry.
Thus, replacing a portion of the activated carbon adsorbent ** Approximated from published literature.
with silica gel resulted in a decrease in the overall adsorptive
capacity provided by the columns for the hydrocarbons. Some
relevant properties of the adsorbents used, the dimensions of
the columns, and some other experimental details are given The fourth hydrocarbon impurity in the feed, butane, is
in Table 1. added to the extended Langmuir isotherm model in order to
predict the multicomponent equilibrium. As with the other
three components, the isotherm parameters for butane have
also been taken from single-component measurements.
Exit from bed The linear driving force (LDF) approximation to model the
particle uptake is also retained. As shown in Malek and Fa-
rooq (1997a) and Malek (1996), for the adsorption of C,-C,
in activated carbon adsorbent, D,,,, = 0.0085 cm2/s, while for
the same in silica gel adsorbent, D,,,,, = 0.0018 cm2/s.
The equilibrium isotherm parameters and mass- and wall
heat-transfer parameters used in (Malek and Farooq, 1997b)
Activated carbon and compiled in Table 2 of that publication are used in the
adsorbent present study without any change. The additional equilibrium
and mass-transfer parameters required in the extended dual-
sorbent model presented here are given in Table 2.
Other than the switch in process parameters at a dimen-
sionless bed length of 0.245 from the inlet, the solution
scheme for the dual-sorbent process is essentially similar to
x = 0.245 the procedure adopted for solving the single-sorbent process
Silica gel model described in the earlier study (Malek and Farooq,
1997b). It is pertinent to note that there are boundary con-
x = 0.0 adsorbent straints of continuous first- and second-order derivatives for

t
Feed to PSA Bed
mass and heat flow at the interface between the two adsor-
bent types (i.e., at x = 0.245) for the dual-sorbent PSA model.
These constraints are intrinsically satisfied if that interface is
between two collocation points. This was achieved with a
choice of 19 collocation points. This choice was also found to
Figure 1. Dual-sorbent PSA bed. provide a suitable compromise between minimizing the mag-

1986 September 1998 Vol. 44, No. 9 AJChE Journal

Table 2. Additional Equilibrium and Kinetic Parameters
Necessary for the Extended Dual-Sorbent Model
Langmuir Isotherm Parameters'
Silica Gel
Methane 4: 1.760
= h, = 5.050X lo-' ( - AH,)= 12.804X lo3
Ethane 4: = 1.611 b,=2.340X10-s (-AH,)=25.6071X103

--
Propane 4: =3.765 b,=1.840X10-4 (-AH,)=20.120X103
Butane 4; -2.344 b,=1.589X10W5 (-AH,)=31.0112X103
Activated Carbon
Butane 9: =3.635 b,=5.568X10-7

'Malek (1996).
Kinetic Parameterf
Silica gel: D,
(-AH,)=44.412X103

= 0.0018
0 5 10 I5
Cycle No.
20 25 30

?, 1
'Malek and Farooq (1997a).
Other equilibrium, mass-, and heat-transfer parameters used in the pre-
sent simulation are same as those listed in Table 2 of Malek and Farooq -s
1
100
t, = 720 s; PH= 9.8 bar
Run: PAC7 8 SACl8; L/vt = 69.9 s
(1997b). (b)

nitude of oscillations in the numerical solution of the differ-

ential equations and the overall computation. The required
CPU time for the solution of the nonisothermal, multicompo-
nent, dual-sorbent PSA process on the Cray 5916 Supercom- L

puter (100 MHz) was expectedly much higher, being about

1,300 CPU s/cycle. On a SGI Power Challenge workstation
running at 200 MHz, the same solution required about 850 0 5 10 15 20 25 30 35
CPU s/cycle. The increase in CPU time was mostly due to an Cycle No.
increase in the number of impurities in the feed and not so
much due to increase in the number of adsorbents. Cyclic Figure 2. Comparison of experimental and simulation
steady state was attained in the simulation in about 130-150 results for (a) single-sorbent and (b)
cycles. dual-sorbent PSA performance with four ad-
sorbable components.
Experimental and Simulation Results : product concentration; +
: product recovery; A : vent
concentration; 0 : vent flow. Lines with open symbols are
Figure 2 shows the experimental and simulation results of corresponding simulation results. Operating conditions are
the six-bed PSA process with four adsorbable components in given in Table 3.
the feed (methane, ethane, propane, and butane). Figure 2a
shows the results for a single-sorbent PSA process (activated vated carbon adsorbent, so much so that it is hardly desorbed
carbon adsorbent), while Figure 2b shows the results for a under the regeneration conditions adopted here. Continuous
dual-sorbent process (silica gel and activated carbon layers). operation under such a condition would naturally degrade the
The operating conditions for the two runs are otherwise the PSA performance, since butane would then accumulate in the
same and are given in Table 3. Experimental run number beds (and penetrate deeper) and the total adsorption capac-
PAC6 is for a single-sorbent operation, while run number ity provided by the beds for the other hydrocarbon compo-
PAC7 is for a dual-sorbent operation. The main difference in
the results of these two experimental runs is that the latter
Table 3. Operating Conditions of Single- and Dual-Sorbent
gives a somewhat earlier breakthrough. This is to be expected PSA Experimental and Simulation Runs*
since less activated carbon adsorbent is available in the dual-
sorbent process to retain the lighter hydrocarbon compo-
nents in the feed. However, of particular importance is that
throughout the experimental run the analysis of the vent
stream (i.e., the blowdown and purge gas) composition for PAC6 9.8 2.8 0.573 0.185/0.067/0.014/0.005 35/50/35
the single-sorbent process showed only a small presence of PAC7 9.8 2.8 0.573 0.185/0.067/0.014/0.005 35/50/35
butane, whereas for the dual-sorbent process, butane was re- Numerical Simulation
jected more appreciably, together with the other hydrocar- SAC17 9.8 2.8 0.573 0.185/0.067/0.014/0.005 35/50/35
bons, during the blowdown and purge steps. Figure 3 shows a SAC18 9.8 2.8 0.573 0.185/0.067/0.014/0.005 35/50/35
breakdown of the experimentally measured vent stream hy- SAC19 9.8 2.8 0.427 0.185/0.067/0.010/0.005 35/50/35
SAC20 9.8 2.8 0.715 0.185/0.067/0.010/0.005 35/50/35
drocarbon composition for the two runs as a function of cycle SAC21 14.0 2.8 0.427 0.170/0.065/0.010/0.005 35/50/35
number as determined by analyzing the collected samples us- SAC22 18.0 2.8 0.427 0.170/0.065/0.010/0.005 35/50/35
ing a Perkin Elmer Autosystem gas chromatograph fitted with SAC23 22.0 2.8 0.427 0.170/0.065/0.010/0.005 35/50/35
a TCD and a Supelco Carboxen-1004 micropacked column.
*PAC6 and SAC17 involved single-sorbent PSA operation while all the
Absence of butane in the vent stream for the single-sorbent other runs involved a dual-sorbent process. All runs conducted with feed
run indicates that butane is strongly adsorbed by the acti- and column wall temperature of 299.15 K.

AIChE Journal September 1998 Vol. 44, No. 9 1987

 3 100 100

-
x"
2.5

2
n n n
E
$
90
1
t f l L 90 9
P
0

5?
B8
1.5 80

J l
.
a
80

-2.
u^
U
0.5
2 70 70
a
0
60 60
4 8 12 16 20 25 30 5 10 15 20 25
Cycle number PH[bar1

Figure 4. Product recovery and purity obtainable using

3 , the dual-sorbent, 6-bed PSA process as func-
r- I I
tion of PH.
Runs SAC19, SAC21. SAC22, and SAC23. Solid lines: non-
isothermal operation; dashed lines: isothermal operation.
Operating conditions are given in Table 3.
21.5

x" 1

0.5
for the four-component, dual-sorbent PSA system generally
show a slightly lower product purity for the same operating
0 pressure as compared to the three-component, single-sorbent
4 a 12 16 10 25 system. On the other hand, product recovery is about the same
Cycle number for the two operations. The purity results are expected since
in the dual-sorbent operation, there is slightly earlier break-
Figure 3. Breakdown of hydrocarbon composition in through of the lighter hydrocarbons due to a reduction in the
vent stream as a function of cycle number for total adsorptive capacity in the bed as a result of partial re-
(a) single-sorbent (Run PACG) and (b) placement of activated carbon with silica gel. It is also appar-
dual-sorbent (Run PAC7) PSA processes. ent that the amount of silica gel added in the column is just
sufficient to prevent penetration of butane into the activated
carbon section which could thereby cause rapid degradation
nents would diminish with time. In contrast, silica gel desorbs of the separation characteristics.
butane relatively readily. Ease or difficulty of desorption de- It is interesting to consider, at this point, the effect of
pends on the isotherm shape and butane isotherm is more heat-transfer dynamics on the PSA performance. All the sim-
favorable on activated carbon than on silica gel (Malek, 1996). ulation results obtained earlier were for a nonisothermal col-
Hence, the dual-sorbent system is more appropriate for oper- umn operation. In other words, there is a finite heat-transfer
ations where heavier molecular weight hydrocarbons are pre- resistance at the column wall. As a result, the bed tempera-
sent. ture profile varies with the cycle step. The simulation results
The simulation results obtained (Figure 2) indicate that the for the product recovery and purity obtainable for the six-bed
mathematical model can closely predict the experimental PSA PSA process under isothermal conditions are also shown in
performance of the dual-sorbent system with four adsorbing Figure 4. In this case, there is no temperature variation in
components in the feed. the bed, as it remains constant at all points and at every cycle
As a further study of the dual-sorbent PSA process, addi- step. The bed temperature is set equal to the feed tempera-
tional simulations were conducted (Table 3) to investigate the ture. The PSA simulation model can be easily configured to
effects of various process parameters. Generally, the dual- satisfy this condition by setting the temperature derivative
sorbent system has an earlier breakthrough time as compared equation to zero.
to the single-sorbent system (Malek and Farooq, 1997b). Under isothermal conditions, a higher product purity is
However, when the feed stream contains heavier hydrocar- generally obtainable as compared to nonisothermal opera-
bons, the merit of the dual-sorbent system, as manifested in tion. The difference in purity is largest at low operating pres-
Figure 3, far exceeds the marginal loss in total bed capacity. sure. The purity for the isothermal case gradually approaches
Figure 4 shows the simulated results of the effect of the that of the nonisothermal condition at higher operating pres-
high operating pressure, PH, on product recovery and purity sure. This is expected since at the higher operating pressure
for the four-sorbate, dual-sorbent PSA system. These plots there is very little breakthrough of hydrocarbons in both cases.
should be compared with the corresponding results for a feed The product recovery is slightly lower at low operating pres-
gas containing three adsorbable components as shown in our sure ratio for the isothermal case, but quickly exceeds that
previous publication (Figure 15 in Malek and Farooq, 1997b). obtained under nonisothermal operation as the operating
Figure 4 shows that the product recovery drops from about pressure is increased. The difference in the plots for the
80% at 9.8 bar to just above 70% at about 16.0 bar, after isothermal and nonisothermal operations is a clear indication
which it levels off as the pressure is increased. There is a of the importance of heat effects in the PSA system being
gradual increase in product purity with pressure. The results studied.

1988 September 1998 Vol. 44, No. 9 AIChE Journal

Performance of the Industrial PSA Unit
Methane Breakthrough
The six-bed PSA process in operation at a refinery in Sin- maamo o a
Helium Carrier
gapore utilizes three different adsorbents, namely, activated 0.8 -- 0 Hydrogen Carrier
alumina, silica gel, and activated carbon. The activated alu-
mina layer, which amounts to less than 5% of bed volume, is -x"
x'
0.8 -- 9

used essentially to remove water vapor in the feed stream if

any is present. In this study, it is assumed that dry feed en-
0.4 --

0.2 -- 9
..
0
0

ters the bed and that the activated alumina layer does not a
affect the bed capacity for the other hydrocarbon con-
stituents in the feed stream to any appreciable extent. Thus
only two layers of adsorbents are considered, namely, silica
gel and activated carbon.
Figure 6. Comparison of methane breakthrough in acti-
With respect to the cycle sequence, the operation of the vated carbon with helium and hydrogen as
industrial H,-PSA unit follows the same steps and flow con- carriers.
figuration used in the laboratory study. The simulation model
for the industrial PSA process is the same as that for the
dual-sorbent, four-component operation described earlier. perature. The isotherm is, however, clearly linear over the
Hydrocarbon constituents in the feed stream other than C,-C, range of interest. Figure 6 shows a comparison of two
are ignored in this study since they are present in trace methane breakthrough curves obtained under the same oper-
amounts, if at all. Apart from the bed dimensions, two other ating conditions, but with helium as carrier in one and hydro-
important differences that need to be considered in the in- gen as carrier in the other. In this case, the two curves are
dustrial operation are the heat-transfer characteristic and the essentially similar, indicating a negligible effect of hydrogen
fact that hydrogen forms the bulk of the gas phase, instead of on the equilibrium or kinetics of hydrocarbon adsorption on
helium. Due to the small ratio of heat-transfer area to vol- activated carbon. Based on these studies, the following as-
ume, the industrial PSA process operates more like an adia- sumptions are made for the simulation model of the indus-
batic process. As far as the simulation model is concerned, trial PSA process with hydrogen as the bulk gas:
the required modification to account for this difference lies a. Hydrogen has a very low adsorption capacity in acti-
in the value of the wall heat-transfer coefficient, h,. For adi- vated carbon. The isotherm is linear and does not affect the
abatic operation. h , is set to zero in the simulation model. capacity of the hydrocarbons, which are more strongly ad-
The use of helium in place of hydrogen as the carrier gas sorbed.
in the laboratory experiments is the second important depar- b. Hydrogen is practically inert to silica gel.
ture from the industrial operation. Limited experiments were c. Mass-transfer resistance for hydrogen is essentially
conducted to study the impact of the presence of hydrogen negligible in both silica gel and activated carbon.
on both the equilibrium and kinetics of hydrocarbon adsorp- d. Hydrogen does not affect the adsorption kinetics of
tion. Figure 5 shows the equilibrium adsorbed amount of hy- the hydrocarbons in both activated carbon and silica gel.
drogen in activated carbon. The equilibrium amount was de- Hence, in the simulation model, the extended Langmuir
termined using both the dynamic column breakthrough and model is used for computing multicomponent equilibria of
constant flow equilibrium desorption (DCBT and CFED) ex- hydrocarbon adsorption on both silica gel and activated car-
periments (Malek and Farooq, 1996a). Although the ad- bon, with a linear isotherm model used for hydrogen adsorp-
sorbed amount is small, there is some adsorption capacity for tion on activated carbon. For silica gel, only the macropore
hydrogen in the sorbent. This is clear from the fact that the volume capacity of hydrogen was taken into account, as was
calculated adsorbed amount is higher than the amount pre- done for helium in all previous simulations. The Henry's Law
sent in the particle voids under the given pressure and tem- parameters for the hydrogen-activated carbon adsorption
equilibrium are given in Table 4. The Henry's Law equation
is given as follows:

: 26.0 'C
0.6 .. "65.0°C

As with the other components, the LDF model is adopted

to characterize particle uptake. Since mass-transfer resist-
ance for hydrogen adsorption is assumed to be negligible, an
equilibrium model is used for hydrogen uptake as follows:

Table 4. Henry's Law Isotherm Parameters for Adsorption of

Figure 5. Adsorption equilibrium data of hydrogen on Hydrogen on Activated Carbon
activated carbon at two different tempera-
Isotherm Parameters
tures.
Lines: Henry's Law fit. K-, = 7.690X lo-' ( - A H A ) , =5.19OX1O3

AIChE Journal September 1998 Vol. 44, No. 9 1989

 Table 5. Operating Conditions for Results Shown in Figure 7

18.0 69.0 311.15 0.20/0.07/0.06/0.03 35/58/35

Nov. 22, 1992 18.0 71.4 306.15 0.15/0.04/0.02/0.01 32/52/32
Nov. 23, 1992 18.0 74.1 307.15 0.15/0.06/0.03/0.016 32/52/32
with yH = 1 for the hydrogen component. Thus, in the nu- Simulation
merical solution, Eq. 2 above provides an additional term in 9.8 - 22.0 69.8 308.15 0.17/0.065/0.02/0.01 35/50/35
solid loading calculation.
Similar to the simulation model for the laboratory unit, it
is assumed that there is no interaction among the diffusing Figure 4 for the laboratory-scale six-bed, dual-sorbent PSA
molecules of the adsorbable components. The pressure unit. The general features of the plots for the industrial sys-
changing steps are modeled using linear driving force, with tem are similar, showing an increasing trend in product pu-
the proportionality coefficient chosen to fit measured plant rity and a drop in recovery with increasing presmre, with the
bed pressure profiles. PH and PL are defined operating pa- recovery leveling off at just above 70% at pressures exceeding
rameters. The same cycle sequence as described earlier is 15 bar. In the industrial system, however, the product purity
adopted for the industrial process. The substage durations is slightly lower than that for the nonisothermal, laboratory-
for the simulations were taken to be equal to the durations scale operation. For the industrial H,-PSA unit, it is also ap-
used in the actual H,-PSA operation. parent that no advantage exists in operating the unit beyond
Based on these assumptions, a series of numerical simula- a maximum pressure of about 18.0 bar, since the improve-
tions of the H,-PSA unit were conducted. The required CPU ment in product purity beyond that pressure is marginal.
time for the simulation of the industrial, adiabatic, multicom- Figure 7 also includes the isothermal model results for the
ponent, dual-sorbent H,-PSA process on the Cray 5916 Su- industrial unit. As seen for the laboratory unit, the isother-
percomputer was about 1,500 CPU s/cycle. mal model also generally predicts higher product recovery and
Figure 7 shows the results of the simulations in terms of purity for the industrial unit. It is interesting to note that the
product recovery and purity. The design specification for the recovery and purity data for the design case are closer to the
industrial H,-PSA unit as well as two sets of actual operating isothermal simulation results, although the predicted recov-
data obtained from the plant are also included in the figure ery is somewhat lower. The distinction is less definite for the
for comparison. Table 5 provides the operating conditions for limited actual plant data, but it is perhaps not unreasonable
the plant data and simulation results. According to the de- to conclude that the isothermal model will most likely lead to
sign specification, the industrial unit should be operated at a an underdesign of the PSA unit.
PH of 18.0 bar and a cycle time, t,, of about 750 s. However, In the actual industrial PSA process, the two other impor-
more often than not, the cycle time is reduced in actual PSA tant operating variables are the feed velocity to the columns,
operations in order to contain the hydrocarbon fronts within uf,and the cycle time (specifically the duration of the high-
the beds. The simulation model predictions, which are based pressure adsorption step). It would, therefore, be beneficial
on approximately the mean of these operating conditions, are to study the effects of these variables on the PSA perform-
in good agreement with the actual performance. ance as quantified by the product recovery and purity. Figure
The recovery and purity plots shown in Figure 7 should be 8 shows the variations in the recovery and purity as functions
compared with the nonisothermal simulation results shown in of L/ur and the PSA cycle time, t, . The operating conditions
for the simulation results in the figure are given in Table 6.
The simulation results show that the product recovery in-
creases with feed velocity (decreasing L/uf ratio), as ex-
pected. The recovery is also generally a monotonically in-
creasing function of cycle time. On the other hand, the de-
pendence of product purity on the feed velocity and cycle
time is slightly more intricate. Increasing cycle time sharpens
the concentration front and increases product purity when
operating far from breakthrough. In this range, the bed is not
fully utilized, which reduces the practical importance of the
gain in purity. Increasing both feed velocity and cycle time
reduces the product purity in the chosen operating range. The
quantitative decrease in purity is dependent on the extent of
breakthrough of the hydrocarbon components. For the lower
Figure 7. Comparison of product recovery and product
velocity run, there is insignificant breakthrough of the hydro-

.:
purity from plant design specification, mea-
carbons, and, thus, the purity drops only marginally, remain-
sured plant data and simulation model.
ing above 95%. At higher velocities, the hydrocarbons begin
recovery from design specifications; + : recovery from
plant data (11/23/92); A : recovery from plant data to break through to a larger extent, as shown by the plots for
(11/22/92). Open symbols: product purity; solid lines: simu- L/ur = 93.7 s and L/uI = 71.4 s. For the former case, there is
lation results of adiabatic operation; dashed lines: simula-
tion results of isothermal operation. Operating conditions
a rapid drop in the purity due to the sharp breakthrough of
are given in Table 5. methane as the cycle time is increased. For the latter condi-

1990 September 1998 Vol. 44, No. 9 AIChE Journal

 Slllca gel Activated carbon Methane
section 0 Ethane
Propane
.- 0.6
C A
Butane

s 0.2 0 Cycle No. 4

0 7
A
= , - = , 5 -
- , --
0 0.2 0.4 0.6 0.8 I
Dimensionless axial position

I , I

0 0.2 0.4 0.6 0.8 1

500 600 700 800 900 1000 Dimensionless axial position
tc [SI
I
Figure 8. Effect of cycle time, t,, on (a) product recov- Cycle No. 50
ery and (b) product purity at different Llv, ra- *-0.8
UI
~~

tios for dual-sorbent PSA process. C

5 0.6
Operating conditions are given in Table 6. 0
-
0
e
0.4
-
tion, on the other hand, there is substantial breakthrough of ::0.2
hydrocarbons, with complete methane breakthrough, even at 0
a short cycle time as a result of the high feed velocity. Hence, 0 0.2 0.4 0.6 0.8 1
the purity for this case is consistently low. Figure 8 provides a
Dimensionless axial position
useful guide for the proper operation of the industrial PSA
process in terms of maintaining high product recovery as well
as purity by varying the relevant operating conditions. In par-
ticular, it is possible to achieve very high ( 2 99.9%) product Cycle N O ~ O
~~

purity by operating at low cycle time and feed velocity. 6

C
The movements of the concentration wave fronts in the
adsorbed phase as a function of cycle number are shown in
e
Table 6. Operating Conditions for Dual-Sorbent PSA
g 0.2 a 0

0 0
O E i Z
Simulation Runs in Figure 8* 04 A - D

0 0.2 0.4 0.6 0.8 1

pti PL L/Uf
Run No. (bar) (bar) (S) Xf d X f d X f d X f 4 tcl/tc2/tc3 Dimensionless axial position
SAC24 18.0 2.8 140.4 0.17/0.065/0.03/0.01 35/30/35
SAC25 18.0 2.8 140.4 0.17/0.065/0.03/0.01 35/50/35
SAC26 18.0 2.8 140.4 0.17/0.065/0.03/0.01 35/70/35 Figure 9. Penetration of hydrocarbon fronts in the PSA
SAC27 18.0 2.8 140.4 0.17/0.065/0.03/0.01 35/90/35 bed.
SAC28 18.0 2.8 93.7 0.17/0.065/0.03/0.01 35/30/35 Run SAC25; operating conditions are given in Table 6; dot-
SAC29 18.0 2.8 93.7 0.17/0.065/0.03/0.01 35,50/35 ted line indicates the interface between silica gel and acti-
SAC30 18.0 2.8 93.7 0.17/0.065/0.03/0.01 35/70/35 vated carbon sections.
SAC31 18.0 2.8 93.7 0.17/0.065/0.03/0.01 35/90/35
SAC32 18.0 2.8 71.4 0.17/0.065/0.03/0.01 35/30/35
SAC33 18.0 2.8 71.4 0.17/0.065/0.03/0.01 35/50/35 Figure 9 for one of the simulation runs with a low feed veloc-
SAC34 18.0 2.8 71.4 0.17/0.065/0.03/0.01 35/70/35
2.8 71.4 0.17/0.065/0.03/0.01
ity. In this case, the butane front is well retained within the
SAC35 18.0 35/90/35
silica gel section. Furthermore, there is no breakthrough of
'All runs conducted with feed and column wall temperature of 308.15 K. methane from the PSA bed. On the other hand, some pene-

AIChE Journal September 1998 Vol. 44, No. 9 1991

tration of the butane front into the activated carbon section more realistically modeled as an adiabatic system since the
as well as partial breakthrough of the methane front at the heat-transfer area to volume ratio of the bed is relatively
bed exit were observed when the velocity was increased (for small. The second modification concerns the effect of hydro-
example, run SAC29 in Table 6). gen as the carrier gas in the industrial operation as compared
It is interesting to note that the main effect of increasing to the use of helium in laboratory experiments. Helium had
feed velocity to the PSA unit for the same cycle time is the been used in the experiments as a matter of safety. Experi-
increasing distance penetrated by the various hydrocarbon mental studies show that the presence of hydrogen has little
concentration fronts in the bed during the high-pressure ad- effect on the equilibrium and kinetics of hydrocarbon adsorp-
sorption step. In particular, for the H,-PSA system studied tion on activated carbon. However, hydrogen has a small ad-
here, it is highly desirable to ensure that the butane front sorption capacity in activated carbon and its isotherm varies
does not penetrate beyond the silica gel section and that the linearly with the partial pressure. Predictions of plant per-
methane front is retained within the activated carbon section formance from the simulation model are in very good agree-
as the cycle progresses. As discussed earlier, butane adsorp- ment with operating plant data. In addition, comparison with
tion on activated carbon is almost irreversible and significant isothermal simulation results also reveal the importance of
penetration of the front beyond the silica gel section would, heat effects in the industrial operation. The simulation model
therefore, diminish the capacity of activated carbon for re- will be useful for general PSA process studies as well as for
taining the other (lighter) hydrocarbons as the PSA operation the optimiLation of the plant performance.
continues. Also, as shown earlier, the methane front in the
bed is relatively sharp and a breakthrough of the front would Notation
result in a tremendous drop in product purity. Thus, a proper
b, =equilibrium isotherm parameter, bar-’
way of operating the H,-PSA system for treatment of refin- D,,,,, =average effective diffusivity, cm’/s
ery fuel gas is to ensure the containment of the butane (and h , =effective wall heat-transfer coefficient, J/cm’. s . K
other heavier molecular weight hydrocarbons) within the sil- K,, =model parameter for linear equilibrium isotherm: mmol/g
ica gel section and the methane front (and, hence, the other bar
hydrocarbon components) within the activated carbon sec- L =column length, cm
q =adsorbed phase concentration, mmol/g
tion. 4* =equilibrium adsorbed phase concentration, mmol/g
q: =saturated equilibrium adsorbed phase concentration,
mmol/g
Conclusions qT, =saturated equilibrium adsorbed phase concentration of
A comprehensive study of the performance of a multicom- methane in activated carbon, mmol/g
ponent, six-bed, dual-sorbent PSA process for hydrogen pu- R = universal gas constant, = 8.311 J/mol. K
t,l,t,2,t,3 =duration of stages in a PSA cycle, s
rification from refinery fuel gases has been conducted. The T =temperature, K
PSA beds consist of an initial layer of silica gel used to trap u =interstitial gas velocity, cm/s
heavier hydrocarbons and a second layer of activated carbon X =adsorbate mole fraction in gas phase
adsorbent to trap the lighter hydrocarbons. A simulation Y =4/q:, dimensionless solid phase concentration
,y =dimensionless axial distance
model of the process has been developed and verified based
Y* =qE”/sT1
on some experiments performed using a laboratory-scale six- 0 = T/Tf, dimensionless temperature
bed PSA unit. 7 = uff/L, dimensionless time
Comparison between the single-sorbent (i.e., activated car- ( - A H A )= heat of adsorption, J/mol
bon only) and dual-sorbent operations indicates the advan-
tage of the latter system where higher molecular weight hy- SUbSCriptS
drocarbons like butane are present in the feed. Butane is very B =adsorbent bed
strongly adsorbed on activated carbon and is difficult to de- 1,2,3,4 =methane, ethane, propane, and butane, respectively
sorb by simple pressure reduction. This is, however, not the f =feed or inlet conditions
i, k =adsorbate components in multicomponent adsorption
case with its adsorption on silica gel. Hence, the initial layer
of silica gel effectively prevents the gradual degradation of
the activated carbon layer. Literature Cited
Other parametric studies show that the dual-sorbent sys- Malek, A,, “A Study of Hydrogen Purification from the Refinery Fuel
tem results in earlier breakthrough of the lighter hydrocar- Gas By Pressure Swing Adsorption,” PbD Thesis, Dept. Chem.
Eng., National University of Singapore (1996).
bons in contrast to single-sorbent operations due to a reduc- Malek, A,, and S. Farooq, “Determination of Equilibrium Isotherms
tion in the total adsorptive capacity compared to the former Using Dynamic Column Breakthrough and Constant Flow Equilib-
system. In particular, it is pertinent to control the operating rium Desorption,”J. Chem. Eng. Data., 41, 25 (1996a).
conditions so as to retain the butane front within the silica Malek, A,, and S. Farooq, “Comparison of Isotherm Models for Hy-
gel layer and the methane front within the activated carbon drocarbon Adsorption on Activated Carbon,” AIChE J., 42, 3191
(1996b).
layer. Furthermore, comparison with isothermal simulation Malek, A., and S. Farooq, “Kinetics of Hydrocarbon Adsorption on
results reveals the importance of heat effects in the PSA beds Activated Carbon and Silica Gel,” AIChE J., 43, 761 (1997a).
for this process. Malek, A,, and S. Farooq, “A Study of a 6-Bed Pressure Swing Ad-
As a further study, the dual-sorbent model was modified to sorption Process,” AIChE J., 43, 2509 (1997b).
simulate an actual industrial six-bed H,-PSA system in oper-
ation at a refinery in Singapore. Two important modifications
were considered. The first is that the industrial operation is Manuscript receiued Nou. 3, 1997, and retiision receitied June 17, 1998

1992 September 1998 Voi. 44, No. 9 AIChE Journal