You are on page 1of 24

Adsorption and its Applications in Industryand EnvironmentalProtection

Studies in Surface Science and Catalysis,Vol. 120


A. Dabrowski (Editor) 347
9 1998Elsevier ScienceB.V. All rights reserved.

Nitrogen separation from air by pressure swing adsorption

N. O. Lemcoff

The BOC Group, 100 Mountain Avenue, Murray Hill, NJ 07974-2064, USA

1. ABSTRACT

Pressure swing adsorption is commercially used for the separation of air as an


alternative to the conventional cryogenic separation process. A zeolite based
process, in which the adsorbent shows preferential adsorption of nitrogen over
oxygen under equilibrium conditions, is used for the production of oxygen. On the
other hand, the separation of air for the production of nitrogen is carried out by
pressure swing adsorption over a carbon molecular sieve. The separation is
kinetically based, since the equilibrium adsorption of both oxygen and nitrogen is
very similar, but the oxygen is adsorbed faster. In conventional pressure swing
adsorption systems, both the productivity and yield decrease significantly as
nitrogen purities are increased beyond 99.9 %. Although higher nitrogen purities
were achievable, product flows were so low that it was not an economically viable
process. Only recently, through both improvements in the manufacture of carbon
molecular sieves and new cycle developments, higher performance is achieved in
the high purity region. A theoretical dynamic model developed for the kinetic
separation of nitrogen from air is presented, and the effect of the different process
variables and cycle steps on the process performance is analyzed. A comparison
between the experimental performance and the model predictions is discussed.

2. INTRODUCTION

Gaseous nitrogen is used in the chemical and petroleum industries for storage
tank blanketing and vessel inerting applications. It is also used extensively by
the electronics and metals industries for its inert properties. It is also widely used
in the food industry to retain food freshness longer. Commercial nitrogen is
produced by a variety of air separation processes, including cryogenic liquefaction
and distillation, adsorption separation, and membrane separation.
All adsorption separation processes involve two principal steps: adsorption,
during which one or more species are preferentially adsorbed, and regeneration,
during which these species are removed from the adsorbent. When the
regeneration step is carried out through the reduction of the total pressure, the
348

process is called pressure swing adsorption (PSA), and it has become a subject of
interest in gas separations because of its low energy r e q u i r e m e n t and cost. The
basic steps involved in a cycle are typically pressurization, high-pressure
adsorption, equalization, countercurrent blowdown and low-pressure desorption.
PSA processes have found widespread application in hydrogen purification, air
drying, and air separation [1,2]. In the first two cases, the separation takes place
due to the difference in the equilibrium adsorption isotherms of the components
of the gas mixture. On the other hand, the separation of nitrogen from air on a
carbon molecular sieve, is kinetically controlled. The adsorption equilibrium
isotherms of oxygen and nitrogen are almost identical, and the preferential
sorption of oxygen in these adsorbents results from the faster adsorption of t h a t
species on the carbon molecular sieve.
The objective of this chapter is to study the effect of different process variables
on the performance of a rate induced PSA process, and to compare the
experimental results with the predictions of a theoretical model. The effect of
process variables and cycle steps on the t e m p e r a t u r e and concentration profiles is
also studied, and the relationship between the profiles and the process
performance is discussed.

3. CARBON MOLECULAR SIEVES

Carbon molecular sieves (CMS) have played a critical role in the


commercialization of the pressure swing adsorption process for the separation of
nitrogen from air. They differ from activated carbon mainly in the pore size
distribution and surface area. While activated carbons have a broad range of
pores, with a typical average pore diameter of 20 A, carbon molecular sieves have
a more narrow pore size distribution, with pore sizes in the range of 3 - 5 A. A
molecular probe method is one of the best approaches to determine the effective
micropore size distribution of carbon molecular sieves [3,4]. Typical surface areas
for a carbon molecular sieve are in the range of 250-400 m2/g, while the micropore
volume is about 0.15-0.25 cm3/g [2,5].
One of the earliest publications on the preparation of carbon molecular sieves
was done by Walker et al. [6]. They started from a S a r a n copolymer, but
currently m a n y different raw materials are being used either for research or
commercial purposes. The main sources of carbon molecular sieves for
commercial applications are coal and coconut shell. An extensive list of the work
done in the preparation of carbon molecular sieves can be found elsewhere [7,8].
The typical steps in the preparation of carbon molecular sieves from coal are
shown in Fig. 1 [8]. It can be seen t h a t the main steps are oxidation,
carbonization, followed either by steam activation or pyrolysis. During steam
activation, pores are opened and the resulting carbon molecular sieve can be used
in hydrogen purification, while during pyrolysis carbon deposition takes place,
and a narrower pore size distribution is achieved.
349

Coal ~ Pulverizing ~ Oxidation by Air ~ Oxicoal

Pelletizing ~ Binder

Carbonization ~ Scrubbing

Steam Activation ~ Uniform Initial Material ~ Pyrolysis

5 5

CMS H2 CMS N2

Figure 1. Schematic diagram showing the processes involved in the manufacture of carbon
molecular sieve adsorbents.

A recent development was to prepare carbon based molecular sieves which are
funcionalized with inorganic oxides and supported metals. The main objective is
to combine the molecular sieving properties of the carbon with surface chemical
and physical properties of the inorganic oxides in one composite structure.
S h a r m a and Seshan [9] reported copper modified CMS for the selective removal
of oxygen at t e m p e r a t u r e s below 200oC. At higher t e m p e r a t u r e s , oxygen can be
reversible adsorbed only if present in trace amounts.

4. A D S O R P T I O N RATE

The separation of nitrogen from air is based on the different uptake rate of
nitrogen with respect to oxygen. Typical adsorption isotherms and uptake curves
on carbon molecular sieves are shown in Figs. 2 and 3. Selectivities for carbon
molecular sieves are in the range between 20 and 50 [7,10,11]. This difference in
adsorption kinetics is thought to be related to the different molecular size. The
kinetic d i a m e t e r of oxygen (3.5 A) is slightly smaller t h a n t h a t of nitrogen
(3.6/k).
350

.4 --

=2 -- I
o
o

w_ .~ ,= ~
"0
.0 A

tll
0.8

0.6
--
j" , ..

0.4
0
E 0.2

0 I I I I

5 10 15 20
Pressure (atm)

Figure 2. Adsorption isotherms on carbon molecular sieve at 25~


oxygen nitrogen.

0.8

== 0.6

-~ 0.4 .~ .-'"
0 ~ ~

:~ 0.2 ..'~
U.
0 ~ ~176 I I I

0 50 100 150

Time (s)

Figure 3. Adsorption kinetics on carbon molecular sieve at 25~


oxygen nitrogen.
351

The adsorption kinetics of oxygen and nitrogen on CMS were studied using
different experimental methods: volumetric, gravimetric, chromatographic,
t h e r m a l desorption, and isotope exchange [11]. From the analysis of the
literature it follows t h a t the transport process controlling the uptake rate seem to
obey to two different mechanisms, either a surface barrier at the micropore
entrances, or a diffusional molecular transport within the bulk of the micropores.
R u t h v e n [12] analyzed the diffusion of nitrogen and oxygen in carbon
molecular sieves t h a t respond to Fickian diffusion. They found that, outside the
linear region of the adsorption isotherm, the diffusivity increases strongly with
concentration, following the Darken equation. The actual values of the corrected
limiting diffusivities for oxygen and nitrogen at 300K are Do/r 2 = 2.3 x 10 .3 and
6 x 10 -~ l/s, respectively [12-14]. The corresponding activation energies are 25.1
and 27.2 kJ/mole. An order of magnitude smaller diffusivity values were
m e a s u r e d by other researchers [15], and this was attributed to a difference in the
activation t e m p e r a t u r e of the sieve samples.
Fitch et al. [16] studied the uptake of oxygen and nitrogen in a non-Fickian
carbon molecular sieve. It is a s s u m e d t h a t carbon crystallites are small in size
and are loosely cross-linked with other crystallites to form an aperture-cavity
structure. The aperture has a slit shape which serves as an entrance to the
cavity. It is suggested t h a t the sieving action is produced by the interaction of
Van der Waals potential forces in the slit and the molecules. An experimental
characteristic of these sieves is t h a t the half-time of the adsorption process
depends on the equilibrium coverage, and hence, depends on the adsorption
pressure. The slit potential activation energies determined by these studies were
0.10 and 15.5 kJ/mole, respectively. These values are close to the energy barrier
values calculated by Rao et al. [17] assuming a two graphite layer as a model for
the slit.
Recently, different approaches to describe the kinetics of oxygen and nitrogen
sorption in carbon molecular sieves have been presented. It has been shown t h a t
they can lead, as particular cases, to the two above mentioned mechanisms.
Srinivasan et al. [18] proposed a surface barrier model of diffusion under a
chemical potential gradient, while Trifonov and Golden [19] presented a
theoretical approach based on a hopping mechanism, in which the molecules
jump from occupied to free adsorption sites. Seaton et al. [20] carried out
molecular dynamics simulation of diffusion in individual pores and in pore
networks. The experimental selectivity can be reproduced both at the level of
individual pores, as well as with pore networks having a wide range of pore sizes.
Carbon surfaces exhibit a strong affinity for oxygen even at room temperature.
Aging of the carbon molecular sieve has been attributed to the oxidation of the
carbon surface, which leads to a reduction in the pore or slit size, and a reduction
in the gas uptake rates. Also, sorption of water at or near the pore slits of a CMS
can affect the diffusion rate of gases during the adsorption and desorption steps,
and therefore significantly change its performance in a pressure swing adsorption
352

system. Passivation of active sites has been suggested to achieve a stable CMS
surface [21).

5. P R E S S U R E SWING A D S O R P T I O N P R O C E S S

In actual operation, pressure swing adsorption systems are composed of two


identical beds containing carbon molecular sieve that perform the separation
process continuously. Compressed air enters the bottom of the first bed where the
sieve adsorbs oxygen and other impurities, letting high-purity nitrogen pass
through. When this bed approaches saturation it switches to a regenerative
phase, while the second bed automatically begins the separation process. The
regeneration is accomplished by reducing the pressure, allowing the pelletized
bed material to vent the released gases. After a few minutes, the non-nitrogen
components have been vented from the bed and it is ready to separate air again.
In this way, high purity nitrogen is produced without an interruption. PSA
systems are used primarily for applications that require relatively steady flows of
gaseous nitrogen at rates of approximately 30 - 3,000 Nm3/h. Typical operation
conditions of nitrogen pressure swing adsorption systems are: adsorption
pressure, 8 atm, regeneration pressure, 1 atm, half cycle time, 1-2 min.
The PSA performance is determined in terms of the specific product and yield
at a certain nitrogen product purity. The specific product is defined as the
product flow rate per unit adsorbent volume, while the yield is the ratio between
the net flow rate of nitrogen produced and the flowrate of nitrogen fed to the
system. The specific feed, in turn, is defined as the feed flow rate per unit of
adsorbent volume.
In order to avoid deterioration of the carbon molecular sieve due to water
adsorption, a drying system is generally used. The drying system can be either
external or internal to the vessels. In the first case, a chiller is used to reduce
moisture. In the second case, a desiccant is used which dries out the gas
internally, without the need of bulky external dryers. However, the materials
used as desiccants are all polar materials and, specially for zeolites, have a
selective adsorption of nitrogen over oxygen. Golden et al. [22] developed a
composite adsorbent capable of both dehydrating air and selectively adsorbing
oxygen over nitrogen. An increase in the process performance was achieved by
using LiC1 and silica impregnated carbon molecular sieves. The dehydrating
agents reside in the macropores where water adsorption takes place.
Up to recently, nitrogen purities from commercial pressure swing adsorption
processes were below 99.95%. In order to get higher purities, an additional step of
reduction of the oxygen impurities was required. The low purity nitrogen product
is fed together with a fuel, such as hydrogen, to a deoxygenation unit, where the
oxygen impurities are converted over a catalyst. Palladium or platinum
impregnated alumina are generally used. Purities as high as 99.9999% can be
achieved, but the disadvantage is that humidity and excess fuel are introduced
into the nitrogen product. Although this may not be a problem in some
353

applications, such as reducing atmospheres, it still requires the availability of the


fuel. An alternative way of achieving higher nitrogen purities is the use of
vacuum during the regeneration of the adsorbent beds. As the ratio between the
adsorption and desorption pressures increases substantially, both productivity
and yield increase significantly. However, the addition of another rotating unit is
a disadvantage not only from the capital cost, but also from the maintenance
point of view.
Oxygen selective sorbents have been developed by researchers in different
institutions over the past decades. These metal complexes mimic certain
biological oxygen carriers that selectively remove oxygen from air. Cobalt, iron
and manganese based materials have been developed with capacities of between
1 and 10% oxygen by weight. However, the most serious drawback is their
chemical instability caused by autoxidation reactions. Either the ligand or the
central atom is oxidized, or peroxo-bridged dimers are formed, and their oxygen
adsorption capacity is lost [23].
Nitrogen PSA performance has improved considerably in the last 20 years.
Since 1980, the power consumption has dropped to a half and the productivity
doubled, resulting in both smaller air compressors, and a reduction in the size of
the units. These advances, which have allowed to commercialize pure pressure
swing adsorption systems supplying nitrogen with purities higher than 99.999%,
arise from two different areas: molecular sieve developments and process
improvements [24].
Probably, the most important advances have come as a result of carbon
molecular sieve improvements, but there have been also significant
improvements due to the better use of the sieve by modifications to the PSA
process. Early improvements comprise the incorporation of additional steps to the
basic cycle. In the pressure equalization step, both beds are interconnected to
transfer gas from the high pressure bed to the low pressure bed, and therefore
save both air feed and compression energy [14,25]. During this step, the gas at
the product end of the bed is pushed back, acting like a purge, and therefore
improving both purity and recovery. The pressure equalization step plays an
important and beneficial role in establishing the shape of the solid mass transfer
zone, and it is specially important for cycles with short feed and desorption times
[26].
A better regeneration of the beds is achieved through purging with product gas
during the desorption step and/or the use of vacuum [27]. Although the purge
helps cleaning the bed and a higher purity can be achieved, it can adversely affect
the recovery. An increase in the performance is also observed when the adsorbent
beds are repressurized by feed gas at an initially slow rate followed by a rapid
feed second stage. A combination of product and feed repressurization of the bed
before going into production can also improve the process performance [28,29].
The total cycle time is another important variable in the process. As the cycle
time increases, the performance goes through a maximum. At short cycle times,
both the yield and the specific product go to zero because the nonproductive steps
354

in the cycle have a significant effect. On the other hand, at long cycle times, the
bed becomes saturated, and no separation is possible. For fast cycles the system
behavior depends on the kinetic selectivity, while for slow cycles it is a function of
the equilibrium selectivity [27,30,31].
Other practices that can help achieve a higher productivity are the use of a
dense packing technique for the filling of the beds, as well as the use of layers
with different quality sieve [24].
The separation of nitrogen from air can also be done using zeolite molecular
sieves. Although the kinetic selectivity is similar to that of carbon molecular
sieves, the equilibrium selectivity favors oxygen. Shin and Knaebel [32,33] found
that there is a trade off between purity and yield, which is characteristic of the
diffusion induced separations.
In order to better understand the system behavior, several theoretical models
have been developed describing the process. Different mechanisms are
considered, and assumptions are made in order to predict the performance of the
pressure swing adsorption cycle. These theoretical studies also allow gain an
insight into the effect of the cycle steps and operating variables.
One of the advantages of the theoretical studies is the possibility of studying
the sensitivity of the process performance to both operating variables and
adsorbent properties. For instance, if the gases uptake rates is assumed to
increase, while maintaining the same kinetic selectivity, the productivity
increases. On the other hand, the effect on the yield depends whether a constant
or variable cycle time is assumed [31]. In general it is observed that, while a
change in the oxygen uptake rate will mainly affect the purity, a change in the
nitrogen uptake rate will mainly affect the recovery [14,34].
Good agreement between experimental data for air separation with results
from simulation studies using different gas uptake models has been observed.
Although the linear driving force model can correctly predict the trends of the
experimentally observed behavior, the pore diffusion model shows a much closer
agreement [34].
Several authors have also studied and analyzed the temperature and
concentration profiles in the bed during a PSA cycle. Raghavan and Ruthven [35]
carried out a numerical simulation of the nitrogen pressure swing adsorption
process over a carbon molecular sieve. They run a simple Skarstrom type cycle,
with no equalization step, and analyzed the evolution of the gas and solid phase
composition profiles using a linear isotherm and a linear driving force model. The
mass transfer zones are of an S-shape, but do not show maximum or minimum
inside the bed. More realistic simulation studies, as well as experimental
measurements, have shown the roll-up effect due to displacement of oxygen by
nitrogen, as well as the reduced working capacity between adsorption and
pressurization due to the gradual concentration profiles [10,26]. It was also
confirmed that the oxygen mass transfer zone is very broad under favorable
operating conditions, and spans over a large portion of the bed.
355

In the low purity region, the purification of the product is relatively


independent of cycle time [36]. This could only be so if the shape of the mass
transfer zone (MTZ) is insensitive to the cycle time [31]. Except for very long cycle
times, it would be expected that, due to the high throughputs and the small
amount of 02 adsorbed, the MTZ would span the length of the adsorbent bed.
Recently, Lemcoff et al. [36,37] studied both experimentally and theoretically
the nitrogen pressure swing adsorption process in the high purity region, and
found that the effect of cycle time on the process performance becomes more
significant as the purity increases.

6. THEORETICAL MODEL

Numerical simulations were carried out using the Dynamic Adsorption Process
Simulator (DAPS), developed by LaCava et al. [38]. The following assumptions
were made in the development of the simulator:
- The system is non-isothermal, with the total pressure remaining constant
during production and purge. The pressure is assumed to increase linearly during
pressurization, and follow an exponential decay during blowdown.
- The flow pattern is described by the axial dispersion model. The axial
dispersion is variable point to point along the bed, a function of pressure and
velocity.
- In the equalization step, the detailed composition profile for the
pressurizing gas is taken from the corresponding profile of the depressurizing gas
of the other bed (point by point model).
- The rate of uptake of gases by the non-Fickian carbon molecular sieve is
described in terms of the slit-potential rate model [39], which considers the slit-
cavity structure of the carbon molecular sieve. Under certain conditions, the
Langmuir adsorption rate expression is a good approximation to the slit-potential
rate model.

6.1. Model e q u a t i o n s
The mass balance for component i in the fluid is given by:

c~ci c~2ci c~(VzCi) (1 a)R i (1)


=D L . . . . .
~t ~z 2 ~z

where the rate of adsorption can be expressed in terms of the slit potential rate
model:

ks,imiEKici(1 0kl 0il


Ri = c~-~--- Ki ( 1 - 0 i )
356

At very high coverage, or when the slit potential does not control the rate, the
rate equation reduces to the classical Langmuir form:

= ~t =kdiqmi Eici(l_XOk/_Oil (3)

The heat balance in the adsorber bed is given by:

[pgCpgs +p Cps(1- s)] c~T


s c~t
(4)
c~2T
cg(vzT)+(1 . a)~--~Ri(_AHi)
. . . 2h(T T w)
K L c~z2 - pgCpg Oz r

The boundary conditions for the production step are:

DL-~z0Ci(0, t)=-Vz (0, t f i (0- , t)-ci (0, t)] (5a)

K L __~zC3T
(0, t)= _Vz (0 , t)p g C pg IT(0- , t)-T(0 , t)] (5b)

c9 C i (L, t)= 0 c9T (L, t ) : 0 (5c)


5z 5z

For the pressurization step, the only different boundary condition is the
velocity at the top of the bed:

vz(L,t): 0 (6)

During the vent or blowdown step, the dispersion effect is neglected, and the
boundary conditions are:

0 c i (L, t)= 0 v z (L, t)= 0 (7a)


~.z

c~T (0, t)= 0 c~T (L, t)= 0 (7b)


c~z c~z

The initial conditions are:


357

ci (z,0) = 0 qi (z,0) = 0 T (z,0)= To for a clean bed (Sa)

r162
ci (z,0) = Cio qi (z,0) = qi T (z,0)= To for a s a t u r a t e d bed (Sb)

The numerical integration of the partial differential equations is carried out by


using the numerical method of lines. Finite differences are used in the spatial
discretization procedure, and the Adams multistep method as the time
integrator.
Kinetic and equilibrium data were obtained using a batch column device
described elsewhere [39,40]. At 25oC, the oxygen and nitrogen monolayer capacity
is 1.8 mole/kg CMS, and the adsorption and desorption kinetic constants are,
respectively, 76.84 cm3/(mole s) and 0.0176 1/s for oxygen, and 3.11 cm3/(mole s)
and 0.000585 1/s for nitrogen.
Shirley and Lemcoff [36] estimated the adsorption half-time from results at
different cycle times and product purities. As the oxygen concentration in the
product goes to 0 ppm, the feed per cycle goes to some finite, limiting value. These
values reflect the amount of air needed to bring the adsorber vessels up to the
production pressure, including the amount needed for adsorption as well as that
needed to fill the interstitial and pore volumes in the CMS. At short cycle times,
the majority of the limiting feed is due to the gas needed to fill the interstitial
space in the adsorbent beds, such that very little nitrogen purification occurs,
while at long cycle times, the adsorption of both nitrogen and oxygen is near
equilibrium levels with little nitrogen purification, again. Therefore, when these
values are plotted against the part cycle time, the limiting feed per cycle is a
gross measure of the adsorption kinetics of air. Interpolation of the data gives an
adsorption half-time of about 1600 sec, which matches closely that for air at the
same pressures from batch adsorption experiments.

7. EXPERIMENTAL PART

The experiments were carried in a commercial-scale two-bed nitrogen PSA


unit built by The BOC Group, Inc. A schematic of the unit is shown in Fig. 4.
These units typically run a series of four processing steps in cyclic fashion. In the
first step of a cycle, one bed is pressurized with air while cocurrently producing
nitrogen gas while the second bed is regenerated by venting countercurrently to
the atmosphere or to a vacuum. At the end of this first step, the beds are brought
into fluid communication, causing gases from the bed at high pressure to flow
into the regenerated bed until pressure equalization occurs. During regeneration,
the low-pressure bed is sometimes purged with a small fraction of the purified
product stream. Additional steps, such as product backfill can be included in the
cycle [41]. In this case, before the regenerated bed is pressurized with air, a
fraction of the nitrogen product is used to partially pressurize the bed
countercurrently.
358

HIGH-PRESSURE PRODUCT

T
,,[~~]

[~~ ~!~ VENT

Figure 4. Schematic of nitrogen pressure swing adsorption plant.

Different cycle times (full cycle time between 120 and 1200 s) and product
flowrates (specific product between 5 and 60 Nm3/m3/h) at adsorption pressures
between 6.5 and 9.0 atm (6.6x10 ~ and 9.1x10 ~ Pa) were tested. Regeneration was
done at atmospheric pressure in all cases. The change in the cycle time was
achieved by changing simultaneously the duration of the production and venting
steps. The duration of the remaining steps, pressurization, blowdown, and
equalization, were not modified. The purge flowrate was also varied.
Pressurization of the adsorber beds was controlled according to the optimal
method disclosed by Shirley and LaCava [29].

8. RESULTS

The predictions of the theoretical model have been compared with the
experimental results. Oxygen concentration breakthrough curves were measured
experimentally by sending to an oxygen analyzer a small fraction of the product
gas produced at the exit of the bed. The oxygen analyzer response was about
1 sec, and readings were taken every second. There is a period when the unit is
not producing, namely between the end of the production step in one bed, and the
end of the pressurization step in the other bed. The experimental results are
359

140
A

E
o. 120
o.

~"
0 100

=- 80
c
r 60
0
40
c
ol
>, 20
x
0
0
0 20 40 60 80 100 120

T i m e (s)

Figure 5. Experimental oxygen breakthrough curve. Product purge during desorption step.

shown in Fig. 5. The peak corresponds to the equalization step, and is related to
the concentration of gas being transferred between the beds. As the mass transfer
concentration zone shifts, the concentration increases towards the end of the
production step.
The predicted oxygen breakthrough curves for the simple 4-step cycle, with
and without nitrogen purge, are shown in Fig. 6. A good agreement between the
experiments and the simulation is observed. When no purge with nitrogen
product is done during the vent step, a more linear response is obtained, and the
average oxygen concentration in the half-cycle is higher.
In order to better understand the behavior of the PSA process, oxygen
concentration profiles were obtained from the simulation studies for different
steps in the cycle. We will consider first the case with nitrogen purge during the
vent step. The oxygen mole fraction in the gas phase at different times during the
production step are shown in Fig. 7. It can be seen that the shift in the oxygen
concentration profile is about one third of the bed length. At the same time it can
be seen that, at the beginning of the step, a concave up profile exits, but it
immediately changes to an s shape, and to a concave down profile. This rollover
effect is due to the fact that oxygen is adsorbed faster than nitrogen, so that
initially the adsorbed concentration of oxygen in the carbon molecular sieve is
high. As the step proceeds, nitrogen starts to be adsorbed to a greater degree, and
displaces the oxygen. Therefore, the oxygen concentration in the gas phase
becomes greater than the value in air. Ng et al. [26] studied the shape and
360

movement of the oxygen mass transfer zone at four axial positions in a laboratory
N2 PSA unit, and concluded t h a t the concave down profile is achieved only at the
end of the production step. The different conclusion can be a t t r i b u t e d to the
relative large distance between their m e a s u r e m e n t points, which was about one
quarter of the bed length.

E
100,
90-
~- 80-
O
= 70
60
so
=~ 4o
o 30-
zo-
=~ lo-
o I I t I I I

0 20 40 60 80 100 120
Time (s)

Fig. 6. Predicted oxygen breakthrough curves.


Purge No purge.

0.2500

0.1875
t_

m .
......... ~ "'\ "\
r 0.1250 ......... "~ ",\ \,
e-
'""...... \~,, "'\ \,\
03

~' 0.0625
O

0 0.25 0.50 0.75 ' 1.0t]


Distance/Bed Length

Figure 7. Evolution of gas phase oxygen concentration profiles during the production step.
Nitrogen purge during vent step. Oxygen concentration in product" 0.05%.
361

A
O~
O
"6 0.00030
E
C
O
.=.
"~ 0.00022 \ ".......,
........... "\\\ "'X,, ,\
C
O \ - ,, .\ \ \
O
c-
OO 0.0001s \ -~ e \ \ \
O \ ....... ',,, ' , " , \
U)
'". \ \ ,
.C '...... \\ ",\ \,
_ .. \\ ,, \
0.00007 '..... \\ \, "\
.,,=
"-...... ,,, "\ "\
O "...... ",,,, ",\ "\,
e~

r 0
>,
X
0 0.25 0.50 0.75 ' 1.0lJ
o Distance/Bed Length

Figure 8. Evolution of solid phase oxygen concentration profiles during the production step.
Nitrogen purge during vent step. Oxygen concentration in product 0.05%.

The corresponding oxygen solid phase concentration profiles are shown in


Fig. 8. We can see t h a t the profiles change from concave up to an s-shape. As
discussed above, the adsorbed oxygen concentration near the feed end increases
at the beginning of the step due to the fast oxygen adsorption, but then decreases
due to the displacement by nitrogen. During the production step the solid phase
concentration front also shifts by about a third of the bed length.
The predicted bed t e m p e r a t u r e profiles at different times during the
production step are shown in Fig. 9. A m a x i m u m t e m p e r a t u r e difference of about
7~ is observed near the bed center. It can be seen t h a t during the cycle the
profiles seem to shift vertically throughout the bed.
The oxygen concentration profiles in the gas and solid phases at different times
during the vent step are represented in Figs. 10-11. They span over the whole bed
length and are almost proportional to each other. As the step proceeds, both
adsorbed gases are being desorbed. In addition, the nitrogen product purge helps
the oxygen concentration profiles become flatter. The corresponding t e m p e r a t u r e
profiles show an almost uniform decrease in t e m p e r a t u r e (Fig. 12).
The concentration and t e m p e r a t u r e profiles for the case with no nitrogen
product purge during the vent step are very similar to the case just described.
The major differences are at the bed outlet, during the production step, and at
the bed feed end during the vent step. Bed t e m p e r a t u r e s are slightly higher for
the no purge case, while the oxygen concentration profiles during the vent step
are slightly higher near the bed feed end.
362

308.00
A

==
305.25
L

302.50
F- .r ...... ..i ' I ""'--_ //\
"0
299.75

297.000
o.2g o.s6 o.Tg 1.o6
Distance/Bed Length

Figure 9. Evolution of bed temperature profiles during the production step. Nitrogen purge
during vent step. Oxygen concentration in product 0.05%.

0.500 r-

0.375

0.250

(~ 0.125

0 01 . . . 0.25
. . 0.50; ....... 7---u--~0.75. . . . . - - - ' 1.0()
Distance/Bed Length

Figure 10. Evolution of gas phase oxygen concentration profiles during the purge step.
Nitrogen purge during vent step. Oxygen concentration in product: 0.05%.
363

A
O1
o
"6 0.00014
E
c-
O
0.00010
e..
(D
o
c
o 0.00007
o
u~

,,c

"o 0.00003

(D
O1
,,
x
o o is o go o.is ~do
O Distance/Bed Length
Figure 11. Evolution of solid phase oxygen concentration profiles during the purge step.
Nitrogen purge during vent step. Oxygen concentration in product" 0.05%.

306.0
A

L
302.5

~- 299.0
j / j .

~m 295.5 //"
. / j . . ~
"~

, , ! , ,
292.0
o o2s oso 0.75 1.0'0
Distance/Bed Length
Figure 12. Evolution of bed temperature profiles during the purge step. Nitrogen purge during
vent step. Oxygen concentration in product: 0.05%.

The effect of the backfill step on the bed concentration profiles and the process
performance is shown in Figs. 13-14. The oxygen concentration profiles in the bed
at the end of the production, equalization and purge steps, for a cycle without
backfill, are represented in Fig. 13. The rollover effect is observed at the end of
364

0.300 -

g \ \

0.225
"'\,."~'\,\. N\~\\
0.150
~'\'\, ~\\\
==
\~. K \ \\\
0.075 \ \ -,\

0
0 o2's os'o ors loo
Distance/Bed Length

Figure 13. Oxygen concentration profiles at the end of the steps. Cycle with no backfill.
Product oxygen concentration: 270 ppm.
- - Production Equalization (down)
.... Purge Equalization (up).

0.300

t-
o
0.225 9 \\
I,,,,
NI,-

0 0.150
E
c
~,.\ ~'\'\, ~\\\
D'J
x 0.075
O . \, \ \ "-,.

0 ' ' i ~- - -,~.. = _. ~ ~ r-~- . . . . . . . . . . _ "" , -" ~ . . ,. .

0 o2s o s'o ~ o is 1.o'o


Distance/Bed Length

Figure 14. Oxygen concentration profiles at the end of the steps. Cycle with no backfill.
Product oxygen concentration: 100 ppm.
Production Equalization (down) . . . . Purge
Equalization (up) Backfill.
365

the production step. During the equalization step, gas is transferred from the
high pressure to the low pressure bed through both inlet and outlet. As the
decreasing equalization proceeds, the oxygen concentration profile is spread out
further and, due to the oxygen desorption, a higher concentration t h a n that in air
is observed at the inlet. During the vent step, helped by the countercurrent
purge, the oxygen concentration decreases significantly. As the increasing
equalization step proceeds, gas enters the bed from both ends, and the mass
transfer zone is reduced. At the end of the increasing equalization step, the
oxygen mole fraction at the inlet is again higher t h a n its value in air. During the
following pressurization step, the oxygen concentration profile shifts to the left,
and turns into concave up.
As the backfill is introduced, the oxygen concentration profile shifts towards
the feed end, specially in the high concentration region. Now the oxygen mole
fraction at the feed end is substantially lower t h a n its value in air. During the
following pressurization step, the oxygen concentration profile shifts to the right,
and turns into concave up as in the case above. The oxygen concentration at the
feed end has its value in air. As the production step proceeds, the profile shifts to
the right and becomes steeper, and higher nitrogen purities can be achieved
(Fig. 14).
However, there is a limit to the process improvement. The effect of the
duration of the backfill step on the purity of the nitrogen product is shown in
Fig. 15. As the backfill increases, since more product gas is used, the yield tends
to decrease. At the same time, the feed flowrate has to increase to achieve the

~" 300
Q.
Q.
250
r
O
,..

"" 200
L

r
9 150
or
o~

O
o 100 .~ .o

c
(b
O1 50
x
o 0 I I I I I I
0 0.5 1 1.5 2 2.5 3

Backfill step (sec)

Figure 15. Effect of backfill step on nitrogen purity at constant product flow.
Experimental . . . . . Predicted.
366

same product flow, and an increase in the oxygen concentration in the product is
observed for long backfill steps. A good a g r e e m e n t is observed between the
experimental results and the model predictions.
Other way of achieving higher nitrogen purities is the use of cryogenic back-up
t a n k vent gas for purging the regenerating bed [42]. The reported results to
produce a gas containing 0.1% oxygen are s u m m a r i z e d in Table 1. Case I
represents the base case, when the purge step is carried out with the same
product gas. Cases II and III correspond to the situation when a high purity gas
is used to purge the beds during the regeneration step. A significant
improvement in the performance, which increases with the a m o u n t of purge gas,
can be observed.

Table 1
Comparison of purge methods
Case I Case II Case III
Product i m p u r i t y (% 02) 0.1 0.1 0.1
Purge concentration (% 02) 0.1 < 0.001 < 0.001
Purge to feed ratio (%) 11.2 10.6 13.0
Productivity increase (%) base case 15.1 20.5
Yield increase (%) base case 8.1 11.8

Similar behavior is observed at higher nitrogen purities, as shown in the


simulation studies carried out for a system using an external source for the
nitrogen purge. The oxygen concentration profiles at the end of the different cycle
steps, at constant product flowrate, are shown in Figs. 16-18. From the analysis
of the figures it follows t h a t the main differences in the profiles are observed for
the purge step. It can be seen that, as the purge to feed ratio increases from 0.05
to 0.10, the oxygen concentration profile at the end of the purge step is flatter,
and the oxygen concentration in the product decreases from 500 to 200 ppm (Figs.
16-17). At the higher purge flowrate, a better cleaning of the bed is achieved, and
a higher purity product can be obtained. On the other hand, when the purity of
the external purge gas is increased, so t h a t the oxygen concentration in the purge
gas decreases from 10 ppm to 1 ppm oxygen, no significant change in the product
oxygen concentration is observed (Figs. 16 and 18). The profiles also remain
unchanged. It can be concluded t h a t the purity of the gas used as external purge
has no significant effect of the final product purity, when its purity is more than
one order of m a g n i t u d e higher t h a n the purity of the product gas.
367

0.300
r
O
.m
\ \\
"~ 0.225
\,\ "~ \\
z._
N=,.

m \'\,\ ~\\\
o 0.150
"\,\ ~\\\
o1 "\,\, ~\\\\
o
~' 0.075 \, ~ \\
_",.. . . . . .-_. ~ . . . . . . . . .

' 0.2'5 ' ' 0.5'0 ' ' 0.7'5 ' ' 1.0'0
Distance/Bed Length

Figure 16. Oxygen concentration profiles at the end of the steps. Cycle time: 240s. Product
oxygen concentration: 0.05%. Purge to feed ratio: 0.05. Purge purity: 10 ppm 02.
Production Equalization down
Purge . . . . . . . . . Equalization up.

r 0.300 - ___
O
,m

(3
0.225
N-,

G)
\'\, "~ \\
O \\\ N~\\\
E 0.150
C "\,\ ~\\\
O3
0.075
'\,\. X~\\ \\

- - -" . . . . . . . . . . . '-"="--- - , 4 ,.-..=_. { , ,~ ~ .--~.

0.25 0.50 0.75 1.00


Distance/Bed Length

Figure 17. Oxygen concentration profiles at the end of the steps. Cycle time: 240 s. Product
oxygen concentration: 0.02%. Purge to feed ratio: 0.10. Purge purity: 10 ppm O2.
Production Equalization down
..................................Purge Equalization up.
368

0.300

e,,
o
o,,.,

ot~ 0.225
\\. \\\
N-
- \,\, "~\\\
O
E o.15o
"\'\'\, NNN\
c
\'\ N'\\
r
X 0.075
"\,
\'\,
NN\\\' \ \ \
o

' o~s ' o go ' ols ~do


Distance/Bed Length

Figure 18. Oxygen concentration profiles at the end of the steps. Cycle time: 240s. Product
oxygen concentration: 0.02%. Purge to feed ratio: 0.05. Purge purity: 1 ppm 02.
Production Equalization down
................................. Purge Equalization up.

9. CONCLUSIONS

The pressure swing adsorption process for the separation of nitrogen from air
over carbon molecular sieves was described, and both experimental and
simulation studies were carried out. Two types of adsorbents, with different
behavior in the gases uptake are commonly used, and the process characteristics
for a non-Fickian type sieve were specially analyzed.
The effects of the different variables and cycle steps on the production of
nitrogen were discussed and, in particular, the influence of the purge and backfill
steps on the oxygen concentration profiles was analyzed. The shape and
movement of these profiles help understand the relationship between the process
variables and the pressure swing adsorption performance. A good agreement
between the predictions of the theoretical model and experimental results was
also observed.

NOMENCLATURE
C gas phase concentration, mole/m 3
Cp specific heat, kJ/kg K
DL axial dispersion coefficient, m2/s
AH enthalpy change of adsorption, kJ/mole
369

h heat transfer coefficient, kJ/m 2 s K


K Langmuir equilibrium constant, m3/mole
KL effective thermal conductivity, kJ/m s K
kd Langmuir desorption constant, 1/s
ksl slit potential rate constant, 1/s
L bed length, m
q adsorbed phase concentration, mole/m 3
q lll monolayer capacity, mole/m 3
R rate of adsorption, mole/(m 3 s)
r bed radius, m
T temperature, K
t time, s
Vz axial gas velocity, m/s
Z axial distance, m

Greek letters
bed voidage
9 density, kg/m 3
0 q/qm

Subscripts
g gas
i,k components i,k
o initial
s solid
w wall

REFERENCES

1. R.T. Yang, Gas Separation by Adsorption Processes, Butterworths, Boston,


1987.
2. D.M. Ruthven, S. Farooq and K.S. Knaebel, Pressure Swing Adsorption, VCH,
New York, 1994.
3. J.E. Metcalfe III, M. Kawahata and P.L. Walker, Fuel, 42 (1963) 233.
4. J.D. Moyer, T.R. Gaffney, J.N. Armor and C.G. Coe, Microporous Materials,
2 (1994) 229.
5. I.P. O'koye, M. Benham and K.M. Thomas, Langmuir, 13 (1997) 4054.
6. P.L. Walker, L.G. Austin and S.P. Nandi, in: Chemistry and Physics of
Carbon, P.L. Walker (ed.), Marcel Dekker, NY, vol. 2,257, 1966.
7. R.V. Jasra, N.V. Choudary and S.G.T. Bhat, Sep. Sci. Tech., 26 (1991) 885.
8. H. Jiintgen, K. Knoblauch and K. Harder, Fuel, 60 (1981) 817.
9. P.K. Sharma and P.K. Seshan, Gas Sep. Purif., 4 (1990) 203.
370

10. R.M. Thorogood, Gas Sep. Purif., 5 (1991) 83.


11. R.M. Rynders, M.B. Rao and S. Sircar, AIChE Journal, 43 (1997) 2456.
12. D.M. Ruthven, Chem. Eng. Sci., 47 (1992) 4305.
13. D.M. Ruthven, N.S. Raghavan and M.M. Hassan, Chem. Eng. Sci.,
41 (1986) 1325.
14. S. Farooq and D.M. Ruthven, Chem. Eng. Sci., 46 (1991) 2213.
15. Y.D. Chen, R.T. Yang, P. Uawithya, AIChE Journal, 40 (1994) 577.
16. F.R. Fitch, M. Bulow and A.I. LaCava, Gas Sep. Purif., 8 (1994) 45.
17. M.B. Rao, R.G. Jenkins and W.A. Steele, Langmuir, 1 (1985) 137.
18. R. Srinivasan, S.R. Auvil and J.M. Shorck, Chem. Eng. J., 57 (1995) 137.
19. Y.Y. Trifonov and T.C. Golden, J. Porous Materials, 3 (1996) 5.
20. N.A. Seaton, S.P. Friedman, J.M.D. MacElroy and B.J. Murphy, Langmuir,
13 (1997) 1199.
21. S.K. Verma and P.L. Walker, Jr., Carbon, 30 (1992) 837.
22. T.C. Golden, P.J. Battavio, Y.C. Chen, T.S. Farris and J.N. Armor, Gas Sep.
Purif., 7 (1993) 274.
23. G.Q. Li and R. Govind, Ind. Eng. Chem. Research, 33 (1994) 755.
24. A. Schulte-Schulze Berndt, MUST '96, 185, 1996.
25. N.O. Lemcoff, S.J. Doong and A.I. LaCava, in: Fundamentals of Adsorption,
M. Suzuki (ed.), Kodansha, Tokyo, 357, 1993.
26. M. Ng, J.M. Schork and K.R. Fabregas, Gas Sep. Purif., 7 (1993) 159.
27. N.O. Lemcoff and A.I. LaCava, Gas Sep. Purif., 6 (1992) 9.
28. M.M. Hassan, N.S. Raghavan and D.M. Ruthven, Chem. Eng. Sci.,
42 (1987) 2037.
29. A.I. Shirley and A.I. LaCava, Ind. Eng. Chem. Research, 32 (1993) 906.
30. M.M. Hassan, N.S. Raghavan and D.M. Ruthven, Chem. Eng. Sci.,
41 (I 986) 1333.
31. J.M. Schork, R. Srinivasan and S.R. Auvil, Ind. Eng. Chem. Research,
32 (1993) 2226.
32. H.S. Shin and K.S. Knaebel, AIChE Journal, 33 (1987) 654.
33. H.S. Shin and K.S. Knaebel, AIChE Journal, 34 (1988) 1409.
34. S. Farooq and D.M. Ruthven, Chem. Eng. Sci., 45 (1990) 107.
35. S. Raghavan and D.M. Ruthven, AIChE Journal, 31 (1985) 2017.
36. A.I. Shirley and N.O. Lemcoff, AIChE Journal, 43 (1997) 419.
37. A.I. LaCava and N.O. Lemcoff, Gas Sep. Purif., I0 (1996) 113.
38. A.I. LaCava, J.A. Dominguez and J. Cardenas, in: Adsorption: Science and
Technology, A.E. Rodrigues, M.D. LeVan and D. Tondeur (eds.), NATO ASI
Series, vol. 158, 323, 1989.
39. A.I. LaCava, V.A. Koss and D.A. Wickens, Gas Sep. Purif., 3 (1989) 180.
40. V.A. Koss, D.A. Wickens, P. Cucka and A.I. LaCava, Proc. Carbon 86, 388,
Baden-Baden, Germany, July 1986.
41. N.O. Lemcoff, Pressure swing adsorption process, US Patent 5 520 720
(1996).
42. R. Jain, PSA employing high purity purging, US Patent 5 090 973 (1992).

You might also like