See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/276026690

Modelling and Optimization of Multibed Pressure Swing Adsorption

Flowsheets

Conference Paper · August 2006

CITATIONS READS

2 327

1 author:

Dragan Nikolic
DAE Tools Project
29 PUBLICATIONS   217 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

DAE Tools View project

All content following this page was uploaded by Dragan Nikolic on 09 May 2015.

The user has requested enhancement of the downloaded file.

 MODELLING AND OPTIMIZATION OF MULTIBED
PRESSURE SWING ADSORPTION FLOWSHEETS

Dragan Nikolic1, Michael C. Georgiadis2, Eustathios S. Kikkinides1,*

1
University of Western Macedonia, Department of Engineering and Management of Energy
Resources, Kozani 50100, Greece

2
Imperial College London, Centre for Process Systems Engineering, Department of Chemical
Engineering, London SW7 2AZ, UK

This work presents a generic modelling framework for the separation of gas mixtures using
multi-bed Pressure Swing Adsorption (PSA) systems. The most important features of the
model include various mass, heat and momentum transport mechanisms, gas valve models,
complex boundary conditions and realistic operating steps. All models have been imple-
mented in PSE’s gPROMS environment.

1. MATHEMATICAL MODELLING

The modelling of large-scale PSA processes has to take into account the simultaneous
mass, heat and momentum transport phenomena both in the bulk gas and within the adsorbent
particles. The developed modelling framework consists of:
¾ detailed adsorbent bed model,
¾ gas valve model,
¾ general multi-bed pressure swing adsorption plant model,
¾ c++ helper program which for a given number of the beds and sequence of operat-
ing steps in one bed, generates operating procedure for the whole plant.

1.1. Governing equations for adsorbent bed model

Typical model of the adsorbent bed consists of coupled set of mixed algebraic and partial
differential equations for mass, heat and momentum balance, as well as equilibrium isotherm
equations. Here, the objective is to develop a generic adsorption column model which pro-
vides the flexibility to make a selection among several alternative operating modes. Following
options are available:
¾ Different mass balance equations (local equilibrium theory, linear driving force
theory, surface diffusion theory, pore diffusion theory),
¾ Different heat balance equations (isothermal operating mode, nonisothermal
operating mode, adiabatic operating mode),

*
To whom correspondence should be addressed: e-mail: [email protected]
 ¾ Different momentum balance equations (Darcy equation, Ergun equation),
¾ Multicomponent gas-adsorbate phase equilibrium (both T and P dependant):
• Henry’s low isotherm,
• Extended Langmuire type isotherm,
¾ Equations of state for multicomponent gas mixtures:
• Ideal gas equation of state,
• Real gas equation of state (using compressibility factor),
¾ Equations for prediction of:
• bulk gas overall mass and heat transfer coefficients,
• mass and heat axial dispersion coefficients,
• effective diffusivity coefficient,
• molecular, knudsen and surface diffusivities.

The general balance equation takes the form:

DΨ .
= Ψ g + ∇D∇Ψ (1)
Dt
DΨ
where Ψ is the concentration of mass, heat or momentum, D is the diffusivity, is the
Dt
.
substantial derivative, Ψ g is the generation term and ∇D∇Ψ is the diffusion term.
The equilibrium isotherm is given with following general equation:
Q* = f ( T p , C p ) (2)

where C p is the molar concentration of gas phase within particles and T p is the particle tem-
perature.
The substantial derivative is given by:
DΨ ∂ ( uΨ ) ∂ ( vΨ ) ∂ ( wΨ ) ∂Ψ
= + + +
Dt ∂x ∂y ∂z ∂t
and the Laplacian ∇ 2 Ψ depends on the coordinate system used:
¾ In rectangular coordinates:
∂2Ψ ∂2Ψ ∂ 2Ψ
∇2Ψ = + 2 + 2
∂x 2 ∂y ∂z
¾ In spherical coordinates:
1 ∂ ⎛ 1 ∂Ψ ⎞ 1 ∂ ⎛ ∂Ψ ⎞ 1 ∂ 2Ψ
∇2Ψ = ⎜ ⎟ + ⎜ sin θ ⎟ +
r 2 ∂r ⎝ r 2 ∂r ⎠ r sin θ ∂θ ⎝ ∂θ ⎠ r 2 sin 2 θ ∂φ2
The following assumptions have been made:
¾ For the bulk gas flow, convection and diffusion in x and y directions are neglected
(plug flow),
¾ For particles, transport in θ and φ directions is neglected,

Page 2 of 16
Applying the general balance equations to the mass and heat transfer for bulk gas flow we
obtain:
¾ Mass transfer
∂ ( uCi ) ∂Ci . ∂ 2C
εbed + εbed + (1 − εbed ) C ig = εbed Dz ,i 2 i , ∀z ∈ ( 0, L ) , i = 1, N comp
∂z ∂t ∂z
∂ ( uCi ) ∂Ci 1 − εbed . ∂ 2Ci
+ + C ig = Dz ,i 2 , ∀z ∈ ( 0, L ) , i = 1, N comp (3)
∂z ∂t εbed ∂z
¾ Heat transfer
∂ ( uT ) ∂T . ⎛ ∂ 2T ⎞
εbed ρc p + εbed ρc p = (1 − εbed ) T g + εbed λ ⎜ 2 ⎟ , ∀z ∈ ( 0, L ) , i = 1, N comp
∂z ∂t ⎝ ∂z ⎠
∂ ( uT ) ∂T 1 − εbed . ⎛ ∂ 2T ⎞
ρc p + ρc p = T g + λ ⎜ 2 ⎟ , ∀z ∈ ( 0, L ) , i = 1, N comp (4)
∂z ∂t εbed ⎝ ∂z ⎠
Applying the general balance equations to the particle mass and heat transfer:
¾ Mass transfer
∂Cip .
⎛ ∂ 2Cip 1 ∂Cip ⎞ ⎛ ∂ 2Qip 1 ∂Qip ⎞
εp = (1 − ε p ) Cigp + ε p De ⎜ + 2 + ε
⎟ p s⎜D + 2 ⎟
∂t ⎝ ∂r 2 r ∂r ⎠ ⎝ ∂r 2 r ∂r ⎠
∂Cip 1 − ε p . p ⎛ ∂ 2Cip 1 ∂Cip ⎞ ⎛ ∂ 2Qip 1 ∂Qip ⎞
= Cig + De ⎜ + 2 ⎟ + Ds⎜ + 2 ⎟ (5)
∂t εp ⎝ ∂r 2 r ∂r ⎠ ⎝ ∂r 2 r ∂r ⎠

∀z ∈ [ 0, L ] , ∀r ∈ ( 0, R p ) , i = 1, N comp

¾ Heat transfer
∂T p . ⎛ ∂ 2T p 1 ∂T p ⎞
ρ p c pp = T g + λp ⎜ + 2 ⎟ (6)
∂t ⎝ ∂r 2 r ∂r ⎠
∀z ∈ [ 0, L ] , ∀r ∈ ( 0, R p ) , i = 1, N comp

The boundary conditions depend on the operating step. The well known Danckwert’s bound-
ary conditions have been applied.
Mass transfer in the bulk fluid:
¾ for the stream inlet into the bed
∂Ci
u ( Ci − Ciin ) = Dz ,i , z = 0 or z = L, i = 1, N comp (7)
∂z
¾ for the stream outlet from the bed
∂Ci
= 0, z = 0 or z = L, i = 1, N comp (8)
∂z
Mass transfer in particles:
∂Cip
(
k f Ci − Ci p
r = Rp ) = De
∂r
r = Rp , r = R p , i = 1, N comp (9)

Page 3 of 16
 ∂Cip
= 0, r = 0, i = 1, N comp (10)
∂r
Heat transfer in the bulk fluid:
¾ for stream inlet into the bed
T = T in , z = 0 or z = L (11)
¾ for stream outlet from the bed
∂T
= 0, z = 0 or z = L (12)
∂z
Heat transfer in particles:
⎛ ∂T p ⎞
(
kh T − T p
r = Rp ) =λ ⎜
p

⎝ ∂r ⎠
⎟ r = Rp , r = Rp (13)

∂T p
= 0, r = 0 (14)
∂r
Interstitial velocity:
¾ At z=0 end
∂u
= 0 z =0 (15)
∂z
¾ At z=L end
∂u
= 0 z=L (16)
∂z
Interstitial velocity is controlled by the pressure difference on the gas valves at the bed ends.
Pressure drop along the bed could be described by the following two equations:
¾ Darcy’s equation

180μbc u (1 − εbed )
2
∂P
=− , ∀z ∈ [ 0, L ] (17)
∂z ( 2 Rp ) εbed
2 3

¾ Ergun’s equation
∂P (1 − εbed ) 2 μbc u u 1− ε ρ u
− = 150 + 1.75 3 bed bc (18)
∂z εbed
3
dp 2
εbed d p

The bulk gas mass and heat transfer coefficient could be assumed to be constant, or predicted
using equations for the Sherwood and Nusselt numbers (Wakao et al, 1978, 1979):
¾ Mass transfer coefficient

kf =
Dm
L
(
2.0 + 1.1Sc 0.33 Re0.6 ) (19)

¾ Heat transfer coefficient

Page 4 of 16
 λ
kh =
L
(
2.0 + 0.6 Pr 0.33 Re0.5 ) (20)

The bulk gas mass and heat axial dispersion coefficient could be assumed to be constant, or
predicted using following equations (Wakao et al, 1978, 1979):
¾ Mass axial dispersion coefficient
Dm
Dz = ( 20 + 0.5ScRe ) (21)
εbed
¾ Heat axial dispersion coefficient
λ z = λ bc ( 7 + 0.5 PrRe ) (22)

The effective diffusivity coefficient within the particles can be assumed constant or predicted
using following equation:
ε p Dm Dk
Dep = (23)
τ p Dm + Dk
Molecular diffusivity can be taken as constant or (for the binary mixture) predicted using one
of the following two correlations:
¾ Chapman-Enskog equation
1
T3
Dm = 1.8583 × 10−3 MW (24)
Pσ 2AB Ω AB

where σ AB and ε AB are constants in the Lennard-Jones potential-energy function for

the pair AB, Ω AB is collision integral, a function of k BT ε AB and k B is
Boltzmann’s constant.
¾ Fuller correlation
1
Dm = 1.013 × 10−7 T 1.75 MW (25)
2
PX AB

where X AB consists of contributions of each molecule.

Knudsen diffusivity may be assumed constant or predicted using Kauzmann (1966) correla-
tion:
T
Dk = 9.7 ×103 R pore (26)
MW
The boundary conditions for twelve different operating steps complement the model taking
into account:
¾ closed system,
¾ co-current pressurization,
¾ adsorption,
¾ pressure equalization (depressurization to the other bed),

Page 5 of 16
 ¾ pressure equalization (repressurization from the other bed),
¾ pressure equalization at both bed ends (depressurization to the other bed),
¾ pressure equalization at both bed ends (repressurization from the other bed),
¾ co-current depressurization,
¾ counter-current blowdown,
¾ purge,
¾ counter-current pressurization,
¾ purge with the strong adsorbate.

1.2. Gas valve model

The model describes one-way valve, that is the gas flow is possible only in one way, from
the in towards out port. The reverse flow is impossible to happen under any circumstances.
The purpose of one way valve is to force the flow only in desired directions and to avoid any
unwanted flows.
The gas valve equations used in the mathematical model are:
¾ If: Pout > 0.53Pcrit
2
⎛P ⎞
1 − ⎜ out ⎟
F = Cv Stem _ position Pin ⎝ Pin ⎠ (27a)
∑ xi MWi T

¾ Otherwise:

1 − ( Pcrit )
2

F = Cv Stem _ position Pin (27b)

∑ x MW
i i T

1.3. Pressure swing adsorption plant model

This is the main model for pressure swing adsorption process. It consists of arrays of the
interconnected adsorbent beds and gas valves, and it is general enough to support all possible
operating steps in every known PSA system. Several tasks have been developed in gPROMS
for easier PSA operation. The sample multi-bed PSA system containing five adsorbent beds is
given on Figure 1.

1.4. Helper application

The helper program which generates inter-bed connections for a given number of the beds and
sequence of operating steps in one bed has been developed. The helper generates source code
of a gPROMS task (the operating procedure of the PSA process) according to the corre-
sponding operating steps. The generated operating procedure opens and closes valves in the
proper order using the given time schedule. Over 40 different configurations have been gener-
ated (using from two up to twelve adsorbent beds).

Page 6 of 16
Figure 1 – Sample four-bed PSA flow-sheet

Page 7 of 16
2. MODEL VALIDATION

2.1. Linear driving force model

The work of Nilchan and Pantelides (1998) has been used to validate a single bed linear
driving force model (LDF). In order to validate LDF model the first part of their work has
been chosen (rapid pressure swing process - rPSA).
The model used for comparison is a slight modification of the developed model presented
in section 1. More specifically, all gas valve equations were removed and boundary condi-
tions were changed to be identical as the ones presented in the work of Nilchan and Pantelides
(1998). All the other equations and boundary conditions are the same except the LDF coeffi-
cient. This equation differs from the corresponding one employed by Nilchan and Pantelides
and has the following form:
∂Qi 15
∂z
( )
= 2 Dep ,i Qi* − Qi , ∀z ∈ ( 0, L ) , i = 1, N comp
Rp
Where:
ε p Dm ,i Dk ,i
Dep ,i = , ∀z ∈ ( 0, L ) , i = 1, N comp
τ p Dm ,i + Dk ,i
Where Dm and Dk have been calculated from Chapman-Enskog and Kauzmann correlations
respectively.
Axial domain is discretized using orthogonal collocation on finite elements of 3rd order
with 50 elements in all simulations (both Nilchan and this work). The column is initially at
atmospheric pressure (1 bar) and saturated with air (yN2=79%, yO2=21%).
The underlying PSA process is to perform separation of oxygen from air using zeolite 5A.
Here, nitrogen is preferably the adsorbed component. Nilchan and Pantelides have investi-
gated the effect of 4 parameters on the purity and recovery of O2: feed pressure, column
length, particle radius, and cycle time.
The rPSA configuration is simple and consists of one bed and two operational steps: pressuri-
zation with the feed and depressurization. Input parameters and Comparison of the simulation
results with the results of Nilchan and Pantelides are given in Table 1.
The following table summaries the absolute deviation of Nilchan and Pantelides results
from the ones obtained in this work. The results cleanly indicate that due to the employment
of a different LDF equation differences in prediction of the two models are significant in term
of product purity.

This work
Overall average
vs.
absolute deviation
Nilchan & Pantelides
In purity 7.43%
In recovery –

Page 8 of 16
 Table 1 – Comparison of the simulation results with the results of
Nilchan and Pantelides

Nilchan and
This work Absolute deviation
Pantelides (1998)
No. Pfeed, Pa O2 Purity, % O2 recovery, % Power, W O2 Purity, % O2 Purity, %
1 1.60E+05 66.13 6.23 25.45 62.00 4.13
2 1.80E+05 80.71 5.46 45.09 76.00 4.71
3 2.00E+05 89.19 4.65 70.16 85.00 4.19
4 2.12E+05 92.21 4.20 87.73 89.35 2.86
5 2.20E+05 93.69 3.93 100.48 91.00 2.69
6 2.40E+05 96.09 3.35 135.98 94.00 2.09

No. Rp, m O2 Purity, % O2 recovery, % Power, W O2 Purity, % O2 purity, %

7 1.00E-04 71.58 5.04 56.73 67.00 4.58
8 1.50E-04 92.21 4.20 87.73 89.35 2.86
9 2.00E-04 96.85 3.21 120.51 93.00 3.85
10 2.50E-04 97.36 2.54 153.29 87.00 10.36
11 3.00E-04 96.43 2.14 180.51 73.00 23.43
12 3.50E-04 94.21 1.88 199.89 56.00 38.21

No. Length, m O2 Purity, % O2 recovery, % Power, W O2 Purity, % O2 purity, %

13 0.6 81.20 3.55 91.39 83.00 -1.8
14 0.7 86.32 3.80 90.77 87.00 -0.68
15 0.8 89.44 3.99 89.57 88.80 0.64
16 0.9 91.24 4.12 88.48 89.50 1.74
17 1.0 92.21 4.20 87.73 89.35 2.86
18 1.1 92.72 4.25 87.31 88.30 4.42
19 1.2 92.91 4.26 87.14 87.40 5.51
20 1.3 92.94 4.27 87.11 85.80 7.14
21 1.4 92.89 4.26 87.15 84.00 8.89
22 1.5 92.79 4.25 87.20 82.10 10.69
23 1.6 92.67 4.24 87.25 80.00 12.67

No. τcycle, s O2 Purity, % O2 recovery, % Power, W O2 Purity, % O2 purity, %

24 1 94.38 2.50 150.97 75.80 18.58
24 2 93.71 3.51 106.68 86.30 7.41
24 3 92.21 4.20 87.73 89.35 2.86
24 4 89.96 4.68 76.91 87.30 2.66
24 5 87.39 5.02 69.58 82.80 4.59

2.2. Pore diffusion model

The single bed pore diffusion model has been validated against the results of Shin and
Knaebel (1987, 1988). The first part of their work presents a pore diffusion model for a single
column. The following assumptions have been employed:

Page 9 of 16
 ¾ Isothermal bed operation,
¾ Negligible pressure drop through the bed,
¾ Linear adsorption equilibrium isotherm (Henry’s law),
¾ Axially dispersed plug flow.
Further, they have developed a dimensionless representation of the overall and component
mass balance for the bulk gas and the particles, and they have applied Danckwert’s boundary
conditions. Boundary conditions in the present work are the same as the ones presented by
Shin and Knaebel.
The effective diffusivity, axial dispersion coefficient and mass transfer coefficient have
been assumed constant and taken by Shin and Knaebel. Phase equilibrium is given as a linear
function of gas phase concentration (Henry’s law).
The axial domain is discretized using orthogonal collocation on finite elements of 3rd order
with 20 elements. The radial domain is discretized using the same method of 3rd order with 5
elements. Shin and Knaebel were used the same method with 6 collocation points for discreti-
zation of axial and 8 collocation points for radial domain. The bed is initially assumed clean
(99.999% N2).
The target is to separate nitrogen from air using molecular sieve RS-10. Here oxygen is the
preferably adsorbed component due to higher diffusivity.
The effects of various operational variables on purity and recovery of N2 was investigated:
¾ The duration of adsorbent step for a constant feed velocity,
¾ Feed velocity (for a fixed duration of adsorption step),
¾ The duration of adsorbent step for a fixed amount of feed,
¾ The duration of blowdown step,
¾ Purge gas velocity for a fixed duration of purge step,
¾ The duration of purge step for a constant purge gas velocity,
¾ The duration of purge step for a fixed amount of purge gas,
¾ The duration of pressurization step,
¾ Feed / purge step pressure ratio,
¾ Column geometry (length/diameter ratio).
The PSA configuration consists of one bed and four operational steps: pressurization with the
feed, high pressure adsorption, counter-current blowdown, counter-current purge. Comparison
of experimental and simulation results is given in Table 2. It is seen that this generic model
predicts satisfactorily the behaviour of the process and the overall average absolute deviation
is limited to approximately 2% for both the purity and the recovery.

Shin and This

Overall average absolute deviation
Knaebel work
In purity 1.81% 1.97%
In recovery 2.52% 2.23%

Page 10 of 16
 Table 2 – Experimental and simulation results

Absolute deviation from exp. data, %

Recovery, % Purity, % Power, W In recovery In purity
Exp. Shin & This Exp. Shin & This Shin & This Shin & This
No. This work
data Knaebel work data Knaebel work Knaebel work Knaebel work
1 6.44 6.38 6.96 99.75 100.00 99.31 6.18 -0.06 0.52 0.25 -0.44
2 8.16 8.64 8.50 99.55 99.89 98.92 5.63 0.48 0.34 0.34 -0.63
3 10.65 10.92 9.94 99.00 99.09 98.55 5.31 0.27 -0.71 0.09 -0.45
4 12.39 12.15 11.30 98.35 98.59 98.15 5.01 -0.24 -1.09 0.24 -0.20
5 7.07 8.25 8.52 99.60 99.88 98.95 5.18 1.18 1.45 0.28 -0.65
6 9.41 9.85 9.95 98.85 99.41 98.54 5.31 0.44 0.54 0.56 -0.31
7 11.67 11.72 10.72 98.30 98.77 98.33 5.45 0.05 -0.95 0.47 0.03
8 12.99 12.13 11.41 97.80 98.49 98.11 5.46 -0.86 -1.58 0.69 0.31
9 25.84 17.02 21.23 90.80 95.39 94.47 5.52 -8.82 -4.61 4.59 3.67
10 21.36 16.09 21.21 90.70 96.23 94.75 4.54 -5.27 -0.15 5.53 4.05
11 22.74 15.50 21.20 91.40 96.18 94.78 3.57 -7.24 -1.54 4.78 3.38
12 22.18 22.74 21.08 91.20 96.02 94.77 2.64 0.56 -1.10 4.82 3.57
13 9.39 8.35 8.45 98.25 99.70 99.74 6.20 -1.04 -0.94 1.45 1.49
14 7.40 6.32 7.85 98.90 99.82 99.91 6.19 -1.08 0.45 0.92 1.01
15 5.10 5.00 7.38 99.00 99.88 99.97 6.12 -0.10 2.28 0.88 0.97
16 9.19 8.59 9.82 98.45 99.63 99.60 6.29 -0.60 0.63 1.18 1.15
17 7.07 7.62 8.56 98.50 99.76 99.79 6.26 0.55 1.49 1.26 1.29
18 6.43 6.99 7.42 98.70 99.83 99.89 6.19 0.56 0.99 1.13 1.19
19 5.41 6.13 6.12 98.80 99.93 99.96 6.65 0.72 0.71 1.13 1.16
20 11.51 12.44 11.28 97.95 98.71 98.25 5.79 0.93 -0.23 0.76 0.30
21 9.72 9.67 9.43 98.50 99.42 99.40 6.12 -0.05 -0.29 0.92 0.90
22 6.74 6.59 7.26 98.85 99.79 99.91 6.33 -0.15 0.52 0.94 1.06
23 3.37 4.08 5.46 99.00 99.96 99.99 6.33 0.71 2.09 0.96 0.99
24 16.63 14.33 10.86 95.85 97.47 98.69 5.85 -2.30 -5.77 1.62 2.84
25 10.76 11.55 9.62 96.80 98.72 99.57 6.52 0.79 -1.14 1.92 2.77
26 8.57 8.47 8.81 97.30 99.15 99.84 6.69 -0.10 0.24 1.85 2.54
27 8.26 7.21 8.44 97.65 99.10 99.91 6.62 -1.05 0.18 1.45 2.26
28 7.35 8.56 11.35 98.75 99.87 98.22 6.11 1.21 4.00 1.12 -0.53
29 11.95 14.27 11.17 97.70 97.92 98.31 5.50 2.32 -0.78 0.22 0.61
30 13.30 16.07 11.20 97.00 96.57 98.24 4.89 2.77 -2.10 -0.43 1.24
31 41.56 37.64 26.38 82.30 82.27 89.23 1.46 -3.92 -15.18 -0.03 6.93
32 22.46 16.07 15.11 93.70 94.97 96.70 5.45 -6.39 -7.35 1.27 3.00
33 12.54 8.00 8.99 98.20 100.00 99.06 12.62 -4.54 -3.55 1.80 0.86
34 13.83 13.09 9.86 93.80 94.45 96.06 0.99 -0.74 -3.97 0.65 2.26
35 23.05 16.73 14.25 93.70 97.58 97.75 14.13 -6.32 -8.80 3.88 4.05
42 22.40 16.46 20.51 89.00 94.40 94.96 5.95 -5.94 -1.89 5.40 5.96
43 22.37 16.15 20.06 82.30 89.72 90.12 5.42 -6.22 -2.31 7.42 7.82
Max negative deviation, % -8.82 -15.18 -0.43 -0.65
Max positive deviation, % 2.77 4.00 7.42 7.82
Overall average deviation, % 2.52 2.23 1.81 1.97

Page 11 of 16
3. MULTIBED PSA SIMULATION

As a basis for comparison, pore diffusion model with the data from Shin & Knaebel work
has been chosen. PSA flow-sheets with one, four, eight and twelve beds have been used.
Steps chosen are the same as in original Shin & Knaebel work, except the corresponding
number of pressure equalization steps were introduced in 4, 8 and 12-bed PSA. Whence, four-
bed configuration contains one, eight-bed contains two and 12-bed contains three pressure
equalization steps.
Two different sets of simulations were carried out:
¾ The first with a fixed cycle time (60 seconds) and fixed feed and purge gas
flowrates (with two different series of feed and purge gas flowrates).
¾ The second with a fixed amount of feed and purge gas and the same duration of
adsorption and purge steps.
Results are given in the tables 6.1 to 6.3. It could be seen that increase in bed number leads to
increase in product purity and recovery, decrease in power consumption and decrease in ad-
sorbent productivity. The results show typical trade-off in choosing between PSA plants with:
a) Higher number of the beds (higher capital costs) but better separation perform-
ance (increase in product purity and recovery – more valuable products), lower
energy consumption (lower operating costs) and lower adsorbent productivity
(due to longer cycle times),
b) Less number of the beds (lower capital costs) but poorer separation (decrease in
product purity and recovery), higher energy consumption (higher operating costs)
and better adsorbent productivity.

Table 3 – PSA multibed simulation results: Case 1a

Parameter 1 bed 4 beds 8 beds 12 beds

Purity, % 89.92 92.25 95.97 92.50

Recovery, % 24.11 26.43 20.89 26.70
Power per cycle, W 5.27 3.66 2.45 2.40
Productivity, mol/(kgs) 2.63x10-3 1.83x10-3 1.22x10-3 1.20x10-3
Cycle time, s 60 60 60 60

Page 12 of 16
 Table 4 – PSA multibed simulation results: Case 1b

Parameter 1 bed 4 beds 8 beds 12 beds

Purity, % 99.34 99.88 99.98 ~99.2

Recovery, % 9.98 10.93 7.61 ~13
Power per cycle, W 5.35 3.63 2.65 2.07
Productivity, mol/(kgs) 2.69x10-3 1.82x10-3 1.32x10-3 1.03x10-3
Cycle time, s 60 60 60 60

Table 5 – PSA multibed simulation results: Case 2

Parameter 1 bed 4 beds 8 beds 12 beds

Purity, % 86.96 91.02 ~95.5 92.50

Recovery, % 20.62 23.64 ~24 26.70
Power per cycle, W 5.84 4.00 2.23 2.40
Productivity, mol/(kgs) 2.91x10-3 2.00x10-3 1.11x10-3 1.20x10-3
Cycle time, s 30 40 80 60

4. CONCLUSIONS

The generic modelling framework for the separation of gas mixtures using multi-bed
Pressure Swing Adsorption (PSA) systems has been developed. The core of the framework is
highly detailed adsorbent model which takes into account coupled set of mixed algebraic and
partial differential equations for mass, heat and momentum balance, equilibrium isotherm
equations and operating steps dependant boundary conditions. The complex inter-bed con-
nections are controlled by the gas valves.
The developed models have been implemented in gPROMS modelling environment and
successfully validated against experimental and simulation data available in literature.
The modelling framework has successfully been applied in separation of nitrogen from the
air using multibed PSA systems. The results show that increase in bed number leads to in-
crease in product purity and recovery, decrease in power consumption and decrease in ad-
sorbent productivity, that is the typical trade-off between lower operating and bigger capital
costs.

Page 13 of 16
Nomenclature:
L, Bed length, m
Rp, Pellet radius, m
εbed, Porosity of the bed, -
εp , Porosity of the pellet, -
u, Interstitial velocity, m/s
Q, Adsorbed amount, mol/kg
Qp, Adsorbed amount in the pellet, mol/kg
Q*, Adsorbed amount in equilibrium state with gas phase, mol/kg
C, Molar concentrations of gas phase in bulk gas, mol/m3
Cp, Molar concentrations of gas phase in particles, mol/m3
T, Temperature of bulk gas, K
Tp, Temperature of particles, K
P, Total pressure in adsorbent bed, Pa
ρbc, Density of bulk gas, kg/m3
μbc, Viscosity of bulk gas, Pas
cpbc, Heat capacity of bulk gas, J/(kgK)
λbc, Thermal conductivity of bulk gas, J/(mK)
ρbed, Density of the bed, kg/m3
ρp , Density of the particles, kg/m3
cpp, Heat capacity of the particles, J/(kgK)
λp , Thermal conductivity of the particles, J/(mK)
Hads, Heat of adsorption, J/mol
Dz, Axial dispersion coefficient, m2/s
Dm, Molecular diffusion coefficient, m2/s
Ds, Surface diffusion coefficient, m2/s
De, Effective diffusivity coefficient, m2/s
Qm, Langmuire constant, mol/kg
B, Langmuire parameter, m3/mol
H, Henry’s parameter, m3/kg
kf, Mass transfer coefficient, m2/s
kh, Heat transfer coefficient, J/(m2Ks)
Z, Compressibility factor, -

Page 14 of 16
 References
1. Kawazoe, K., Takeuchi, Y. (1974). Mass transfer in adsorption on bidisperse porous
materials, Macro and micro-pore series diffusion model. Journal of chemical engineering
of Japan, Vol. 7, No.6.
2. Ruthven, D. M. (1984). Principles of adsorption and adsorption processes. John Wiley &
Sons, New York.
3. Yang, R. T. (1987). Gas separation by adsorption processes. Butterworths, Boston.
4. Raghavan, N.S., Hassan, M.M., Ruthven, D.M. (1985). Numerical simulation of a PSA
system—Part I: isothermal trace component system with linear equilibrium and finite
mass transfer resistance. A.I.Ch.E. Journal 31 (3), 385–392.
5. Raghavan, N., Ruthven D. (1985). Pressure swing adsorption. Part III: numerical
simulation of a kinetically controlled bulk gas separation. AlChE J. 31, 2017.
6. Ruthven, D. M., Raghavan, N. S., Hassan, M. M. (1986). Adsorption and Diffusion of
Nitrogen and Oxygen in a Carbon Molecular Sieve. Chem. Eng. Sci., 41, 1325.
7. Raghavan, N., Hassan M., Ruthven, D.M. (1986). Numerical simulation of a PSA system
using a pore diffusion model. Chem. Eng. Sci. 41, 2787.
8. Doong, J., Yang, R.T. (1986). Bulk Separation of Multicomponent Gas Mixtures by
Pressure Swing Adsorption: Pore/Surface Diffusion and Equilibrium Models. AIChE J.,
32, 397.
9. Yang, R.T., Doong, S.J. (1986). Bulk separation of multicomponent gas mixtures by
pressure swing adsorption: Pore/surface diffusion and equilibrium models. AIChE
Journal, 32, 397.
10. Yang, R.T., Doong, S.J. (1986). Parametric study of the pressure swing adsorption
process for gas separation: A criterion for pore diffusion limitation. Chem Eng.
Commun., Vol. 41, 163-180.
11. Doong, S.J., Yang, R.T. (1987). Bidisperse pore diffusion model for zeolite pressure
swing adsorption. AlChE J. 33, 1045.
12. Hassan, M.M., Raghavan, N.S., Ruthven, D.M. (1987). Numerical simulation of a
pressure swing air separation system a comparative study of finite diference and
collocation methods. Canadian Journal of Chemical Engineering 65, 512–516.
13. Raghavan, N., Hassan, M., Ruthven, D.M. (1987). Pressure swing air separation on a
carbon molecular sieve- II. Investigation of a modified cycle with pressure equalization
and no purge. Chemical Engineering Science, 42, 2037-2043.
14. Shin, H.S., Knaebel, K.S. (1987). Pressure swing adsorption: An theoretical study of
diffusion-induced separation. AIChE Journal, 1987, Vol.33, No.4.
15. Shin, H.S., Knaebel, K.S. (1988). Pressure swing adsorption: An experimental study of
diffusion-induced separation. AIChE Journal, Vol.34, No.9.
16. Ruthven, D.M., Farooq S. (1990). Air separation by pressure swing adsorption. Gas
Separation Purification 4, 141.
17. Lemcoff, N.O., LaCava, A.I. (1992). Effect of regeneration pressure level in kinetically
controlled pressure swing adsorption. Gas separation and purification, Vol 6 No 1.
18. Ruthven, D. M. (1992). Diffusion of Oxygen and Nitrogen in Carbon Molecular Sieve.
Chem. Eng. Sci., 47, 4305.

Page 15 of 16
 19. Ruthven, D. M., Farooq, S., Knaebel, K. S. (1994). Pressure swing adsorption. VCH
Publishers, New York.
20. Hartzog, D.G., Sircar, S. (1994). Sensitivity of PSA process performance to input
variables. Air Products and Chemicals Inc.
21. Nilchan, S., Pantelides, C.C. (1998). On the optimisation of periodic adsorption
processes. Adsorption 4, 113-147.
22. Pantelides, C.C (2000). The mathematical modelling of the dynamic behaviour of process
systems. Imperial College of Science, Technology and Medicine, London.
23. Chahbani, M.H., Tondeur D. (2000). Mass transfer kinetics in pressure swing adsorption.
Separation and Purification Technology 20 (2–3), 185–196.
24. Rajasree, R., Moharir, A.S. (2000). Simulation based synthesis, design and optimization of
pressure swing adsorption (PSA) processes. Computers and chemical engineering 24,
2493-2505.
25. Valve sizing. Technical bulletin, Swagelok company, July 2002.
26. Thomas, S.Y. Choong, Peterson, W.R., Scott, D.M. (2002). Development of novel
algorithm features in the modelling of cyclic processes. Computers and chemical
engineering 26, 95-112.
27. Jiang, L., Biegler, L. T., Fox, V. G. (2003). Simulation and optimization of pressure-swing
adsorption systems for air separation. AIChE J., 49 (5), 1140-1157.
28. Ko, D., Siriwardane, R., Biegler, L. T. (2003). Optimization of a pressure-swing
adsorption process using zeolite13X for CO2 sequestration. Ind. Eng. Chem. Res.f 42,
339-348.
29. Cruz, P., Santos, J.C., Magalhaes, F.D., Mendes A. (2003). Cyclic adsorption separation
processes: analysis strategy and optimization procedure. Chemical engineering science.
30. Cruz, P., Santos, J.C., Magalhaes, F.D., Mendes, A. (2003). Cyclic adsorption separation
processes: analysis strategy and optimization procedure. Chemical Engineering Science
58, 3143-3158.
31. Jiang, L., L. T. Biegler, and V. G. Fox (2004). Simulation and optimal design of multiple-
bed pressure swing adsorption systems. AIChE J., 50 (5), 2904-2917.
32. Ko, D., Siriwardane, R., Biegler, L. T. (2004). Simulation and Optimization of FVPSA for
CO2 Recovery. CAPD Annual Review Meeting, Carnegie Mellon University, Pittsburgh,
PA, Mar. 8.
33. Kim, M.B., Jee, J.G., Bae, Y.S., Lee C.H. (2004). Parametric study of pressure swing
adsorption process to purify oxygen using carbon molecular sieve. American Chemical
Society.
34. Chang, D., Min, J., Moon, K., Park, Y. K., Jeon, J.K., Ihm, S.K. (2004). Robust numerical
simulation of pressure swing adsorption process with strong adsorbate CO2. Chemical
Engineering Science 59, 2715-2725.
35. Jiang, L., Biegler, L. T., Fox, V. G. (2005). Design and optimization of pressure swing
adsorption systems with parallel implementation. Computers and Chemical Engineering
29 (5), 393-399.
36. Jee, J., Kim, M., Lee, C. (2005). Pressure Swing Adsorption Processes to Purify Oxygen
Using a Carbon Molecular Sieve. Chemical Engineering Science 60, 869.

Page 16 of 16

View publication stats