Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract: In this paper a rigorous rate-based model for the selective absorption and desorption
of sour gases in packed towers within a coke oven gas purication process (VACASULFwprocess) is presented. It considers multi-component mass transfer of CO2, H2S, NH3 and
HCN in aqueous potassium hydroxide (KOH) or potash (K2CO3) solutions and the inuence of
chemical reactions on mass transfer using enhancement factors. A modied Pitzer model is
used to estimate activity coefcients in the electrolyte solution for the calculation of the chemical
and the phase equilibria as well as the reaction kinetics (Bronsted-Bjerrum equation). Using data
from the literature this thermodynamically consistent approach was successfully validated for the
phase equilibria and the reaction kinetic calculation. Since no experimental data for the abovementioned chemical system in packed towers is available in the literature, own experiments were
carried out in a pilot plant packed tower. A newly developed experimental method using redundant measurements and a Data Reconciliation procedure improved the accuracy of concentration proles for the gas and the liquid phase. Using eight absorption and eight desorption
experiments, as well as industrial on-site measurements of the entire plant, the model was successfully validated. In addition, the model was used for global optimization using evolutionary
algorithms, revealing an optimization potential of 30% in operating costs.
Keywords: packed towers; reactive absorption; gas purication; sour gas treatment; mass transfer.
INTRODUCTION
Correspondence to:
R. Thiele, BASF AG, GCT/D
L540, 67056, Ludwigshafen,
Germany.
E-mail: robin.thiele@
basf.com
DOI: 10.1205/cherd06091
02638762/07/
$30.00 0.00
Chemical Engineering
Research and Design
Trans IChemE,
Part A, January 2007
# 2007 Institution
of Chemical Engineers
74
HCO
3
! H3 O OH
! NH
4 OH
K3
! CO2
3
H3 O
(R:2)
K4
! HS H3 O
K5
! S2 H3 O
HS H2 O
K6
(R:1)
K2
H2 O
H2 S H2 O
! H CN
HCN H2 O
(R:3)
(R:4)
(R:5)
(R:6)
K6
! CN H3 O
K7
! HCO
3
K8
HCO
3
(R:7)
! H2 NCOO H2 O
(R:8)
Strong electrolyte:
KOH
K9
! K OH
(R:9)
Kj
Electrolyte System
K1
NH3 H2 O
HCN
RATE-BASED MODEL
75
nc
Y
i1
ai i,j
nc
Y
mi i,j gi i,j
(1)
i1
where ai is the activity of component i, ni,j is the stoichiometrical coefcient of component i in reaction j and gi is the
activity coefcient taking non-idealities into account.
76
THIELE et al.
Figure 2. Comparison experiment-model (rate-based and equilibrium), absorption in 32 g l21KOH, F 1.8 Pa1/2, B 9 m3 m22 h21.
Phase Equilibrium
The equilibrium at the interface between the molecular
species is calculated using the Henry coefcient for dissolved
diluted gases as the standard-state fugacity.
xi gi HeH2 O,i yi Fi P
(2)
considered separately regarding balances for mass, components and enthalpy. Mass transfer between both phases
over the interfacial area is evaluated with the lm theory,
which assumes one-dimensional diffusion in the lm and a
completely mixed bulk phase. The cross-interactions of the
General Maxwell Stefan theory can be neglected due to
the low concentrations (Peters, 1997). The convective term
is also neglected.
The mass transfer rate J of component i for segment j is
then calculated by
Ji,j ki,jv rvj aj (yi,jb yi,jint )
ki,jl rlj aj Ei,j (xi,jint xi,jb )
(3)
The mass transfer coefcients kvi,j and kli,j for the gas and
liquid lm are calculated from correlations developed by
Billet and Schultes (1999). The interfacial area aj is calculated
from the same contributions. The increase in concentration
gradients at the interfacial area due to chemical reaction
leads to an enhancement of mass transfer which is
accounted for by the enhancement factor Ei,j, calculated
from correlations also applied by Hoogendoorn et al. (1988).
He used enhancement factors for reversible reactions of
rst order and pseudo-rst order derived from the surface
renewal theory by Danckwerts (1970).
Analogous to the mass transfer models, enthalpy
balances are formulated separately for the liquid and the
vapour phase and the interface, thus, considering the heat
transfer over the interface. Liquid and vapour enthalpies
are calculated from Pawlikowski and Newman (1982) and
Cox et al. (1989) and the liquid and vapour side heat transfer
coefcients are obtained from the weighted corresponding
component mass transfer coefcients using the Lewis
equation.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 7487
(4)
(5)
(6)
gCO2 gOH
gHCO3
(7)
Since the activity coefcients can also be calculated for complex solutions, the inuence of the non-ideality of the liquid
phase on the rate constant is accounted for by the thermodynamic model which is also used for the calculation of the
chemical equilibrium and the phase equilibrium. Thus a consistent thermodynamic approach is maintained throughout
the model. Haubrock et al. (2005) used a similar equation
as (7), however he only included the activity coefcients for
the reactants. Using this approach he was able to decrease
the inuence of the ionic strength, however a ratio of 2 for
k exp
OH2/kOH2 for intermediate ionic strength still persists.
77
78
THIELE et al.
PILOT PLANT
For the experimental validation of the simulation a pilot
plant at the TU Berlin was used which has been designed
for both absorption and desorption experiments under
atmospheric pressure and vacuum conditions. It consists
Figure 5. Flowsheet of the pilot plant in absorption mode. Desorption equipment is hidden.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 7487
79
Figure 6. Flowsheet of the pilot plant in desorption mode. Absorption equipment hidden.
(8)
L liquid flow,
j envelope 1, 2 and 3,
y gas concentration
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 74 87
80
THIELE et al.
MODEL VALIDATION USING PILOT
PLANT EXPERIMENTS
zyurt
Instead a robust estimator, the Fair-function is used (O
and Pike, 2004). The Data Reconciliation problem can then
be formulated as follows:
min J min
nmv
X
i1
c2
j1i j
j1i j
ln 1
c
c
xi x^ i
1i
xi [ V, L, c, y
si
s.t. f (x) 0
Lb x Ub
(9)
The Fair function introduces simultaneous gross error detection into the data reconciliation (DR) process. This can be
shown by the inuence function (Figure 8). The contributions
from the Fair function for larger standard errors are signicantly lower than for the weighted least squares (WLS)
function.
The resulting optimization problem [equations (8) and (9)]
consists of 31 variables and 12 equality constraints and is
non-linear. It can be solved using an NLP-Solver like the
SQP-Method. The DR problem including the solver method
has been implemented in MATLAB (Faber et al., 2004).
By applying the DR method the quality of the measurement
data is expected to be improved, which will be validated in the
next section.
Table 1. Experimental parameters for the three desorption experiments (D1, D7 and D8 from Thiele, 2007).
Case 1 (D1)
Process data
Vapour
Flow (kg h21)
Temp. (8C)
Liquid
Flow (l h21)
Temp. (8C)
CO2 (g l21)
H2S (g l21)
NH3 (g l21)
KOH (g l21)
Plant data
Top pres. (mbar)
Pres. drop (mbar)
F-factor (Pa0.5)
Case 2 (D7)
Case 3 (D8)
Feed
Outlet
Feed
Outlet
Feed
20
56.87
53.62
15
56.37
53.45
34
103.22
99.98
45.04
27.52
35.73
101.9
54.58
23.55
33.89
99.78
45.14
19.89
2.12
4.93
30.98
99.35
54.59
18.08
0.049
0.096
30.3
99.82
80.2
20.15
2.53
5.11
29.90
148.8
9.7
2.16
152.9
5.9
1.61
Outlet
102.05
103.52
103.0
14.58
0.624
0.0017
28.09
1123.8
3.12
1.47
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 7487
81
Table 2. Experimental parameters for three absorption experiments, results of the DR in brackets (A2, A6, A8 from Thiele, 2007).
Case 4 (A2)
Process data
Vapour
Flow (Nm3 h21)
Temp. (8C)
Presat. temp (8C)
CO2 (Vol%)
H2S (Vol%)
NH3 (Vol%)
Liquid
Flow (l h21)
Temp. (8C)
CO2 (g l21)
H2S (g l21)
NH3 (g l21)
KOH (g l21)
Plant data
Top pres. (mbar)
Pres. drop (mbar)
F-factor (Pa0.5)
Feed
47.09 (47.07)
25.38
23.77
2.383 (2.209)
Case 5 (A6)
Outlet
Feed
46.55 (47.33)
25.06
24.35
2.393 (2.335)
0.410 (0.404)
26.92
1.754 (1.749)
Case 6 (A8)
Outlet
Feed
46.88 (48.84)
25.00
19.78
2.012 (2.131)
0.514 (0.561)
0.9 (1.26)
24.88
2.369 (2.323)
0.0100 (0.0035)
69.95 (69.82)
25.02
5.97 (5.97)
70.37 (70.50)
27.87
12.04 (12.04)
69.95 (70.00)
25.02
12.38 (12.38)
0.25 (0.25)
70.83 (70.75)
25.81
12.59 (12.59)
4.32 (4.32)
35.84 (34.42)
33.42 (34.09)
28.48 (28.58)
28.35 (28.28)
1123
6.34
1.80
1094
6.45
1.80
69.94
24.98
17.33
0.87
0.00
31.98
(70.00)
(17.33)
(0.92)
(0.00)
(32.19)
Outlet
25.39
2.173 (2.072)
0.095 (0.088)
0.0050 (0.0025)
70.59
28.93
18.45
6.94
7.27
31.92
1110
6.11
1.80
(70.47)
(18.44)
(5.92)
(6.60)
(31.97)
82
THIELE et al.
83
INDUSTRIAL PLANT
In addition to our own pilot plant experiments, validation of
the model was also done using measurement data from an
industrial VACASULF process of Citizens Gas & Coke Utility
in Indianapolis, Indiana. The plant has been described in
Currey (1995). Measurements were taken in 1997 during a
start-up phase by Uhde GmbH. The data was then processed
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 74 87
84
THIELE et al.
exp
MaxDR Maxexp
0.37 1.24
1.09
3.37 3.08 12.24
4.94 6.85 13.96
6.66 37.66 18.44
8.25
12.16
32.24
54.57
exp
MaxDR Maxexp
1.28 4.53
3.27
21.41 36.91a 63.92
14.66
168.41
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 7487
85
heights) where chosen as decision variables for the optimization routine. The objective function was based on annualized costs comprising both investment and operating costs.
Costs functions were derived using condential cost data
from the industrial partner for running plants regarding operating costs and from realised projects regarding investment
costs. Investment cost functions are simple linear or degressive functions. Parameters for the evolutionary algorithm
were 10 individuals per generation, out of which two parents
were selected which gave by linear recombination the new
parent. The new generation was obtained by normal distributed mutation of the parent.
For an operated industrial VACASULFw plant similar to
Figure 15 the entire optimization resulted in a decrease of
the operation costs by 30% as can be seen from Figure 17.
The optimizer decreased the concentration of the potash
Table 4. Validation using data from the industrial process.
Deviations
Simulation $ Process
Rich potash solution
CCO2
CH2S
CHCN
Normal stripped
Potash solution
CCO2
CH2S
CHCN
213%
6%
65% (low conc.)
Highly stripped
Potash solution
CCO2
CH2S
CHCN
219%
219%
113% (low conc.)
29%
1%
23%
RESULTS
For the design and the optimization of the VACASULFwprocess, a selective absorption and desorption-process, a
new model has been presented. The model was successfully
validated by vapour liquid equilibrium data and reaction
kinetics from the literature and by the conducted eight
desorption experiments and eight absorption experiments in
the packed tower. The developed experimental procedure
included redundant measurements and a data reconciliation
method to improve the accuracy of the measurement data
and also automatically detected gross errors. The model
uses a thermodynamically consistent approach for both the
chemical and phase equilibria as well as the reaction kinetics.
For the rst time, it was shown that the Bronsted Bjerrum
equation is able to reproduce the non-ideality in the liquid
phase regarding the reaction kinetics for the formation of
bicarbonate using a modied Pitzer approach.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 74 87
86
THIELE et al.
NOMENCLATURE
ai
aj
E
J
k
K
L
mi
P
V
xi
yi
activity of component I
effective interfacial area for segment j, m2 m23
enhancement factor
mass transfer rate, mol s21
reaction velocity constant, 1 s21
equilibrium constant
liquid molar ow, mol h21
molality of component I, mol kg21
pressure, Pa
vapour molar ow, mol h21
mole fraction of component i in liquid
mole fraction of component i in vapour
Greek symbols
1
standard error
gi
activity coefcient of component I
stoichiometric coefcient of component i in reaction j
ni,j
r
density, kg m23
REFERENCES
Benson, H.E., Field, J.H. and Jimeson, R.M.; 1954, CO2 absorption:
employing hot potassium carbonate solutions, Chemical Engineering Progress, 50(7): 356.
Bjerrum, N., 1924, Zur Theorie der chemischen Reaktionsgeschwindigkeit, Zeitschrift fur physikalische Chemie, 108: 82100.
Billet, R. and Schultes, M., 1999, Prediction of mass transfer columns
with dumped and arranged packings, Chem Eng Res Des, 77:
498 504.
Brettschneider, O., 2003, Process Synthesis and Optimisation of the
Coke Oven Gas Process Synthese und Optimierung der Kokereigasreinigung (VDI-Verlag, TU Berlin, Germany).
Brettschneider, O., Thiele, R., Faber, R., Thielert, H. and Wozny, G.,
2004, Experimental investigation and simulation of the chemical
absorption in a packed column for the system NH3-CO2-H2SNaOH-H2O, Sep Pur Techn, 39(3): 139159.
Bronsted, J.N., 1922, Zur Theorie der chemischen Reaktionsgeschwindigkeit, In Zeitschrift fur physikalische Chemie, 102: 169207.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 7487
87
ACKNOWLEDGEMENTS
The support provided by VDKF e.V. and Chemstations Inc. is
gratefully acknowledged, as well as founding by Deutsche
Forschungsgemeinschaft for the partner project DFG LI806/4-3
that partially supported this work. The work presented in this paper
has been conducted during the previous position of Robin Thiele at
the TU Berlin. Statements and opinions presented in this paper are
not done in the name of BASF AG.
The manuscript was received 11 July 2006 and accepted for
publication after revision 24 October 2006.
Trans IChemE, Part A, Chemical Engineering Research and Design, 2007, 85(A1): 74 87