You are on page 1of 10

Ind. Eng. Chem. Res.

2001, 40, 4187-4196 4187

Kinetic Modeling of the Methanol to Olefins Process. 2.


Experimental Results, Model Discrimination, and Parameter
Estimation
Tae-Yun Park† and Gilbert F. Froment*,‡
Laboratorium voor Petrochemische Techniek, Universiteit Gent, Krijgslaan 281, B-9000 Gent, Belgium

The methanol to olefin process on ZSM-5 was studied in an integral tubular reactor at
atmospheric pressure and over a temperature range of 360-480 °C. Eight kinetic models based
upon the elementary steps of the conversion of methanol over dimethyl ether into olefins were
tested. They contained more than 30 parameters. The model parameters were transformed to
include the physicochemical constraints in the parameter estimation itself, instead of accounting
for these a posteriori. The estimation was performed by the genetic algorithm, followed by the
Levenberg-Marquardt routine, but in combination with sequential quadratic programming to
account for the constraints. The finally retained model corresponds to a mechanism that proceeds
over oxonium methylide formed from a methoxy ion interacting with a basic site of the catalyst.
The ylide subsequently reacts with dimethyloxonium ions to generate in parallel the primary
products ethylene and propylene. Through steps of carbenium ion chemistry, the latter lead to
higher olefins but also, to a lesser extent, to paraffins and aromatics.

Introduction microscopy. The particle size was less than 10 µm. The
surface area was on the order of 400 m2/g. The catalyst
In the first part of this work, the chemical pathways powder was pelletized by pressing it into wafers at 375
leading from methanol to olefins were detailed in their kg/cm2. It was then crushed and screened to 0.5-1.0
elementary steps and eight rival kinetic models were mm.
developed based on the steps leading to the formation
of the primary products: methane, ethylene, and pro-
pylene out of methanol and dimethyl ether (DME). Experimental Setup and Procedure
Higher olefins, paraffins, and aromatics are formed by The experimental setup is shown in Figure 1. During
a very large number of elementary steps of carbenium reaction the reactor is immersed in a molten salt bath,
ion chemistry. The kinetics of these steps were written but for the conditioning of the catalyst, it is lifted by an
in terms of single events so as to limit the number of oil system into an infrared-heated furnace. This com-
independent parameters. In this second part, the model bined unit has been described in detail by Lox et al.2
discrimination and parameter estimation is dealt with, The reactor tube itself has a length of 0.27 m and an
starting from experimental data obtained with a ZSM-5 internal diameter of 0.0214 m. For the experiments at
catalyst. temperatures above 450 °C, a titanium reactor was used
to avoid methanol decomposition. The catalyst bed was
Catalyst Synthesis and Characteristics diluted with inert oxide beads (5 vol. of beads inert/1
vol. of catalyst). The catalyst was pretreated by heating
ZSM-5 was synthesized as described by Jacobs and
it at 120 °C/h to 550 °C in a nitrogen flow (30 mL/min).
Martens.1 Reagents were silicon oxide (Aerosil, De-
Achieving steady state after a change in operating
gussa), NaAl2O2H2O (BHD Chemicals), sodium hydrox-
conditions took something like an hour. Catalyst de-
ide (Baker analyzed reagent), tetrapoly(ammonium
activation was slow and not noticed before at least 5 h
bromide) (TPABr; Fluka), and concentrated sulfuric acid
of operation. The catalyst was carefully regenerated in
(Baker). The specific silicon/aluminum ratio was ob-
air at 550 °C to avoid excessive hot spots. The initial
tained by adjusting the aluminum content in the
activity was completely recovered, even after some 50
synthesis mixture (silicon/aluminum ) 200). The crys-
regenerations.
talline product was dried at 80 °C and calcined in static
air at 550 °C for 12 h. NaHZSM-5 was acidified by ion The reactor effluent was analyzed on-line by means
exchange with NH4NO3 (1.0 N) for 3 h under total of a Carle CGC 500. The C4+ hydrocarbons were
reflux. The zeolite was identified as highly crystalline separated on a capillary column RSL-160 with a length
ZSM-5 by X-ray diffraction and by scanning electron of 30 m and identified by means of a flame ionization
detector. Light hydrocarbons, water, and the internal
standard, nitrogen, were separated by a series of packed
* To whom correspondence should be addressed. columns and detected by a thermal conductivity detector
† Present address: Catalytica Energy Systems, Inc., 430
(TCD). Hydrogen was quantitatively transferred from
Ferguson Drive, Mountain View, CA 94043-5272. Tel: +1-650-
940-6217. Fax: +1-650-618-1454. E-mail: taeyun@mail.com. the carrier gas, helium, into a nitrogen flow and detected
‡ Present address: Department of Chemical Engineering, by a separate TCD. Peak processing was carried out by
Texas A&M University, College Station, TX 77843-3122. Tel: a process and a personal computer. For more detailed
+1-979-845-3361. Fax: +1-979-845-6446. E-mail: g.froment@ analysis of the isomers, a liquid sample of the C6+
ChE.tamu.edu. hydrocarbons was analyzed by GC-MS. The use of an
10.1021/ie000854s CCC: $20.00 © 2001 American Chemical Society
Published on Web 06/16/2001
4188 Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001

Figure 1. Experimental fixed-bed setup for the kinetic study of the MTO process.

Figure 2. Methanol conversion vs space time at various temper-


atures.

internal standard allowed mass balances to be checked.


In all cases the closure exceeded 97%.
The kinetic data were collected at various settings of Figure 3. Selectivities for DME, olefins, and parafins + aromatics
temperature, inlet partial pressures of methanol, and as a function of methanol conversion at 440 °C and a total pressure
of 1.04 bar. Feed: methanol.
space time, τ ()W/FMeOH°). The experiments were
performed at five different temperatures in the range
of 360-480 °C. Depending on the temperature, the experiments amounted to 222. A few were duplicated,
space time ranged from 0.1 to 7 gcat‚h/mol. The total mainly for the purpose of collecting statistical informa-
pressure inside the reactor, pt, was 1.04 bar for all of tion. For the selected experimental conditions, internal
the experiments. The initial partial pressure of metha- and external heat- and mass-transfer limitations were
nol was varied by diluting the methanol feed with insignificant, mainly because of the small particle size
bidistilled water and nitrogen. The total number of of the catalyst. The test calculations3 were based on gas
Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001 4189

Figure 4. Yields of MTO products vs space time.

Table 1. Definition of the Parameters To Be Estimated after the Reparametrization of Rate and Equilibrium Constants
Pi definition Pi definition
P1 ∆SPr°(MeOH)/R - ∆HPr°(MeOH)/RTm P 18 +
Esr(OM;DMO+:R2)/R
P2 ∆HPr°(MeOH)/R P 19 ln Asr(OM;DMO+:R+ + +
3 )-Esr(OM;DMO :R3 )/RTm
P3 ∆SHyd°(R+ +
1 )/R - ∆HHyd°(R1 )/RTm
P 20 Esr(OM;DMO+:R+ 3 )/R
+
P4 ∆HHyd°(R1 )/R P 21 ln APr(O2) - EPr(O2)/RTm
P5 ln AC(R+ +
1 ) - EC(R1 )/RTm
P 22 EPr(O2)/R
P6 EC(R+1 )/R P 23 ∆S̃Pr/R
P7 ln AF(DME) - EF(DME)/RTm P 24 ∆HPr°(O2)/R
P8 EF(DME)/R P 25 ∆HPr°(O3)/R
P9 ∆SPr°(DME)/R - ∆HPr°(DME)/RTm P 26 ∆HPr°(O4r)/R
P 10 ∆HPr°(DME)/R P 27 ∆HPr°(O5r)/R
P 11 ln AF(CH4) - EF(CH4)/RTm P 28 ∆HPr°(O6r)/R
P 12 EF(CH4)/R P 29 ∆HPr°(O7r)/R
P 13 ln Asr(R+ +
1 ;bs) - Esr(R1 ;bs)/RTm
P 30 ∆HPr°(O8r)/R
P 14 Esr(R+1 ;bs)/R P 31 t
ln(CH +Ã)

P 15 ln Asr(OM;H+) - Esr(OM;H+)/RTm P 32 R
P 16 Esr(OM;H+)/R P 33 E°/RR
P 17 ln Asr(OM;DMO+:R+ + +
2 ) - Esr(OM;DMO :R2 )/RTm

mixtures containing methanol, DME, and olefins with Reparametrization


up to eight carbon atoms.
To reduce the correlation between preexponential
Experimental Results factors and activation energies, reparametrization has
been applied.6 The Arrhenius form of the rate coefficient
Figure 2 shows the conversion of methanol as a is given by
function of space time, represented by τ, and Figure 3
typical selectivities (number of moles formed per 100 ki ) Ai exp(-Ei/RT) (1)
mol of methanol reacted) of DME, olefins, (from C3 to
C8), and aromatics (from C6 to C10) as a function of After introduction of the mean temperature, Tm, the rate
methanol conversion. DME is rapidly formed out of coefficients for the formation of primary products can
methanol but also rapidly consumed. The major prod- be written as

[( ) ( )]
ucts are olefins, but the selectivity toward olefins
reaches a maximum around a methanol conversion of Ei Ei 1 1
ki ) exp ln Ai - - - (2)
95%. The production of paraffins and aromatics is RTm R T Tm
negligible as long as the methanol conversion is kept
below 70%. Figure 4 shows yields (grams formed per Note that the total concentration of acid and/or basic
100 g of methanol fed) versus the space time based upon t t
sites (CH + or Cbs) is incorporated into the rate coef-
methanol. The main product is propylene, while the ficient given in eqs 1 and 2.
ethylene yield levels off at high space times. Although The rate coefficient (2) can also be written as
it has been reported that methane is one of the main
products at low methanol conversion,4 this is not the
case in this work. Calculations showed that at high
space times the composition of the olefin fraction was
[
ki ) exp P i - P k (1
-
T Tm
1
)] (3)

at equilibrium, confirming literature information.5 in which the temperature-independent parts (ln Ai -


4190 Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001

Table 2. List of Physicochemical Constraints for the Parameters of Mechanism a′′


number constraints physical meaning
1 P 25 < P 24 ∆HPr°(O3) < ∆HPr°(O2)
2 P 26 < P 25 ∆HPr°(O4r) < ∆HPr°(O3)
3 P 27 < P 26 ∆HPr°(O5r) < ∆HPr°(O4r)
4 P 28 < P 27 ∆HPr°(O6r) < ∆HPr°(O5r)
5 P 29 < P 28 ∆HPr°(O7r) < ∆HPr°(O6r)
6 P 30 < P 29 ∆HPr°(O8r) < ∆HPr°(O7r)
7 P 3 + P 4/Tm < 0 ∆SHyd°(R+ 1) < 0
8 P 5 + P 6/Tm - ln(kBTminCH t
+/h) < 0 ∆SC°q(R+ 1) < 0
9 P 7 + P 8/Tm - ln(kBTminCH t
+/h) < 0
∆SC°q(DME) < 0
10 P 21 + P 22/Tm - ln(kBTminCH t
+/h) < 0
∆SPr°q(O2) < 0
11 and 12 -S°(MeOH)/R < P 1 + P 2/Tm < -41.8/R -S°(MeOH) < ∆SPr°(MeOH) < -41.8
13 and 14 -S°(DME)/R < P 9 + P 10/Tm < -41.8/R -S°(DME) < ∆SPr°(DME) < -41.8
15 and 16 -S°(O2)/R - ln(σO R2+ O2 R2+
gl /σgl ) < P 23 < - 41.8/R - ln(σgl /σgl )
2 -S°(O2) < ∆SPr°(O2) < -41.8
17 1.4 × 10-3P 2 - (P 1 + P 2/Tm) < 51.04/R 1.4 × 10-3∆HPr°(MeOH) - ∆SPr°(MeOH) < 51.04
18 1.4 × 10-3P 10 - (P 9 + P 10/Tm) < 51.04/R 1.4 × 10-3∆HPr°(DME) - ∆SPr°(DME) < 51.04
19-25 1.4 × 10-3Pj - (P 23 + ln(σO Rir’+
gl /σgl )) < 51.04/R, j ) 22 + i
ir 1.4 × 10-3∆HPr°(Oir) - ∆SPr°(Oir) < 51.04, i ) 2, 3, ..., 8
26∼116 ∆HMe(i,j;P k) < 0, i ) category index, j ) reaction index ∆HMe(i,j) < 0, i ) category index, j ) reaction index
117∼168 ∆HOl(i,j;P k) < 0, i ) category index, j ) reaction index ∆HOl(i,j) < 0, i ) category index, j ) reaction index
169∼259 |∆HMe(i,j;P k)|/R - P 33 < 0, i ) category index, j ) reaction index EMe(i,j) > 0, i ) category index, j ) reaction index
260∼311 |∆HOl(i,j;P k)|/R - P 33 < 0, i ) category index, j ) reaction index EOl(i,j) > 0, i ) category index, j ) reaction index

Ei/RTm) and Ei/R are represented by P i and P k, product of R and ∆HMe(i,j). It is known that nonlinear
respectively. These are the parameters to be estimated, regression becomes difficult in the presence of nonlinear
rather than Ai and Ei. constraints.8 To avoid this problem, the rate coefficient
In a similar way the equilibrium constants involved represented in eq 6 was reparametrized as follows:
in the formation of primary products can be written as

[
Ki ) exp P l - P m (1
-
T Tm
1
)] (4) kMe(i,j) ) ne(i,j) exp P k - { Pl
T (
Pm-
|∆HMe(i,j)|
R )}
(7)
with P l and P m based on the enthalpy and entropy
changes between reactant(s) and product(s) of the given where P k ) ln(CH t
+Ã), P l ) R, and P m ) E°/RR. The
reaction, i.e. constraint for satisfying positive activation energy now
∆S° ∆H° ∆H° becomes linear:
Pl ) - , Pm) (5)
R RTm R
|∆HMe(i,j)|
The rate coefficients for the formation of higher olefins Pm> (8)
were reparametrized in a different way. They were R
written in terms of the single-event approach and the
Evans-Polanyi relation.7 For instance, the rate coef- The protonation equilibrium constant for the various
ficient for the ith category and the jth methylation is reference olefins is expressed as follows:
given by
σO ir
1
KPr(Oir) )
gl

σRglir'
+ ( )
exp P j - P k , i ) 2, 3, ..., 8
T
(9)

where P j ) ∆S̃Pr/R and P k ) ∆HPr°/R.


According to the differences in stability between
cations established in carbenium ion chemistry, the
protonation enthalpies of the various reference olefins
have to satisfy the following relationship:
The E° - R|∆HMe(i,j)| should be positive, because it
represents the activation energy for the given elemen- ∆HPr°(O8r) < ∆HPr°(O7r) < ... < ∆HPr°(O3) <
tary step. In addition, the heat of formation, ∆HMe(i,j), ∆HPr°(O2) (10)
contains the protonation enthalpies of the reference
olefins, which have to be estimated also. The reparam-
etrization described in eqs 3-6 was applied to the rate The enthalpy differences in the various protonation
and equilibrium constants of the eight rival kinetic processes are mainly determined by the differences in
models. As an example, the definition of the 33 param- the heat of formation of the various carbenium ions,
eters P i of the kinetic model based upon mechanism while the olefins are highly stable species, so that the
a′′ is shown in Table 1. contribution of the difference in the heat of formation
for the various olefins is negligible compared with that
of the carbenium ions.
Physicochemical Constraints
The physicochemical relationship given in eq 10 can
The condition that E° - R|∆HMe(i,j)| should be positive readily be expressed in terms of the parameters defined
leads to a nonlinear constraint, resulting from the in eq 9. For the kinetic model based on mechanism a′′
Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001 4191
t
i.e., k ) CH +k′, the following relationship, obtained by
comparing eq 3 with eq 12, imposes a negative ∆S°q:
t
Pk kBTminCH +
Pj+ - ln <0 (13)
Tm h

If the constraint (13) is satisfied at the minimum


temperature, Tmin, it is also satisfied over the complete
temperature range.
For the parameters involved in the protonation steps,
Boudart’s criteria9 define a rigorous set of constraints
for the standard protonation enthalpy (∆HPr°) and
Figure 5. General flow diagram of the application of the hybrid entropy (∆SPr°)
GA to the constrained parameter estimation.
∆HPr° > 0
of part 1, the relationship (10) can be written using the
parameters defined in Table 1 as 0 < -∆SPr° < Sg°
P30 < P29 < ... < P25 < P24 (11) 41.8 < -∆SPr° < 51.04 + 1.4 × 10-3(-∆HPr°) (14)

The nature of the elementary step provides an ad- where Sg° is the standard entropy of the molecule in
ditional relationship for the entropy of activation, ∆S°q. the gas phase. Combining the second and third inequali-
For elementary steps of the type A + B f [A‚‚‚B]qf AB, ties provides the range for the protonation entropy:
∆S°q is negative. From the transition state theory, the
rate coefficient for the elementary step can be written -Sg° < ∆SPr° < -41.8 (15)
as

( ) ( )
With the remaining relationship, i.e., ∆SPr° < 51.04 +
kBT ∆S°q E
k′ ) exp exp - (12) 1.4 × 10-3(-∆HPr°), the area of the parameters that
h R RT satisfy the Boudart criteria can be defined. In the case
of the kinetic model based on the mechanism a′′, for
t
If the total concentration of acid sites, CH +, is incorpo- instance, it can be shown from Boudart’s criteria that
rated into the rate coefficient of the elementary step, the parameters related to the protonation of methanol

Figure 6. Performance of the hybrid GA in the estimation of the parameters of the kinetic model based upon mechanism a′′.
4192 Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001

Table 3. List of Final Parameters for Retained Mechanism a′′


Pi lower limita estimate upper limita |t value| parameterb values unit
P1 -4.6433 × 100 -3.9404 × 100 -3.2374 × 100 11.21 ∆SPr°(MeOH) -1.3391 × 102 J‚mol-1‚K-1
P2 -8.5333 × 103 -8.3354 × 103 -8.1375 × 103 84.25 ∆HPr°(MeOH) -6.9305 × 101 kJ‚mol-1
P3 -4.9843 × 100 -4.2179 × 100 -3.4514 × 100 11.01 ∆SHyd°(R+ 1)
-8.5028 × 101 J‚mol-1‚K-1
P4 -4.1758 × 100 -4.1168 × 103 -4.0578 × 103 139.50 ∆HHyd°(R+ 1)
-3.4229 × 101 kJ‚mol-1
P5 3.1244 × 100 3.8080 × 100 4.4916 × 100 11.14 AC′(R+1)
9.3907 × 105 s-1‚bar-1
P6 5.8509 × 103 5.9878 × 103 6.1247 × 103 87.47 EC(R+ 1)
4.9786 × 101 kJ‚mol-1
P7 1.8134 × 101 1.8561 × 101 1.8987 × 101 87.05 AF′(DME) 1.2343 × 109 s-1‚bar-1
P8 7.7518 × 100 8.0004 × 102 8.2492 × 102 64.34 EF(DME) 6.6520 × 100 kJ‚mol-1
P9 -3.8957 × 100 -3.3374 × 100 -2.7792 × 100 11.96 ∆SPr°(DME) -8.6317 × 101 J‚mol-1‚K-1
P 10 -4.9289 × 103 -4.8263 × 103 -4.7237 × 103 94.08 ∆HPr°(DME) -4.0128 × 101 kJ‚mol-1
P 11 2.8721 × 10-1 7.1901 × 10-1 1.1508 × 100 3.33 AF′(CH4) 1.3978 × 1010 s-1‚bar-1
P 12 1.4284 × 104 1.4687 × 104 1.5090 × 104 72.89 EF(CH4) 1.2212 × 102 kJ‚mol-1
P 13 8.3370 × 100 9.0938 × 100 9.8569 × 100 23.83 Asr′(R+1 ;bs)
7.1483 × 1017 s-1
P 14 1.6298 × 104 1.6574 × 104 1.6850 × 104 120.20 Esr(R+1 ;bs)
1.3781 × 102 kJ‚mol-1
P 15 8.2304 × 100 9.0800 × 100 9.9296 × 100 21.38 Asr′(OM;H+) 2.3491 × 1014 s-1
P 16 1.0904 × 104 1.1088 × 104 1.1273 × 104 120.31 Esr(OM;H+) 9.2193 × 101 kJ‚mol-1
P 17 7.9092 × 100 8.3949 × 100 8.8806 × 100 34.57 Asr′(OM;DMO+:R+2)
7.7433 × 108 s-1
P 18 2.8230 × 103 2.9090 × 103 2.9950 × 103 67.64 Esr(OM;DMO+:R+ 2)
2.4187 × 101 kJ‚mol-1
P 19 7.2316 × 100 7.7135 × 100 8.1955 × 100 32.01 Asr′(OM;DMO+:R+3)
9.7050 × 1011 s-1
P 20 8.1202 × 103 8.2635 × 103 8.4068 × 103 115.33 Esr(OM;DMO+:R+ 3)
6.8707 × 101 kJ‚mol-1
P 21 -8.7386 × 100 -8.1645 × 100 -7.5950 × 100 28.45 APr′(O2) 8.5785 × 105 s-1‚bar-1
P 22 1.3916 × 104 1.4129 × 104 1.4342 × 104 132.87 EPr(O2) 1.1748 × 102 kJ‚mol-1
P 23 -8.9252 × 100 -8.4304 × 100 -7.9357 × 100 34.08 ∆S̃Pr -7.0095 × 101 J‚mol-1‚K-1
P 24 -2.3453 × 103 -2.3008 × 103 -2.2563 × 103 103.40 ∆HPr°(O2) -1.9130 × 101 kJ‚mol-1
P 25 -9.4422 × 103 -9.3039 × 103 -9.1656 × 103 134.55 ∆HPr°(O3) -7.7358 × 101 kJ‚mol-1
P 26 -9.9335 × 103 -9.9335 × 103 -9.6959 × 103 81.59 ∆HPr°(O4r) -8.0616 × 101 kJ‚mol-1
P 27 -1.0488 × 104 -1.0488 × 104 -1.0371 × 104 176.73 ∆HPr°(O5r) -8.6230 × 101 kJ‚mol-1
P 28 -1.4485 × 104 -1.4235 × 104 -1.3985 × 104 113.98 ∆HPr°(O6r) -1.1836 × 102 kJ‚mol-1
P 29 -1.4767 × 104 -1.4585 × 104 -1.4404 × 104 160.61 ∆HPr°(O7r) -1.2127 × 102 kJ‚mol-1
P 30 -1.4962 × 104 -1.4626 × 104 -1.4290 × 104 87.06 ∆HPr°(O8r) -1.2161 × 102 kJ‚mol-1
P 31 1.4451 × 101 1.5390 × 101 1.6328 × 101 32.80 Ã′ 1.6111 × 107 s-1‚bar-1
P 32 3.2348 × 10-2 3.4304 × 10-2 3.6259 × 10-2 35.09 R 3.4304 × 10-2 dimensionless
P 33 3.3719 × 105 3.4183 × 105 3.4647 × 105 147.40 E° 9.7496 × 101 kJ‚mol-1
a Approximate 95% confidence limit. b Original parameters included in the reparametrized form.

defined in Table 1 should satisfy the following inequali- should be satisfied:


ties:

-
S°(MeOH)
< P1 +
P2
<-
41.8 Pk -
Pl
T
Pm-(|∆HMe(i,j)|
R
>0 ) (18)
R Tm R
A similar constraint can be written for the activation
-3
1.4 × 10 P2 - P1 + (
P2
Tm
<
51.04
R ) (16)
energy for oligomerization. In the case of β scission, the
relationship (18) is not necessary, because it has been
shown in part 1 that the rate coefficients for the
elementary steps of β scission are not independent: they
The range of ∆SPr° for methanol protonation can be
are calculated from the rate coefficients and the equi-
obtained from eq 16, but generally the range of ∆HPr°
librium constants of protonation. For the kinetic model
is not available. A range extending from 0 to -200 kJ/
based upon mechanism a′′, a total of 311 linear physi-
mol has been selected. The Boudart criteria have been
cochemical constraints, given in Table 2, has been
applied to all of the protonation processes occurring in
implemented.
MTO elementary steps, which include the protonation
of methanol, DME, and various olefins.
The heat effect of the elementary steps of methylation Objective Function
and oligomerization should be negative, and this has
been imposed as another physicochemical constraint. The objective function to be minimized in the estima-
Because the calculation of the heats of formation of tion of the kinetic parameters was based upon the
those elementary steps involves the protonation enthal- difference between experimental and calculated yields
pies of olefins, this constraint leads to of the MTO products:
nrespnresp nexp
∆HMe(i,j,P k) < 0
FM(P j) ) ∑∑ ∑ σhk
(yih - ŷih(P j))(yik - ŷik(P j))
∆HOl(i,j,P k) < 0 (17) h)1 k)1
i)1
(19)
Because of the large number of elementary steps, This is the generalized least-squares criterion, derived
eq 17 generates a large number of constraints. from the maximum likelihood criterion, as reviewed by
The activation energies of each elementary step of Froment and Hosten.10 It is thereby assumed that the
methylation should be positive. Therefore, with the differences between experimental and calculated yields
parameters defined in eq 7, the following inequality are normally distributed with zero mean and that those
Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001 4193

associated with the hth and kth responses are inde-


pendent. The σhk are usually unknown, but they can be
estimated for each response from replicated experi-
ments.3 In the present study, however, the σhk are
considered to be weighting factors which more or less
equilibrate the contributions of the different yields in
the objective function. The weighting factor for the hth
response is defined as

nexp

( ∑
i)1
yih)-$
(σhh)-1 ) (20)
nresp nexp

∑ (∑yik)
k)1 i)1
-$

Figure 7. Experimental (points) and calculated ethylene yields


For $ ) 1, the weighting factors express the relative for mechanisms a′, a′′, and b′ (curves).
importance of the responses, while for $ ) 0, all of the
responses are equally weighted. In the present study,
a value of 0.3 has been used for O7, O8, and methane.
For the rest of the responses, $ has been set to 1. The
predicted responses, ŷi(P j) in eq 19, are calculated from
the continuity equations for the components i in the
integral plug-flow reactor, dFi/dW ) Ri. Because of the
stiff character of the set, resulting from the very fast
rate of formation of DME, Gear’s method11 was used for
the integration of the set.

Constrained Parameter Estimation


To avoid getting trapped in local minima, the param-
eters of the various rival models were estimated in a
first step using the hybrid genetic algorithm (GA)
developed by Park and Froment.12 To increase their
accuracy, the parameter values thus obtained were used
in a second step as initial guesses for the local optimizer.
Because the Levenberg-Marquardt algorithm is an
unconstrained optimization algorithm, a sequential
quadratic program, called FFSQP,13 was used to account
for the constraints listed in Table 2, but the ultimate
parameter estimation was carried out by the Leven-
berg-Marquardt technique.
Figure 5 shows a flow diagram of the complete
estimation procedure. Before the set of parameters
obtained from the GA search is inserted as an initial
guess into the constrained optimizer FFSQP, it is
checked if the parameters satisfy the physicochemical
constraints listed in Table 2. The final estimation is
performed by the Levenberg-Marquardt algorithm,
after which it is checked whether the statistical criteria
(F test on the model and t test on the parameters) and
the physicochemical constraints are still satisfied. If this
is not the case, the procedure returns to the initial
search by the GA.
Figure 6 illustrates the application of the procedure
to the kinetic model derived from mechanism a′′. Three
steps enter in the generation of the initial guess by the Figure 8. Experimental (points) and calculated yields (curves)
GA, as shown in part a of Figure 6, illustrating the for various MTO products vs space time.
evolution of the objective function for the best set of
parameters as a function of the number of GA iterations. lower than 105, is the corresponding set of parameters
In the upper region, the GA is searching for the set of accepted as an initial guess for the local optimization.
parameters satisfying all of the physicochemical con- This is shown in the lower region (part a) of Figure 6.
straints listed in Table 2. Once such a set is found, the In part b of this figure, the performance of the local
GA starts solving the coupled nonlinear differential optimization by the FFSQP and Levenberg-Marquardt
equations (21) to obtain the values of the objective routines is illustrated. Sets of parameters are retained
functions for each set of parameters. Only when the as initial guesses for the local optimizers only when the
value of the objective function is sufficiently low, i.e., current value of the objective function has been suf-
4194 Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001

Figure 9. Parity plots for various products for temperatures ranging from 360 to 480 °C, a total pressure of 1.04 bar, and space time
ranging from 0.2 to 2 gcat‚h/mol. Feed: MeOH (+ N2 + H2O).

ficiently improved. The number of iterations performed ing that more than half of the sets of parameters in the
by the local optimizers is also listed in Figure 6. group is still distributed over the whole parameter
Although the set of parameters estimated from the space. The off-line performance shown in part c of
initial guess (5) shown in part b was found to satisfy Figure 6 evolves quite satisfactorily over the GA search,
all of the physicochemical constraints as well as the comprising 120 iterations. The number of parameter
statistical tests, the hybrid procedures were continued sets used in this run amounted to 1000. The crossover
until the selected number of GA iterations was com- and mutation probability were 0.10 and 0.005, respec-
pleted. This policy was chosen to confirm that the tively. Because of the large amount of calculation
parameter estimates derived from the initial guess (5) required for the evaluation of the objective function, not
truly correspond to the global minimum. More than 100 all of the 222 data sets were used in the model
different initial guesses generated by the GA have been discrimination and parameter estimation. A careful
tried, but none outperformed the initial guess (5). The selection reduced this set to 31, thus yielding a degree
bias curve shown in part c of Figure 6 did not reach a of freedom of 246 (31 experiments × 9 responses - 33
value higher than 0.7 until 120 GA iterations, confirm- parameters).
Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001 4195

Results and Discussion but also sequential quadratic programming to account


for the 311 physicochemical constraints. These were
Table 3 shows the set of parameters involved in the introduced into the optimization proper rather than
kinetic model based on the mechanism a′′, which is the verified a posteriori. The procedure led to a clear-cut
retained model. The parameter estimates satisfy all of discrimination between eight rival models and retained
the physicochemical constraints listed in Table 2. As a model in which the formation of the primary products
shown in Table 3, the calculated t values confirm that ethylene and propylene resulted from a reaction be-
all of the parameters are statistically significant. The tween oxonium methyl ylide and dimethyloxonium ion.
calculated F value was 1.1 × 106. The absolute values The production of higher olefins proceeds over elemen-
of the binary correlation coefficients between param- tary steps of carbenium ion chemistry, and their kinetic
eters were generally lower than 0.3. The rate of forma- formulation made use of the single-event concept and
tion of DME, out of DMO+, itself formed by methylation the Evans-Polanyi relationship. The model yields an
of methanol (viz., step i.4 of Table 1 of part 1), is excellent fit of the experimental data, as evidenced by
extremely rapid, even at low temperature (E ) 6652 the parity plots and the statistical F test. The param-
J/mol). The rate coefficient of R+ 2 formation out of eters all satisfied the statistical t test.
oxonium ylide, OM, and DMO+ (step iii.a′′.2 in Table 1 A tool is now available for an optimal design and
of part 1) exceeds that of R+ 3 formation out of the same operation of the MTO reactor and for a more oriented
reactants (viz., step iii.a′′.4 in Table 1 of part 1) at low specification of the desired properties of an MTO
temperatures, but not any more above 485 °C, because catalyst.
of the higher activation energy of the latter step (68.71
kJ/mol versus 24.19 kJ/mol for the former). The forma-
tion of methane is slow and requires an activation Notation
energy of 122 kJ/mol. The ∆HPr° of the reference olefins
evolves from -19.13 kJ/mol for O2, over -77.3 kJ/mol Ai ) preexponential factor of an elementary step i incor-
for O3, to -121.61 kJ/mol for O8r. porating Ctbs and/or CHt
+

The difference between the rival kinetic models Ãi ) single-event preexponential factor of reaction type i
originates from different pathways for the formation of Ctbs ) total concentration of basic sites, mol/gcat
t
the primary products of the MTO process. Ethylene is CH + ) total concentration of acid sites, mol/gcat

the common primary product for all of the kinetic Ea° ) intrinsic activation barrier in the Evans-Polanyi
models, whereas in some models propylene is a second- relation, J/mol
ary product only, formed out of ethylene. The evolution Ei ) activation energy of reaction type i
of the experimental yield of ethylene as a function of FM ) multiresponse objective function
space time is S-shaped. Only three kinetic models out h ) Plank constant: 1.841 × 10-37, J‚h
of eight were found to predict ethylene yields that come Ki ) equilibrium constant of an elementary step i
close to the experimental value, and these are given in kB ) Boltzmann constant: 1.381 × 10-23, J/K
Figure 7. The kinetic models based on trimethyloxonium ki′ ) rate coefficient of an elementary step i
as a central intermediate could not reproduce such a kMe(i,j) ) rate coefficient of methylation in category i, and
shape, as exemplified by the kinetic model based upon reaction j, incorporating CHt
+
mechanism b′. The calculated ethylene yield curves kMe′(i,j) ) rate coefficient of methylation in category i and
reflect the experimental trend only when the kinetic reaction j
model is based on the oxonium methyl ylide mechanism ki ) rate coefficient of step i, incorporating CH t t
+ and/or Cbs
[(a-a′-a′′) of part 1]. Among them, the model based on k̃ ) single-event rate coefficient
the mechanism a′′ gave the best fit of the experimental ne ) number of single events
ethylene yield. Although the kinetic model based upon nexp ) number of experiments
mechanism a, which is the original mechanism sug-
nresp ) number of independent responses
gested by Hutchings and Hunter,4 leads to an S-shaped
nprm ) number of parameters to be estimated
curve, the predicted values of the ethylene yield were
Oij ) olefin with carbon number i (i ) 2, 3, ..., 8) and isomer
too low. The kinetic model based upon the mechanism
index j
a′′ with the set of parameters given in Table 3 also
P ) parameters to be estimated after reparametrization
yielded an excellent fit of the complete product spec-
of rate and equilibrium constants
trum, as shown by way of example for the data obtained
at 440 °C in Figures 7 and 8. Figure 9, finally, presents R+ ij ) carbenium ion with carbon number i (i ) 1, 2, ..., 8),
and isomer index j
parity plots of the various MTO products for the
complete range of experimental conditions covered by Ri ) net reaction rate for gas-phase species i, mol/gcat‚h
the data used for the model discrimination and param- R ) gas constant: 8.314, J/mol‚K
eter estimation. r(i) ) reaction rate for species i, mol/gcat‚h
ri(j) ) reaction rate of type i for species j, mol/gcat‚h
ri(j,k) ) reaction rate of an elementary step of type i at jth
Conclusion category and
kth reaction, mol/gcat‚h
A carefully selected set of experiments allowed one S°(i) ) standard entropy of component i, J/mol‚K
to significantly determine 33 parameters in each of a T ) temperature, K
number of kinetic models derived from a detailed Tm ) mean temperature, K
mechanistic description of the MTO process. The pa- Tmin ) minimum temperature, K
rameter estimation involved the minimization of a W ) amount of catalyst, gcat
multiresponse objective function by nonlinear regres- yih ) experimental yield of response h for the ith experi-
sion. This is a substantial effort comprising a combina- ment
tion of the GA and the Levenberg-Marquardt routine ŷih ) calculated yield of response h for the ith experiment
4196 Ind. Eng. Chem. Res., Vol. 40, No. 20, 2001

Greek Letters Observation Concerning the Mechanism of Formation of the


Primary Products. Catal. Today 1990, 6, 279.
R ) transfer coefficient in the Evans-Polanyi relation (5) Quann, R. J.; Green, L. A.; Tabak, S. A.; Krambeck, F. J.
∆HPr° ) heat of protonation, J/mol Chemistry of Olefin Oligomerization over ZSM-5 Catalyst. Ind.
σhk ) element of the inverse of nresp × nresp error covariance Eng. Chem. Res. 1988, 27, 565.
matrix (6) Kitrell, J. R. Mathematical Modeling of Chemical Reactions.
∆S°q ) standard entropy of activation, J/mol‚K Adv. Chem. Eng. 1970, 8, 97.
∆S̃Pr ) protonation entropy excluding symmetry contribu- (7) Vynckier, E.; Froment, G. F. Modeling of the Kinetics of
tion, J/mol‚K Complex Processes based upon Elementary Steps. In Kinetic and
wi ) molecular weight of species i, g/mol Thermodynamic Lumping of Multicomponent Mixtures; Astarita,
G., Sandler, S. I., Eds.; Elsevier Science Publishers BV: Amster-
Subscripts dam, The Netherlands, 1984; p 131.
bs ) basic site (8) Press: W. H.; Flannery, B. P.; Teukolsky, S. A.; Vetterling,
g ) gas phase W. T. Numerical Recipes; Cambridge University Press: New York,
1986.
M ) multiresponse
(9) Boudart, M.; Mears, D.; Vannice, M. A. Ind. Chim. Belg.
m ) mean
1967.
Me ) methylation
(10) Froment, G. F.; Hosten, L. H. Catalytic Kinetics: Model-
Ol ) oligomerization ling. In Catalysis Science and Technology; Anderson, J. R.,
Pr ) protonation Boudart, M., Eds.; McGraw-Hill: New York, 1978.
t ) total (11) Gear, G. W. Numerical Initial Value Problems in Ordinary
Superscripts Differential Equations; Prentice-Hall: Englewood Cliffs, NJ, 1971.
(12) Park, T.-Y.; Froment, G. F. A Hybrid Genetic Algorithm
t ) total for the Estimation of the Parameters in Detailed Kinetic Models.
q ) transition state Comput. Chem. Eng. 1998, 22, S103.
(13) Zhou, J. L.; Tits, A. L.; Lawrence, C. T. A Fortran Code
Literature Cited for Solving Constrained Nonlinear (MinMax) Optimization Prob-
lems, Generating Iterations Satisfying All Inequality and Linear
(1) Jacobs, P. A.; Martens, J. A. Synthesis of High-Silica Constraints. User’s Guide for FFSQP, Version 3.6; Electrical
Aluminosilicate Zeolites. Stud. Surf. Sci. Catal. 1987, 33, 19. Engineering Department, University of Maryland: College Park,
(2) Lox, E.; Coenen, F. R. V.; Froment, G. F. A versatile Bench- MD, 1996.
Scale Unit for Kinetic Studies of Catalytic Reactions. Ind. Eng.
Chem. Res. 1988, 27, 576. Received for review September 21, 2000
(3) Froment, G. F.; Bischoff, K. B. Chemical Reactor Analysis Revised manuscript received January 30, 2001
and Design, 2nd ed.; Wiley: New York, 1990. Accepted January 31, 2001
(4) Hutchings, G. J.; Hunter, R. Hydrocarbon Formation from
Methanol and Dimethyl ether: A Review of the Experimental IE000854S

You might also like