Selective cracking of natural gasoline over HZSM-5 zeolite

F U E L P R O CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8 ) 8 1 9–8 2 7
w w w. e l s e v i e r. c o m / l o c a t e / f u p r o c
Selective cracking of natural gasoline over HZSM-5 zeolite
Marcelo J.B. Souzaa , Fabiano A.N. Fernandesb,⁎, Anne M.G. Pedrosac , Antonio S. Araujoc
a
Universidade Federal de Sergipe, Departamento de Engenharia Química, Cidade Universitária Prof. José Aloísio de Campos,
CEP 49100-000, Aracaju/SE, Brazil
b
Universidade Federal do Ceará, Departamento de Engenharia Química, Campus do Pici, Bloco 709, 60455-760, Fortaleza/CE, Brazil
c
Universidade Federal do Rio Grande do Norte, Departamento de Química, Campus Universitário, 59078-970, Natal/RN, Brazil
AR TIC LE I N FO
ABS TR ACT
Article history:
This work presents a study on the catalytic cracking of natural gasoline (extracted from
Received 29 October 2007
natural gas) over HZSM-5 zeolite. A factorial planning was carried out to evaluate the effect
Received in revised form
of temperature and W/F ratio on the cracking of natural gasoline, analyzing their effects on
13 December 2007
conversion and product distribution using an analysis based on surface response
Accepted 31 December 2007
methodology. The process was optimized focusing on the maximization of the mass
fractions and the production of specific products such as ethene, propene and butanes. The
Keywords:
results have shown that the maximum selectivity and hourly mass production of ethene is
Natural gasoline
obtained at high temperature (450 °C) and low catalyst weight to flow rate ratio (W/F) (7.2 to
Cracking
8.2 gcat h/mol). Maximum selectivity of propene is obtained at 350 °C and 7.0 gcat h/mol, while
Zeolite HZSM-5
the best condition for maximum mass production is found at 421 °C and 5.7 gcat h/mol. The
Optimization
highest mass production of butanes is favored by high temperature (450 °C) and mid range
Neural network
W/F ratios (12.1 gcat h/mol), while the highest selectivity is found at low temperature (350 °C).
© 2008 Elsevier B.V. All rights reserved.
1.
Introduction
Natural gas is a mixture of hydrocarbons such as methane
(main component), ethane, propane, butane and nature gasoline (pentane, hexane and heptane). During the processing of
natural gas, natural gasoline cut is condensed and separated
from the gaseous products. It can be sold as such or can be
cracked selectively to produce other hydrocarbons such as
ethene and propene to the plastic industry or as propane and
butane to LPG (liquefied petroleum gas) production.
Hydrocracking is an important process in the petroleum
refining industry to produce feedstocks. Current hydrocracking processes are based on commercial catalysts that are
effective above 450 °C. The development of new catalysts that
can present high activity under lower temperatures may
increase the operational and economical benefits of hydrocracking processes [1].
Hydrocracking is also the industrial process with the
largest consumption of catalyst and even small achievements
in yield and selectivity will have a very significant economical
and ecological impact. These improvements can be obtained
at many levels, from the design of new and improved catalysts
to the optimization of reaction conditions [2]. In this work a
new zeolite-based catalyst was developed and the best
operating condition for the reaction was investigated.
Zeolites are used in catalytic and separation technologies,
especially, in processing of hydrocarbons; therefore, their
adsorptive properties are considered important. Adsorption of
n-pentane, n-hexane and n-heptane on MFI-type molecular
sieves (HZSM-5 zeolites) has attracted attention of many
researchers because of the complex adsorption profiles
observed in this system [3–8]. The adsorption isotherms of
n-pentane and n-hexane on MFI show a step at the loading
corresponding to four molecules per unit cell [3–5]. The
adsorption of n-hexane observed on MFI molecular sieves
follows a two-step adsorption pathway due to two different
adsorption sites in the micropore system, fact supported by
satisfactory fitting of adsorption profiles with functions
⁎ Corresponding author. Tel.: +55 85 33669611; fax: +55 85 33669610.
E-mail addresses: [email protected] (M.J.B. Souza), [email protected] (F.A.N. Fernandes).
0378-3820/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.fuproc.2007.12.014
820
F U E L P RO CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8) 8 19 –8 2 7
Fig. 1 – Scheme of the catalytic reactor. Where: G1 = nitrogen; G2 = helium; V1, V2, = 4 way valves; CF1, CF2, CF3 = flux valves;
B = saturator; P = 10 way pneumatic valve; CT1, CT2 = temperature controllers; T1, T2 = thermocouples; S = vent; C = chromatogram;
CG = gas chromatograph; F = oven e R = reactor with catalyst.
derived from the dual-site Langmuir adsorption model. It is
assumed that the molecules are adsorbed either in the straight
or in the zigzag channels, since they have smaller diameter
than the intersections, ensuring stronger interaction of the
adsorbed molecules with the zeolite lattice [8,9]. The stronger
interaction of adsorbed n-hexane and also of n-pentane with
the zeolite lattice can be an advantage for cracking reactions,
reason for which this zeolite was selected as a catalyst the
cracking of natural gasoline.
Catalytic cracking involves a very complex reaction scheme
operating on the solid catalyst. The complexity of real feeds
and the number of reactions involved requires a large volume
of calculations in addition to the problem of the number
of unknown kinetic parameters that requires estimation.
To overcome the difficulties in dealing with such detailed
models, the optimization of the reaction condition was carried
out using stacked neural networks.
2.
Materials and methods
2.1.
Catalyst Preparation
2.2.
Natural gasoline hydrocracking reaction
The experiments were carried out in a catalytic fixed-bed reactor
operating at continuous flow rate and atmospheric pressure. A
reactor made of quartz with 0.15 m in length and 0.015 m in diameter
was used. A scheme of the reaction apparatus is shown in Fig. 1.
Prior to the reaction, the catalyst (HZSM-5 zeolite) was
activated in situ at 450 oC for 2 hours under a flow of dry nitrogen
(20 mL/min). The fine powder catalyst sample (0.4 g) was loaded
into the micro-flow reactor and a thermocouple was placed on the
center of the catalyst bed to monitor the reaction temperature.
Fine powder catalyst was used to avoid any mass transfer effects.
The reaction was performed diluting C5+ vapor (natural
gasoline) with nitrogen under constant flow rate. Liquid natural
gasoline was fed into the saturator and was totally gasified
previous to reaction to ensure constant composition of the feed
stream. The C5+ vapor to nitrogen ratio was maintained at 1.6 in
all runs. An experimental planning was designed to study the
effects of temperature and catalyst weight to flow rate ratio (W/F)
over the cracking selectivity and the conversion. Temperatures
between 350 and 450 °C and W/F ranging from 5 to 17 gcat h/mol
were applied in the experiments. The operating conditions used
in the experimental planning are presented in Table 1. The
composition of the natural gasoline used in the experiments is
Table 1 – Experimental planning
The NaZSM-5 Zeolite was prepared by hydrothermal crystallization [10] in a steel autoclave at 423 K by 7 days, under
autogenous pressure. The gel of synthesis was prepared from a
mixture containing sodium hydroxide (Merck), aluminum sulphate (Inlab), silica gel (Riedel de Häen AG), sulfuric acid (Merck)
and tetrapropylamoniun bromide (Jansen Chimica), to direct the
structure of the ZSM-5 zeolite [11,12]. The gel molar composition
was 48.8 SiO2. 1.0 Al2O3: 14.3 Na2O: 2.4 (TPA)2O: 180 H2O. The
synthesis product was filtered, washed with distilled water and
dried for 10 h at 393 K. The sample was calcined at 773 K in
nitrogen flow by 15 h, and in dry air by 10 h, at the same
temperature. The HZSM-5 was obtained by ion exchange of a
NaZSM-5 zeolite, with 0.6 M ammonium chloride solution, and
subsequently calcined [12]. The HZSM-5 zeolite was characterized
by atomic absorption (Varian AA-175), X-ray diffraction (Rigaku)
and Scanning Electron Microscopy (SEM, Zeiss).
Run
1
2
3
4
5
6
7
8
9
10
11
12
Experimental condition
Temperature (°C)
W/F (gcat h/mol)
350
350
350
350
400
400
400
400
450
450
450
450
16.56
9.74
6.90
5.18
16.56
9.74
6.90
5.18
16.56
9.74
6.90
5.18
F U E L P R O CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8 ) 8 1 9–8 2 7
Table 2 – Chemical composition of natural gasoline,
provided by Petrobras-UPGN Guamaré (components
mass fractions)
n-Hexane
n-Pentane
2-methyl-pentane
2-methyl-butane
Cyclopentane
Cyclobutane
0.124
0.404
0.094
0.305
0.028
0.036
shown in Table 2 and were provided by Petrobras (UPGN Guamaré,
RN, Brazil).
Under the given conditions, C5+ underwent cracking to C2, C3
and C4 paraffins and C2 and C3 olefins. The reaction products
were analyzed by an online gas chromatograph with mass
spectrum (Shimadzu model QP 5000) using a Chromopack CPAl2O3/KCl capillary column (50 m long, 0.32 m in diameter). The
chromatograph was operated under the following conditions:
column temperature of 100 °C, interface temperature of 200 °C and
injector temperature of 200 °C, using helium as mobile phase. The
conversion rate of the feedstock (natural gasoline) was calculated
by the summation of the formation rate of the products.
2.3.
Optimization procedure
The optimization procedure used to obtain the best operating
condition to enhance the productivity and selectivity of some
product derived from natural gasoline was based on the use of
neural networks. Neural networks have attracted great interest as
predictive models, as well as for pattern recognition. Neural
networks have the ability of learning the behavior of the process
and the relationships between variables, without needing a model
of the phenomenological laws that rule the system. The success in
obtaining a reliable and robust network depends strongly on the
choice of process variables involved, as well as the available sets of
821
data and the domain used for training purposes [13–15]. The
optimization procedure is summarized in Figs. 2 and 3. Fig. 2
shows schematically the procedure to select the neural networks
and Fig. 3 shows the procedure used to find the best operating
conditions with the trained set of neural networks.
In this work, the back propagation algorithm was used, as it is
the most extensively adopted algorithm for neural networks and
performs well. The available data were split in two sets. One set
was used to train the network and the other to test its prediction
capability. The activation sigmoid function used in the neural
network is given by Eq. (1). A random selected bias was used, and
weights were updated by a Hessian approach.
y¼
1
P
1 þ expð xÞ
ð1Þ
The amount of data available from the experiments is scarce to
train a unique neural network. The neural network (NN) would be
trained but the associated error would be high and the neural
network would not be reliable to be used in an optimization
procedure. In general, a neural network requires a great amount of
data (usually more than 100 data points) to generate reliable
predictions, especially if the process being evaluated is complex.
To optimize the cracking process herein, a stacked neural network
(SNN) was used because they provide better predictions when few
data points are available to train the neural network. Furthermore,
a stacked network which combines a number of NNs can improve
overall representation, accuracy and robustness [13,14].
The procedure used herein consisted in training 5 neural
networks with different topologies, and only part of the data
available (9 different operating conditions) was presented to each
NN. The prediction of the SNN was given by the mean value of the
prediction of each individual NN of the stacked neural network.
Since each NN behave differently in different regions of the
training range (or space), the combination of the results from two
or more NNs can be more accurate, since a bad result from one NN
can be compensated by good results from other NNs [13,15].
Fig. 2 – Flow chart of the neural network selection.
822
F U E L P RO CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8) 8 19 –8 2 7
Table 4 – Cracking selectivity and product distribution of
HZSM-5 catalyst at 400 oC
Catalyst to flow rate ratio (W/F) [gcat h/mol]
5.18
6.90
9.74
16.56
FC5+ [mol/h]
Conversion
0.0728
35.41
0.0579
41.55
0.0410
44.22
0.0241
45.11
Products [%]
Ethane
Ethene
Propane
Propene
iso-Butane
n-Butane
3.80
8.87
44.59
15.27
13.38
14.16
2.84
6.01
47.89
11.70
15.51
16.03
2.90
4.79
49.16
8.85
17.28
17.07
7.40
7.12
52.80
7.90
12.81
11.95
language was developed to maximize the production of these
main components, following the scheme presented in Fig. 3. The
maximization was carried out using the quasi-Newton method with
gradient calculated by finite difference. The estimates for the
operating conditions were presented to the neural network which
returned the values of the cracking products molar fractions at each
step of the iteration.
Fig. 3 – Flow chart of the optimization procedure.
3.
The overall output of the stacked neural network is a weighted
sum of the individual NNs outputs:
Results and discussion
Where, e is the relative error of the neural network prediction
[%].
The optimization was carried out searching for the best
operating conditions that would allow the highest selectivity and
the highest production of ethane, propene and butane (objective
functions) because these products have the highest commercial
value among the cracking products. A program in FORTRAN
The characterization of the HZSM-5 zeolite by the hydrothermal method showed that the obtained catalyst had a welldefine crystalline structure, presenting a XRD diffractogram
characteristic of MFI (Mobil Five) [16]. The chemical composition of the calcined material was: (Na0.32H4.84Al5.16Si90.84)O192
presenting a SiO2/Al2O3 ratio of 17.6. Scanning Electron
Microscopy (SEM) showed that the HZSM-5 presented orthorhombic crystallites of ca. 2 μm.
The results obtained with the experimental planning showed
that the main products of the cracking of natural gasoline over
HZSM-5 are propane and butane (iso- and n-butane). Tables 3–5
show the cracking activity and selectivity of the studied catalyst
under different reaction conditions. During reaction the HZSM-5
catalyst did not present reduction of activity or formation of
coke. After the reaction no noticeable changes were observed on
the catalyst structure.
The results indicate that conversion was directly affected by
temperature and slightly affected by the catalyst weight to flow
rate ratio (W/F) which only becomes important at high
Table 3 – Cracking selectivity and product distribution of
HZSM-5 catalyst at 350 oC
Table 5 – Cracking selectivity and product distribution of
HZSM-5 catalyst at 450 °C
Catalyst to flow rate ratio (W/F) [gcat h/mol]
Catalyst to flow rate ratio (W/F) [gcat h/mol]
Y¼
X
wi yi
ð2Þ
Where, Y is the stacked NN predictor, yi is the ith NN predictor
and wi is the stacking weight for the ith neural network.
After training, each neural network was validated using the
remaining 3 experimental data points that were not used in the
training stage. The neural networks that provided predictions
with less than 3% of error were chosen to compose the stacked
neural network. Deviations were calculated by Eq. (3).
e¼
j
j
ðexperimental dataÞ ðNN predictionÞ
100½k
ðexperimental dataÞ
ð3Þ
5.18
6.90
9.74
16.56
FC5+ [mol/h]
Conversion
0.0728
10.24
0.0579
11.10
0.0410
11.23
0.0241
12.81
Products [%]
Ethane
Ethene
Propane
Propene
iso-Butane
n-Butane
0.0
4.19
45.87
9.91
18.93
21.09
0.0
2.64
38.99
19.39
19.01
19.85
0.0
2.51
43.72
5.62
21.51
26.63
0.0
2.05
42.16
7.29
21.46
26.99
5.18
6.90
9.74
16.56
FC5+ [mol/h]
Conversion
0.0728
74.38
0.0579
62.22
0.0410
54.43
0.0241
48.11
Products [%]
Ethane
Ethene
Propane
Propene
iso-Butane
n-Butane
7.06
7.92
28.92
14.77
20.34
20.99
12.67
10.30
33.05
11.90
11.59
20.47
10.15
8.74
32.49
12.33
17.80
18.47
10.09
6.55
46.98
7.20
14.03
15.12
F U E L P R O CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8 ) 8 1 9–8 2 7
823
of butanes and decreasing the amount of ethene and propene in
the final product. A simple model proposed by Kotrel et al. [20]
provides a plausible explanation that is consistent with the
results observed herein. Olefins formed in the primary cracking
of C5 may adsorb in the zeolite forming carbonium ions which
can go through oligomerization (reactions 9 and 10) and
bimolecular cracking (reactions 11 and 12).
C2¼ þ HZ→C2HZ
Fig. 4 – Temperature and W/F effect on reaction conversion.
C2¼ þ C2HZ→C4HZ
ð10Þ
C5 þ C2HZ→C4 þ C3HZ
ð11Þ
C3 þ C4HZ→C4 þ C3HZ
ð12Þ
As proposed by Tran et al. [21] the carbonium ion can also
go through a hydride transfer reaction producing paraffins:
C4HZ→C4 þ HZ
temperatures (above 400 °C) (Fig. 4). Conversions up to 75% were
obtained at high temperature (450 °C) and high W/F (20.0 gcat h/
mol).
As expected for a selective cracking reaction, low temperatures favor the production of hydrocarbons with higher
molecular weight as iso-butane and n-butane, while low
molecular weight hydrocarbons, such as ethane, are favored
by higher temperatures. At high temperatures part of the butane
molecules produced by cracking of pentane and hexane begins
to be consumed forming propane or ethane. This results show
that the cracking reaction is selective and the relative reactivity
of the hydrocarbons are a function of the number of carbons of
the hydrocarbon chain, where longer chains crack easier.
According to proposed mechanisms for monomolecular
cracking [17] the reaction proceeds via the formation of a
transition state that resembles a surface coordinated carbonium
ion. Cleavage of the bonds associated with n-pentane and 2methyl-butane atoms (main natural gasoline components) would
lead to the formation of primary cracking reaction products:
C5→H2 þ C5¼
ð4Þ
C5→CH4 þ C4¼
ð5Þ
C5→C2 þ C3¼
ð6Þ
C5→C2¼ þ C3
ð7Þ
ð9Þ
ð13Þ
The production of ethane is observed only at temperatures
higher than 400 °C (Fig. 5). As the mass fraction of ethane
increases, a proportional decrease in butane mass fraction is
observed, showing that ethane formation is caused by direct
cracking of iso- and n-butane, which undergo cracking only at
elevated temperatures. The production of propane showed a
complex behavior, being significantly affected by the W/F
ratio, especially at high temperatures, where the mass fraction
of propane increases to 58% at 430 °C and W/F = 20 gcat h/mol.
At the same temperature but at a lower W/F (5 gcat h/mol) the
mass fraction of propane decreases to less than 30% (Fig. 6).
Ethene and propene, two important feedstock to the plastic
industry, presented low selectivity and only up to 10% of ethene
and 15% of propene were produced (mass fraction) (Figs. 7 and
8). As propene and ethene are primary cracking products of C5
(n-pentane and 2-methyl-butane) the low concentration of
these two components in the product indicate further oligomerization of these components into heavier hydrocarbons and
their participation in bimolecular cracking. The highest production of ethene and propene was obtained operating at high
temperatures and low W/F. High temperatures directly affect
The chromatographic results had not detected any pentene
nor butene, ruling out reaction 4 and 5, which would be in
accordance with the fact that the cracking reaction occurs
preferably at inner central C–C bonds [18,19]. This primary
cracking of pentane and 2-methyl-butane explains the large
amounts of propane but do not explain the formation of
large quantities of iso- and n-butane. The cracking of hexane
would lead to the formation of butanes through reaction 8
but the small mass fraction of n-hexane and 2-methyl-pentane
(21.8%) in the natural gasoline also cannot explain the formation
of up to 50% of iso- and n-butane in the final product.
C6→C4 þ C2¼
ð8Þ
Therefore, secondary reactions, especially further reactions
of primary cracking products must be leading to the formation
Fig. 5 – Production of ethane as a function of temperature
and W/F.
824
F U E L P RO CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8) 8 19 –8 2 7
Fig. 6 – Production of propane as a function of temperature
and W/F.
the cracking rate of butane and other heavy products into low
weight hydrocarbons such as ethane, ethene, propane and
propene increasing their mass fraction in the reaction product
Fig. 9 – Production of iso-butane as a function of temperature
and W/F.
mixture. At high temperature the cracking reaction rate is
higher than the rate of oligomerization and bimolecular
cracking and an increase in the amount of ethene and propene
is observed. Low W/F ratio reduces the apparent rate of
oligomerization because the concentration of complexed olefin
(CnHZ) does not reach the steady state [20], thus more ethene
and propene will be observed in the product distribution.
The production of iso- and n-butane is favored by low
temperature (350 °C) and mild W/F ratio (12 gcat h/mol) as
shown in Figs. 9 and 10. The mass fraction of these two
components in the final product is up to 50% and their formation
is linked to the selective cracking of n-hexane and 2-methylpentane which present high relative reactivity, and due to
oligomerization of low molecular weight olefins and bimolecular
cracking, which has a positive net rate at low temperatures.
4.
Fig. 7 – Production of ethene as a function of temperature
and W/F.
Fig. 8 – Production of propene as a function of temperature
and W/F.
Optimization
Few kinetic models are available in the literature describing
natural gasoline cracking (heptane, hexane and pentane) over
Fig. 10 – Production of n-butane as a function of temperature
and W/F.
Table 6 – Preditions of two neural networks used to compose the stacked neural network and predictions of the stacked neural network
Run
Experimental Data
Cnv
C2
C2=
C3
Predicted Data
Prediction Error [%]
iC4
nC4
Cnv
C2
C2=
C3
C3=
iC4
nC4
Cnv
C2
C2=
C3
C3=
iC4
nC4
7.29
5.62
19.39
9.91
7.90
8.85
11.70
15.27
7.20
12.33
11.90
14.77
21.46
21.51
19.01
18.93
12.81
17.28
15.51
13.38
14.03
17.80
11.59
20.34
26.99
26.63
19.85
21.09
11.95
17.07
16.03
14.16
15.12
18.47
20.47
20.99
11.80
11.23
11.10
10.24
45.11
44.22
40.82
35.41
74.38
63.83
54.43
48.11
0.03
0.00
0.00
0.00
7.40
2.90
2.79
3.80
10.09
9.51
12.67
7.06
2.08
2.51
2.64
4.19
7.12
4.79
6.12
8.87
6.55
8.76
10.30
7.92
41.99
73.72
38.99
45.87
52.80
49.16
48.02
44.59
46.98
32.10
33.05
28.92
7.17
5.62
19.39
9.91
7.90
8.85
11.73
15.27
7.20
12.23
11.90
14.77
21.53
21.51
19.01
18.93
12.81
17.28
15.66
13.38
14.03
17.78
11.59
20.34
26.61
26.63
19.85
21.09
11.95
17.07
15.81
14.16
15.12
18.81
20.47
20.99
7.88
0.02
0.04
0.01
0.00
0.00
1.77
0.00
0.00
2.59
0.00
0.00
–
0.00
0.00
0.00
0.00
0.00
1.90
0.03
0.00
6.33
0.01
0.00
1.56
0.00
0.00
0.00
0.00
0.00
1.81
0.00
0.00
0.23
0.00
0.00
0.40
0.00
0.00
0.00
0.00
0.00
0.27
0.00
0.00
1.21
0.00
0.00
1.62
0.00
0.00
0.00
0.00
0.00
0.28
0.00
0.00
1.21
0.00
0.00
0.34
0.00
0.00
0.00
0.00
0.00
0.96
0.00
0.00
0.08
0.00
0.00
1.42
0.00
0.00
0.00
0.00
0.00
1.38
0.00
0.00
1.81
0.00
0.00
Neural Network trained without runs 2, 10 and 12
1
12.81
0.00
2.05
42.16
7.29
2
11.23
0.00
2.51
43.72
5.62
3
11.10
0.00
2.64
38.99
19.39
4
10.24
0.00
4.19
45.87
9.91
5
45.11
7.40
7.12
52.80
7.90
6
44.22
2.90
4.79
49.16
8.85
7
41.55
2.84
6.01
47.89
11.70
8
35.41
3.80
8.87
44.59
15.27
9
74.38
10.09
6.55
46.98
7.20
10
62.22
10.15
8.74
32.49
12.33
11
54.43
12.67
10.30
33.05
11.90
12
48.11
7.06
7.92
28.92
14.77
21.46
21.51
19.01
18.93
12.81
17.28
15.51
13.38
14.03
17.80
11.59
20.34
26.99
26.63
19.85
21.09
11.95
17.07
16.03
14.16
15.12
18.47
20.47
20.99
12.80
11.52
11.11
10.23
45.11
44.22
41.56
35.40
74.37
61.69
54.43
49.10
0.00
0.00
0.00
0.00
7.40
2.90
2.84
3.80
10.09
10.25
12.67
7.24
2.03
2.65
2.64
4.20
7.12
4.78
6.03
8.85
6.55
8.70
10.31
7.99
41.15
44.01
39.99
45.88
52.83
49.14
47.95
44.55
46.97
32.30
33.07
30.89
7.28
6.12
19.39
9.92
7.91
8.34
11.72
15.25
7.19
12.06
11.90
15.01
21.45
21.72
19.01
18.93
12.81
17.27
15.53
13.36
14.03
17.33
11.60
20.41
26.97
26.90
19.85
21.10
11.97
17.05
16.08
14.11
15.11
18.52
20.48
21.35
0.05
2.58
0.13
0.12
0.00
0.01
0.02
0.03
0.01
0.86
0.01
2.06
0.00
0.00
0.00
0.00
0.01
0.03
0.00
0.08
0.02
0.96
0.02
2.61
0.78
5.46
0.08
0.14
0.04
0.21
0.40
0.28
0.03
0.42
0.08
0.85
0.03
0.68
0.01
0.02
0.05
0.05
0.13
0.09
0.02
0.60
0.07
6.80
0.18
8.84
0.02
0.07
0.09
0.14
0.21
0.12
0.13
2.20
0.03
1.66
0.07
0.98
0.01
0.03
0.03
0.03
0.14
0.16
0.03
2.64
0.09
0.34
0.07
1.01
0.02
0.04
0.18
0.12
0.31
0.32
0.05
0.25
0.02
1.70
Stacked neural network
1
12.81
0.00
2
11.23
0.00
3
11.10
0.00
4
10.24
0.00
5
45.11
7.40
6
44.22
2.90
7
41.55
2.84
8
35.41
3.80
9
74.38
10.09
10
62.22
10.15
11
54.43
12.67
12
48.11
7.06
21.46
21.51
19.01
18.93
12.81
17.28
15.51
13.38
14.03
17.80
11.59
20.34
26.99
26.63
19.85
21.09
11.95
17.07
16.03
14.16
15.12
18.47
20.47
20.99
12.66
11.15
10.97
10.29
45.14
44.06
41.62
35.36
73.04
62.43
54.36
48.32
0.05
0.00
0.00
0.00
7.35
2.94
2.81
3.88
10.13
9.81
12.41
7.05
2.05
2.53
2.63
4.20
7.16
4.76
6.04
8.74
6.46
8.41
10.19
7.86
42.16
43.86
38.94
45.88
52.56
49.27
47.75
44.57
47.14
32.47
33.05
29.32
7.26
5.84
18.78
9.89
7.89
8.84
11.77
15.22
7.16
12.28
11.90
14.83
21.47
21.55
19.04
18.93
12.80
17.31
15.54
13.35
14.07
17.64
11.69
20.33
26.91
26.81
19.80
21.10
12.09
16.99
16.05
14.13
15.09
18.70
20.48
21.07
1.16
0.69
1.19
0.48
0.07
0.37
0.16
0.16
1.80
0.34
0.12
0.44
–
0.00
0.00
0.00
0.74
1.24
1.22
2.07
0.37
2.37
2.06
0.11
0.27
0.68
0.45
0.15
0.52
0.52
0.55
1.42
1.37
2.77
1.10
0.81
0.01
0.33
0.12
0.03
0.46
0.21
0.28
0.05
0.34
0.07
0.00
1.37
0.35
2.88
2.12
0.20
0.18
0.15
0.61
0.32
0.50
0.41
0.02
0.41
0.04
0.18
0.14
0.01
0.08
0.19
0.22
0.20
0.30
0.91
0.84
0.07
0.30
0.66
0.27
0.04
1.16
0.46
0.11
0.20
0.20
1.28
0.06
0.38
Neural network trained without runs 1, 7 and 10
1
12.81
0.00
2.05
42.16
2
11.23
0.00
2.51
43.72
3
11.10
0.00
2.64
38.99
4
10.24
0.00
4.19
45.87
5
45.11
7.40
7.12
52.80
6
44.22
2.90
4.79
49.16
7
41.55
2.84
6.01
47.89
8
35.41
3.80
8.87
44.59
9
74.38
10.09
6.55
46.98
10
62.22
10.15
8.74
32.49
11
54.43
12.67
10.30
33.05
12
48.11
7.06
7.92
28.92
2.05
2.51
2.64
4.19
7.12
4.79
6.01
8.87
6.55
8.74
10.30
7.92
42.16
43.72
38.99
45.87
52.80
49.16
47.89
44.59
46.98
32.49
33.05
28.92
7.29
5.62
19.39
9.91
7.90
8.85
11.70
15.27
7.20
12.33
11.90
14.77
825
⁎ Cnv is the conversion [%]; C2 is ethane; C2= is ethane; C3 is propane; C3= is propene; iC4 is i-butane and nC4 is n-butane.
F U E L P R O CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8 ) 8 1 9–8 2 7
C3=
826
F U E L P RO CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8) 8 19 –8 2 7
Table 7 – Optimum operating conditions for the
production of ethene, propene and C4
Production of
Temperature [°C]
W/F [gcat h/mol]
Optimization aiming highest selectivity
Ethene
450
Propene
352
n-butane
350
iso-butane
350
350
C4 (n- + iso-butane)
7.2
7.0
13.6
13.5
13.5
Optimization aiming highest mass
Ethene
Propene
n-butane
iso-butane
C4 (n- + iso-butane)
8.2
5.7
12.0
12.3
12.1
production
450
421
450
450
450
zeolites [1,2]. Estimation of the kinetic parameter for the
available models presents a high degree of difficulty and
might not fully represent cracking over other zeolites because
the kinetic mechanism or the limiting step may be different.
Thus, as we currently do not have a reliable kinetic model, the
optimization of the reaction studied herein was carried out
using stacked neural networks (SNN).
The experimental planning resulted in a set of 12 data
points. From this set, three random data points were removed
from the data set presented to the neural networks (NNs) and
were used as quality control for the stacked neural network
predictions, being presented to the NNs only at the testing
stage, never on the training stage. The remaining data points
were split into 45 groups, each containing 9 data points. Each
group was used to train a NN with 2 hidden layer and 20
neurons in the first hidden layer and 25 neurons in the second
hidden layer. After training, 11 NNs outputted prediction with
less than 3% of error and were selected to compose the stacked
neural network. Table 6 shows the predictions of two neural
networks selected to comprise the stacked neural network and
the predictions of the stacked neural network. The mean
prediction errors were 0.58, 1.40, 0.97, 0.27, 0.85, 0.27, 0.43%
respectively for conversion, ethane, ethane, propane, propene,
iso-butane and n-butane molar fractions. The mean errors as
well as the individual errors were all below 3% making the
stacked neural network suitable to be used in the optimization
of the process.
The optimization was carried out searching for the best
operating conditions that would allow the highest selectivity
and the highest production of ethane, propene and butanes
which have the highest commercial value among the cracking
products, so two objectives functions were employed. One to
search for the optimum operating condition aiming higher
selectivity of a individual or group of components, and one to
search for the optimum operating condition aiming higher
mass production of a specific or group of components. The
optimization problem solved to search for the highest selectivity of a component was given by:
Box 1
Find: T and W/F
Maximize: ϕi, where i = ethene, propene and butane
Within ranges of operating conditions:
350 ≤ T ≤ 450 °C
5.0 ≤ W/F ≤ 16.0 gcat h/mol
Where ϕi is the selectivity of component i calculated by
dividing the flow rate of component i by the total flow rate
of all cracking products.
Table 8 – Product distribution (mass fractions) for the reaction operating at the optimum conditions to obtain highest
selectivity
Production of
Ethene
Propene
iso-butane
n-Butane
C4
Conversion
Mass fraction
(%)
Ethane
Ethene
Propane
Propene
iso-butane
n-butane
55.0
11.7
11.0
11.0
11.0
0.125
0.000
0.000
0.000
0.000
0.102
0.027
0.022
0.022
0.022
0.334
0.397
0.427
0.427
0.427
0.116
0.190
0.061
0.061
0.061
0.117
0.189
0.217
0.217
0.217
0.206
0.197
0.273
0.273
0.273
Data in bold emphasis refers to the highest selectivity of the product.
Table 9 – Product distribution (mass fractions) for the reaction operating at the optimum conditions to obtain highest
production (mol/h)
Production of
Ethene
Propene
iso-butane
n-Butane
C4
Conversion
Mass fraction
(%)
Ethane
Ethene
Propane
Propene
iso-butane
n-butane
58.1
50.3
68.0
68.5
68.2
0.121
0.087
0.084
0.084
0.084
0.100
0.094
0.066
0.065
0.066
0.331
0.355
0.375
0.382
0.377
0.121
0.159
0.096
0.094
0.095
0.127
0.138
0.193
0.191
0.193
0.200
0.167
0.186
0.184
0.185
Data in bold emphasis refers to the highest yield of the product.
F U E L P R O CE SS I NG T EC H NOL O G Y 8 9 (2 0 0 8 ) 8 1 9–8 2 7
The results for the optimization of component selectivity
are presented in Tables 7 and 8.
When the highest mass production of a component is to be
optimized, the conversion of the feedstock and its flow rate
needs to be taken into account and the problem to be solved is
given by:
Box 2
Find: T and W/F
Maximize: Fi, where i = ethene, propene and butane
Within ranges of operating conditions:
350 ≤ T ≤ 450 °C
5.0 ≤ W/F ≤ 16.0 gcat h/mol
Where Fi is the flow rate of component i
The results for the optimization of component selectivity
are presented in Tables 7 and 9.
As shown and discussed in Fig. 1, the temperature is the
factor that mostly affects the cracking reaction rate and thus
conversion of the feedstock. As such its effect on conversion
also influences mass production and as a result most
components will have its highest hourly mass production at
higher temperatures but with a lower selectivity (Table 9).
5.
Conclusion
The results showed that the main products of the cracking of
natural gasoline over HZMS-5 zeolite are propane, iso-butane
and n-butane. The production of ethane occurs only at high
temperatures, as the production of ethene. The surface
analysis method allows concluding that the cracking reactor
over HZMS-5 zeolite is selective and that the relative reactivity
of cracking is a function of the carbon number of the
hydrocarbon molecule.
Among the main commercial products, the results have
shown that the maximum selectivity and hourly mass
production of ethene is obtained at high temperature
(450 °C) and low W/F ratio (7.2 to 8.2 gcat h/mol). Maximum
selectivity of propene is obtained at 350 °C and 7.0 gcat h/mol,
while when its mass production is to be optimized the best
condition is found at 421 °C and 5.7 gcat h/mol. The highest
mass production of butanes is favored by high temperature
(450 °C) and mid range W/F ratios (12.1 gcat h/mol), while the
highest selectivity is found at low temperature (350 °C).
Since current hydrocracking processes are based on
commercial catalysts that are effective above 500 °C, the
developed catalyst have presented good potential to be used
commercially because it presents good activity and selectivity
under lower temperatures (350 to 450 °C) for the production of
ethene, propene and butane.
Acknowledgement
The authors would like to thank CNPq-CTPetro e ANP for the
financial support of this study and the award of scholarship.
827
REFERENCES
[1] K.I. Alhumaizi, V.M. Akhmedov, S.M. Al-Zahrani, S.H.
Al-Khowaiter, Low temperature hydrocracking of n-heptane
over Ni-supported catalysts: study of global kinetics, Appl.
Catal. 219 (2001) 131–140.
[2] R.R. Pinto, P. Borges, M.A.N.D.A. Lemos, L. Lemos, F.R. Ribeiro,
Kinetic modelling of the catalytic cracking of n-hexane and
n-heptane over zeolite catalyst, Appl. Catal. 272 (2004) 23–28.
[3] R.E. Richards, L.V.C. Rees, Sorption and packing of n-alkane
molecules in ZSM-5, Langmuir 3 (1987) 335–340.
[4] F. Eder, J.A. Lercher, Alkane sorption in molecular sieves: the
contribution of ordering, intermolecular interactions, and
sorption on Bronsted acid sites, Zeolites 18 (1997) 75–81.
[5] W. Zhu, F. Kapteijn, B. van der Linden, J.A. Moulijn,
Equilibrium adsorption of linear and branched C6 alkanes on
silicalite-1 studied by the tapered element oscillating
microbalance, Phys. Chem. Chem. Phys. 3 (2001) 1755–1761.
[6] W. Makowski, D. Majda, Temperature-programmed
equilibrated desorption of n-hexane as tool for
characterization of the microporous structure of zeolites,
Thermochim. Acta 412 (2004) 131–137.
[7] B. Millot, A. Methivier, H. Jobic, Adsorption of n-alkanes on
silicalite crystals. A temperature-programmed desorption
study, J. Phys. Chem. 102 (1998) 3210–3215.
[8] B. Millot, A. Methivier, H. Jobic, I. Clemencon, B. Rebours,
Adsorption of branched alkanes in silicalite-1: a
temperature-programmed-equilibration study, Langmuir
15 (1999) 2534–2539.
[9] W. Makowski, D. Majda, Equilibrated thermodesorption
studies of adsorption of n-hexane and n-heptane on zeolites
Y, ZSM-5 and ZSM-11, Appl. Surf. Sci. 252 (2005) 707–715.
[10] R.J. Argauer, G.R. Landolt, Conversion with ZSM-5 family of
crystalline aluminosilicate zeolites. US Patent RE 29857 (1978).
[11] M.J.B. Souza, A.S. Araujo, V.J. Fernandes, A.O.S. Silva,
Beneficiamento da fração C5+ do pólo de Guamaré a partir de
reações de craqueamento catalítico sobre zeólitas ácidas, Rev.
Tec. En. Petr. Gás 1 (2002) 74–78.
[12] A.S. Araujo, M.J.B. Souza, Conversão catalítica de metanol em
hidrocarbonetos usando a zeólita HZSM-5 modificada por
nióbio, An. Assoc. Bras. Quim. 45 (1996) 40–46.
[13] J. Zhang, Inferential estimation of polymer quality using
bootstrap aggregated neural networks, Neural Netw. 12 (1999)
927–938.
[14] J. Zhang, E.B. Martin, A.J. Morris, C. Kiparissides, Inferential
estimation of polymer quality using stacked neural networks,
Comp. Chem. Eng. 21 (1997) S1025–S1030.
[15] F.A.N. Fernandes, L.M.F. Lona, Neural network applications in
polymerization processes, Braz. J. Chem. Eng. 22 (2005) 401–418.
[16] R.J. Argauer, G.R. Landolt, Crystalline zeolite ZSM-5 and
method of preparing the same. US Patent 3,702,886 (1972).
[17] S.M. Babitz, B.A. Williams, J.T. Miller, R.Q. Snurr, W.O. Haag,
H.H. Kung, Monomolecular cracking of n-hexane on Y, MOR,
and ZSM-5 zeolites, Appl. Catal. 179 (1999) 71–86.
[18] L. Pellegrini, S. Locatelli, S. Raselle, S. Bonomi, V. Calemma,
Modeling of Fischer–Tropsch products hydrocracking, Chem.
Eng. Sci. 59 (2004) 4781–4787.
[19] S.T. Sie, M.M.G. Senden, H.M.H. van Wechem, Conversion of
natural gas to transportation fuels via the shell middle distillate
synthesis process (SMDS), Catal. Today 8 (1991) 371–394.
[20] S. Kotrel, M.P. Rosynek, J.H. Lunsford, Origin of first-order
kinetics during the bimolecular craking of n-hexane over
H-ZSM-5 and H-β zeolites, J. Catal. 191 (2000) 55–61.
[21] M.T. Tran, N.S. Gnep, G. Szabo, M. Guisnet, Comparative study
of the transformation of n-butane, n-hexane and n-heptane
over H-MOR zeolites with various Si/Al ratios, Appl. Catal. 170
(1998) 49–58.