See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/245419221

Kinetic Study on a Commercial Amorphous

Hydrocracking Catalyst by Weighted Lumping
Strategy

ARTICLE in INTERNATIONAL JOURNAL OF CHEMICAL REACTOR ENGINEERING JANUARY 2010

Impact Factor: 0.79 DOI: 10.2202/1542-6580.2193

CITATIONS DOWNLOADS VIEWS

8 24 85

3 AUTHORS:

Sepehr Sadighi Arshad Ahmad

Research Institute of Petroleum Industry (RIPI) Universiti Teknologi Malaysia
104 PUBLICATIONS 100 CITATIONS 76 PUBLICATIONS 127 CITATIONS

SEE PROFILE SEE PROFILE

Akbar Irandoukht
Research Institute of Petroleum Industry (RIPI)
14 PUBLICATIONS 32 CITATIONS

SEE PROFILE

Available from: Sepehr Sadighi

Retrieved on: 01 August 2015
 I NTERNATIONAL J OURNAL OF C HEMICAL
R EACTOR E NGINEERING
Volume 8 2010 Article A60

Kinetic Study on a Commercial Amorphous

Hydrocracking Catalyst by Weighted
Lumping Strategy
Sepehr Sadighi Arshad Ahmad
Akbar Irandoukht

Universiti Teknologi Malaysia, sadighi sepehr@yahoo.com

Universiti Teknologi Malaysia, arshad@fkkksa.utm.my

Research Institute of Petroleum Industry, irandoukhta@ripi.ir
ISSN 1542-6580
Copyright 2010
c The Berkeley Electronic Press. All rights reserved.
 Kinetic Study on a Commercial Amorphous
Hydrocracking Catalyst by Weighted Lumping
Strategy
Sepehr Sadighi, Arshad Ahmad, and Akbar Irandoukht

Abstract
Hydrocracking is an important upgrading process in the petroleum refinery,
and it is generally used to process feedstocks ranging from vacuum gas oil (VGO)
to vacuum residue. In this work, hydrocracking of VGO using a dual functional
amorphous catalyst was carried out at a pilot scale unit under the following re-
action conditions: liquid hourly space velocity (LHSV) from 1 to 1.5 hr1 and
reaction temperatures of 360-440 C at the constant pressure and hydrogen to oil,
156 bar and 1780 Nm3 /m3 , respectively. The effluent of the reactor was character-
ized to dry gas, naphtha, kerosene, diesel and unconverted VGO or residue. The
pilot tests demonstrated that performing experiments beyond the temperature, rec-
ommended by catalyst vendor, lead the process to unstable hydrocracking. To de-
scribe the yield of hydrocracking products a five-lump discrete lumping approach
with ten reactions was proposed. At first, the kinetic model contained twenty
kinetic constants which were estimated by using the conventional objective func-
tion. The estimated parameters showed that the tendency of the catalyst to convert
VGO to gas and naphtha was negligible whilst rate constants for hydrocracking of
VGO to middle distillates were considerably high which was compatible with the
nature of amorphous hydrocracking catalysts. After evaluating the magnitude of
reaction rates and eliminating the ignorable constants, the network was reduced
to six reactions in which only nine parameters were needed. The predictions in-
dicated that the latter network could fit the yield of products more acceptable as
if the average absolute deviation between experimental and calculated yields was
descended from 16.25% to 12.6%. Then, to have a better prediction, a weighted
objective function was used in which weight factors were calculated by a proposed
weighted least square expression. The results confirmed that this approach could
reduce average absolute deviation of model to 10.75%, and it created a fairly even
distribution of deviation between hydrocracking products.

KEYWORDS: hydrocracking, vacuum gas oil, lump kinetic model, amorphous

catalyst, weight factors
Sadighi et al.: Weighted Lumped Kinetic Estimation 1

1. Introduction
The increasing demand for light petroleum feedstocks like gasoline and diesel
provides an incentive for upgrading of residual streams in a refinery to more
valuable products. The heavy feedstock can be converted to lighter ones using
thermal and/or catalytic processing in the absence or presence of hydrogen
pressure (Gary and Handwerk, 2001; Meyers, 1986). Hydrocracking is one of the
most important processes in modern refineries to produce low sulfur diesel. The
versatility and flexibility of the process makes it economically attractive to
convert different types of feedstocks into various products including gas, LPG,
naphtha, kerosene and diesel, leading to its widespread applications. Typical of
industrial processes, optimal operation is required to guarantee profitability and
such a task necessitates the use of process models. These models are used to
predict the product yields and qualities, and are useful for sensitivity analysis, so
that the effect of operating parameters such as reactor temperature, pressure,
space velocity, and others on yields and qualities of products can be understood.
The models can also be used for process optimization and control, design of new
plants and selection of suitable hydrocracking catalysts (Valavarasu et al, 2005).
However, the complexity of a hydrocracking feed makes it extremely difficult to
characterize and describe its kinetic at a molecular level (Ancheyta et al, 2005).
One approach to simplify the problem is to consider a partitioning of the
components into a few equivalent classes called lumps or lumping technique, and
then to assume each class as an independent entity (Astarita and Sandler, 1991).
As it is mentioned in literatures, developing simple kinetic models (e.g., power-
law model) for complex catalytic reactions is a common approach that can give
basic information for catalyst screening, reactor design and optimization
(Ancheyta et al., 2005). In this field, many works were reported in literatures in
which hydrocracking models with three-lump (Yui and Sanford, 1989; Callejas
and Martinez, 1999; Aoyagi et al., 2003), four-lump (Aboul-Gheit, 1989;
Valavarasu et al., 2005), five-lump (Ayasse et al., 1997; Ancheyta et al., 1999;
Almeida and Guirardello, 2005; Singh et al., 2005) and six-lump (Sadighi et al.,
2010a; Sadighi et al., 2010b) partitions were developed.
The first aim of this work is studying the behavior of a commercial
amorphous catalyst within and beyond conventional operating conditions and
developing an optimized five-lump kinetic model for hydrocracking of vacuum
gas oil for a fixed-bed pilot scale reactor. The next, to estimate more accurate
kinetic parameters in this wide range of LHSVs and temperatures, a new weighted
least square expression is defined in which the needed weight factors are
estimated from pilot results. It is confirmed that the suggested technique shows
less average absolute deviation than the conventional method of fitting used in
previous investigations and the deviation is distributed evenly between all lumps.

Published by The Berkeley Electronic Press, 2010
2 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

2. Experimental
2.1 Device
The experiment was conducted in the pilot catalyst system (Geomechanique BL-
2) which its process flow diagram is presented in Fig. 1. This plant that belongs
to catalysis research center (CRC) of research institute of petroleum industry
(RIPI), can tolerate temperature and pressure up to 500C and 300 bar,
respectively. The reactor temperature was maintained at the desired level by a
jacket heater around the reactor, which was supposed to provide an isothermal
temperature along the active reactor section. All experiments were supported and
conducted by CRC (Kianpoor, 1999). The physical properties of the feed and
product samples were determined according to the ASTM standard procedures.

Figure 1. Simplified process flow diagram of hydrocracking set up. FC, flow controller; PC,
pressure controller; TIC, temperature indicator and controller; TI, temperature indicator;
HP, high pressure separator; LP, low pressure sampling vessel;
FI, flow indicator; GC, gas chromatograph

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 3

2.2 Feed and catalyst

Hydrocracking tests were performed with the fresh vacuum gas oil (VGO) which
was taken directly from vacuum tower of a 100,000 barrel per day refinery. The
VGO physical properties are shown in Table 1. To do the experiments, the reactor
bed was loaded with 33 g (50 cm3) catalyst which its characterization are
tabulated in Table 2. The under study catalyst was a commercial amorphous dual
functional type which performs hydrocracking and hydrotreating reactions
simultaneously. About 25mm above the catalyst bed and also 25mm below the
catalyst bed were filled with inert particles only.

Table 1. Properties of fresh vacuum gas oil

Density at 15C g /cm3 0.908

Sulfur wt% 1.73
Total Nitrogen wt% 0.09
Aromatics wt% 27.3
Non-Aromatics Wt% 72.6
Distillation analysis ( ASTM D1160 )
IBP C 332
10% C 402
30% C 427
50% C 447
70% C 464
90% C 490
FBP C 515

Table 2. Catalyst specifications of hydrocracking process

Shape - Spherical
Mesh - 10-20
3
Bulk Density kg/m 654
Density(solid) kg/m3 2500
Surface Area m2/g 270
App. porosity cm3/g 1.13
Chemical Properties
Base Silica Alumina
SiO2 wt % 55.39
Al2O3 wt % 9.27
WO3 wt % 24.53
NiO wt % 3.55
Cao wt % 0.46
Fe2O3 wt % Trace

Published by The Berkeley Electronic Press, 2010
4 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

2.3 Test conditions

Hycrocracking was carried out under the following process conditions:
1. H2/HC = 1780 Nm3/m3
2. LHSV = 1,1.2 and 1.5 hr-1
3. Temperature = 360C, 380C, 400C,420C and 440C
4. Pressure = 156 bar
The pressure and H2/HC were selected according to the recommendation
of catalyst vendor. For the selected catalyst, in a commercial reactor, the normal
working LHSV is normally between 1 and 1.2 hr-1; but for the hydrocracking of
VGO, the recommended LHSV is between 1 and 1.5 hr-1 (HSU and Robinson,
2006); The SOR and EOR temperatures for this type of amorphous catalyst are
about 390C and 420C, respectively; therefore the experiments were performed
in the wider range. Since the strategy for modeling was lumping of feed and
products, hydrogen was neglected in the mass balance.

3. Hydrocracking kinetic model

This work considers five lumps, i.e., unconverted VGO or residue, diesel,
kerosene, naphtha and gas to match main products of pilot plant reactor. Fig. 2
illustrates reaction pathways associated with the strategy. Note that if all pathways
of reactions were considered, the model would include twenty kinetic parameters.
These parameters should be estimated by using experimental data, and it was too
laborious. Some judgments are normally welcomed to reduce the model
complexity, without sacrifying the accuracy. Upon close scrutiny of the system
under consideration, the model could be reduced which required less kinetic
parameters. The reduction of parameters can be done according to the order of
magnitude of rate constants in comparison to the highest one in the average
process temperature, as also discussed later.

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 5

Figure 2. The complete 5-lump kinetic model

3.1 Kinetic model

Mathematical models for a trickle-bed catalytic reactor can be complex due to

many microscopic and macroscopic effects occurring inside the reactor, including
flow patterns of both phases, size and shape of a catalyst particles, wetting of the
catalyst pores with liquid phase, pressure drop, intraparticle gradients, thermal
effects and, of course, kinetics on the catalyst surface (Sertic-Bionda, 2005). It is
therefore more practical to reduce the complexity of the reactor, focusing only on
momentous process variables. This suggests a development of simpler models
that incorporates less number of parameters.
For each reaction, a kinetic expression ( R ) is formulated as the function of mass
concentration ( C ) and kinetic parameters ( k0 , E ).
The following assumptions have been made in the development of the
present model:
1- Hydrocracking is a first order hydrocracking reaction and since hydrogen
is present in excess, the rate of hydrocracking can be taken to be
independent of the hydrogen concentration (Mohanty, 1991).
2- The pilot reactor operates under isothermal conditions.
3- A plug flow pattern exists in the trickle bed reactor.
4- Hydrogen feed is pure.
5- The petroleum feed and the products are in the liquid phase in the reactor.
6- The pilot unit is in steady state operation.
7- Catalyst activity does not change with time; therefore simulation is only
valid for start of run conditions

Published by The Berkeley Electronic Press, 2010
6 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Based on these assumptions, the kinetic constants of proposed model are

as follows:

EFj
Vacuum gas oil or Feed ( F ): k Fj = k0 Fj exp( ) (1)
RT

Note j in Eq. 1 represents diesel ( D ), kerosene ( K ), naphtha ( N ) and gas ( G )

lumps.

EDj '
Diesel ( D ): k Dj ' = k0 Dj ' exp( ) (2)
RT

where j ' in Eq. 2 represents kerosene ( K ), naphtha ( N )and gas ( G ) lumps.

EKj ''
Kerosene ( K ): k Kj '' = k0 Kj '' exp( ) (3)
RT

where j ' ' in Eq. 3 are naphtha ( N ) and gas ( G ) lumps.

E NG
Naphtha ( N ): k NG = k0 NG exp( ) (4)
RT

Here, T and R are the absolute value of bed temperature and ideal gas constant,
respectively.
The reaction rates ( R ) can be formulated as the following:

G
Vacuum gas oil reaction ( RF ): RF = k Fj C F (5)
j=D

G
Diesel ( RD ): RD = k FD C F k Dj 'C D (6)
j '= K

G
Kerosene ( RK ): RK = k FK C F + k DK C D k
j ''= N
Kj '' CK (7)

Naphtha ( RN ): RN = k FN C F + k DN C D + k KN C K k NG C N (8)

Gas ( RG ): RG = k FG C F + k DG C D + k KG C K + k NG C N (9)

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 7

3.2 Mass balance

The mentioned kinetic model is assumed to be valid for an isothermal plug flow
reactor. The mass balance equations and kinetic model should be solved
simultaneously to evaluate the product yields ( Y j ). The former is as the following:

1 (C j )
R j = 0 (10)
V
In the Eq. 10, a positive sign indicates products and a negative sign indicates
reactant (feed or VGO).
( )
=0 (11)
V
G
Fm = C j (12)
j=F

C j .
Yj = (13)
Fm

Here , , V , Fm , C j , , Y j are the catalyst effectiveness factor, catalyst

volume fraction, bed volume, stream mass flow rate, mass concentration of
products, volume flow rate through reactor and yield of product The effectiveness
factor for spherical catalyst in trickle bed regime is considered 0.8 (Mills et al.,
1999).
The only unknown variable in the above equations is density of the stream
inside the reactor which can be calculated as follows:

1 G Yj
= (14)
0 j=F j

Here j is the density of lumps (Table 3). It should be noted that the
density evaluated by Eq. 14 ( 0 ) is standard density which can be used to
determine density at the reactor condition ( ) by the Standing-Katz correlation
(Ahmed, 1989). In deviating from the SI system we give the equation with the
original units:

( p, T ) = 0 + p T (15)

Published by The Berkeley Electronic Press, 2010
8 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

where 0 represents the density at standard conditions in lb/ft3. The pressure

dependence can be evaluated by:

p
p = [0.167 + 16.181 10 0.0425.0 ].[ ]
1000
(16)
p 2
0.01 [0.299 + 263 10 0.0603. 0 ].[ ]
1000

where p is the pressure in psia. Since the density drops with ascending
temperature, a temperature correction with the temperature T in 0R is needed:

T = [0.0133 + 152.4 ( 0 + p ) 2.45 ].[T 520]

0.764.( 0 + p )
(17)
[8.1 10 6 0.0622 10 ].[T 520]2

3.3 Parameter estimation

For parameter estimation, the sum of squared errors, SQE1 , as given below, was
minimized:
Nt G
meas pred 2
SQE1 = w j .(Y jn Y jn ) (18)
n =1 j = F

meas pred
In Eq. 18, N t , Y jn and Y jn are the numbers of test runs, the measured product
yield and the predicted one by model, respectively. Where w j in Eq. 18 is the
weight factor of lumps which is considered one in most of the kinetic estimation
works, called conventional strategy in this paper. But, to balance the yield of
products with higher values, like diesel, with the lower ones, like naphtha, this
parameter roles an important effect. In this work, at first the required weight
factors in Eq. 18 were estimated by the following objective function:

F Nt Nt
SQE2 = ( w j Ynj wref Yref ) 2 Subject to w j ,ref > 0 (19)
j =G n=1 n =1

The subscript ref in Eq. 19 refers to the lump with the lowest measured
yield. The reaction and mass balance expressions according to Eqs. 1 to 17 were
coded and solved simultaneously using the Aspen Custom Modeler (ACM)
programming environment (AspenTech) to evaluate the product yields ( Ynj ).
Then, to estimate kinetic parameters, Eq. 18 was minimized by sequential

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 9

application of the NL2Sol and Nelder-mead algorithms, which are both found in
the Aspen Custom Modeler software. NL2Sol algorithm is a variation on
Newton's method in which part of the Hessian matrix is computed exactly and
part is approximated by a secant (quasi-Newton) updating method. To promote
convergence from a poor initial point, a trust-region is used along with a choice of
model Hessian. Hence, the approximate region is found with NL2Sol; then to fine
tune the parameters, Nelder-Mead simplex method is used.
The estimated parameters and the adequacy of the regressions were
checked with an analysis of variance (ANOVA) using R-adjusted and the Fischer
test with 99% probability (Clarke and Kempson, 1997; Montgomery, 2001). The
ANOVA of all strategies discussed in the next section were studied and their
adequacy was tested using the static Fischer test ( F ) with a 1% critical level.
Additionally, to compare the simulated and measured product values, absolute
average deviations (AAD) (Marafi et al., 2001) were calculated by the following
equation:

meas pred 2
Nt G (Y jn Y jn )

n=1 j = F meas 2
Y jn
AAD% = 100 % (20)
Nt

4. Results and discussion

4.1 Lump characteristics
The main products of the process were gas (G) including dry gases and LPG,
naphtha (N), kerosene (K), diesel (D) and unconverted VGO (F). The average
density and boiling point range of these products are presented in Table 3. Based
on the resulted data, it was found that the distillation property of the residue or
unconverted oil is close to that of VGO feedstock. Therefore considering them as
one lump was not an irrational assumption.

Published by The Berkeley Electronic Press, 2010
10 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Table 3. Average properties of hydrocracking product

Sp.gr IBP-FBP
@15.5 C (C)
Gas 0.35 40-
Naphtha 0.744 40-141
Kerosene 0.796 141-260
Diesel 0.823 260-370
Residue 0.908 370+

4.2 Pilot test results

Experimental data of product yields for gas, naphtha, kerosene, diesel and
unconverted VGO at 380C, 400C, and 420C versus space velocity (LHSV)
were carried out. Performing mass balance around the reactor showed that the
error for all experiments was less than 1 % which was normalized on all lumps. It
was essential to say that at 440oC, approximately 20oC upper than the maximum
working temperature of catalyst, producing of all lumps especially gas, naphtha
and kerosene in all LHSVs were completely unstable as if reaching to steady state
condition to take sample was impossible. It was supposed that at this point, the
depositing of coke on the catalyst was accelerated; therefore, it affected the yield
of hydrocracking. Moreover, the yield of products especially gas and naphtha at
3600C in all LHSVs were very low and approximately zero. Therefore, the
mentioned phenomena can justify avoiding from these operating conditions in the
refinery units.

Figure 3. Gas yield vs. Space velocity where H2 /Oil=1780 Nm3/Sm3, Pressure=156 bar
() T=380C, () T=400C and () T=420C

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 11

Figure 4. Naphtha yield vs. Space velocity where H2 /Oil=1780 Nm3/Sm3, Pressure=156 bar
() T=380C, () T=400C and () T=420C

Figure 5. Kerosene yield vs. space velocity where H2 /Oil=1780 Nm3/Sm3, Pressure=156 bar
() T=380C, () T=400C and () T=420C

Published by The Berkeley Electronic Press, 2010
12 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Figure 6. Diesel yield vs. space velocity where H2 /Oil=1780 Nm3/Sm3

Pressure=156 bar, () T=380C, () T=400C and ( )T=420C

Figure 7. Unconverted yield vs. space velocity where H2 /Oil=1780 Nm3/Sm3,

Pressure=156 bar, () T=380C, () T=400C and () T=420C

Figs. 3, 4 and 5 illustrated that the yield of gas, naphtha and kerosene was
decreased by increasing the LHSV which was in agreement with the role of the
residence time-the smaller the LHSV, the better hydrocracking. As it was
expected, for gas, naphtha and kerosene the temperature strengthen the
hydrocracking paths so that the yield of them was promoted.
As it can be understood from Fig. 7, for the unconverted VGO or residue,
these reasons acted reversely; therefore temperature decreased the yield of
residue, and the LHSV increased it.
In Fig. 6, for diesel, from 380oC to 400oC, the sensitivity of diesels yield
to the LHSV was the same as lighter products; but it was ascended with LHSV at

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 13

420oC, like residue. It was supposed that at this temperature, the catalyst changed
its behavior and it had tendency to convert VGO to diesel more than others.
Additionally, as it can be found from this figure, at a fixed LHSV, by increasing
the temperature, the diesel production was decreased sharply. It was supposed that
at higher temperature, the diesel lump was cracked to lighter products. Therefore,
the more production of lighter products such as naphtha and kerosene at higher
temperatures at a fixed LHSV was explainable.
Moreover, from Fig. 3, it can be found that it was a sharp rising in
production of gas from 400oC to 420oC. Therefore, for the understudy catalyst,
hydrocracking can be divided to two imaginary regimes which were the low
severity below 400oC and the high severity at 420oC. The similar phenomenon
was reported in the previous work (Botchway et al., 2004) for hydrotreating of
Bitumen-derived heavy gas oil from Athabasca over a commercial NiMo/Al2O3
catalyst in a trickle-bed reactor.
As it can be seen from presented figures, the yield of diesel and kerosene,
named middle distillates, was considerably higher than naphtha, a reasonable
phenomenon for the amorphous catalyst which was reported in literatures
(Scherzer and Gruia, 1996; Ali et al., 2002).
Additionally, Figs. 3 to 7 demonstrate that for the LHSV between 1 and
-1
1.5 hr , the yield of products was fairly stable and had slight slope in all
temperatures which can be a good reason for working at this region in commercial
plant.

4.3 Modeling results

The twenty kinetic parameters for the assumed model (Fig. 2) were estimated by
using measured pilot data in which all weights of Eq. 18 were assumed one. Table
4 shows the estimated values of apparent activation energies and frequency
factors. In this table, the rate constants for all reactions were evaluated in the
average operating temperature (4000C). After applying conventional estimation
method, to compare the predicted and measured yield values, average absolute
deviation ( AAD% ) was calculated which its average for yield of all lumps was
about 16.25%. Also, the average absolute deviations for all lumps were calculated
and presented in Table 5 under the name of complete network.
Then ANOVA analysis for the complete model was carried out to
establish the adequacy of the model. This proved that the complete model could
predict the yield of the products with an R 2 (adjusted) of ~0.989, which was
acceptable. Moreover, the adequacy of the fitted model was tested using the static
Fischer test ( F ) with a 1% critical level. The value of F = 99.31 was higher
than Fcritical (19,25,0.01) ~ 2.75 , confirming that the model well fitted the observed
yields.

Published by The Berkeley Electronic Press, 2010
14 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Table 4. Kinetic parameters for the complete network

Frequency Factor Activation Energy Rate constant
(m3.hr-1.m3 cat-1) (kcal/mol) (m3.hr-1.m3 cat-1)
k0FD 2.14E+07 EFD 20.22 kFD 5.79
k0Fk 4.00E+12 EFk 37.23 kFk 3.22
k0FN 0.001 EFN 27.70 kFN 1.39E-12
k0FG 0.089 EFG 0.01 kFG 0.09
k0DK 2.46E+08 EDK 33.55 kDk 3.12E-03
k0DN 1.11E+10 EDN 28.40 kDN 6.61
k0DG 0.0716 EDG 33.17 kDG 1.20E-12
k0KN 4.67 EKN 5.19E-07 kKN 4.67
k0KG 0.0012 EKG 0.42 kKG 1.23E-03
k0NG 28.60 ENG 2.14E-03 kNG 28.55

Table 5. The AAD % for the different strategies for parameter estimation

Complete Reduced Reduced-weighted

Lump
network network network
Gas 18.36 15.96 13.88
Naphtha 25.61 15.92 10.17
Kerosene 14.49 8.40 6.68
Diesel 8.27 9.90 11.41
Un.VGO 14.65 12.84 11.64
Ave. 16.25 12.60 10.75

Data in Table 4 revealed that rate constants for conversion of feed to

middle distillates (kFD, kFK) were higher than the feed to lighter products whilst
the catalyst has low tendency to convert feed to naphtha (kFN). Furthermore, it can
be found that the rate constant for hydrocracking naphtha to gas was
tremendously high which was reasonable because of the simplicity of cracking
light chains of naphtha in comparison to heavier and more stable ones of diesel
and kerosene. It means that middle distillates had a negligible role in producing
gas in the understudy hydrocracking process. The mentioned phenomena can
justify the lowest yield for naphtha produced by amorphous hydrocracking
catalysts which was consistent with the literatures (Scherzer and Gruia, 1996; Ali
et al., 2002) that reported this type of catalyst has tendency to produce higher
amounts of middle distillates and less amounts of naphtha cuts. Moreover the
higher rate of feed to diesel than feed to kerosene can be good reason for yielding
more diesel in this process.

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 15

Finally from Table 4, it can be concluded that the path of converting feed
to naphtha, diesel to gas, diesel to kerosene and kerosene to gas can be ignored.
Also the activation energy for conversion of feed to gas, kerosene to naphtha and
naphtha to gas, because of their low values, can be considered roughly zero
during tuning procedure.
After eliminating the mentioned least possible paths and parameters, and
rerun the estimation program, AAD% of reduced model was descended to 12.6%
which was more satisfying in comparison to complete network. The AAD% of all
lumps resulted by reduced kinetic network are presented in Table 5 under the
name of the reduced network. The ANOVA proved that the reduced model can
predict product yields with an R 2 (adjusted) of ~0.997. In addition, the value of
F = 381.3 was higher than F (8,36,0.01) = 3.04 , demonstrating that the reduced
model well fitted the observed yields.
Therefore, it is evident from above discussions that data prediction by
using the reduced model is more accurate when it is compared with the complete
one. This also confirms the same situation for the previous works (Singh et al.,
2005; Sadighi et al., 2010a; Sadighi et al., 2010b).
Table 6 shows the estimated values of apparent activation energies and
frequency factors for the reduced kinetic network. Therefore, a reduced model
with six reaction paths and nine kinetic parameters is resulted, depicted in Fig. 8.
In this network, nine kinetic parameters should be tuned by using forty-five
observations so that it had acceptable degree of freedom for parameter estimation.

Table 6. Kinetic parameters for the reduced network

Frequency Factor Activation Energy Rate constant
(m3.hr-1.m3 cat-1) (kcal/mol) (m3.hr-1.m3 cat-1)
k0FD 1.87E+08 EFD 23.02 kFD 6.25
k0Fk 1.33E+14 EFk 42.22 kFk 2.57
k0FN - EFN - kFN -
k0FG 0.32 EFG - kFG 0.32
k0DK - EDK - kDk -
k0DN 3.20E+11 EDN 32.73 kDN 7.47
k0DG - EDG - kDG -
k0KN 2 EKN - kKN 1.57
k0KG - EKG - kKG -
k0NG 25.98 ENG - kNG 25.98

Published by The Berkeley Electronic Press, 2010
16 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Figure 8. The reduced 5-lump kinetic model

The next try at parameter estimation was done by using the factors
presented in Table 7, calculated from minimizing the Eq. 19 which should be
substituted in Eq. 18 as weight factors. The estimated kinetic constants for this
approach were presented in Table 8. After predicting the yields by applying them,
the average AAD% was descended to 10.75% exhibiting more accuracy than
conventional strategies that they did not use weight factors. Moreover, the
adequacy of the fitted model was tested using static Fischer ( F ) with 1% critical
level. The value of F = 369.96 is higher than F (8,36,0.01) = 3.04 , demonstrated
that the regressed model fitted well the observed values. Also, this approach could
predict product yields with an R 2 (adjusted) of ~0.997.
This deviation was considered acceptable for a kinetic model in this wide
range of operating conditions. To compare the predicted and measured yield
values by the reduced-weighted approach, average absolute deviation for all
lumps is presented in Table 5 with the name of the reduced-weighted. It can be
understood that the AAD % of the hydrocracking lumps resulted by weighted
least square expression, especially for naphtha, was less than the complete or
reduced networks. Only the exception was the diesel that its deviation was 3%
more than the complete network, still less than the average value.

Table 7. Estimated factors for weighted estimation

Lump Weight factor

Gas 0.81
Naphtha 1.43
Kerosene 0.62
Diesel 0.41
Unconverted 0.28

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 17

Table 8. Kinetic parameters for the weighted network

Frequency Factor Activation Energy Rate constant
(m3.hr-1.m3 cat-1) (kcal/mol) (m3.hr-1.m3 cat-1)
k0FD 1.02E+09 EFD 25.30 kFD 6.16
k0Fk 6.93E+14 EFk 44.46 kFk 2.50
k0FN - EFN - kFN -
k0FG 0.48 EFG - kFG 0.48
k0DK - EDK - kDk -
k0DN 2.21E+13 EDN 38.55 kDN 6.63
k0DG - EDG - kDG -
k0KN 1.35 EKN - kKN 1.35
k0KG - EKG - kKG -
k0NG 24.39 ENG - kNG 24.39

In Fig. 9, the AAD % of all lumps and the average values predicted by
complete, reduced and reduced-weighted model are demonstrated. In this figure,
it was clear that naphtha and gas lumps had the highest deviation when they were
predicted by complete model. This problem was abated after predicting the yields
by the reduced approach. But, it was observed an even deviation between lumps
when reduced-weighted strategy was applied.
In Fig. 10 the parity plot for the measured data and the predicted by the
reduced-weighted model is presented, certifying the acceptable agreement
between them. This figure indicates the least agreement for the diesel, supposed to
be created by the changes in the process severity above 4000C, discussed before.
Because one of the purposes for performing the pilot tests is developing
appropriate kinetic model to scale up the reactor and to evaluate the catalyst
performance in commercial reactors, the predictions resulted from complete or
reduced network can be illusive because of high deviations for the strategic
products, especially naphtha. These approaches can be really hazardous when the
main products have the fewer yields in comparison to the others. Even an
inaccurate value for gas or unconverted residue, which is not precious, can create
design problems for the equipment like condenser of the distillation tower and the
recycle pump, respectively. Therefore, the presented strategy can be beneficial to
have more reliable results because of its higher accuracy and more even deviation
between lumps. It is thought that applying the proposed weighted least-square in
this research can be efficient for the other similar processes.

Published by The Berkeley Electronic Press, 2010
18 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Figure 9. AAD% of lumps compared with average value predicted by the different strategies

Figure 10. Comparison between the measured yields and the predicted yields

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 19

5. Conclusions
In this research, nine experiments in different LHSVs and bed temperatures,
performed at a pilot scale vacuum gas oil hydrocracker with a commercial dual
functional amorphous catalyst, were reported. The experiments showed that the
temperatures upper and lower than operating conditions recommended by catalyst
vendor, created abnormal behaviors for the catalyst as if it can lead the process
into unstable hydrocracking; therefore, they should be avoided in commercial
reactors. After that, a 5-lump kinetic base model was proposed to predict the yield
of products. For under study amorphous catalyst, it was confirmed that the
magnitude for the rate constant of feed to middle distillate was higher than feed to
naphtha cut. The rate constants of feed to diesel, feed to kerosene, feed to gas and
naphtha to gas at the average bed temperature (4000C) were 6.16, 2.5, 0.48 and
24.39, respectively whilst the one for feed to naphtha was negligible. Therefore,
they can be the main reasons for higher productivity of middle distillate by
amorphous catalyst in hydrocracking reactor which was also reported in the
previous researches. Moreover, at the operating conditions, naphtha had the
highest tendency to be converted to gas created the lowest yield for naphtha
between all lumps.
Consequently, reaction paths with low rate constants (feed to naphtha,
diesel to gas and kerosene as well as feed to gas) and ignorable activation energies
(feed to gas kerosene to naphtha and naphtha to gas) were eliminated, leaded to
have a reduced kinetic network with six reaction paths and nine kinetic
parameters. After re-estimating parameters of the model, it was concluded that
this network can predict the yield of products with better accuracy, reducing
average absolute deviation from 16.28% to 12.6%.
Finally, to estimate the kinetic constants, a weighted function was applied
which could decrease the average absolute deviation of the prediction to 10.75%.
Additionally, applying the proposed weight factors in this research not only
decreased the absolute average deviation of the prediction, but also created an
even distribution of error between lumps which was supposed to be efficient for
the accurate evaluation of the catalyst by the model.

Published by The Berkeley Electronic Press, 2010
20 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

6. Nomenclature
6.a Notations

AAD Absolute Average Deviation, %

C Mass concentration, kg/m3
D Diesel

E Apparent activation energy, kcal/mol

Fm Stream mass flow rate, kg/hr
G Gas
k Reaction rate constant, m3.hr-1.m3 cat-1
k0 Frequency factor, m3.hr-1.m3 cat-1
K Kerosene
LHSV Liquid Hourly Space Velocity, hr-1
N Naphtha
Nt Number of experiments
p Pressure, bar or psia
R Ideal gas constant, 1.987 kcal.kmol-1.K-1
Rj Reaction rate of lump j, kg.hr-1.m3 cat-1
T Temperature, K or R
V Volume of catalyst, m3

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 21

6.b Greek letters

Catalyst void fraction

p Deviation of density from standard density at operating pressure,
kg/m3
T Deviation of density from standard density at operating
temperature, kg/m3
Effectiveness factor
Volume flow rate, m3/hr
Density, kg/m3
0 Standard density, kg/m3

6.c Subscripts

Dj ' Diesel to lighter lumps

DG Diesel to gas
DK Diesel to kerosene
DN Diesel to naphtha
FD Feed to naphtha
FG Feed to gas
Fj Feed to lighter lumps
FK Feed to kerosene
FN Feed to naphtha
FD Feed to diesel
i Feed, diesel, kerosene, naphtha and gas lumps
j Diesel, kerosene, naphtha and gas lumps
j' kerosene, naphtha and gas lumps
j' ' naphtha and gas lumps
n Number of experiment
Kj ' ' Kerosene to lighter lumps
ref Reference lump

Published by The Berkeley Electronic Press, 2010
22 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

References
Aboul-Gheit K., "Hydrocracking of Vacuum Gas Oil (VGO) for Fuels
Production-Reaction Kinetics", Erdol Erdgas Kohle, 1989, 105 (7/8), 319-320.

Ahmed, T., "Hyrocarbon Phase Behavior", Gulf Publishing Co. (Houston),

(1989).

Ali M. A., Tatsumi T. and Masuda T., "Development of heavy oil hydrocracking
catalyst using amorphous silica-alumina and zeolites as catalyst supports",
Applied Catalysis A: General, 2002, 233, 77-90.

Almeida R. M. and Guirardello R., "Hydroconversion kinetics of Marlim vacuum

residue", Catalysis Today, 2005, 109, 104-111.

Ancheyta J., Lopez F. and Aguilar E., "5- Lump kinetic model for gas oil catalytic
cracking", Applied Catalysis A: General, 1999, 177, 227-235.

Ancheyta J., Sanchez S. and Rodriguez M. A., "Kinetic modeling of

hydrocracking of heavy oil fractions: A review", Catalysis Today, 2005, 109, 76-
92.

Aoyagi K., McCaffrey W. C. and Gray M. R., "Kinetics of Hydrocracking and

Hydrotreating of Coker and Oilsands Gas Oils", Petroleum Science Technology,
2003, 21(5), 997-1015.

Astarita G. and Sandler, S. I., "Kinetics and thermodynamics lumping of

multicomponent mixtures", 1991, Elsevier: Amsterdam, 111-129.

Ayasse A.R., Nagaishi H. and Chan E.W., "Lumped kinetics of hydrocracking of

bitumen", Fuel, 1997, 76(11), 1025-1033.

Botchwey C., Dalai A.K., and Adjaye J., "Kinetics of Bitumen-Derived Gas Oil
Upgrading Using a Commercial NiMo/Al2O3 Catalyst", Can. J. Chem. Eng.,
2004, 82, 478-487.

Callejas M. A. and Martinez M. T., "Hydrocracking of a Maya Residue Kineic

and Product Yield Distributions", Ind. Eng. Chem. Res., 1999, 38, 98-105.

http://www.bepress.com/ijcre/vol8/A60
Sadighi et al.: Weighted Lumped Kinetic Estimation 23

Clarke G. M. and Kempson R. E., "Introduction to the design and Analysis of

Experiments", 1997, Arnold, London.

Gary J.H. and Handwerk G.E.," Petroleum Refining Technology and Economics",
2001, Marcel Dekker Publication, 4th Edition.

HSU C.S. and Robinson P.R., "Practical Advances in Petroleum Processing",

2006, Vol. I, Springer Publication, 1st Edition.

Kianpoor, P., "Determination of stoichiometric coefficients of hydrocracking

reactions", 1999 Master thesis, Research Institute of Petroleum Industry &
Amirkabir University of Technology.

Marafi A., Kam E. and Stanislaus A., "A kinetic study on non-catalytic reactions
in hydroprocessing Boscan crude oil", Fuel, 2008, 87, 2131-2140.

Mary G., Chaouki J. and Luck F.,"Trickle-Bed Laboratory Reactors for Kinetic
Studies", International J. of Chemical Reactor Engineering, 2009, 7.

Meyers, Robert A., "Handbook of Petroleum Refining Processes", 1986, New

York: McGraw-Hill, Second Edition.

Mills P.L. and Dudukovic M.P., "A Dual-Series Solution for the Effectiveness
Factor of Partially Wetted Catalysts in Trickle-Bed Reactors", 1976, Ind. Eng.
Chem. Fund., 18(2), 139-149.

Mohanty S., Saraf D. N. and Kunzro D., "Modeling of a hydrocracking reactor",

Fuel Processing Technology, 1991, 29, 1-17.

Montgomery D. C., "Design and Analysis of Experiments", 2001, John Wiley &
Sons, New York.

Sadighi S., Arshad A. & Mohaddecy S. R., "6-Lump Kinetic Model for a
Commercial Vacuum Gas Oil Hydrocracker", 2010a, International J. of Chemical
Reactor Engineering, 8.

Published by The Berkeley Electronic Press, 2010
24 International Journal of Chemical Reactor Engineering Vol. 8 [2010], Article A60

Sadighi, S., Arshad, A., & Irandoukht, A., "Modeling a Pilot Fixed-bed
Hydrocracking Reactor via a Kinetic Base and Neuro-Fuzzy Method", 2010b,
Journal of Chemical Engineering Japan, 43 (2), 174-185.

Scherzer J. and Gruia A. J., "Hydrocracking Science and Technology", 1996,

Marcel Dekker, Inc., New York.

Sertic-Bionda K., Gomzi Z. and Saric T., "Testing of Hydrosulfurization process

in small trickle-bed reactor", Chem. Eng. J., 2005, 106, 105-110.

Singh J., Kumar M. M., Saxena A. K. and Kumar S., "Reaction pathways and
product yields in mild thermal cracking of vacuum residues: A multi-lump kinetic
model", Chem. Eng. J., 2005, 108, 239-248.

Valavarasu G., Bhaskar M. and Sairam B., "A Four Lump Kinetic Model for the
Simulation of the Hydrocracking Process", Petroleum Science and Technology,
2005, 23, 1323-1332.

Yui S. M. and Sanford E. C., "Mild hydrocracking of bitumen-drived coker and

hydrocracker heavy gas oils: kinetic product yield and product properties" Ind.
Eng. Chem. Res., 1989, 28, 319-320.

http://www.bepress.com/ijcre/vol8/A60