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

net/publication/225830500

4-Lump Kinetic Model for Vacuum Gas Oil Hydrocracker Involving Hydrogen
Consumption

Article in Korean Journal of Chemical Engineering · July 2010

DOI: 10.1007/s11814-010-0172-0

CITATIONS READS

61 6,208

3 authors:

Sepehr Sadighi Arshad Ahmad

Research Institute of Petroleum Industry (RIPI) Universiti Teknologi Malaysia
184 PUBLICATIONS 643 CITATIONS 214 PUBLICATIONS 3,525 CITATIONS

SEE PROFILE SEE PROFILE

Mehdi Rashidzadeh
Research Institute of Petroleum Industry (RIPI)
77 PUBLICATIONS 836 CITATIONS

SEE PROFILE

All content following this page was uploaded by Arshad Ahmad on 11 October 2014.

The user has requested enhancement of the downloaded file.

Korean J. Chem. Eng., 27(4), 1099-1108 (2010)
DOI: 10.1007/s11814-010-0172-0
RAPID COMMUNICATION

4-Lump kinetic model for vacuum gas oil hydrocracker involving

hydrogen consumption
Sepehr Sadighi*,†, Arshad Ahmad*, and Mehdi Rashidzadeh**

*Universiti Teknologi Malaysia, Faculty of Chemical and Natural Resources Engineering,

81310 UTM Skudai, Johor Bahru, Malaysia
**Catalysis Research Center, Research Institute of Petroleum Industry, P.O. Box 14665-137, Tehran, Iran
(Received 9 September 2009 • accepted 4 November 2009)

Abstract−A 4-lump kinetic model including hydrogen consumption for hydrocracking of vacuum gas oil in a pilot
scale reactor is proposed. The advantage of this work over the previous ones is consideration of hydrogen consumption,
imposed by converting vacuum gas oil to light products, which is implemented in the kinetic model by a quadratic
expression as similar as response surface modeling. This approach considers vacuum gas oil (VGO) and unconverted
oil as one lump whilst others are distillate, naphtha and gas. The pilot reactor bed is divided into hydrotreating and
hydrocracking sections which are loaded with different types of catalysts. The aim of this paper is modeling the hy-
drocracking section, but the effect of hydrotreating is considered on the boundary condition of the hydrocracking part.
The hydrocracking bed is considered as a plug flow reactor and it is modeled by the cellular network approach. Initially,
a kinetic network with twelve coefficients and six paths is considered. But following evaluation using measured data
and order of magnitude analysis, the three route passes and one activation energy coefficient are omitted; thus the num-
ber of coefficients is reduced to five. This approach improves the average absolute deviation of prediction from 7.2%
to 5.92%. Furthermore, the model can predict the hydrogen consumption for hydrocracking with average absolute de-
viation about 8.59% in comparison to those calculated from experimental data.
Key words: Vacuum Gas Oil, Hydrotreating, Hydrocracking, Lump Kinetic Model, Hydrogen Consumption

INTRODUCTION sumed hydrogen are the share of HDT in the first stage which should
be considered during kinetic modeling of hydrocracking reactions
Crude oils contain a large fraction of heavy products for which in the second stage.
only few outlets exist. Indeed, the world demand for light and middle Typical of industrial processes, optimal operation is required to
distillate continually increases, while at the same time, the available guarantee profitability, and such a task necessitates the use of pro-
crude oil becomes heavier [1]. Therefore, upgrading of heavy crude cess models. These models are used to predict the product yields
oil fractions to more useful lighter products is indispensable. Hydro- and qualities, and are useful for sensitivity analysis, so that the effect
cracking is one of the most important processes in a modern re- of operating parameters such as reactor temperature, pressure, space
finery to produce low sulfur diesel. The versatility and flexibility of velocity, as well as others on product yields and qualities can be
the process makes it economically attractive to convert different understood. The models can also be used for process optimization
types of feedstock into various yields including gas, LPG, naphtha, and control, design of new units and selection of suitable hydroc-
kerosene and diesel, leading to its widespread applications. Among racking catalysts [4]. However, the complexity of hydrocracking
all the commercially proven technologies for heavy fraction hydro- feed makes it extremely difficult to characterize and describe its
cracking, those using fixed-bed reactors in series charged with dif- kinetics at a molecular level [5]. One way of simplifying the prob-
ferent functionalities are very favorable. But, the main disadvantage lem is to consider the partition of the species into a few equivalent
of fixed-bed reactors is the loss of catalyst activity over time as a classes, the so-called lumps or lumping technique, and then assume
result of catalyst deactivation which reduces drastically the length each class is an independent entity [6]. This approach is attractive
of run [2]. In the particular case of vacuum gas oils (VGO), a pre- for kinetic modeling of complex mixtures because of its simplicity
vious HDT stage, first stage, for removing nitrogen, sulfur and metal [7].
compounds as well as saturation of polynuclear aromatics (PNA) Mosby et al. [8] reported a model that describes the performance
to preserve catalyst from fast deactivation is required [3]. During of a residue hydrotreater using lump first-order kinetics. The pro-
the HDT process a portion of the hydrogen, dependent on HDS and posed model divides residue into lumps that are “easy” and “hard”
HDN reactions, is consumed and most of the heavy sulfur and ni- crack. This lumping scheme was used by Aboul-Gheit [9] to deter-
trogen compounds are converted to lighter products. Therefore, it mine the kinetic parameters of vacuum gas oil (VGO) hydrocrack-
can be concluded that a part of the desirable products and con- ing, expressing composition in molar concentration. In his four-lump
kinetic model, VGO was converted to gases, gasoline, and middle
To whom correspondence should be addressed.
†
distillates. The model had eight kinetic constants that were esti-
E-mail: [email protected] mated by experiments performed in a fixed-bed plug flow micro
1099
1100 S. Sadighi et al.

reactor. Ayasse et al. [10] fitted experimental product yields from These parameters were estimated from experimental data obtained
catalytic hydrocracking of Athabasca bitumen obtained in a contin- in a fixed-bed down-flow reactor with Maya heavy crude. The lumps
uous-flow mixed reactor. In this model, the data were all obtained at adopted for the study were residue, VGO, distillates, naphtha and
a constant temperature, so the Arrhenius parameters of the rate con- gases. This kinetic model predicted well the hydrocracking process
stants were only valid in a narrow operating range. Moreover, the of Maya heavy oil at moderate pressure and temperatures in a down-
model was stoichiometric base and was significantly model-depen- flow experimental reactor.
dent. Therefore, the model was not recommended for more than Singh et al. [16] also adopted a 5-lump modeling strategy in pre-
five lumps. Cellejas and Martinez [11] studied the kinetics of Maya dicting the yield of mild thermal cracking of vacuum residue. The
residue in a perfectly mixed reactor in continuous operation in the lumping scheme chosen was based on the most value added prod-
presence of a hydrotreating commercial catalyst. They used a first- ucts, i.e., gas, gasoline, light gas oil and vacuum gas oil. In this case,
order kinetic model with a 3-lump configuration which was atmo- the model had seven kinetic parameters.
spheric residue, lights oil and gases. The six kinetic constants in Almeida et al. [17] presented a 5-lump kinetic model for hydro-
the model were estimated by experimental data. The model repre- conversion of Marlim vacuum residue in which by utilizing four-
sented good validation at 375 and 400 oC, but at 415 oC the fits were teen experiments in batch reactor, 26 coefficients were estimated
bad. Another kinetic model for gas oil hydrocracking was pro- for the kinetic model.
posed by Yui and Sandford [12]. In this case, pilot scaled experi- One of the recent works using the lumping method was the dy-
ments were performed in a trickle-bed reactor at various different namic simulation of hydrotreating operation presented by Remesat
operating conditions. Their model was also a 3-lump kinetic model, et al. [18]. This work confirmed the validity of lumping strategy
and similarly, the four kinetic constants were estimated from experi- even for the hydrotreating process. The usage of lumping strategy
mental data. Anchyeta et al. [13] proposed a 5-lump kinetic model is not limited to hydrocracking and some works have been done in
for catalytic cracking of gas oil in which the deactivation of catalyst the similar fields like modeling of fluid catalytic cracking process
was considered as an exponential law with one decay parameter [19], hydroconversion [20] and catalytic [21,22] as well as thermal
depending on the time-on stream. The model had eight kinetic con- cracking [23] of heavy oils which the latter is in the field of petro-
stants, including one for catalyst deactivation which was estimated chemical processes.
from experiments obtained in a microactivity reactor (MAT). Prod- In the present study, a model for a dual bed pilot scale hydroc-
uct yields predicted by this model showed a good agreement with racking unit has been developed. The first section bed of the reactor
experimental data with average deviation less than 2%. Aoyagi et is charged with hydrotreating catalyst to eliminate the amount of
al. [14] studied the kinetics of hydrotreating and hydrocracking of impurities in the VGO to produce low sulfur and nitrogen products
conventional gas oil, coker gas oils and gas oils derived from Atha- and preserve the hydrocracking catalyst loaded in the second bed.
basca bitumen. The model used a first-order expression and 3-lump The model presented in this paper only considers the hydrocrack-
kinetic network to describe the reaction of heavy gas oil. Their ex- ing part, but the effect of hydrotreating section has also been con-
periments were performed in fixed condition (constant temperature, sidered in the boundary conditions of the hydrocracking catalyst.
pressure and LHSV) and the model predicted the results accept- Although there are only four lumps involved in this investigation,
ably. Sanchez et al. [15] proposed a five-lump kinetic model with the main advantage of that over those previously reported in the
10 kinetic parameters for moderate hydrocracking of heavy oils. literature is consideration of the hydrogen consumption in kinetic

Fig. 1. Simplified process flow diagram of hydrocracking set up.

July, 2010
 4-Lump kinetic model for vacuum gas oil hydrocracker involving hydrogen consumption 1101

equations and overall mass balance. The present model according Table 2. Properties of fresh VGO and recycle feed
to 4-lump approach is simplistic, but as it can be concluded later, Property Fresh VGO Recycle feed
the model can predict the yield of hydrocracking products and hy-
[email protected] C o
0.8777 0.8738
drogen consumption reasonably adequately.
Distillation range (vol%)
EXPERIMENTAL ASTM D1160 C
o o
C
IBP 329.7 287.8
1. Pilot Plant Device 10% 390.6 390.7
Hydrotreating (HDT) and Hydrocracking (HDC) of VGO are 30% 423.2 430.1
performed in a high pressure test plant (‘BASF’) (Fig. 1). The reac- 50% 445.6 452.9
tor is designed as a tube with an inside diameter of 16 mm and total 70% 475.1 478.3
length of 2,160 mm which is subdivided into four sections. The first 90% 523.7 517.1
section, having a length of 100 mm, is packed with inert SiC parti- End point 567.1 561.3
cles. This entrance section is used to provide a uniform distribution Nitrogen (ppmwt) 800 200
of gas and liquid. The two following sections with a length of 355 Sulfur (wt%) 1.4 0.03
mm and 865 mm are loaded with 63.5 cm3 hydrotreating and 152 Asphalt & Resin (wt%) <0.1 <0.1
cm3 hydrocracking catalysts, respectively. The final section is also
contained with 50mm of inert. The SiC diameter lies in the 1.5-2.5
Table 3. Average properties of hydrocracking product
mm range.
In all experiments reported in this paper a single charge of both Lump Sp.gr@15 oC IBP-FBP (oC)
catalysts is used. The temperature along the reactor bed is con- Gas 0.350 40−
trolled by use of four thermocouples. Therefore, an isothermal con- Naphtha 0.750 40-160
dition is maintained along the active reactor section. Distillate 0.823 161-370
2. Catalyst
In the present investigation two types of commercial hydrotreat-
ing and hydrocracking (zeolite-based) of VGO are used which are point range of others are presented in Table 3. All properties of the
loaded in separate beds. The usage of zeolite based hydrocracking feed and product samples are determined according to the ASTM
catalysts for upgrading of residue has been reported before [24]. The standard procedures.
characteristics of HDT and HDC catalysts are presented in Table 1. 4. Test Conditions
Before loading, both catalysts are heated to 130 oC and are held at Hydrocracking is performed under the following process condi-
this temperature for about 6 hr for drying. Then, they are sulfided tions:
with an appropriate agent according to the manual of the catalyst 1. H2/HC=1,357 Nm3/Sm3
vendor. 2. LHSV=0.8, 0.9 and 1.05 hr−1
3. Feed and Product Characterization 3. Temperature=360 oC, 370 oC, 380 oC and 390 oC
The hydrocracking feed is prepared by blending of the fresh VGO 4. Pressure=146 bar
and recycle feed (unconverted oil) taken from commercial Isomax The pressure and H2/HC are selected as recommended by the
unit which is located in Iran. The feed properties are shown in Table catalyst vendor. The LHSV and start of run temperature in a com-
3. Mixing ratio of fresh and recycle feed is 83.3 vol% and 16.7 vol%, mercial reactor are normally around 1 hr−1 and 380 oC; therefore,
respectively. wider conditions for these variables are selected.
The main products of the process included gas (G), naphtha (N),
distillate (D) and unconverted oil whilst its properties are assumed CHEMICAL REACTIONS
the same as VGO. According to Table 2, the latter assumption is
not far from reality because of the negligible difference between 1. Hydrotreating Section
distillation properties of these lumps. The average density and boiling The following reactions are major reactions promoting in the hy-
drotreating of VGO [25]:
1-1. Hydrodesulfurization (HDS)
Table 1. Characteristics of HDT&HDC catalyst As a lumped sulfur compound in VGO, 4,6-dimethyl-dibenzothio-
Property HDT HDC phene is selected to take part in HDS reactions [26]. To remove sulfur
from this compound, 2 molecules of hydrogen are required per sulfur
Size & Shape 1/16'' & Quadralobe 1/16'' & Cylindrical
atom to convert it to a hydrocarbon within the boiling range of diesel
Color Green Brown
cut [27]. Therefore, by HDS reaction, the sulfur lump in VGO is
Bulk density (kg/m3) 750 850
converted to diesel and H2S.
BET surface area (m2/g) 186.5 199.4 1-2. Hydrodenitrogenation (HDN)
Langmuir surface area 259.2 273.7 Organonitrogen compounds in petroleum feedstock undergo hy-
(m2/g) drodenitrogenation to form ammonia and liquid hydrocarbon [28,
Average pore diam (A0) 89.09 69.14 29]. Between alkyl amines, quinoline is one of the most relevant
Main Ingredients Mo, Ni, Ti Zr, W, Ni, Si, Al nitrogen compounds in VGO of which seven molecules of hydro-
Korean J. Chem. Eng.(Vol. 27, No. 4)
1102 S. Sadighi et al.

gen per nitrogen atom are needed to convert it to a hydrocarbon 4. Hydrogen feed is pure
liquid within the boiling range of naphtha cut [27]. 5. The petroleum feed and the products are in the liquid phase in
1-3. Hydrodearomatization (HDA) the reactor
The aromatic compounds present in the feed oil are grouped into 6. The pilot unit is in steady state operation
mono-, di-, tri- and polyaromatics [30]. It was reported that mono- 7. Catalyst activity does not change with time; therefore simulation
aromatics are significantly more difficult to saturate [31] and the is only valid for start of run conditions.
amount of tri- and poly aromatics in VGO is considerably lower 1. Axial Dispersion
than di-aromatics [32]. Therefore, saturation of di-aromatics is the To ensure that the reactor works at plug flow regime we have to
most possible HDA reaction in VGO hydrotreating which needs survey whether backmixing can be neglected or not. A criterion re-
two molecules of hydrogen [27]. ported by Mears [34] can be used to estimate the required mini-
2. Hydrocracking Section mum bed length Lb, so that backmixing effects can be ignored:
In the hydrocracking part, catalyst converts complex ring com-
L 20n 1
pounds into light products by the following sequence of hydrocrack- -----b > --------- ⋅ ln ---------
- (1)
dp Pez 1− x
ing reactions [27]. All of them consume hydrogen which should be
included in the product of the reactor to have an exact mass balance In Eq. (1), Lb, catalyst bed length; dp, particle diameter; n, order of
for the hydrocracking model. reaction; x, fractional conversion; Pez, Peclet number, can be esti-
mated as a function of Reynolds number. Depending on the corre-
MODELING APPROACH FOR HYDROCRACKING lation used for determining Pez, published by Froment and Bischoff
SECTION [35] or Baerns et al. [36], the minimum ratio of Lb/dp varies between
450 and 3,850, whereas the experimental ratio is 400.
This work considers the 4-lump mathematical model, i.e., VGO, For the current case, the particle diameter and the bed length are
distillate, naphtha and gas, to match main products in the pilot. The 0.211 mm and 122 cm, respectively. Consequently, Lb/dp is about
kerosene and diesel, also light and heavy naphtha are lumped together 578 for understudy reactor, enough higher than the minimum value.
as distillate and naphtha cuts, respectively. Moreover, it is assumed Hence, assuming the reactor as a plug reactor can be an acceptable
that VGO (as hydrocracking feed) should consume hydrogen to be assumption.
cracked to lighter cuts. Fig. 2 illustrates the process pathways asso- 2. Kinetic Expression
ciated with the mentioned strategy. Note that if all pathways of reac- For each reaction, a kinetic expression (R) is formulated as the
tions are considered, the model would include twelve kinetic param- function of mass concentration and kinetic parameters (k0, E). Based
eters which should be estimated using experimental data. on these assumptions, the kinetic constants of the proposed model
Mathematical models for a trickle-bed catalytic reactor can be are as in the following:
complex due the many microscopic and macroscopic effects occur-
−E
ring inside the reactor: flow patterns of both phases, size and shape Vacuum gas oil or Feed (F): kFj = k0Fj exp ⎛ ----------Fj⎞ (2)
⎝ RT ⎠
of a catalyst particles, wetting of the catalyst pores with liquid phase,
pressure drop, intraparticle gradients, thermal effects and, of course, Note thatin Eq. (2) represents diesel (D), naphtha (N) and gas (G)
kinetics on the catalyst surface [30]. It is therefore more practical to
− EDj'⎞
reduce the complexity of the reactor, focusing only on momentous Diesel (D): kDj' = k0Dj' exp ⎛ ----------- (3)
process variables. This suggests the development of simpler models ⎝ RT ⎠
that incorporates the fewest possible parameters. The following as- j' in Eq. (3) represents naphtha (N) and gas (G).
sumptions have been made in the development of the present model:
1. Hydrocracking is a first-order hydrocracking reaction. Since hy- − ENG⎞
Naphtha (N): kNG = k0NG exp ⎛ -----------
- (4)
drogen is present in excess, the rate of hydrocracking can be taken to ⎝ RT ⎠
be independent of the hydrogen concentration [33].
In Eqs. (2) to (4), T and R are the bed temperature and ideal gas
2. A cell network pattern exists in the trickle bed reactor
constant, respectively.
3. The pilot reactor operates under isothermal conditions
The reaction rates (R) can be formulated as the following:
G

Vacuum gas oil reaction (RF): RF = ∑ kFj CF (5)

j=D

CF in Eq. (5) is the mass concentration of VGO.

As depicted in Fig. 2, it is assumed for converting of VGO to
hydrocracking products, hydrogen is added to the related path. There-
fore, the net reaction rate for them can be described as the following:
G

Diesel (RD): RD = kFD CF( 1+ α) − ∑ kDj'CD (6)

j'=N

Naphtha (RN): RN=kFNCF(1+α)+kDNCD−kNGCNG (7)

Fig. 2. The complete 4-lump kinetic model. Gas (RG): RG=kFGCF(1+α)+kDGCD+kNGCN (8)

July, 2010
 4-Lump kinetic model for vacuum gas oil hydrocracker involving hydrogen consumption 1103

In Eqs. (6) to (8), α shows the consumed unit mass of hydrogen cell (Eq. (11)). Finally, Eqs. (10) to (13) for each cell should be solved
per unit mass of converted VGO which is added to the molecular simultaneously to calculate the hydrocracking products.
structure of products (diesel, naphtha and gas) during hydrocrack-
Cj(i−1)ν (i−1)±η.ε.Rj(i)×Vcat(i)=Cj(i)ν (i) (10)
ing reactions. This coefficient is modeled by a quadratic polynomial
equation [37] to predict the consumed hydrogen during the hydroc- F (i)
ν (i) = -----------
m
(11)
racking process as a response function of temperature (T) and LHSV ρ(i)
as the following: G

Fm ( i ) = ∑ C j ( i ) ν ( i ) (12)
α =β0+β1T+β2.LHSV+β11.T2+β22.LHSV2+β12.T.LHSV (9) j=F

In this equation, β0 is the intercept coefficient, β1 and β2 are the linear V

Vcat( i) = ------b (13)
terms, β11 and β22 are the squared terms and β12 is the interaction term. N
These coefficients are estimated from experimental data and the ade-
In the above equations, the “−” is for reactant (feed or VGO), and
quacy of regression is checked with analysis of variance (ANOVA)
the “+” sign is for the products; j, lumps from feed (F) to gas (G);
using R-squared and Fischer F-test [37,38].
C, the mass concentration of lumps; η, effectiveness factor; ε, bed
3. Mass Balance
void fraction; Fm(i), mass flowrate in each cell; Vcat(i) is the volume
Plug flow for fixed-bed reactors is assumed in many reported
of hydrocracking catalyst in each cell; Vb is the volume of hydroc-
pilot scale reactor models that consist of a set of ordinary differential
racking catalyst and N is the number of cells. The effectiveness factor
equation (ODEs) with defined boundary conditions. In this paper,
for spherical catalyst in trickle bed regime and the bed void fraction
to model the hydrocracking section, we implemented a cell network
is 0.7 [44] and 0.35, respectively.
approach, the accuracy of which was confirmed for trickle bed reac-
The only unknown variable in the above equations is density of
tors [39]. As shown in Fig. 3, the hydrocracking catalytic bed from
feed and product stream inside the reactor which can be calculated
the inlet to the outlet is divided into a number (N=200) of well-mixed
as follows:
cells that are grouped along the longitude direction. Mixing only oc-
G
curs within each cell and backmixing is not accounted for between
ρ0 ( i ) = ∑ Y j ( i ) ρj (14)
the adjacent cells. It is obvious that this approach is equivalent to a j=F
one-parameter non-ideal reactor model [40] that was adopted in
In Eq. (14), j is from feed (F) to gas (G), ρj is density of lumps. It
some of the previous works in reactor modeling [41-43] in which by
should be noted that the density evaluated by this equation is stan-
increasing the series reactor we can simulate the plug flow behavior.
dard density, which can be determined at reactor condition by the
In the interest of improving accuracy of the developed model,
Standing-Katz correlation [45]. In deviating from the SI system we
the volumetric flow rate in the reactor (ν) is considered variable
give the equation with the original units:
and it is calculated according to the density of output stream of each
ρj(p, T)=ρ0(i)+∆ρp(i)−∆ρT(i) (15)

Where ρ0(i) represents the density at standard conditions in l b/ft3.

The pressure dependence can be evaluated by:
−0.0425. ρ 0 ( i ) p
∆ρp (i) = [0.167 +16.181 × 10 ]. ------------
1000
2
−0.0603. ρ0 ( i ) p
− 0.01 × [0.299 + 263 × 10 ]. ------------ (16)
1000
Where p is the pressure in psia. Since the density drops with ascend-
ing temperature, a temperature correction with the temperature T
in 0R is needed:
∆ρT(i)=[0.0133+152.4×(ρ0(i)+∆ρp(i))−2.45].[T−520]
∆ρT(i)=−[8.1×10−6−0.0622×10−0.764.(ρ (i)+∆ρ (i))].[T−520]2 0 p
(17)
For parameter estimation, the sum of squared error, SQE, as given
below, is minimized:
Nt G
meas pred 2
SQE = ∑ ∑ (Yjk − Yjk ) (18)
k=1 j=F

meas pred
In Eq. (18), Nt, Yjk and Yjk are the number of test runs, meas-
ured product yield and the predicted by model, respectively.
The hydrocracking reaction model according to Eqs. (2) to (17) is
coded and solved simultaneously by using Aspen Custom Modeler
(ACM) programming environment (AspenTech, 2001) to evaluate
Fig. 3. Schematic representation of series mixed cells. the product yields (Yi). Then Eq. (18) is minimized by sequencing
Korean J. Chem. Eng.(Vol. 27, No. 4)
1104 S. Sadighi et al.

NL2Sol and Nelder-mead algorithm which are both in the Aspen

Custom Modeler software. NL2Sol algorithm is a variation on New-
ton’s method in which part of the Hessian matrix is computed exactly
and part is approximated by a secant (quasi-Newton) updating meth-
od. To promote convergence from a poor initial point, a trust-region
is used along with a choice of model Hessian. Hence, the approxi-
mate region is found with NL2Sol; then to fine tune the parame-
ters, Nelder-Mead simplex method is used.
To evaluate the estimated kinetic parameters, average absolute
deviation of predictions (AAD%) [46] is calculated by using the
following expression.
Fig. 4. Aromatic content vs. temperature. H2/oil=1,357 Nm3/Sm3,
Nt meas pred 2 pressure=146 bar (◆) LHSV=0.8 hr−1, (■) LHSV=0.9 hr−1,
(Y − Y )
∑ --------------------------------
k
meas
2
k
(▲) LHSV=1.05 hr−1.
k=1 Yk
AAD% =100 -----------------------------------------
-% (19)
Nt

RESULTS AND DISCUSSIONS

The effect of temperature on the yields of gas, naphtha, distillate

and unconverted VGO (hydrocracking products) at three constant
LHSVs (0.8, 0.9 and 1.05 hr−1) was studied. We found that there
are no abnormalities in the hydrocracking behavior of the catalyst
in the experienced range. As expected, for gas, naphtha and distil-
late the temperature promotes the hydrocracking paths so that the
yields of these products are increased. For the unconverted VGO
or residue, temperature acts reversely so that it decreases the yield Fig. 5. Hydrogen consumption vs. temperature. H2/oil=1,357 Nm3/
of residue. The effect of LHSV is in agreement with the rule of the Sm3, pressure=146 bar (◆) LHSV=0.8 hr−1, (■) LHSV=
smaller LHSV, the better the hydrocracking. Thus, converting of 0.9 hr−1, (▲) LHSV=1.05 hr−1.
feed to gas, naphtha and diesel is always better in low LHSV. For
the unconverted VGO or residue this reason acts reversely.
For all experiments (12 tests) it is found that the sulfur and nitrogen increased sharply with temperature, but LHSV does not have any
content of the hydrocracking product (mixture of naphtha, distillate sensible variation on it. Even, in temperatures 380 oC and 390 oC,
and unconverted oil) are less than 100 and 50 ppmwt, respectively. the hydrogen consumption is a little higher at lower LHSVs, dis-
Therefore, it is concluded that the hydrotreating catalyst can pro- cussed later.
vide diesel with the sulfur content lower than 50 ppmwt which is Table 4 shows the yield of hydrocracking products, convention-
desirable for a transport fuel. ally calculated on the basis of VGO input feed. The gas in this table
The aromatic content of hydrocracking product in different LHSVs is only composed of C1, C2, C3, C4 and a little amount of C5, repre-
versus temperature is depicted in Fig. 4. As it can be concluded from senting only hydrocracking products. As it can be seen, in most of
this figure, the total aromatic content of the liquid products is reduced the experiments we have more than 1% error in mass balance, mainly
by temperature. But, LHSV has the reverse effect on that. resulting from hydrotreating (HDS, HDN and HDA) reactions which
The mass flow rate of hydrogen consumption for each experi- are mainly performed in the hydrotreating section Therefore, the
ment is calculated by performing a mass balance around the sys- effect of hydrotreating reactions is on the boundary of the hydroc-
tem. We expect that the hydrogen consumption has been increased racking section, second catalytic bed, which can be considered ac-
noticeably with raising the temperature and LHSV. But, as it was cording to the reactions mentioned in section 2.4.1.
revealed from Fig. 5, our expectation is only satisfied for the tem- To do this, at first hydrogen consumption is categorized to four
perature. It means the hydrogen consumption in constant LHSV main groups involved of HDS, HDN, HDA and HDC. According

Table 4. Yield percent of hydrocracking process based on the fresh VGO feed
T (oC) 360 370 380 390 360 370 380 390 360 370 380 390
−1
LHSV(hr ) 0.8 0.8 0.8 0.8 0.9 0.9 0.9 0.9 01.05 01.05 01.05 01.05
Gas% 01.71 02.54 02.81 03.23 01.37 01.66 02.23 02.86 01.17 01.30 01.64 01.99
Naphtha% 07.13 08.92 11.96 13.93 05.96 07.91 09.56 12.17 05.58 06.82 08.23 10.28
Diesel% 22.33 24.41 26.77 31.55 19.84 20.91 22.96 27.65 18.66 19.93 21.39 24.56
Residue% 67.53 63.12 57.67 50.94 71.47 68.33 64.22 56.55 73.24 70.63 67.56 62.08
Total 98.71 98.99 99.21 99.66 98.63 98.82 98.96 99.23 98.65 98.68 98.82 98.91

July, 2010
 4-Lump kinetic model for vacuum gas oil hydrocracker involving hydrogen consumption 1105

Table 5. Hydrogen consumption categorized to main reaction groups

T (oC) 360 370 380 390 360 370 380 390 360 370 380 390
−1
LHSV(hr ) 0.800 0.800 0.800 0.800 0.900 0.900 0.900 0.900 1.050 1.050 1.050 1.050
HDS (gr/hr) 0.370 0.370 0.370 0.370 0.420 0.420 0.420 0.420 0.490 0.490 0.490 0.490
HDN (gr/hr) 0.620 0.620 0.620 0.620 0.700 0.700 0.700 0.700 0.810 0.810 0.810 0.810
HDA (gr/hr) 0.230 0.280 0.330 0.370 0.240 0.290 0.340 0.390 0.220 0.290 0.320 0.400
HDC (gr/hr) 0.760 1.060 1.360 2.030 0.620 0.810 1.120 1.840 0.570 0.770 1.030 1.570
Total 1.980 2.335 2.681 3.392 1.977 2.209 2.575 3.349 2.089 2.360 2.656 3.280

Table 6. Product rates of the hydrotreating process

T (oC) 360 370 380 390 360 370 380 390 360 370 380 390
−1
LHSV(hr ) 0.80 0.80 0.80 0.80 0.90 0.90 0.90 0.90 1.05 1.05 1.05 1.05
Gas (gr/hr) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Naph. (gr/hr) 5.65 5.65 5.65 5.65 6.35 6.35 6.35 6.35 7.41 7.41 7.41 7.41
Dis. (gr/hr) 16.95 16.95 16.94 16.94 19.07 19.07 19.06 19.06 22.27 22.26 22.25 22.25
Res. (gr/hr) 131.51 131.56 131.61 131.65 147.94 147.98 148.03 148.09 172.61 172.68 172.72 172.80

to the assumptions presented in section 2.4.1, the required hydro-

gen for reduction of sulfur and nitrogen as well as aromatization is
shown in Table 5.
The consumed hydrogen for hydrocracking is calculated by sub-
tracting the hydrogen consumption for hydrotreating from total con-
sumed hydrogen. Table 5 revealed this phenomenon that in a con-
stant LHSV, the hydrogen consumption for hydrocracking is in-
creased by temperature, but in constant temperature it is decreased
by LHSV. Therefore, we suppose that a hydrocracking reaction is
reversely affected by LHSV and it can be the reason for nuance vari-
ation of total hydrogen consumption by LHSV as shown in Fig. 5.
Fig. 6. Hydrogen consumption per converted hydrocarbon vs. tem-
As it was discussed in the chemical reactions section, hydrotreat-
perature H2/oil=1,357 Nm3/Sm3, pressure=146 bar (◆)
ing has a share in producing naphtha and diesel after converting LHSV=0.8 hr−1, (■) LHSV=0.9 hr−1, (▲) LHSV=1.05 hr−1.
sulfur and nitrogen compounds to H2S and NH3, respectively. For
each corresponding reaction, a specific amount of hydrogen is con-
sumed and H2S, NH3, naphtha and diesel products are determined Table 7. Coefficient values for the responses α in Eq. (9)
by the amount of sulfur and nitrogen value in the VGO feed. There- Variable β0 β1 β2 β11 β22 β12
fore, the feed emitted from hydrotreating, entered into the hydroc- T.LHSV 3441.319 −10.762 −18.906 0.00854 5.452 −0.00858
racking bed is composed of naphtha, diesel and purified VGO which
are presented in Table 6. The produced naphtha and diesel should
be subtracted from the final products to determine the performance directly by temperature, but LHSV has a reverse effect on that.
of hydrocracking catalyst. The α value is fitted to Eq. (9) using the regression analysis tool
In Table 6, the production of gas in the hydrotreating section has of Excel software (Microsoft office version 2007). The calculated
been neglected, which can be a source of error, especially at high coefficients are shown in Table 7. After that an ANOVA test is per-
temperatures which is discussed later. formed to evaluate the validity of the model. It confirmed that the
Up to now, the hydrogen consumption caused by hydrotreating quadratic model can predict α with an R2 (R square) of about 0.9735
reactions is considered in mass balances and the deflection result- which is acceptable. Moreover, the adequacy of the fitted model is
ing from them is healed. But, as it was mentioned before, a part of tested using static Fischer (F) with 1% critical level. The value of
the hydrogen will be entered in the molecular structure of products F=43.40 is higher than F(5, 6, 0.01)=8.47, demonstrating that the
during the hydrocracking reaction. To model this effect, considered regressed model fitted well the observed values. The parity plot for
by (α) in hydrocracking mathematical expression, the hydrogen measured and predicted by the quadratic model is presented in Fig. 7.
consumed per converted VGO should be evaluated from experi- After achieving a model to predict α, twelve kinetic parameters
ments. needed for the 4-lump kinetic network (Fig. 1) are estimated by us-
Fig. 6 shows the variation of hydrogen consumed (mgr) per con- ing measured pilot data. Table 8 shows the estimated values of ap-
verted VGO (gr) during hydrocracking reactions, representing by α, parent activation energies and frequency factors. In this table, the
in the second catalyst bed. As discussed before, this value is increased rate constants for all reactions are evaluated in the average operat-
Korean J. Chem. Eng.(Vol. 27, No. 4)
1106 S. Sadighi et al.

Fig. 8. The reduced 4-lump kinetic model.

Table 10. Kinetic parameters for the reduced network

Frequency factor Activation energy
Rate order
(m3 ·hr−1 ·m3 cat−1) (kJ/mol)
k0FD 4.26E+07 EFD 22.43 kFD 0.83
k0FN 1.44E+10 EFN 30.66 kFN 0.54
k0FG 0 EFG 0 kFG 0
k0DG 0 EDN 0 kDN 0
k0DN 0 EDG 0 kDG 0
Fig. 7. Parity plots for quadratic model of hydrogen consumption
for hydrocracking H2/oil=1,357 Nm3/Sm3, pressure=146 bar k0NG 2.06 ENG 0 kNG 1
(◆) LHSV=0.8 hr−1, (■) LHSV=0.9 hr−1, (▲) LHSV=1.05
hr−1.

Table 8. Kinetic parameters for the complete network

Frequency factor Activation energy
Rate order
(m3 ·hr−1 ·m3 cat−1) (kJ/mol)
k0FD 6.64E+07 EFD 23.01 kFD 0.55
k0FN 1.04E+08 EFN 24.27 kFN 0.33
k0FG 0 EFG 12.96 kFG 0
k0DG 0 EDN 0.26 KDN 0
k0DN 5.10E+03 EDG 23.75 kDG 2.4E-5
k0NG 2.7 ENG 1.1E-6 KNG 1

Table 9. The for the different strategies in the plug flow reactor Fig. 9. Parity plot for gas resulted by reduced kinetic network H2/
Lump Complete network Reduced network oil=1,357 Nm3/Sm3, pressure=146 bar.
Gas 12.41 9.32
Naphtha 08.81 7.17 justify the higher yield of distillate in hydrocracking process. Also,
Distillate 04.52 4.31 we suppose that the high value for the rate of naphtha to gas is be-
Un.VGO 03.05 2.88 cause of the simplicity of the cracking of naphtha’s light chains in
Ave. 7.2 5.92 comparison to heavy chains of distillate and VGO, which seems a
reasonable phenomenon.
Therefore, three paths which are VGO to gas, distillate to gas
ing temperature (375 oC). To compare the simulated and measured and distillate to naphtha can be ignored. Moreover, the activation
product values, average absolute deviation is calculated and presented energy of naphtha to gas is very low so that it can be omitted too.
in Table 9 with the name of the complete network. As it is found Finally, a reduced kinetic network results, depicted in Fig. 8. At this
from this table, the for prediction of all yields by the resulting kinetic time, there are only five remaining kinetic constants which should
factors in Table 9 is about 7.2%, which we think it is acceptable in be estimated. But, for that, there are forty-eight observations, making
the wide range of testing conditions. The main source of error is acceptable degree of freedom for parameter estimation procedure.
outstanding deviation for prediction of gas, which will be discussed The reduced model is then estimated again using measured data,
later. producing new coefficients as presented in Table 10. Upon com-
Data in Table 8 reveal that rate constants for the conversion of paring the measured data against the model predictions, the aver-
feed to distillate (kFD) and naphtha (kFN) are not far from each other. age AAD% for the reduced model is 5.92% which is dramatically
This phenomenon is consistent with the literature [47,48] that re- improved in comparison to the complete network. This confirms
ported zeolite hydrocracking catalysts have tendency to produce naph- the same situation for the claim of the previous work for thermal
tha, not more than distillate, but higher than amorphous catalysts. cracking that stated predictions using the reduced parameter model
Moreover, it can be concluded that the lower tendency of the cata- were more accurate [49].
lyst to convert distillate to naphtha as well as distillate to gas can The AAD% of all lumps resulted by reduced kinetic network is
July, 2010
 4-Lump kinetic model for vacuum gas oil hydrocracker involving hydrogen consumption 1107

Fig. 10. Parity plot for naphtha resulted by reduced kinetic net-
work H2/oil=1,357 Nm3/Sm3, pressure=146 bar.
Fig. 13. Parity plot for hydrogen consumption of hydrocracking
resulting from reduced kinetic network, H2/oil=1,357 Nm3/
Sm3, Pressure=146 bar.

than predicted by the model. We think that during the hydrotreat-

ing of VGO in the first catalytic section, there is the possibility for
hydrocracking of light chains to gas, which is not included in the
model, and the concentration of gas in the boundary of hydrocrack-
ing bed is assumed zero. Therefore, at high temperatures, fortifying
this possibility, we have more deviation for gas prediction, which
increases the AAD% of the model. The same reason can be inter-
preted for naphtha that some hydrocracking reactions can consume
or produce naphtha in the hydrotreating section, not included in the
model and they create more AAD% for this lump in comparison to
Fig. 11. Parity plot for distillate resulted by reduced kinetic net- diesel and VGO. We found that the mass balance error of the model
work H2/oil=1,357 Nm3/Sm3, pressure=146 bar.
for all predictions is less than 0.05%, confirming the exactness of
the discussed approaches to decrease the error created by reducing
impurities and adding hydrogen consumption. We think that the
nuance error may be caused by nickel, vanadium or coke formation
which was not included in the model.
Finally, the parity plot for hydrogen consumption in the hydroc-
racking section can be observed in Fig. 13. The AAD% for that is
about 8.59%. We think that this error is natural because of the depen-
dency of hydrogen consumption to yield of VGO, the latter of which
has 2.88% error. Also, the quadratic model applied for calculation of
needed hydrogen to VGO conversion has intrinsically a little devi-
ation which is transferred to the kinetic model predictions.

CONCLUSIONS

Fig. 12. Parity plot for residue resulted by reduced kinetic net- In this research, a 4-lump kinetic model for gas oil hydrocrack-
work H2/oil=1,357 Nm3/Sm3, pressure=146 bar. ing was developed to predict the yield of gas, naphtha, distillate and
hydrogen consumption of VGO hydrocracking in a pilot scale reac-
presented in Table 9 with the title of the reduced network. tor charged with zeolite-base catalyst. Experiments were carried out
The parity plots for measured data and model predictions are pres- in temperatures from 360 oC to 390 oC, LHSV from 0.9 to 1.05, H2
ented in Figs. 9 to 12, certifying the acceptable agreement between pressure at 146 bar and H2/oil at 1,357 Nm3/m3. The active zone of
experimental and predicted values by the developed model. It is the reactor was divided into hydrotreating and hydrocracking sec-
evident from these figures that the yield prediction of the reduced tions, of which the effect of hydrotreating was considered accord-
kinetic network is acceptable. ing to HDS, HDN and HDA reactions reported in the literature. The
Furthermore it can be understood from Fig. 9 that the measured hydrogen consumption of the hydrocracking reactions was imple-
values for the yield of gas, related to high temperatures, are higher mented in the kinetic model of hydrocracking by use of a quadratic
Korean J. Chem. Eng.(Vol. 27, No. 4)
1108 S. Sadighi et al.

response surface model which modeled the unit mass of hydrogen (2006).
consumed per unit mass of converted VGO. The data generated in 22. X. Meng, Ch. Xu, J. Gao and L. Li, Catal. Communications, 8, 1197
the pilot plant reactor were used to estimate kinetic parameters for (2007).
each involved hydrocracking reaction and coefficients of hydrogen 23. S. Zahedi, J. Towfighi, R. Karimzadeh and M. R. Omidkhah, Korean
consumption model. The results showed that the reduced kinetic J. Chem. Eng., 25, 4 (2008).
network could predict the experimental data with average deviation 24. J. W. Choi, W. S. Choi, K. H. Lee and B. H. Ha, J. Korean Institute
about 5.92%, which was 1.28% more accurate than the complete Chem. Eng., 32, 5 (1994).
network. Moreover, the hydrogen consumption for hydrocracking 25. M. A. Rodriguez and J. Ancheyta, Energy & Fuels, 18, 789 (2004).
of VGO to light products was predicted with 8.59% average abso- 26. N. Kunisada, K. H. Choi, Y. Korai and I. Mochida, Appl. Catal. A:
lute deviation. Gen., 260, 185 (2004).
The advantage of this model over the previous ones is imple- 27. C. S. Hsu and P. R. Robinson, Practical advances in petroleum pro-
menting of hydrogen in the mass balance equation. This is impor- cessing, Volume I, Springer Publication, 1st Ed. (2006).
tant not only to have a less error yield prediction, but also to evaluate 28. F. S. Mederos, M. A. Rodriguez, J. Ancheyta and E. Arce, Energy
the hydrogen consumption of hydrocracking catalyst which is mo- & Fuels, 20, 996 (2006).
mentous as an economical variable. 29. R. J. Angelici, Polyhedron, 16, 18, 3073 (1997).
30. M. Bhaskar, G. Valavarasu, B. Sairman, K. S. Balaraman and K,
REFERENCES Balu, Ind. Eng. Chem. Res., 43, 6654 (2004).
31. G. M. Sanzhez and J. Ancheyta, Appl. Catal. A: Gen., 207, 407
1. J. J. Vestraete, K. Le Lannic and I. Guibard, Chem. Eng. Sci., 62, (2001).
5402 (2007). 32. R. G. Tailleur, Comput. Chem. Eng., 29, 2404 (2005).
2. A. Alvarez and J. Ancheyta, Appl. Catal. A: Gen., 351, 148 (2008). 33. S. Mohanty, D. N. Saraf and D. Kunzru, Fuel Processing Tech., 29,
3. A. Alvarez and J. Ancheyta, Chem. Eng. Sci., 63, 662 (2008). 1 (1991).
4. G. Valavarasu, M. Bhaskar and B. Sairam, Petrol. Sci. Technol., 23, 34. D. E.Mears, Chem. Eng. Sci., 26, 1361 (1971).
1323 (2005). 35. G. F. Froment and K. B. Bischoff, Chemical reactor analysis and
5. J. Ancheyta-Juarez, F. Lopez-Isunza and E. Aguilar-Rodriguez, design, 2nd Ed., Wiley, New York (1990).
Appl. Catal. A: Gen., 177, 227 (1999). 36. M. Baerns, H. Hofmann and A. Renken, Chemische reaktionstech-
6. F. J. Krambeck, Kinetics and thermodynamics lumping of multicom- nik, George Thieme Verlag, Stuttgart (1987).
ponent mixtures, Elsevier, Amsterdam, 111 (1991). 37. D. C. Montgomery, Design and analysis of experiments, John Wiley
7. J. Ancheyta, S. Sanchez and M. A. Rodriguez, Catal. Today, 109, 76 & Sons, New York (2001).
(2005). 38. G. M. Clarke and R. E. Kempson, Introduction to the design and
8. F. Mosby, R. D. Buttke, J. A. Cox and C. Nikolaids, Chem. Eng. Sci., analysis of experiments, Arnold, London (1997).
41, 989 (1986). 39. G. Jing, Y. Jiang and M. H. Al-Dahhan, Chem. Eng. Sci., 63, 751
9. K. Aboul-Gheit, Erdol Erdgas Kohle, 105, 1278 (1989). (2008).
10. R. Ayasse, H. Nagaishi and E. W. Chan, Fuel, 76, 1025 (1997). 40. O. Levenspiel, Chem. Reaction Eng., 3rd Ed., John Wiley & Sons
11. M. A. Callejas and M. T. Martinez, Ind. Eng. Chem. Res., 38, 98 Inc. (2001).
(1999). 41. V. K. Pareek, Z. Yap, M. P. Brungs and A. A. Adesina, Chem. Eng.
12. S. M. Yui and E. C. Sanford, Ind. Chem. Res., 28, 319 (1989). Sci., 56, 6063 (2001).
13. J. Ancheyta, F. Lopez and E. Aguilar, Appl. Catal. A: Gen., 177, 227 42. S. N. Palaskar, J. K. De and A. B. Pandit, Chem. Eng. Tech., 23
(1999). (2001).
14. K. Aoyagi, W. C. McCaffrey and M. R. Gray, Petrol. Sci. Technol., 43. A. Kumar, G. M. Ganjyal, D. Jones and M. A. Hanna, J. Food Eng.,
21, 997 (2003). 84, 441 (2008).
15. S. Sanchez, M. A. Rodriguez and J. Ancheyta, Ind. Eng. Chem. Res., 44. P. L. Mills and M. P. Dudukovic, Ind. Eng. Chem. Fund., 18, 2
44, 9409 (2005). (1979).
16. J. Singh, M. M. Kumar, A. K. Saxena and S. Kumar, Chem. Eng. 45. T. Ahmed, Hydrocarbon phase behavior, Gulf Publishing, Hous-
J., 108, 239 (2005). ton (1989).
17. R. M. Almeida and R. Guirardello, Catal. Today, 109, 104 (2005). 46. A. Marafi, E. Kam and A. Stanislaus, Fuel, 87, 2131 (2008).
18. D. Remesat, B. Young and W. Y. Surcek, Chem. Eng. Res. Design. 47. M. A. Ali, T. Tatsumi and T. Masuda, Appl. Catal. A: Gen., 233, 77
In press. (2002).
19. F. van landeghem, D. Nevicato, I. Pitault, M. Forissier, P. Turlier, C. 48. J. Scherzer and A. J. Gruia, Hydrocracking Sci. and Tech., CRC
Derouin and J. R. Bernard, Appl. Catal., A, 138, 381 (1996). Press (1996).
20. R. M. Almeida and R. Guirardello, Catal. Today, 109, 104 (2005). 49. J. Singh, M. M. Kumar, A. K. Saxena and S. Kumar, Chem. Eng.
21. X. Meng, Ch. Xu, J. Gao and L. Li, Appl. Catal. A: Gen., 301, 32 J., 108, 239 (2005).

July, 2010

View publication stats