Fuel 90 (2011) 21652181

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Kinetic model development and simulation of simultaneous hydrodenitrogenation

and hydrodemetallization of crude oil in trickle bed reactor
Aysar T. Jarullah, Iqbal M. Mujtaba , Alastair S. Wood
School of Engineering, Design and Technology, University of Bradford, Bradford BD7 1DP, UK

a r t i c l e

i n f o

Article history:
Received 3 August 2010
Received in revised form 17 January 2011
Accepted 18 January 2011
Available online 3 February 2011
Keywords:
Hydrodenitrogenation
Hydrodemetallization
Trickle bed reactor
Mathematical modelling
Parameter estimation

a b s t r a c t
One of the more difcult tasks in the petroleum rening industries that have not been considered largely
in the literature is hydrotreating (HDT) of crude oil. The accurate calculations of kinetic models of the relevant reaction scheme are required for obtaining helpful models for HDT reactions, which can be condently used for reactor design, operating and control. In this work, an optimization technique is employed
to evaluate the best kinetic models of a trickle bed reactor (TBR) process utilized for hydrodenitrogenation (HDN) and hydrodemetallization (HDM) that includes hydrodevanadization (HDV) and hydrodenickelation (HDNi) of crude oil based on pilot plant experiments. The minimization of the sum of the
squared errors (SSE) between the experimental and estimated concentrations of nitrogen (N), vanadium
(V) and nickel (Ni) compounds in the products is used as an objective function in the optimization problem to determine the kinetic parameters.
A series of experimental work was conducted in a continuous ow isothermal trickle bed reactor, using
crude oil as a feedstock and the commercial cobaltmolybdenum on alumina (CoMo/c-Al2O3) as a catalyst.
A three-phase heterogeneous model based on twolm theory is developed to describe the behaviour
of crude oil hydroprocessing in a pilotplant trickle bed reactor (TBR) system. The hydroprocessing reactions have been modelled by power law kinetics with respect to nitrogen, vanadium and nickel compounds, and with respect to hydrogen. In this work, the gPROMS (general PROcess Modelling System)
package has been used for modelling, simulation and parameter estimation via optimization. The model
simulations results were found to agree well with the experiments carried out in a wide range of the
studied operating conditions. The model is employed to predict the concentration proles of hydrogen,
nitrogen, vanadium and nickel along the catalyst bed length in three phases.
2011 Elsevier Ltd. All rights reserved.

1. Introduction
The technologies for upgrading petroleum fractions are some of
the most important processes in the rening industry because of
the growing market demands for different crude oil derivatives
and decreasing availability of light oils [1]. Therefore, it is essential
to increase the productivity of distillates with high quality. Among
these technologies, hydrotreatment operation, which has the
capacity for increasing the distillates production and to remove
the impurities such as sulfur, nitrogen, metals (Ni and V) and
asphaltenes [2].
The presence of nitrogen compounds in crude oil or oil fractions
has a detrimental effect for rening industries. Nitrogen compounds are responsible for catalyst poisoning and reducing catalyst
activity. Furthermore, nitrogen compounds have toxic effects on

Corresponding author. Fax: +44 (0)1274 235700.

E-mail address: I.M.Mujtaba@brad.ac.uk (I.M. Mujtaba).
0016-2361/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.fuel.2011.01.025

the storage stability of oil products and affect the colour of oil
products [3]. Andari et al. [4] have shown the impact of
nitrogen and sulfur compounds through their studies of Naphtha,
Kerosene and Diesel oils derived from Al-Kuwait crude oil and they
proved that these compounds showed unwanted inuence on the
stability of fuel in addition to the environmental pollution. Kaernbach et al. [5] conrmed that the nitrogen compounds signicantly
affect the catalyst activity through their works on the vacuum
residue.
The metallic compounds in crude oil have also been of great
interest to researchers in this area because of the problems caused
by these compounds. The existence of metallic compounds in
crude oil and its fractions has harmful effects. These compounds
have a very bad inuence on the HDT efciency, plug the pores
of catalysts used, cause rapid deactivation for the hydroprocessing
catalyst, where they tend to deposit on the catalyst, and seem to
act to reduce HDT activity by decreasing catalyst surface area
[68]. Also, the presence of vanadium and nickel in addition to iron
and copper affects the activity of cracking catalysts and causing an

2166

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Nomenclature
a
aL
aS
AC
A0j
API
C LH2
C Li
C SH2
C Si
Dei
dc
DLH2
DK
DLi
DR
ds
EAj
GL
hH2
kj
K LH2
K SH2
K Si
L
Lc
LHSV
mj
nj
Mw
p
PGH2
r
rg
rj
R
Sg
Sp
SSE
Sp.gr15.6
T

dimensionless number
gasliquid interfacial area, cm1
liquidsolid interfacial area, cm1
surface area, cm2
pre-exponential
factor
for
reaction
j,
(mol/
cm3)1n (cm3/g s) (mol/cm3)m
American Petroleum Institute
concentration of hydrogen in the liquid phase, mol/cm3
concentration of i compound in the liquid phase, mol/
cm3
concentration of H2 in the solid phase, mol/cm3
concentration of i compound in the solid phase, mol/
cm3
effective diffusivity of i compound in the pores of catalyst, cm2/s
diameter of cylindrical catalyst particle, cm
molecular diffusivity of H2 in the liquid, cm2/s
Knudsen diffusivity, cm2/s
molecular diffusivity of i compound in the liquid, cm2/s
reactor diameter, cm
diameter of spherical catalyst particle, cm
activation energy for j process, J/mol
liquid mass velocity, g/cm2 s
Henrys coefcient for hydrogen, MPa cm3/mol
reaction rate constant for j reaction, (mol/cm3)1n (cm3/
g s) (mol/cm3)m
gasliquid mass transfer coefcient for hydrogen, cm/s
liquidsolid mass transfer coefcient for H2, cm/s
liquidsolid mass transfer coefcient for i compound,
cm/s
length of particle, cm
length of cylindrical catalyst particle, cm
liquid hourly space velocity, h1
order of reaction of hydrogen in reaction j
order of reaction of i compound in reaction j
molecular weight, kg/kg mole
reactor total pressure, psia
partial pressure of hydrogen, MPa
particle radius, cm
pore radius, cm
chemical reaction rate of j reaction per unit mass of the
catalyst, mol/g s1
universal gas constant, J/mol K
specic surface area of particle, cm2/g
total geometric external area of particle, cm2
sum of square errors
specic gravity of oil at 15.6 C
reaction temperature

increase in the level of coal deposited. Also, the presence of these

compounds, especially vanadium in the fuel used in the high
power machines as gaseous turbines lead to the formation of some
sediment on the turbine, which can lead to the change in balance
[9,10]. Furthermore, the ash resulting from the combustion of fuels
containing sodium and particularly vanadium reacts with refractory furnace linings to lower their fusion points and hence cause
their destruction [3].
The process of crude oil hydrotreating is a new challenge and
new technology which has not been considered previously, where
all hydrotreating processes are carried out on each oil cuts separately, and not on the full crude oil (i.e. after the separation of
crude oil to its derivatives, such as gasoline, kerosene, light and
heavy gas oil). This means that a large amount of the impurities,

TmeABP
TBR
ug
uL
v
Vg
V H2
Vp
z

mean average boiling point, R

trickle bed reactor
velocity of the gas, cm/s
velocity of the liquid, cm/s
volume, cm3
pore volume per unit mass of catalyst, cm3/g
molar gas volume of H2 at standard conditions, Nl/mol
total geometric volume of catalyst, cm3
reactor bed length, cm

Greek letters
qB
bulk density of the catalyst particles, g/cm3
qL
liquid density at process conditions, lb/ft3
q15.6
density of oil at 15.6 C, g/cm3
q20
density of the oil at 20 C, g/cm3
qo
density of oil at 15.6 C and 101.3 kPa, lb/ft3
qp
particle density, g/cm3
gj
catalyst effectiveness factor j reaction
e
void fraction of the catalyst bed
lL
liquid viscosity at process conditions, mPa.s
tLC
critical specic volume of liquid feedstock, cm3/mol
i
tC
critical specic volume of i compound, ft3/mol
tL
molar volume of liquid feedstock, cm3/mol
ti
molar volume of i compound, cm3/mol
solubility coefcient of H2, Nl kg1 MPa1
kH2
DqP
pressure dependence of liquid density, lb/ft3
DqT
temperature correction of liquid density, lb/ft3
/i
Thiele Modulus i compound
h
particle porosity
s
tortuosity factor
Superscripts
0
degree
G
gas phase
H2
hydrogen
L
liquid phase or gasliquid interface
S
solid phase or liquidsolid interface
i
compound (crude oil, H2, N, V or Ni)
Subscripts
0
at the rst reactor length
c
cylindrical
f
at the nal reactor length
g
gas
H2
hydrogen
i
compound (N, V, Ni or H2)
j
reaction (HDN, HDV or HDNi)
L
liquid
s
spherical

namely, sulfur, nitrogen, metals, aromatics and asphaltenes will

be deposited at the bottom of the atmospheric and vacuum distillation column. In addition, hydrotreating process each section separately is fairly easy due to the ability to control the reaction, the
knowledge of physical and chemical properties, kind of reaction
and its condition. Hydrotreating of crude oil is regarded as a big
and difcult challenge since crude oil involves a lot of compounds
and multiple phases, in addition to difcult structures. Additionally, hydrotreating of crude oil in the existence of asphaltenes that
contain a large amount of these impurities, especially metals that
close the active sites on the catalyst is one of the more difcult
and signicant problems. The expected benets of directly hydrotreating crude oil are increasing of middle distillates productivity
due to conversion of heavy compounds and long molecules that

2167

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

is concentrated in heavy fractions to light compounds as a result of

hydrotreating of crude oil before distillation process. In contrast to
conventional processes that are carried out for each fraction separately, which means that the heavy compounds and long molecules
will be deposited at the bottom of the atmospheric and vacuum
distillation column, which is difcult to hydrotreating them using
normal operations and conditions.
Furthermore, the mathematical modelling of the hydrotreating
of crude oil is a hard task in view of the intricate physiochemical
changes that are undergone in the feed together with transport
phenomena and mechanisms of catalyst deactivation in the reaction system, the major challenge being the evaluation of the kinetic
models accurately, which can accurately predict the product compounds at different process conditions. For HDN, HDV and HDNi
reactions, the development of such kinetic models is a hard task
according to the great variety of structures. Thus, this paper is focused upon calculating the parameters of kinetic models applied to
the hydrodenitrogenation, hydrodevanadization and hydrodenickelation of crude oil based on detailed experimental data. The models utilized have taken from the literature and the kinetic
parameters are estimated via minimizing sum of the squared error
between experimental data and model prediction. Finally, the
model is used for simulation of the HDN, HDV and HDNi processes
carried out using gPROMS software [11].

2. Mathematical model of TBR for HDN, HDV and HDNi reactions

A mathematical model is a set of variables and a set of equations that build relationships among the variables for describing
some aspects of the behaviour of the system under investigation.
Process models are very protable. It has been employed for operator training, safety systems design, design of operation as well as
operation control systems designs. The improvement of faster
computer and advanced numerical methods has enabled modelling
and solution of the whole process [12].
In the present study, a threephase heterogeneous model has
been developed for describing the behaviour of pilot-plant trickle
bed reactors applied to the HDN, HDV and HDNi of crude oil. The
model based on two lm theory and includes correlations for calculating mass-transfer coefcients, oil density, Henrys coefcients,
solubility of hydrogen, oil viscosity, diffusivity, molar volume, specic surface area, etc. under the operating conditions, using information presented in the literature [1315].
There are three phases in the reactor: gas phase (hydrogen frequently), liquid phase (feedstock oil) and solid xed particles (catalyst-bed), where the reactions take place. Trickle bed reactor
process is marked by the simultaneous existence of gas and liquid,
over and through a third catalyst solid phase in a cocurrent ow
mode [1618].
Mathematical modelling of HDT process is a hard task due to
the complex physical and chemical changes that the feed undergoes along with the mass transfer phenomena in the reaction system. Kinetic aspects are a major factor of reactor modelling, but in
this case, the conversion of a large amount of nitrogen, vanadium
and nickel compounds made it a huge problem. The following
assumptions were used to create the mathematical models for
HDN, HDV and HDNi processes using TBR:







No radial concentrations gradients.

Steady-state operation of the reactor.
One-dimensional heterogeneous model.
Isothermal and constant pressure operation of the reactor.
Complete wetting of catalyst.
No change of catalyst activity with time (thus the effect of catalyst deactivation on kinetic parameter is negligible).

The required data and available tools with the assumptions for
modelling and simulation processes for crude oil hydrotreating are
tabulated in Fig. 1.
2.1. Mass balance equations
Mass balance equations in the trickle bed reactor for HDN, HDV
and HDNi processes are described with the following set of differential and algebraic equations.
(i) Gas phase
G

Hydrogen :

dP H2

dz

PGH2
RT L
kH2 aL
 C LH2
ug
hH2

!
1

(ii) Liquid phase

"
#
!


PGH2
1 L
S
L
L
S
k aL
 C H2  kH2 aS C H2  C H2

u L H2
dz
hH2
L

Hydrogen :

dC H2

2
The above equations represent the mass balance equations for
the gaseous compounds (H2), while the mass balance equation
for the liquid compounds (nitrogen, vanadium and nickel) can be
written by equating their liquid-phase concentration gradients to
their mass transfer between the liquid-phase and the solid phase.
The mass balance equations can be written as:


dC i
1 S 
i N; Vand Ni:
 ki aS C Li  C Si
uL
dz
L

(iii) Solid phase

The solution of Eqs. (1)(3) need surface concentrations of
hydrogen, nitrogen, vanadium and nickel. At steady-state, the compounds transported between the liquid phase and the solid phase
are consumed or produced through the chemical reaction. By
equating the liquidsolid interfacial mass transfer of H2, N, V and
Ni components with their reaction rates, we get the following
equations:



X
S
kH2 aS C LH2  C SH2 qB
gj rj


S
ki aS C Li  C Si qB gj r j

For H2

4
5

i = N, V and Ni and j = HDN, HDV and HDNi.

The above equations can be solved using the boundary conditions at z = 0 as follows:

PGH2 z 0 P GH2
C Li z 0 C Li

initial
initial

6
7

i = H2, N, V and Ni.

2.2. Chemical reaction rate
Development of kinetic models for hydrotreating of crude oil
reactions is a difcult task owing to the complexities of crude oil
composition and its analysis. Heteroatoms are found in more than
one form in crude oil, for instance, metals compounds can be found
as porphyrine, vanadyl and non-vanadyl, whereas nitrogen compounds are occurred as pyridine, quinoline, isoquinoline, pyrrole, indole and carbazole [19,20]. Each form is described by its own
reactivity and complex reaction ways, which are specic to each
feed. In order to estimate for such complexity of feed, the rate of
chemical reaction is usually lumped into a single power law

2168

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Fig. 1. Required data and available tools for modeling and simulation of HDN, HDV and HDNi reactions.

reaction [21]. The HDN, HDV and HDNi reactions are irreversible under usual operation conditions. HDN, HDV and HDNi reactions are
modelled by the power law models with respect to the concentration of nitrogen, vanadium and nickel and with hydrogen as follows:

r j K j C Si nj C SH2 mj

i = N, V and Ni and j = HDN, HDV and HDNi.

Reaction rate constant for HDN, HDV and HDNi reactions (Kj)
can be determined for each reaction using the Arrhenius equation
as follows:



EAj
K j A0j exp 
RT

GL qL uL

11

The liquidsolid mass transfer coefcients can be calculated by

the Van KrevelenKrekels equation as follows:

K Si
DLi aS


1:8

GL
aS lL

0:5

DLi 8:93  108

2.3. Reactor performance

K LH2 aL
DLH2

 0:4
GL
7

lL

lL
qL DLH2

!0:5
10

!1=3
12

i = H2, N, V and Ni.

In order to determine the liquidsolid and gasliquid mass
transfer coefcients, it is necessary to know the molecular diffusivity of H2, N, V and Ni in the liquid. The diffusivity can be calculated
by a Tyn-Calus equation:

j = HDN, HDV and HDNi.

The trickle bed reactor contains a number of parameters: mass

transfer coefcients, oil density, oil viscosity, solubility of hydrogen, diffusivities, effectiveness factor and others. These parameters
are estimated using the correlations presented in the literature
[15,16,2233] as follows.
The equation used to determine the gasliquid mass transfer
coefcient for H2 is:

lL
qL DLi

t0:267
T
L
l
t0:433
L
i

13

i = H2, N, V and Ni
The molar volume of crude oil (L), H2, N, V and Ni can be calculated by the following equation:

 1:048

ti 0:285 tiC

14

i = crude oil, H2, N, V and Ni.

The critical specic volume of liquid (crude oil) is estimated by
a RiaziDaubert correlation:

tLC 7:5214  103 T meABP 0:2896 q15:6 0:7666 Mw

15

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Henry coefcients of H2 can be obtained from solubility

coefcients:

hH2

V H2
kH2 qL

VP
SP

16

qP

Korsten and Hoffmann [34] have presented the following equation for the solubility of hydrogen in hydrocarbon mixtures:

T
kH2 0:559729  0:42947  103 T 3:07539  103
q20
!
0:835783
2
6
1:94593  10 T
q20 2

qL q DqP  DqT
h
i  P
 0:01
DqP 0:167 16:181  100:0425q 
1000


2
h
i
P
 0:299 263  100:0603q 
1000
h

qB

19

29

Dei

"
!#0:5

n1
qP
n 1 K j Ci
2
Dei

0
h@

1
1 A
1
D1K
DL

20

Glasos equation has used as a generalized mathematical equation for oil viscosity. The equation has the following form:

30

31

i = N, V and Ni and j = HDN, HDV and HDNi.

The tortuosity factor (s) generally has a value of 27 [35].
Generally, the tortuosity factor is assumed to be 4 according to
literature reports [29,35,38].
Knudsen diffusivity factor (DK) is evaluated as follows:

DK 9700rg

DqT 0:0133 152:4q DqP 2:45  T  520

h
i
 8:1  106  0:0622  100:764q DqP  T  5202

28

For j reaction and i compound, Thiele Modulus can be stated as:

17

18

!#0:5

n 1 KC n1
As qP
2
De

1e

VP
/i
SP

The oil density (qL) as a function of temperature and pressure is

estimated by the Standing-Katz equation:

"

2169

0:5
T
Mw

32

2h
Sg q P

33

h qP V g

34

rg

lL 3:141  1010 T  4603:444 log10 APIa

21

To determine the values of gi, the following equation is used

[37,39,40]:

a 10:313log10 T  460  36  447

22

gj

API

141:5
 131:5
sp:gr15:6

2.4. Kinetic parameters of the models

AC 1  e

24

For cylindrical particle

as

2prL
2
41  e
1  e 1  e
r
d
pr 2 L

25

The bed void fraction of the catalyst (e) is calculated by the following equation. This equation has been developed for packed bed
of spheres:

e 0:38 0:0736
41

2 3
2 7
5
 2

 
DR
ds

dc
ds dc Lc
2

The accurate estimation of kinetic parameters of the relevant

reactions scheme are required in order to obtain a useful model,
which can be condently used for reactor design and process optimization. In the model presented above, the reaction orders of
nitrogen, vanadium and nickel compounds (nj), hydrogen compound order (mj), reaction rate constants
(Kj), activation energies
 
(EAj) and pre-exponential factors A0j parameters of Eqs. (8) and
(9) are such signicant parameters for the HDN, HDV and HDNi
processes. The major focus of this paper is to accurately calculate
these parameters.
3. Parameter estimation techniques

26

DR
ds

For cylindrical particles, the equivalent spherical diameter is given by the equation:

"

35

23

The surface area of the particles per unit volume of the bed is
described as:

as

tanh /i
/i

!#1=2
27

The terms of the catalyst effectiveness factor (g) are usually referred to internal diffusion limitations [35]. It was observed that an
increasing in the particle size, the chemical reaction rate decreases.
In the literature, the effectiveness factor has been found to be in
the range of 0.00571 [36,37]. Thieles Modulus (/) is utilized for
calculating the catalyst effectiveness factor (g) because of the particle size of catalyst is small [38]. The generalized Thiele Modulus
for nth-order irreversible reaction is:

Parameter estimation is required to ensure accurate model predictions and good model based decision. It is a key problem in the
improvement of process models either steady state or unsteady
state, and hence is an important issue in both process design and
control. The features and accuracy of the model utilized estimate
the realism with which the actual process can be represented.
Using a suitable model is helpful not only in nding optimum operation conditions of the process and in developing process analysis,
but also in the control strategies design for the system at process
conditions [41]. For the purpose of process optimization, design
of reactor, process control and selection of catalyst, it is important
to develop kinetic models that can accurately predict the concentration of product under process conditions. For complex hydrocarbon mixtures (like crude oil), the development of such kinetic
models is a difcult task owing to the existence of a huge variety
of structures [42].

2170

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Parameters estimation is necessary in several elds of science

and engineering as many physiochemical processes are described
by systems equations with unknown parameters. Recently, the
benets of developing kinetic models for chemical engineers with
accurate parameter calculations have increased owing to the
developed control technologies and optimization of process, which
can apply fundamental models. Estimation of kinetic parameters is
an important and difcult step in the development of models,
Calculations of unknown kinetic parameters can be achieved by
utilizing experimental data. When estimating kinetic parameters
of the models, the goal is to calculate appropriate parameter values
so that errors between experimental and theoretical data (based on
mathematical model) are minimized. On the other hand, the predicted values from the model should match the experimental data
as closely as possible [43]. The scope of parameter estimation techniques is vast, and some of the parameter estimation techniques
are presented here.
Tatiraju and Soroush [44] have used model inversion for parameter estimation that includes a left inverse of the process model
and at each time instant calculates least-squared error estimates
of parameters by using readily available on-line measurements.
The parameter estimation by state estimation technique, is usually
utilized in chemical and biochemical engineering and needs a dynamic model for each of the unknown parameters to be determined [4547]. Kinetic parameters in Calorimetric technique
method are determined via simple mass and energy balances
[4850]. Optimization technique is employed as well for parameter
calculation, where sum of squared errors between the measurements and estimated values is minimized [51,52]. This technique
is very popular and has largely been utilized in the chemical operations to evaluate many kinetic parameters.
In recent years, several optimization techniques have been
developed, related to the parameter estimation problems. Among
them, the regularization methods, the augmented Lagrangian
methods and the level set approaches have been improved in order
to approximate discontinuous parameters and to reduce the sensitivities of the optimization schemes [53]. Linear or non-linear
regressions are utilized to estimate kinetic rate constant from measured rates and concentrations, and several computer programs
are available for this issue [54,55]. Despite the popularity of the
linear programming methods due to their ability to handle several
parameters, they cannot be applied to any operation owing to the
requirements of linear objective function in terms of the adjustable
parameters [55]. Non-linear optimization is the most popular entity and is commonly employed for calculating the best values of
kinetic parameters. Many methods can be employed for non-linear
optimization techniques, such as Stochastic methods [56],
maximum likelihood estimation [57], Newton method [58],
LevenbergMarquardt method [59,60], Genetic algorithm (GA)
[6164], evolutionary algorithm [65], adaptive GA [66], differential
evolution [67] and Successive Quadratic Programming [68] have
been reported widely to obtain the parameter estimation problems. For HDT processes, the LevenbergMarquardt method and
Successive Quadratic Programming (SQP) methods are particularly
suitable [33,38,6973].
In order to evaluate the best values of kinetic parameters in this
study, two approaches have been employed depending on the
nitrogen, vanadium and nickel content in hydrotreated products
under varies operating conditions. They are as follows:
A. Non-linear regression to obtain simultaneously the reaction
orders of nitrogen, vanadium and nickel compound (n),
hydrogen compound order (m) and reaction rate constants
(Kj), then linear regression with Arrhenius equation to estimate

 activation energy (EAj) and pre-exponential factor
A0j for each process.

B. Non-linear regression to determine nj, mj, EAj and

simultaneously for each process.

A0j

Both approaches use the following objective function based on

the minimization of the sum of squared errors (SSE) between
 the

experimental concentrations
of N, V and Ni compound C exp


i;y
and calculated C cal
i;y , in the products:

SSE

N
Data
X

cal 2
C exp
i;y  C i;y 

36

y1

where i = N, V and Ni
3.1. Optimization Problem Formulation for Parameter Estimation
The optimization problem can be described as follows:
Given

Reactor conguration, the feedstock, the catalyst,

reaction temperature, hydrogen pressure and
liquid hourly space velocity
For the rst approach: the reaction order of N, V
and Ni compounds (nj), hydrogen compound order


(mj) and reaction rate constants K 1j ; K 2j ; K 3j for

Optimize

each reaction at different temperatures (335 C,

370 C, 400 C, respectively)
For second approach: the reaction order of N, V
and Ni compound (nj), hydrogen compound order
(mj), the activation energy (EAj) and pre 
exponential factor A0j (for each reaction).
So as to
minimize
Subject
to

The sum of squared errors (SSE).

Process constraints and linear bounds on all
optimization variables in the process.

Mathematically, using the rst approach, the problem can be

presented as:
Min SSE
nj ; mj ; K j j HDN; HDV; HDNi
s:t: f z; xz; xz; uz; v 0; z0 ; zf  model; equality constraints
nLj 6 nj 6 nUj
inequality constraints
mLj 6 mj 6 mUj
K iLj

K ij

K iU
j ;

inequality constraints
i 1; 2; 3

inequality constraints

Table 1
Properties of commercial catalyst (CoMo/c-Al2O3).
Chemical specication
NiO (wt.%)
MoO3 (wt.%)
Na2O (wt.%)
SiO2 (wt.%)
SO2 (wt.%)
Fe (wt.%)
Al2O3

3
15
0.07
1.1
2
0.04
Balance

Physical specication
Form
Pore volume (cm3/g)
Surface area (m2/g)
Mean particle diameter (mm)
Mean particle length (mm)
Bulk density (g/cm3)

Extrude
0.5
180
1.8
4
0.67

2171

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Fig. 2. General scheme of the hydrotreating pilot-plant unit.

Table 2
Experimental data for HDN of crude oil and simulation data of the pilot plant-TBR using two approaches (linear and non-linear regression).
Operating conditions

Experimental results

LHSV
(h1)

P
(MPa)

T
(C)

Nitrogen
(wt.%)  104

Conversion
(%)

Nitrogen (wt.%)-nonlinear
regression  104

Simulated results
Conversion
(%)

Absolute
error%

Nitrogen (wt.%)-linear
regression  104

Conversion
(%)

Absolute
error%

0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5

4
4
4
4
4
4
4
4
4
7
7
7
7
7
7
7
7
7
10
10
10
10
10
10
10
10
10

335
335
335
370
370
370
400
400
400
335
335
335
370
370
370
400
400
400
335
335
335
370
370
370
400
400
400

526.9
713.2
783.5
327.9
512.7
618.9
203.7
362.8
485.3
472.5
649.0
738.1
261.0
449.1
571.9
155.3
305.5
418.6
424.9
605.3
687.9
234.2
418.5
518.3
130.5
274.4
387.1

47.31
28.68
21.65
67.21
48.73
38.11
79.63
63.72
51.47
52.75
35.10
26.19
73.90
55.09
42.81
84.47
69.45
58.14
57.51
39.47
31.21
76.58
58.15
48.17
86.95
72.56
61.29

519.623
698.148
780.410
322.029
518.731
629.256
194.645
369.570
487.651
462.619
650.617
741.917
270.200
461.001
575.914
155.760
313.755
428.905
425.902
618.017
714.779
239.307
423.793
540.139
133.936
279.811
391.446

48.04
30.18
21.19
67.79
48.13
37.07
80.53
63.04
51.23
53.74
34.94
25.81
72.98
53.90
42.41
84.42
68.62
57.11
57.41
38.20
28.52
76.07
57.62
45.99
86.61
72.20
60.85

1.38
2.11
0.39
1.79
1.18
1.67
4.44
1.87
0.48
2.10
0.25
0.52
3.52
2.65
0.70
0.29
2.70
2.46
0.23
2.10
3.91
2.18
1.26
4.21
2.63
1.97
1.12

515.199
695.499
778.546
317.414
515.319
626.614
190.118
365.107
483.610
461.368
650.718
742.335
268.583
460.959
576.447
153.690
312.679
428.413
426.672
620.101
716.934
239.331
425.896
542.859
133.082
280.665
393.177

48.48
30.45
22.14
68.26
48.47
37.34
80.99
63.49
51.64
53.86
34.93
25.77
73.14
53.90
42.35
84.63
68.73
57.16
57.33
37.99
28.31
76.07
57.41
45.71
86.69
71.93
60.68

2.22
2.48
0.63
3.20
0.51
1.25
6.67
0.63
0.35
2.35
0.26
0.57
2.90
2.64
0.79
1.04
2.35
2.34
0.42
2.44
4.22
2.19
1.77
4.74
1.98
2.28
1.57

Model prediction

1.0
0.75
1.25
0.5
1.5

P
(MPa)

T
(C)

Nitrogen content-wt.%  104 (using 2nd approach)

Conversion%

10
10
4
7
5.5

385
400
370
385
335

347.610
211.025
580.986
206.803
759.106

65..24
78.89
41.90
79.32
24.09

2172

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Using the second approach, the problem can be expressed as:

problem and is solved using a Successive Quadratic Programming

(SQP) method within gPROMS software.

Min SSE
0

nj ; mj ; EAj ; Aj
s:t:

4. Experimental work

j HDN; HDV; HDNi

f z; xz; xz; uz; v 0; z0 ; zf  model; equality constraints

nLj 6 nj 6 nUj
inequality constraints
mLj 6 mj 6 mUj

inequality constraints

L
EAj

inequality constraints

6 EAj 6

U
EAj

0
0U
A0L
j 6 Aj 6 Aj

inequality constraints

f z; xz; 
xz; uz; v 0, represents the process model presented in
Section 2, where z is the independent variable (length of the bed
reactor), x(z) gives the set of all differential and algebraic variables,
xz denotes the derivative of differential variables with respect to
length of the bed reactor, u(z) is the control variables, and v represents the design variables (length independent constant parameters). The length interval of interest is [z0, zf] and the function f: is
assumed to be continuously differentiable with respect to all its
arguments [74,75].
The solution method of optimization employed by gPROMS is a
two-step method known as feasible path approach. The rst step
performs the simulation to converge all the equality constraints
(described by f) and to satisfy the inequality constraints. The second step performs the optimization (updates the values of the
decision variables such as the kinetic parameters) [76]. The optimization problem is posed as a Non-Linear Programming (NLP)

4.1. Materials
Iraqi crude oil has been used as a feed for hydrotreating studies.
It contains 2.0 wt.% of sulfur, 0.1 wt.% of nitrogen, 26.5 ppm of
vanadium, 17 ppm of nickel and 1.2 wt.% of asphaltene. The catalyst used for the HDN, HDV and HDNi processes in this work was
the commercial cobaltmolybdenum on alumina (CoMo/cAl2O3) type catalyst. The properties of the catalyst used are listed
in Table 1.
4.2. Equipment and procedure
A schematic diagram of the hydrotreating pilot plant is shown
in Fig. 2. Generally, the pilot plant can be divided into four sections: the feed section, the reactor section, the products sections
and gases section.
The feed supply module primarily includes a liquid feed tank
and a feed pump. A cylindrical tank with a capacity of 2 l of the
feedstock is the feed tank. In order to introduce the feed oil into
the reactor, a high-pressure dosing pump has employed for this
purpose. The feedstock and hydrogen passes through the reactor
in a concurrent ow mode. The length and the diameter of the
reactor were 2 cm and 65 cm, respectively, and the reactor tube

Table 3
Experimental data for HDV of crude oil and simulation data of the pilot plant-TBR using two approaches (linear and non-linear regression).
Operating conditions

Experimental results

Simulated results

LHSV
(h1)

P
(MPa)

T
(C)

Vanadium
(ppm)

Conversion
(%)

Vanadium(ppm)-nonlinear
regression

Conversion
(%)

Absolute
error%

Vanadium (ppm)-linear
regression

Conversion
(%)

Absolute
error%

0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5

4
4
4
4
4
4
4
4
4
7
7
7
7
7
7
7
7
7
10
10
10
10
10
10
10
10
10

335
335
335
370
370
370
400
400
400
335
335
335
370
370
370
400
400
400
335
335
335
370
370
370
400
400
400

8.33
14.34
17.79
5.54
11.70
15.50
3.21
8.40
12.10
5.60
11.80
15.10
3.15
8.66
12.50
1.59
5.46
9.13
3.84
9.42
13.23
1.98
6.63
10.27
0.891
3.95
6.89

68.57
45.89
32.87
79.09
55.85
41.51
87.89
68.30
54.34
78.87
55.47
43.02
88.11
67.32
52.83
94.00
79.40
65.55
85.51
64.45
50.00
92.53
74.98
61.24
96.64
85.10
74.00

8.6997
14.8727
17.9405
5.4227
11.5011
15.0354
3.3230
8.7468
12.4326
5.6621
11.7733
15.2785
3.0128
8.2686
11.9514
1.5874
5.7026
9.1982
3.9996
9.7013
13.3593
1.8883
6.3105
9.8807
0.8908
4.0329
7.1805

67.17
43.88
32.30
79.54
56.60
43.26
87.46
66.99
53.08
78.63
55.57
42.34
88.63
68.80
54.90
94.01
78.48
65.29
84.91
63.39
49.59
92.87
76.19
62.27
96.64
84.78
72.90

4.44
3.71
0.84
2.12
1.70
2.99
3.52
4.13
2.75
1.11
0.23
1.18
4.35
4.52
4.39
0.16
4.44
0.75
4.16
2.99
0.98
4.63
4.82
3.79
0.02
2.10
4.22

8.7776
14.9471
18.0027
5.4468
11.5347
15.0676
3.3159
8.7436
12.4320
5.7592
11.8885
15.3831
3.0503
8.3341
12.0203
1.5954
5.7279
9.2308
4.0972
9.8389
13.4929
1.9251
6.3899
9.9718
0.9009
4.0692
7.2312

66.88
43.60
32.06
79.45
56.47
43.14
87.49
67.00
53.09
78.27
55.14
41.95
88.49
68.55
54.64
93.98
78.38
65.17
84.54
62.87
49.08
92.73
75.89
62.37
96.60
84.64
72.71

5.37
4.23
1.19
1.68
1.41
2.79
3.30
4.09
2.74
2.84
0.75
1.87
3.16
3.76
3.84
0.34
4.91
1.10
6.70
4.45
1.99
2.77
3.62
2.90
1.11
3.02
4.95

Model prediction

1.0
0.75
1.25
0.5
1.5

P
(MPa)

T
(C)

Vanadium content-ppm (using 2nd approach)

Conversion%

10
10
4
7
5.5

385
400
370
385
335

5.0913
2.3578
13.4918
2.2116
16.4829

80.79
91.10
49.09
91.65
37.80

2173

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

was made of stainless steel. The length of the reactor has been
divided into three sections. The rst section, having a length of
20 cm, was packed with inert particles (glass beads with 4 mm
diameter). This entrance section has been used to heat up the mixture to the required temperature, to ensure homogeneous ow distribution of gas and liquid and to avoid end effects. The following
section with a length of 27.8 cm contained a packing of 60.3 g catalyst. The bottom part (17.2 cm) was packed with inert particles to
ensure to serve as disengaging section. The reactor was operated in
isothermal mode by independent temperature control of ve zone
electric furnaces, which provided an isothermal temperature along
the active reactor section.
The product part includes of low and high gasliquid separator
and products storage tank. The reactor outlet is led to the high
pressure separator where the liquid and gas are separated. Finally,
the gases section where the gas is exiting from is passed through a
gas ow meter before being released.
Before starting up any run, a leak test must be conducted. The
leak test is done with nitrogen (N2) at 13 MPa for 12 h. Once the
unit passes the leak tests, the catalyst presulding process (see further details on experimental procedure and presulding in Jarullah
et al. [77]) will start. Calibration has been carried out on all pilot
plant equipments (such as pump, instrumentation and control)
for ensuring the measurements accuracy.
4.3. Experimental runs
The main hydrotreating reactions in this work are hydrodenitrogenation (HDN) and hydrodemetallization (HDM) that includes

hydrodevanadization (HDV) and hydrodenickelation (HDNi) reactions. The data obtained from these experiments were used in
the development of models that can represent the HDN, HDV and
HDNi reactions to determine kinetic parameters and to validate
the model under different operating conditions.
Note, all analytical techniques that have been used for the specications of the feedstock and the products were accurate, fast and
repeatable. Product analysis has been repeated twice for each run
at each operating condition for ensuring the accuracy of the results.
Average results have been taken into considerations for each sample with maximum deviation of 2% among all runs. IP-285 method
was used to calculate the vanadium and nickel content in the feedstock and the products samples. While the nitrogen content in the
feedstock and products were estimated by using ASTM: D-4629
method.

5. Results and discussions

5.1. Experimental results
Experimental works for hydrodenitrogenation, hydrodevanadization and hydrodenickelation of crude oil were carried out under
the following operating conditions:





Reaction temperature: 335400 C.

Liquid hourly space velocity (LHSV): 0.51.5 h1.
Hydrogen pressure: 410 MPa.
H2/Oil ratio: 250 l/l.

Table 4
Experimental data for HDNi of crude oil and simulation data of the pilot plant-TBR using two approaches (linear and non-linear regression).
Operating conditions

Experimental results

Simulated results

LHSV
(h1)

P
(MPa)

T
(C)

Nickel
(ppm)

Conversion
(%)

Nickel (ppm)-nonlinear
regression

Conversion
(%)

Absolute
error%

Nickel (ppm)-linear
regression

Conversion
(%)

Absolute
error%

0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5

4
4
4
4
4
4
4
4
4
7
7
7
7
7
7
7
7
7
10
10
10
10
10
10
10
10
10

335
335
335
370
370
370
400
400
400
335
335
335
370
370
370
400
400
400
335
335
335
370
370
370
400
400
400

2.01
4.80
6.80
1.11
3.10
4.95
0.73
2.28
3.85
1.15
3.22
5.10
0.62
2.02
3.50
0.36
1.33
2.48
0.81
2.50
4.07
0.42
1.48
2.71
0.24
0.93
1.87

88.18
71.76
60.00
93.47
81.76
70.88
95.70
86.59
77.35
93.23
81.06
70.00
96.35
88.12
79.41
97.88
92.18
85.41
95.23
85.29
76.06
97.53
91.29
84.06
98.59
94.53
89.00

1.9332
4.6836
6.7673
1.1222
3.2016
5.0490
0.6948
2.2498
3.8010
1.1526
3.2596
5.1186
0.6232
2.0689
3.5699
0.3673
1.3708
2.5508
0.7995
2.4918
4.1416
0.4153
1.5075
2.7584
0.2380
0.9624
1.8978

88.63
72.45
60.19
93.40
81.17
70.30
95.91
86.76
77.59
93.22
80.82
69.89
96.33
87.83
79.00
97.84
91.94
84.99
95.30
85.34
75.64
97.56
91.13
83.77
98.60
94.34
88.84

3.82
2.42
0.48
1.10
3.28
2.00
4.82
1.32
1.27
0.23
1.23
0.36
0.52
2.42
1.99
2.03
3.07
2.85
1.29
0.33
1.76
1.12
1.86
1.78
0.83
3.48
1.49

1.9001
4.6368
6.7195
1.1128
3.1903
5.0393
0.6950
2.2577
3.8353
1.1318
3.2251
5.0804
0.6183
2.0635
3.5668
0.3680
1.3784
2.5667
0.7847
2.4649
4.1103
0.4123
1.5046
2.7583
0.2388
0.9690
1.9127

88.82
72.72
60.47
93.45
81.23
70.36
95.91
86.72
77.44
93.34
81.03
70.11
96.36
87.86
79.02
97.83
91.89
84.90
95.38
85.50
75.82
97.57
91.15
83.77
98.59
94.30
88.75

5.47
3.40
1.18
0.25
2.91
1.80
4.79
0.98
0.38
1.58
0.16
0.38
0.27
2.15
1.91
2.22
3.64
3.49
3.12
1.40
0.99
1.83
1.66
1.78
0.50
4.19
2.28

Model prediction

1.0
0.75
1.25
0.5
1.5

P
(MPa)

T
(C)

Nickel content-ppm (using 2nd approach)

Conversion %

10
10
4
7
5.5

385
400
370
385
335

1.2064
0.5562
4.1712
0.4783
5.8171

92.90
96.73
75.46
97.19
65.78

2174

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Nitrogen content in product (ppm)

900

(a)

800
700
600
500
400
300

T=608K
T=643K
T=673K

200
100

Simulated
Simulated
Simulated

0
0.4

0.6

0.8

1.2

1.4

1.6

-1

Nitrogen content in product (ppm)

LHSV (hr )
800.4

(b)

700.4
600.4
500.4
400.4
300.4
200.4

T=608K
T=643K
T=673K

100.4

Simulated
Simulated
Simulated

0.4
0.4

0.6

0.8

1.2

1.4

1.6

Nitrogen content in product (ppm)

LHSV (hr-1)
800

(c)

700
600
500
400
300
200
100

T=608K

Simulated

T=643K

Simulated

T=673K

Simulated

0
0.4

0.6

0.8

1.2

1.4

1.6

LHSV (hr-1)
Fig. 3. Experimental data (points) and simulated (lines) variation of outlet nitrogen content vs. liquid hourly space velocity at different reactor temperature and at pressure
(a) 4 MPa, (b) 7 MPa, (c) 10 MPa.

The experimental results for HDN, HDV and HDNi processes in

tabular form are shown in Tables 24 (also in Figs. 35 with model
predictions).
It has been observed from the experimental results (Tables 24,
also in Figs. 35) that the nitrogen, vanadium and nickel content in
all products decreased with increasing in temperature and pressure and decreasing in liquid hourly space velocity. Similar attitude
has also been found by several investigations for HDN, HDV and
HDNi processes using different oily feedstocks (but not on the full
crude oil) [6,33,7880].
The increasing in nitrogen and metals removal at high temperature may be attributed to several reasons: at high reaction temperature, the unreactive nitrogen and metal compounds or the
compounds containing these impurities become activated enough

to react with hydrogen. Also, the large molecules are decomposed

into smaller molecules, which can more easily diffuse inside the
catalyst pores and reach the inner active sites where the HDT reactions occur. Oil diffusivity increases through the catalyst pores due
to decreases in the oil viscosity. Furthermore, the increase in temperature raises the activation energy leading to increase the number of particles of these compounds interacted. As a result, the
long-nitrogen and metal compounds will cleavage and spread
within the catalyst [6,81]. As liquid hourly space velocity decreased, denitrogenation and demetallization increase because of
the contact time (residence time) increases between the molecules
of reactants and catalyst, and provide sufcient time for the reaction process [6,82,83]. Whereas the reason for increasing in nitrogen, vanadium and nickel removal by increasing in hydrogen

2175

Vanadium content in product (ppm)

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

20

(a)

18
16
14
12
10
8

T=608K
T=643K
T=673K

6
4

Simulated
Simulated
Simulated

2
0
0.4

0.6

0.8

1.2

1.4

1.6

Vanadium content in product (ppm)

LHSV (hr-1)
18

(b)

16
14
12
10
8
6
4
2

T=608K

Simulated

T=643K
T=673K

Simulated
Simulated

0
0.4

0.6

0.8

1.2

1.4

1.6

Vanadium content in product (ppm)

LHSV (hr-1)
16
14

(c)

12
10
8
6
4

T=608K
T=643K
T=673K

Simulated
Simulated
Simulated

0
0.4

0.6

0.8

1.2

1.4

1.6

-1

LHSV (hr )
Fig. 4. Experimental data (points) and simulated (lines) variation of outlet vanadium content vs. liquid hourly space velocity at different reactor temperature and at pressure
(a) 4 MPa, (b) 7 MPa, (c) 10 MPa.

pressure due to the contact between the hydrogen and hydrocarbons and the catalyst [79].
It is interesting to note that the conversion of HDNi is found to
be higher than HDV (Tables 3 and 4). These observations although
not reported in the public domain, are not uncommon in industries
[84,85]. This could be due to varying composition and properties of
different crude oils and due to variation in the actual amount of Ni
and V present in these crude oils.
5.2. Estimation of kinetic parameters
The kinetic parameters for crude oil hydrodenitrogenation, hydrodevanadization and hydrodenickelation presented in the present

work have been estimated depending on the experimental data

using TBR model.
In the rst approach, the reaction order of N, V and Ni
(nj, j=HDN, HDV, HDNi
 ), hydrogen order (mj) and reaction rate constants
K ij ; i 1; 2; 3 for each reaction were estimated simultaneously.
Linearization process is then used for estimating
  the activation energy (EAj) and the pre-exponential factor A0j for each reaction. To
estimate activation energies and pre-exponential factors for each
reactions, the Arrhenius equation described previously (Eq. (9)) is
used for this purpose. The Arrhenius-based dependence of the kinetic model is demonstrated in Fig. 6 for all processes. Plot of ln Kj
versus 1/T gives a straight line with a slope equal to EAj/R and
intercept equal to ln A0j . In the second approach, the activation

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

Nickel content in product (ppm)

2176

8
7

T=608K
T=643K
T=673K

Simulated
Simulated
Simulated

5
4
3
2

(a)

1
0
0.4

0.6

0.8

1.2

1.4

1.6

-1

LHSV (hr )

Nickel content in product (ppm)

6
T=608K
T=643K
T=673K

Simulated
Simulated
Simulated

4
3
2

(b)

1
0
0.4

0.6

0.8

1.2

1.4

1.6

-1

LHSV (hr )

Nickel content in product (ppm)

4.5
T=608K
T=643K
T=673K

4
3.5

Simulated
Simulated
Simulated

3
2.5
2
1.5
1

(c)

0.5
0
0.4

0.6

0.8

1.2

1.4

1.6

-1

LHSV (hr )
Fig. 5. Experimental data (points) and simulated (lines) variation of outlet nickel content vs. liquid hourly space velocity at different reactor temperature and at pressure (a)
4 MPa, (b) 7 MPa, (c) 10 MPa.

 
energies (EAj), pre-exponential factors A0j , reaction order of N, V
and Ni (nj) and hydrogen order (mj) were determined simultaneously. The generated kinetic parameters for HDN, HDV and HDNi
processes are presented in Table 5 for both approaches,
respectively.
It is has been observed from Table 5 that the second approach
gives sum of squared errors (SSE) less than rst approach in all
reactions. It can be concluded depending on the objective function
(SSE) that parameter estimation with the second approach is more
accurate. In other words, determine of activation energy and preexponential factor by linearization process of Arrhenius equation
gives higher error in comparison to those obtained with simultaneous estimation of kinetic parameters via non-linear regression.
The best reaction order of nitrogen, vanadium and nickel were

1.6723, 1.2514 and 1.6884, respectively, while, the best order of

H2 for HDN, HDV and HDNi were 0.3555, 0.6337 and 0.5667,
respectively, which is typical for lumped kinetics. Several investigators have extensively studied the reaction kinetics of hydroprocessing processes of several distillate cuts (not of crude oil which is
the focus of this work) and showed that most HDT reactions follows half to second order kinetics and zero to one order kinetics
for hydrogen [15,30,33,78,80,86,87]. Note, Jarullah et al. [88] had
a similar observation in estimating kinetic parameters for hydrodesulfurization of crude oil.
The established kinetic parameters for HDN, HDV and HDNi
reactions in the present study, has been applied to simulate the
performance of a pilot plant trickle bed reactor. The partial pressure of hydrogen, molar concentration proles of N, V and Ni and

2177

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

13

5
4.5

11

lnk (HDN & HDNi)

3
7

2.5
2

lnk (HDV)

3.5

1.5
1

0.5
1

1.46

1.48

1.5

1.52

1.54

1.56

1.58

1.6

1.62

1.64

1.66

1000/T (K)
Fig. 6. Linear representation of Arrhenius equation for HDN, HDV and HDNi of crude oil.

Table 5
Comparison of kinetic parameter values estimated with two approaches for HDN,
HDV and HDNi models.
HDV

HDNi

First approach (linear)

9.06245
Ki; i 1

HDN

11.6142

51827.2

K ij ; i 2

20.0872

17.5322

77768.1

K ij ; i 3

34.3759

28.8485

103539

m
n
EAj
A0j

0.3325
1.6302
69973.95
9435596.91

0.6267
1.2482
47172.80
127388.9139

0.5633
1.6819
36288.95
0.683  108

SSE

3.5442  106

0.24521638

0.01000841

0.6337
1.2514
46181.6
126566.0

0.5667
1.6884
37678.3
1.045  108

0.2225156

0.007773

Second approach (non-linear)

m
0.3555
n
1.6723
EAj
71775.5
2.85  107
A0
j

2.8957  106

800
700

Nitrogen (ppm)- Simulated

SSE

H2 in the liquid phase and in the solid phase were calculated using
the correlations given earlier, which contain a number of
parameters.
A comparison between experimental results and model prediction results for HDN, HDV and HDNi of crude oil were plotted in
Figs. 35 (using second approach, where the experimental data
are represented in the form of points; while the simulation results
are represented in the form of curves (each curve representing
three simulated points)), and given in Tables 24 (for both approach). As can be noticed from the results, the model was found
to simulate the performance of the pilot plant TBR very good agreement in the range of operating conditions studied between both
concentrations with average absolute error less than 5% by using
second approach. It has also been noted from these gures that
the nitrogen, vanadium and nickel removal increase with increasing in temperature, pressure and decreasing in liquid hourly space
velocity. These increases happened due to the kinetics parameters
used to describe HDN, HDV and HDNi processes in this model are
affected by the operating conditions. The reaction temperature of

600
500
400
300
200
100
100

200

300

400

500

600

700

Nitrogen (ppm)- Experimental

Fig. 7. Comparison between experimental and calculated concentrations of nitrogen.

800

2178

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

the reactor impacts upon the mass velocity of the gases and liquids,
diffusivities of the components, mass transfer coefcient at the
gasliquid and at the liquidsolid interfaces, solubility and Henrys
coefcients of hydrogen in addition to viscosity and density of the
compounds.
The temperature also inuences the rate constant of HDN, HDV
and HDNi processes. Increasing the reaction temperature lead to an
increase in reaction rate constants dened by the Arrhenius equations. As a result, the reaction rates of these reactions will increase.
These results supported the fact that the operating temperature is
very effective for enhancing the degree of denitrogenation and
demetallization.
Liquid hourly space velocity (LHSV) is also a signicant operational factor that calculates the severity of reaction and the efciency of hydrotreating. As the liquid hourly space velocity
decreased, the quantity of the reactions rates will be signicant.
On the other hand, decreasing liquid hourly space velocity described by liquid velocity, means increasing the residence time
and hence the reaction severity will increase [6,16,83,89].
In addition, the hydrogen partial pressure has an effect on the
reactions used in this study. The mechanisms utilized to describe
HDN, HDV and HDNi reactions used a kinetic equation with the
order of the hydrogen concentration at the catalyst surface less than
1. Therefore, conversion of N, V and Ni compounds increases with

pressure. The effect of pressure above 100 atm can be neglected

owing to the viscosity of the oil feedstock increase with pressure,
and the diffusivity and mass transfer coefcient will decrease with
the pressure. Hence, at high operating pressures, the pressure impact upon N, V and Ni conversions becomes important [34,83,90].
Figs. 79 show a parity plot of the model for HDN, HDV and
HDNi reactions studied in this study (each point represents simulated (Y-axis) and experimental (X-axis) values at the same time
with the same operating conditions for each point). The plot between the experimental results and simulated nitrogen, vanadium
and nickel contents in all products appears to be straight line with
a slope close to 1.0, indicating very good agreement between the
measurement results and the predicted. The model can be used
to describe the behaviour of the pilot plant trickle bed reactor at
different operating conditions for which experimental data are
not available.
5.3. Simulation of the HDN, HDNi and HDV Pilot Plant Reactor
The model developed now can be applied for describing the
behaviour of the pilot plant trickle bed reactor at different
operating conditions for which experimental data are not available.
Tables 24 show model predictions in terms of nitrogen conversion, vanadium conversion and nickel conversion at operating

20

Vanadium (ppm)- Simulated

18
16
14
12
10
8
6
4
2
0
0

10

12

14

16

18

20

Vanadium (ppm)- Experimental

Fig. 8. Comparison between experimental and calculated concentrations of vanadium.

Nickel (ppm)- Simualted

7
6
5
4
3
2
1
0
0

Nickel (ppm)- Experimental

Fig. 9. Comparison between experimental and calculated concentrations of nickel.

2179

6.57E-04

10.00001

6.57E-04

10
9.99999

6.57E-04

9.99998
6.57E-04

Liquid phase
Solid phase
Gas phase

6.57E-04

9.99997
9.99996

6.57E-04

9.99995

6.57E-04

H2 concentration (Mpa)

H2 concentration (mol/cm3)

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

9.99994
9.99993

6.57E-04
0

10

15

20

25

30

Reactor bed length (cm)

Fig. 10. Concentration proles of H2 in liquid, solid and gas phase down through the reactor.

9.0E-08

3.5E-06

N concentration (mol/cm3)

N in liquid phase
N in soild phase
V in liquid phase
V in solid phase
Ni in liquid phase
Ni in solid phase

2.5E-06

7.0E-08
6.0E-08

2.0E-06

5.0E-08

1.5E-06

4.0E-08
3.0E-08

1.0E-06
2.0E-08
5.0E-07

V and Ni concentration (mol/cm3)

8.0E-08

3.0E-06

1.0E-08

0.0E+00
0

10

15

20

25

0.0E+00
30

Reactor bed length (cm)

Fig. 11. Concentration proles of N, V and Ni in liquid and solid phase down through the reactor.

conditions other than experimental conditions. Using the model,

the concentration prole of hydrogen in the gas, liquid and solid
phases generated under maximum process condition (T = 400 C,
P = 10 MPa and LHSV = 0.5 h1) is presented in Fig. 10. It is noticed
that the hydrogen partial pressure in gas phase decreased along
the catalyst bed length as a result of hydrogen consumption.
Whereas, the concentration prole of hydrogen in the liquid phase
and solid phase increased a long the catalyst bed length. As is well
known, their forms are estimated by a balance between chemical
reaction and gasliquid mass transfer, decreased initially because
of the high reaction rate in this part of the reactor and then increased substantially along the reactor bed length. This behaviour
can be attributed to the difference in mass transfer rate at gas
liquid and liquidsolid, and reaction kinetics. When mass transfer
at a liquidsolid interface becomes predominant, the H2 concentration decreases in both the solid and liquid phases, and when the
mass transfer from liquid to gas becomes important, the liquid
phase concentration also solid phase concentration increases
[15,16,33,91].

The concentration proles of N, V and Ni along the catalyst bed

length in the liquid and solid phase at maximum conditions is presented in Fig. 11. As can be seen from this gure, the concentration
prole of these compounds reduced in both liquid phase and solid
surface along the reactor bed length. In addition, there is a concentration gradient between both phases. This gradient is governed by
liquidsolid mass transfer rate calculated from the equations used
in this model, which is based mainly on the physical properties of
the liquid, such as density and viscosity, and also liquid mass
velocity. Therefore, the feedstock becomes lighter and thus physical properties are improved and mass transfer of liquidsolid will
enhance reducing this concentration gradient [15,16].

6. Conclusions
Estimation of the kinetic parameters in trickle bed reactor applied for HDN, HDV and HDNi of crude oil is required to develop

2180

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

an accurate model, so that the model can be effectively used for

simulation, optimization and control.
The kinetic parameters estimations of trickle bed reactor model
for HDN, HDV and HDNi reactions of crude oil have been calculated
using pilot plant experimental data and an optimization technique.
In order to evaluate the best values of kinetic parameters, two approaches, linear and non-linear regression have been applied for
estimating the best values of kinetic parameters of trickle bed reactor process for HDN, HDV and HDNi of crude oil. In the rst approach, the reaction order of N, V and Ni (nj), hydrogen order
(mj) and reaction rate constants K ij for each reaction were estimated simultaneously. Linearization process is then applied to
estimate
the activation energy (EAj) and the pre-exponential factor


the activation enerA0j for each reaction. In the second
 approach,

gies (EAj), pre-exponential factors A0j , reaction order of N, V and
Ni (nj) and hydrogen order (mj) were determined simultaneously.
Based on the objective function (SSE), the parameters estimated
with the second approach is found to be more accurate and the
simulation results showed very well correspondence with the
experimental data with an average absolute error of less than 5%.
The effect of reactor temperature (T), partial pressure of hydrogen (P) and liquid hourly space velocity (LHSV) upon the N, V and
Ni conversion and upon the concentration proles along the reactor bed length were studied using the process model. It has been
observed that the inuence of these operating conditions in HDN,
HDV and HDNi of crude oil conrming that high temperature, pressure and low liquid hourly space velocity improve the nitrogen,
vanadium and nickel conversion. The model can now be condently applied to reactor design, operation and control, as well as
to predict the concentration proles of any compound at any
conditions.
Finally note, sulfur and asphaltene are also regarded as important contaminants in crude oil. Studying the effect of these impurities were beyond the scope of this paper, but have been
reported elsewhere.
References
[1] Rana MS, Samano V, Ancheyta J, Diaz JAI. A review of recent advances on
process technologies for upgrading of heavy oils and residua. Fuel
2007;86:121631.
~ eda LC, Alonso F,
[2] Ancheyta J, Betancourt G, Marroqun G, Centeno G, Castan
et al. Hydroprocessing of Maya heavy crude oil in two reaction stages. Appl
Catal A 2002;233:15970.
[3] Speight JG. The desulfurization of heavy oils and residues. New York: Marcel
Dekker, Inc; 2000.
[4] Andari MK, Behbehani H, Stanislaus A. Sulfur compound type distribution in
Naphtha and gas oil fractions of Kuwaiti crude. Fuel Sci Technol Int
1996;14:93961.
[5] Kaernbach W, Kisielow W, Warzecha L, Miga K, Klecan R. Inuence of
petroleum nitrogen compounds on hydrodesulphurization. Fuel 1990;69:
2214.
[6] Abbas AS. Low sulfur feedstock from basrah reduced crude oil for coke
production. M.SC. Thesis. Iraq: University of Baghdad; 1999.
[7] Bartholdy J, Cooper BH. Metal and coke deactivation of resid hydroprocessing
catalysts. ACS symposium on resid upgrading Denver, USA: CO; 1993. p. 39686.
[8] Pereira CJ, Cheng JW, Suarez WC. Metal deposition in hydrotreating catalyst.
Ind Eng Chem Process Des Dev 1990;29:5201.
[9] Ali LH, Abdul-Karim E. The oil, origin, composition and technology. Iraq:
AlMosul University; 1986.
[10] Gary JH, Handwerk GE. Petroleum rening: technology and economics. 3rd
ed. New York: Marcel Dekker; 1994.
[11] gPROMS. Introductory user guide. Process System Enterprise Ltd (PSE). http://
www.psenterprise.com/gproms/; 2005.
[12] Khalfhallah HA. Modelling and optimization of oxidative desulfurization
process for model sulfur compounds and heavy gas oil. Ph.D. thesis. UK:
University of Bradford; 2009.
[13] Froment GF, Depauw GA, Vanrysselberghe V. Kinetic modeling and reactor
simulation in hydrodesulfurization of oil fractions. Ind Eng Chem Res
1994;33:297588.
[14] Jimenez F, Nunez M, Kafarov V. Modeling of industrial reactor for
hydrotreating of vacuum gas oils Simultaneous hydrodesulfurization,
hydrodenitrogenation and hydrodearomatization reactions. Chem Eng J
2007;134:2008.

[15] Mederos FS, Ancheyta J. Mathematical modeling and simulation of

hydrotreating reactors: cocurrent versus countercurrent operations. Appl
Catal A: General 2007;332:821.
[16] Bhaskar M, Valavarasu G, Meenakshisundaram A, Balaraman KS. Petrol Sci
Technol 2002;20:25168.
[17] Avraam DG, Vasalos IA. HdPro: a mathematical model of trickle-bed reactors
for the catalytic hydroprocessing of oil feedstocks. Catal Today 2003;79
80:27583.
[18] Shokri S, Marvast MA, Tajerian M. Production of ultra low sulfur diesel:
simulation and software development. Petrol Coal 2007;49:4859.
[19] Wauquier JP. Crude oil petroleum products process owsheets. Paris: Institute
Francais du Petrole; 1995.
[20] Mara A, Fukase S, Al-Marri M, Stanislaus A. A comparative study of the effect
of catalyst type on hydrotreating kinetics of Kuwaiti atmospheric residue.
Energy Fuels 2003;17:6618.
[21] Lababidi HMS, Shaban HI, Al-Radwan S, Alper E. Simulation of an atmospheric
residue desulfurization unit by quasi-steady state modeling. Chem Eng
Technol 1998;21:193200.
[22] Mears DE. The role of axial dispersion in trickle-ow laboratory reactors. Chem
Eng Sci 1971;26:13616.
[23] Carberry JJ. Chemical and catalytic reaction engineering. New York: McGrawHill; 1976.
[24] Shah YT. Gasliquidsolid reactors design. New York: McGraw-Hill Inc.; 1979.
[25] Raid RC, Prausnitz JM, Poling BE. The properties of gases & liquids. 4th ed. New
York: McGraw-Hill; 1987.
[26] Ahmed T. Hydrocarbon phase behaviour. Houston: Gulf Publishing; 1989.
[27] Froment GF, Bischoff KB. Chemical reactor analysis and design. 2nd ed. New
York: Wiley; 1990.
[28] Dudukovic MP, Larachi F, Mills PL. Multiphase catalytic reactors: a perspective
on current knowledge and future trends. Catal Rev 2002;44:123246.
[29] Macas MJ, Ancheyta J. Simulation of an isothermal hydrodesulfurization small
reactor with different catalyst particle shapes. Catal Today 2004;98:24352.
[30] Rodriguez MA, Ancheyta J. Modeling of hydrodesulfurization (HDS),
hydrodenitrogenation (HDN), and the hydrogenation of aromatics (HDA) in a
vacuum gas oil hydrotreater. Energy Fuels 2004;18:78994.
[31] Shokri S, Zarrinpashne S. A mathematical model for calculation of effectiveness
factor in catalyst pellets of hydrotreating process. Petrol Coal 2006;48:2733.
[32] Mederos FS, Rodrguez MA, Ancheyta J, Arce E. Dynamic modeling and
simulation of catalytic hydrotreating reactors. Energy Fuels 2006;20:93645.
[33] Alvarez A, Ancheyta J. Modeling residue hydroprocessing in a multi-xed-bed
reactor system. Appl Catal A: General 2008;351:14858.
[34] Korsten H, Hoffmann U. Three-phase reactor model pilot trickle-bed for
hydrotreating in reactors. AIChE J 1996;42:135060.
[35] Sattereld CN. Trickle-bed reactors. AIChE J 1975;21:20928.
[36] Scamangas A, Papayannakos N, Marangozis J. Catalytic hydrodesulfurization of
a petroleum residue. Chem Eng Sci 1982;37:18102.
[37] Li C, Chen W, Tsai C. Highly restrictive diffusion under hydrotreating reactions
of heavy residue oils. Ind Eng Chem Res 1995;34:898905.
[38] Marroquin G, Ancheyta J, Esteban C. A batch reactor study to determine
effectiveness factors of commercial HDS catalyst. Catal Today 2005;104:705.
[39] Aris R. The mathematical theory of diffusion and reaction in permeable
catalysts. Oxford: Clarendon Press; 1975.
[40] Chang J, Liu J, Li D. Kinetics of resid hydrotreating reactions. Catal Today
1998;43:2339.
[41] Gau Ch, Stadtherr MA. Reliable nonlinear parameter estimation using interval
analysis: error-in-variable approach. Comput Chem Eng 2000;24:6317.
[42] Khorasheh F, Chan EC, Gary MR. Development of a continuous kinetic model
for catalytic hydrodesulfurization of bitumen. Petrol Coal 2005;47:408.
[43] Poyton AA, Varziri MS, McAuley KB, McLellan PJ, Ramsay JO. Parameter
estimation in continuous-time dynamic models using principal differential
analysis. Comput Chem Eng 2006;30:698708.
[44] Tatiraju S, Soroush M. Parameter estimator design with application to a
chemical reactor. Ind Eng Chem Res 1998;37:45563.
[45] Stephanopoulos G, San KY. Studies on on-line bioreactor identication. I:
theory. Biotechnol Bioeng 1984;26:117688.
[46] Kosanovich KA, Piovoso MJ, Rokhlenko V, Guez A. Nonlinear adaptive control
with parameter estimation of a CSTR. J Proc Control 1995;5:13748.
[47] Soroush M. Nonlinear state-observer design with application to reactors. Chem
Eng Sci 1997;52:387404.
[48] Carloff R, Probss A, Reichert KH. Temperature oscillation calorimetry in
stirred tank reactors with variable heat transfer. Chem Eng Technol
1994;17:40613.
[49] Barandiaran MJ, De Arbina LL, De La Cal JC, Gugliotta LM, Ausa JM. Parameter
estimation in emulsion copolymerization using reaction calorimeter data. J
Appl Polym Sci 1995;55:12319.
[50] Regnier N, Defaye G, Caralp L, Vidal C. Software sensor based control of
exothermic batch reactors. Chem Eng Sci 1996;51:512536.
[51] Muske KR, Rawlings JB. Nonlinear moving horizon state estimation. In: Berber
R, editor. Methods of model based process control. Netherlands: Kluwer; 1995.
[52] Robertson DG, Lee JH, Rawlings JB. A moving horizon based approach for least
squares estimation. AIChE J 1996;42:220924.
[53] Ahn J, Ch-Ki Cho, Kang S. An efcient numerical parameter estimation scheme
for the space-dependent dispersion coefcient of a solute transport equation
in porous media. Math Comput Model 2010;51:16776.
[54] Kuzmic P. Program DYNAFIT for the analysis of enzyme kinetic data:
application to HIV proteinase. Anal Biochem 1996;237:26073.

A.T. Jarullah et al. / Fuel 90 (2011) 21652181

[55] Mendes P, Kell D. Non-linear optimization of biochemical pathways:
applications to metabolic engineering and parameter estimation.
Bioinformatics 1998;14:86983.
[56] Rinnooy Kan AHG, Timmer GT. Global optimization: a survey. Int Ser Numer
Math 1989;87:13355.
[57] Tjoa J, Biegler LT. Simultaneous solution and optimization strategies for
parameter estimation of differential-algebraic equation systems. Ind Eng
Chem Res 1991;30:37685.
[58] Fletcher R. Practice methods of optimization. 2nd ed. Chichester: John Wiley
and Sons; 1987.
[59] Levenberg K. A method for the solution of certain nonlinear problems in least
squares. Quart Appl Math 1944;2:1648.
[60] Marquardt DW. An algorithm for least-squares estimation of non-linear
parameters. SIAM J 1963;11:43141.
[61] Holland JH. Adaptation in natural and articial systems. Ann Arbor (MI): The
University of Micigan Press; 1975.
[62] Goldberg DE. Genetic algorithms in search, optimization and machine
learning. Reading (MA): Addison-Wesley; 1989.
[63] Alonge F, Dippolito F, Ferrante G, Raimondi FM. Parameter identication of
induction motor model using genetic algorithms. IEE Proc Control Theory Appl
1998;145:58793.
[64] Orlowska KT, Lis J, Szabat K. Identication of the induction motor parameters
at standstill using soft computing methods. Comput Math Electr Electron Eng
2006;25:18194.
[65] Nangsue P, Pillay P, Conry S. Evolutionary algorithms for induction motor
parameter determination. IEEE Trans Energy Convers 1999;14:44753.
[66] Abdelhadi B, Benoudjit A, Nait Said N. Identication of induction machine
parameters using an adaptive genetic algorithm. Electr Power Compon Syst
2004;32:76784.
[67] Ursem RK, Vadstrup P. Parameter identication of induction motors using
differential evolution. The 2993 congress on evolutionary computation, vol. 2.
CEC03; 2003. p. 7906.
[68] Tjoa TB, Biegler LT. Reduced successive quadratic programming strategy for
errors-in-variables estimation. Comput Chem Eng 1992;16:52333.
[69] Tsamatsoulis D, Papayannakos N. Investigation of intrinsic hydrodesulphurization kinetics of a VGO in a trickle bed reactor with backmixing effects. Chem
Eng Sci 1998;53:344958.
[70] Sanchez S, Rodrguez MA, Ancheyta J. Kinetic model for moderate
hydrocracking of heavy oils. Ind Eng Chem Res 2005;44:940913.
[71] Trejo F, Ancheyta J. Kinetics of asphaltenes conversion during hydrotreating of
Maya crude. Catal Today 2005;109:99103.
[72] Resendiz E, Ancheyta J, Rosales-Quintero A, Marroquin G. Estimation of
activation energies during hydrodesulfurization of middle distillates. Fuel
2007;86:124753.
[73] Sanchez S, Ancheyta J. Effect of pressure on the kinetics of moderate
hydrocracking of Maya crude oil. Energy Fuels 2007;21:65361.
[74] Morrison KR. Optimal control of processes described by systems of
differential-algebraic equations. Ph.D. thesis. UK: University of London; 1984.

2181

[75] Ekpo EE, Mujtaba IM. Performance analysis of three controllers for the
polymerisation of styrene in a batch reactor. Chem Prod Process Model
2007;2:129.
[76] Mujtaba IM. Batch distillation: design and operation. Series on chemical
engineering, vol. 3. London: Imperial College Press; 2004.
[77] Jarullah AT, Mujtaba IM, Wood AS. Improvement of the middle distillate yields
during crude oil hydrotreatment in a trickle-bed reactor. Energy Fuels 2011.
doi:10.1021/ef101327d.
[78] Maria AC, Martinez MT. Hydroprocessing of a Maya Residue. Intrinsic kinetics
of sulfur-, nitrogen-, nickel-, and vanadium-removal reactions. Energy Fuels
1999;13:62936.
[79] Ancheyta J, Rivera GB, Sanchez GM, Arellano AMP, Maity SK, Cortez MT, et al.
Exploratory study for obtaining synthetic crudes from heavy crude oils via
hydrotreating. Energy Fuels 2001;15:1207.
[80] Bhaskar M, Valavarasu G, Sairam B, Balaraman KS, Balu K. Three-phase reactor
model to simulate the performance of pilot-plant and industrial trickle-bed
reactors sustaining hydrotreating reactions. Ind Eng Chem Res 2004;43:
665469.
[81] Isoda T, Kusakabe K, Morooka Sh, Mochida I. Reactivity and selectivity for the
hydrocracking of vacuum gas oil over metal-loaded and dealuminated yzeolites. Energy Fuels 1998;12:493502.
[82] Kim LK, Choi KS. Hydrodesulfurization over hydrotreating catalysts. Int Chem
Eng 1987;27:34057.
[83] Al-Humaidan FS. Modelling hydrocracking of atmospheric residue by discrete
and continuous lumping. MS.C. thesis. Kuwait: Kuwait University; 2004.
[84] Abdul-Halim A-KM. Private communications. Professor, University of
Baghdad-Iraq; 5 Jauary 2011.
[85] Riyadh HH. Private communications. Manager, North Reneries CompanyBaiji, Iraq; 5 Jauary 2011.
[86] Kumar VR, Balaraman KS, Rao VSR, Ananth MS. Performance study of certain
commercial catalysts in hydrodesulfurization of diesel oils. Petrol Sci Technol
2001;19:102938.
[87] Cheng Z, Fang X, Zeng R, Han B, Huang L, Yuan W. Deep removal of sulfur and
aromatics from diesel through two-stage concurrently and countercurrently
operated xed-bed reactors. Chem Eng Sci 2004;59:546572.
[88] Jarullah AT, Mujtaba IM, Wood AS. Kinetic parameter estimation and
simulation of trickle-bed reactor for hydrodesulfurization of crude oil. Chem
Eng Sci 2011;66:85971.
[89] Areff HA. The effect of operating conditions on vacuum gas oil hydrotreating
on sulfur and aromatics content. M.SC. thesis. Iraq: University of Tikrit; 2001.
[90] Jimenez F, Nunez, Kafarov V. Study and modeling of simultaneous
hydrodesulfurization, hydrodenitrogenation and hydrodearomatization on
vacuum gas oil hydrotreatment. Comput Aid Chem Eng 2005;20:61924.
[91] Van Hasselt BW, Lebens PJM, Calis HPA, Kapteijn F, Sie ST, Moulijn JA, et al. A
numerical comparison of alternative three-phase reactors with a conventional
trickle-bed reactor. The advantages of countercurrent ow for hydrodesulfurization. Chem Eng Sci 1999;54:47919.