Академический Документы
Профессиональный Документы
Культура Документы
Fuel
journal homepage: www.elsevier.com/locate/fuel
Modeling of CSTR and SPR small-scale isothermal reactors for heavy oil T
hydrocracking and hydrotreating
⁎
Cristian J. Calderóna, Jorge Ancheytab,
a
Facultad de Química, UNAM, Ciudad Universitaria, Ciudad de México 04510, Mexico
b
Instituto Mexicano del Petróleo, Eje Central Lázaro Cárdenas Norte152, Col. San Bartolo Atepehuacan, Ciudad de México 07730, Mexico
G RA P H I C A L AB S T R A C T
A R T I C L E I N F O A B S T R A C T
Keywords: The dynamic modeling and simulation of a continuous slurry phase reactor for catalytic hydrocracking and
Modeling hydrotreating of an atmospheric residue (312 °C+) are reported. The reactor model is based on an axial dis-
Simulation persion. The hydrocracking kinetic model takes into account a five-lump model previously reported in the lit-
Slurry-phase reactor erature. The hydrotreating reactions simulated are: hydrodesulfurization (described by Langmuir–Hinshelwood
Heavy oil
kinetics), hydrodenitrogenation (for basic and non-basic nitrogen), hydrodeasphaltenization and hydro
Hydrocrackig
Hydrotreating
Conradson carbon removal (modeled with power-law approach). All the intrinsic kinetic parameters and cor-
relations were taken from the literature. The performance for the slurry-phase reactor was compared with a
continuous stirred tank reactor. Dynamic simulations and steady-state predictions agreed with the expected
behavior of the heavy fractions and impurities hydroprocesing.
⁎
Corresponding author.
E-mail address: jancheyt@imp.mx (J. Ancheyta).
https://doi.org/10.1016/j.fuel.2017.11.089
Received 11 September 2017; Received in revised form 19 November 2017; Accepted 22 November 2017
0016-2361/ © 2017 Elsevier Ltd. All rights reserved.
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
Also, these reactors experience high pressure drop of the bed which Recent investigations concerning to kinetic models for hydro-
makes difficult to maintain a normal operation, therefore shut-down is cracking in slurry-phase have been reviewed [4]. Due to heavy oil
more frequent due to short catalyst life and unstable operation of the consists of a large amount of components, different approximations are
reactor [2]. used to represent hydrocracking reactions, being the most common the
To address the disadvantages in HDC in conventional technologies, lumping techniques. The method consists of lumping various com-
slurry-phase hydrocracking processes have been developed. This tech- pounds in a few pseudocompounds that differ by boiling temperature
nology consists of mixing the oil feed, hydrogen, and dispersed catalysts range. This approximation is easy to implement in a reactor model
together going through the reactor. It has the same processing as because it reduces the number of kinetic equations and parameters to
thermocracking but also reduces coking due to the presence of hy- be estimated. On the other hand it is known that as long as the reaction
drogen and catalyst that promote hydrogenation reactions [3]. The rate coefficients have been obtained under kinetic regime, the kinetic
catalyst also acts as a support for the low amount of coke that could be model can be implemented in the reactor model independently if those
formed, and due to the catalyst leaves the reactor continuously this parameters were obtained in a different type of reactor.
effect has no greater relevance in the operation. Depending on the oil to Moreover, the literature concerning mathematical modeling of hy-
be treated, the catalysts are designed for the removal of impurities such drocracking in SPR’s is scarce, as shown in a previous review work [5].
as sulfur, nitrogen, metals and asphaltenes, even for the saturation of Most of the reports are based on computational fluid dynamics models
aromatics and olefins. All these reactions occur simultaneously and are (CFD) formulated in steady-state and focused on the performance of the
known as hydrotreating reactions (HDT). hydrodynamic variables inside the reactor and not on residue conver-
While slurry-phase reactors (SPR) for hydrocracking of heavy oil are sion [6–10]. It should be pointed out that due to the difficulty for ob-
a promising technology, plants at a commercial scale have not been taining proper information, all those reactor models were validated
developed or are still in the design stage due to high catalyst cost and with air–water systems. On the other hand, there are recent modeling
elevated operating conditions necessary to achieve the conversion of the reports dealing with hydrocracking in industrial and pilot plant units
heaviest fractions of the feed [4]. In addition, SPR technology has some [11,12]. However, such models are based on ideal plug-flow patterns in
major disadvantages compared with fixed, moving, or ebullated bed steady-state.
reactors, such as the difficult separation of catalyst and the liquid pro- Recent models for SPR’s have been published. Those reactor models
duct, uncertain scale-up, catalyst sedimentation and agglomeration, as were formulated for different reacting systems, such as Fischer–Tropsch
well as the hard understanding of reaction kinetics and flow patterns [5]. synthesis, methanol synthesis, dimethyl ether synthesis, and diesel
853
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
hydrodesulfurization [13–18]. Such models are based on axial disper- t = 0, 0 ⩽ Z ⩽ L x i = x iin (2)
sion (ADM), despite of the well-known uncertainty in the parameters
especially the axial dispersion coefficient. In a previous investigation, ∂x i
0 ⩽ t ⩽ tf , Z = 0 D = u (x i−x i0)
modeling of a SPR for hydrotreating of gas oil was reported, which uses ∂Z (3)
a kinetic model for mild hydrocracking reactions for the light fractions
∂x i
[19]. The model was successfully validated versus the experimental 0 ⩽ t ⩽ tf , Z = L =0
∂Z (4)
data of a hydrodesulfurization system and was compared with the
predictions of the reactor model of Khadem-Hamedani et al. [18]. In the case of the CSTR, the well known consideration of perfect
The objectives of this investigation are to adapt this model for hy- mixing is assumed, then the reactor model approximates to
drocracking and hydrotreating of heavy oil in a slurry phase reactor, as C R
well as to present a comparison between the dynamic behavior and ∂x i W
= x iin−x i + P ∑∑ νi ri,k
steady-state predictions of a SPR and a continuous stirred tank reactor ∂t mT i k (5)
(CSTR) based on intrinsic reaction rate coefficients and activation en-
with the initial condition
ergies taken from the literature with the aim of finding the best oper-
ating conditions and to explore the influence of catalyst deactivation x i (0) = x iin (6)
and effectiveness factors in the reactor performance.
x i stands for the total mass composition of the component i in feed
and product (gases + hydrocrabon), the subscript i represents hy-
2. Model formulation
drogen, hydrogen sulfide, sulfur, nitrogen, residue, vacuum gas oil,
distillates, naphtha and light gases in feed and product, mT represents
2.1. Reactor models
the total mass (gases + hydrocarbon) in the reactor, u the feed velocity,
WP the amount of catalyst in the feed and D the axial dispersion coef-
A schematic representation of the small-scale reactors to be modeled
ficient, which was estimated through the correlation proposed by
is presented in Fig. 1. In order to compare the performance of both C R
Deckwer et al. [13]. The therm ∑i ∑k νi ri,k represents the rate of re-
reactors, the same operating conditions were assumed: An isothermal
action of each component i in reaction k. Being νi the stoichiometric
operation, 891 std m3/ m3 of H2−hydrocarbon ratio, 100 ml of catalyst
coefficient and ri,k the intrinsic reaction rate.
loaded in the reactors, and a constant pressure of 9.8 MPa. The di-
mensions of the SPR are 38 cm of length and 7.7 cm of diameter while
2.2. Reaction kinetics
the CSTR total volume is 1.7 Lt. It is also assumed that in both reactors
the heavy oil feed is preheated and mixed with hydrogen and then
Heavy oils are a complex mixture of large number of components,
passed through the reactor.
therefore the study of all the reactions involved in hydrocracking and
Axial dispersion model was selected to approximate the SPR per-
hydrotreating is quite complicated. For the particular case of hydro-
formance. The model is based on previous considerations reported in
cracking, there are different types of kinetic models that have been
the literature, such as dynamic regime, isothermal operation, and a
reported to approximate all the reactions [20]. Despite the dependence
pseudo–homogeneous slurry-phase [13–19]. It is also assumed that the
of the kinetic parameters with properties of the feed and the use of an
gas phase is present in excess, then the mathematical model can be
invariant distillation range of products, a pseudo-component based
approximated with an homogeneous model. Based on these con-
model will be used to represent the hydrocracking kinetic due to its
siderations, the dynamic equations of the ADM are as follows:
simplicity and utility in simulation and control applications. The kinetic
C R
∂x i ∂x ∂ 2x W model is presented in Fig. 2, this model was proposed by Sanchez et al.
= −u i + Da 2i + P ∑∑ νi ri,k [21] and consists of five pseudo-components: residue (538 °C+), va-
∂t ∂Z ∂Z mT i k (1)
cuum gas oil (343–538 °C), middle distillates (204–343 °C), naphtha
With the initial and boundary conditions: (IBP-204 °C) and gases, and 10 reaction paths for hydrocracking of all
854
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
n nC
xS CS x H2 H2 n
rHDS = φS ηS kCS + kTS xS TS
(1 + K H2 S x H2 S )2 (12)
Asph n Asph nT
rHDAsph = φAsph ηAsph kCAsph xAsph + kTAsph xAsph (13)
nCCR nTCCR
rHDCCR = φCCR ηCCR kCCCR x CCR + kTCCR x CCR (14)
nNBN nTNBN
rHDNNBN = φNBN ηNBN kCNBN xNBN + kTNBN xNBN (15)
nBN n
rHDNBN = φBN ηBN kCBN xBN + kTBN xBNTBN (16)
nNBN TNBN n
−φNBN ηNBN kCNBN xNBN −kTNBN xNBN (17)
Catalyst deactivation is originated by coke deposition, however the
estimation of this phenomenon is based on deactivation functions. For
hydrocracking reactions a time-dependant non-selective expression was
used, which means that the selectivity of the hydrocracking reactions is
not affected by coke deposition, then the deactivation constant k DHDC is
the same for all the reactions. On the contrary for hydrotreating reac-
tions a time-dependant selective model was used, in this case it is ne-
cessary to take into account the effect of coke deposition on each hy-
drotreating reaction, then there would be a deactivation constant for
each reaction, (k DS,k DAsph,k DCCR,k DNBN ,k DBN ). In both cases the deactiva-
tion function is presented in Eq. 18 where mi is the deactivation order
for reaction i, (HDC ,HDS,HDAsph,HDCCR,HDNBN ,HDBN ).
1
φ=
(1 + k Di t )mi (18)
In the case of the CSTR Eq. 18 can be used directly in the reactor
Fig. 2. Five-lump kinetic model for hydrocracking reported by Sanchez et al. [21]. model due to the catalyst does not leave the reactor during the opera-
tion, however for the SPR it is considered that the catalyst does not
lumps, as well as 10 reaction rate coefficients to be calculated. Recently remain in the reactor, on the contrary it enters and leaves the reactor
Martínez and Ancheyta [22] conducted an experimental study based on along with the feed, then the deactivation function changes to Eq. 19.
the kinetic model presented in Fig. 2 to obtain the 10 intrinsic reaction 1
rate coefficients in a CSTR. The kinetic expressions for hydrocracking φ=
reactions are presented in Eqs. (7)–(11). Although most literature re-
(1 + k Di ( ))
Z
u
mi
(19)
ports propose that hydrocracking reactions of heavy oils follow a first- Z
Where represents the variation of time that the catalyst remains in
u
order reaction, in this work the hydrocracking reaction of vacuum re- the reactor. Also for the SPR, due to the catalyst particles are very small,
sidue was considered to be of second-order while the rest of the hy- the effectiveness factor is considered to be the unity for all the reactions
drocracking reactions were assumed to be of first-order [23,22]. The (ηi = 1).
second-order reaction is attributed to the broad distribution of reaction
rate values of compounds present in the residue fraction, which gen- 2.3. Model equations
erates that the reaction rate declines faster than that of a simple first-
order reaction due to the most-reactive compounds disappear first and The generalized SPR model is presented in Eq. 1, along with the
fast, leaving the most-refractory compounds in the residue [23]. initial and boundary conditions given by Eqs. (2)–(4). These equations
All the reaction rate expressions involve the calculation of effec- need to be formulated for each one of the components involved in the
tiveness factors and catalyst deactivation function, due to the large hydrocracking and hydrotreating system considering the reaction rate
particle size and the deposition of coke on the catalyst surface. expressions presented in Eqs. (7)–(17), then the set of partial differ-
rR = −φHDC η (k1 + k2 + k3 + k 4 ) yR2 (7) ential equations (PDE) is obtained as shown in Eqs. (20)–(31). That
system of PDE needs to be reduced into an ordinary differential equa-
rVGO = φHDC η [k1 yR2 −(k5 + k6 + k 7 ) yVGO ] (8) tion (ODE) system in order to find the numerical solution with a MA-
TLAB subroutine, particularly the ode23s. The orthogonal collocation
rD = φHDC η [k2 yR2 + k5 yVGO −(k8 + k 9 ) yD ] (9) method was selected to perform the calculations.
855
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
∂x G ∂x ∂ 2x W
= −u G + D G2 + P φHDC [k 4 xR2 + k 7 xVGO + k 9 xD + k10 xN ]
∂t ∂Z ∂Z mT
(24)
n nx
∂xS ∂x ∂ 2x W ⎛ xS xS x H2 H2
n ⎞
= −u S + D 2S − P ⎜φS k xS + kTS xS TS ⎟
∂t ∂Z ∂Z mT (1 + K H2 S x H2 S )2 (25)
⎝ ⎠
∂xBN ∂x ∂ 2x W n
= −u BN + D BN + P (φBN kCBN xBN
nBN
+ kTBN xBNTBN
∂t ∂Z ∂Z 2 mT
nNBN TNBN n
−φNBN k xNBN xNBN −kTNBN xNBN ) (28)
∂xNBN ∂x ∂ 2x W nTNBN
= −u NBN + D NBN + P (φNBN k xNBN xNBN
nNBN
+ kTNBN xNBN )
∂t ∂Z ∂Z 2 mT
(29)
∂x H2 ∂x H2 ∂2x H2 W
= −u +D + P rH2
∂t ∂Z ∂Z 2 mT (30)
n nx
∂x H2 S ∂x H2 S ∂2x H2 S W ⎛ x H2xSS x H2 H2 n ⎞
= −u +D + P φS k xS + kTS xS TS
∂t ∂Z ∂Z 2 mT ⎜ (1 + K H2 S x H2 S )2 ⎟
⎝ ⎠
Fig. 3. Comparison of reactor performance of heavy oil hydrocracking. (–) SPR, (- -)
(31) CSTR, (Symbols) Experiment. (o) Residue, (*) Vacuum gas oil, (+) Middle distillates, (Δ )
For the CSTR, the general model is based on Eq. 2, then including Naphtha, (∇) Gases.
the reaction rate expression (7)–(17) the CSTR model is obtained as
shown in Eqs. (32)–(43). These equations form an ordinary differential dxNaph WP
equation system which can be solve directly as well in MATLAB with
in
= xNaph −xNaph + φ [k3 xR2 + k6 xVGO + k8 xD−k10 xN ]
dt mT HDC (35)
the subroutine ode23s.
dx G
dxR W = xGin−x G + WP φHDC [k 4 xR2 + k 7 xVGO + k 9 xD + k10 xN ]
= xRin−xR− P φHDC [k1 + k2 + k3 + k 4 ] xR2 dt (36)
dt mT (32)
n nx
dxS W ⎛ xS xS x H2 H2 n ⎞
dxVGO in
= xVGO
W
−xVGO + P φHDC [k1 xR2−(k5 + k6 + k 7 ) xVGO] = xSin−xS − P ⎜φS k xS + kTS xS TS ⎟
dt mT (33) dt mT (1 + K H2 S x H2 S )2 (37)
⎝ ⎠
Table 1
Reaction rate coefficients and feed properties reported by Martínez and Ancheyta [22,24] for hydrocracking and hydrotreating reactions at 380 °C.
856
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
dx CCR W nTCCR
Fig. 4. Comparison of reactor performance of heavy oil hydrotreating. (–) SPR, (- -) CSTR, in
= x CCR −x CCR + P (φCCR k xCCR x CCR
nCCR
+ kTCCR x CCR )
(Symbols) Experiment. (o) Sulfur, (*) Asphaltenes, (+) CCR, (Δ ) Basic Nitrogen, (∇) Non dt mT (39)
Basic Nitrogen.
dxBN W n
in
= xBN −xBN + P (φBN kCBN xBN
nBN nNBN
+ kTBN xBNTBN −ϕNBN k xNBN xNBN
dt mT
n
TNBN
−kTNBN xNBN ) (40)
dxNBN W nTNBN
in
= xNBN −xNBN + P (φNBN k xNBN xNBN
nNBN
+ kTNBN xNBN )
dt mT (41)
dx H2 W
= x Hin2 −x H2 + P rH2
dt mT (42)
n nx
dxS W ⎛ xS xS x H2 H2 n ⎞
= xSin−xS + P ⎜φS k xS + kTS xS TS ⎟
dt mT (1 + K H2 S x H2 S )2 (43)
⎝ ⎠
857
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
858
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
Fig. 10. Yield of hydrocracked lumps as a function of residue conversion in the CSTR.
859
C.J. Calderón, J. Ancheyta Fuel 216 (2018) 852–860
simulations have shown similar behavior in the CSTR and SPR. As ex-
pected higher conversion of residue is reached in the SPR (effectiveness
factor equals one) and deactivation of catalyst is not taken into account.
The results of steady-state simulation have shown that it is possible to
approximate a slurry-phase reactor for hydrocracking with an axial
dispersion model with appropriate reaction rate coefficients.
As expected, for experimental proposes the CSTR has some ad-
vantages compared with the SPR. It promises lower cost in the ex-
perimental framework due to the amount of catalyst used. It was also
found that for modeling purposes, the experimental SPR can be con-
veniently approximated by means of an ADM, however to reduce its
complexity, the SPR can be approximated to an ideal CSTR model. On
the other hand for industrial applications the SPR promises to obtain
better results in the conversion of residues to light fractions.
Acknowledgement
The authors thank Instituto Mexicano del Petróleo for its financial
support. C.J.C also thanks CONACyT for his Ph.D. scholarship.
References
860