Modeling of vacuum residue hydrotreatment Impact of the feedstock description on model output Pedro Manuel Mendes Rivotti Dissertação para obtenção do Grau de Mestre em Engenharia Química Júri Presidente: Prof. João Carlos Moura Bordado Orientadores: Prof. Fernando Manuel Ramôa Cardoso Ribeiro Dr. Vitor Raul Lameiras Franco da Costa Vogal: Prof. Francisco Manuel da Silva Lemos Setembro de 2009
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Modeling of vacuum residue hydrotreatment
Impact of the feedstock description on model output
Pedro Manuel Mendes Rivotti
Dissertação para obtenção do Grau de Mestre em
Engenharia Química
Júri
Presidente: Prof. João Carlos Moura Bordado
Orientadores: Prof. Fernando Manuel Ramôa Cardoso Ribeiro
Dr. Vitor Raul Lameiras Franco da Costa
Vogal: Prof. Francisco Manuel da Silva Lemos
Setembro de 2009
i
Resumo
Neste trabalho é apresentado o estudo de um modelo de hidrotratamento de resíduos de vácuo
desenvolvido no IFP.
O modelo em estudo simula a operação de um reactor de hidrodesmetalização e de uma unidade
completa, envolvendo as secções de hidrodesmetalização e hidrodessulfuração. A difusão no interior
do catalisador é descrita pelas equações de Stefan-Maxwell adaptadas a um sistema com moléculas
com grande diferença de dimensões. O elevado número de compostos considerados e as equações
utilizadas para descrever a difusão no interior do catalisador levam a que uma rotina de estimação de
parâmetros possa convergir para um mínimo local sem significado físico, se os parâmetros iniciais
forem introduzidos aleatoriamente.
Neste estudo é apresentada a metologia utilizada para obter estimativas iniciais dos parâmetros
do modelo, nomeadamente das constantes cinéticas e de adsorção. A obtenção das estimativas
iniciais foi feita utilizando como base um resíduo do Médio Oriente (Buzurgan). Com os parâmetros
obtidos, a resposta do modelo foi simulada para outros resíduos, com diferentes origens geográficas
(Arabian Light, Djeno e Oural). O estudo consistiu igualmente numa análise de sensibilidade aos
diferentes valores a introduzir no modelo.
A resposta do modelo a variações nos seus parâmetros e valores de entrada pôde ser explicada,
na maioria dos casos, utilizando argumentos baseados nos fenómenos físicos e químicos envolvidos
no processo catalítico. As simulações efectuadas para resíduos com diferentes origens geográficas
apenas evidenciaram pequenos desvios em relação aos dados experimentais.
Table 15 - Kinetic constants optimized for the HDS section, using Buzurgan residue as reference. ... 59
Table 16 – Density of Buzurgan residue for different residence times in the HDM and HDS reactor. . 67
Table 17 – Percentage of each SARA fraction in Buzurgan residue, for different residence times in the
HDM and HDS reactor. .......................................................................................................................... 67
Table 18 – Heteroatoms composition of Buzurgan residue for different residence times in the HDM
and HDS reactor. ................................................................................................................................... 67
Table 19 – Density of Arabian Light residue for different residence times in the HDM and HDS reactor.
Table 20 – Percentage of each SARA fraction in Arabian Light residue, for different residence times in
the HDM and HDS reactor..................................................................................................................... 68
Table 21 – Heteroatoms composition of Arabian Light residue for different residence times in the HDM
and HDS reactor. ................................................................................................................................... 68
Table 22 – Density of Djeno residue for different residence times in the HDM and HDS reactor. ....... 68
Table 23 – Percentage of each SARA fraction in Djeno residue, for different residence times in the
HDM and HDS reactor. .......................................................................................................................... 69
Table 24 – Heteroatoms composition of Djeno residue for different residence times in the HDM and
Figure 15 – Collision between molecules with different sizes, as described by Fornasiero (Fornasiero
et al., 2005). ........................................................................................................................................... 20
Figure 16 – Schematic representation of the different lumps considered in the model. ....................... 27
Figure 17 – Discretization adopted for the spherical catalyst and the plug flow reactor. ...................... 39
Figure 18 – Possible method of dividing asphaltenes in 4 fractions for an Oural residue feedstock. ... 43
Figure 19 - Possible method of dividing asphaltenes in 4 fractions for a Buzurgan residue feedstock,
based on the overlap of 4 Lorentz distributions. ................................................................................... 44
Figure 20 – Effect of variations in the adsorption constant ka1 in the model simulated nitrogen removal
The adsorption constants presented in Table 9 suggest that asphaltenes are the SARA fraction
with the strongest adsorption on the catalyst. Since there is no physical explanation for having such a
difference between the adsorption constant for asphaltenes and resins, the high value observed for
asphaltenes may result from purely mathematical arguments. Also, the adsorption constants for NH3
and H2S would be expected to have higher values, owing to the strong interaction between these
compounds and the catalyst.
4.2.2.2 Kinetic constants
Contrary to the adsorption constants, the kinetic constants implemented in the model are in
sufficient number to allow varying one constant without affecting a large number of curves. For
example, by changing the kinetic constant k1, (chapter 3.2.3) only the curve concerning the nickel
removal from asphaltenes would be affected. Table 10 shows which kinetic constants affect each of
the heteroatom removal and lump fraction curves.
Table 10 – Kinetic constants affecting each of the heteroatom removal and lump fraction curves.
Asphaltenes Resins Aromatics Saturates
S N Ni V
Lump fraction
k3, k4 k5 k1 k2
k4, k6, k7
k10, k11 k12 k8 k9
k7, k11, k13, k14, k15
k16 - - -
k7, k14, k15
- - - -
k13, k15, k4, k11
According to Table 10, variations in the constants k4 and k11 affect both the sulfur removal and the
lump fraction curves of asphaltenes and resins. Therefore, it was chosen to use these constants only
to adjust the sulfur removal curves, while the lump fraction curves were adjusted with the other
relevant constants. It should also be pointed out that constants k3 and k4 for asphaltenes and k10 and
k11 for resins were given the same value, since there is no evidence on how the saturation removal
influences the reaction rate.
The optimization methodology for the kinetic constants is exemplified in Figure 21 for the nickel
removal from asphaltenes curve. In accordance with Table 10, the kinetic constant used to adjust this
curve was k1. It can be seen that higher values of k1 imply higher nickel removal from asphaltenes,
which would be the expected behavior since k1 represents the nickel removal from asphaltenes
reaction rate.
48
The methodology used to adjust the other curves was analogous to the one used for the nickel
removal from asphaltenes. In the case of the lump fraction curves, since cracking reactions always
affect more than one lump, several combinations of these parameters had to be tested in order to find
the ones that led to the minimal difference between experimental data and model output.
Figure 21 - Effect of variations in the kinetic constant k1 on the model simulated nickel removal from asphaltenes.
As emphasized in chapter 4.2.1, although at this point it was important to obtain a reasonable
adjustment between experimental data and model output, the main goal was to confirm whether the
model was sensitive to variations on these parameters and if the variations could be explained using
arguments based on the phenomena involved.
Table 11 presents the set of kinetic constants resulting from this optimization.
Table 11 – Kinetic constants optimized for Buzurgan residue feedstock.
Kinetic constant Reaction Value 1 AspNi � Asp + Ni 4.20x10-2 2 AspV � Asp + V 5.25x10-2 3 AspS ansm � Asp ansm + H2S n = 1,2 ; m = 1,2 5.20x10-4 4 AspS ans2 � Asp ans1 + H2S + Sat n = 1,2 5.50x10-4 5 AspN � Asp + NH3 4.00x10-2 6 Asp ans2 –> Asp ans1 + Sat n = 1,2 5.72x10-2
7 Asp a2sm � Asp a1sm + Aro m = 1 , 2 Asp a1sm � Res a2s2 + Aro m = 1 , 2
5.24x10-2
8 ResNi � Res + Ni 6.66x10-2 9 ResV � Res + V 7.98x10-2 10 ResS ansm � Res ansm + H2S n = 1,2 ; m = 1,2 4.25x10-2 11 ResS ans2 � Res ans1 + H2S + Sat n = 1,2 4.60x10-2 12 ResN � Res + NH3 3.00x10-2 13 Res ans2 –> Res ans1 + Sat n = 1,2 2.09x10-2 14 Res a2sm � Res a1sm + Aro m = 1 , 2 2.96x10-2 15 Res � Aro + Sat 2.81x10-2 16 AroS � Aro + H2S 4.00x10-2 17 Asp / Res � coke 6.11x10-4
49
The kinetic constants presented in Table 11 suggest that for asphaltenes, the rate at which the
saturation level is lowered (k6) is higher than the rate for aromaticity lowering (k7). For the resins, the
opposite behavior is observed, according to the values of the kinetic constants k13 and k14. There is no
physical evidence for this difference between asphaltenes and resins and, as shown in chapter
4.3.2.1, other sets of kinetic constants can be used without observing this behavior. As expected, the
kinetic constant for the removal of aromaticity and saturation from asphaltenes (k15) is intermediary
between k13 and k14.
The kinetic constants obtained for desulfurization reactions suggest that this reaction occurs with a
greater rate in resins and aromatics (k10, k11 and k16) than in asphaltenes (k3 and k4). This observation
arises from the fact that sulfur in asphaltenes is present in compounds more refractory than those of
resins and aromatics.
The kinetic constant k17, corresponding to the coke formation reaction, did not have impact in any
of the curves simulated by the model. Therefore, it was not optimized and a wide range of values
could be used for this parameter.
As referred in chapter 4.2.1, different combinations of parameters may serve as a solution.
Therefore, the kinetic constants here presented should be analyzed carefully to avoid drawing false
conclusions on the reactivity of a residue.
4.2.2.3 Estimation routine results
The model parameters obtained for the Buzurgan residue (Table 10 and Table 11) were
afterwards used as initial values for the estimation routine implemented on the model. With the
maximum iterations number set to 250 iterations, the sum of square differences between experimental
data and model output could be decreased by 5%.
For the same number of maximum iterations, two modifications were tested in the estimation
routine. A first attempt consisted in giving more importance to the sulfur in asphaltenes in the sum of
square difference between experimental data and model output, since this was the curve that showed
more deviations (Figure 23). A weight 1000 times higher was given to its square difference. For
another attempt, all the experimental data and model output were normalized (divided by its initial
values) and the routine was set to minimize the square difference between the normalized values.
However, none of these modifications improved the estimation routine results. To further decrease the
sum of square differences experimental data and model output, the computation time would have to
be increased, by increasing the maximum number of iterations.
4.2.2.4 Optimized results
The results from a simulation using the optimized kinetic and adsorption constants of Table 10 and
Table 11 are presented in the following figures (Figure 22 to Figure 24) along with the corresponding
experimental data.
50
As seen in these figures, an optimization based only on the kinetic and adsorption constants
resulted in a good agreement between experimental data and model output for a given residue.
However, a tendency towards high fractions was observed in the sulfur removal from asphaltenes
curve and towards low fractions in the nitrogen removal from resins. Different sets of kinetic and
adsorption constants could have been tested, to obtain a better fit for these curves. However, to test
the performance of the implemented estimation routine (chapter 4.2.2.3), the parameters should not
be set to values that lead to a perfect adjustment, since any possible improvements obtained with the
routine would be imperceptible.
(a) (b)
Figure 22 – Optimized model output vs. experimental data (Buzurgan) for: (a) nickel removal; (b) vanadium removal.
(a) (b)
Figure 23 - Optimized model output vs. experimental data (Buzurgan) for: (a) sulfur removal; (b) nitrogen removal.
51
(a) (b)
Figure 24 - Optimized model output vs. experimental data (Buzurgan) for: (a) total lump fractions; (b) overall heteroatom removal.
4.2.3 Simulations for different residues
Following the optimization scheme presented in chapter 4.2.1, the kinetic and adsorption
constants optimized for Buzurgan residue were used to simulate the other residues: Arabian Light,
Djeno and Oural. The repartition and distribution coefficients were kept at the same values used for
Buzurgan residue, except for the repartition coefficients of total asphaltenes and resins, which are
determined using the GPC curves (chapter 4.1). The values used for the ns numbers of aromatics and
saturates of each residue were based on the value for Buzurgan residue. This value was multiplied by
the ratio of molecular weights of the fraction in the two residues.
The simulations for different residues allow understanding how the model responds to variations in
experimental input. The input to be changed when a new residue is introduced includes the feedstock
density, the fraction of each SARA lump and the analysis of heteroatoms in asphaltenes, resins and
aromatics.
The obtained results (lump fraction and overall heteroatom removal) for Arabian Light are
presented in Figure 25. As shown in this figure, good adjustments were obtained for the overall
heteroatom removal and for the lump fractions, although in the latter slight tendencies towards low
fractions were observed for saturates and aromatics and towards high values for resins.
52
(a) (b)
Figure 25 – Model output vs. experimental data for Arabian Light residue, using the parameters optimized for Buzurgan residue. (1) Lump fractions; (2) Overall heteroatom removal.
Figure 26 presents the lump fraction and overall heteroatom removal simulated for Djeno residue.
In this case, the overall sulfur removal is underestimated by the model and slight tendencies towards
high fractions can be observed in the case of resins and towards low values in the case of aromatics.
(a) (b)
Figure 26 - Model output vs. experimental data for Djeno residue, using the parameters optimized for Buzurgan residue. (1) Lump fractions; (2) Overall heteroatom removal.
The lump fraction and overall heteroatom removal simulated for Oural residue are presented in
Figure 27. From the observation of this figure it is not clear whether the model simulations are in
agreement with experimental data, since only two experimental points are available for each curve.
However, the curves for sulfur and vanadium removal in Figure 27-b) appear to be deviated from the
corresponding experimental points.
53
(a) (b)
Figure 27 - Model output vs. experimental data for Oural residue, using the parameters optimized for Buzurgan residue. (1) Lump fractions; (2) Overall heteroatom removal.
4.2.4 Sensitivity analysis
As referred in chapter 4.2.1, the sensitivity analysis here presented intends to determine how the
model responds to variations in the repartition coefficients, distribution coefficients and ns numbers. It
was carried out for Buzurgan residue using the optimized kinetic and adsorption constants (chapter
4.2.2).
4.2.4.1 Repartition coefficients
The repartition coefficients are model inputs defined for total asphaltenes and resins, for
asphaltenes and resins with vanadium and for asphaltenes and resins with nickel. They represent the
fraction of these compounds that are of types a2s2, a2s1, a1s2 and a1s1.
As an example, Figure 28 presents the model simulated asphaltenes fraction resulting from
variations in the total asphaltenes repartition coefficients. As seen in Figure 28-a), the model did not
show much sensitivity to these variations. This observation was general and no curves showed
significant variations.
Figure 28-b) corresponds to a zoom of Figure 28-a) and intends to show in which order the small
variations observed occur.
54
(a) (b)
Figure 28 – Effect of variations in the total asphaltenes repartition coefficients on the model simulated asphaltenes fraction. (b) corresponds to a zoom of (a).
As expected, the model predicts less conversion for asphaltenes when a bigger weight is given to
the heavier fractions (a2s2). On the other hand, when the lighter fractions (a1s1) are favored, more
conversion for asphaltenes is simulated by the model. It can also be observed that the model predicts
more conversion for compounds with less degree of saturation (compounds of the type axs1).
4.2.4.2 ns numbers
As described in chapter 3.2.6, two ns numbers are required as model inputs, one for aromatics
and one for saturates. These numbers represent the ratio between the molar volume of a molecule
and the molar volume of hydrogen (unitary molecule).
As an example, Figure 29 presents the model simulated vanadium removal from asphaltenes,
resulting from variations in these two ns numbers. The chosen values intended to test the influence of
having bigger molecules (ns Sat = 29; ns Aro = 30), smaller molecules (ns Sat = 9; ns Aro = 10) and the
influence of separately having bigger saturates (ns Sat = 30; ns Aro = 20) or bigger aromatics
(ns Sat = 19; ns Aro = 30). Figure 29-b) corresponds to a zoom of Figure 29-a).
Figure 29 – Effect of variations in the ns numbers on the model simulated vanadium removal from asphaltenes. (b) corresponds to a zoom of (a).
The effect observed in Figure 29-b) is in agreement with the expected, with more vanadium
removal occurring when smaller molecules are considered. Also, for bigger molecules, the model
predicts less removal. As for the impact of separately having bigger saturates or bigger aromatics, the
two curves corresponding to this situation only slightly differ. However from the order observed in
Figure 29-b), less removal is predicted when bigger saturates are considered. This tendency is similar
to that observed in chapter 4.2.4.1 where a higher degree of saturation limited the asphaltenes
conversion.
A limitation of the model was also observed while varying the ns numbers, since the simulated
sulfur removal was higher for large molecules and lower for smaller molecules, contrary to the
expected. Figure 30 shows the model simulated sulfur removal from asphaltenes when the previously
mentioned variations were carried out. The order of the curves in this case is inverse to that observed
in Figure 29.
(a) (b)
Figure 30 - Effect of variations in the ns numbers on the model simulated sulfur removal from asphaltenes. (b) corresponds to a zoom of (a).
56
4.2.4.3 Distribution coefficients
The distribution coefficients are model inputs defined for the sulfur or nitrogen in asphaltenes and
resins. For each of these lumps, there are 4 distribution coefficients (one for each fraction, a2s2, a2s1,
a1s2 or a1s1) and they represent the total number of sulfur atoms in the lump fraction.
In general, the model output showed little sensitivity to variations in these coefficients. For
example, when the 4 distribution coefficients were changed from 15 to 10 and 50, no significant
changes were observed in any curve (Figure 31-a). However, when the kinetic constant k11 (Table 11)
was set to higher values, from 4.6x10-2 to 5x10-1, and the same variations in the distribution
coefficients were performed, a variation was observed in the total resins fraction curve, as presented
in Figure 31-b). According to the figure below, the resins conversion is favored when less sulfur atoms
are present in this lump, which would be expected since compounds with less sulfur atoms will have
smaller sizes and thus will react more easily.
This example points out that a sensitivity analysis carried out with fixed values for the kinetic and
adsorption constants can lead to false conclusions. For example, in this case, the number of sulfur
atoms in the resins lump only impacts the reactivity if the desulfurization reaction with resins cracking
is given enough importance (high values for the kinetic constant k11). However, analyzing the
combined effect of variations in all parameters and model inputs would not be possible, since it would
involve a large amount of simulations.
(a) (b)
Figure 31 – Effect of variations in the 4 distribution coefficients for sulfur in resins on the model simulated total resins fraction using the kinetic constants: (a) k11 = 4.6x10-2 (b) k11 = 5x10-1.
57
4.3 HDM + HDS model
4.3.1 Optimization strategy
For the complete model, with both HDM and HDS sections, the number of kinetic and adsorption
constants is twice as that of the HDM model. However, the principle of a hydrotreatment unit consists
on having separate sections for HDM and HDS and therefore the parameters for each section can be
optimized separately.
The parameter optimization scheme used is analogous to the one presented for the HDM section
(chapter 4.2.1). The only difference is that, in this case, the initial conditions for the HDS model are
determined by the model simulations in the HDM section for residence times of 1h and 2h. Therefore,
the parameter optimization for the HDM section should be carried out together with the HDS section,
so that an overall minimum is found.
The sensitivity analysis to variations in the repartition coefficients, distribution coefficients and ns
numbers, as the one presented in chapter 4.2.4 for the HDM section, was not carried out for this
model since the results tendency was expected to be similar.
4.3.2 Parameter optimization
4.3.2.1 HDM section parameters
The kinetic and adsorption constants optimized for the HDM section of the model are presented in
Table 12 and Table 13, respectively.
Table 12 – Adsorption constants optimized for the HDM section, using Buzurgan residue as reference.
Group Adsorption constant
Asphaltenes 1.00x10-3
Resins 5.63x10-4
Aromatics 3.30x10-4
Saturates 8.94x10-6
NH3 3.00x10-2
H2S 3.19x10-2
As referred in chapter 4.3.1, the adsorption constants presented in Table 12 should be similar to
those optimized for the HDM model (Table 9). In fact, the adsorption constants for the four SARA
fractions are the same as those optimized for that model. However, in this case the adsorption
constants optimized for NH3 and H2S have higher values, in accordance with the strong interaction
expected between these compounds and the catalyst.
58
Table 13 – Kinetic constants optimized for the HDM section, using Buzurgan residue as reference.
Kinetic constant Reaction Value
1 AspNi � Asp + Ni 3.00x10-2
2 AspV � Asp + V 5.00x10-2
3 AspS ansm � Asp ansm + H2S n = 1,2 ; m = 1,2 7.00x10-3
4 AspS ans2 � Asp ans1 + H2S + Sat n = 1,2 7.00x10-3
5 AspN � Asp + NH3 8.00x10-2
6 Asp ans2 –> Asp ans1 + Sat n = 1,2 5.00x10-2
7 Asp a2sm � Asp a1sm + Aro m = 1 , 2 Asp a1sm � Res a2s2 + Aro m = 1 , 2
5.00x10-2
8 ResNi � Res + Ni 1.60x10-1
9 ResV � Res + V 2.00x10-1
10 ResS ansm � Res ansm + H2S n = 1,2 ; m = 1,2 1.10x10-1
11 ResS ans2 � Res ans1 + H2S + Sat n = 1,2 1.10x10-1
12 ResN � Res + NH3 8.00x10-2
13 Res ans2 –> Res ans1 + Sat n = 1,2 4.00x10-2
14 Res a2sm � Res a1sm + Aro m = 1 , 2 4.00x10-2
15 Res � Aro + Sat 1.50x10-1
16 AroS � Aro + H2S 7.00x10-2
17 Asp / Res � coke 1.00x10-4
As expected, the kinetic constants presented in Table 13 are similar to those optimized for the
HDM model (Table 11). The differences between the two sets of kinetic constants may arise from the
inherent uncertainty of the parameters. Also, as referred in chapter 4.3.1, the HDM kinetic constants
had to be readjusted since they have influence on the initial values of the HDS section.
The values obtained for the desulfurization rates in asphaltenes (k3 and k4), resins (k10 and k11)
and aromatics (k16) indicate a lower rate for asphaltenes, in accordance with the observations of
chapter 4.2.2.2.
As mentioned in chapter 4.2.2.2, the kinetic constant k17, corresponding to the coke formation
reaction, did not have impact in any of the curves simulated by the model. Therefore, it was not
optimized and several different values could have been presented in Table 11.
It should be re-emphasized that different combinations of parameters can serve as a solution.
Therefore, the kinetic constants here presented should be analyzed carefully to avoid drawing false
conclusions on the reactivity of a residue.
59
4.3.2.2 HDS section parameters
The kinetic and adsorption constants optimized for the HDS section of the model are presented in
Table 14 and Table 15, respectively.
Table 14 - Adsorption constants optimized for the HDS section, using Buzurgan residue as reference.
Group Adsorption constant
Asphaltenes 3.00x10-4
Resins 2.29x10-4
Aromatics 8.95x10-5
Saturates 8.94x10-6
NH3 3.00x10-2
H2S 3.19x10-2
As expected, the adsorption constants in Table 14 differ from those optimized for the HDM section
(Table 12), since the two sections use different catalysts. The agreement between the adsorption
constants of saturates, NH3 and H2S in both sections results only from the fact that the values
optimized for the HDM section were used as a starting point for the optimization in the HDS section. It
does not have physical significance. The observations concerning the relative values of the constants
in Table 14 are analogous to those pointed out for the HDM section, with the NH3 and H2S adsorption
constants having the highest values.
Table 15 - Kinetic constants optimized for the HDS section, using Buzurgan residue as reference.
Kinetic constant Reaction Value 1 AspNi � Asp + Ni 1.00x10-15 2 AspV � Asp + V 5.00x10-2 3 AspS ansm � Asp ansm + H2S n = 1,2 ; m = 1,2 2.00x10-3 4 AspS ans2 � Asp ans1 + H2S + Sat n = 1,2 2.00x10-3 5 AspN � Asp + NH3 5.00x10-3 6 Asp ans2 –> Asp ans1 + Sat n = 1,2 5.00x10-1
7 Asp a2sm � Asp a1sm + Aro m = 1 , 2 Asp a1sm � Res a2s2 + Aro m = 1 , 2
1.00
8 ResNi � Res + Ni 1.00 9 ResV � Res + V 2.50 10 ResS ansm � Res ansm + H2S n = 1,2 ; m = 1,2 2.50x10-1 11 ResS ans2 � Res ans1 + H2S + Sat n = 1,2 2.50x10-1 12 ResN � Res + NH3 5.00x10-3 13 Res ans2 –> Res ans1 + Sat n = 1,2 1.00 14 Res a2sm � Res a1sm + Aro m = 1 , 2 1.00 15 Res � Aro + Sat 2.50 16 AroS � Aro + H2S 1.55 x101 17 Asp / Res � coke 1.00x10-1
60
The very low value of k1 in Table 15 corresponds to the lower limit that kinetic constants can
assume in the model. A reaction with such low kinetic constant should be considered not to occur in
significant extent. However, this value should not be related to an inferior demetallization catalytic
activity. Instead, it probably reflects the diffusion limitations of asphaltenes that cause small
concentrations of asphaltenes with nickel inside the catalyst, thus lowering the reaction rate. This
effect is not observed for asphaltenes with vanadium (k2), which may be due to the higher
concentrations of vanadium in the Buzurgan residue.
The kinetic constants of desulfurization reactions indicate higher rates in resins (k10 and k11) and
aromatics (k16) than those observed in the HDM section, which would be expected considering the
type of catalyst used.
The values obtained for the asphaltenes cracking (k6 and k7) and resins cracking (k13, k14 and k15)
indicate a higher rate for these reactions in the HDS section. This behavior would be expected since
the catalyst used in this section has a higher acidity.
4.3.2.3 Optimized results
Figure 32 to Figure 34 present the results of a simulation using the kinetic and adsorption
constants optimized for the HDM + HDS model, along with the corresponding experimental data. In
these figures, the solid lines represent the model output for the HDM section while the dashed lines
represent the model output for the HDS section. The experimental data is represented by the filled and
unfilled marks, for the HDM and HDS sections, respectively.
The results presented in these figures reveal that the model output satisfactorily represents the
HDM experimental data. The slight deviations observed in the nitrogen removal curve (Figure 33-b)
could have been readjusted by further tuning the kinetic constants.
As for the HDS section, the results in general are considered satisfactory. However, some of the
experimental data like the nickel fraction in asphaltenes or the sulfur fraction in resins does not appear
to be correctly simulated by the model. In other cases, for example for the vanadium in asphaltenes, it
is difficult to find a set of parameters that gives good predictions, simultaneously for the two residence
times in study.
61
(a) (b)
Figure 32 – HDM +HDS model: optimized model output vs. experimental data (Buzurgan) for: (a) nickel
removal; (b) vanadium removal.
(a) (b)
Figure 33 – HDM +HDS model: optimized model output vs. experimental data (Buzurgan) for: (a) sulfur removal; (b) nitrogen removal.
62
(a) (b)
Figure 34 - HDM +HDS model: optimized model output vs. experimental data (Buzurgan) for: (a) total lump fractions; (b) overall heteroatom removal.
4.3.3 Simulations for different feedstocks
The simulations for different residue feedstocks were carried out in the same way as for the HDM
model (chapter 4.2.3), i.e., the kinetic and adsorption constants were set to the optimized values
(chapter 4.3.2) and for each feedstock, the only model inputs changed were the total repartition
coefficients of asphaltenes and resins, and the ns numbers of aromatics and saturates.
The experimental data to be introduced is the same as for the HDM model and corresponds to the
analysis of the feedstock (density, fraction of each SARA lump and analysis of heteroatoms in
asphaltenes, resins and aromatics). Having the HDS section incorporated in the model does not imply
the need of additional experimental data, since the initial conditions for this section are determined by
the output of the HDM section.
The obtained results (lump fraction and overall heteroatom removal) for the Arabian Light vacuum
residue are presented in Figure 35. The results for the HDM section are similar to those obtained for
this feedstock in the HDM model (chapter 4.2.3), which would be expected since similar sets of kinetic
and adsorption constants are used. However, in Figure 35-b), a slight tendency towards low fractions
can be observed for the sulfur removal. Also, Figure 35-a) the conversion of aromatics and resins
simulated by the model slightly differs from the experimentally observed. For the HDS section, no
significant deviations were observed.
63
(a) (b)
Figure 35 – HDM + HDS model: Model output vs. experimental data for Arabian Light residue, using the parameters optimized for Buzurgan residue. (1) Lump fractions; (2) Overall heteroatom removal.
Figure 36 presents the results of the simulation carried out for the Djeno residue. For the HDM
section, the model simulated curves have a similar behavior to those presented for the HDM model.
However, some additional deviations towards low fractions can be observed for the nickel and
nitrogen removal curves. As for the HDS section, the results can be considered satisfactory, except for
the nickel removal curve (Figure 36-b), which is dislocated towards low fractions.
(a) (b)
Figure 36 – HDM + HDS model: Model output vs. experimental data for Djeno residue, using the parameters optimized for Buzurgan residue. (1) Lump fractions; (2) Overall heteroatom removal.
The results obtained for Oural residue feedstock are presented in Figure 37. As mentioned in
chapter 4.2.3, it is difficult to compare the HDM model simulations with experimental data for this
64
feedstock, since only two experimental points are available. However, the general tendency of
experimental points in Figure 37 appears to be well simulated by the model.
(a) (b)
Figure 37 – HDM + HDS model: Model output vs. experimental data for Oural residue, using the parameters optimized for Buzurgan residue. (1) Lump fractions; (2) Overall heteroatom removal.
4.4 Results summary
The results presented in the above chapters concern a residue hydrotreatment kinetic model that
simulate the operation of a reactor with a single HDM section and afterwards the operation of the unit
with both HDM and HDS sections.
Based on the parameter optimization carried out for Buzurgan residue, it may be concluded that it
is possible to satisfactorily adjust the model output to experimental data by optimizing the kinetic and
adsorption constants. To improve the slight deviations observed, further simulations with different
parameters would have to be carried out or the obtained set of parameters would have to be
introduced in the estimation routine implemented. However, it was shown that for a maximum of 250
iterations, the sum of square differences between experimental data and model output only decreased
by 5%.
The values obtained for the kinetic and adsorption constants not always could be explained based
on the phenomena involved. However, it should be noted that of values with different physical
meaning can sometimes serve as a solution for the problem, based on purely mathematical
arguments.
The kinetic and adsorption constants optimized for the reference residue can be used to simulate
for other residues. The obtained model simulations only show slight deviations from experimental
data, which could be improved by changing the repartition coefficients, distribution coefficients and ns
numbers.
65
According to the sensitivity analysis carried out, the impact of varying the repartition coefficients,
distribution coefficients and ns numbers may be, in most cases, explained using arguments based on
the phenomena involved. However, an inconsistency was found while varying the ns numbers since
the model simulated a higher sulfur removal for large molecules and lower removal for smaller
molecules, contrary to the expected.
The results presented in the sensitivity analysis also pointed out that the model response to
variations in a certain parameter may depend on the values set for other parameters. Therefore, the
sensitivity analysis results should be analyzed carefully, to avoid drawing false conclusions.
Although many of the results presented are focused on finding a good adjustment between
experimental data and model output, the main goal of this study was to determine the model’s
sensitivity to variations in its parameters and how it responds for feedstocks with different origins.
66
5 Conclusion and perspectives
The main purpose of this study was to analyze the sensitivity of the model output to variations in
its parameters and to verify if these variations were coherent with the phenomena occurring in a
residue hydrotreatment reactor. In most cases, the observed results could be explained based on the
phenomena occurring with some exceptions.
One of the problems perceived during the parameter optimization concerned the lack of
experimental data. To ensure that the optimized parameters correctly describe the experimental
observations, more data should be obtained, especially for the Oural residue in both sections.
The kinetic and adsorption constants optimization was carried out for a chosen reference
feedstock, in this case Buzurgan residue. Using the obtained parameters to simulate for other
residues (Arabian Light, Djeno and Oural), the results are mainly satisfactory although some deviation
were observed between experimental data and model output. For future work, it would be interesting
to optimize these parameters for a different feedstock and see how this would impact the model
simulations for residues with different origins.
It was pointed out that using the implemented estimation routine, significant computation time is
required to decrease the sum of square differences between experimental data and model output by
only 5%. For future work it would be interesting to analyze the results of long estimation run, with a
high number of maximum iterations. The set of kinetic and adsorption constants obtained in this study
could be used as an initial value for this estimation.
Although this model was developed with the purpose of understanding the reactivity of residue
feedstocks with different origins, it could be used as a base for a model describing the full operation of
an industrial residue hydrotreatment unit. This would imply taking into account the series of reactors
used in each section, the intermediary quenching operations and the effect of variations in the
operating conditions of the unit such as the reactor temperature.
67
6 Appendices
6.1 Detailed analysis for the different residue fee dstocks
The following tables present the detailed analysis available for the different residue feedstocks
(Buzurgan, Arabian Light, Djeno, Oural). For each residue, the available data includes the feedstock
density, the percentage of each SARA fraction and the composition in heteroatoms (sulfur, nitrogen,
nickel and vanadium). All available data is presented for the different residence times studied in each
reactor (HDM and HDS).
6.1.1 Buzurgan
Table 16 – Density of Buzurgan residue for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 1 2 1 2
Density (kg/m 3) 1037 1034 1026 1013 998
Table 17 – Percentage of each SARA fraction in Buzurgan residue, for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 1 2 1 2
S 9.6 11.1 11.1 13.3 15.4
A 38.5 39.3 40.8 43.7 45.2
R 37 34.7 35.2 30.1 27.4
A 14.9 14.9 12.8 12.9 12
Table 18 – Heteroatoms composition of Buzurgan residue for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time 0 1 2 1 2
Asphaltenes
S (%wt) 8.04 8.33 8.07 8.34 8.03
N (%wt) 1.09 1.02 0.84 0.99 0.98 Ni (wppm) 251 207 207 249 252 V (wppm) 801 626 600 590 630
Resins
S (%wt) 7.15 6.09 5.74 5.52 4.6
N (%wt) 0.78 0.75 0.76 0.8 0.8 Ni (wppm) 56 55 45 47 30 V (wppm) 172 163 132 161 77
Aromatics S (%wt) 4.87 4.53 4.04 2.29 1.98
68
6.1.2 Arabian Light
Table 19 – Density of Arabian Light residue for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 1 2 1 2
Density (kg/m 3) 1017 996 995 990 989
Table 20 – Percentage of each SARA fraction in Arabian Light residue, for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 1 2 1 2
S 10.8 12.3 14.3 14.6 16.6
A 42.5 45 47.1 47.6 47.8
R 37.7 34.5 30.7 30.3 27.7
A 9 8.2 7.9 7.5 7.8
Table 21 – Heteroatoms composition of Arabian Light residue for different residence times in the HDM and HDS reactor.
6.1.3 Djeno
Table 22 – Density of Djeno residue for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 1 2 1 2
Density (kg/m 3) 989 984 984 984 989
HDM reactor HDS reactor
Residence time 0 1 2 1 2
Asphaltenes
S (%wt) 6.84 7.02 6.78 6.74 5.72
N (%wt) 1.16 0.92 0.88 0.96 0.89 Ni (wppm) 202 158 160 202 157 V (wppm) 628 488 463 497 370
Resins
S (%wt) 6.05 5.42 4.99 4.69 4.12
N (%wt) 0.75 0.71 0.79 0.76 26 Ni (ppm) 25 33 33 27 78 V (wppm) 107 106 94 93 0.71
Aromatics S (%wt) 3.87 3.78 3.66 2.54 1.69
69
Table 23 – Percentage of each SARA fraction in Djeno residue, for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 1 2 1 2
S 18.8 20.2 20.7 21.3 21.4
A 28 30.8 31.7 33.7 34.9
R 47.9 44.5 41.1 43.1 39.2
A 5.2 4.5 4.4 3.9 4.4
Table 24 – Heteroatoms composition of Djeno residue for different residence times in the HDM and HDS reactor.
6.1.4 Oural
Table 25 – Density of Oural residue for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 2 1 2
Density (kg/m 3) 1004 995 990 989
HDM reactor HDS reactor
Residence time 0 1 2 1 2
Asphaltenes
S (%wt) 0.52 0.52 0.49 0.48 0.45
N (%wt) 1.44 1.5 1.52 1.58 1.43 Ni (wppm) 479 414 413 500 374.4 V (wppm) 47 50 55.5 39 56.5
Resins
S (%wt) 0.65 0.4 0.38 0.37 0.29
N (%wt) 1.25 1.2 1.21 1.27 1.16 Ni (wppm) 125 120 108 112 96 V (wppm) 16 13 11 12 11
Aromatics S (%wt) 0.52 0.27 0.31 0.19 0.38
70
Table 26 – Percentage of each SARA fraction in Oural residue, for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time (h) 0 2 1 2
S 11.1 13.6 15.5 18.2
A 43.5 44.8 46.5 47.7
R 39.8 37.1 33.2 29.6
A 5.7 4.5 4.8 4.5
Table 27 – Heteroatoms composition of Oural residue for different residence times in the HDM and HDS reactor.
HDM reactor HDS reactor
Residence time 0 2 1 2
Asphaltenes
S (%wt) 3.24 2.53 3.16 2.84
N (%wt) 1.37 1.31 1.38 1.38 Ni (wppm) 336 287 374 259.9 V (wppm) 1006 765 886 591.8
Resins
S (%wt) 3.17 2.55 2.52 1.9
N (%wt) 1.18 1.12 1.17 1.22 Ni (wppm) 77 52 52 41 V (wppm) 283 144 154 114
Aromatics S (%wt) 3.11 3.1 2.11 1.31
71
6.2 GPC curves
The following figures present the GPC curves for the resins and asphaltenes in each residue
The following chapters present the results of simulations for different residues using the
parameters optimized for the HDM + HDS model with Buzurgan used residue as a reference. In these
figures, the solid lines represent the model output for the HDM section while the dashed lines
represent the model output for the HDS section. The experimental data is represented by the filled and
unfilled marks, for the HDM and HDS sections, respectively.
6.3.1 Arabian Light
(a) (b)
Figure 42 – HDM +HDS model: optimized model output vs. experimental data (Arabian Light) for: (a) nickel removal; (b) vanadium removal.
(a) (b)
Figure 43 - HDM +HDS model: optimized model output vs. experimental data (Arabian Light) for: (a) sulfur removal; (b) nitrogen removal.
74
(a) (b)
Figure 44 - HDM +HDS model: optimized model output vs. experimental data (Arabian Light) for: (a) total lump fractions; (b) overall heteroatom removal.
6.3.2 Djeno
(a) (b)
Figure 45 – HDM +HDS model: optimized model output vs. experimental data (Djeno) for: (a) nickel removal; (b) vanadium removal.
75
(a) (b)
Figure 46 – HDM +HDS model: optimized model output vs. experimental data (Djeno) for: (a) sulfur removal; (b) nitrogen removal.
(a) (b)
Figure 47 - HDM +HDS model: optimized model output vs. experimental data (Djeno) for: (a) total lump fractions; (b) overall heteroatom removal.
76
6.3.3 Oural
(a) (b)
Figure 48 – HDM +HDS model: optimized model output vs. experimental data (Oural) for: (a) nickel removal; (b) vanadium removal.
(a) (b)
Figure 49 – HDM +HDS model: optimized model output vs. experimental data (Oural) for: (a) sulfur removal; (b) nitrogen removal.
77
(a) (b)
Figure 50 - HDM +HDS model: optimized model output vs. experimental data (Djeno) for: (a) total lump fractions; (b) overall heteroatom removal.
78
Reference List
1 Ancheyta, J. et al. (2003). Changes in Asphaltenes Properties during Hydrotreating of Heavy Crudes. Energy & Fuels, 17, 1233-1238.
2 Ancheyta, J., Speight, J.G. (2007). Hydroprocessing of heavy oils and residua. Boca Raton : CRC press,
3 Andersen, S.I., Speight, J.G., (2001). Petroleum resins: Separation, character, and role in petroleum. Petroleum Science and Technology, 19, 1-34.
4 Bellos, G.D. et al. (2005). Modelling of the performance of industrial HDS reactors using a hybrid neural network approach. Chemical Engineering and Processing, 44, 505-515.
5 Callejas, M., Martinez, M., (1999). Hydrocracking of Maya Residue. Kinetics and Product Yield Distributions. Industrial & Engineering Chemistry Research, 38, 3285-3289.
6 Castex, H., (1985). Résines et Asphaltènes: évolution en fonction des types de matière organique et de leur enfouissement. Revue de l'Institut Français du Pétrole, 40, 169-189.
7 Cooper, B.H. et al. (1986). Technology: hydroprocessing conditions affect catalyst shape selection. Oil Gas J., 8, 39-44.
8 Corella, J. et al. (1988). Some intrinsic kinetic equations and deactivation mechanism leading to deactivation curves with a residual activity. Industrial & Engineering Chemistry Research, 27, 375-381.
9 Egorova, M. (2003). Study of Aspects of Deep Hydrodesulphurization By Means of Model Reactions, DISS. ETH nº 15376. Swiss Federal Institute Of Technology, Zurich.
11 Ferreira, C. et al. (2009). Modeling residue hydrotreating. CHEMICAL ENGINEERING SCIENCE, In Press, Corrected Proof,
12 Fornasiero, F. et al. (2005). Multicomponent diffusion in highly asymmetric systems. An extended Maxwell-Stefan model for starkly different-sized, segment-accessible chain molecules. Macromolecules, 38, 1364-1370.
13 Froment, G.F., Bischoff, K. (1990). Chemical reactor analysis and design.
14 Furimsky, E., Massoth, F.E., (1999). Deactivation of hydroprocessing catalysts. Catalysis Today, 52, 381-495.
15 Gonzalez, G. et al. (2006). Asphaltenes precipitation from crude oil and hydrocarbon media. Energy & Fuels, 20, 2544-2551.
16 Haulle, F.X. (2002). Thèse de l'université Paris VI - Modélisation cinétique de l'hydrotraitement en lit fixe des résidus pétroliers - Étude de la réactivité des composés soufrés.
17 Haulle, F. X. and Kressmann, S. Kinetic Modeling of Residue Desulfurization: lumping approach on sulfur compounds in heavy oil fractions. 2002 AIChE Spring Meeting, New Orleans, LA, March, 10-14. Presentation at the 2002 AIChE Spring Meeting, New Orleans, LA, March, 10-14 . 2002. Ref Type: Conference Proceeding
79
18 Iliuta, I. et al. (2006). Simulating simultaneous fines deposition under catalytic hydrodesulfurization in hydrotreating trickle beds - does bed plugging affect HDS performance? CHEMICAL ENGINEERING SCIENCE, 61, 1321-1333.
19 Jaffe, S.B. et al. (2005). Extension of structure-oriented lumping to vacuum residua. Industrial & Engineering Chemistry Research, 44, 9840-9852.
20 Jost, K. et al. (1985). Chromatographia, 20, 700-
21 Kressmann, S. et al. (1998). Recent developments in fixed-bed catalytic residue upgrading. Catalysis Today, 43, 203-215.
22 Krishna, R., Wesselingh, J.A., (1997). The Maxwell-Stefan approach to mass transfer. CHEMICAL ENGINEERING SCIENCE, 52, 861-911.
23 Le Page, J.F. et al. (1992). Resid and heavy oil processing. Éditions Technip, Paris, pp.
24 Leinekugel-le-Cocq, D. (2004). Contribution a la modélisation dynamique simplifiée d'un procédé d'adsorption modulée en pression (P.S.A). Thesis of Université Claude Bernard Lyon 1.
25 LeLannic, K and Guibard, I. Kinetic modelling of residue hydrodesulfurization. Conference on Petroleum Phase Behavior and Fouling, Amsterdam . 2005. Ref Type: Conference Proceeding
26 LeLannic, K. (2006). Désulfuration profonde de résidus pétroliers: Elaboration d'un modèle cinétique. Thesios of École Normale Supérieure de Lyon.
27 Lemos, F. et al. (2002). Reactores Químicos. IST Press,
28 Leprince, P. (2001). Conversion Processes. Éditions Technip, Paris, pp.
29 Li, J., Carr, P.W., (1997). Accuracy of Empirical Correlations for Estimating Diffusion Coefficients in Aqueous Organic Mixtures. Analytical Chemistry, 69, 2530-2536.
30 Merdrignac, I., Espinat, D., (2007). Physical-chemistry characterization of petroleum fractions: the state of the art. Oil & Gas Science and Technology, 62, 7-32.
31 Merdrignac, I. et al. (2004). Size exclusion chromatography: Characterization of heavy petroleum residues. Application to resid desulfurization process. Petroleum Science and Technology, 22, 1003-1022.
32 Murgich, J., (2003). Molecular simulation and the aggregation of the heavy fractions in crude oils. Molecular Simulation, 29, 451-461.
33 Neurock, M., Libanati, C., and Klein, M. T. Modeling asphaltenes reaction pathways: Intrinsic chemistry. AIChE Symposium Series. Fundamentals of Resid upgrading 85[273]. 1989. Ref Type: Conference Proceeding
34 Oelderik, J.M. et al. (1989). Progress in the catalysis of the upgrading of petroleum residue. A review of 25 years of R&D on Shell's residue hydroconversion technology. Applied catalysys, 47, 1-24.
36 Oyekunle, L.O., Kalejaiye, B.O., (2003). Kinetic modeling of hydrodesulphurization of residual oils. I. Power law model. Petroleum Science and Technology, 21, 1475-1488.
37 Pfeiffer, J. P., and P. N. Saal. 1940. J.Phys.Chem. 44.
80
38 Rodriguez, M.A., Ancheyta, J., (2004). Modeling of hydrodesulfurization (HDS), hydrodenitrogenation (HDN), and the hydrogenation of aromatics (HDA) in a vacuum gas oil hydrotreater. Energy & Fuels, 18, 789-794.
39 Sheu, E.Y., (2006). Small angle scattering and asphaltenes. Journal of Physics-Condensed Matter, 18, S2485-S2498.
40 Speight, J.G. (1999). The Chemistry and Technology of Petroleum. New York, pp.
41 Takatsuka, T. et al. (1996). A catalyst deactivation model for residual oil hydrodesulfurization and application to deep hydrodesulfurization of diesel fuel, Deactivation and Testing of Hydrocarbon-Processing Catalysts, ACS Symp., in: 414-427.
42 Tayakout, M. Note Interne IFP - Caractérisation des charges de résidus. Limitation Diffusionnelle. Cas de l'hydrodémétallisation. 2006. Ref Type: Report
43 Toulhoat, H. et al. (2005). THERMIDOR: A new model for combined simulation of operations and optimization of catalysts in residues hydroprocessing units. Catalysis Today, 109, 135-153.
44 Verstraete, J.J. et al. (2008). Molecular Reconstruction of Heavy Petroleum Residue Fraction. CHEMICAL ENGINEERING SCIENCE, Submitted,
45 Web Article, dieslnet. com. Emission Standards » European Union. http://www.dieselnet.com/standards/eu/ld.php . 2009. Ref Type: Internet Communication
46 Yen, T.F., (1972). Present status of the structure of petroleum heavy ends and its significance to various technical applications. American Chemical Society, Division of Fuel Chemistry, Preprints,
47 Zhao, B., Shaw, J.M., (2007). Composition and size distribution of coherent nanostructures in Athabasca bitumen and Maya crude oil. Energy & Fuels, 21, 2795-2804.