Você está na página 1de 9

Chemical Engineering Science 63 (2008) 1683 1691

www.elsevier.com/locate/ces

Modeling of mass transfer and thermal cracking during the coking of


Athabasca residues
Ramin Radmanesh a , Edward Chan b , Murray R. Gray c,
a DuPont Canada Inc., Research and Business Development, Kingston, Ont., Canada K7L 5A5
b Syncrude Canada Ltd., Edmonton Research Centre, 9421- 17th Avenue, Edmonton, Alta., Canada T6N 1H4
c Department of Chemical and Materials Engineering, University of Alberta, Edmonton, Alta., Canada T6G 2G6

Received 25 March 2007; received in revised form 16 August 2007; accepted 5 November 2007
Available online 21 November 2007

Abstract
The kinetics of thermal cracking of lms of vacuum residue from Athabasca bitumen in the temperature range of 457.530 C was modelled
with liquid-phase mass transfer, reaction-dependent uid properties, and coke formation by reaction of cracked products in the liquid phase.
Previous investigations on the thermal cracking of vacuum residue in thin lms showed that at low lm thickness ( 20 2 m) the coke yield
was insensitive to the temperature and heating rate for thin lms of bitumen. The coke yield increased with the thickness of the initial lm,
in the range from 20 to 80 m (2 m). At the same time, the viscosity of the reacting liquid increased rapidly with time, which would slow
down the diffusion of products inside the lm. This coupling of transport and reaction would enhance the formation of coke by increasing
the rate of recombination reactions. The concept of intrinsic coke is used in a new kinetic model to account for the minimum observed coke
formation in thin lms. With increasing lm thickness, the increasing yield of extrinsic coke is modelled through the change in uid properties
as a function of extent of reaction, which reduces the rate of diffusion in the reacting liquid phase. The model was able to properly account
for the insensitivity of coke yield in thin lms to reaction temperature and the dependence of coke yield on the thickness of the liquid lm.
2007 Elsevier Ltd. All rights reserved.
Keywords: Heavy oil; Bitumen; Thermal cracking; Coking; Mass transfer; Diffusion

1. Introduction
The number of reactions involved in thermal cracking of bitumen and vacuum residues are enormously large due to the
existence of numerous distinct molecules in the petroleum feedstock. Modern analytical measurements have shown that number of unique molecules in a vacuum residue is in the order of
105 106 (Rahimi and Gentzis, 2006). Lumped kinetic schemes
are useful in aggregating a large number of molecules into few
categorically dened lumps. These lumps can be classied according to their boiling point, solubility, or adsorption properties. Despite its simplistic approach, lumped kinetic models are
still used in predicting the behaviour of the petroleum resids
during chemical processing. For instance, Gray et al. (2004)
showed that micro carbon residue (MCR) content in the vapour
Corresponding author. Tel.: +1 780 492 7965; fax: +1 780 492 2881.

E-mail address: murray.gray@ualberta.ca (M.R. Gray).


0009-2509/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2007.11.019

product correlated with the content of material boiling over


650 C.
Delayed and uid-bed coking continue to be the major process route for producing distillates from vacuum residues. Delayed coking is characterized by less severity in temperature
and thus lower liquid product yields. Fluid coking offers rapid
heating rate and higher severity conditions, which, in turn, results in lower coke yield and higher liquid and gas production.
Most studies on coke formation and kinetics of coking have
been carried out at a temperature range and operating conditions which are more relevant to delayed coking (Wiehe, 1993;
Schabron et al., 2002; Yue et al., 2004). The coke formation in
delayed coking is dominated by phase separation, which results
in an induction period in the formation of coke. Induction time
in delayed coking is an inherent characteristic of the resid that
varies from minutes to hours depending on the temperature.
Wiehe (1993) developed a kinetic model that included phase
separation of asphaltenes.

1684

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

While the reactionseparation kinetic models have been


proven to work well for delayed coking, the role of the induction time is much less signicant at the more severe conditions
of uidized-bed coking where the reactions occur in thin lms
on the surface of heat-carrier particles in a uidized bed (Gray,
2002). In thin lms of vacuum residue at 500 and 530 C,
rapid vaporization, rapid cracking, and rapid coke formation
combined to eliminate the observable induction period (Gray
et al., 2004). Gray et al. (2004) proposed a reaction scheme
based on the lumped approach for devolatilization and cracking of the bitumen lm at severe reaction conditions of up to
530 C; but the mass transfer in the kinetic model was highly
simplied and was only limited to vapour phase mass transfer.
Recent experimental data on adhesion forces and viscosity of
Athabasca vacuum residue during thermal cracking revealed
a dramatic increase in viscosity by four orders of magnitude
after only 10 s of cracking at 503 C (Aminu et al., 2004). This
substantial increase in viscosity implies a concomitant drop
in the diffusivity in the reacting lm as the cracking reaction
proceeds. Consequently, the mass transfer within the liquid
phase must be important during the coking reactions.
Previous reports (Gray et al., 2004, 2007) showed that at
low lm thickness (=20 2 m) the coke yield was insensitive
to the temperature in the range of 457530 C. This limiting
lm thickness was selected from experimental considerations;
thinner lms are possible but yields could not be determined
accurately. It also revealed that by increasing the initial lm
thickness from 20 to 80 m the coke yield increases, due to the
retention of components in the coke phase. This coupling of
coke yield and lm thickness reveals the importance of mass
transfer inside the reacting liquid lm. Under these conditions,
the role of heating rate was also insignicant.
This paper presents an improved revised kinetic model, building on the simple lumped approach of Gray et al. (2004), to account for important experimental observations: that coke yield
from thin lms is insensitive to temperature, that the physical
and transport properties of the reacting liquid change rapidly
with time, and that the yield of coke increases with the thickness of the liquid lm.
2. Theory and model denition
Viscositydiffusion relationship: The viscosity of liquid bitumen lm during the coking increases rapidly (Aminu et al.,
2004). The diffusivity of a solute in a solution is known to
be inversely proportional to the viscosity of solution. Fig. 1
demonstrates the variation of experimental viscosity and estimated diffusivity of the heavy residue fraction (boiling over
650 C) inside a reacting lm of Athabasca vacuum residue at
503 C. The diffusivity was estimated using a modied form of
the WilkeChang equation (Reid et al., 1977):
DAB = 7.4 108

(M)1/2 T
2/3

m VA0.6

(1)

where M is the average molecular weight of the feed, T is


the temperature in Kelvin, m is the viscosity of the reacting

Fig. 1. Viscosity variation of a reacting AVR lm of 20 2 m thickness with


time during the coking at T =503 C (data of Aminu et al., 2004). Diffusivity
of the heavy residue inside of a reacting bitumen lm was calculated by
modied version of WilkeChang equation (1). Lines show the trends of the
data.

Fig. 2. Schematic presentation of a bitumen lm subjected to thermal cracking.

lm and VA is the molar volume of each diffusing lump. The


exponent of 23 for viscosity was selected for viscous liquids
(Hiss and Cussler, 1973; Tyrrell, 1981; Kulkarni and Mashelkar,
1983). The rapid decrease in the diffusivity illustrated in Fig. 1
implies that the rate-controlling step for release of products
from the liquid lm will shift from initial kinetic control to
liquid-phase mass transfer.
Controlling transport processes: The experiments on
Athabasca vacuum residue were conducted on thin lms of liquid supported on a at surface of a Curie-point alloy (Fig. 2).
The Biot number indicates the relative importance of internal
mass (or heat) transfer to the external mass (or heat) transfer:
Bim =

KG 
,
D

(2)

h
,
(3)

where Bim and Bit are the Biot numbers for mass and heat transfer, respectively,  is the thickness of the reacting lm, KG and h
are the heat and mass transfer coefcient in the vapour phase, respectively, D is the diffusion coefcient of a lump component in
the reacting lm and  are the thermal conductivity coefcient
of the reacting lm. Small Biot numbers (Bi>1) indicate that
Bit =

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

1685

The second mechanism responsible for coke formation is


polymerization and recombination of lighter fractions (light
residue and gas oil) to form a second class of coke or extrinsic
coke. These recombination reactions would lead to formation
of large aromatic cores by consolidation of smaller aromatic
groups. This mechanism was demonstrated by Dutta et al.
(2001) who used 13 C-labelled model compounds with boiling
points in the gas oil range. The incorporation of this 13 C increased with the thickness of the reacting liquid lm, due to recombination reactions. The amount of this extrinsic coke would
increase as the residence time of the cracked products in the
liquid lm increases.
Incorporating these observations gave the kinetic network
illustrated in Fig. 4. All the cracking reactions are assumed to
be rst order, while the recombination to form extrinsic coke
was second order. The reaction rates for each of these lumped
fractions indicated in Fig. 4 can then be expressed as follows:
Fig. 3. The coke yield for different feeds vs the amount of heavy residue
(+650 C; data of Gray et al., 2004). The gure shows that the amount of
coke for a reacting bitumen lm of 20 2 m thickness is correlated with
the initial heavy residue fraction.

external mass or heat transfer are rate-limiting. Estimated parameters indicated that Bim increased from the initial value of
Bi = 5.3500 after 18 s, due to the rapid drop in the diffusion
coefcient. This rise of Bim suggests that internal mass transfer becomes the rate-limiting step soon after commencing the
cracking reactions. In contrast, the thermal Biot number (Bit )
for a thin lm of 20 2 m was of order Bit = 2 104
at all times, because the thermal conductivity of heavy oil
(0.11.0.12 W/m/K, Speight, 2001) would decrease little during the formation of coke (0.4.0.9 W/m/K, Michaelian et al.,
2002). In this regime of heat transfer, therefore, the lm temperature will be uniform at the temperature of the Curie-point
alloy.
2.1. Kinetic model
Intrinsic and extrinsic coke formation: Following the description of Wiehe (2000) on the mechanism of coke formation during the coking process, coke formation is considered
to originate from two different mechanisms. Intrinsic coke formation results from large aromatic cores in the feed, which do
not crack and which are too large to vaporize. Aromatic species
with more than ve rings in their structure have a high boiling point and the energy required for bond cleavage is quite
large. The heavy residue (+650 C) is likely to have more aromatics than the lower-boiling fractions, therefore, this fraction
gives rise to intrinsic coke. This assumption is also supported
by Fig. 3, which shows that nal coke yield in thin-lm experiments is correlated with the heavy residue fraction for the
three different feeds listed in Table 1. The correlation between
the intrinsic coke yield and the fraction of heavy residue shows
a slope of k = 0.26. Studies on other crude oils would be required to determine if the yield of intrinsic coke were sensitive
to chemical structure of the heavy residue.

r1 = k1 (w1 mcore ),

(4)

r2 = S12 k1 (w1 mcore ) k2 w2 k6 w2 w3 ,

(5)

r3 = S13 k1 (w1 mcore ) + S23 k2 w2 k6 w2 w3 ,

(6)

r4 = k1 S14 (w1 mcore ) + k2 S24 w2 ,

(7)

r5 = k5 (mcore w5 ),

(8)

r6 = k6 w2 w3 .

(9)

In these equations mcore = Y w1,0 , representing the aromatic


core fraction in the heavy residue that gives rise to intrinsic
coke. The quantity of mcore is correlated with the heavy residue
fraction in the feed, as portrayed in Fig. 4 which gives the value
of Y . S is the selectivity coefcient for different reactions, w is
the weight fraction of each lump fraction and k is the rst order
reaction constant. Eq. (5) is a mathematical description of the
explanations given previously on the mechanism of intrinsic
coke formation. Since mcore cannot be thermally cracked, it
should be excluded from total mass of heavy residue fraction
which undergoes thermal cracking. Eqs. (8) and (9) describe
the rate of coke production through intrinsic and extrinsic coke
formation, respectively.
The estimates for the mass-transfer Biot number indicated
that the internal mass transfer inside the sample quickly becomes dominant as the reaction proceeds. Therefore, the general form of the mass transfer-reaction was used for the cracking
of the components in the liquid lm:


jwi
j
jwi
=
Di
+ ri ,
(10)
jt
j
jz
where i represents each of the lumped fractions in the kinetic
model, wi is the weight fraction of component i in the liquid
phase and Di is the diffusivity of that fraction in the solution.
The term ri is the net rate of formation of component i and
is evaluated at the Curie-point temperature of the substrate for
the reacting lm (Bit >1). Eq. (10) can be simplied for coke

1686

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

Table 1
Properties of feeds derived from Athabasca bitumen

MCR, wt.%
Molecular weight
Density, kg/m3
Boiling fraction
Light gas oil
75 C
Heavy gas oil
75.524 C
Light residue
524.650 C
Heavy residue
+650 C

Fluid coker scrubber bottoms


vacuum residue (SBR)

Athabasca vacuum residue (AVR)

24.6
620
1102

27.8
700
1087

0
30
37
33

0
10
40
50

Fluid coker pilot vacuum


residue (CPR)

600
1060
0
15
60
25

interface is given by:


Ni = KGy (yi yi ),

(12)

where yi is the mole fraction of component i in the nitrogen


stream, yi is the concentration of the component at the interface
of the two phases and KGy is the mass transfer coefcient in the
gas phase estimated from the correlation of Tosun (2002) for
Sherwood number for mass transfer to gas owing over a at
plate. Since the small volume of vapour products leaving the
liquid phase was immediately diluted in nitrogen ow, yi 0.
yi is related to the mole fraction of component i in the liquid
phase by the equilibrium constant, which indicates volatility of
each lump:
Ki =
Fig. 4. Reaction network for thermal cracking of bitumen.

(lumped components 5 and 6) where D5 = D6 = 0. The diffusivity is calculated by viscosity data and using WilkeChang
equation as presented by Eq. (1). Viscosity and consequently
diffusivity change with time as more coke is produced. The following empirical correlation was introduced to relate the viscosity of the solution to coke concentration in the liquid phase:
r =


= 1 + []Xc .
0

(11)

In Eq. (11), Xc is the coke concentration in the liquid lm, 


and 0 are viscosity at any coke concentration and at Xc = 0,
respectively, and [] and  are constant values calculated by
tting the viscosity data of Aminu et al. (2004). The initial
viscosity (0 ) was approximately 2 mPa s (Aminu et al., 2004),
and was assumed to be insensitive to reaction temperature in
comparison to the effect of conversion (Fig. 1). Additional data
would be required to determine any temperature dependence of
[] in Eq. (11). The best t parameters to the data of Fig. 1 in
a plot of ln(r 1) vs ln(Xc ) were [] = 8.08 104 mPa s and
 = 1.8 (R 2 = 0.99).
Evaporation of the product from the liquid phase to the gas
phase takes place at the interface of the liquid and gas phases
as illustrated in Fig. 2. The mass ux leaving the lm at the

yi
.
xi

(13)

After substituting Eq. (13) into Eq. (12) and converting to


a mass basis, the boundary condition can be simplied to
Eq. (14) for the ux at z = :
Fi,z= = Di

jwi
= KGy,i Ki wi Mi ,
jz

(14)

where  is the density of the feed. The mass ux of all components was zero at the bottom interface (Fi,z=0 = 0).
As the thermal cracking of heavy material in the lm proceeds and more vapour products leave the sample, the bitumen
lm starts to shrink. A total mass balance for the reaction lm
gives the rate of shrinkage:
d 
(KGy Kw i (t)Mi ).
=
dt
4

(15)

i=1

This equation should be solved along with Eq. (10) for all the
lumped fractions in the kinetic scheme to provide the mass
fraction prole of each component in the reacting lm. This
prole can then be integrated throughout the lm to obtain an
average mass fraction at each time step.
The kinetic parameters were estimated by minimizing the objective function, which was square of errors between the model
and experimental value. This objective function was expressed

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

as follows:
SSQ1 =

presented as


2
 w model w exp
i
i
exp

1687

Max(wi )

(16)

exp

where wimodel and wi denotes the calculated and experimental values of w, respectively. The objective function was
normalized by the maximum value of each fraction in the experimental data in order to have an unbiased optimization. The
Pattern search function of MATLAB, in conjunction with the
non-linear transient solver of COMSOL 3.2, were employed simultaneously to nd a set of kinetic parameters that minimized
the aforementioned objective function. The MATLAB function
served as the optimization routine, while COMSOL solved the
governing equations (Eqs. (10)(15)) at each iteration.
3. Results and discussion
3.1. Model parameters
The model parameters were rst calculated for the thermal
cracking of scrubber bottoms residue (SBR, Table 1). Table 2
presents the estimated parameters that gave the best t of the
model to the experimental data from Gray et al. (2004). The activation energy for the cracking of heavy residue and heavy gas
oils were in the reported range for heavy hydrocarbons thermal
cracking (Khorasheh and Gray, 1993). The kinetic constants for
extrinsic coke formation (E6 and A6 ) were further adjusted using data for thick lms (20.80 m) as discussed below, in order
to obtain a consistent t to both thin-lm and thick-lm data.
Table 2 also lists the equilibrium constant for each lump
fraction. The equilibrium constants were initially calculated
by ash calculation at the reaction temperature using the
PengRobinson equation of state (Gray et al., 2004). However,
the distribution of lumped components in the vapour phase predicted by the model was not satisfactory when using the equilibrium constants given by PengRobinson equation of state,
possibly due to errors in extrapolating the estimates for critical
properties and acentric factor to the non-distillable heavy fractions. In order to improve the results, the equilibrium constants
for distillates (K4 ) were calculated from the PengRobinson
equation of state, then the other Ki values were treated as
adjustable parameters in the model. As the initial trial using
equilibrium constants derived from PengRobinson equation
of state revealed, these parameters only affect the distribution
of products in the vapour phase and have a minor effect on the
coke yield and reactant yield inside the reacting lm.
The range for the kinetic parameters presented in Table 2
shows the 97.7% condence limit. The optimum value of the
square of error for the parameters presented in Table 2 was
SSQ1 = 0.38. The 97.7% condence was calculated based on
the analysis of variance (Constantinides and Mostou, 1999).
According to this criterion, the variance due to the lack of
t, which was presented by Eq. (16), must be small compared to the variance of the experimental error. A normalized
form of this latter error which is consistent with SSQ1 can be

SSQ2 =

 exp
exp 2
p 
n

wij wi,ave
exp

i=1 j =1

wi,max

(17)

In Eq. (17), p is the number of experimental points for each


component and n is the number of repeated experiments. SSQ2
for the experimental data at T = 503 C was 0.73 which was
larger than the SSQ1 . A computational test was performed on a
selected range of kinetic parameters in the vicinity of the optimum point to identify an acceptable range for the kinetic parameters based on the above mentioned criterion. A set of kinetic
parameters were chosen within this selected range randomly
using a random number generator with a uniform distribution
(RAND routine of MATLAB). The model was then solved for
each set of the randomly chosen parameters. This procedure
(i.e. random selection of a set of parameters and solving the
model for that set) was repeated 1000 times and the number of
times that resulted in SSQ1 > SSQ2 was identied. The condence for that range was dened as the number of times when
the condition of SSQ1 > SSQ2 was violated to the total number of random trials. If the condence was less than 95% the
selected range for each parameter was narrowed down.
3.2. Model predictions
The model parameters calculated at the intermediate temperature of T = 503 C for SBR were used to predict yields
of different fractions at other temperatures and for other feeds.
Fig. 5 compares the model result with experimental data on
coke fractional yield of coke at three different temperatures for
SBR feed. The fractional yield shown in this gure is the summation of intrinsic and extrinsic coke. While intrinsic coke remains independent of temperature, the extrinsic coke increases
with increasing temperature. As Fig. 5 shows, however, the nal total yield of coke at thin-lm thicknesses does not signicantly increase with increasing temperature, which means that
coke forms through intrinsic mechanism as described above.
The rate of coke formation was sensitive to temperature, but not
the nal yield at long reaction times. At higher lm thicknesses,
however, the extrinsic coke formation becomes important due
to higher residence time of gas oils in the liquid lm and this
results in higher total coke yield. As mentioned above, this is
supported by other experimental work (Dutta et al., 2001). The
higher coke yield of delay cokers compared to the uid cokers
is also another practical example that shows how mass transfer
limitations inside the reacting liquid can promote the coke formation reactions. The other parameter that can affect the yield
of the extrinsic coke is the residence time of the reactants inside the lm. This parameter is controlled by the mass transfer
across the lm and thus is very sensitive to the diffusivity of
reactants inside the sample and the lm thickness.
One of the major features of this kinetic approach is that it is
consistent with the insensitivity of the nal coke yield to temperature at low lm thicknesses. As Fig. 5 shows, the nal coke
yield at different temperatures remains almost invariant. The
model is less successful at the low temperature of T = 457 C.

1688

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

Table 2
Estimated kinetic and thermodynamic parameters
Reaction

Apparent activation energy,


Ea,i (kJ/mol)

Apparent pre-exponential factor,


log(Ai ), (s1 )

Stoichiometric coefcients

Cracking of heavy residue (1)

230 1

14.00 0.5

Cracking of light residue (2)

188 1

11.00 0.5

Intrinsic coke
formation (5)
Extrinsic coke formation (6)

33.7 0.4

1.0 0.2

S12 = 1
S13 = S14 = 0
S23 = 0.2
S24 = 0.8

99.6 1

5.0 0.2

T = 475 C
0.0226
0.18
0.5
26

T = 503 C
0.08
0.6
1.7
28

T = 530 C
0.2
0.90
2.7
31

Equilibrium constants
K1
K2
K3
K4 a
a Equilibrium

constant determined using PengRobinson equation of state in HYSYS software.

Fig. 5. Coke yield at different temperatures for a SBR bitumen lm of 20 m.


Symbols are the experimental data (Gray et al., 2004) and lines are model
results.

An induction time in the formation of coke at this temperature can be observed, similar to the kinetics of delayed coking
(Wiehe, 1993). The model, however, works well at T > 500 C,
which is the temperature range of uid-bed cokers.
Fig. 6 shows the fractional yield for heavy residue, light
residue and heavy gas oil in the vapour and extract phases at
T = 503 C. The boiling point of the heavy residue is much
higher than the reaction temperature. This lump is therefore
more prone to thermal cracking than to evaporation.
3.3. Model predictions for different feeds
Fig. 7 shows the prediction of the model for coke formation
for three feeds listed in Table 1 at T = 503 C. The kinetic
parameters were derived for SBR as presented in Table 2. The
data of Fig. 8 show the experimental data and model prediction

Fig. 6. Fractional yield of heavy residue, light residue and heavy gas oil
liquid (extract) and vapour (distillate) product for SBR feed at T = 503 C.
Symbols are the experimental data (Gray et al., 2004) and lines show the
model prediction.

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

1689

Fig. 7. Coke fractional yield prediction for three different feeds at T =503 C.
Symbols are the experimental data (Gray et al., 2004) and lines show the
model prediction.

on the total liquid and vapour yield for CPR. Although, this
feed had a different thermal history, the parameters derived
from tting of data for SBR were able to predict the yield of
coke and different lumped products.
3.4. Model results on reactant concentration prole
The model predictions for the concentration of each fraction
inside the reacting lm illustrate the importance of concentration gradients as the lm thickness is increased. Fig. 9 shows
the weight fraction prole of the light residue for two different lm thicknesses at different times inside the reacting lm.
In both thin- and thick-lm cases, as the reaction proceeds the
escape of reactants from the lm becomes more difcult due
to the abrupt drop in diffusivity, which leads to a concentration
gradient almost immediately upon starting the reactions. This
result is in accord with large estimated value of Bim . The larger
diffusion length at lm thickness of  = 80 m leads to a lower
evaporation rate and even greater gradients (Fig. 9b).
3.5. Model prediction for thick lms
As mentioned earlier, as lm thickness increases, the dominant mechanism of mass transfer inside the lm shifts from
diffusion to a bubbling regime. By increasing the ambient pressure, the growth of the vapour bubbles can be retarded and
the bubbling transport suppressed. This technique was used
by Gray et al. (2007) to measure coke yield as a function of
lm thickness without bubble formation. The experimental data
presented in Fig. 10 show a higher coke yield at higher reactor pressure for AVR feed when the bubble formation has
been suppressed and the dominant mass transfer mechanism is
diffusion.
The kinetic parameters presented in Table 2 were determined
by tting the model to data for coking of a thin lm of SBR,
where the effect of mass transfer is minimized. In the thin-lm
condition, the extrinsic coke formation does not play an impor-

Fig. 8. Model results on total liquid (extract) and vapour (distillate) for CPR
feed at different temperature. Symbols are the experimental data (Gray et al.,
2004) and lines show the model prediction.

tant role; instead the coke is formed through the intrinsic coke
formation mechanism, which is insensitive to the temperature.
The kinetic parameters for the extrinsic coke obtained for thin
lms should therefore be determined from data for reaction of
thick lms, where the extrinsic coke formation becomes important. The extrinsic coke kinetic parameters, as presented in
Table 2, gave the best t to both thin- and thick-lm conditions. The model prediction shown in Fig. 10 is derived with
the model parameters calculated for the thin lm of SBR, with
adjustment of the rate parameters for extrinsic coke formation.
The increase in the coke yield is due to the extrinsic coke formation, resulting from the increased trapping of light residue and
gas oils inside the lm, as shown in Fig. 9b. The model gives
reasonable quantitative agreement to the high-pressure data.
3.6. Model sensitivity on diffusion coefcient
The WilkeChang equation is commonly used to estimate
diffusion coefcient in dilute solutions. To evaluate the effect
of error in the estimate of the diffusion coefcient on the model

1690

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

Fig. 10. Coke yield as a function of lm thickness for VTB feed at T =503 C
and 0.65 MPa. Points represent the experimental data (Gray et al., 2007) and
the line is the model result.

extrinsic coke formation improved the tting of the model to


data for the yield of coke and liquid product fractions. The combination of mass transfer with time-dependent diffusion coefcient and the modied reaction scheme gave predictions that
were consistent with the increase in coke yield as the initial
lm thickness of vacuum residue was increased.
Notation

Fig. 9. Light residue fraction prole inside a reacting lm of SBR bitumen


(a) when the initial lm thickness is  = 20 m and (b) when the initial lm
thickness is  = 80 m.

prediction, a sensitivity analysis on the effect of diffusivity


given by modied version of WilkeChang equation (Eq. (1))
on the coke yield calculation by the model was performed. A
40% change in the diffusion coefcient changed the coke yield
by less than 0.5%. When the diffusion coefcient was reduced
by 80%, then the change in the coke yield was much larger, at
+7%. The model results, therefore, were relatively insensitive
to the estimates of the diffusion coefcients within the predicted
range of the WilkeChang equation.
4. Conclusions
Mass transfer inside the lm of reacting vacuum residue
was included in a general model for coke formation and product evolution. A reaction scheme which based on intrinsic and

A
Bim
Bit
D
Ea
F
h
k
K
KGy
mcore
M
n
p
S
SSQ
t
T
VA
w
Xc
y
Y

pre-exponential value, s1
mass transfer Biot number
thermal Biot number
diffusion coefcient
apparent activation energy, kJ/mol/K
ux of material on a mass basis, g/m2 /s
convective heat transfer coefcient, W/m2 /K
reaction constant, s1
equilibrium constant
mass transfer coefcient in gas phase expressed
as mole fraction, mol/m2 /s
mass fraction of aromatic core in heavy residue
molecular mass, g/mol
number of repeated experiments in Eq. (17)
number of experimental points in Eq. (17)
stoichiometric coefcient
sum of square errors as presented by Eq. (34)
time, s
reaction temperature, K
molar volume of solute in WilkeChang equation
mass fraction of a component in liquid phase
mass fraction of coke in the reacting lm
mole fraction of lump component in gas phase
amount of heavy fraction that converts to intrinsic
coke, slope in Fig. 3

R. Radmanesh et al. / Chemical Engineering Science 63 (2008) 1683 1691

Greek letters


[]
r


0


constant in Eq. (11)


thickness of the lm, m
constant in Eq. (24)
relative viscosity in Eq. (24)
heat conduction coefcient, W/m/K
viscosity of the bitumen lm, Pa s
viscosity of bitumenlm at t = 0
density of bitumen, kg/m3

References
Aminu, M.O., Elliott, J.A.W., McCaffrey, W.C., Gray, M.R., 2004. Fluid
properties at coking process conditions. Industrial & Engineering
Chemistry Research 43 (12), 29292935.
Constantinides, A., Mostou, N., 1999. Numerical methods for chemical
engineers with MATLAB applications. In: Prentice Hall International
Series in the Physical and Chemical Engineering Sciences.
Dutta, R.P., McCaffrey, W.C., Gray, M.R., Muehlenbachs, K., 2001. Use of
13 C tracers to determine mass-transfer limitations on thermal cracking of
thin lms of bitumen. Energy and Fuels 15 (5), 10871093.
Gray, M.R., 2002. Fundamentals of bitumen cooking processes analogous to
granulations: a critical review. Canadian Journal of Chemical Engineering
80 (3), 393401.
Gray, M.R., McCaffrey, W.C., Huq, I., Le, T., 2004. Kinetics of cracking
and devolatilization during coking of Athabasca residues. Industrial and
Engineering Chemistry Research 43 (18), 54385445.
Gray, M.R. , Le, T., Wu, X.A., 2007. Role of pressure in coking of thin lms
of bitumen. Canadian Journal of Chemical Engineering 85 (5), 773780.
Hiss, T.G., Cussler, E.L., 1973. Diffusion in high viscosity liquids. A.I.Ch.E.
Journal 19 (4), 698703.

1691

Khorasheh, F., Gray, M.R., 1993. High-pressure thermal cracking of


n-hexadecane. Industrial & Engineering Chemistry Research 32 (9),
18531863.
Kulkarni, M.G., Mashelkar, R.A., 1983. A unied approach to transport
phenomena in polymeric media. I. Diffusion in polymeric solutions, GELS
and melts. Chemical Engineering Science 38 (6), 925939.
Michaelian, K.H., Hall, R.H., Bulmer, J.T., 2002. Photoacoustic infrared
spectroscopy and thermophysical properties of Syncrude cokes. Journal of
Thermal Analysis and Calorimetry 69 (1), 135147.
Rahimi, P.M., Gentzis, T., 2006. The chemistry of bitumen and heavy oil
processing. In: Hsu, C.S., Robinson, P.R. (Eds.), Practical Advances in
Petroleum Processing, vol. II. Springer, Berlin, pp. 149186.
Reid, R.C., Prausnitz, J.M., Sherwood, T.K., 1977. The Properties of Gases
and Liquids. McGraw-Hill, New York.
Schabron, J.F., Pauli, A.T., Rovani Jr., J.F., 2002. Residual coke formation
predictability maps. Fuel 81 (17), 22272240.
Speight, J.G., 2001. Handbook of Petroleum Analysis. Wiley Interscience,
New York.
Tosun, I., 2002. Modeling in Transport Phenomena: A Conceptual Approach.
Elsevier Science B.V,
Tyrrell, H.J.V., 1981. Diffusion and viscosity in the liquid phase. Science
Progress 67 (266), 271293.
Wiehe, I.A., 2000. The enhanced microcarbon tester and other ideal
laboratory cokers. In: Symposium on Residuum/Asphaltene/Coke/Solids
Characterization in Petroleum Processing 220th National Meeting.
American Chemical Society.
Wiehe, I.A., 1993. Phase-separation kinetic model for coke formation.
Industrial & Engineering Chemistry Research 32 (11), 24472454.
Yue, C., Watkinson, A.P., Lucas, J.P., Chung, K.H., 2004. Incipient coke
formation during heating of heavy hydrocarbons. Fuel 83 (1112),
16511658.

Você também pode gostar