Chemical Engineering Science 61 (2006) 8028 – 8039

www.elsevier.com/locate/ces

A generalized model to predict the liquid–liquid equilibrium in the systems

furfural + lubricating oils
Baudilio Coto ∗ , Rafael van Grieken, José L. Peña1 , Juan J. Espada
Department of Environmental and Chemical Technology, School of Experimental Sciences and Technology (ESCET), Rey Juan Carlos University,
c/Tulipán s/n, 28933, Móstoles (Madrid), Spain

Received 22 May 2006; received in revised form 13 September 2006; accepted 14 September 2006
Available online 23 September 2006

Abstract
In the lubricating oil manufacturing process, the aromatic content of vacuum distillates is reduced by solvent extraction, frequently with
furfural. These mixtures present very complex composition which makes difficult the description of the liquid–liquid equilibrium involved. In
previous studies, the possibility to describe such equilibrium by using a reduced number of pseudo-components and the NRTL model has been
stated. In this work, a generalized model to describe the liquid–liquid equilibrium in the systems furfural + lubricating oils is presented. The
generalized model is based on a correlation of pseudo-component properties (specific gravity, density, refractive index and sulfur content) and
NRTL parameters with the oil average boiling temperature which has been developed from the model description for SPD and HND mixtures.
This method allows to set up a suitable thermodynamic model for any lubricant oil cut by using as experimental information only the average
boiling point and three physical properties (density, refractive index and sulfur content) of the distillate used as feed. The accuracy of the model
was checked by simulating single-stage extractions carried out with different lube oil cuts within the range SPD–HND in different experimental
conditions. Calculated and experimental yields, furfural content and physical properties of the raffinates and extracts were compared and a
good agreement was obtained.
䉷 2006 Elsevier Ltd. All rights reserved.

Keywords: Aromatic extraction; Furfural; Lubricating oils; Liquid–liquid equilibrium

1. Introduction The aromatic extraction process requires a good solvent

In the process of manufacturing lubricating oil base stocks,
removal of aromatic hydrocarbons from vacuum distillates is
required to improve several lubricating properties (Singh and
Kishore, 1978; Sequeira Jr., 1994). The residue from the crude
oil atmospheric distillation (long residue) is transferred to a
vacuum distillation column and separated into different lube oil
cuts, characterized by their boiling range and viscosity. Five raw
cuts are commonly obtained, called spindle distillate (SPD),
light neutral distillate (LND), medium neutral distillate (MND),
heavy neutral distillate (HND) and bright stock distillate (BSD).
∗ Corresponding author. Tel.: +34 91 4887089; fax: +34 91 4887068.
E-mail address: baudilio.coto@urjc.es (B. Coto).
1 Alfonso Cortina Technology Centre, REPSOL-YPF, 28933, Móstoles
(Madrid), Spain.
0009-2509/$ - see front matter 䉷 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2006.09.037
which maximize differences in properties between the liquid
phases. Furfural is one of the most widely used solvents. Its
selectivity towards aromatic compounds is high enough, it
decreases slowly when increasing temperature and it is accept-
able for both light and heavy vacuum distillates as reported
by different authors (Nelson, 1978; Mcketta, 1989; De Lucas
et al., 1993).
The design and simulation of solvent extraction processes
have been done mostly by fully empirical methods (Rahman
et al., 1984) that require wide experimental determination. The
application of such approach is limited because the experi-
mental information is not usually available. A more rigorous
modeling of the extraction process would be a suitable tool to
simulate changes in operating conditions or feed quality. To
properly describe the liquid–liquid equilibrium (LLE) estab-
lished in the extraction operation, a consistent thermodynamic
model should be based on the equilibrium data of the studied
 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039
systems. Solvent + lubricating oils are especially difficult sys-
tems because the high number of components in these mixtures,
and therefore, simplified methods of calculations are needed.
The pseudo-components approach based on distillation
curves, which is widely used to represent the complex compo-
sition of petroleum fractions in the vapor–liquid equilibrium
calculations, is not effective in LLE modeling because the
chemical structure of the compound has much bigger effect
than boiling temperatures. De Lucas et al. (1993) have suc-
cessfully applied the pseudo-component approach to predict
the LLE for the systems furfural+lubricating oils.
Recently, Briesen and Marquardt (2004) have reported an
adaptive multi-grid strategy based on a discretization to con-
struct correction terms using composition representation. The
obtained model, based on molecular information, can be used
to describe the LLE.
The composition of such complex mixtures is frequently
determined by standard test methods, such as ASTM D2007
(Annual Book of Standards, 2003). However, these methods
are time and money consuming and, therefore, composition
data are not usually available. For that reason, the composition
of petroleum mixtures is related with their physical properties,
few of them experimentally determined while the remaining are
estimated by known correlations. Different authors have pro-
posed accurate relations to calculate the composition of paraf-
fins, naphthenes and aromatics from the refractive index and
the viscosity of the mixture (Riazi and Daubert, 1986). How-
ever, the use of most of the developed correlations is limited
and, therefore, new methods are needed especially for heavy
petroleum fractions. Specific gravity (SG), density (D), refrac-
tive index (RI) and sulfur content (S%) were selected in this
work, since these properties can be successfully correlated with
the composition of the lubricant mixtures, obtaining good re-
sults for light and heavy lubricant oils (Van Grieken et al., 2005;
Coto et al., 2006).
In order to describe the LLE in the system solvent+lubricating
oils, different thermodynamic models can be used (Poling et al.,
2001; Prausnitz et al., 2000; Raal and Mühlbauer, 1998).
NRTL (Renon and Prausnitz, 1968) and UNIQUAC (Abrams
and Prausnitz, 1975) methods have been used to correlate LLE
data for the system furfural + hydrocarbon systems (Letcher
et al., 2003; Morawski et al., 2003).
Recently, a method based on three pseudo-components (sat-
urates, aromatics and polars) together with the NRTL model
has been reported to describe LLE in the systems furfural+light
(SPD) and furfural+heavy (HND) lubricating oils (Coto et al.,
2006; Van Grieken et al., 2005). The application of such model
to describe the extraction operation and to predict properties of
petroleum fractions (feed, raffinates and extracts) was also re-
ported. The results obtained in the prediction of several physical
properties showed similar or even higher accuracy than those
calculated by other methods reported in the literature. However,
the application of such models is limited to extraction experi-
ments carried out with light or heavy lubricating oils.
In this work, a generalized model was developed from the
studies reported for HND and SPD (Coto et al., 2006; Van
Grieken et al., 2005). Such generalization is based on the rela-
8029
tion between the pseudo-components properties and the NRTL
parameters with the average boiling point of the mixture used
as feed. The relations between pseudo-component properties
and average boiling point can be used to calculate the composi-
tion of the lubricant oil, if it is not experimentally available, by
means of an iterative method based on the pseudo-components
properties, and developed in this work. Thus, the LLE can be
described by means of a reduced number of parameters and
using as experimental information only the characterization of
the lube oil cut used as feed. To check the accuracy of the pro-
posed model, single-stage extraction experiments carried out
with lubricating oil cuts (obtained from light Arabia crude oil
and within the range SPD–HND) were simulated with Aspen
Plus, using the generalized model. Good results were obtained
for the prediction of yields and furfural content of the involved
mixtures. The predictions of physical properties (SG, D, RI and
S%) calculated by using the generalized model showed a good
agreement with the experimental values.
2. Experimental section
SPD, LND and MND were provided by REPSOL-YPF from
their refinery in Cartagena (Spain). Such mixtures were ob-
tained from light Arabia crude oil. Furfural was also supplied
by REPSOL-YPF and distilled before use to remove the oxi-
dation products formed due to air contact.
Experimental results for the single-stage extraction experi-
ments were obtained by using a 2L cylindrical stirred glass re-
actor following the procedure reported by Van Grieken et al.
(2005). Temperature was set by recirculating silicone oil from
a thermostatic bath, and controlled within ±0.1 ◦ C. A gentle
stream of nitrogen was passed to prevent furfural decomposi-
tion. Agitation at 430 rpm was maintained for 1 h, followed by
settling for another hour to achieve a good separation of the
two phases. Furfural was removed from extracts and raffinates
by vacuum distillation. Table 1 shows the temperature at which
50% of the lubricating oil is distilled (T50% ) following ASTM
D-1160. (SG) and D at 343 K (D343 ) (ASTM D1298), RI at
343 K (RI 343 ) (ASTM D1747) and S% (ASTM D4294) of feed,
raffinates and extracts were determined by means of the men-
tioned ASTM procedures (Annual Book of Standards, 2003).
Each mixture is denoted by a letter (R, raffinate; E, extract),
followed by the corresponding number of the experiment.
From the results shown in Table 1, for experiments at the
same temperature, lower yields of extract were obtained when
decreasing the furfural/feed ratio: species undergo less dissolu-
tion in furfural, and therefore, the efficiency of the extraction is
lower. Consequently, the yield of raffinate is higher. When ex-
periments with the same furfural/feed ratio and different tem-
perature are compared, higher temperatures increase not only
aromatic extraction, but also other compounds, reducing the
selectivity of the process.
3. Generalized model
A generalized model to describe the extraction operation was
developed from the models previously reported for SPD and
8030 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039

Table 1
Experimental results for the extractions carried out with SPD, LND and MND

Fraction Furfural/feed T Yield Furfural SG D343 RI 343 %S

(v/v) (K) (wt%) (wt%) (g/ml) (wt%)

SPD (T50% = 651 K)

Feed – – – – 0.9123 0.8745 1.4889 2.76
R-1 1 328 74.1 10.7 0.8829 0.8447 1.4694 1.75
E-1 1 328 25.9 82.5 1.0104 0.9738 1.5558 5.58
R-2 4 328 53.3 9.0 0.8599 0.8214 1.4550 0.88
E-2 4 328 46.7 91.7 0.9839 0.947 1.5356 4.98
R-3 7 328 43.8 7.4 0.8518 0.8132 1.4496 0.59
E-3 7 328 56.2 94.2 0.9676 0.9305 1.525 4.49
R-4 1 348 67.9 16.0 0.8834 0.8452 1.4693 1.75
E-4 1 348 32.1 78.8 0.9839 0.9470 1.5363 4.97
R-5 4 348 41.6 11.4 0.8572 0.8187 1.4531 0.76
E-5 4 348 58.4 89.8 0.9566 0.9195 1.5185 4.25
R-6 7 348 28.0 10.4 0.8482 0.8095 1.4481 0.56
E-6 7 348 72.0 92.7 0.9342 0.9020 1.5070 3.90
R-7 1 368 53.8 10.4 – – – –
E-7 1 368 46.2 92.7 0.9456 0.9082 1.5118 3.90
R-8 4 368 20.7 14.3 0.8545 0.8159 1.4521 0.76
E-8 4 368 79.3 86.6 0.9286 0.8910 1.5002 3.34
R-9 7 368 0.8 – – – 1.4509 –
E-9 7 368 99.2 90.1 0.9215 0.8747 1.4934 2.77

LND (T50% = 717 K)

Feed – – – 0.9236 0.8857 1.4960 2.48
R-1 1 333 80.0 10.4 0.9009 0.8694 1.4800 1.80
E-1 1 333 20.0 85.2 1.0304 0.9872 1.5710 5.31
R-2 4 333 60.3 7.9 0.8805 0.8498 1.4670 1.11
E-2 4 333 39.7 92.5 0.9991 0.9634 1.5460 4.63
R-3 7 333 51.5 7.1 0.8710 0.8414 1.4620 0.74
E-3 7 333 48.5 – – – – –
R-4 1 353 75.3 15.4 0.8991 0.8647 1.4800 1.71
E-4 1 353 24.7 81.8 1.0060 0.9677 1.5530 4.74
R-5 4 353 51.4 11.5 0.8772 0.8422 1.4650 0.98
E-5 4 353 48.6 91.0 0.9791 0.9394 1.5340 4.07
R-6 7 353 39.5 9.9 – 0.8277 1.4600 0.68
E-6 7 353 60.5 93.4 – 0.9277 1.5235 3.65
R-7 1 373 66.6 23.4 0.8994 0.8630 1.4800 1.71
E-7 1 373 33.4 75.7 0.9729 0.9444 1.5310 3.87
R-8 4 373 34.1 15.6 0.8738 0.8377 1.4630 0.85
E-8 4 373 65.9 88.1 0.9520 0.9149 1.5140 3.35
R-9 7 373 16.5 12.7 – 0.8243 1.4580 –
E-9 7 373 83.5 91.2 – 0.8982 1.5047 2.84

MND (T50% = 745K)

Feed – – – – 0.9341 0.8947 1.4992 2.46
R-1 1 333 83.2 10.3 0.9135 0.8738 1.4843 1.88
E-1 1 333 16.8 87.2 1.0494 1.0112 1.5800 5.31
R-2 4 333 66.3 8.6 0.8949 0.8550 1.4723 1.29
E-2 4 333 33.7 93.6 1.0235 0.9850 1.5601 4.77
R-3 7 333 57.8 8.0 0.8868 0.8468 1.4666 1.04
E-3 7 333 42.2 95.3 1.0092 0.9706 1.5488 4.48
R-4 1 353 78.7 14.4 0.9122 0.8725 1.4838 1.83
E-4 1 353 21.3 83.8 1.0247 0.9863 1.5631 4.75
R-5 4 353 57.0 10.9 0.8919 0.8520 1.4698 1.21
E-5 4 353 43.0 91.9 0.9987 0.9600 1.5432 4.18
R-6 7 353 45.8 10.3 0.9128 0.8419 1.4682 0.85
E-6 7 353 54.2 94.1 0.9956 0.9463 1.5369 3.74
R-7 1 373 72.1 21.6 – 0.8731 1.4842 1.85
E-7 1 373 27.9 78.7 – 0.9568 1.5410 4.10
R-8 4 373 42.4 15.7 – 0.8468 1.4722 1.04
E-8 4 373 57.6 89.4 – 0.9336 1.5287 3.50
R-9 7 373 28.3 12.9 – 0.8372 1.4663 0.74
E-9 7 373 71.7 92.3 – 0.9205 1.5192 3.10
 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039 8031

HND lubricating oil cuts (Coto et al., 2006; Van Grieken et al., 3.1. Pseudo-component properties
2005). Such models were based on the pseudo-component prop-
erties and the NRTL model. The pseudo-component properties The parameters of Eq. (1) for the different properties
for SPD and HND lubricating oil cuts previously reported are (SG, D343 , RI 343 and %S) of each pseudo-component are
shown in Table 2. NRTL parameters calculated for such lubri- shown in Table 4. According to such correlations, the pseudo-
cating oil cuts are presented in Table 3. components properties of any vacuum distillate lubricating oil
In this work, both pseudo-component and NRTL parameters with T50% within the range SPD–HND (640–801 K) can be
for SPD and HND were linearly correlated with T50% as fol- easily calculated.
lows: The following mixing rule was assumed to relate the physical
properties of any lubricating oil mixture to the properties and
P = A · T50% + B, (1) composition of pseudo-components:
 
XS XA XP
where P denotes a pseudo-component property or a NRTL bi- Pi,calc = P i,S + P i,A + P i,P , (2)
100 100 100
nary interaction parameter, and A and B are linearly adjustable
parameters.
Fig. 1 shows, as an example, the assumed linear dependence 2.0
between pseudo-component SG (for saturates, aromatics and
polars) and T50% of the lubricant oil. In such figure it is possible
to check how the T50% range between the previously studies
for HND and SPD is covered.
1.6

Table 2
SG

Pseudo-component properties previously reported for SPD (Coto et al., 2006)

and HND (Van Grieken et al., 2005)
1.2
Saturates Aromatics Polars

SPD
T50% (K) 640 640 640
SG 0.8050 1.0212 1.9300
D343 (g/ml) 1.4110 1.5860 1.6000 0.8
RI 343 0.7670 1.0000 1.3000
%S (wt%) 0 4.6 20.0

HND 640 680 720 760 800

T50% (K) 801 801 801 T50% / K
SG 0.8361 0.9764 1.9228
D343 (g/ml) 0.7930 0.9900 1.2232
Fig. 1. Relation between SG and T50% for the studied lubricant oils. Line: —–
RI 343 1.4311 1.5816 1.5830
Saturates, ..... Aromatics, - - - - Polars; Symbols:  SPD (Coto et al., 2006)
%S (wt%) 0 2.10 23.3
and HND (Van Grieken et al., 2005),  SPD,  LND, 䊉 MND, this work.

Table 3
NRTL binary interaction parameters previously reported for SPD (Coto et al., 2006) and HND (Van Grieken et al., 2005)

Component i Furfural Furfural Saturate Saturate Aromatic Polar

Component j Saturate Aromatic Aromatic Polar Polar Furfural

SPD
aij 0.70 −1.65 11.13 67.00 9.50 6.62
aj i −3.85 110.00 −6.39 9.47 −7.27 −17.00
bij 1310 755 −4472.87 −26000 −3110 1315
bj i 1421.75 −644 1900 −4000 3030 −2300
ij 0.23 0.10 0.30 0.13 0.30 0.20

HND
aij 0.77 0.17 10.49 −125.49 −33.84 −10.19
aj i −4.81 11.15 −7.7 −4.01 −0.1 −2.92
bij 1361.55 −592.67 −6070.78 46708.69 13210.45 2266.9
bj i 1707.14 4646.93 1937.2 −627.88 −3012.05 −1446.52
ij 0.29 0.30 0.20 0.20 0.20 0.20
8032 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039

Table 4 0.90
Correlation between pseudo-component properties and T50% a

Property (P ) Saturates Aromatics Polars

A × 10 4
B A × 10 4
B A × 104 B

Calculated D343 / gml-1

SG 1.9 0.6814 2.8 1.1991 −0.5 1.9586
D (g/ml) 1.6 0.6637 0.6 1.0398 −4.8 1.6052
RI 1.3 1.3311 −0.3 1.6035 −1.1 1.6676
%S (w%) – – −152 14.29 205 6.88 0.88
aP = A · T50% + B.

where Pi,calc is the calculated value of property i; Pi,S , Pi,A and

Pi,P the calculated values of the property i for saturate, aro-
matic and polar pseudo-component, respectively, and XS , XA
and XP the values of composition in saturates, aromatics and
polars. 0.86
In this work Eq. (2) was used straightforward to calcu- 0.86 0.88 0.90
late physical properties of any lubricating oil mixture from Experimental D343 / gml-1
calculated pseudo-component properties and composition. In
addition, such equation was used in a reversed way in order Fig. 2. Experimental and calculated D343 for the lubricant oils used in this
to calculate the composition values in saturates, aromatics and work. Symbols:  SPD,  LND, 䊉 MND.
polars for any lubricating oil cut used as feed in the extraction
operation.
Table 5
The calculation procedure to obtain the composition of the Correlation between ij and T50% a
feed involves only three properties, D343 , RI 343 and %S. For
a given unknown lubricating oil cut, the pseudo-component Interactionb A × 104 B × 102
properties are obtained from Eq. (1) and values for Xs , XA and f-s 4.3 −4.9
XP are those which minimize the following objective function: f-a 13 −70.6
 f-p −0.2 21.5
F= |Pi,exp − Pi,calc |, (3) s-a −6.2 69.7
i
s-p 4.6 −16.8
a-p −6.2 69.7
where Pi,exp is the experimental value for the considered prop- a = A · T
ij 50% + B.
erty, i, and Pi,calc is the corresponding one calculated from b f, Furfural; s, Saturate; a, Aromatic; p, Polar.
Eq. (2). The sum is extended over the three above-mentioned
properties (D343 , RI 343 and %S). Fig. 2 shows, as an example,
the comparison between the experimental and predicted values bij
ij = aij + , (6)
of D343 for the distillates considered in this work. As it can be Text
observed, no deviation was found, and therefore the calculated
values of composition can be successfully considered. where xi is the molar fraction of component i; ij is a model
parameter related with the tendency of species i and j to be ran-
domly distributed, and ij is the model parameter which must
3.2. NRTL model be considered for the interaction between pairs of molecules i
and j. ij is assumed to be temperature dependent (Text ) accord-
The NRTL equation is based on the local composition con- ing to Eq. (6), for which two binary parameters, aij and bij ,
cept and considers only binary interactions. The NRTL expres- are defined (Demirel and Gecegörmez, 1991). As ii = 0, aij =
sion for activity coefficient is given by Raal and Mühlbauer aj i , bij  = bj i and ij = j i , five binary parameters should be
(1998), Prausnitz et al. (2000), Poling et al. (2001): considered to characterize each binary interaction.
c NRTL parameters for SPD and HND shown in Table 3 were
j =1 xj j i Gj i
ln i = c used to obtain a T50% -dependent generalized model. Thus, ij
k=1 xk Gki parameters, previously calculated for SPD and HND were cor-
c   c  related with T50% by means of expression (1) and shown in
 xj Gij m=1 xm mj Gmj
+ c ij −  c , Table 5. According to expression (6), temperature-dependent
j =1 k=1 xk Gkj k=1 xk Gkj parameters ij and j i were estimated for SPD and HND at dif-
(4) ferent temperatures, in this work values of 323, 343 and 363 K
were selected for Text as usual temperatures for the extraction
Gij = exp(−ij ij ), (5) operation. Eq. (1) was used to correlate ij and j i at fixed Text
 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039 8033

Pseudo-components
properties

T50% of the lube oil

NRTL parameters Single-stage
used as feed
extractions

Composition of
the lubricating oil
used as feed

Simulation by
AspenPlus®

Fig. 3. Scheme for the application of the generalized model.

Table 6 Table 7
Correlation between ij and T50% a Pseudo-component properties for the selected lubricating oils

Interactionb T = 323 K T = 343 K T = 363 K Saturates Aromatics Polars

A × 10 3
B A × 10 3
B A × 10 3
B SPD
T50% (K) 651 651 651
f-s 1.5 3.8 1.4 3.6 1.3 3.5 SG 0.8071 1.0181 1.9295
f-a −15 10.0 −13 8.9 −12 7.9 D343 (g/ml) 0.7688 0.9993 1.2947
f-p 104 −90.6 103 −89.5 102 −88.6 RI 343 1.4124 1.5857 1.5989
s-a −35 19.0 −33 19.2 −31 18.8 S% (wt%) 0 4.39 20.23
s-p 202 −142.7 120 −85.9 48 −35.3
a-p 44 −28.6 26 −16.4 9.9 −5.4 LND
s-f −0.4 0.8 −0.8 0.8 −1.1 0.7 T50% (K) 717 717 717
a-p −512 435.9 −520 439.8 −520 443.2 SG 0.8199 0.9997 1.9265
p-f −86 65.8 −87 66.2 −88 66.6 D343 (g/ml) 0.7794 0.9952 1.2633
a-s −7.4 4.2 −7.5 3.9 −7.5 3.6 RI 343 1.4206 1.5839 1.5919
p-s −19 9.2 −23 12.3 −26 15.1 S % (wt%) 0 3.39 21.58
p-a −72 47.9 −65 43.1 −59 38.7
MND
a = A · T50% + B. T50% (K) 745 745 745
ij
b f, Furfural; s, Saturate; a, Aromatic; p, Polar. SG 0.8253 0.9920 1.9253
D343 (g/ml) 0.7840 0.9935 1.2499
RI 343 1.4291 1.5832 1.5890
S% (wt%) 0 3.0 22.1
values with T50% . Obtained values are shown in Table 6. From
the dependence of ij against 1/Text according to Eq. (6), values
for aij and bij can be obtained. Such procedure was also appli-
mental conditions given in Table 1 were simulated by means of
cable to j i . These relations allow calculate NRTL parameters
Aspen Plus䉸 using the generalized model and the results for all
for any lubricant oil using as experimental information only the
extraction experiments are shown in Table 9. The composition
T50% . The range of application is for lubricating oils with T50%
of each cut used as feed was estimated following the method
values within the range SPD–HND (640–801 K). Fig. 3 sum-
described above from the pseudo-components properties and
marizes the scheme of application of the generalized model.
physical properties of the different lubricating oils (SPD, LND
and MND).
4. Results and discussion In order to determine the accuracy of the different predic-
tions, absolute average deviation () was calculated as follows:
In order to check the quality of the generalized model de-  
i |Pexp − Pcalc |
scribed before, several experiments were carried out with the = = , (7)
lubricating oil cuts (within the range SPD–HND) presented in N N
Table 1. Pseudo-components properties and NRTL parameters where N represents the number of mixtures considered, Pexp
of such lubricating oil cuts were calculated from their T50% is the experimental value of a property and Pcalc the calculated
(Table 1) according to the coefficients listed in Tables 4, 5 and one. Table 10 lists the values of  for different properties related
6. Calculated properties of each pseudo-component for SPD, with the single-stage extractions for SPD, LND and MND.
LND and MND lubricating oil are shown in Table 7. Calculated Experimental and predicted yields for all lubricating oil cuts
NRTL parameters for such cuts are shown in Table 8. Experi- are compared in Fig. 4. As it can be observed, no systematic
8034 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039

Table 8
NRTL parameters calculated for the selected lubricating oils

Component i Furfural Furfural Saturate Saturate Aromatic Polar

Component j Saturate Aromatic Aromatic Polar Polar Furfural

SPD
aij 0.70 −1.52 11.10 53.84 6.54 5.47
aj i −3.92 103.24 −6.48 8.55 −6.77 −16.05
bij 1313.60 661.84 −4584.83 −21029.00 −1995.90 1380.80
bj i 1441.20 −282.96 1902.50 −3769.50 2615.80 −2237.90
ij 0.23 0.10 0.30 0.10 0.30 0.20

LND
aij 0.74 −0.78 10.83 −25.07 −11.22 −1.42
aj i −4.31 62.72 −7.01 3.02 −3.84 −10.28
bij 1334.70 109.21 −5239.88 8776.9 4694.40 1771.0
bj i 1558.20 1885.70 1917.8 −2387.2 138.87 −1887.40
ij 0.26 0.20 0.25 0.16 0.25 0.20

MND
aij 0.75 −0.46 10.72 −58.54 −18.76 −4.35
aj i −4.47 45.50 −7.24 0.68 −2.59 −7.83
bij 1343.70 −125.20 −5517.78 21422 7532.8 1936.50
bj i 1607.80 2805.70 1924.30 −1800.8 −911.95 −1738.70
ij 0.27 0.23 0.23 0.17 0.23 0.20

deviations were found in any case, obtaining the expected dis-
tribution of phases. Thus, at the lowest temperature and fur-
fural/feed ratio 79.5%, 84.5% and 85.7% in raffinate was pre-
dicted for SPD, LND and MND, respectively, which is in good
concordance with 74.1%, 80.0% and 83.2% experimentally de-
termined for such cuts. When increasing the temperature or the
solvent/feed ratio, the yield becomes lower as stated before.
Thus, at the same temperature and highest solvent/feed ratio
a raffinate yield of 44.0%, 54.1% and 58.1% was calculated
for SPD, LND and MND, respectively, showing again a good
agreement with the 43.8%, 51.5% and 57.8% experimentally
obtained. Absolute average deviations of 2.6%, 3.0% and 1.2%,
were obtained for the prediction of extraction yields for SPD,
LND and MND, respectively.
The comparison between experimental and predicted con-
tent of furfural for raffinates and extracts is shown in Fig. 5.
The model exhibits good accuracy in all experimental condi-
tions, even at the higher values of temperature or furfural/feed
ratio. For instance, the predicted values for furfural content
in raffinate for MND was 9.1% at the lowest temperature
and furfural/feed = 7 and 19.2% at higher temperature and
furfural/feed ratio = 1, which fit well with the experimental
ones, 8.0% and 21.6%, respectively. Absolute average devia-
tions of 0.6%, 1.5% and 1.1% were obtained for the prediction
of furfural content in raffinate for SPD, LND and MND,
respectively.
Physical properties (SG, D, RI and S%) of each mixture (feed,
raffinates and extracts) for SPD, LND and MND were calcu-
lated according to expression (2), using the pseudo-components
properties shown in Table 7 and the compositions in saturates,
aromatics and polars (XS , XA and XP ) obtained by simulation
presented in Table 9. As an example, experimental and calcu-
lated values for SPD experiments are summarized in Table 11.
The comparison between experimental and calculated values of
such properties is shown in Figs. 6–8 for SPD, LND and MND
extraction experiments, respectively. As it can be seen, a good
agreement between predicted and experimental values was ob-
tained in all cases. In all cases absolute average deviations were
< 0.015 for SG, < 0.007 for D343 , < 0.005 for RI 343 and < 0.6
for %S.
As it can be observed in Figs. 6–8 the prediction of the S%
shows systematic deviations. This can be due to a limitation
of the model since it is assumed that the S% is zero for satu-
rates, which is true if only tiophenic sulfur (present in aromatic
and polar species) is present. It works well for the lightest lube
oil cut (SPD). However, for heavier lube oil cuts, other sulfur
species present in the saturate compounds should be consid-
ered. Although the proportion of this kind of sulfur is not very
high, it could affect the predictions obtained by using the pro-
posed model and a new distribution of the S% for the pseudo-
components would be needed.
In spite of these limitations, the obtained results reported by
Coto et al. (2006) using the simplified distribution of the S%
present similar accuracy than those obtained using a method
proposed by Riazi et al. (1999).
The experiments carried out at highest temperatures and
furfural/feed ratio could not be simulated because the system
is near the miscibility region. This is a limitation in the appli-
cation of the model previously indicated (Van Grieken et al.,
2005) for the system furfural + HND lubricating oils and also
found for the system furfural + SPD lubricating oil (Coto
et al., 2006). No accurate predictions are obtained in these con-
ditions; however, this limitation is not a big model lack because
in the practice these severe conditions are not very usual.
 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039 8035

Table 9
Simulation results for the single-stage extraction experiments

Fraction Furfural/feed T Yield Furfural XS XA XP

(w/w) (K) (wt%) (wt%) (wt%) (wt%) (wt%)

SPD
Feed – – – – 50.4 45.8 3.8
R-1 1 328 79.5 11.0 60.0 37.6 2.4
E-1 1 328 20.5 85.3 13.2 77.3 9.5
R-2 4 328 57.1 9.0 71.4 27.2 1.4
E-2 4 328 42.9 92.2 22.5 70.4 7.1
R-3 7 328 44.0 8.3 76.3 22.6 1.1
E-3 7 328 56.0 94.1 30.1 63.9 6.0
R-4 1 348 70.8 15.4 62.5 35.2 2.3
E-4 1 348 29.2 79.9 21.1 71.4 7.5
R-5 4 348 42.3 11.8 75.4 23.3 1.3
E-5 4 348 57.7 89.8 32.1 62.2 5.7
R-7 1 368 24.4 10.8 80.2 18.9 0.9
E-7 1 368 75.6 92.2 40.8 54.5 4.7
R-8 4 368 22.7 16.0 77.5 21.4 1.1
E-8 4 368 77.3 86.9 42.5 52.9 4.6

LND
Feed – – – – 47.6 48.5 3.9
R-1 1 333 84.5 10.7 54.8 41.9 3.3
E-1 1 333 15.5 88.1 8.4 84.4 7.2
R-2 4 333 65.6 9.7 63.9 33.4 2.7
E-2 4 333 34.4 93.5 16.7 77.2 6.1
R-3 7 333 54.1 9.4 68.0 29.5 2.5
E-3 7 333 45.9 95.0 23.7 70.8 5.5
R-4 1 353 77.7 14.6 57.1 39.5 3.4
E-4 1 353 22.3 83.4 14.6 79.7 5.7
R-5 4 353 52.5 12.7 67.7 29.7 2.6
E-5 4 353 47.5 91.2 25.5 69.3 5.2
R-7 1 373 70.3 20.5 58.0 38.5 3.5
E-7 1 373 29.7 78.1 23.1 72.1 4.8
R-8 4 373 35.7 16.7 69.9 27.5 2.6
E-8 4 373 64.3 88.4 35.2 60.2 4.6

MND
Feed – – – – 45.5 50.1 4.4
R-1 1 333 85.7 9.9 54.8 41.9 3.3
E-1 1 333 14.3 89 8.4 84.4 7.2
R-2 4 333 68.2 9.3 63.9 33.4 2.7
E-2 4 333 31.8 93.9 16.7 77.2 6.1
R-3 7 333 58.1 9.1 68.0 29.5 2.5
E-3 7 333 41.9 95.4 23.7 70.8 5.5
R-4 1 353 79.6 13.8 57.1 39.5 3.4
E-4 1 353 20.4 84.6 14.6 79.7 5.7
R-5 4 353 56.4 12.4 67.7 29.7 2.6
E-5 4 353 43.6 91.9 25.5 69.3 5.2
R-7 1 373 73.0 19.2 58.0 38.5 3.5
E-7 1 373 27.0 79.9 23.3 72.0 4.7

Table 10
Absolute average deviations for predictions of single-stage extractions

Yield Furf raff Furf ext SG × 102 D343 × 102 RI 343 × 102 S%

SPD 2.6 0.6 0.8 0.8 0.6 0.5 0.5

LND 3.0 1.5 1.4 0.5 0.5 0.3 0.5
MND 1.2 1.1 0.7 1.4 0.7 0.3 0.6
8036 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039

100 100

Calculated yield / w %
Calculated yield / w %
80 80

60 60

40 40

20 20

0 0
0 20 40 60 80 100 0 20 40 60 80 100
(a) Experimental yield / w % (b) Experimental yield / w %

Calculated yield / w % 100

80

60

40

20

0
0 20 40 60 80 100
(c) Experimental yield / w %

Fig. 4. Comparison between experimental and predicted values for extraction yields for the single-stage extraction experiments: (a) SPD; (b) LND; (c) MND.
Symbols:  Raffinates;  Extracts.

100 100
Calculated furfural % / w %

Calculated furfural % / w %

80 80

60 60

40 40

20 20

0 0
0 20 40 60 80 100 0 20 40 60 80 100
(a) Experimental furfural % / w % (b) Experimental furfural % / w %

100
Calculated furfural % / w %

80

60

40

20

0
0 20 40 60 80 100
(c) Experimental furfural % / w %

Fig. 5. Comparison between experimental and predicted values for the furfural content for the single-stage extraction experiments: (a) SPD; (b) LND;
(c) MND. Symbols:  Raffinates; 䊉 Extracts.

According to all the obtained results, it is possible to erties of the involved mixtures. Therefore, the generalized
correlate the pseudo-components properties and NRTL pa- model can be used to describe the LLE in the systems
rameters with the T50% of the lubricating oil cut to pre- furfural + lubricating oil cuts with T50% within the range
dict extraction yields, furfural content and physical prop- SPD–HND.
 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039 8037

Table 11
Experimental and calculated values of the studied properties for SPD extraction experiments

Fraction Furfural/feed T SG D343 RI 343 %S

(w/w) (K) (g/ml) (wt%)

Experimental Calculation Experimental Calculation Experimental Calculation Experimental Calculation

Feed – 0.9123 0.9144 0.8745 0.8752 1.4889 1.4871 2.76 2.78

R-1 1 0.8829 0.8888 0.8447 0.8512 1.4694 1.4712 1.75 2.13
E-1 1 1.0104 1.0277 0.9738 0.9814 1.5558 1.5576 5.58 5.27
R-2 4 0.8599 0.8629 0.8214 0.8255 1.4550 1.4532 0.88 1.48
E-2 4 0.9839 0.9922 0.9470 0.9505 1.5356 1.5384 4.98 4.49
R-3 7 0.8518 0.8526 0.8132 0.8151 1.4496 1.4458 0.59 1.21
E-3 7 0.9676 0.9688 0.9305 0.9282 1.5250 1.5236 4.49 3.99
R-4 1 0.8834 0.8836 0.8452 0.8456 1.4693 1.4671 1.75 2.01
E-4 1 0.9839 0.9972 0.947 0.9548 1.5363 1.5411 4.97 4.61
R-5 4 0.8572 0.8547 0.8187 0.8170 1.4531 1.4471 0.76 1.27
E-5 4 0.9566 0.9629 0.9195 0.9226 1.5185 1.5197 4.25 3.87
R-7 1 – 0.8447 – 0.8070 – 1.4400 – 1.01
E-7 1 0.9456 0.9390 0.9082 0.8993 1.5118 1.5038 3.90 3.34
R-8 4 0.8545 0.8505 0.8159 0.8128 1.4521 1.4440 0.76 1.17
E-8 4 0.9286 0.9345 0.891 0.8949 1.5002 1.5007 3.34 3.24

1.04 1.00
Calculated D343 / g ml-1

1.00 0.96
Calculated SG

0.96 0.92

0.92 0.88

0.88 0.84

0.84 0.80
0.84 0.88 0.92 0.96 1.00 1.04 0.80 0.84 0.88 0.92 0.96 1.00
Experimental SG Experimental D343 / g ml-1

1.58 6
1.56
Calculated S % / w %

5
Calculated RI343

1.54
4
1.52
3
1.50
2
1.48
1.46 1

1.44 0
1.44 1.46 1.48 1.50 1.52 1.54 1.56 1.58 0 1 2 3 4 5 6
Experimental RI343 Experimental S % / w %

Fig. 6. Comparison between experimental and predicted properties for SPD mixtures (feed, raffinates and extracts).

5. Conclusions the activity coefficients. In order to reach a predictive model,

A thermodynamic model, previously developed for the de-
scription of the LLE established in the extraction operation
of lubricating oils with furfural, has been extended to other
feeds within the range SPD–HND yielding a generalized model
version.
Only three pseudo-components (saturates, aromatics and po-
lars) are considered and the NRTL model is used to determine
a procedure has been developed to determine both the pseudo-
component properties and the NRTL parameters. The T50%
value was chosen as representative of the lubricating oil mix-
ture and a linear dependence was established between pseudo-
components properties and NRTL parameters and T50% .
The feed composition was estimated from their physical
properties and that estimated for the pseudo-components. In-
volved properties were D343 , RI 343 and %S.
8038 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039

1.04 1.00

Calculated D343 / g ml-1

1.00 0.96

Calculated SG
0.96 0.92

0.92 0.88

0.88 0.84

0.84 0.80
0.84 0.88 0.92 0.96 1.00 1.04 0.80 0.84 0.88 0.92 0.96 1.00

Experimental SG Experimental D343 / g ml-1

1.58 6
1.56

Calculated S % / w %
5
Calculated RI343

1.54
4
1.52
3
1.50
2
1.48
1.46 1

1.44 0
1.44 1.46 1.48 1.50 1.52 1.54 1.56 1.58 0 1 2 3 4 5 6
Experimental RI343 Experimental S % / w %

Fig. 7. Comparison between experimental and predicted properties for LND mixtures (feed, raffinates and extracts).

1.04 1.04
Calculated D343 / gml-1

1.00 1.00
Calculated SG

0.96 0.96

0.92 0.92

0.88 0.88

0.84 0.84
0.84 0.88 0.92 0.96 1.00 1.04 0.84 0.88 0.92 0.96 1.00 1.04

Experimental SG Experimental D343 / g ml-1

1.60 6
1.58 5
Calculated S % / w %
Calculated RI343

1.56
4
1.54
3
1.52
2
1.50
1.48 1

1.46 0
1.46 1.48 1.50 1.52 1.54 1.56 1.58 1.60 0 1 2 3 4 5 6
Experimental RI343 Experimental S % / w %

Fig. 8. Comparison between experimental and predicted properties for MND mixtures (feed, raffinates and extracts).

The generalized model was applied to a single-stage extrac- extraction experiments as well as for the physical properties
tion experiments carried out with vacuum distillates SPD, MND (SG, D343 , RI 343 and %S) of the involved mixtures (raffinates
and LND, all of them within the range SPD–HND. Accurate and extracts). Absolute average deviations were always lower
predictions were obtained for yields and furfural content in the than 3.0% for the yields and lower than 1.5% for the furfural
 B. Coto et al. / Chemical Engineering Science 61 (2006) 8028 – 8039 8039

content in raffinates and extracts. Likewise, absolute average through the project “Desarrollo de modelos de simulación de
deviations for the studied properties were lower than 0.014 for procesos de refino”.
SG, around 0.006 for D343 , lower than 0.005 for RI 343 and
around 0.5 for the %S. References
S% predictions show systematic deviations revealing that it
Abrams, D., Prausnitz, J.M., 1975. Statistical thermodynamics of liquid
is necessary to obtain a new distribution of the S% for the
mixtures: a new expression for the excess Gibbs energy of partly or
pseudo-components. No systematic deviations were obtained completely miscible systems. A.I.Ch.E. Journal 21, 116–128.
for the predictions of the other properties in the application ASTM Annual Book of Standards, 2003. American Society for Testing and
of the model to the different feeds, thus concluding that the Materials, vol. 5. West Conshohocken, Pennsylvania.
differences between these feeds can be well established through Briesen, H., Marquardt, W., 2004. New approach to refinery process simulation
with adaptive composition representation. A.I.Ch.E. Journal 50, 633–645.
the different physical properties involved and the T50% value.
Coto, B., Van Grieken, R., Peña, J.L., Espada, J.J., 2006. A model to
predict physical properties for light lubricating oils and its application
Notation to the extraction process by furfural. Chemical Engineering Science 61,
4381–4392.
De Lucas, A., Rodríguez, L., Sánchez, P., Carnicer, A., 1993. Extraction of
a NRTL interaction parameter aromatic compounds from heavy neutral distillate lubricating oils by using
A parameter of linear correlation furfural. Separation Science and Technology 28, 2465–2477.
b NRTL interaction parameter Demirel, Y., Gecegörmez, H., 1991. Simultaneous representation of excess
B parameter of linear correlation enthalpy and vapor–liquid equilibrium data by the NRTL and UNIQUAC
models. Fluid Phase Equilibria 65, 111–133.
Comp. compound
Letcher, T.M., Zondi, S., Naicker, K., 2003. Liquid–liquid equilibria for
D343 liquid density at 343 K, g/ml mixtures of (Furfural + an aromatic hydrocarbon + an alkane) at T =
E extract 298.15 K. Journal of Chemical and Engineering Data 48, 23–28.
f furfural Mcketta, J.J., 1989. Encyclopedia of Chemical Processing and Design. Marcel
F objective function Dekker, New York.
Morawski, P., Letcher, T.M., Naicker, P.K., Domanska, U., 2003. Liquid–liquid
HND heavy neutral distillate
equilibria for mixtures of (furfural+a chlorinated aromatic compound +an
LND light neutral distillate alkane) at T = 298.15 K. Journal of Chemical and Engineering Data 48,
p polar 822–826.
P studied property such as D, SG, RI or S% Nelson, W.L., 1978. Petroleum Refining Engineering. fourth ed. McGraw-
P average property of pseudo-component Hill, New York.
Poling, B.E., Prausnitz, J.M., O‘Connell, J.P., 2001. The Properties of Gases
R raffinate and Liquids. McGraw-Hill, New York.
RI 343 refractive index at 343 K Prausnitz, J.M., Lichtenthaler, R.N., De Azevedo, E.G., 2000. Molecular
s saturate Thermodynamics of Fluid-Phase Equilibria. Prentice-Hall, New Jersey.
SG specific gravity Raal, J.D., Mühlbauer, A.L., 1998. Phase Equilibria. Measurement and
SPD spindle distillate Computation. Taylor and Francis, Washington.
Rahman, M., Mikitenko, P., Asselineau, L., 1984. Solvent extraction of
S% sulfur content, wt% aromatics from middle distillates. Chemical Engineering Science 39,
T temperature of extraction, K 1543–1558.
T50% average temperature of the feed lubricating oil, K Renon, H., Prausnitz, J.M., 1968. Local compositions in thermodynamic
v/v volume/volume excess functions for liquid mixtures. A.I.Ch.E. Journal 14, 135–144.
x mole fraction Riazi, M.R., Daubert, T.E., 1986. Prediction of molecular-type analysis of
petroleum fractions and coal liquids. Industrial and Engineering Chemistry
X composition of each mixture, wt% Process Design and Development 25, 1009–1015.
Greek letters Riazi, M.R., Nasimi, N., Roomi, Y.A., 1999. Estimation of sulfur content of
petroleum Products and crude oils. Industrial and Engineering Chemistry
 NRTL interaction parameter Research 38, 4507–4512.
 NRTL interaction parameter given by Eq. (4) Sequeira Jr, A., 1994. Lubricant Base Oil and Wax Processing. Marcel Dekker,
New York. pp. 81–82.
Subscripts Singh, H., Kishore, K., 1978. Solvent refining of medium viscosity distillate
and changes in group chemical composition. Journal of Applied of
A aromatics Chemistry and Biotechnology 28, 617–625.
i, j component i, j Van Grieken, R., Coto, B., Romero, E., Espada, J.J., 2005. Liquid–liquid
P polars equilibrium in the system furfural + heavy neutral distillate. Industrial and
Engineering Chemistry Research 44 (21), 8106–8112.
S saturates

Acknowledgments

The authors thank REPSOL-YPF for providing the lubricant

oil samples used in this work and for the financial support