Application of the Bergman-Sutton

Method for Determining Blend Viscosity

Robert P. Sutton, Marathon Oil Company, and David F. Bergman, BP America

Summary where µ2 is the viscosity of the solvent hydrocarbon (i.e., the

As the petroleum industry accesses more low-gravity-oil resources, component with the lower viscosity). This response is typical of
modification of viscosity by blending lighter hydrocarbons has mixtures with a similar chemical nature (Watson K factor). As the
become a necessity in order to attain bulk properties that will flow chemical nature of the components begins to differ, the complex-
though a pipeline. In the more conventional oil reservoirs, the ity of the interaction grows and the resulting trend of normalized
need to estimate the viscosity of oil blends occurs when reservoir viscosity with blend fraction can result in a near-linear or con-
fluids are contaminated with oil-based muds or when production cave-downward trend (Curves 6 and 7). This behavior though is
streams from different reservoirs or fields are commingled in a less common, and the concave-upward behavior is prevalent for
single pipeline. Several methods have appeared in the literature for most blends. Viscosity ratio also has a greater impact on the final
estimating blend viscosity. All of these methods require a measured relationship than Watson K factor, as evidenced by the behavior of
viscosity for each component of the blend. The number of viscosity Curve 8 in Fig. 1. This curve shares the same difference in Watson
measurements is compounded when the viscosity of the blended K factor, as Curves 6 and 7, but the higher viscosity ratio results in
mixture is required at several temperatures. Of the viscosity cor- the typical concave-upward behavior exhibited by most blends.
relations published, the Bergman and Sutton method has the widest In 2009, Bergman and Sutton presented a method (hereafter
range of temperature and oil API gravity and has been consistently referred to as the Bergman and Sutton method) to predict dead-oil
demonstrated to provide accurate results over these conditions. viscosity accurately. Using a large data set, the method yielded
This method requires the component specific gravity, the Watson statistical accuracies that are summarized in Table 1.
characterization factor (Watson K factor), and temperature to The authors believe that this method should also be applicable
estimate viscosity. By using the proper mixing rules, an estimate to determine the viscosity of hydrocarbon blends. Therefore, a
of blend viscosity can be made with comparable or improved database of blend-viscosity data was created and evaluated with
accuracy over the “best” published methods without the need for several published empirical blend-viscosity correlations to estab-
individual-component viscosity measurements. A database of 2,059 lish an expected level of calculation accuracy. Calculations using
blend-viscosity measurements from more than 800 mixtures was the Bergman and Sutton method were then compared to these
created to compare the accuracy of the various methods. Viscosity results. Considering that measured viscosity data can be in error
measurements of the individual components in the blends studied by 15%, these results are considered excellent. Furthermore, the
exceeded 7,600 data points. A diverse group of mixtures, ranging database used to evaluate blend viscosity in this work is different
from light alkane or aromatic pure components to bitumen, diesel, from that used in the development of the Bergman and Sutton dead
biodiesel, condensate, crude, and assay fractions, was included in oil viscosity method (Bergman and Sutton 2009). The comparable
this database. Blends comprised the typical binary mixtures but results (Table 1) obtained here provide further confirmation of the
ranged up to a maximum of eight components in the mixture. robustness of this method and the extension of the Bergman and
Sutton method for use in determining blend viscosity.
Introduction
Database
The prediction of the blend viscosity resulting from the mixture
of two or more oils has historically been considered a rather com- A database of viscosity data from a total of 2,059 mixtures con-
plex problem. As a result, numerous empirical methods have been sisting of 7,646 components was created to ascertain the accuracy
published, which are used to predict the viscosity of hydrocarbon of the blend-viscosity methods. The data were derived primarily
mixtures. The viscosity of a blend typically does not vary linearly from literature sources and some in-house studies representing data
as a function of component concentration (Roelands 1966). This collected on 827 different oils or pure components (Al-Besharah
point is illustrated in Fig. 1, which depicts the behavior of the et al. 1987, 1989; Al-Besharah 1989; Anhorn and Badakhshan
normalized change in viscosity with the volume fraction of the 1994; Barrufet and Setiadarma 2003; Barrufet et al. 1999; Chevron
lighter component in the blend. Curve 1 illustrates the behavior 2005; Díaz et al. 1986; Drayer 1985; Estrada-Baltazar et al. 1998a,
of an ideal mixture of the components where the mixture viscos- 1998b; Exxon 2005; Ha and Koppel 2008; Hernández-Galván
ity varies linearly with the light-component fraction of the blend. et al. 2007; Iglesias-Silva et al. 1999; Irving 1977b; Joshi and Pegg
However, this is not the case in real life, so Curve 1 essentially 2007; Orbey and Sandler 1993; Ramakrishnan et al. 2003; Reid
provides only a point of reference. The correct relationship results et al. 1987; Shigemoto et al. 2006; Shu 1984; Tat and Van Gerpen
in either a concave-up or a concave-down response with the light- 1999, 2000, 2002; Wu et al. 1998, 1999; Yuan et al. 2005). The
component fraction. The more common trend for oil mixtures pure-component data included the alkane compounds, n-pentane
is concave up, so the blend viscosity is impacted more by the through n-hexadecane, and ringed compounds such as benzene,
less-viscous component. As the viscosity ratio defined by Eq. 1 toluene, and ethyl benzene. Oils included lube oils, diesel, biodie-
increases, the relationship exhibits more curvature, as indicated sel, condensate, naphtha, dead crude oil, and bitumen. Oil based
by Curves 2 through 5 in Fig. 1. mud components representing the ester, olefin, and mineral oil
families were included, and crude oil assay fractions were also
1 included. A summary of the data is provided in Table 2.
rat = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
2 Before adding any data to the database, they were checked to
ensure consistency in the reported measured viscosity. The Bergman
plot (Fig. 2) was used in this process. Data displayed on this plot
Copyright © 2012 Society of Petroleum Engineers should exhibit a linear trend that is roughly parallel to pure-component
This paper (SPE 117711) was accepted for presentation at the SPE Eastern Regional/ data. Deviation from these established trends is an indication of incon-
AAPG Eastern Section Joint Meeting, Pittsburgh, Pennsylvania, USA, 11–15 October 2008, sistent measurements or non-Newtonian behavior. An example of this
and revised for publication. Original manuscript received for review 5 August 2008. Revised
manuscript received for review 31 December 2008. Paper peer approved 23 January
type of data problem was presented by Degiorgis et al. (2001). Data
2009. exhibiting these types of trends were excluded from the database.

106 February 2012 SPE Production & Operations

 Mixture Viscosity

1– µrat=1
2– µrat=1.5
3– µrat=5
6
4– µrat=10
5– µrat=100
6–Dif Char ( µrat=1.5)
7
7–Dif Char ( µrat=2)
8–Dif Char ( µrat=5)
Normalized Viscosity

8
1
3 2
4

0.0 1.0
Concentration

Fig. 1—Change in blend viscosity with concentration.

TABLE 1—ACCURACY OF BERGMAN AND SUTTON METHODS

Method # Pts % AE Std Dev % AAE Std Dev

Bergman and Sutton (2009) 9024 –5.1 21.4 16.6 14.5

Current work 2059 1.5 15.6 11.0 11.2

The relationship between API gravity, Watson K factor, and 20°API because the calculated viscosity is very sensitive to this
dead-oil viscosity as established by Bergman and Sutton is parameter.
depicted graphically in Fig. 3. At a specified temperature, oil Bergman and Sutton updated the relationship (Eq. B-4 in Appen-
viscosity increases with decreasing API gravity and increasing dix B) between specific gravity, molecular weight, and the Watson
Watson K factor. Oils must be characterized using both the API or characterization factor with application for low-to-high API gravity
specific gravity and Watson K factor in order to predict viscosity oils less than 60°API. For lighter oils, the relationship proposed by
accurately. The Watson K factor may be estimated from specific Whitson (Whitson 1983; Riazi 2005) (Eq. B-3 in Appendix B) is
gravity and molecular weight. Care must be exercised when esti- used. Eqs. B-3 and B-4 are summarized in Appendix B, which pro-
mating the Watson K factor for heavy oils with gravities less than vides details for the Bergman and Sutton dead oil viscosity method.
Fig. 4 illustrates the relationship between molecular weight and
TABLE 2—DESCRIPTION OF BLEND VISCOSITY
API gravity where the Watson K factor is assumed to vary based
DATABASE on average trends with API gravity as presented by Bergman and
Sutton (2009). The deviation at lower API gravity is the result of
Property Minimum Maximum the improved accuracy of the Bergman and Sutton method, while
the deviation at higher API gravity is the result of the more-accurate
Component oil gravity (°API) 0.45 149.5
results from Whitson’s equation (Whitson 1983; Riazi 2005).
Blend oil gravity (°API) 8.38 89.2 The molecular weight is readily defined for pure components
3
Blend density (g/cm ) 0.58 1.00 but was unavailable for 45 of the oils included in the database.
Blend molecular weight 76.6 677 For these oils, only the API gravity was specified, so the oil was
incompletely characterized. In this instance, the Watson K factor
Component Watson K factor 9.7 14.6
was inferred to match the viscosity profile of the component oil,
Blend Watson K factor 9.8 13.3 and then molecular weight was calculated from the Watson K factor
Blend dead oil viscosity (cp) 0.157 464,000 and the specific gravity, resulting in a fully characterized oil.
14
Component dead oil viscosity (cp) 0.0596 5.28 10
Empirical Methods
Temperature (°F) –4 346.9
All of the empirical methods require the viscosity of each com-
Components in blend 2 8 ponent to be blended as well as other physical properties such as

February 2012 SPE Production & Operations 107

 Viscosity of Pure Hydrocarbons by Family
3
n-Paraffins
Aromatics
Cyclohexanes
16.9 API
10,000 cp Naphthalenes
Olefins
2 1000 cp 8.6 API
Wax Crude Oils
SPE 69420
100 cp 25.9 API
32.4 API
Ln Ln (Viscosity,cp + 1)

35.0 API
1 Inconsistent
10 cp Measurement 16.1 API
39.9 API
3 cp
0
1 cp

0.5 cp
–1
0.3 cp

–2
0.1 cp

0°F 50°F 100°F 150°F 200°F 250°F 300°F 400°F

–3
5.6 5.8 6 6.2 6.4 6.6
Ln (Temperature, °F + 310)

Fig. 2—Bergman plot to test validity of data.

volume, weight, or mole fraction. The general form of equation It is readily apparent that these simple techniques for deter-
used by these methods is mining the mixture viscosity are insufficient and very inaccurate;
however, a thorough and more-detailed review of these results is
n
m = ∑ X i f ( i ), . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
warranted. Fig. 5 provides this detail for these simple techniques
i =1
as a function of viscosity ratio. This was found to be a very useful
parameter for benchmarking the performance and accuracy of the
n
 m = ∑ X i f ( i ), . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3)
blend equations.
i =1
Examination of Fig. 5 reveals that an accurate determination of
blend viscosity is obtained for viscosity ratios less than 3; however,
where Xi is the volume, weight, or mole fraction of each component the errors become quite large as the viscosity ratio increases. This
and f(µi) or f(i) is a function of absolute (dynamic) viscosity or is shown further in Fig. 6, which clearly illustrates the increasing
kinematic viscosity, respectively. The function could be as simple errors with viscosity ratio for simple volumetric averaging of com-
as just the viscosity, although this is normally not the case because ponent viscosity. Only volumetric averaging is presented, but the
the resulting errors in predicted blend viscosity can be quite large. same conclusions also apply to simple weight and mole-fraction
Computer models used to solve multiphase-flow pressure-drop averaging methods. Fig. 6 displays error over a limited range of
problems often use a volume fraction average to mix oil and water viscosity ratio and shows how the accuracy declines as this quantity
to develop an equivalent liquid viscosity. A test of this philosophy increases. In contrast, the Bergman and Sutton method shown in
applied to the entire database of blended oils is summarized in Fig. 7 provides consistent results over this range of viscosity ratio
Table 3. and beyond. An increased understanding of this behavior lies in

Fig. 3—Change in viscosity with API gravity and Watson K factor.

108 February 2012 SPE Production & Operations

 Correlation Comparison Irving (1977a) provides a comprehensive review of blend-
1000 viscosity methods published before the early 1970s. A review of
MW Whitson more-recent literature identifies methods overlooked by Irving
900 MW BS
and newly developed methods. A collection of these methods is
800 summarized in Appendix A (see Table A-1). Only methods that
are predictive using measured component viscosity are included.
Molecular Weight

700
Methods that require the fitting of a constant to data or those
600 that require viscosity measurements at multiple temperatures are
500 excluded from the table.
A review of the published blend-viscosity methods shows equa-
400 tions that require the volume, weight, or mole fraction of each com-
300 ponent in the mixture. The literature data typically provide blends
by volume or weight fraction, with the specific gravity of the each
200
component provided. Molecular weight is defined for mixtures of
100 pure components and is sometimes reported for the oils. Occasion-
0
ally, the Watson K factors are also reported for the oils.
0 20 40 60 80 100
Mixing Rules
API Gravity
In order to evaluate the accuracy of the various literature blend-
Fig. 4—Comparison of methods to calculate molecular weight. viscosity methods and test the applicability of the Bergman and
Sutton method to blends, various mixing rules are required. If
only the volume fraction of a blend was reported, it is necessary
Fig. 1, which simply depicts the behavior of blend viscosity with to determine the equivalent weight or mole fraction.
concentration of the lighter component. For an ideal case of mix- Weight fraction for each component in an “n” component mix-
tures with similar viscosity, specific gravity, and character, the blend ture can determined from volume fraction and the specific gravity
viscosity tends to be a linear function of component concentration of each component by
in the mixture. As viscosity ratio increases, the more common
trend for oil mixtures is the concave-upward response. The higher Vi i
wi = n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4)
the viscosity ratio, the more concave the relationship becomes.
Therefore, for low viscosity ratios, simple linear relationships could ∑V  i i
i =1
be used while still achieving acceptably accurate results. However,
the authors do not recommend this for common practice because Likewise, volume fraction can be determined from weight fraction
more-general and higher-accuracy procedures exist. and specific gravity:

TABLE 3—ACCURACY OF BLEND VISCOSITY

CALCULATED USING SIMPLE AVERAGING TECHNIQUES

>10%
Method # Pts % AE Std Dev % AAE Std Dev Error Count
10 12 10 12
Volume average 2,059 6.91 10 2.92 10 6.91 10 2.92 10 1,535
10 12 10 12
Mole average 2,059 3.06 10 1.30 10 3.06 10 1.30 10 1,283
10 12 10 12
Weight average 2,059 7.81 10 3.31 10 7.81 10 3.31 10 1,564

Effect of Viscosity Ratio

50
All Data
45 1.0 < µrat < 1.5
1.5 < µrat < 2.0
40 2.0 < µrat < 3.0
Average Absolute Error, %

3.0 < µrat < 5.0

35
5.0 < µrat < 10
10 < µrat < 100
30
100 < µrat < 500
25 500 < µrat < 2000
2000 < µrat < 10,000
20 µrat > 10,000

15

10

0
Volume Average Mole Average Weight Average

Fig. 5—Accuracy of simple blend-viscosity techniques.

February 2012 SPE Production & Operations 109

 Volume Average Bergman and Sutton
100 100

75 75

% Error
50 50
% Error

25 25

0
0

–25
–25
1 2 3 4 5 6 7 8 9 10
1 2 3 4 5 6 7 8 9 10
Viscosity Ratio Viscosity Ratio

Fig. 6—Effect of viscosity ratio on error resulting from volu- Fig. 7—Effect of viscosity ratio on error from Bergman and
metric averaging. Sutton.

wi  i 141.5
Vi = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5)  API =
− 131.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)
n
m
∑w i i
The molecular weight of the mixture is a mole-fraction-weighted
i =1
average of the component molecular weights:
The calculation of mole fraction from weight fraction follows n
M m = ∑ xi Mi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
wi Mi
xi = n , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6) i =1

∑w i Mi The Watson K factor for each component oil can be determined

from true boiling point (TBP) analysis (UOP 1986). Alternatively,
i =1
it can be estimated from the component specific gravity and
while the calculation of mole fraction from volume fraction molecular weight. The latter procedure is more commonly used.
requires both component specific gravity and mole weight: The average boiling point temperature (BPT) can then be readily
determined.
Vi i Mi
xi = n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7) The calculation of both the mixture specific gravity and the
∑ Vii Mi molecular weight are well defined by the literature; however,
different criteria could be used to determine an average Watson
i =1
K factor for the mixture. By definition, the Watson K factor is
The empirical methods from the literature use volume, weight, calculated by
or mole fraction together with the component viscosity to predict
blend viscosity. The Bergman and Sutton method can be used to Tb1 3
predict the viscosity of the blend; however, this method requires Kw = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11)
only the specific gravity and Watson K factor of the mixture along 
with temperature. The Watson K factor provides a convenient way Options investigated to determine the Watson K factor for the
to characterize the oil while providing insight into the paraffin mixture were
content of the oil. The real benefit of using the Bergman and Sut- 1. Calculate Watson K factor directly from the mixture gravity
ton method is that the viscosity of each component in the blend and molecular weight using Eqs. B-3 and B-4.
does not need to be measured at each temperature specified for the 2. Mole-fraction-weighted average BPT.
blend-viscosity determination. 3. Weight-fraction-weighted average BPT
The specific gravity of the mixture is derived from a volume- 4. Volume-fraction-weighted average BPT.
weighted average of the component specific gravities as expressed Each option was tested using the Bergman and Sutton method to
here: calculate the blend viscosity for each mixture, and then the result
n was compared with the measured blend viscosity. The results, sum-
 m = ∑ Vi i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8) marized in Table 4, clearly indicate that volume-fraction-weighted
i =1
average boiling point temperature should be used to determine the
And the API gravity of the mixture is calculated from its definition: average Watson K factor for the mixture.

TABLE 4—RESULTS OF DIFFERENT TECHNIQUES FOR AVERAGING BPT

WITH BERGMAN AND SUTTON VISCOSITY METHOD
Std % Std >10% Error
Method # Pts % AE Dev AAE Dev Count

BPT from Kw derived from

mixture MW and specific 2,059 –1.3 117.2 34.9 111.8 1,501
gravity with Eqs. B-3 and B-4
Mole fraction average BPT 2,059 –22.9 29.1 26.4 25.9 1,225
Weight fraction average BPT 2,059 9.2 20.3 14.9 16.5 964
Volume fraction average BPT 2,059 1.5 15.6 11.0 11.2 795

110 February 2012 SPE Production & Operations

 TABLE 5—ACCURACY OF BLEND VISCOSITY METHODS

>10% Error
Method # Pts % AE Std Dev % AAE Std Dev Count

Arrhenius 2,059 –12.9 27.0 20.0 22.2 1,049

Bingham 2,059 –33.7 32.6 34.4 31.8 1,334
Kendall 8 10 8 10
2,059 5.65 10 2.45 10 5.65 10 2.45 10 1,001
and Monroe
Cragoe 2,059 15.7 30.8 18.7 29.0 846
Refutas 1,978 1.4 19.2 10.4 16.2 585
Maxwell 1,978 –8.4 18.8 13.2 15.9 780
Koval 2,059 –5.5 20.8 12.5 17.5 715
Chevron 1 2,059 –4.7 17.2 11.5 13.7 757
Chevron 2 2,059 2.7 16.7 9.8 13.7 627
Shu 2,059 2.9 23.2 15.5 17.5 1,012
Wallace
2,059 –2.3 17.2 10.3 13.9 673
and Henry
Barrufet 2,059 29.3 417.5 41.4 416.5 1,077
Parkash 2,059 –3.3 17.1 10.4 14.0 669
Bergman
2,059 1.5 15.6 11.0 11.2 795
and Sutton

Results viscosities (less than 1.77 cSt) result in excessive error for this
Methods described in Appendix A and the Bergman and Sutton method. Therefore, in order to compare methods, a subset (517
method (Appendix B) were evaluated against the full database. data points) of the database was created using only data with
The results are provided in Table 5 and indicate a wide range component kinematic viscosity greater than 1.77 cSt. The results
of outcomes for the complete database. The top five methods as are presented in Table 6. Several methods provide more accurate
ranked by lowest average absolute error are results than the McAllister method, and because of its limited
1. Chevron 2 applicability; it was eliminated from further comparisons. The
2. Wallace and Henry Refutas and Maxwell methods are limited to blends with compo-
3. Refutas nent kinematic viscosity greater than 0.2 cSt, which resulted in a
4. Parkash minimal number of data points (81) that had to be excluded from
5. Bergman and Sutton evaluation. The resulting errors of all methods were plotted against
The spread in absolute error for these methods is approximately blend viscosity and are summarized in Figs. 8 through 22.
1%, and the Bergman and Sutton method is more consistent than Various metrics were evaluated to aid in assessing the accu-
the other methods, as indicated by its lower standard deviation. racy of the various viscosity blending methods. These included
As discussed in Appendix A, some of the published methods the ranges of blend viscosity and API gravity, difference in API
have limited applicability. The method offered by McAllister gravity, difference in Watson characterization factor, fraction of the
is limited by the defined range of the Saybolt viscosity scale. light component in the mixture, viscosity ratio, and temperature.
Attempts to evaluate mixtures with low component kinematic For mixtures with more than two components, the fraction of light

TABLE 6—ACCURACY OF BLEND VISCOSITY METHODS ON REDUCED DATA SET

FOR COMPARISON WITH THE MCALLISTER (2005) METHOD

>10% Error
Method # Pts % AE Std Dev % AAE Std Dev Count

Arrhenius 517 5.3 23.3 13.9 19.4 188

Bingham 517 –24.0 25.4 25.2 24.2 331
Kendall
517 154.2 1,593.7 156.2 1,593.5 225
and Monroe
Cragoe 517 7.7 20.1 10.7 18.7 130
Refutas 517 4.5 17.1 8.9 15.3 112
Maxwell 517 0.5 15.5 8.4 13.0 111
Koval 517 –1.5 18.8 10.1 15.9 137
Chevron 1 517 3.8 16.3 8.6 14.3 112
Chevron 2 517 3.4 17.3 9.0 15.2 116
Shu 517 15.2 24.3 19.9 20.6 313
Wallace
517 2.9 17.4 9.0 15.1 115
and Henry
Barrufet 517 0.5 29.3 17.0 23.9 271
Parkash 517 0.9 15.7 8.5 13.3 108
McAllister 517 2.5 30.9 18.0 25.3 229
Bergman
517 3.6 15.0 11.4 10.4 212
and Sutton

February 2012 SPE Production & Operations 111

 Arrhenius Method Bingham Method
200 200

150 150

100 100

% Error
% Error

50
50

0
0

–50
–50

–100
–100 0.1 1 10 100 1,000 10,000
0.1 1 10 100 1,000 10,000 Viscosity, cp
Viscosity, cp
Fig. 9—Error in calculated blend viscosity from Bingham.
Fig. 8—Error in calculated blend viscosity from Arrhenius.

Kendall and Monroe Cragoe

200 200

150 150

100 100
% Error
% Error

50 50

0 0

–50 –50

–100 –100
0.1 1 10 100 1,000 10,000 0.1 1 10 100 1,000 10,000
Viscosity, cp Viscosity, cp

Fig. 10—Error in calculated blend viscosity from Kendall and Fig. 11—Error in calculated blend viscosity from Cragoe.
Monroe.

component was defined as the sum of components with API gravity have a low viscosity and/or the concentration of the high-viscos-
greater than the API gravity of the blend. ity component must be small. In either case, the deviation from
The results in Fig. 23 compare correlation error for different an ideal mixture is minimal. As the blend viscosity increases, the
ranges of blend viscosity. For a mixture viscosity less than 0.5 opportunity exists for a larger difference in component viscosity
cp, all of the methods provide reasonable results; however, for and character. A larger error is observed for many of the methods,
the mixture to have a viscosity this low, the components must although the change is minimized for the five better methods iden-

Refutas Maxwell
200 200

150 150

100 100
% Error
% Error

50 50

0 0

–50 –50

–100 –100
0.1 1 10 100 1,000 10,000 0.1 1 10 100 1,000 10,000

Viscosity, cp Viscosity, cp

Fig. 12—Error in calculated blend viscosity from Refutas. Fig. 13—Error in calculated blend viscosity from Maxwell.

112 February 2012 SPE Production & Operations

 Koval Chevron 1
200 200

150 150

100 100
% Error

% Error
50 50

0
0

–50
–50

–100
–100
0.1 1 10 100 1,000 10,000
0.1 1 10 100 1,000 10,000
Viscosity, cp
Viscosity, cp
Fig. 14—Error in calculated blend viscosity from Koval.
Fig. 15—Error in calculated blend viscosity from Chevron
Method 1.

Chevron 2 Shu
200
200

150
150

100
100
% Error
% Error

50
50

0
0

–50
–50

–100
–100
0.1 1 10 100 1,000 10,000
0.1 1 10 100 1,000 10,000
Viscosity, cp
Viscosity, cp
Fig. 17—Error in calculated blend viscosity from Shu.
Fig. 16—Error in calculated blend viscosity from Chevron
Method 2.

tified previously. The error resulting from the Bergman and Sutton The correlation error with changing mixture API gravity is
method is comparable to or slightly larger than that of the other shown in Fig. 24. The conclusions derived from this graph are
four methods identified. This is unexpected because Bergman and essentially the same as those drawn from Fig. 23. However, in this
Sutton is completely predictive, while the other methods require case, mixtures with lower API gravity will have higher viscosity
actual measured component-viscosity data for the calculation of compared to mixtures with high API gravity. Therefore, the larger
blend viscosity. errors are associated with the lower API gravity blends.

Barrufet and Setiadarma Wallace and Henry

200 200

150 150

100 100
% Error

% Error

50 50

0 0

–50 –50

–100 –100
0.1 1 10 100 1,000 10,000 0.1 1 10 100 1,000 10,000
Viscosity, cp Viscosity, cp

Fig. 18—Error in calculated blend viscosity from Barrufet and Fig. 19—Error in calculated blend viscosity from Wallace and
Setiadarma. Henry.

February 2012 SPE Production & Operations 113

 McAllister Parkash
200 200

150 150

100 100
% Error

% Error
50 50

0 0

–50 –50

–100 –100
0.1 1 10 100 1,000 10,000 0.1 1 10 100 1,000 10,000

Viscosity, cp Viscosity, cp

Fig. 20—Error in calculated blend viscosity from McAllister. Fig. 21—Error in calculated blend viscosity from Parkash.

Bergman and Sutton Fig. 25 shows correlation error as affected by differences in

200 the component API gravity. A larger difference in the component
API gravity indicates a greater contrast in the liquids blended,
150 which tests the correlations’ ability to model the data. Of the
five methods identified, the Bergman and Sutton method is most
100 consistent, which not only validates the proposed mixing rules but
also demonstrates the robustness of the method. Fig. 26 provides
% Error

50
correlation error with differences in the component Watson K fac-
tors. A larger difference in this parameter indicates a greater dis-
0
similarity or change in the paraffin content of the liquids blended.
All methods show an increase in error as the blended liquids
become more dissimilar.
–50
Fig. 27 depicts the correlation error as affected by the volume
fraction of the lighter component. Errors increase as the volume
–100
0.1 1 10 100 1,000 10,000
fraction approaches 0.5, which corresponds to the larger vol-
ume difference in components. The error reduces as the fraction
Viscosity, cp
increases to unity.
Fig. 28 shows correlation error as the viscosity ratio between com-
Fig. 22—Error in calculated blend viscosity from Bergman and
Sutton. ponents changes. This parameter is defined as the ratio of the heavy-

Comparison of Viscosity Blending Methods

50
All Data
45 µ < 0.5 cp
0.5 < µ < 1.0 cp
40 1.0 < µ < 5 cp
Average Absolute Error, %

5 < µ < 10 cp
35 10 < µ < 100 cp
µ > 100 cp
30

25

20

15

10

0
e

a
1

ry
l

u
s

as
m

sh

n
l
el
e
ro

va

rm
iu

Sh

to
en
d

d
go
ha

w
ut

a
on

ho

ho
en

Ko

t
da

rk
ax

Su
H
ra

ef
ng

et
et
rh

Pa
tia
M
C

d
Bi

M
Ar

d
an
nd

Se

an
on

on
la

ce

nd

an
vr

vr
al

la
he
he

ta

gm
nd

al
C

fe
Ke

r
Be
rru
Ba

Fig. 23—Summary of blend-viscosity methods by final blend viscosity.

114 February 2012 SPE Production & Operations

 Error in Blend Viscosity by Blend °API Range
50
All Data
45 °API < 20
20 < °API < 30
40
30 < °API < 40
Average Absolute Error, %

35 40 < °API < 60

°API > 60
30

25

20

15

10

h
m

a
oe

as

2
us

el

n
va

ry

s
m
Sh
go

tto
w
ha

od

od
ut

ka
ni

en
r

Ko

da
on

ax
ra

Su
he

ef
ng

r
H

Pa
tia
M

et

et
M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba
Fig. 24—Summary of blend-viscosity methods by final blend API gravity.

component viscosity to the light-component viscosity and is probably generally show better or equal accuracy and greater consistency when
one of the better indicators of correlation performance. As the ratio compared to the other four methods. Figs. 29 through 33 provide
increases, the blended liquids become more dissimilar. For low values more details on these five methods and illustrate the consistency
of viscosity ratio (i.e., values less than 10), the errors observed from and accuracy of the Bergman and Sutton method at higher viscosity
the top four procedures [Chevron 2; Wallace and Henry; Refutas; ratios. A closer examination of these charts shows a bias in the Wal-
and Parkash] are less than those reported for Bergman and Sutton. lace and Henry and Parkash methods. At viscosity ratios greater than
This is the result of these methods’ use of the actual measured com- 100, these methods show a tendency to underpredict blend viscosity
ponent viscosity while the Bergman and Sutton method is entirely while the errors resulting from Chevron 2, Refutas, and Bergman and
predictive. At viscosity ratios greater than 10, Bergman and Sutton Sutton are more closely centered around zero error.

Error in Blend Viscosity by Difference in Component °API Gravity

50
All Data
45 0 < °API < 5
5 < °API < 10
40
10 < °API < 25
Average Absolute Error, %

35 25 < °API < 50

50 < °API < 75
30 °API > 75

25

20

15

10

0
l

h
u
m

a
oe

as

2
us

el

n
va

ry

s
m
Sh
go

tto
w
ha

od

od
ut

ka
ni

en
r

Ko

da
on

ax
ra

Su
he

ef
ng

r
H

Pa
ia
et

et
M

M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba

Fig. 25—Summary of blend-viscosity methods by difference in component API gravity.

February 2012 SPE Production & Operations 115

 Error in Blend Viscosity by Difference in Watson Factor
50
All Data
45 0.0 < Kw < 0.25
0.25 < Kw < 0.5
40 0.5 < Kw < 0.75
Average Absolute Error, %

0.75 < Kw < 1.0

35
1.0 < Kw < 1.5
Kw > 1.5
30

25

20

15

10

0
l

h
u
m

a
oe

as

2
us

el

n
va

ry

s
am
Sh
go

tto
w
ha

od

od
ut

ka
en
ni

Ko
on

ax
ra

Su
he

ef

d
ng

r
H

Pa
tia
et

et
M

M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba
Fig. 26—Summary of blend-viscosity methods by difference in component Watson K factor.

Fig. 34 provides the variation in error with temperature. At least exhibit physically incorrect behavior with changing temperature.
one other investigator (Degiorgis et al. 2001) has reported changes The Bergman and Sutton method was designed to provide consis-
in correlation error with changes in temperature. The five methods tent results with changing temperature. The behavior of the method
identified in this work show relatively small changes in error with with this data set confirms this observation.
changing temperature. It is important to note that the Chevron 2,
Wallace and Henry, Refutas, and Parkash methods require measured Volumetric Shrinkage
component viscosity for each temperature evaluated, while the When low- and high-density oils are mixed, the resulting volume is
Bergman and Sutton method requires no measured viscosity data. less than the sum of the individual volumes. The discrepancy in vol-
Furthermore, many of the predictive dead oil viscosity methods ume occurs on the molecular level when large and small molecules

Error in Blend Viscosity by Fraction of Light Component

50
All Data
45 0.0 < Vlight < 0.1
0.1 < Vlight < 0.2
40 0.2 < Vlight < 0.3
Average Absolute Error, %

0.3 < Vlight < 0.4

35
0.4 < Vlight < 0.5
0.5 < Vlight < 0.6
30
0.6 < Vlight < 0.7
25 0.7 < Vlight < 0.8
0.8 < Vlight < 0.9
20 0.9 < Vlight < 1.0

15

10

0
l

h
u
m

a
oe

as

2
us

el

n
va

ry

s
m
Sh
go

tto
w
ha

od

od
ut

ka
ni

en
r

Ko

da
on

ax
ra

Su
he

ef
ng

r
H

Pa
tia
et

et
M

M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba

Fig. 27—Summary of blend-viscosity methods by volume fraction of light component in blend.

116 February 2012 SPE Production & Operations

 Effect of Viscosity Ratio
50
All Data

45 1.0 < µrat < 1.5

1.5 < µrat < 2.0
40 2.0 < µrat < 3.0
Average Absolute Error, %

3.0 < µrat < 5.0

35 5.0 < µrat < 10
10 < µrat < 100
30 100 < µrat < 500
500 < µrat < 2000
25 2000 < µrat < 10,000
µrat > 10,000
20

15

10

h
u
m

a
oe

as

2
us

el

n
va

ry

s
am
Sh
go

tto
w
ha

od

od
ut

ka
ni

en
r

Ko
on

ax
ra

Su
he

ef

d
ng

r
H

Pa
tia
et

et
M

M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba
Fig. 28—Summary of blend-viscosity methods by viscosity ratio (heavy-component viscosity/light-component viscosity).

Chevron 2 Wallace and Henry

150 150

100 100
% Error

% Error

50 50

0 0

–50 –50

–100 –100
1 10 100 1,000 10,000 100,000 1,000,000 1 10 100 1,000 10,000 100,000 1,000,000
Viscosity Ratio Viscosity Ratio

Fig. 29—Accuracy of Chevron Method 2 with changing viscosity Fig. 30—Accuracy of Wallace and Henry method with changing
ratio. viscosity ratio.

Parkash
Refutas 150
150

100
100
% Error

50
% Error

50

0 0

–50 –50

–100
–100
1 10 100 1,000 10,000 100,000 1,000,000
1 10 100 1,000 10,000 100,000 1,000,000
Viscosity Ratio Viscosity Ratio

Fig. 31—Accuracy of Refutas method with changing viscosity Fig. 32—Accuracy of Parkash method with changing viscosity
ratio. ratio.

February 2012 SPE Production & Operations 117

 Bergman and Sutton biodiesel, condensate, naphtha, dead crude, and assay fractions.
150 Blends comprised the typical binary mixtures but ranged up to
a maximum of eight components in the mixture.
• Several correlations from the literature for calculating blend
100
viscosity were used to benchmark expected accuracies for calcu-
lated blend viscosity for this database. It is important to note that
all of these methods require the viscosity of each component in
% Error

50
the blend to be measured. Therefore, for the 2,059 blends stud-
ied in this paper, a total of 7,646 components were mixed that
0 required a total of 4,340 individual viscosity measurements in
order to evaluate these methods. Use of the Bergman and Sutton
method would eliminate the required viscosity measurements.
–50
• Simple procedures such as volume-, weight-, or mole-fraction-
weighted averaging techniques can be used to estimate blend vis-
–100 cosity, provided the ratio of component viscosity is sufficiently low.
1 10 100 1,000 10,000 100,000 1,000,000 Results for this database would indicate an upper limit of three. The
Viscosity Ratio authors do not recommend these procedures for general application
because the errors about this limit become quite large. Other more-
Fig. 33—Accuracy of Bergman and Sutton method with chang- general methods are available that avoid this problem.
ing viscosity ratio. • All of the methods evaluated provide excellent results when
the viscosity ratio is sufficiently low. The Kendall and Mon-
roe method, which has been recommended by API, is another
are mixed. The smaller molecules in the low-density liquid fill the example of a method that performs well at low viscosity ratios.
voids between the larger molecules in the more-dense liquid. This At viscosity ratios greater than five, the errors from this method
phenomenon has been modeled by API MPMS 12.3 (2006): begin to grow and quickly become unacceptably large. This
S = 4.86 × 10 −6 V2 ⎡⎣100 (1 − V2 ) ⎤⎦ ( ) method should be avoided with high-viscosity-ratio blends.
0.819 2.28
API 2 −  API1 , . . . . . (12) • Methods reported in the literature including Chevron 2; Wallace and
Henry; Refutas; and Parkash can be used to calculate blend-viscos-
where  API 2 >  API1 and V2 is the volume fraction of the light com- ity in the traditional manner for a wide range of conditions.
ponent. Referring to Eq. 12, the amount of shrinkage is governed • Viscosity ratio provides one of the best metrics for comparing
by the difference in API gravity between the two liquids and the the accuracy of blend-viscosity correlations.
volume fraction of the lighter liquid, as illustrated in Fig. 35. • Mixture properties of the blend should be determined using the
Application of Eq. 12 with the database showed shrinkage ranging following techniques:
0–0.87%, with an average of 0.1%. This reduction in volume has o Specific gravity: volume-weighted average of the component
a small and negligible effect on the results of the blend- viscosity property
methods, as illustrated in Fig. 36. o Molecular weight: mole-fraction-weighted average of the
component property
Conclusions o BPT: volume-weighted average of the component property
• A large, diverse database of blended-hydrocarbon-viscosity • The application of a volumetric-shrinkage correction resulting
data was compiled. Components in the blends had vastly dif- from blending two or more oils with different API gravities
ferent physical and chemical characteristics and ranged from does not materially impact the calculation of blend viscosity.
light-alkane or aromatic pure components to bitumen, diesel, Therefore, this correction can be ignored.

Effect of Temperature
50
All Data
45 T < 75°F
75 < T < 100°F
40
Average Absolute Error, %

100 < T < 150°F

35 150 < T < 200°F
T > 200°F
30

25

20

15

10

0
l

h
u
m

a
oe

as

2
us

el

n
va

ry

s
m
Sh
go

tto
w
ha

od

od
ut

ka
en
ni

Ko

da
on

ax
ra

Su
he

ef
ng

r
H

Pa
tia
et

et
M

M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba

Fig. 34—Summary of blend-viscosity methods by different temperature environments.

118 February 2012 SPE Production & Operations

 1.5 1.5

1.25 1.25

Vol % Shrinkage
1

Vol % Shrinkage
0.75
0.75

0.5
0.5

0.25
0.25

0
102030 0
Vol % 4050 60 70 50 60 70
Light 80 90 0 20 30 40
10 0 10 ffere nce
API Gravity Di

Fig. 35—Behavior of blend volume-shrinkage equation.

• The Bergman and Sutton method for calculating dead oil viscosity Statistical Quantities
can be used to estimate the viscosity for hydrocarbon blends. The AE = average error, %,
mixing rules identified in Conclusion 7 must be used in order to
calculate the viscosity of the blend using the mixture specific grav- 100 N X icalc − X imeas
ity, mixture Watson K factor (Eq. 11), and temperature. The Berg- AE = ∑ X
N i =1 imeas
man and Sutton method is predictive and does not require compo-
nent-viscosity measurements for each temperature evaluated. For AAE = average absolute error, %,
viscosity ratios in excess of 10, the accuracy of the Bergman and
Sutton method equals or exceeds the accuracy of methods that 100 N X icalc − X imeas
require measured component viscosity. Therefore, the Bergman AAE = ∑ X
N i =1 imeas
and Sutton method can be a valuable procedure for calculating
blend viscosities over a very wide range of conditions. SD = standard deviation,
N

∑( X − X)
2
Acknowledgments
i
The authors would like to thank the management of Marathon Oil SD = i =1
Company and BP America for permission to publish this paper. N −1
Finally, the primary author would like to thank his wife, Nancy.
Without her patience and understanding, this paper would never X = generic dependent variable
have been written. N = number of observations

Effect of Blend Shrinkage

50
No Shrink
45 Shrink

40
Average Absolute Error, %

35

30

25

20

15

10

0
l

h
u
m

a
oe

as

2
us

el

n
va

ry

s
m
Sh
go

tto
w
ha

od

od
ut

ka
ni

en
r

Ko

da
on

ax
ra

Su
he

ef
ng

r
H

Pa
tia
et

et
M

M
C

R
r

Bi

d
d
M

M
Ar

Se
nd

an
an
on

on
a

nd

an
ce
vr

vr
ll

ta
da

gm
la
he

he

al

fe
n

r
Ke

Be
rru
Ba

Fig. 36—Summary of blend-viscosity methods comparing the use or omission of volume shrinkage.

February 2012 SPE Production & Operations 119

Nomenclature Barrufet, M.A. and Setiadarma, A. 2003. Reliable heavy oil-solvent viscos-
ity mixing rules for viscosities up to 450 K, oil-solvent viscosity ratios
IC1 = Chevron blend index (Method 1)
up to 4 x 105, and any solvent proportion. Fluid Phase Equilibria 213
IC2 = Chevron blend index (Method 2)
(1–2): 65–79. http://dx.doi.org/10.1016/S0378-3812(03)00285-1.
ICr = Cragoe blend index Barrufet, M.A., Hall, K.R., Estrada-Baltazar, A., and Iglesias-Silva, G.A.
IM = Maxwell blend index 1999. Liquid Viscosity of Octane and Pentane + Octane Mixtures from
IP = Parkash blend index 298.15 K to 373.15 K Up To 25 MPa. J. Chem. Eng. Data 44 (6):
IR = Refutas blend index 1310–1314. http://dx.doi.org/10.1021/je990043z.
IWH = Wallace and Henry blend index Bergman, D.F. and Sutton, R.P. 2007. A Consistent and Accurate Dead-
Kw = Watson characterization factor Oil-Viscosity Method. Paper SPE 110194 presented at the SPE Annual
S = volumetric shrinkage, fraction Technical Conference and Exhibition, Anaheim, California, USA,
T = temperature, °F 11–14 November. http://dx.doi.org/10.2118/110194-MS.
Chevron Crude Oil Assays. Chevron Crude Oil Marketing, http://
Tabs = temperature, °R
crudemarketing.chevron.com/.
Tb = average boiling point temperature, °R
Cragoe, C.S. 1933. Changes in the Viscosity of Liquids with Temperature,
Tco = n-alkane critical temperature, °R Pressure and Composition. Proc., 1st World Petroleum Congress, Lon-
V = volume fraction don, 18–24 July, Paper WPC 201, 529–541.
w = weight fraction Degiorgis, G.L., Maturano, S., Garay, M., Galliano, G.R., and Fornes, A.
x = mole fraction 2001. Oil Mixture Viscosity Behavior: Use in Pipeline Design. Paper
 = reduced boiling point temperature SPE 69420 presented at the SPE Latin American and Caribbean Petro-
60 = coefficient of thermal expansion at a base tem- leum Engineering Conference, Buenos Aires, Argentina, 25–28 March.
perature of 60°F http://dx.doi.org/10.2118/69420-MS.
API = API gravity, ° Díaz, R.M., Berardo, I., Fernández, A.M., and Folgueras, M.B. 1986. Pre-
o = oil specific gravity diction of the viscosity of lubricating oil blends at any temperature. Fuel
75 (5): 574–578. http://dx.doi.org/10.1016/0016-2361(95)00289-8.
 = kinematic viscosity, cSt
Drayer, D.E. 1985. Viscosities of Crude Oil Mixtures. Internal Report,
SSU = Saybolt Universal viscosity, seconds
Marathon Oil Company, Houston, Texas (21 March 1985).
 = absolute or dynamic viscosity, cp Estrada-Baltazar, A., Alvarado, J.F.J., Iglesias-Silva, G.A., and Barrufet,
od = dead-oil viscosity, cp M.A. 1998a. Experimental Liquid Viscosities of Decane and Octane +
rat = viscosity ratio, high-viscosity component/low- Decane from 298.15 K to 373.15 K and Up to 25 MPa. J. Chem. Eng.
viscosity component Data 43 (3): 441–446. http://dx.doi.org/10.1021/je970233e.
 = density, g/cm3 Estrada-Baltazar, A., Iglesias-Silva, G.A., and Barrufet, M.A. 1998b.
o = oil density, g/cm3 Liquid Viscosities of Pentane and Pentane + Decane from 298.15 K
o60 = oil density at 60 °F, g/cm3 to 373.15 K and Up To 25 MPa. J. Chem. Eng. Data 43 (4): 601–604.
oT = oil density at temperature T, g/cm3 http://dx.doi.org/10.1021/je9702697.
x, f1, f2, 1, 2 = correlating parameters Exxon Crude Oil Assays. ExxonMobil, http://www.exxonmobil.com/apps/
crude_oil/index.html.
Subscripts Gary, J.H., Handwerk, G.E., and Kaiser, M.J. 2007. Petroleum Refin-
ing: Technology and Economics, fifth edition, 262–263. Boca Raton,
1 = heavy component
Florida: CRC Press.
2 = light component
Ha, H.Z. and Koppel, P. 2008. Accurately Predict Viscosity of Syncrude
100 = property at 100°F Blends. Hydrocarbon Processing (Jul-08): 87–92.
210 = property at 210°F Hernández-Galván, M.A. García-Sánchez, F., and Macías-Salinas, R. 2007.
i = component Liquid Viscosities of benzene, n-tetradecane, and benzene + n-tetra-
m = mixture decane from 313 to 393 K and pressures up to 60 MPa: Experiment
and modeling. Fluid Phase Equilibria 262 (1–2): 51–60. http://dx.doi.
References org/10.1016/j.fluid.2007.08.010.
Al-Besharah, J.M. 1989. The Effect of Blending on Selected Physical Prop- Iglesias-Silva, G.A., Estrada-Baltazar, A., Hall, K.R., and Barrufet, M.A.
erties of Crude Oils and Their Products. PhD dissertation, University 1999. Experimental Liquid Viscosity of Pentane + Octane + Decane
of Ashton, Birmingham, Alabama (February 1989). Mixtures from 298.15 to 373.15 K up to 25 MPa. J. Chem. Eng. Data
Al-Besharah, J.M., Akashah, S.A., and Mumford, C.J. 1989. The effect 44 (6): 1304–1309. http://dx.doi.org/10.1021/je990044r.
of temperature and pressure on the viscosities of crude oils and Irving, J.B. 1977a. Viscosity of Binary Liquid Mixtures: A Survey of Mix-
their mixtures. Ind. Eng. Chem. Res. 28 (2): 213–221. http://dx.doi. ture Equations. Report 630, National Engineering Laboratory, Glasgow,
org/10.1021/ie00086a014. Scotland (February 1977).
Al-Besharah, J.M., Salman, O.A., and Akashah, S.A. 1987. Viscosity of Irving, J.B. 1977b. Viscosity of Binary Liquid Mixtures: The Effectiveness
crude oil blends. Ind. Eng. Chem. Res. 26 (12): 2445–2449. http:// of Mixture Equations. Report 631, National Engineering Laboratory,
dx.doi.org/10.1021/ie00072a010. Glasgow, Scotland (February 1977).
Anhorn, J.L. and Badakhshan, A. 1994. MTBE: A Carrier for Heavy Oil Joshi, R.M. and Pegg, M.J. 2007. Flow properties of biodiesel fuel blends
Transportation and Viscosity Mixing Rule Applicability. J. Cdn. Pet. at low temperature. Fuel 86 (1–2): 143–151. http://dx.doi.org/10.1016/
Tech. 33 (4): 17–21. j.fuel.2006.06.005.
API MPMS 12.3, Manual of Petroleum Measurement Standards, Chapter Koval, E.J. 1963. A Method for Predicting the Performance of Unstable
12: Calculation of Petroleum Quantities—Section 3: Calculation of Miscible Displacement in Heterogeneous Media. SPE J. 3 (2): 145–154;
Volumetric Shrinkage Resulting From Blending Light Hydrocarbons Trans., AIME, 228. SPE-450-PA. http://dx.doi.org/10.2118/450-PA.
With Crude Oil. 2006. Washington, DC: API. Maxwell, J.B. 1950. Data Book On Hydrocarbons: Application to Pro-
API Technical Data Book—Petroleum Refining, sixth edition, Chap. 11. cess Engineering, Sec. 9. Princeton, New Jersey: D. Van Nostrand
1997. Washington, DC: API. Company, Inc.
ASTM D2161-05, Standard Practice for Conversion of Kinematic Viscos- McAllister, E.W. ed. 2005. Pipeline Rules of Thumb Handbook, sixth edition,
ity to Saybolt Universal Viscosity or to Saybolt Furol Viscosity. 2005. Chap. 12, 346. Oxford, UK: Elsevier (Gulf Professional Publishing).
West Conshohocken, Pennsylvania: ASTM International. http://dx.doi. Orbey, H. and Sandler, S.I. 1993. The Prediction of the Viscosity of Liq-
org/10.1520/D2161-05. uid Hydrocarbons and Their Mixtures as a Function of Temperature
Baird, C.T. IV. 1989. Guide To Petroleum Product Blending, Chap. 8. and Pressure. Canadian Journal of Chemical Engineering 71 (June):
Flagstaff, Arizona: HPI Consultants, Inc. 437–446.

120 February 2012 SPE Production & Operations

Parkash, S. 2003. Refining Processes Handbook, Chap. 11, 327–328. The Arrhenius method was originally specified using a volume-
Oxford, UK: Elsevier (Gulf Professional Publishing). fraction-weighted averaging scheme. Kendall and Monroe pointed
Ramakrishnan, H.R., Khataniar, S. Dandekar, A.Y., Patil, S.L., Chukwu, out that using weight or mole-fraction-weighted average could
G.A., Kamath, V.A., Haslebacher, W.F., and Hackworth, J.H. 2003. produce more-accurate results. On the basis of the results below,
Characterization of Gas-to-Liquid (GTL) and Alaska North Slope the mole-fraction-weighted average scheme was selected for use
Crude (ANSC) Oil Blend Properties for Flow Through the Trans Alaska elsewhere in this paper.
Pipeline System (TAPS) by Density and Viscosity Measurements. The API Technical Data Book (1997) notes that the Kendall
Petroleum Science and Technology 21 (1–2): 301–314. http://dx.doi. and Monroe method is applicable to blends where the compo-
org/10.1081/LFT-120016950. nents are similar in nature. On the basis of this observation, it is
Reid, R.C., Prausnitz, J.M., and Poling, B.E. 1987. The Properties of Gases expected that this method will yield poor results for mixtures as
and Liquids, fourth edition, 476. New York: McGraw-Hill, Inc. the viscosity ratio between the components increases. Referring
Riazi, M.R. 2005. Characterization and Properties of Petroleum Fractions, to Fig. 28, the accuracy of this method suffers when the viscosity
Chap. 8. West Conshohocken, Pennsylvania: ASTM. ratio exceeds five.
Roelands, C.J.A. 1966. Correlational Aspects of the Viscosity-Tempera- The Chevron 1 and Parkash methods exist in the literature
ture-Pressure Relationship of Lubricating Oils. PhD thesis, University as tables. The tables were curve fitted and the resulting equa-
of Delft, Delft, The Netherlands. tions found in Table A-1 were used for all further analysis. Data
Shigemoto, N., Al-Maamari, R.S., Jibril, B.Y., and Hirayama, A. 2006. in the database exceeded limits found in the original tables. No
A Study of the Effect of Gas Condensate on the Viscosity and Stor- constraints were placed on the equations, and reasonable, accurate
age Stability of Omani Heavy Crude Oil. Energy & Fuels 20 (6): results were obtained from both methods.
2504–2508. http://dx.doi.org/10.1021/ef060074h. Viscosity methods, such as Shu or Barrufet and Setiadarma
Shu, W.R. 1984. A Viscosity Correlation for Mixtures of Heavy Oil, Bitu- are formulated to calculate the blend viscosity of a two-compo-
men, and Petroleum Fractions. SPE J. 24 (3): 277–282. SPE-11280-PA. nent mixture. For blends with more than two components, the
http://dx.doi.org/10.2118/11280-PA. components are ordered from high to low viscosity. Components
Tat, M.E. and Van Gerpen, J.H. 1999. The Kinematic Viscosity of Biodiesel 1 and 2 are blended, forming an intermediate mixture, which is
and Its Blends with Diesel Fuel. Journal of the American Oil Chemists’ then blended with Component 3 and so on. There was no bias
Society (JAOCS) 76 (12): No. J9166, 1511–1513. in error resulting from blends with more than two components,
Tat, M.E. and Van Gerpen, J.H. 2000. The Specific Gravity of Biodiesel which illustrates this as a viable process for using these methods
and Its Blends with Diesel Fuel. Journal of the American Oil Chemists’ originally designed for two-component blends.
Society (JAOCS) 77 (2): No. 9143, 115–119. Wallace and Henry examined Cragoe’s method and proposed a
Tat, M.E. and Van Gerpen, J.H. 2002. Physical Properties and Composi- modified version as detailed in Table A-1. The constant (0.01) in
tion Detection of Biodiesel-Diesel Fuel Blends. Paper ASAE 02-6084 their equation was determined because it minimized the average
presented at the American Society of Agricultural Engineers (ASAE) error when applied to their data set of heavy-oil, bitumen, and
2002 Annual Meeting, Chicago, Illinois, USA, 28–31 July. solvent binary mixtures. Provisions were mentioned for updating
UOP375-86 Calculation of UOP Characterization Factor and Estimation this constant for specific applications; however, the original 0.01
of Molecular Weight of Petroleum Oils. 1986. West Conshohocken, value was used for the evaluation presented in this document.
Pennsylvania: ASTM. Several methods use a viscosity blend index (VBI). These meth-
Wallace, D. and Henry, D. 1987. A Correlation for Correcting the Viscos- ods include correlations by Cragoe, Refutas, Maxwell, Chevron 1,
ity of Solvent-Extracted Bitumen to a Solvent-Free Basis. AOSTRA J. Chevron 2, Wallace and Henry, and Parkash. A comparison of these
Res. 3 (1987): 239–249. methods is provided in Figs. A-1 and A-2. The VBI is calculated
Whitson, C.H. 1983. Characterizing Hydrocarbon Plus Fractions. SPE J. 23 from the component viscosity and then averaged according to the
(4): 683–693. SPE-12233-PA. http://dx.doi.org/10.2118/12233-PA. respective method to determine the mixture VBI. The equation
Wu, J.W., Nhaesi, A.H., and Asfour, A.A. 1999. Viscometric Properties of used for the component VBI is then rearranged to solve for viscos-
Eight Binary Liquid n-Alkane Systems at 308.15 and 313.15 K. Fluid ity using the mixture VBI.
Phase Equilibria 164 (2): 285–293. http://dx.doi.org/10.1016/S0378- The McAllister method determines blend viscosity using vis-
3812(99)00250-2. cosity measured in Saybolt Seconds Universal (SSU). ASTM
Wu, J.W., Shan, Z., and Asfour, A.A. 1998. Viscometric Properties of D2161-05 (2005) provides details for converting between centi-
Multicomponent Liquid n-Alkane Systems. Fluid Phase Equilibria 143 stokes and SSU, as provided in
(1–2): 263–274. http://dx.doi.org/10.1016/S0378-3812(97)00269-0.
Yuan, W., Hansen, A.C., Zhang, Q., and Tan, Z. 2005. Temperature-Depen-  SSU = ⎡⎣1.0 + 0.000061( T − 100 ) ⎤⎦
dent Kinematic Viscosity of Selected Biodiesel Fuels and Blends with
Diesel Fuel. Journal of the American Oil Chemists’ Society (JAOCS) ⎡ ⎤
82 (3): No. J10974, 195–199. ⎢ ⎥
⎢ 1.0 + 0.03264 ⎥
× ⎢ 4.6324 + ⎥. . . . . (A-2)
⎛ 3930.2 + 262.7 ⎞
Appendix A—Blend-Viscosity Methods ⎢ ⎜ +23.97 2 + 1.646 3 ⎟ × 10 ⎥⎥
−5

⎢⎣ ⎝ ⎠ ⎦
Methods commonly used throughout the industry to evaluate the
resulting viscosity of a blended mixture are summarized in Table
A-1. The methods included in this table are limited to correlations The minimum defined value for SSU viscosity is 32 seconds,
that are predictive of blend viscosity without the need for precorre- which corresponds to a kinematic viscosity of 1.77 cSt. Attempts
lated constants or for techniques requiring viscosity measurements to extrapolate to lower values of viscosity result in increased error
at multiple temperatures. in blend-viscosity calculations. Therefore, this method is limited
Depending on the method, the blend viscosity is determined to component kinematic viscosities greater than 1.77 cSt. Table 6
using component absolute (dynamic) viscosity or kinematic vis- compares the methods to a common data set of component viscosi-
cosity. The density of the mixture at the temperature at which ties suitable for use with the McAllister method.
the viscosities were measured is necessary to convert from
kinematic to absolute viscosity (Eq. A-1). If blend viscosity is Appendix B—Bergman and Sutton
required at a different temperature, then viscosity measurements Viscosity Method
for each component are required at that temperature. Changes The density of the oil at 60°F is determined from the API gravity
in density with temperature can be calculated accurately using using Eq. B-1. The constant, 0.999012, converts the specific grav-
Eq. B-2. ity to density. It is normally neglected as it is essentially unity,
but is included in this equation because results for heavy oils are
m =  m m. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-1) affected.

February 2012 SPE Production & Operations 121

 Original
Method Publication Equations

122 February 2012 SPE Production & Operations

 TABLE A-2—RESULTS OF DIFFERENT AVERAGING TECHNIQUES WITH
ARRHENIUS-FORM EQUATION

Method # Pts % AE Std Dev % AAE Std Dev

Volume fraction weighting 2,022 218.6 4,123.2 219.9 4,123.2

Mole fraction weighting 2,022 –12.3 26.5 19.6 21.7
Weight fraction weighting 2,022 634.0 15,919.9 634.8 15,919.9

141.5 −1
o60 = 0.999012 . . . . . . . . . . . . . . . . . . . . . . . (B-1) ⎛ 0.533272 + 1.91017 × 10 −4 Tb ⎞
131.5 +  API ⎜ ⎟
+7.79681 × 10 −8 Tb2
Tco = Tb ⎜⎜ ⎟ , . . . . . . . . . . . (B-5a)
⎟
The thermal expansion coefficient for generalized crude oils with −2.84376 × 10 −11 Tb3
a base temperature of 60°F is calculated using Eq. B-2a. This is ⎜ ⎟
⎜⎝ +9.59468 × 10 T
27 − 13 ⎟⎠
an update to the traditional ASTM D1250-80 method. b

60 =
( 2.5042 × 10 −4
+ 8.302 × 10 −5 o60 ). . . . . . . . . . . . . (B-2a)  = 1−
Tb
, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-5b)
o260 Tco

The volume correction factor (VCF) resulting from the change in

temperature from 60°F is given by Eq. B-2b: ln ( 2 − 0.152995) = 2.40219 − 9.59688
+3.45656 2 − 143.632 4 , . . . . . . . (B-5c)
VCF = e{ 60
−  ( T − 60 ) ⎡⎣1+ 0.860 ( T − 60 ) ⎤⎦}
. . . . . . . . . . . . . . . . . . . . . (B-2b)

ln (1) = 0.701254 + 1.38359 ln ( v 2 ) + 0.103604 ⎡⎣ ln ( 2 ) ⎤⎦ ,

2
The resulting density as affected by a temperature change can be
determined from the density at 60°F and the VCF:
. . . . . . . . . . . . . . . . . . . . . (B-5d)
oT = o60 VCF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-2c)
 oo = 0.843593 − 0.128624 − 3.36159 3 − 13749.512 ,
Whitson (1983) presented the following equation relating molecu-
lar weight and specific gravity to the Watson characterization
factor: . . . . . . . . . . . . . . . . . . . . . (B-5e)

K w = 4.5579 M o0.15178  o−0.84573. . . . . . . . . . . . . . . . . . . . . . . . (B-3)  o =  o −  oo , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-5f)

Bergman and Sutton found the resulting error in this relationship

to be excessive for low-gravity, high-molecular-weight oils and x = 2.68316 − 62.0863 Tb0.5 , . . . . . . . . . . . . . . . . . . . . (B-5g)
recommended the following relationship:

⎡ ⎛ b2 M o + b3 o ⎞ b5 b6 ⎤ b8
b7
f2 = x  o − 47.6033 o2 Tb0.5 , . . . . . . . . . . . . . . . . . . . (B-5h)
K w = ⎢ b1 exp ⎜ M o  o ⎥  o , . . . . . . . . . . . . (B-4)
⎣ ⎝ + b4 M o o ⎟⎠ ⎦ 2
⎛ 1 + 2 f2 ⎞
where
( ) (
ln  210 + 232.442 Tb = ln  2 + 232.442 Tb ⎜
⎝ 1 − 2 f2 ⎟⎠
, )
b1 = 2012.84 b5 = 0.589485 . . . . . . . . . . . . . . . . . . . . . (B-5i)
b2 = −1.8519×10−3 b6 = 3.36211
b3 = −3.70833 b7 = 0.3333
b4 = 1.31441×10−3 b8 = −1.0 . f1 = 0.980633 x  o − 47.6033 o2 Tb0.5 , . . . . . . . . . . . (B-5j)

VBI Method Comparison VBI Method Comparison

1000 1000

100
100
Cragoe
Blend Index

Blend Index

10 Wallace and Henry

10
1

1
0.1 Maxwell
Refutas
0.01 Chevron 1 0.1
Chevron 2
Parkash
0.001 0.01
0.1 1 10 100 1000 10000 100000 1000000 0.1 1 10 100 1000 10000 100000 1000000
Kinematic Viscosity, cSt Dynamic Viscosity, cp

Fig. A-1—VBI methods that use kinematic viscosity. Fig. A-2—VBI methods that use absolute viscosity.

February 2012 SPE Production & Operations 123

 ( ) (
ln 100 + 232.442 Tb = ln 1 + 232.442 Tb ⎜
⎛ 1 + 2 f1 ⎞
) ,
2

B=
( ) ( )
ln ⎡⎣ ln od210 + 1 ⎤⎦ − ln ⎡⎣ ln od100 + 1 ⎤⎦
⎝ 1 − 2 f1 ⎟⎠
. . . . . . . . . . . . (B-6b)
ln ( 5200 ) − ln ( 410 )
. . . . . . . . . . . . . . . . . . . . . (B-5k)
The viscosity at any temperature, T, can then be determined:

o100 = 0.999012 o VCF100, . . . . . . . . . . . . . . . . . . . . . . . . (B-5l) ⎛ ⎧ln ⎡ ln 

⎪
od = exp ⎜ exp ⎨ ⎣
( ⎤
od100 + 1 ⎦ ) ⎫⎞
⎪
⎬⎟ − 1. . . . . (B-6c)
⎜⎝ +
⎩⎪ ⎣B ⎡ ln ( T + 310 ) − ln ( 41
1 0 ) ⎦ ⎭⎪⎟⎠
⎤
od100 = 100 o100 , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-5m)

Rob Sutton is a senior technical consultant for Marathon Oil

o210 = 0.999012 o VCF210 , . . . . . . . . . . . . . . . . . . . . . . . . (B-5n) Company in Houston where he works in the Reservoir Performance
Group under the Upstream Technology organization. He holds a
BS degree in petroleum engineering from Marietta College and
od210 =  210 o210 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-5o) an MS degree in petroleum engineering from the University of
Louisiana at Lafayette.
Bergman’s method is defined by Dave Bergman holds a PhD degree in chemical engineering
from joined The University of Michigan. He joined Amoco as
ln ⎡⎣ ln ( od + 1) ⎤⎦ = A + B ln ( T + 310 ). . . . . . . . . . . . . . . . . (B-6a) a research engineer in 1976. He then worked for BP since the
merger in 1999 until he retired in 2008. During that time, he was
active in laboratory analyses and procedures and equation of
The slope, B, is determined from known viscosity at the two tem- state modeling of reservoir fluids. In his retirement, he is a consul-
peratures, 100 and 210°F: tant for fluid properties and other PVT areas.

124 February 2012 SPE Production & Operations