Accepted Manuscript

Measurements and predictions of Middle Eastern heavy crude oil viscosity using
compositional data

Abdulrahim Kamel, Osamah Alomair, Adel Elsharkawy

PII: S0920-4105(18)30930-6
DOI: https://doi.org/10.1016/j.petrol.2018.10.062
Reference: PETROL 5425

To appear in: Journal of Petroleum Science and Engineering

Received Date: 18 May 2018

Revised Date: 23 September 2018
Accepted Date: 18 October 2018

Please cite this article as: Kamel, A., Alomair, O., Elsharkawy, A., Measurements and predictions of
Middle Eastern heavy crude oil viscosity using compositional data, Journal of Petroleum Science and
Engineering (2018), doi: https://doi.org/10.1016/j.petrol.2018.10.062.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Measurements and Predictions of Middle Eastern Heavy Crude Oil Viscosity

Using Compositional Data

Abdulrahim Kamel, Osamah Alomair, and Adel Elsharkawy

PT
Petroleum Engineering Department, College of Engineering & Petroleum, Kuwait University P.O. Box
5969, Safat 13060, Kuwait.

RI
SC
ABSTRACT
Determining the viscosity of heavy crude oil is important in the field of petroleum science, and reservoir

U
simulators are routinely used to achieve this goal. In this study, a new empirical model for predicting heavy-oil
AN
viscosity is presented. The model was developed using the compositional data of 28 heavy crude oil samples from
the Middle East. Viscosity measurements from 196 data points in the 20–80 °C temperature range were used during
model development, and another 47 viscosity measurements were used to validate the model. The compositional and
M

viscosity measurements were also used to evaluate the accuracy of previously published methods for predicting
heavy-oil viscosities. These methods include corresponding-states models, residual viscosity models, and equations-
of-state-based models. The study reveals that none of the previously published methods adequately predict the
D

viscosities of heavy crude oils. On the other hand, it shows that the new empirical model is able to predict the
TE

viscosities of heavy oils with an average relative error of 3.8%. The proposed model is much simpler and easier to
use than existing methods.
EP

KEYWORDS: Heavy Oil, Viscosity, Correlation, Corresponding State, Equation of State

C
AC

CORRESPONDING AUTHOR

Osamah Alomair

Petroleum Engineering Department, College of Engineering & Petroleum, Kuwait University

P.O. Box 5969, Safat, 13060, Kuwait

Tel.: +965 24987456

1
 ACCEPTED MANUSCRIPT

Email: dr-alomair@hotmail.com

1. INTRODUCTION

Reservoir simulators are frequently used to predict and optimize oil recovery from oil fields. These

PT
simulators require the reservoir fluids properties as functions of pressure P, temperature T, and composition. The
precision of these data can seriously affect the simulation results. The required fluid properties include densities ρ,
viscosities µ, dissolved gas-oil ratios, and formation volume factors. These properties are usually measured at

RI
reservoir temperature; however, in many cases measured data are unavailable. In such cases, crude oil viscosity
models are used to estimate viscosities available at reservoir temperatures as well as other temperatures during the

SC
design of tubing and pipelines. Typically, before a certain model is developed for the prediction of crude-oil
viscosity for a particular zone or a region, the accuracies of previously published models should be evaluated. If
none of the published models reasonably predict the viscosities for that region, the experimentally measured crude-

U
oil viscosity data are used to develop a new local empirical model.
AN
1.1. Compositional Based Viscosity Models
M

Compositional based methods for predicting oil viscosity are classified into three categories that are based on:

(i) Empirical residual concepts, as used by Lohrenz, Bray, and Clark (LBC) (Lohrenz et al. 1964) and Al-Syabi et
D

al. (2001).
TE

(ii) The corresponding-states (CSs) principle with one or two reference fluids, as developed by Pedersen and
Fredenslund, (Pedersen et al. 1987), Aasberg-Petersen et al. (1991), and Lindeloff et al. (2004).
EP

(iii) Cubic equations of state (EOSs) used by Guo et al. (2001) that are based on the Peng and Robinson EOS (Peng
et al. 1976) and Little and Kennedy (1968).
C

Table 1 summarizes the models that have previously been used to calculate heavy-oil viscosities along with
their respective limitations.
AC

2
 ACCEPTED MANUSCRIPT

Table 1. Summary of composition-based viscosity models.

Group Model Year Range Note

Light to moderate
Based on residual viscosity
concept and corresponding

Used 260 oil samples (black and highly

LBC 1964 0.4812–0.903g/cc (Pr: volatile).
0.02–3)

PT
Light to moderate Used 2789 data points and accounted for
state theory

(0.01–15 cP)
Al-Syabi 2001 thermal and structural effects (molecular

RI
(Pr: 0.004–4) weights and temperatures).

Seven oils from the North Sea (162 data

SC
Pedersen 1984 0.2–5 cP points). Methane was used as the reference
fluid.
Based on corresponding state theory

Six oil samples from the North Sea (34 data

U
Aasberg- points). Two references, methane and n-
1991 0.3–1.23 cP
Petersen decane. The author mentions that Pedersen
AN
(1984) is more accurate.

Eight oil samples from the North Sea (129

data points). Methane was not used as a
M

Lindeloff 2004 3.2–8500 cP

reference fluid when the methane reference
temperature was below 65 K.
D

Little and 828 oil samples (3349 data points). Applied

Kennedy 1968 0.071–58.52 cP to both liquids and gases. Similar form to
TE

(LK) the van der Waals EOS.

Based on EOS

22 oil samples (126 data points). Cubic

Guo (EOS)-
2001 <5 cP viscosity equations of state were found to be
MPR
EP

inapplicable to heavy petroleum fluids.

C
AC

1.2. Models Based on Empirical Residual (ER) Concepts

This group of models, based on ER concepts, consists of the most commonly used correlations for
simulating compositional reservoirs in petroleum engineering. These correlations are obtained by subtracting the
diluted gas viscosity µ* from the total viscosity of the fluid μ and are extremely sensitive to fluid density, especially
for highly viscous fluids. The empirical LBC is based on the residual viscosity concept of Jossi et al. (1962) and was
developed by Lohrenz et al. (1964) using 260 samples of gases and light oil mixtures. Many modifications and
evaluations have been carried out to improve the predictabilities of these models for highly viscous fluids. The

3
 ACCEPTED MANUSCRIPT

model of Al-Syabi et al. (2001) is an example of one such modification that was developed using 2789 data points
for samples with reduced densities greater than 2.5. In order to improve its fluid viscosity prediction, this correlation
includes reduced temperature to account for thermal effects, and molecular weight to account for structural effects.

1.3. Models Based on Corresponding States (CSs)

PT
Models based on corresponding states are the most commonly used methods for predicting the viscosities
of fluids (Zéberg-Mikkelsen 2001). The methods are based on one or more reference fluids and rely on the principle
that under the same reduced conditions, the same reduced viscosity is obtained for any of the fluids in a group. This

RI
method predicts viscosities as functions of temperature, pressure, composition, pseudo-critical properties, and the
viscosity of a reference fluid at a reference temperature and pressure. As such, it can only be related to two of the

SC
following properties: the reduced temperature Tr, reduced pressure Pr, reduced density ρr, or reduced volume vr.
Pedersen et al. (1984) introduced a method for predicting the viscosities of oil and gas as reduced viscosities, which
are functions of reduced pressure and temperature, using one reference fluid. Methane was chosen as the reference

U
fluid because it is one of the most studied fluids in terms of its viscosity and density. The structure of this model is
similar to that of Ely and Hanley (Ely et al. 1981) that requires an acentric factor and the critical compressibility of
AN
the reference fluid and each component. However, Pedersen et al. (1984) introduced a new parameter, namely the
rotational coupling coefficient α, which is related to the effect of molecular size and density.
M

One of the major difficulties associated with using methane as a reference is that it will freeze if the
reduced temperature is below 0.4 (T = 76 K), which is above the freezing point of many hydrocarbons. For this
reason, Pedersen et al. (1987) modified the viscosity model by adding a fourth viscosity term that ensures
D

calculational continuity of the reference viscosity above and below the freezing point of methane (Tf = 95 K).
TE

However, this model gives inaccurate predictions when it is applied to crude oils with higher molecular-weight
components than the reference fluid (Raut et al. 2008). To overcome this limitation, Aasberg-Petersen et al. (1991)
proposed a different model with two reference components (methane and n-decane) based on the corresponding
EP

states method of Teja and Rice (1981a) but with molecular weight included as the interpolation parameter instead of
an acentric factor. This change was introduced because the acentric factor is sensitive to the nature of the heavy fluid
fraction; in particular, as the molecular weight increases, the acentric factor decreases. As the density of methane
C

calculated by the method proposed by McCarty (1974) is not valid at temperatures below 60 K, Rønningsen (1993)
proposed a semi-empirical correlation for stabilized crude oil viscosities that allows the viscosity to be used as a
AC

starting point. This correlation applies at atmospheric pressure if there is no wax precipitation, and is valid at
temperatures above 30 °C (303 K), conditions under which the fluid exhibits Newtonian behavior. Lindeloff et al.
(2004) modified the Rønningsen equation by classifying crude oils into stabilized oils and live oils and then
extended it to heavy-oil applications.

1.4. Models Based on Equations of State (EOSs)

4
 ACCEPTED MANUSCRIPT

The EOS concept involves a mathematical relationship between volume, pressure, and temperature that
includes compositional information when dealing with mixtures. It can be used for different states of matter but is
mostly used for gases, and no EOS application has been found for petroleum fluids (Motahhari 2013). Little and
Kennedy (1968) (LK) developed the first cubic equation of state for predicting viscosity which is an empirical
correlation for predicting the viscosities of hydrocarbons as functions of pressure, temperature, and composition.
This correlation can be applied to both liquids and gases and is based on the van der Waals (VDW) EOS (Van der

PT
Waals 1873). Guo et al. (1997) proposed two viscosity models based on the Peng and Robinson (1976) equation of
state (PR-EOS) and the Patel and Teja (1982) equation of state (PT-EOS), but poor results were obtained for

RI
complex fluids. A modified PR-EOS (MPR) was developed by Guo et al. (2001), which changed the mixing rules
and functional forms of the pure-compound parameters, (Motahhari 2013).

SC
In this paper, we introduce a new, simple, and easy-to-implement model. This model is a function of easily
measured parameters, such as composition, density, molecular weight, and temperature. The density of the fluid is
used as an input parameter because viscosity predictions for heavy oils were found to be very sensitive to the

U
estimated density. This model can be incorporated into compositional reservoir simulations to predict heavy-oil
viscosity instead of EOS-based models that use EOS-calculated densities as input parameters.
AN
M

2. MATERIALS AND METHODS

2.1. Physical property measurements

D
TE

Twenty-eight samples of dead oils from several reservoirs were collected from various wells. These
samples were stored under dry conditions in sealed dark 2.5 L bottles. All samples were stored in thermostatically
controlled cabinets at 15 °C (288 K). The samples were mixed with an emulsion breaker immediately after the
EP

bottles were opened, to remove traces of basic sediments and water cut according to ASTM D4007.

The density and API gravity of each oil sample were measured with an Anton-Paar DMA 4500M density
meter according to ASTM D5002; the instrument was composed of one oscillating U-tube designed for measuring
C

density based on the relationship between density and oscillation period. Temperature control within the unit was
AC

reached with a built-in combined thermostat with cascaded Peltier elements and a Pt-100 thermometer. The accuracy
of the thermostat and its uncertainty are specified by the manufacturer to be 1 mK and 10 mK, respectively, with a
repetition of 0.01%. The instrument was adjusted following the manufacturer's procedures over the entire
temperature range using at least two known standards, namely dry air and freshly degassed ultra-pure water. The
instrument was calibrated to validate density-measurement accuracy at the temperature of interest using dry air and
certified ultrapure water supplied by a manufacturer (S. No. 78169, S.H. Kalibrier, GmbH products), at temperatures
ranging from 20 to 80 °C (293 to 353 K). For each sample, density measurements were performed in triplicate, and
the results were averaged. Before and after any measurement, it is important that the measurement cells are

5
 ACCEPTED MANUSCRIPT

systematically cleaned and dried to avoid sample contamination, after which a single standard check was performed
prior to each series of measurements in order to verify that the measurement cells are clean and to validate that the
adjustments are within the relative deviation of ±0.0001 g/cc. The precision value of the instrument was ±5×10-5
g/cc, and the uncertainty of the measured density was within ±6×10-5 g/cc.

The dynamic viscosity (μ) of each crude oil sample was determined at temperatures ranging from 20 to 80

PT
°C (293 to 353 K) using an Anton Paar Stabinger SVM 3000 viscometer, which is a cylindrical rotational
viscometer that operates on the basis of a modified Couette principle, with a swiftly rotating outer tube and an inner
measuring bob. The instrument can be used in the 0–105 °C temperature range and has a built-in thermoelectric

RI
thermostat system; the uncertainty in the measured temperature stated by the manufacturer is 20 mK. The instrument
precision complies with ASTM D7042. The measuring ranges of the entire instrument were set automatically with

SC
the assistance of a built-in program and four certified standard samples supplied by the manufacturer (Canon
Instruments Co.) that were used for instrument calibration in triplicate in the 20–80 °C temperature range. The
precision value of the instrument was ± 0.1% and the estimated uncertainty in the measured sample viscosities was ±

U
0.12%.
AN
In order to ensure that the measured samples meet the required quality criteria, effective quality-control
procedures were implemented with an uncertainty assessment approach similar to that described by Laesecke et al.
(2012). The 28 samples were grouped into four sets, with each set containing seven samples for density and
M

viscosity measurements at elevated temperatures. Frequently, at the end of each set, one sample was re-measured for
verification purposes. In addition, a single standard check was performed prior to each set of measurements in order
D

to ensure that the sample measurements remain valid. The repeatability check was performed with a certified
standard at elevated temperatures. A minimum of three sequential measurements were taken at each temperature; the
TE

data were averaged and standard deviations calculated. Density and viscosity measurements matched the
manufacturer’s specifications of 0.001% and 0.15%, respectively, to within confidence intervals (CIs) of 95%. On
the other hand, reproducibility at 95% CI determined from deviations between repeated measurements was 0.0039%
EP

for density, and 1.98% for viscosity. To determine whether or not the instrument was performing to within
specifications, estimates of combined reference and instrument uncertainties are recommended. The average
experimental errors for density and viscosity measurements were 0.005% and 0.095%, respectively. The estimated
C

absolute average deviations were approximately 0.00006 (g/cc) for density and 0.43 cP for viscosity measurements,
AC

with an average tolerance zone of ±0.00006 g/cc and ±0.56 cP. Using a coverage factor of two (95% CI), the results
revealed that all deviations are within the combined and expanded uncertainties of 0.00004 g/cc and 0.00008 g/cc
for density, and 0.34% and 0.68% for viscosity.

The molecular weight of the entire sample was measured using a wide-range cryette, which uses the
principle of freezing-point depression (cryoscopy), and was calibrated with benzene and n-C9. Directly following
calibration, the cryette was checked using n-C14, with a known molecular weight of 198.40 g/mol, for final
adjustment.

6
 ACCEPTED MANUSCRIPT

A Shimadzu GC-2014 chromatograph and the LabSolutions distillation GC analysis software were used for
analyses. This system is compliant with ASTM D5307, and combines operational comfort with high-level
functionality. Standard solutions were used to calibrate the system. Two standard solutions were analyzed, one a
mixture of n-C3 to n-C9, and the other a complete mixture of hydrocarbons from n-C10 to n-C44. The column
temperature ranged from -30 to 350 °C, and a carrier gas flow rate of 40 mL/min (nitrogen), an injection port
temperature of 350 °C, and an FID temperature of 350 °C were used. Each component was quantified according to

PT
the recovery factor intrinsic to each identified component.

The C1 to C6 factions were examined with a front-end DHA (detailed hydrocarbon analyzer) using a 50-

RI
meter PONA column with specific pressure-temperature programs; the recovery was calculated using dedicated
Carburane software. 2,2,4-Trimethylpentane (2 wt%), as an internal standard, was spiked into the sample and the

SC
recovery of each fraction was calculated based on true boiling point. The mole fraction of each component was
calculated from recovered wt%, molecular weight, and density, where the molecular-weight fraction of each
component was calculated using a defined database. The C7+ fraction was calculated by mathematical regression,

U
where the overall measured molecular weight minus the sum of all molecular-weight fractions tended to zero.
AN
The analysis results from Data Set 1 are provided in Table 2 for researchers interested in calibrating their
models.
M

2.2. Sensitivity Analyses

Sensitivity analyses were performed in order to understand the relationship between input variables and
D

viscosity; %C4–C7+, molecular weight of C7+ components MWc7+, specific gravity of C7+ components SGc7+,
temperature, API, and density were all found to be related to the viscosity of the oil (See Fig A-1 through A-6 in
TE

Appendix A). Fig. 1 displays the correlation coefficient of the relationship between each parameter and viscosity,
which visually shows the statistical strength of each relationship. This figure reveals that viscosity is directly related
to the density of the oil, the C7+ percentage, the molecular weight, and the C7+ specific gravity, and it is inversely
EP

related to temperature, C4–C6 compositions, and API. In addition, density, temperature, and API are most correlated
with viscosity; consequently, they are the main parameters to be considered in the development of any new equation.
C
AC

7
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M

Fig. 1. Correlation coefficients for the relationships between viscosity and various parameters.
D
TE
C EP
AC

8
 ACCEPTED MANUSCRIPT

Table 2. Data Set 1 (sample compositions Ci, densities, and viscosities).

API at ρ (g/cc) µ (cP) at µ (cP) at µ (cP) at µ (cP) at µ (cP) at µ (cP) at µ (cP) at

No. %C4 %C5 %C6 %C7+ MWC7+ SGC7+
15 °C at 20 °C 20 °C 30 °C 40 °C 50 °C 60 °C 70 °C 80 °C
1 20.4 1.02 2.57 3.37 93.04 326.40 0.94 0.930 200.59 106.07 62.6 40.177 27.268 19.367 14.258

PT
2 20.0 0.04 0.44 1.62 97.90 331.22 0.93 0.933 243.11 131.9 76.875 48.461 32.351 22.637 16.5
3 19.6 0.25 1.57 3.94 94.24 343.10 0.94 0.935 322.46 171.13 96.462 58.975 38.663 26.759 19.225
4 19.5 0.00 0.12 1.16 98.72 295.70 0.94 0.935 312.40 158.25 88.722 54.503 35.498 24.334 17.456

RI
5 18.7 0.08 1.13 2.09 96.70 342.65 0.94 0.941 232.12 126.17 74.171 46.59 30.978 21.568 15.598
6 18.6 0.35 1.62 3.21 94.81 338.41 0.95 0.941 196.73 108.36 64.539 41.088 27.617 19.408 14.192

SC
7 18.0 0.46 1.23 2.06 96.25 353.31 0.95 0.945 319.57 262.18 140.28 81.78 51.562 34.484 24.369
8 18.0 0.35 1.14 1.95 96.56 353.44 0.95 0.945 285.66 151.65 87.573 54.331 35.767 24.71 17.801

U
9 18.0 1.05 1.53 1.92 95.50 347.26 0.95 0.945 396.52 171.08 97.209 59.809 39.16 26.93 19.336
10 17.9 0.15 1.65 3.72 94.49 357.24 0.95 0.945 516.01 205.44 115.72 70.104 45.172 30.612 21.64

AN
11 17.6 0.18 1.16 2.73 95.93 343.69 0.95 0.947 309.50 162.45 92.869 56.966 37.234 25.471 18.215
12 17.4 0.46 0.49 1.95 97.10 367.35 0.95 0.949 494.24 249.43 137.83 81.972 52.014 34.807 23.512

M
13 17.1 0.00 0.23 1.19 98.58 345.72 0.95 0.951 891.90 431.67 209.77 118.36 72.911 47.536 32.358
14 16.8 0.17 0.73 2.23 96.87 354.55 0.96 0.952 374.80 192.84 108.31 65.532 42.155 28.601 20.13
15 16.7 0.16 0.96 1.87 97.00 351.13 0.96 0.953 382.14 197.41 110.22 66.462 42.623 28.759 20.242

D
16 16.4 0.09 0.43 1.21 98.27 372.04 0.96 0.955 521.79 259.03 141.78 83.945 53.008 35.422 24.717

TE
17 16.2 0.00 0.09 0.40 99.51 380.46 0.96 0.956 1320.2 607.26 291.99 159.21 95.079 60.171 40.503
18 16.0 0.09 1.25 1.42 97.23 378.29 0.96 0.958 552.91 272.47 147.95 86.964 54.608 36.261 25.209
19 15.9 0.09 0.34 1.04 98.54 367.17 0.96 0.958 535.47 264.89 143.88 85.346 53.702 35.705 24.913
EP
20 15.8 0.18 0.68 2.00 97.14 335.27 0.96 0.959 541.02 266.69 144.7 84.987 53.336 35.294 24.511
21 14.9 0.15 0.32 1.37 98.16 369.29 0.97 0.965 788.58 375.15 196.19 111.31 68.012 43.975 29.95
C

22 14.9 0.09 0.39 1.21 98.31 377.01 0.97 0.965 767.13 363.2 191.05 108.93 66.528 46.093 29.308
23 14.8 1.01 1.95 2.62 94.42 412.79 0.97 0.966 1294.6 420.23 217.33 123.11 74.758 48.121 32.683
AC

24 14.8 0.69 1.32 1.91 96.09 361.14 0.97 0.965 895.64 591.89 299.11 165.25 98.236 61.867 41.218
25 14.8 0.11 0.21 0.96 98.72 364.62 0.97 0.966 746.11 353.31 185.22 105.68 64.809 42.081 28.698
26 14.5 0.13 0.25 1.05 98.58 362.00 0.97 0.968 1022.4 430.23 221.74 124.31 74.883 47.925 32.261
27 14.3 0.21 0.18 1.45 98.16 355.03 0.97 0.969 930.79 466.73 238.63 132.32 79.264 50.546 33.914
28 13.4 0.05 0.05 0.39 99.52 384.46 0.98 0.975 1430.6 632.54 312.66 170.03 99.442 62.18 41.074

9
 ACCEPTED MANUSCRIPT

3. RESULTS AND DISCUSSION

3.1. Evaluation of Existing Viscosity Models

Compositional-based models for predicting the viscosities of heavy crude oils were evaluated using the
Middle Eastern oil-field viscosity data provided in Data Set 1. The resulting database contains 196 viscosity data

PT
points covering a wide range of temperatures. Using these data, most of the published viscosity models failed to
accurately predict the viscosities of the various samples. Given that significant errors were observed, the performance

RI
of the published models is discussed in the following paragraph.

All models were coded in Excel VBA and worked perfectly, except for the model of Pedersen et al. (1987), which

SC
required the use of the PVTi simulation software. The LBC model (1964) and that of Aasberg-Petersen et al. (1991)
were both executed using the PVTi simulator to crosscheck the results of the coded calculations; these models use
EOS-calculated densities as input parameters. We found that the errors in the EOS-calculated densities impacted

U
greatly on the predicted viscosities. Therefore, we used measured densities as input parameters for the LBC (1964)
and Aasberg-Petersen et al. (1991) models in order to improve their predictiveness for heavy oil, instead of
AN
performing density matching within the PVTi simulator using constant-volume shifts or temperature dependencies.

As mentioned previously, the viscosities calculated using the LBC and Al Syabi models are very sensitive to density
M

as they have the same equation structure. The results greatly improve when the experimentally measured densities are
used instead of the predicted densities. Moreover, both models are very sensitive to any conversion factors used in the
equation for the C7+ critical-volume parameters. Therefore, error analyses for the compositional-based published
D

models using the current data were performed. Table 3 provides a summary of the error analysis for each model
TE

including the maximum, minimum, absolute average relative error (AARE), and standard deviation (SD). The overall
AARE was found to be about 70% for the LBC model, with an associated SD of 21%. This error is ascribed to the
fact that the LBC model is parameterized using light compounds. The minimum error (22%) occurs at a temperature
EP

of 80 °C (353 K), as the sample viscosity is less than 10 cP at higher temperatures. Al Syabi et al. (2001) model on
the other hand, overestimates viscosities with the highest error percentage. None of the EOS correlations (MPR and
LK) is considered applicable for heavy oils, as both show AAREs close to 100%. In contrast, the CS model of
C

Lindeloff et al. (2004) (hereafter referred to as the “Lindeloff model”) has the lowest AARE of around 49%, followed
by that of Aasberg-Petersen et al. (1991) with an AARE of 59%. The earlier model of Pedersen et al. (1987) over-
AC

predicted viscosities, resulting in a high AARE.

10
 ACCEPTED MANUSCRIPT

Table 3. Summarizing the errors in the compositional-based modeling of Data Set 1.

Aasberg-
Error% LBC Al Syabi Pedersen Lindeloff MPR LK
Petersen

PT
Max 97 1158 923 93 86 99 100

Min 22 3 2 10 20 66 93

RI
AARE 70 339 158 59 49 88 99

SD 21 250 137 20 16 8 1

U SC
3.2. Modifications of Existing Models and the Development of New Viscosity Models
AN
The previous evaluation identified the need for a more accurate viscosity model appropriate to heavy oils
over a wide range of temperatures and compositions. This can be achieved either by modifying an existing model or
M

by developing a new viscosity model. In either event, the ability to accurately predict viscosities is important,
especially when laboratory analysis is unavailable or difficult to perform due to sample unreliability.
D

3.3. Modifying Existing Compositional Models

TE

The coefficients used in the LBC (1964) model and the Lindeloff (2004) correlation were adjusted using the
experimental viscosity and density data in the 20–80 °C (293–325 K) temperature range. These two correlations were
EP

selected for modification because of their direct methods of calculation and simplicities. Excel Solver was used to
adjust the coefficients and provide minimum error, as shown in the modified LBC and Lindeloff equations,
respectively:
C

1
( µ − µ * ) ξ + 10−4  4 = 0.07835 + 0.01982 ρr + 0.00734 ρr2 + 0.06839 ρr3 + 0.02056 ρr4
AC

(1)
 

349.35874 7.22258 Mw
log 10 ( µ ) = −0.0138642 (Mw ) − + (2)
T (K ) T (K )

Where ξ is the viscosity reducing parameter, µ* is the low-pressure gas mixture viscosity (dilute gas viscosity), and
Mw is the molecular weight.

11
 ACCEPTED MANUSCRIPT

Error analyses indicate that the percentage errors of both models were reduced by more than 30%; the AARE for the
LBC correlation was reduced from 70 to 47% (with an SD of 26.7%), while that for the Lindeloff correlation was
reduced from 49.1 to 20.7% (with an SD of 15.77%). Although the error of the modified LBC was reduced, the
correlations shown in Fig. 2 reveal large deviations between experimental and calculated viscosities. In the modified
Lindeloff model most of the viscosity calculations begin to deviate when the viscosity exceeds 200 cP; for lower
viscosities the correlation coefficient between the experimental and calculated values is 92.22%.

PT
RI
U SC
AN
M
D
TE
C EP
AC

12
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
C EP
AC

Fig. 2. (a) Calculated vs. experimental viscosities using the original and modified LBC and Lindeloff equations; (b) log-log plot
of the data in (a).

13
 ACCEPTED MANUSCRIPT

3.4. Development of a New Compositional-based Viscosity Model

Modifying the coefficients used in the LBC model (1964), the model of Aasberg-Petersen et al., or any other model,
is a standard method used in the PVTi software to match experimental viscosity data and create a representative PVT

PT
model for a given fluid in a reservoir simulation. However, such a tuning process creates a model that can only be
used for a specific fluid. In this study all viscosity data were used to modify the coefficients in the original LBC
model, which resulted in a reduction in the AAE from 70 to 47%, while modifying the coefficients in the Lindeloff

RI
model resulted in a reduction in the AAE from 49 to 20%. Despite this, the modified models exhibited significant
scatter and underestimated the viscosities of heavy oils above 200 cP (Figs. 2a and 2b). Therefore, numerous

SC
functional forms were examined using the XLSTAT and SPSS software packages based on sensitivity analyses
discussed previously in section 2.2; this process eventually led to equation (3), which had the lowest AARE and SD,
and the highest correlation coefficient. The optimum value for each parameter was determined by multiple nonlinear

U
regression analyses of the experimental viscosity data. In this manner, an equation for the viscosity of heavy crude oil
was developed.
AN
The current empirical model is a function of the composition, density, temperature, and molecular weight of
the crude oil. In this equation, C4, C5, and C6 are combined as an intermediate component. The density was used
M

instead of the API because it is a function of temperature and was measured experimentally at different temperatures.
The molecular weight of the mixture was introduced to the equation as a new independent variable to improve the
prediction of the viscosity. Finally, the equation contains six independent variables, as shown:
D
TE

, (3)

where
EP

= C4 + C5 + C6 , (4)

n
MW mix = ∑x i MW ci ,
C

(5)
i =1
AC

and MWci is the molecular weight of component Ci.

As shown in Figs. 3a and 3b, the predicted and experimental data are in good agreement using this new
empirical model, with a correlation coefficient of 96%. Fig. 4 reveals that the residuals are normally distributed with
low variation and are frequently distributed around 30.7 cP. Some minor deviations were observed but only for a few
samples.
14
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
C EP
AC

Fig. 3. (a) Calculated vs. experimental viscosities using the current empirical model; (b) log-log plot of the data in (a).

15
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D

Fig. 4. Residual histogram of the results using the current empirical model.
TE

3.5. Model Accuracy

EP

In this work, the best-modified compositional-based model was selected for comparison with the newly
developed model. The performance and accuracies of the models for predicting heavy-crude-oil viscosity are
C

displayed in Figs. 5 and 6. It is clear from these figures that the proposed empirical model provides data that are in
good agreement with those obtained experimentally, especially for very heavy crude-oil samples. The modified
AC

Lindeloff model, on the other hand, provides data that are in good agreement with experiment when the viscosities
are below 150 cP. The new empirical model provides the best predictions for the heavy-oil-sample viscosities
considered in this study and performs very well against the other models. Table 4 provides a summary of the errors
produced by the two proposed models and their accuracies in predicting the viscosities of the heavy crude oils
considered in this study. It is clear that the modified Lindeloff model has the highest AARE (20.73%) with an SD of
16%, and the lowest correlation coefficient (0.92). The newly developed empirical model, on the other hand, has an
AARE of 18.86%, which is lower than that of the modified Lindeloff model, with the highest correlation coefficient
16
 ACCEPTED MANUSCRIPT

(0.96); the root square mean error (RSME) of the new method is the lowest (19.84). Consequently, the current model
produces less scatter for viscosities above 300 cP.

PT
RI
U SC
AN
M
D
TE
C EP
AC

17
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
Fig. 5. (a) Calculated vs. experimental viscosities for the current and modified Lindeloff models applied to Data Set 1; (b) log-
M

log plot of the data in (a).

D
TE
C EP
AC

18
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
C EP
AC

Fig. 2. (a) Calculated vs. experimental viscosities for the current and modified Lindeloff models at viscosities below 200 cP;

(b) log-log plot of the data in (a).

19
 ACCEPTED MANUSCRIPT

Table 4. Summaries of the errors generated by the two proposed models when applied to Data Set 1.

Current Empirical Model Modified Lindeloff

ARE% 3.83 11.38

AARE% 18.86 20.73

PT
SD (for AARE) 0.13 0.16

Max% (Abs. error) 53.1 70.26

RI
Correlation coefficient 0.96 0.92

RSME 19.84 544.39

SC
Sum of residuals 277.7 7621.41

R² 0.92 0.85

U
AN
3.6. Model Validation
M

Validation is an important step in the model building process, as it confirms performance and accuracy. The
proposed models, which were developed using Data Set 1, were validated using another 47 data points from Data Set
D

2. Table 5 summarizes the ranges in compositions, temperatures, densities, and viscosities for this independent data
TE

set. The validation results and associated statistics summarized in Table 6 reveal that the new empirical model
provides the best predictions of the viscosities of heavy crude oils. The current empirical model has the lowest AARE
(12.28%) with a correlation coefficient of 0.98, whereas the modified Lindeloff model has the highest AARE. The
EP

good performance of the current empirical model with heavy oils is related to the fact that it is a function of the oil
density at different temperatures and the molecular weights of the mixture, as well as the oil composition and
temperature. The cross plot shown in Fig. 7 depicts the calculated and experimental viscosities for Data Set 2. This
C

figure clearly indicates that the newly developed empirical model exhibits superior performance to that of the
modified Lindeloff model.
AC

20
 ACCEPTED MANUSCRIPT

Table 5. Independent data from different fields (Data Set 2).

PT
API at ρ (g/cc) µ (cP) µ (cP) µ (cP) µ (cP) µ (cP) µ (cP) µ (cP)
No. C4 C5 C6 C7+ MWC7+ SGC7+ at at at at at at
15 °C at 20 °C at 20 °C

RI
30 °C 40 °C 50 °C 60 °C 70 °C 80 °C

1 19.58 0.006 0.028 0.034 0.932 379.385 0.942 0.933 274.58 146.84 85.538 53.158 34.972 24.021

SC
2 19.53 0.009 0.036 0.040 0.915 368.718 0.938 0.928 208.19 114.65 68.41 43.491 29.153

U
3 19.34 0.010 0.042 0.042 0.906 382.840 0.947 0.936 271.07 145.17 84.792 52.823 35.038

AN
4 18.67 0.007 0.034 0.032 0.927 374.896 0.948 0.939 379.76 196.12 110.99 67.072 42.829

5 18.57 0.012 0.032 0.033 0.923 385.901 0.948 0.939 372.14 193.18 109.03 62.307

M
6 17.75 0.043 0.036 0.038 0.884 418.224 0.958 0.946 524.35 263.96 145.73 86.411 55.436

7 17.63 0.010 0.037 0.032 0.921 391.993 0.956 0.945 420.58 215.94 121.42 73.182 47.545

D
8 17.56 0.009 0.039 0.040 0.912 400.445 0.957 0.947 567.13 282.52 154.97 91.285 57.671

9 16.06 0.001 0.018 0.029 0.952 365.678

TE
0.963 0.957 523.01 256.58 139.61 81.975 51.541 34.149 23.588
C EP
AC

21
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
C EP
AC

Fig. 7. (a) Calculated vs. experimental viscosities for the current and modified Lindeloff models based on Data Set 2;

22
 ACCEPTED MANUSCRIPT

(b) log-log plot of the data in (a).

Table 6. Summaries of the errors generated by the two proposed models when applied to Data Set 2.

Current
Modified Lindeloff

PT
Empirical Model

ARE% 8.73 -40.20

RI
AARE% 12.28 42.50

SD (for AARE) 8.47 22.51

SC
Max% (Abs. error) 32.27 82.29

Min% (Abs. error) 0.14 0.95

U
Correlation coefficient 0.99 0.95

RSME 27.40 378.88

AN
Sum of residuals 187.82 -2597.50

R² 0.98 0.90
M
D

4. CONCLUSIONS
TE

The performance of previously published compositional models for viscosity prediction has been evaluated
over a wide range of temperatures using the experimentally determined viscosities of 28 oil samples. This analysis
EP

revealed that none of the models currently used in the oil industry is capable of accurately predicting the viscosities
of heavy crude oils. The majority of the compositional-based models underestimate viscosities, particularly the LBC
C

model. The errors in the viscosities estimated by the LBC correlation were reduced when measured densities were
used instead of those estimated using the modified Peng–Robinson equation. Some of the well-known viscosity
AC

models, such as LBC and Lindeloff, were modified in order to improve their viscosity predictions by adjusting their
coefficients. These modified models were found to perform better than their original versions for heavy crude oils. A
new compositional-based empirical viscosity model was also introduced to predict dead heavy crudes using the
measured viscosity data at elevated temperatures. This new empirical model is a function of the density, molecular
weight of the mixture, the oil composition, and temperature. The new proposed model successfully predicts
viscosities with an AARE of 18.86%. Model validation with an independent data set from different oil fields
containing 47 data points confirmed that this new empirical model provides better performance and accuracy than

23
 ACCEPTED MANUSCRIPT

any of the published models. In addition, this empirical model can easily be tuned and implemented in any reservoir-
simulation software.

PT
5. NOMENCLATURE
µ, Viscosity

RI
ρ, Density

µod, Dead oil viscosity

SC
ξ, Viscosity reducing parameter

µ*, Low pressure gas mixture viscosity (dilute gas viscosity)

U
T, Temperature

P, Pressure
AN
Pr, Reduced pressure

Tf, Freezing point temperature

M

Tr , Reduced temperature

vr , Reduced volume
D

α, Rotational coupling coefficient

TE

ρr , Reduced density

Ci, Mole fraction of hydrocarbon components

MW, Molecular weight

EP

MWc7+, Molecular weight of C7+ components

MWmix, Mixture molecular weight

C

SGc7+, Specific gravity of C7+ components

AC

, Intermediate components (C4 + C5 + C6)

ARE, Absolute relative error

AARE, Absolute average relative error

SD, Standard deviation

RSME, Root square mean error

24
 ACCEPTED MANUSCRIPT

ACKNOWLEDGMENT

The authors acknowledge the support received from Kuwait University Research Grant No. [GE 01/17], through the
Petroleum Fluid Research Center (PFRC).

PT
RI
REFERENCES
Aasberg-Petersen K, Knudsen K, Fredenslund A. Prediction of viscosities of hydrocarbon mixtures. Fluid Phase

SC
Equilibria. 1991; 70:293–308.

Al-Syabi Z, Danesh A, Tohidi, B, Todd AC., Tehrani, DH. A residual viscosity correlation for predicting the

U
viscosity of petroleum reservoir fluids over wide ranges of pressure and temperature. Chem Eng. Sci. 2001;
56:6997–706.
AN
Ely JF, Hanley HJM. A computer program for the prediction of viscosity and thermal conductivity in hydrocarbon
mixtures (TRAPP). National Bureau of Standards Technical Note 1039. Washington: US Government Printing
M

Office; 1981.

Ely JF, Hanley HJM. Prediction of transport properties. 1. Viscosity of fluids and mixtures. Ind Eng Chem Fundam.
D

1981; 20:323–332.
TE

Guo XQ, Wang LS, Rong SX, Guo TM. Viscosity model based on equations of state for hydrocarbon liquids and
gases. Fluid Phase Equilibria. 1997; 139:405–21.
EP

Guo XQ, Sun CY, Rong SX, Chen GJ, Guo TM. Equation of state analog correlations for the viscosity and thermal
conductivity of hydrocarbons and reservoir fluids. J Pet Sci Eng. 2001; 30:15–27.

Jossi JA, Stiel LI, Thodos G. The viscosity of pure substances in the dense gaseous and liquid phases. AIChE J.
C

1962; 8:59–63.
AC

Laesecke A, Fortin TJ, Splett JD. Density, speed of sound, and viscosity measurements of reference materials for
biofuels. Energy Fuels. 2012; 26 1844–1861.

Lindeloff N, Pedersen KS, Rønningsen HP, Milter J. The corresponding states viscosity model applied to heavy oil
systems. J Can Pet Tech. 2004; 43:47–53.

Little JE, Kennedy HTA. Correlation of the viscosity of hydrocarbon systems with pressure, temperature and
composition. SPE J. 1968; 8; 157–62.

25
 ACCEPTED MANUSCRIPT

Lohrenz J, Bray BG, Clark, CR Calculating Viscosities of reservoir fluids from their compositions. J Pet Technol.
1964; 16:1171–76.

McCarty RDA Modified Benedict-Webb-Rubin Equation of state for methane using recent experimental data.
Cryogenics. 1974; 14:276–80.

PT
Motahhari H. Development of viscosity model for petroleum industry applications. PhD Thesis, University of
Calgary, Alberta, Canada, 2013.

Patel NC, Teja AS. A new cubic equation of state for fluids and fluid mixtures. Chem Eng Sci. 1982; 37:463–73.

RI
Pedersen KS, Fredenslund A, Christensen PL, Thoassen P. viscosity of crude oils. Chem Eng Sci. 1984; 39:1011–

SC
16.

Pedersen KS, Fredenslund A. An improved corresponding states model for the prediction of oil and gas viscosities
and thermal conductivities. Chem Eng Sci. 1987; 42:182–86.

U
Peng DY, Robinson DB. A new two-constant equation of state. Ind Eng Chem Fundam. 1976; 15:59–64.
AN
Raut B, Patil SL, Dandekar AY, Fisk R, Maclean B, Hice V. Comparative study of compositional viscosity
prediction models for medium-heavy oils. Int J Oil Gas Coal Technol. 2008; 1:229–49.
M

Rønningsen HP. Prediction of viscosity and surface tension of North Sea petroleum fluids by using the average
molecular weight. Energy Fuel. 1993; 7:565–73.
D

Teja AS, Rice P. Generalized corresponding states method for the viscosities of liquid mixtures. Ind Eng Chem
TE

Fundam. 1981a; 20:77–81.

Teja AS, Rice P. A Multifluid corresponding states principle for the thermodynamic properties of fluid mixtures.
EP

Chem Eng Sci. 1981b; 36:1–6.

Van der Waals JD. On the continuity of the gaseous and liquid states. PhD thesis, University of Leiden, 1873.
C

Zéberg-Mikkelsen CK. Viscosity study of hydrocarbon fluids at reservoir conditions – modeling and measurements.
PhD Thesis, Technical University of Denmark, Kongens Lyngby, Denmark, 2001.
AC

26
 ACCEPTED MANUSCRIPT

Appendix A

The main input parameters that affect viscosity are displayed in the following figures in order to understand the
relationship between the input and output variables of the model. Figures A-1 through to A-6 show relationships
between viscosity and the input variables for equation 3.

PT
RI
U SC
AN
Fig. A-1. Experimental viscosity as a function of C7+ Fig. A-2. Experimental viscosity as a function of the
M

mole percent. C7+ molecular weight.

D
TE
C EP
AC

Fig. A-3. Experimental viscosity as a function of the Fig. A-4. Experimental viscosity as a function of
C7+ specific gravity. intermediate-component percentage, (C4 + C5 + C6)
%.

27
 ACCEPTED MANUSCRIPT

PT
RI
SC
Fig. A-5. Log of experimental viscosity vs. density. Fig. A-6. Experimental viscosity vs. molecular
weight of the mixture.

U
AN
M
D
TE
C EP
AC

28
 ACCEPTED MANUSCRIPT

Highlights

• New compositional based viscosity model

• The proposed model is much simpler and easier to use than existing methods

PT
• The new proposed model successfully predicts viscosities

RI
U SC
AN
M
D
TE
C EP
AC