Novel Density and Viscosity Correlations For Gases and Gas Mixtures Containing Hydrocarbon and Non-Hydrocarbon Components

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Novel Density and Viscosity

Correlations for Gases and Gas


Mixtures Containing Hydrocarbon
and Non-Hydrocarbon Components

E. M. El-M. Shokir
King Saud University

alkanes and impurities such as carbon dioxide and hydrogen sul-


Abstract fide. These impurities may be present in small quantities in natural
Hydrocarbon gas often contains some amounts of heavier hy- gas and are important when modelling the mixture properties(4). In
drocarbon and non-hydrocarbon components that contribute to this paper, the fuzzy logic technique was applied for developing
its properties (i.e. viscosity and density). Prediction of the density new efficient empirical models to estimate density and viscosity
and viscosity values for hydrocarbon gases is necessary in several of pure hydrocarbon gases (from methane to pentane) and hy-
hydrocarbon gas engineering calculations such as the calculation drocarbon gas mixtures (different gas mixtures of methane with
of gas reserves, gas metering, gas compression, estimating the ethane and/or propane,…., n-Decane) containing small concen-
pressure gradient in gas wells and for the design of pipeline and trations of non-hydrocarbon components (i.e. carbon dioxide, ni-
surface facilities. Literature correlations for the density and vis- trogen and helium) over a wide range of temperatures (0-238°C)
cosity of pure hydrocarbon gas such as methane, ethane, propane, and pressures (1-890 bar). The new models are designed to be sim-
butane and isobutene are available. However, wide-ranging and pler and more efficient than the existing equations of state (EOS).
accurate correlations for predicting the gas viscosity and density They eliminate the numerous computations involved in any EOS
are not available for gas mixtures associated with heavier hydro- calculations.
carbon components and impurities components such as carbon
dioxide, nitrogen, helium and hydrogen sulphide.
This paper presents two new models for estimating the den-
sity and viscosity of pure hydrocarbon gases and hydrocarbon
Hydrocarbon Gas Density
gas mixtures containing high amounts of pentane, plus small The calculation of natural gas volume, density or viscosity at el-
concentrations of non-hydrocarbon components (i.e. carbon di- evated pressures and temperatures requires values of Z-factor. Sev-
oxide, nitrogen and helium), over a wide range of temperatures eral authors(5-7) have developed numerous equations of state (EOS)
and pressures on the basis of fuzzy logic approach. The density for that purpose. The real gas Z-factor is presented as an explicit
model was developed using apparent molecular weight, pseudo- function of the pseudo-reduced pressure and temperature based on
reduced temperature and pseudo-reduced pressure. However, the the theory or law of “corresponding states”(8). All the EOS models
viscosity model was developed using density, apparent molec- are implicit in terms of the Z-factor, which means that the Z-factor
ular weight and pseudo-reduced temperature. The fuzzy models is solved as a root of the EOS. In general, EOS are not easy to use,
were derived from 5,350 measurements of density and viscosity not only because of the numerous calculations involved, but basi-
of various pure gases and gas mixtures. The partitioning of the cally due to the lack of knowledge to calculate critical properties,
input space into the fuzzy regions, represented by the individual acentric factors of plus-fraction’s components, and the binary in-
rules, was obtained through fuzzy clustering. Accuracy of the teraction parameters. However, if reliable models or correlations to
new fuzzy models was compared to various literature correla- predict the above properties are at hand, EOS can be used to pre-
tions by blind tests using 1,460 measurements of density and dict density or physical properties of pure hydrocarbons or hydro-
viscosity. The results show that the new fuzzy models are more carbon mixtures. Obeida et al.(9) presented a method to accurately
accurate than the compared correlations. calculate the compressibility factor of pure hydrocarbon, non-hy-
drocarbon gases and gas mixtures. These calculations are based
on correction functions developed from correlations of Z-factor
Introduction data generated by the Peng-Robinson equation of state(10). How-
ever, they mentioned that to calculate the Z-factor of a gas mix-
Accurate determination of the density, viscosity and phase be- ture, the Z-factors of both hydrocarbon and non-hydrocarbon gases
haviour of pure hydrocarbon gases and hydrocarbon gas mixtures should be known. They recommend to use measured data. If mea-
is essential for reliable reservoir characterization and simulation sured data are not available for one or both of them, the Z-factor
and, hence, for optimum usage and exploitation. The variety of of both gases can be calculated using about 16 polynomial equa-
possible natural gas mixtures at different conditions of interest pre- tions with a total of 48 coefficients. In addition, the proposed func-
clude obtaining the relevant data by experimental means alone, tions were developed for certain ranges of temperatures, pressures
thus, requiring the development of prediction methods. Natural and composition of different gases. Dranchuk and Abou-Kassem(6)
gas is a mixture of many components. Wide ranging correlations developed a gas density equation of state (DAK-EOS) based on
for the viscosity of the lower alkanes, such as methane, ethane, a Han-Starling form of the Benedict-Webb-Rubin(5) equation of
propane, butane and iso-butane, have already been developed and state. They reported an average absolute error of 0.486%. How-
are available in the literature(1-3). However, wide-ranging cor- ever, they also noted that this correlation is specifically limited to
relations are often not readily available for many of the higher the following ranges: 1 < TPr < 3 and 0.2 < PPr < 30. In addition,
Peer Reviewed Paper (“Review and Publication Process” can be found on our Website)

October 2008, Volume 47, No. 10 45


this correlation is also recommended for the region PPr < 1 and 0.7 gas density from the new gas density model/correlation was com-
< TPr < 1. The DAK-EOS is considered the current standard corre- pared with the calculated gas density from different EOS based on
lation for the prediction of gas density. Only 1,500 data points, in- the method of calculating the pseudo-critical properties.
cluding pure components and gas mixtures from several different
sources, were used in building the DAK-EOS. Nishiumi(7) devel-
oped an equation of state (NS-EOS) to estimate thermodynamic Hydrocarbon-Gas Viscosity
properties of pure substances. This equation is of the form of a
Benedict-Webb-Rubin(5) equation of state type (BWR-EOS) and Usually the gas properties are measured in the laboratory. Oc-
has 17 coefficients. The authors recommend not using the NS-EOS casionally, experimental data become unavailable and estima-
in the following ranges: 1 ≤ TPr ≤ 1.3 and 1 ≤ PPr ≤ 3. In addition, tions from EOS or empirical correlations becomes a necessity. The
the application of this equation is very complex. most widely used EOS are Soave-Redlich-Kwong(20) (SRK-EOS),
Brill and Beggs(11) presented an explicit relation to estimate/pre- Peng-Robinson(10) (PR-EOS) and Patel-Teja(21) (PT-EOS). Several
dict the Z-factor. They reported that 94 data points were used to es- attempts have been made to improve the accuracy of PR-EOS to
tablish this correlation, which yields an average absolute error of predict phase equilibria of sour gases(22-25). Li and Guo(22) modi-
0.19%. However, their correlation can only be used in the ranges fied the original PR-EOS to predict properties of sour gases by in-
of 1.2 ≤ TPr ≤ 2.4 and 0 ≤ PPr ≤ 10. troducing 33 constants. However, the modification increases the
As mentioned above, the real gas Z-factor is usually predicted numerical computations and makes the equation inconvenient for
from correlations or equations of state as an explicit function of reservoir simulation usage. Guo et al.(26) presented two viscosity
the pseudo-reduced pressure and temperature based on the law models based on PR-EOS and PT-EOS. They found that their
of corresponding states(8) where the pseudo-reduced pressure and EOS-based viscosity model is capable of satisfactorily describing
temperature are defined relative to pseudo-critical pressure and pure component hydrocarbon viscosity, but poorly predicts the vis-
temperature (PPr = P/PPc and TPr = T/TPc). The principle of cor- cosity of hydrocarbon mixtures.
responding states, as proposed by Van der Waals, applies to single On the other hand, McCain(27) indicated that gas viscosity cor-
component gases, but subsequent work by Kay(12) extended it to relations have errors in estimated viscosity in the order of 20%
gas mixtures. For gas mixtures, pseudo-critical pressure and tem- or more for heavy gases. Reichenberg(28) employed a multiparam-
perature are used in place of critical pressure and temperature. The eter correlation technique for calculating the viscosities of dense
pseudo-critical values have no physical significance, but merely gases. The dense gas viscosity is correlated as a function of TPr
provide a means of correlating mixtures. Kay(12) proposed that and PPr. Larger deviations in the calculated viscosities obtained
pseudo-critical pressure and temperature could be calculated using from this method occur at high pressures. Lucas(29) employed a
simple mole average relationships as: method which is similar to that proposed by Reichenberg in some
aspects. This method also employed temperature and pressure as
PPc = ∑ moli Pci
state variables. The Lucas method requires the critical temperature,
critical pressure, critical compressibility factor and the dipole mo-
TPc = ∑ moliTci ment. The error in the calculated viscosities is more than 8% for
...................................................................................... (1)
dense gases. Chung et al.(30) extended their multiparameter cor-
relation to low pressure gas viscosities to account for high pres-
There are two well known problems associated with using either sure gas viscosities by employing an empirical correction factor.
Kay’s rules or the Stewart-Burkhardt-Voo(13) (SBV) rules: 1) the The deviations in the calculated viscosities were less than 5%(31).
pseudo-critical properties of the heptanes plus fraction must be es- At high pressures, larger deviations in the calculated viscosities
timated; and, 2) the principle of corresponding states breaks down occur. For example at a pressure of 140 bar and at a temperature
when non-hydrocarbon gases or impurities are included in the mix- of 407°C, the deviation in the calculated viscosity of n-butane is
ture. The gravity and molar mass of the heptanes plus fraction and more than 10%. Brule and Starling(32) developed a method similar
the Kesler-Lee(14) equations, could be used to estimate the pseudo-
in form to that of Chung et al.(30). The correlation was modified to
critical properties of the heptanes plus fraction. The Wichert and
make its characterization parameters compatible with those of the
Aziz(15) correlation could be used to adjust the pseudo-critical con-
Benedict-Webb-Rubin EOS(5). Larger deviations in the calculated
stants for hydrogen sulfide and carbon dioxide content. Sutton(16)
viscosities for pure gases occur at high pressure. For example, at
related the pseudo-critical pressure and temperature of gas mix-
a pressure of 200 bars, the highest deviation in the calculated re-
tures as a function of gas specific gravity, as shown in the fol-
sults of methane is 16%. However, this technique was developed
lowing equations:
to be more suitable for heavy hydrocarbons rather than for simple
molecules. Carr et al.(33) developed a two-step procedure to esti-
PPc = 756.8 − 131.0 γ g − 3.6 γ g 2 mate hydrocarbon gas viscosity. The first step is to determine the
TPc = 169.2 − 349.5γ g − 74.0 γ g 2 gas viscosity at atmospheric conditions (i.e. a reference condition).
.............................................................. (2) Once estimated, the viscosity at atmospheric pressure is then ad-
justed to conditions at the desired temperature and pressure using
Corredor et al.(17) presents a further modification of the SBV a second correlation. He reported 0.38% average absolute error for
equations which takes into account the effects of the heptane plus the gas viscosity data at 1 atmosphere. However, he used only 30
fraction, acid gases and nitrogen. Elsharkawy and Elkamel(18) pre- data points to develop these equations. Also, he recommended ap-
sented a new correlation to estimate pseudo-critical properties from plication of these equations for cases where the specific gravity
gas gravity. This correlation accounts for the presence of non-hy- varies between 0.55 and 1.55 and the temperature between 38 and
drocarbons without knowing the compositional details. 149°C. Therefore, the validity of these equations could possibly be
Recently, Londono et al.(19) modified the DAK- and NS-EOS by questioned. Jossi et al.(34) developed a relationship for the viscosity
using optimized quadratic relations for the pseudo-critical pressure of pure gases and gas mixtures which includes pure components
and temperature as functions of gas specific gravity. The fuzzy-set such as argon, nitrogen, oxygen, carbon dioxide, sulfur dioxide,
theory is an efficient tool for modelling the kind of uncertainty as- methane, ethane, propane, butane and pentane. He reported ap-
sociated with vagueness, imprecision and/or a lack of information proximately 4% average absolute error and also stated that this
regarding a particular element of the problem at hand. Therefore, correlation should only be applied for values of reduced density
one of the objectives of this paper is to apply the fuzzy modelling below 2.0. Lee et al.(35) developed a correlation to estimate hydro-
approach for developing a new gas density model as a function of carbon gas viscosity using temperature, gas density at a specific
the pseudo-reduced pressure and temperature, and the molecular temperature and pressure, and the molecular weight of the gas. He
weight of the pure and mixture hydrocarbon gases with an accu- reported 2% average absolute error at low pressures and 4% av-
racy comparable with that obtained using EOS and no need of Z- erage absolute error for high pressures for hydrocarbon gases of
factor calculation. The pseudo-reduced pressure and temperature specific gravities below 1.0. For gases of specific gravity above
was calculated using Kay’s(12) method. In addition, the predicted 1.0, this relation is less accurate. The range of pressures used in the
46 Journal of Canadian Petroleum Technology
development of this correlation is between 6.8 and 551 bar. The where Ri(i = 1, 2,…, c) denotes the ith rule, xj(j = 1, 2,…, m) are the
temperature range is between 38 and 171°C. input variables, yi represents the local output variables, Ai1,…, Aim
Recently, Londono et al.(19), developed a new correlation for hy- are the antecedent linguistic variables, and aio, ai1,…, aim are the
drocarbon gas viscosity as a function of the gas viscosity at 1 atmo- model consequent parameters. The final global output of the TSK
sphere, gas density and temperature. He reported a 4.91% average fuzzy model ŷ for a crisp input vector x = (xi,……, xm)T is calcu-
absolute error for natural gas mixtures. However, this correlation lated by the fuzzy mean-weight formula as:
depends on the measured or calculated gas viscosity at atmosphere
c
pressure. He developed a new correlation for gas viscosity at atmo-
sphere pressure based on his database. Therefore, the second ob- ∑βi ( x )i y^ i
jective of this paper is to apply the fuzzy modelling approach for y^ = i=1
c
developing a new gas viscosity model as a function of the gas den- ∑ βi ( x )
sity, pseudo-reduced temperature and the molecular weight of the i=1 . ................................................................................... (7)
pure and mixture hydrocarbon gases.
where βi(x) represents the degree of firing (DOF) of the ith fuzzy
rule, defined by:
Fuzzy Logic Approach
βi ( x ) = min {µ Ai1 ( x1 ), ,µ Aim ( xm )} , 1≤ i ≤ c
.............................. (8)
The basic concept of fuzzy logic, or fuzzy set theory, was first in-
troduced by Zadeh(36). Unlike crisp logic (CL), which a value may where min represents the minimum operator and μAij: R→[0,1] is
or may not belong to one class, fuzzy sets allow partial member- the membership function of the antecedent fuzzy set Aij. The con-
ship. The membership or non-membership of an element x in crisp struction of the TSK fuzzy model from numerical data proceeds in
set A is described by a characteristic function of μA(x), where: three steps: 1) fuzzy clustering; 2) setting the membership func-
tions; and, 3) parameter estimation(39).
1 if x ∈ A
µ A ( x) = 
0 if x ∉ A
................................................................................ (3) Fuzzy Clustering
Partitioning of the input/output space is performed in the first
Fuzzy set theory extends this concept by defining partial mem- step using the selected clustering method. Each obtained cluster
bership which can take values ranging from 0 to 1: represents a certain operating region of the system in which input/
output data values are highly concentrated. The learning or training
µ A : U → 0,1 data, divided into these information clusters, are then interpreted as
......................................................................................... (4) rules. Methods of fuzzy clustering, such as fuzzy c-means (FCM)
(40) and Gustafson-Kessel (GK)(41), are convenient tools for the pro-

where U refers to the universal set defined in specific problem and cess of partitioning the input space or input/output space. Any clus-
A is a fuzzy set(37). If U is a finite set U = {x1, x2,…, xn}, then a tering method can be used. For a comprehensive treatment of the
fuzzy set A in this universe U can be represented by listing each el- different fuzzy clustering techniques, see Bezdek and Pal(42). In
ement and its degree of membership in the set A as: this paper, FCM was used. The process is iterative as follows: for
data set Z, choose the number of clusters 1 < c < N, the fuzzifica-
A = {µ A ( x1 ) / x1 ,µ A ( x2 ) / x2 , ,µ A ( xn ) / xn } tion exponent s > 1 and the termination tolerant ε > 0.
...................................... (5) Initialize the partition matrix U(0) randomly.
Step 1: Compute the cluster means:
The notation μA(x1)/x1 is used only to separate the membership
degree μA(xi) of an element xi from itself and does not refer to a N

normal division. ∑µ(ikl−1) Zks


The rule-based fuzzy models are models in which the relation- Vi( l ) = k =1
i = 1, 2,…… , c
N s
ships between the different variables are represented by means of ∑µ(ikl−1) 
fuzzy implications or fuzzy if/then rules of the form: If (hypoth- k =1 .................................................. (9)
esis) then (conclusion). The form in the expression above typically
expresses an inference mechanism such that if a fact (hypothesis) is Step 2: Compute the cluster covariance matrices:
known, then another fact (conclusion) can be inferred. Fuzzy infer-
ence is the process of formulating from a given input to an output N s
Z − V ( l )  T
using fuzzy logic. There are three types of fuzzy inference systems ∑µ(ikl−1)   k i 
including the linguistic (Mamdani-type), the relational equation Fi = k =1
i = 1, 2, , c
N
and the Takagi, Sugeno and Kang (TSK)(38). In linguistic models, s

both the antecedent and the consequence are fuzzy sets, while in
∑µ(ikl−1) 
k =1 .................................... (10)
the TSK model the antecedent consists of fuzzy sets, but the con-
sequence is made up of linear equations. Fuzzy relational equation
Step 3: Compute the distances:
models aim at building the fuzzy relation matrices according to the
input/output process data determined. Among these methods, the T
TSK fuzzy model(38) is the least general with respect to modelling Dik2 = Z k − Vi( l )  det( Fi )1/( N +1) Fi−1  Z k − Vi( l ) ,
arbitrary nonlinear systems because every rule describes a fuzzy
k = 1, 2, , N i = 1, 2, , c
region in the input domain, where the outputs depend on the input ...............................(11)
in a linear manner. On the other hand, the TSK model is easier to
identify because it needs fewer rules and its parameters can be es- Step 4: Update the partition matrix. If Dik > 0 for:
timated from numerical data using optimization methods such as
least-square algorithms. So, the TSK model will be used. In gen- 1 ≤ i ≤ c, 1 ≤ k ≤ N ,
eral, the TSK fuzzy model consists of a set of fuzzy rules, each rule µ (ikl ) =
1
describing a local linear input/output relationship as: k
∑ ( Dik / D jk )2/( s−1)
j =1
R1 : if xi is Ai1 and  and xm is Aim ..................................................................... (12)
then yi = ai1x1 +  + aim xm + ai0
..................................................... (6) Otherwise,
October 2008, Volume 47, No. 10 47
bership function such as a triangular membership function that is
µ (ikl ) = 0 if Dik > 0, and µ (ikl ) ∈ 0,1
presented(36) as:
k
with ∑µ(ikl) = 1 until U (1) − U ( l −1) < ε   x − v 
 k ij 
i=1 .............................................. (13) µ Aij ( xk ) = max 0,1 −
 bij 
 
........................................................... (14)
at the end of the iterative procedure, the membership values μik and
cluster centres vi are obtained. with the centre coordinates vij and the parameters bij controlling,
respectively, the mean and the spread (variance) of the member-
ship function.
Setting the Membership Function
The fuzzy clusters detected in the input/output product space Parameter Estimation
give information on how the data points are structured in the input The consequent parameters of a TSK model are obtained as a
space. This information, which is captured in the cluster centres least-square approximation in the following way. Let X denote the
and eigenvalues of the fuzzy covariant matrices, is projected into matrix having an ith row with the input vector xi, and let Y denote
the input axes to induce the antecedent fuzzy sets. If vi1, vi2,..., vim the vector column having yi as its ith component. Let Wi denote the
are the input space coordinates of the ith cluster centre, then the N × N real matrix having the normalized degree of firing βij as its
antecedent fuzzy sets of the TSK model are defined by any mem- jth diagonal element(39, 43), where:

Table 1: Summarizes the minimum and maximum values of the all components in the collected gases.

Fluids Groups 1 2 3 4 5
No. of Data 459 370 655 598 568
Statistical
Parameters Min Max Min Max Min Max Min Max Min Max
C1, mole % 100 100 - - - - - - - -
C2, mole % - - 100 100 - - - - - -
C3, mole % - - - - 100 100 - - - -
iC4, mole % - - - - - - - - - -
nC4, mole % - - - - - - 100 100 - -
iC5, mole % - - - - - - - - - -
nC5, mole % - - - - - - - - 100 100
C6, mole % - - - - - - - - - -
C7+*, mole % - - - - - - - - - -
C8, mole % - - - - - - - - - -
C9, mole % - - - - - - - - - -
C10, mole % - - - - - - - - - -
N2, mole % - - - - - - - - - -
CO2, mole % - - - - - - - - - -
H2S, mole % - - - - - - - - - -
He, mole % - - - - - - - - - -
Viscosity, cP 0.01013 0.0406 0.0096 0.1038 0.0075 0.174 0.00675 0.308 0.00709 0.406
Density, gm/cc 0.00038 0.2975 0.00072 0.4851 0.00105 0.585 0.00139 0.646 0.0017 0.685
T, °R 491.67 919.67 559.67 919.67 491.67 919.67 491.67 919.67 491.67 919.67
P, psi 14.7 10000 14.7 10000 14.7 10000 14.7 10000 14.7 8000
Mw 16.04 16.04 30.07 30.07 44.097 44.097 58.123 58.123 72.15 72.15
continued . .
Fluids Groups 6 7 8 9
No. of Data 1,350 965 395 1,450
Statistical
Parameters Min Max Min Max Min Max Min Max
C1, mole % 10 90 10 90 30 70 71.7 92.9
C2, mole % - - - - - - 0.94 14
C3, mole % 10 90 - - - - 0.42 8.3
iC4, mole % - - - - - - 0.18 1.9
nC4, mole % - - 10 90 - - - 0.87
iC5, mole % - - - - - - - -
nC5, mole % - - - - - - 0.06 0.6
C6, mole % - - - - - - 0.06 0.39
C7+*, mole % - - - - - - 0.01 0.94
C8, mole % - - - - - - - -
C9, mole % - - - - - - - -
C10, mole % - - - - 30 70 - -
N2, mole % - - - - - - - 5.2
CO2, mole % - - - - - - - 3.2
H2S, mole % - - - - - - - 0.05
He, mole % - - - - - - - 0.05
Viscosity, cP 0.0091 0.1573 0.01 0.228 0.0555 0.991 0.0113 0.0527
Density, gm/cc 0.0061 0.5595 0.0091 0.6177 0.2836 0.7404 0.0056 0.366
T, °R 559.67 919.67 559.67 919.67 559.67 799.67 559.67 799.67
P, psi 200 10000 200 10000 800 10000 200 8000
Mw 18.85 41.3 20.251 53.915 28.6647 117.0166 16.60136164 22.302464

* Kesler-Lee(14) equations were used to estimate the pseudo-critical properties of the heptanes plus fraction.

48 Journal of Canadian Petroleum Technology


β i ( xi ) where Zk is the kth data point, Vi is the centre of the ith cluster,
βij = and Z is the average of the data. So, the optimum number of clus-
c
∑ βZ ( x j ) ters is determined as a minimum of the functional S(c) as c in-
Z =1 ..................................................................................... (15) creases. The FCM was applied to the training data set for c = 2,
3,...,12, assuming that the nonlinearity in the data could be suffi-
and let Θi = [ai1,…, aim, ai0] denote the vector of consequent pa- ciently approximated by twelve local linear submodels. Figure 1
rameters of the ith rule. To estimate the offset term ai0, a unitary shows the result obtained with the validity criterion S(c). In this
column I is appended to the matrix X to produce the extended ma- case, the optimum number of clusters was found to be 4, with a
trix Xe = [X,I]. Then, the parameter vector Θi is calculated as the fuzzification exponent s of 2. The antecedent membership func-
least-square solution: tions were constructed from the centres and parameters of the
detected fuzzy cluster by using the trapezoidal membership. For
−1
Θi =  X eT .Wi . X e  X eT .Wi .Y example, Figure 2 shows the resulting membership functions that
................................................................. (16) correspond to one input variable (pseudo-reduced temperature).
The consequent parameters were obtained from the training data
of the set of approximate linear equations Y ≈ X.Θi.Y. set by the least-squared method, as described before. The resulted
gas density model is:
The New Gas Density Model
-300
In this study, a Takagi-Sugeno fuzzy inference system was ap-
plied to estimate gas density from pseudo-reduced pressure and -350
temperature and molecular weight data using Matlab software. For

S(c) Validity Function


-400
this purpose, an extensive database (about 6,810 points) for gas
-450
viscosity and gas density data was collected from a variety of lit-
erature sources(8, 33-35, 44-49). Table 1 summarizes the minimum and -500
maximum values of all components in the collected gases. This da- -550
tabase represents 2,839 points for pure gas components and 3,971
-600
points for gas mixtures. About 5,350 data points are randomly se-
lected for building the new viscosity and density models, and the -650
rest of the data points (1,460) are put aside for testing the resulting -700
gas density and viscosity models. Table 2 shows the minimum and
-750
maximum values for the used data in building the new gas density 0 2 4 6 8 10 12
model. The maximum and minimum values of the different param- Number of Clusters (c)
eters constitute the limits of the fuzzy-model domains; extrapo-
lation beyond these values could lead to unrealistic results. The FIGURE 1: Behaviour of cluster validity function [S(c)] for the gas
training data points (TPr, PPr, Mw, ρg), k = 1, 2,...., 5,350, are orga- density model.
nized in a 5,350 × 4 matrix, as shown in Equation (17), and clus-
tered using the FCM algorithm, as mentioned before.
(a)
0.8
T PPr 1 M w1 ρ g1 
 Pr 1  0.7
T Cluster center
ρg 2 
Gas Density (gm/cc)

PPr 2 M w2
Z =  Pr 2  0.6
     
T PPr k M wk ρ gk 
0.5
 Pr k
.............................................................. (17) 0.4

0.3
To determine the number of fuzzy rules, the following fuzzy va-
lidity measure was used(50). 0.2

0.1
N k  2 2
S ( c) = ∑ ∑ (µ jk )m  Zk − Vi − Vi − Z  0
  0 1 2 3 4
j =1 i=1
........................................ (18) Pseudo-Reduced Temperature

(b)
1.2
Table 2: Summarizes the statistical parameters
Membership Degree

for the data used in building the new gas density 1


correlation.
0.8
Parameter Maximum Minimum
0.6
PPr 26.6 0.021
TPr 3.4 0.6 0.4
Mw 119 16.04
Density 0.741 0.00038 0.2

Table 3: Summarizes the statistical parameters for 0


0 1 2 3 4
the data used in testing the new and other existing gas Pseudo-Reduced Temperature
density correlations.
FIGURE 2: (a) The four local linear submodels (cluster centres
Parameter Maximum Minimum are denoted by the black rectangular marks); (b) the trapezoidal
PPr 20.6 0.3 membership functions obtained after projecting the centres and
TPr 3.35 0.6 standard deviations of the fuzzy clusters onto the input pseudo-
Mw 117.2 16.04 reduced temperature variable, at building the new gas density
Density 0.7185 0.0089 model/correlation.

October 2008, Volume 47, No. 10 49


Table 4: Summarizes the statistical error for the calculated gas density from the new and other existing gas
density correlations.

AAER, % AAER, %
@ Calculation of PPc and TPc from @ Calculation of PPc and TPc from
Correlation Mole Average Relationship Gas Specific Gravity
This study 2.37 ---
DAK-EOS(6) 2.77 4.23'
Londono et al.(19) Optimized DAK-EOS 2.81 3.1"
Londono et al.(19) Optimized NS-EOS 21.8 24.2"
Beggs and Brill(11) 4.58 6.2'

R1 : if 0.4 ≤ TPr ≤ 1.154 and 0 < PPr ≤ 25 and 16 ≤ M w ≤ 119 then Table 4 shows that the new gas density model yields the accu-
Gas Density = −4.9879557788 E − 01TPr + 1.47383883796 E − 02 PPr
rate prediction of the experimental gas density for all the tested
gases with the lowest percentage of AAER among all the tested
+9.0073085492 E − 04 M w + 7.813443259854 E − 01
.....(19a) correlations.

R2 : if 1.154 < TPr ≤ 1.512 and 4.675 < PPr ≤ 22.5 and 16 ≤ M w ≤ 119
then Gas Density = −3.7536301831E − 01TPr + 2.25887183348 E
The New Gas Viscosity Model
−02 PPr + 1.3264966273E − 03 M w + 5.916942278815 E − 01 As mentioned before, gas viscosity is a function of temperature,
..... (19b)
molecular weight, pressure and compressibility factor, which, in
turn, is function of the pseudo-reduced pressure and temperature.
R3 : if 1.512 < TPr ≤ 2.506 and 4.675 < PPr ≤ 23 and 16 ≤ M w ≤ 119 Therefore, gas density, pseudo-reduced temperature and molecular
weight were used as independent variables to develop the new gas
and 2.506 ≤ TPr ≤ 3.4 and 10.4 ≤ PPr ≤ 15.5 then
viscosity correlation/model. The same 5,350 data set points that
Gas Density = −8.316619168 E − 02TPr + 2.118041458 E − 02 PPr
were used for building the density model are also used for building
+5.249806233E − 03 M w + 8.086108681E − 02 the gas viscosity model. The same procedures mentioned above
......(19c)
for building the new gas density model by fuzzy logic approach
were applied. Table 5 shows the maximum and minimum values
R4 : if 1.512 < TPr ≤ 2.506 and 0 < PPr ≤ 4.675 and 16 ≤ M w ≤ 119 of the different parameters that constitute the limits of the fuzzy
and 2.506 ≤ TPr ≤ 3.4 and 4.675 ≤ PPr ≤ 10.4 then
gas viscosity model domains; extrapolation beyond these values
could lead to unrealistic results. The training data points (ρg, TPr,
Gas Density = −7.553883745 E − 02TPr + 3.636967857 E − 02 PPr
Mw, GV), k = 1,2,..., 5,350, are organized in a 5,350 × 4 matrix, as
+5.360699055 E − 03 M w + 1.883403033E − 02 shown in Equation (20), and clustered using the FCM algorithm,
..... (19d)
as mentioned before, and the number of fuzzy rules are determined
by the same fuzzy validity measurement in Equation (18). The re-
sulted optimum cluster numbers are equal to 4, as shown in Figure
4. The antecedent membership functions were also constructed
Validations of the New Gas Density Model from the centres and parameters of the detected fuzzy cluster by
To validate and test the new gas density model, the other 1,460 using the trapezoidal membership. For example, Figure 5 shows
points that randomly cut out from the original database were used the resulting membership functions that correspond to the input
for that propose. The minimum and maximum values for these test gas density variable.
data parameters are shown in Table 3. These data points were also
used for comparing the results of the new model with the results ρ TPr 1 M w1 GV1 
of the other density predicting model. The calculated gas density  g1 
from DAK-EOS(6), Londono et al.(19) optimized DAK-EOS and ρ TPr 2 M w2 GV2 
Z =  g2 
Londono et al. optimized NS-EOS were carried out two times. Ini-      
tially, Kay’s(12) concept for calculating the pseudo-reduced pres- ρ TPr k M wk GVk 
sure and temperature was used. Then Sutton’s(16) concept for  gk
....................................................... (20)
calculating the pseudo-reduced pressure and temperature was used
for DAK-EOS(6). Finally, for the other two correlations, Londono The resulting gas viscosity model is:
et al.’s optimized quadratic relations for the pseudo-critical pres-
sure and temperature as a function of gas specific gravity were
R1 : if 0 < ρ g ≤ 0.26 and 0.7 ≤ TPr ≤ 3.4 and 16 ≤ M w ≤ 72.2 then
used. The results are shown in Table 4. This table shows that the
accuracy of the calculated gas density by using the Kay concept is Gas Vis cos ity = 5.6563271E − 02ρ g + 4.9374602 E − 03TPr +
the best. Therefore, the predicted gas density from the new model 4.1949307 E − 05 M w + 2.93978342 E − 03
was compared with the other gas density correlations by using the ....(21a)
concept of Kay. Figure 3(a-e) shows the predicted gas density from
the new model compared to the results of the other gas density cor-
relations versus the measured gas density. It should be noticed that R2 : if 0.26 < ρ g ≤ 0.46 and 0.88 ≤ TPr ≤ 3.4 and 16 ≤ M w ≤ 72.2
all the points right on the x-axis represent overestimated points
then Gas Vis cos ity = 1.3707401E − 01ρ g − 2.72296913E − 03TPr
(out of the scale), and some of them are out of the working range
of the corresponding gas density correlation. +7.64990184 E − 06 M w + 4.050623771E − 03
..... (21b)
As shown in Figure 3(a-e) the new gas density model yields a
good agreement between the predicted and the measured gas den-
sity with absolute average error (AAER, %) equal to 2.37%. In the
order of accuracy, DAK-EOS and Londono et al. optimized DAK- R3 : if 0.46 < ρ g ≤ 0.595 and 0.645 ≤ TPr ≤ 1.5 and 21.6 ≤ M w ≤ 119
EOS come in the second and third order, respectively. However, then Gas Vis cos ity = 4.8834347 E − 01ρ g + 4.437225271E − 02TPr +
DAK-EOS and Londono et al. optimized DAK-EOS correlations 6.57567117 E − 04 M w − 2.002744453E − 01
have 192 and 182 overestimated points, respectively. In addition, ......(21c)
50 Journal of Canadian Petroleum Technology
(a) (b)

0.8 0.8

Calculated Gas Density (gm/cc)

Calculated Gas Density (gm/cc)


0.7 0.7 DAK-EOS(6)
This Study

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.0
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Measured Gas Density (gm/cc) Measured Gas Density (gm/cc)

(c) (d)
0.8 0.8
Calculated Gas Density (gm/cc)

Calculated gas Density (gm/cc)


Beggs & Brill (11)
0.7 Londono et al. - 0.7
Optimized -DAK-
0.6 EOS(19) 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Measured Gas Density (gm/cc) Measured Gas Density (gm/cc)

(e)
0.8
Measured Gas Density (gm/cc)

0.7 Londono et al.-


optimized-NS-EOS(19)
0.6

0.5

0.4

0.3

0.2

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Calculated Gas Density (gm/cc)

FIGURE 3: Predicted gas density from the new gas density model, DAK-EOS(6), Londono et al.(19) optimized DAK-EOS, Beggs and Brill(11)
and Londono et al.(19) optimized NS-EOS vs. the measured gas density in (a) to (e), respectively.

Table 5: Summarizes the statistical parameters and Londono et al. may result from the error in the calculated gas
for the data used in building the new gas viscosity viscosity at atmosphere pressure, whereas the Londono et al. cor-
correlation. relation was developed by using only 261 points.

Parameter Maximum Minimum


Density 0.741 0.00038
TPr 3.4 0.6 Validations of the New Gas Viscosity
Mw 119 16.04 Model
Viscosity 0.991 0.00675
Also, to validate and test the new gas viscosity model, 1,460
R4 : if 0.595 < ρ g ≤ 0.76 and 0.5 ≤ TPr ≤ 1.4 and 35 ≤ M w ≤ 119 points randomly picked from the original database were used. The
minimum and maximum values for these test data parameters are
then Gas Vis cos ity = 1.500143567ρ g + 9.84848797 E − 02TPr +
shown in Table 6. These data points were also used for comparing
1.407003797 E − 03 M w − 8.257944449 E − 01 the results of the new model with the results of the other viscosity
........ (21d)
predicting correlations. Figure 6 (a-d) shows the predicted gas vis-
at atmosphere pressure was calculated by the Londono et al.(19) cosity from the new model compared to the results of the earlier
correlation at 1 atmosphere. It can be concluded that the higher gas viscosity correlations versus the measured gas viscosity. All
percentage of error in the calculated gas viscosity by Jossi et al.(34) the points to the right on the x-axis represent overestimated points
October 2008, Volume 47, No. 10 51
Table 6: Summarizes the statistical parameters for
-500
the data used in testing the new and other existing gas
-550 viscosity correlations.
S(c) Validity Function
-600
Parameter Maximum Minimum
-650

-700
Density 0.7185 0.0089
TPr 3.35 0.6
-750 Mw 117.2 16.04
-800 Viscosity 0.55 0.0088
-850
-900
Table 7: Summarizes the statistical error for the
calculated gas viscosity from the new and other
-950
0 2 4 6 8 10 12 14 existing gas viscosity correlations.
Number of Clusters (c)
Correlation AAER, %
FIGURE 4: Behaviour of cluster validity function [S(c)] for the gas This 2.44
viscosity model. Lee et al.(35) 3.68
Londono et al.(19) 35
Jossi et al.(34) 48
(a)
1.2 viscosity at atmospheric pressure using the Londono et al. correla-
tion, which was developed using only 261 points.
1 Cluster center
Gas Viscosity (cP)

0.8
Conclusions and Recommendations
0.6
1. The presented fuzzy-modelling approach in this paper shows
0.4 a good potential to model complex, nonlinear and multivari-
able natural gas properties.
0.2 2. The fuzzy-modelling approach is capable of estimating the
density and viscosity of pure gases and gas mixtures with
0 high accuracy compared to the experimental values.
0 0.2 0.4 0.6 0.8
3. The new gas density model/correlation yields accurate pre-
Gas Density (gm/cc) dictions with the lowest average absolute error (2.37%)
(b)
among all tested gas density correlations.
1.2 4. The new gas viscosity model/correlation yields accurate
predictions with the lowest average absolute error (2.44%)
Membership Degree

1
among all tested gas viscosity correlations.
0.8 5. The new gas density model can be used as an effective tool to
estimate the density of the natural gas mixtures with no need
0.6
to calculate the gas compressibility factor.
0.4 6. The new developed gas viscosity model can be used to esti-
mate the viscosity of the natural gas mixtures with no need
0.2 to measure or calculate the gas viscosity at atmospheric
pressure as required by most of the early gas viscosity
0
correlations.
0 0.2 0.4 0.6 0.8
7. This work could be extended to develop a universal viscosity
Gas Density (gm/cc) and density correlation considering gas condensate and sour
natural gas mixtures. In addition, investigations of the ex-
FIGURE 5: (a) The four local linear submodels (cluster centres
are denoted by the black rectangular marks); (b) the trapezoidal plicit effects of non-hydrocarbon components, such as water,
membership functions obtained after projecting the centres and nitrogen, carbon dioxide and hydrogen sulfide, on both vis-
standard deviations of the fuzzy clusters onto the input gas density cosity and density of gas mixtures are recommended.
variable, at building the new gas viscosity model/correlation.

(out of the scale), and some of them are out of the working range Acknowledgements
of the corresponding gas viscosity correlation. Table 7 summarizes
The author thanks the Research Centre of the College of Engi-
the percentage of the AAER for all the test gas viscosity corre-
neering at King Saud University for providing the financial sup-
lations. As shown in Figure 6(a-d), the new gas viscosity model port for this study under research grant No. 44/429.
yields a good agreement between the predicted and the measured
gas viscosity with AAER equal to 2.44%. The Lee et al.(35) corre- Nomenclature
lation comes in the order of accuracy with AAER equal to 3.68%
and having zero points out of its range. However, Londono et al.(19) A = a crisp set
and Jossi et al.(34) yield an AAER of more than 35% and having bij = parameter controlling spread (variance) of membership
function
300 and 273 overestimated points, respectively. Both Londono et
c = number of clusters
al. and Jossi et al. gas viscosity correlations are highly dependent
Di = distance between any point (Zk) in the training matrix Z
on the gas viscosity at 1 atmosphere. Due to the lack of the atmo- and the cluster mean vi
spheric gas viscosity at each tested temperature, then the gas vis- Fi = cluster covariance matrices
cosity at atmospheric pressure was calculated by the Londono et GV = gas viscosity
al. correlation at 1 atmosphere. It can be concluded that the high k = sample number
percentage of error in the calculated gas viscosity using Jossi et al. moli = mole fraction of component i
and Londono et al. may be a result of the error in the calculated gas m = input variables
52 Journal of Canadian Petroleum Technology
0.50
0.50

Calculated Gas Viscosity (cP)

Calculated Gas Viscosity (cP)


This Study Lee et.(35)
0.40 0.40

0.30 0.30

0.20 0.20

0.10 0.10

0.00 0.00
0.00 0.10 0.20 0.30 0.40 0.50 0.00 0.10 0.20 0.30 0.40 0.50

Measured Gas Viscosity (cP) Measured Gas Viscosity (cP)

0.50 0.50
Calculated Gas Viscosity (cP)

Calculated Gas Viscosity (cP)


Londono et al.(19) Jossi et al.(34)
0.40 0.40

0.30 0.30

0.20 0.20

0.10 0.10

0.00 0.00
0.00 0.10 0.20 0.30 0.40 0.50 0.00 0.10 0.20 0.30 0.40 0.50

Measured Gas Viscosity (cP) Measured Gas Viscosity (cP)


FIGURE 6: Predicted gas viscosity from the new gas model, Lee et al.(35), Londono et al.(19) and Jossi et al.(34) vs. the measured gas viscosity in
(a) to (d), respectively.

Mw = molecular weight 2. Friend, D.G., ELY, J.F. and Ingham, H., Thermophysical Proper-
N = total number of samples ties of Methane; Journal of Physical and Chemical Reference Data,
PPc = pseudo-critical pressure Vol. 18, No. 2, pp. 583-638, 1989.
Pci = critical pressure of i gas component 3. Vogel, E., Kuchenmeister, C., Bich, E. and Laesecke,
A., Reference Correlation of the Viscosity of Propane; Journal of
PPr = pseudo-reduced pressure Physical and Chemical Reference Data, Vol. 27, No. 5, pp. 947-970,
Ri = fuzzy rules 1998.
s = fuzzification exponent 4. HUBER, M.L., Laesecke, A. and Xiang, H.W., Viscosity Cor-
S(c) = fuzzy validity measuring function relations for Minor Constituent Fluids in Natural Gas: n-octane,
TPc = pseudo-critical temperature n-nonane and n-decane; Fluid Phase Equilibria, Vol. 224, No. 2, pp.
Tci = critical temperature of i gas component 263-270, 2004.
TPr = pseudo-reduced temperature 5. Benedict, M., Webb, G.B. and Rubin, L.C., An Empirical
Equation for Thermodynamic Properties of Light Hydrocarbons and
Vi = cluster means matrix
Their Mixtures: I. Methane, Ethane, Propane, and n-Butane; Journal
Wi = N × N real matrix having the normalized DOF of the ith of Chemical Physics, Vol. 8, No.4, pp. 334-345, 1940.
fuzzy rule as its jth diagonal element 6. Dranchuk, P.M. and Abou-Kassem, J.H., Calculation of Z
X = any matrix having ith row with the input vector xi Factors for Natural Gases Using Equations of State; Journal of Cana-
Xe = extended matrix produced by adding a unitary column I dian Petroleum Technology, Vol. 14, No. 3, pp. 34-36, Ju1y-September
to the matrix X 1975.
Y = any vector column having yi as its ith components 7. Nishiumi, H., An Improved Generalized BWR Equation of State
ŷ = final global output of the TSK fuzzy model with Three Polar Parameters Applicable to Polar Substances; Journal
Z_ = Matrix of training data points of Chemical Engineering of Japan, Vol. 13, No. 3, pp. 178-183,1980.
Z = average value of the input variable data 8. STANDING, M.B. and KATZ, D.L., Density of Natural Gases;
Transactions AIME, Vol. 146, pp. 140-149, 1942.
βi(x) = degree of firing (DOF) of the ith fuzzy rule
9. Obeida, T.A., Heinemann, Z.E. and Kriebernegg, M., Ac-
U = universal set
curate Calculations of Compressibility Factor for Pure Gases and Gas
Θi = vector of the resulting coefficients of the ith fuzzy rule Mixtures; paper SPE 37440 presented at the SPE Production Opera-
vi1 = input space coordinates of the ith cluster centre tions Symposium, Oklahoma City, OK, 9-11 March 1997.
ε = termination tolerant 10. PENG, D.-Y. and ROBINSON, D.B., A New Two Constant Equation
μA = characteristic function representing the membership of of State; Industrial and Engineering Chemistry Fundamentals, Vol.
any element in the crisp set A 15, No. 1, pp. 59-64, 1976.
ρg = gas density 11. Brill, J.P. and Beggs, H.D., Two-Phase Flow in Pipes; Intercomp
γg = gas specific gravity, (air = 1) Course, The Hague, Netherlands, 1974.
12. Kay, W.B., Density of Hydrocarbon Gases and Vapors at High Tem-
perature and Pressure; Industrial and Engineering Chemistry, Vol. 28,
References pp. 1014-1019, September 1936.
1. Friend, D.G, Ingham, H. and ELY, J.F., Thermophysical Prop- 13. STEWART, W.F., BURKHARDT, S.F. and VOO, D., Prediction of
erties of Ethane; Journal of Physical and Chemical Reference Data, Pseudocritical Parameters for Mixtures; paper presented at the Na-
Vol. 20, No. 2, pp. 275-347, 1991. tional Meeting of the AIChE, Kansas City, MO, 18 May 1959.
October 2008, Volume 47, No. 10 53
14. KESLER, M.G. and LEE, B.I., Improved Prediction of Enthalpy 38. Takagi, T. and Sugeno, M., Fuzzy Identification of Systems and
of Fractions; Hydrocarbon Processing, Vol. 55, No. 3, pp. 153-158, Its Applications to Modelling and Control; IEEE Transaction on Sys-
March 1976. tems, Man and Cybernetics, Vol. 15, No. 1, pp. 116-132, 1985.
15. Wichert, E., and Aziz, K., Calculation of Z’s for Sour Gases; 39. Shokir, E.M. EL-M., A Novel Model for Permeability Prediction
Hydrocarbon Processing, Vol. 51, No. 5, pp. 119-122, May 1972. in Uncored Wells; SPE Reservoir Evaluation & Engineering, Vol. 9,
16. SUTTON, R.P., Compressibility Factors for High-Molecular-Weight
No. 3, pp. 266-273, June 2006.
Reservoir Gases; paper SPE 14265 presented at the SPE Annual
Technical Conference and Exhibition, Las Vegas, NV, 22-26 Sep- 40. Yen, J., Wang, L. and Gillespie, C.W., Improving the Interpret-
tember 1985. ability of TSK Fuzzy Models by Combining Global Learning and
17. Corredor, J.H., Piper, L.D. and McCain, W.D., Compress- Local Learning; IEEE Transactions on Fuzzy Systems, Vol. 6, No. 4,
ibility Factors for Naturally Occurring Petroleum Gases; paper SPE pp. 531-537, November 1998.
24864 presented at the SPE Annual Technical Conference and Exhi- 41. Gustafson, D.E. and Kessel, W.C., Fuzzy Clustering With a
bition, Washington, DC, 4-7 October 1992. Fuzzy Covariance Matrix; IEEE-CDC, Vol. 2, pp. 761-766, 1979.
18. ELSHARKAWY, A.M. and ELKAMEL, A., Compressibility Factor
for Sour Gas Reservoirs; paper SPE 64284 presented at the SPE Asia 42. Bezdek, J.C. and Pal, S.K., Fuzzy Models for Pattern Recogni-
Pacific Oil and Gas Conference and Exhibition, Brisbane, Australia, tion: Methods That Search for Structures in Data; IEEE Press, New
16-18 October 2000. York, NY, 1992.
19. Londono, F.E., ARcher, R.A. and Blasingame, T.A., Corre- 43. Wang, L.-X. and Mendel, J.M., Generating Fuzzy Rules by
lations Hydrocarbon-Gas Viscosity and Gas Density – Validation and Learning From Examples; IEEE Transactions on Systems, Man and
Correlation of Behavior Using a Large-Scale Database; SPE Reser- Cybernetics, Vol. 22, No. 6, pp. 1414-1427, 1992.
voir Evaluation & Engineering, Vol. 8. No. 6, pp. 561-572, December
44. Gonzalez, M.H., Eakin, B.E. and Lee, A.L., Viscosity of
2005.
Natural Gases; American Petroleum Institute, Monograph on API Re-
20. SOAVE, G., Equilibrium Constants from a Modified Redlich-Kwong
search Project 65, New York, NY, 1970.
Equation of State; Chemical Engineering Science, Vol. 27, pp. 1197-
1203, 1972. 45. Diehl, J., Gondouin, M., Houpeurt, A., Neoschil, J.,
21. PATEL, N.C. and TEJA, A.S., A New Cubic Equation of State for Thelliez, M., Verrien, J.P. and Zurawsky, R., Viscosity and
Fluids and Fluid Mixtures; Chemical Engineering Science, Vol. 37, Density of Light Parafins, Nitrogen and Carbon Dioxide; Editions
No. 3, pp. 463-473, 1982. Technip, CREPS/Geopetrole, Paris, France, 1970.
22. LI, Q. and GUO, T.M., A Study on the Supercompressibility and 46. Golubev, I.F., Viscosity of Gases and Gas Mixtures, a Handbook;
Compressibility Factors of Natural Gas Mixtures; Journal of Petro- this paper is a translation from Russian by the NTIS (National Tech-
leum Science and Engineering, Vol. 6, pp. 235-247, 1991. nical Information Service), Springfield, VA, 1959.
23. MOHSEN-NIA, M., MODDARESS, H. and MANSOORI, G.A.,
Sour Natural Gas and Liquid Equation of State; paper SPE 26906 47. Stephan, K. and Lucas, K., Viscosity of Dense Fluids; The
presented at the SPE Eastern Regional Meeting, Pittsburg, PA, 2-4 Purdue Research Foundation, Plenum, New York, NY, 1979.
November 1993. 48. Setzmann, U. and Wagner, W., A New Equation of State and
24. HURON, M.J., DUFOUR, G.N. and VIDAL, J., Vapour-Liquid Tables of Thermodynamic Properties for Methane Covering the
Equilibrium and Critical Locus Curve Calculations with the Soave Range from the Melting Line to 625 K at Pressures up to 1000 MPa;
Equation for Hydrocarbon Systems with Carbon Dioxide and Hy- Journal of Physical and Chemical Reference Data, Vol. 20, No. 6, pp.
drogen Sulfide; Fluid Phase Equilibria, Vol. 1, pp. 247-265, 1978. 1061-1155, 1991.
25. EVELEIN, K.A. and MOORE, R.G., Prediction of Phase Equilibria
in Sour Natural Gas Systems Using the Soave-Redlich-Kwong Equa- 49. PoetTmann, H.F. and Carppenter, P.G., The Multiphase Flow
tion of State; Industrial and Engineering Chemistry Process Design of Gas, Oil and Water Through Vertical Flow String with Application
and Development, Vol. 18, No. 4, pp. 618-624, 1979. to the Design of Gas-lift Installations; API Drilling and Production
26. GUO, X.-Q., WANG, L.-S., RONG, S.-X. and GUO, T.-M., Vis- Practice, Vol. 17, pp. 257-263, 1952.
cosity Model Based on Equations of State for Hydrocarbon Liquids 50. Fukuyama, Y. and Sugeno, M., A New Method of Choosing
and Gases; Fluid Phase Equilibria, Vol. 139, Nos. 1-2, pp. 405-421, the Number of Clusters for Fuzzy C Means Method; in the Proceed-
December 1997. ings of the 5th Fuzzy System Symposium, pp. 247-250, Tokyo, Japan,
27. MCCAIN, JR., W.D., The Properties of Petroleum Fluids; Second 1989.
Edition, PennWell Books, Tulsa, OK, 1990.
28. Reichenberg, D., Fluids and Fluids Mixtures; Symposium on
Transport Properties of Fluids and Fluid Mixtures, National Engi-
neering Laboratory, East Kilbride, Glasgow, Scotland, May 1979. Provenance—Original Petroleum Society manuscript, Novel Density and
29. LUCAS, K., Die Berechnung der Eigenschaften von Gasen und Flüs- Viscosity Correlations for Gases and Gas Mixtures Containing Hydro-
sigkeiten nach der Molekularen Theorie  -  Stand des Wissens und carbon and Non-Hydrocarbon Components (2007-161), first presented
Anwendungen in der Verfahrenstechnik; Chemie Ingenieur Technik, at the 8th Canadian International Petroleum Conference (the 58th Annual
Vol. 49, No. 2, pp. 117-125, February 1977. Technical Meeting of the Petroleum Society), June 12-14, 2007, in Cal-
30. Chung, T.-H., LEE, L.L. and Starling, K.E., Application of Ki-
gary, Alberta. Abstract submitted for review December 15, 2006; editorial
netic Gas Theories and Multiparameter Correlation for Prediction of
comments sent to the author(s) June 25, 2008; revised manuscript received
Dilute Gas Viscosity and Thermal Conductivity; Industrial and Engi-
July 14, 2008; paper approved for pre-press July 14, 2008; final approval
neering Chemistry Fundamentals, Vol. 23, No. 13, pp. 8-13, 1984.
September 18, 2008.
31. Reid, R.C., PrausNiTz, J.M. and Poling, B.E., The Properties
of Gases and Liquids; Fourth Edition, McGraw-Hill Book Company,
New York, NY, 1987.
32. Brule, M.R. and Starling, K.E., Thermophysical Properties of
Complex Systems: Applications of Multiproperty Analysis; Indus-
trial and Engineering Chemistry Process Design and Development, Author’s Biography
Vol. 23, pp. 833-845, 1984.
33. Carr, N.L. Kobayashi, R. and Burrows, D.B., Viscosity of Eissa Mohamed El-Moghawry Bayoumi
Hydrocarbon Gases Under Pressure; Transactions AIME, Vol. 201, Shokir is an Associate Professor of Petro-
pp. 264-272, 1954. leum Engineering at King Saud University.
34. Jossi, J.A., Stiel, L.I. and Thodos, G., The Viscosity of Pure Dr. Shokir is an author or co-author of many
Substances in the Dense Gaseous and Liquid Phases; AIChE Journal, published papers in international journals
Vol. 8, No.1, pp. 59-63, March 1962. and technical papers in various subjects in-
35. LEE, A.L., GONZALEZ, M.H. and EAKIN, B.E., The Viscosity of cluding artificial intelligence, formation
Natural Gases; Journal of Petroleum Technology, Vol. 18, No. 8, pp. evaluation, fluid phase behaviour, forma-
997-1000, August 1966.
36. Zadeh, L.A., Fuzzy Sets; Information and Control, Vol. 8, No. 3,
tion damage and drilling. He holds B.Sc.
pp. 338-353, 1965. and M.Sc. degrees from the Cairo Uni-
37. Yager, R.R. and Zadeh, L.A., An Introduction to Fuzzy Logic versity, Egypt, and a Ph.D. degree from
Applications in Intelligent Systems; Kluwer Academic Publishers, the Technical University of Clausthal, Germany, all in petroleum
Norwell, MA, 1992. engineering.
54 Journal of Canadian Petroleum Technology

You might also like