Professional Documents
Culture Documents
H I G H L I G H T S
a r t i c l e i n f o a b s t r a c t
Article history: We, herein, present a statistical method for diagnostics of the outliers in phase equilibrium data
Received 11 March 2012 (dissociation data) of simple clathrate hydrates. The applied algorithm is performed on the basis of the
Received in revised form Leverage mathematical approach, in which the statistical Hat matrix, Williams Plot, and the residuals of
6 June 2012
a selected correlation results lead to define the probable outliers. This method not only contributes to
Accepted 13 June 2012
outliers diagnostics but also identifies the range of applicability of the applied model and quality of the
Available online 27 June 2012
existing experimental data. The available correlation in the literature in exponential form is used to
Keywords: represent/predict the hydrate dissociation pressures for three-phase equilibrium conditions (liquid
Evaluation of data water/ice–vapor-hydrate). The investigated hydrate formers are methane, ethane, propane, carbon
Gas hydrate
dioxide, nitrogen, and hydrogen sulfide. It is interpreted from the obtained results that the applied
Leverage approach
correlation for calculation/estimation of the phase behavior of simple clathrate hydrate systems is
Hat matrix
Williams plot statistically valid and correct, 5 hydrate dissociation data are found to be probable doubtful ones and 10
Outlier diagnostics data points are out of applicability domain of the applied correlation.
& 2012 Elsevier Ltd. All rights reserved.
1. Introduction The first industrial importance of the gas hydrate formation has
been attributed to the blockage of gas/oil transportation lines in
Clathrate (gas) hydrates are inclusion compounds composed of petroleum industry in the early 1930s (Sloan and Koh, 2008;
H2O and guest species (Sloan and Koh, 2008). They are normally Hammerschmidt, 1934). More rigorously, gas hydrate formation
stabilized by the guest molecules (can be either gas or special may occur during the steps of production and processing of natural
liquids) enclathrated in the hydrogen-bonded water cages. At gas when traceable amounts of associated water exist (even in the
relatively high pressures and low temperatures, water molecules form of very low water content). Furthermore, novel technologies
form various crystalline structures generally depending on the utilizing gas hydrates have attracted much attention in the recent
size and shape of the guest molecule(s) (Sloan and Koh, 2008). years (called positive applications) (Sloan and Koh, 2008;
Structures I (sI), II (sII), and H (sH) are known to form as three Eslamimanesh et al., 2012a). For instance, these clathrate structures
common structures of clathrate hydrates. may be used as media for the storage and transportation of natural
gas and hydrogen (Sloan and Koh, 2008; Eslamimanesh et al., 2012a).
With respect to the aforementioned points, accurate and reliable
phase equilibrium data are required to design the most efficient
n
Corresponding author at: MINES ParisTech, CEP/TEP-Centre Énergétique et processes (either to prevent gas hydrate formation or to use it as a
Procédés, 35, Rue Saint Honoré, 77305 Fontainebleau, France.
promising technique for other applications) and to adjust the para-
Tel.: þ33 1 64 69 49 70; fax: þ 33 1 64 69 49 68.
E-mail address: amir-hossein.mohammadi@mines-paristech.fr meters of the developed models (thermodynamic/numerical ones) for
(A.H. Mohammadi). the corresponding phase equilibrium calculation/prediction.
0009-2509/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ces.2012.06.019
A. Eslamimanesh et al. / Chemical Engineering Science 80 (2012) 402–408 403
Experimental phase equilibrium data (mainly dissociation data) a matrix known as Hat matrix composed of the experimental data
for simple clathrate hydrate systems e.g. CH4, C2H6, C3H8, CO2, N2, and the represented/predicted values obtained from a correlation
H2S, etc. have been extensively reported in the literature (Sloan (model) (Rousseeuw and Leroy, 1987; Goodall, 1993; Mohammadi
and Koh, 2008). Different experimental techniques have been thus et al., 2012). Hence, a suitable mathematical model is also required
far applied for obtaining these data (Sloan and Koh, 2008; to pursue the calculation steps of the algorithm.
Eslamimanesh et al., 2012a). The Leverage or Hat indices are determined as a Hat matrix (H)
Although the quantities of these reported data seem to be with the following definition (Rousseeuw and Leroy, 1987; Goodall,
adequate for industrial applications, there may be no general 1993; Mohammadi et al., 2012; Gharagheizi et al., 2012):
method to check their qualities. However, two methods have
been already employed to check the reliability of phase equili- H ¼ XðX t XÞ1 X t ð1Þ
brium data of the systems containing gas hydrates as follows:
where X is a two-dimensional matrix composing n chemicals or data
(1) The sketch of logarithm of dissociation pressure (log(p)) vs. (rows) and k parameters of the model (columns) and t stands for the
reciprocal temperature (1/T) generally exhibits a straight line transpose matrix. The Hat values of the chemicals in the feasible
(which shows the value of the enthalpy of hydrate formation). region of the problem are, as a matter of fact, the diagonal elements of
Thus, the hydrate dissociation data for any of the simple the H matrix.
hydrate formers can be sketched to observe whether they The Williams plot is later sketched for graphical identification
show the same trends or not (Eslamimanesh et al., 2011a). of the suspended data or outliers on the basis of the calculated H
(2) Thermodynamic consistency test is generally applied as a values through Eq. (1). This plot shows the correlation of Hat
reliable method for assessment of the phase equilibrium data indices and standardized cross-validated residuals (R), which are
quality. Very recently, we performed consistency test method defined as the difference between the represented/predicted
for the first time on significant systems including water content values and the implemented data (Rousseeuw and Leroy, 1987;
of methane in equilibrium with gas hydrate, liquid water or ice, Goodall, 1993; Mohammadi et al., 2012; Gharagheizi et al., 2012).
(Eslamimanesh et al., 2011b), solubility data of carbon dioxide A warning Leverage (H) is generally fixed at the value equal to 3 n/
and methane with water inside and outside gas hydrate forma- p, where n is number of training points and p is the number of
tion region, (Eslamimanesh et al., 2011c) and composition of model (correlation) input parameters plus one (Rousseeuw
vapor phase in equilibrium with gas hydrate and liquid water and Leroy, 1987; Goodall, 1993; Mohammadi et al., 2012;
for the carbon dioxideþmethane or nitrogenþwater systems Gharagheizi et al., 2012). The leverage of 3 is normally considered
(Eslamimanesh et al., 2012b). However, the preceding method is as a ‘‘cut-off’’ value to accept the points within73 range (two
generally applicable only in the case of applicability of molar horizontal red lines) standard deviations from the mean (to cover
compositions of equilibrium phases. It is worth pointing out that 99% normally distributed data) (Rousseeuw and Leroy, 1987;
the studied phase equilibrium data include the molar composi- Goodall, 1993; Mohammadi et al., 2012; Gharagheizi et al.,
tions of the phases which are not necessarily the hydrate 2012). Existence of the majority of data points in the ranges
dissociation conditions. Such data are very scarce in the open 0 r H r Hn and 3 r Rr 3 reveals that both model development
literature. As a consequence, this method may not be an efficient and its representations/predictions are done in applicability
technique of our concern in this work (Belandria et al., 2011). domain, which results in a statistically valid model. ‘‘Good High
Leverage’’ points are located in domain of Hn r H and 3 r Rr 3.
The Good High Leverage can be designated as the ones, which are
Hence, it is of much interest to propose a statistically-correct
outside of applicability domain of the applied model (Rousseeuw
method for simultaneous detection of the doubtful data and their
and Leroy, 1987; Goodall, 1993; Mohammadi et al., 2012;
quality along with checking the validity and domain of applicability
Gharagheizi et al., 2012). In other words, the model is not able
of an available correlation (in exponential form) for representation/
to represent/predict the following data at all. The points located
prediction of the phase equilibrium data of simple clathrate
in the range of R o 3 or 3 oR (whether they are larger or smaller
hydrates. In the present study, we use the Leverage approach
than the Hn value) are designated as outliers of the model or
(Rousseeuw and Leroy, 1987; Goodall, 1993; Gramatica, 2007) for
‘‘Bad High Leverage’’ points. These erroneous representations/
this purpose. To the best of our knowledge, this is the first time that
predictions can be attributed to the doubtful data (Rousseeuw
a statistical method is used for evaluation of such data.
and Leroy, 1987; Goodall, 1993; Gharagheizi et al., 2012).
2. Theory
2.2. Correlation
2.1. Leverage approach
As the Leverage approach (Rousseeuw and Leroy, 1987; Goodall,
As mentioned in our previous work, outlier diagnostics 1993; Gramatica, 2007) seems to have been developed for checking
(or detection) may be crucial in developing the mathematical the applicability domain of the models merely based on the numer-
models (Rousseeuw and Leroy, 1987; Goodall, 1993; Mohammadi ical regressions, we have used the following recommended simple
et al., 2012). Outlier detection is to identify individual datum (or correlation to represent/predict the corresponding phase equilibrium
groups of data) that may differ from the bulk of the data present in data (Sloan and Koh, 2008; Kamath, 1984)
a dataset (Rousseeuw and Leroy, 1987; Goodall, 1993;
Mohammadi et al., 2012). The proposed methods for this purpose p=MPa ¼ 103 expða þ b:T 1 =KÞ ð2Þ
normally consist of numerical and graphical algorithms
(Rousseeuw and Leroy, 1987; Goodall, 1993; Mohammadi et al., where p stands for the dissociation pressure, T is the dissociation
2012). One of the efficient and reliable statistical algorithm for temperature, and a and b are the parameters of the correlation, which
outlier diagnostics is the Leverage approach (Rousseeuw and are reported in Table 1. (Sloan and Koh, 2008; Kamath, 1984). It
Leroy, 1987; Goodall, 1993; Gramatica, 2007). The employed should be noted that the range of dissociation temperatures and the
elements of this method includes the values of the residuals (i.e. phase equilibrium regions, for which Eq. (2) has been recommended
the deviations of a model results from the experimental data) and are also presented in Table 1 (Sloan and Koh, 2008; Kamath, 1984).
404 A. Eslamimanesh et al. / Chemical Engineering Science 80 (2012) 402–408
3. Experimental data data are generally acceptable to be used for the Leverage statistical
approach (Rousseeuw and Leroy, 1987; Goodall, 1993; Gramatica,
Selected experimental gas hydrate dissociation data (Mohammadi 2007).
et al., 2005; Nakamura et al., 2003; Kharrat and Dalmazzone, 2003; To pursue our objectives, the H values have been calculated
Yang et al., 2001; Ohmura et al., 2002; Jager and Sloan, 2001; Jager, through Eq. (1) and the Williams plots have been sketched in
2001; Nixdorf and Oellrich, 1997; Hachikubo et al., 2002; Servio et al., Figs. 1–12. All the calculated H and R values are presented in several
2000; Guo et al., 1992; Reamer et al., 1952; Roberts et al., 1940; tables in the Supporting Information. The warning Leverages (H*)
Englezos, 1991; Clarke and Bishnoi, 2000; Avlonitis, 1988; Deaton and have been fixed at 3 n/p for the whole data. In addition, the
Frost, 1946; Galloway et al., 1970; Falabella, 1975; Mooijer-van den recommended cut-off value of 3 has been applied (Belandria et al.,
Heuvel et al., 2002; Robinson and Mehta, 1971; Verma et al., 1975; 2011; Rousseeuw and Leroy, 1987; Goodall, 1993).
Patil, 1987; Kubota et al., 1984; Englezos and Ngan, 1993; Platteeuw The following results are interpreted from application of the
and van der Waals, 1959; Holder and Kamath, 1982; Wendland et al., aforementioned methodology
1999; Fan and Guo, 1999; Fan et al., 2000; Yoon and Lee, 1997; Komai
et al., 1997; Ohgaki et al., 1993; Sugahara et al., 2003; Mohammadi (1) Accumulation of the data points in the ranges 0 rH rH* and
and Richon, 2010; Selleck et al., 1952; Bond and Russell, 1948; Von 3 rR r 3 for each dissociaiton datasets of the investigated
Stackelberg, 1949; Mohammadi and Richon, 2009; Marshall et al., simple hydrate formers reveals that the applied correlation is
1964; Wendland et al., 1999) for the systems containing simple
hydrate formers (CH4, C2H6, C3H8, CO2, N2, H2S) in liquid water–
hydrate–vapor (Lw–H–V) and ice–hydrate–vapor (I–H–V) regions
have been treated in this work. Table 2 reports the ranges of the
experimental data along with their sources.
Table 1
Parameters of Eq. (2) (Kamath, 1984) along with the ranges of applicability
(reproduced from (Sloan and Koh, 2008)).
Equilibrium Temperature a b
region range of
applicability/K
Table 2
Range of experimental data treated in this work.
Fig. 3. Detection of the probable doubtful experimental data (Servio et al., 2000;
Galloway et al., 1970) and the applicability domain of the applied correlation for Fig. 5. Detection of the probable doubtful experimental data (Nixdorf and
the C2H6 clathrate hydrate system in the Lw–H–V region The H* value is 0.077. See Oellrich, 1997; Reamer et al., 1952; Deaton and Frost, 1946; Mooijer-van den
the supporting information for references of the out of leverage and suspended Heuvel et al., 2002; Robinson and Mehta, 1971; Verma et al., 1975; Patil, 1987;
data. Kubota et al., 1984; Englezos and Ngan, 1993) and the applicability domain of the
applied correlation for the C3H8 clathrate hydrate system in the Lw–H–V region.
The H* value is 0.088.
Fig. 4. Detection of the probable doubtful experimental data (Roberts et al., 1940;
Fig. 6. Detection of the probable doubtful experimental data (Deaton and Frost,
Deaton and Frost, 1946; Falabella, 1975) and the applicability domain of the
1946; Platteeuw and van der Waals, 1959; Holder and Kamath, 1982) and the
applied correlation for the C2H6 clathrate hydrate system in the I–H–V region. The
applicability domain of the applied correlation for the C3H8 clathrate hydrate
H* value is 0.545.
system in the I–H–V region. The H* value is 0.316.
statistically correct and valid for representation/prediction of recommended to use/develop other models or correlations on
these experimental values. the basis of different theoretical concepts for their calcula-
(2) The whole hydrate dissociation data points can be declared to tions/estimations in order to avoid estimation through biased
within the applicability domain of the correlation except one model calculations (Mohammadi et al., 2012).
point in the corresponding data of carbon dioxide clathrate (3) The data points located in the range of Ro 3 or 3oR (ignoring
hydrates in LwþV þH region, three points of ethane hydrate their H values) are designated as outliers or bad high leverage
dissociation data in LwþV þH region, one point of hydrogen points, as already mentioned. All the hydrate dissociation data
sulfide hydrate dissociation data in Lw þVþH region, and five for the investigated simple hydrate formers are valid (not
points of methane hydrate dissociation data in Lw þV þH outliers) except one point in the corresponding data of carbon
region. In addition, good high leverage points are accumu- dioxide clathrate hydrates in LwþVþH region, two points of
lated in the domains of H* rHand 3 rR r 3. These points ethane hydrate dissociation data in LwþVþ H region, one point
may be declared to be outside of applicability domain of the of methane hydrate dissociation data in LwþVþH region, and
applied correlation though cannot be assigned as doubtful one point of nitrogen hydrate dissociation data in LwþVþ H
experimental data. It should be noted that, in the case of region. These erroneous representations/predictions can be
encountering with good high leverage points, it is attributed to the doubtful hydrate dissociation data.
406 A. Eslamimanesh et al. / Chemical Engineering Science 80 (2012) 402–408
Fig. 10. Detection of the probable doubtful experimental data (Mohammadi and
Fig. 8. Detection of the probable doubtful experimental data (Hachikubo et al., Richon, 2010) and the applicability domain of the applied correlation for the N2
2002; Wendland et al., 1999) and the applicability domain of the applied clathrate hydrate system in the I–H–V region. The H* value is 1.000.
correlation for the CO2 clathrate hydrate system in the I–H–V region The H* value
is 0.857.
Acknowledgment
Galloway, T.J., Ruska, W., Chappelear, P.S., Kobayashi, R., 1970. Experimental Nakamura, T., Makino, T., Sugahara, T., Ohgaki, K., 2003. Stability boundaries of gas
measurement of hydrate numbers for methane and ethane and comparison hydrates helped by methane—Structure-H hydrates of methylcyclohexane
with theoretical values. Ind. Eng. Chem. Fundam. 9, 237–243. and cis-1,2-dimethylcyclohexane. Chem. Eng. Sci. 58, 269–273.
Hammerschmidt, E.G., 1934. Formation of gas hydrates in natural gas transmission Nixdorf, J., Oellrich, L.R., 1997. Experimental determination of hydrate equilibrium
lines. Ind. Eng. Chem. Res. 26, 851–855. conditions for pure gases, binary and ternary mixtures and natural gases. Fluid
Hachikubo, A., Miyamoto, A., Hyakutake, K., Abe, K., Shoji, H., 2002. Phase Phase Equilib. 139, 325–333.
Equilibrium Studies on Gas Hydrates Formed from Various Guest Molecules Ohmura, R., Kashiwazaki, S., Shiota, S., Tsuji, H., Mori, Y.H., 2002. Structure-I and
and Powder Ice. The Fourth International Conference on Gas Hydrates. structure-H hydrate formation using water spraying. Energy Fuels 16,
Yokohama, Japan, 19–23, 357–360 pp. 1141–1147.
Holder, G.D., Kamath, V.A., 1982. Experimental determination of dissociation Ohgaki, K., Makihara, Y., Takano, K., 1993. Formation of CO2 hydrate in pure and
pressures for hydrates of the cis- and trans-isomers of 2-butene below the
sea waters. J. Chem. Eng. Jpn. 26, 558–564.
ice temperature. J. Chem. Thermodyn. 14, 1119–1128.
Patil, S.L., 1987. Measurement of Multiphase Gas Hydrate Phase Equilibria: Effect
Jager, M.D., Sloan, E.D., 2001. The effect of pressure on methane hydration in pure
of Inhibitors and Heavier Hydrocarbon Components. Master of Science Thesis.-
water and sodium chloride solutions. Fluid Phase Equilib. 185, 89–99.
University of Alaska, Fairbanks, Alaska.
Jager, M.D., 2001. High Pressure Studies of Hydrate Phase Inhibition Using Raman
Platteeuw, J.C., van der Waals, J.H., 1959. Thermodynamic properties of gas
Spectroscopy. Ph.D. Chemical Engineering Thesis. Colorado School of Mines.
Golden, Colorado, USA. hydrates II: phase equilibria in the system H2S–C3H8–H2O at 3 1C. Rec. Trav.
Kamath, V.A., 1984. Study of Heat Transfer Characteristics during Dissociation of Chim. Pays-Bas 78, 126–133.
Gas Hydrates in Porous Media. Ph.D. Dissertation. University of Pittsburgh, Rousseeuw, P.J., Leroy, A.M., 1987. Robust Regression and Outlier Detection.
University Microfilms No. 8417404. Ann Arbor, MI. John Wiley & Sons, New York.
Kharrat, M., Dalmazzone, D., 2003. Experimental determination of stability conditions Reamer, H.H., Selleck, F.T., Sage, B.H., 1952. Some prop. of mixed paraffinic & other
of methane hydrate in aqueous calcium chloride solutions using high pressure olefinic hydrate. Trans. Am. Inst. Min. Metall. Pet. Eng. 195, 197–202.
differential scanning calorimetry. J. Chem. Thermodyn. 35, 1489–1505. Roberts, O.L., Brownscombe, E.R., Howe, L.S., 1940. Constitution diagrams and
Kubota, H., Shimizu, K., Tanaka, Y., Makita, T., 1984. Thermodynamic properties of composition of methane and ethane hydrates. Oil & Gas J. 39, 37–43.
R13 (CClF3), R23 (CHF3), R152a (C2H4F2), and propane hydrates for desalina- Robinson, D.B., Mehta, B.R., 1971. Hydrates in the propane–carbon dioxide–water
tion of sea water. J. Chem. Eng. Jpn. 17, 423–429. system. J. Can. Pet. Technol. 10, 33–35.
Komai, T., Yamamoto, Y., Ikegami, S., 1997. Equilibrium properties and kinetics of Sloan, E.D., Koh, C.A., 2008. Clathrate Hydrates of Natural Gases, third ed. CRC
methane and carbon dioxide gas hydrate formation/dissociation. ACS 42, Press, Taylor & Francis Group, New York.
568–570, Division of Fuel Chemistry Preprints. Servio, P., Englezos, P., Bishnoi, P.R., 2000. Kinetics of ethane hydrate growth on
Marshall, D.R., Saito, S., Kobayashi, R., 1964. Hydrates at high pressures: part I. latex spheres measured by a light scattering technique. Ann. N.Y. Acad. Sci.
Methane-water, argon-water, and nitrogen-water systems. AIChE J. 10, 912, 576–582.
202–205. Sugahara, K., Tanaka, Y., Sugahara, T., Ohgaki, K., 2003. Thermodynamic stability
Mohammadi, A.H., Eslamimanesh, A., Gharagheizi, F., Richon, D., 2012. A novel and structure of nitrogen hydrate crystal. J. Supramolecular Chemistry 2,
method for evaluation of asphaltene precipitation titration data. Chem. Eng. 365–368.
Sci. 78, 181–185. Selleck, F.T., Carmichael, L.T., Sage, B.H., 1952. Phase behavior in the hydrogen
Mohammadi, A.H., Anderson, R., Tohidi, B., 2005. Carbon monoxide clathrate sulfide–water system. Ind. Eng. Chem. 44, 2219–2226.
hydrates: equilibrium data and thermodynamic modeling. AlChE J. 51, Verma, V.K., Hand, J.H., Katz, D.L., 1975. Gas Hydrates from Liquid Hydrocarbons
2825–2833. (Methane–Propane–Water System). GVC/AIChE Joint Meeting. Munich, 10 pp.
Mohammadi, A.H., Tohidi, B., Burgass, R.W., 2003. Equilibrium data and thermo- Von Stackelberg, M., 1949. Solid gas hydrates. Die Naturwissenschaften 36,
dynamic modeling of nitrogen, oxygen, and air clathrate hydrates. J. Chem.
359–362.
Eng. Data 48, 612–616.
Wendland, M., Hasse, H., Maurer, G., 1999. Experimental pressure—temperature
Mohammadi, A.H., Richon, D., 2010. Ice-clathrate hydrate-gas phase equilibria for
data on three- and four-phase equilibria of fluid, hydrate, and ice phases in the
air, oxygen, nitrogen, carbon monoxide, methane, or ethane þwater system.
system carbon dioxide–water. J. Chem. and Eng. Data 44, 901–906.
Ind. Eng. Chem. Res. 49, 3976–3979.
Yang, S.O., Cho, S.H., Lee, H., Lee, C.S., 2001. Measurement and prediction of phase
Mohammadi, A.H., Richon, D., 2009. Equilibrium data of carbonyl sulfide and
hydrogen sulfide clathrate hydrates. J. Chem. Eng. Data 54, 2338–2340. equilibria for waterþmethane in hydrate forming conditions. Fluid Phase
Mooijer-van den Heuvel, M.M., Peters, C.J., De Swaan Arons, J., 2002. Gas hydrate Equilib. 185, 53–63.
phase equilibria for propane in the presence of additive components. Fluid Yoon, J.H., Lee, H., 1997. Clathrate phase equilibria for the water–phenol–carbon
Phase Equilib. 193, 245–259. dioxide system. AlChE J. 43, 1884–1893.