Vous êtes sur la page 1sur 6

Parameter collection for saturated and compressed liquid

density models based on Hankinson-Thomson model

Kari I. Keskinen1,2* and Mika M. Aalto3
Process Engineering, Neste Engineering Oy, P.O. Box 310, FIN-06101 PORVOO,
Finland, e-mail: kari.keskinen@fortum.com
Laboratory for Chemical Engineering and Plant Design, Helsinki University of
Technology, P.O. Box 6100, FIN-02015 HUT, Finland
Bio- and Chemical Technology, National Technology Agency of Finland, P.O. Box
69, FIN-00101 HELSINKI, Finland, e-mail: mika.aalto@tekes.fi
One of the most successful saturated liquid density models is the Hankinson-Thomson (HT) model, often called
COSTALD. It is accurate for pure compounds from the triple point to the near-critical region, and it has been
successfully extended to mixtures. Moreover, a large collection of pure-component parameters has already been
published. HT was developed mainly for hydrocarbons and other organic compounds, but it is also accurate for
many other types of compounds.
The Hankinson-Brobst-Thomson (HBT) model is an extension of the HT model to the compressed liquid region;
HBT is applicable both for pure compounds and mixtures. Later, many new methods for compressed liquid
density calculation have been proposed. Many of those utilize a concept similar to HBT model: First, the
saturated liquid density is obtained, and then the effect of pressure is described using another model. HT has
often been selected for the saturated liquid density predictions in these compressed liquid density models due to
its good accuracy.
We have initiated a project where the parameters needed for HT model and some of the compressed liquid
density models will be determined for as many compounds as possible. Saturated liquid density data, vapor
pressure data and critical constants for pure compounds are collected using databases and literature, evaluated,
and used for parameter optimization. Moreover, this work will give us ideas for improvement of the HT model.
It is already known that some light gaseous compounds and associating compounds need special treatment in the
model. In addition to that, improvement may be needed for liquid mixture calculations.
A large pure-component parameter compilation for the HT model will be published. The parameters will be of
use for anyone interested in calculating accurately the saturated and/or compressed liquid density of a pure
compound or liquid mixture. The parameters will also be available at a web-site.

Keywords: Hankinson-Thomson model, COSTALD, liquid density, saturated liquid,

compressed liquid, liquid mixture, model parameters

1 Introduction
Liquid densities are needed in many engineering problems such as process calculations, simulations, equipment
and pipe design, and liquid metering calculations. A good liquid density correlation should be accurate and
reliable over the whole liquid region from the freezing point to the critical point. In most practical cases, the
fluids of interest are mixtures.
Liquid densities can be calculated from equations of state (EOS). Normally, the accuracy of liquid density
predictions with EOS is not sufficient, and therefore special correlations for liquid density are used. Many
accurate correlations are available for saturated liquid densities of pure fluids. The accuracy of many models is
not good enough for mixtures, mainly due to problems in formulating the mixing rules.
A good discussion of saturated and compressed liquid density calculation methods can be found in Poling et al.
[14]. For density of saturated pure liquids, they recommend the Daubert et al. [8] version of the (modified
Rackett) equation proposed by Spencer and Danner [16]. That model is applicable only for pure compounds.
Reid et al. [15] recommend Spencer and Danner [16] and Hankinson-Thomson [9] (HT) models. They also
recommend compressed liquid densities to be calculated using the Hankinson-Brobst-Thomson (HBT) model
[17], in which HT is used for saturated liquid density. Poling et al. [14] recommend the model by Aalto et al. [1,
2, 3] for accurate compressed liquid density; the model is applicable from triple point up to the critical point for
pure compounds. The model by Aalto et al. [1,2,3] uses the HT model as basis and is applicable both for pure
compounds and mixtures to predict compressed liquid densities. The data used in general model parameter
fitting of that model was limited to 20 MPa as maximum pressure, and therefore systematic errors at extremely
high pressures are obtained. Later, Aalto and Keskinen [5,6] published an improved version of Aalto et al.
[1,2,3] model. The new model is applicable for pure components and liquid mixtures also for extremely high
pressures and gives better density prediction near critical temperature. The Aalto and Keskinen [5,6] model also
uses the HT model as the basis for saturated liquid density.
The saturated liquid density models available in literature for refrigerant systems were reviewed by Nasrifar and
Mosfeghian [12] and they found that the HT model was the most accurate among 18 methods tested. They state
that Iglesias-Silva and Hall [10] model is better for polar compounds and non-hydrocarbons than the HT model.
Later, Nasrifar et al. [11] have published a compressed liquid density calculation model which is based on the
use of the HT model as the basis for saturated liquid density.
Based on the above discussion it can be stated that one of the most successful saturated liquid density calculation
models is the Hankinson-Thomson model. We will review the HT model, what general and pure-component
model parameters are needed, where they are already available and how to the fit the parameters from
experimental data. Then we will review shortly some of the above mentioned compressed liquid calculation
methods based on the HT model.
Our aim is to determine the HT model parameters for as many pure compounds as possible, as well as some
parameters needed by the compressed liquid calculation methods based on the HT model. We give a listing of a
few new HT parameters determined this far. The necessary data for the parameter fitting, data treatment, factors
to be taken into account and the fitting procedure is discussed. Also, we will review some of HT model
deficiencies that might need improvement. Carrying out a large study of parameter fitting would point out some
problem areas and hints for the HT model improvement. Our HT model parameter collection will be available on
a web-site.

2 Hankinson-Thomson model for saturated liquid density

The saturated liquid molar volume, Vs, is computed in the Hankinson-Thomson model [9] from

= VR(0 ) 1 SRKVR( )

V R( 0) = 1 + a (1 Tr )

1/ 3

+ b(1 Tr )

2/ 3

+ c(1 Tr ) + d (1 Tr )

V R( ) = (e + fTr + gTr2 + hTr3 ) / (Tr 100001


4/ 3

, when 0.25< Tr < 0.95

when 0.25 < Tr

< 1.0


Tr =


Tc , HBT

For equations (2) - (3) the general parameters are the following:
a = -1.52816
b = 1.43907
c = -0.81446
d = 0.190454
e = -0.296123
f = 0.386914
g = -0.0427258 h = -0.0480645
The pure-component specific parameters are the following: characteristic volume V*, slightly adjusted critical
temperature Tc,HBT and the SRK-acentric factor SRK. These are tabulated for many compounds in Reid et al.
[15]. Reid et al. [15] recommended the use of these special parameters to get best accuracy. Hankinson and
Thomson [9] also give their original parameter collection. For refrigerant systems Aalto [4] has determined a set
of these parameters.
Although equation (2) was originally stated to apply only at range 0.25< Tr < 0.95, Aalto et al. [1,2,3], Aalto and
Keskinen [5,6] and also Nasrifar et al. [11] and Nasrifar and Mosfeghian [12] found it to hold reasonably well up
to the critical temperature.
Poling et al. [14] discuss the need for the special parameters V*, Tc,HBT and SRK. They state that real Tc can be
used in place of special Tc,HBT. We agree on this, but remind that the same value of Tc should be used with the
model that was used in the fitting of the V* and SRK. Poling et al. [14] also state that with a little loss of accuracy
V* can be set equal to Vc and correspondingly SRK can be set to the normal Riedel acentric factor . Later in this
paper we will see how this simplification affects on the error of density prediction.

3 Compressed liquid densities based on Hankinson-Thomson saturated liquid density

3.1 Hankinson-Brobst-Thomson model
Thomson et al. [17] proposed a method (HBT) to calculate compressed liquid density. This model first calculates
the saturated liquid density using equations (1) - (4). The effect of pressure is then taken into account using the
generalized Tait equation. This method needs the pure component vapor pressure (experimental value or from a
correlation equation like Antoine or Wagner equation) and the pure component critical pressure.
Aalto et al. [2] have shown that this generalized Tait equation has problems above reduced temperature 0.95.
Therefore this method is to be replaced with newer and more accurate methods described next.

3.2 Aalto et al. model

Aalto et al. [1,2] have presented a compressed liquid density calculation method that first calculates saturated
liquid density using HT model, and then the effect of pressure is taken into account by a modification of
equation of Chang and Zhao [7]. The Aalto et al. model uses the same parameters as the HT model: V*, Tc,HBT
and SRK. This model is quite accurate as discussed by Poling et al. [14]. The drawback of this model is that data
used in the general model parameter fitting was limited to 20 MPa as maximum. The model has been found to
have systematic errors (density underprediction) in pressures higher than 25 MPa and it is recommended only for
the pressure range it was initially fitted, i.e. maximum pressure for this model is 20 MPa.

3.3 Aalto and Keskinen model

Aalto and Keskinen [5,6] (AK) improved the model of Aalto et al. [1,2,3]. The data used in fitting the general
model parameters extended up to 800 MPa. A lot of data points were also located near the critical temperature.
Thus the AK model can be used with excellent accuracy up to extremely high pressures. The AK model also
applies with good accuracy to the critical temperature giving also the liquid compression at the critical
temperature when pressure is higher than critical pressure. The AK model was compared to the model of Aalto et
al. [1,2,3] and was found to be of nearly the same accuracy in the pressure range from saturation pressure to 20
MPa. Therefore the AK model is recommended over the model of Aalto et al. The AK model, using Vs from the
HT model, is as follows

A + C ( D Tr ) ( p r p s ,r )

V = Vs

A + C ( p r p s ,r )



A = a0 + a1Tr + a2Tr3 + a3Tr6 + a4 / Tr

B = b0 +

b2 + SRK


C = c1 (1 Tr ) 2 + 1 (1 Tr ) 2 exp[c3 + c4 ( pr p s ,r )]



The general model parameters for equations (5) - (8) are

a0 = 482.85416 a1 = -1154.2977 a2 = 790.09727 a3 = -212.14413
b0 = 0.0264002 b1 = 0.42711522 b2 = 0.5
a4 = 93.4904
c1 = 9.2892236 c2 = 2.5103968 c3 = 0.5939722 c4 = 0.0010895002
D = 1.00001
E = 0.80329503
Reduced temperature is obtained from equation (4), and reduced pressure pr and reduced pressure of saturated
vapor pr,s are

pr =

and p r , s =


The AK model uses the same pure component parameters as the HT model: V*, Tc,HBT and SRK. In addition to
that the critical pressure and saturated liquid vapor pressure are needed.

3.4 Nasrifar et al. Model

Recently, Nasrifar et al. [11] proposed a new compressed liquid density calculation model that is also based on
the HT saturated liquid density. They have published the necessary general model parameters. The pure
component parameters needed are Tc, pc and SRK. The model has also been extended to mixtures and then the
parameter V* is also required, and it is already required by the HT model. The pure component vapor pressure is
also required. The prediction of compressed liquid density also with this method looks quite accurate.

4 Density of saturated and compressed liquid mixtures

All the methods discussed above, HT for saturated liquid density [9], HBT for compressed liquid density [17],
Aalto et al. [1,2,3] for compressed liquid density, AK model for compressed liquid density [5,6] and Nasrifar et
al. [11] have been extended to mixtures using pseudocomponent approach: the pure component parameters are
combined with composition to form parameters for a pseudocomponent. Then the same model equations are
used as for pure compounds and as a result the molar volume and/or density of a mixture is obtained. There is
need to avoid all kinds of binary interaction parameters in the mixing rules. The selected mixing rules are used
without any binary interaction parameters. If binary interaction parameters are to be used, an exhaustive
parameter fitting has to be carried out using a lot of binary density measurements.
The pseudocomponent method has problems that are briefly discussed by Poling et al. [14]. The main problem is
that the vapor pressure of the pseudocomponent is obtained from the generalized Riedel equation, which does
not give the real bubble point curve of the mixture. Another problem is that the pseudocomponent critical point
is not the same (normally it has lower critical temperature and also pressure) as the real critical point, which is
more accurately predicted by using equations of state. Thus, there is normally a gap between the maximum
temperature handled by the models and the real critical temperature. Pokki et al. [13] have presented a more
comprehensive discussion of these problems and tested the use of EOS for obtaining the mixture parameters and
bubble point pressure.

5 Pure-component parameters for HT model and compressed liquid density models

The number of pure component specific parameters for different models should be kept as small as possible. The
pure component parameters used in different models should be: 1) used also in other models and 2) readily

available. Therefore we have chosen to use the real critical temperature Tc instead of Tc,HBT. It should be noted
that there is wide variation in the critical temperature values reported in the literature, especially for heavier
compounds. Thus, the same Tc value that is used in other parameter (V* and SRK) fitting should be used also
when applying the HT model.
Table 1: Some of the new HT parameters already determined in this project. Critical values from other sources.
Tc, K
V*, cm3/mol SRK
pc, MPa
Argon, Ar
Boron trichloride, BCl3






Boron trifluoride, BF3






Bromine, Br2






Nitrosyl chloride, NOCl






Chlorine, Cl2






Phosphorus trichloride, PCl3






Silicon tetrachloride, SiCl4






Deuterium, D2






Fluorine, F2






Nitrogen trifluoride, NF3






Silicon trifluoride, SiF4






Sulfur hexafluoride, SF6






Hydrogen iodide, HI






Hydrogen sulfide, H2S

Ammonia, NH3






Hydrazine, N2H4






Helium, He






Iodine, I2






n-Decane, C10H22






Example 1: The saturated liquid molar volume of n-decane at temperature 25 C is 195.95 cm3/mol (Poling et
al. [14]). The saturated liquid molar volume is calculated using the values from a) Poling et al. [14] as Tc,HBT = Tc
= 617.70 K, V* = Vc = 624.00 cm3/mol, and SRK = = 0.490, b) Reid et al. [15] Tc,HBT = 617.65 K, V* = 619.20
cm3/mol, and SRK = 0.4916, and c) new values in table 1 above Tc,HBT = Tc = 617.70 K, V* = 619.03 cm3/mol,
and SRK = 0.4918.
Results: Using equations (1) - (4) we get

T, K


VR( 0)

VR( )



Vs, cm3/mol

Error, %

Based on the example, it is justified to eliminate the use of separate Tc,HBT value since the use of real Tc gives
nearly identical values. Also, the Poling et al. [14] approximation is useful if separate V* and SRK values are not
available, but some accuracy is then lost, which is clearly seen by example 2.
Example 2: The saturated liquid molar volume of liquid iodine (I2) at temperature 453.2 K is 67.86 cm3/mol
(Poling et al. [14]). Using values Tc,HBT = Tc = 819.00 K, V* = Vc = 155.00 cm3/mol, and SRK = = 0.1115
(from vapor pressure correlation) from Poling et al. [14] we get Vs = 56.47 cm3/mol, which is -16.78 % in error!

The values from table 1 above (Tc,HBT = Tc = 819.00 K, V* = 184.79 cm3/mol, and SRK = 0.1058) yield Vs =
67.42 cm3/mol, which is only -0.65 % in error.

6 Procedure used for parameter fitting

The procedure for obtaining new HT model parameters has been discussed by Aalto [4] when new parameters
for refrigerants were fitted. See also Appendix B in Aalto and Keskinen [6]. Basically, parameter fitting is an
optimization problem where the error between measured and predicted values is minimized. Data selection and
evaluation (graphically and comparing different data sets) are of vital importance.

7 Discussion
The Hankinson-Thomson model is one of the most accurate and most general models for saturated liquid density
prediction. This method is used as a basis in many compressed liquid density calculation methods; saturated
liquid density is obtained first and then the effect of pressure is taken into account. Both the HankinsonThomson model and the compressed liquid density models require a set of special pure-component parameters:
V*, Tc,HBT and SRK. We have chosen to use the real critical temperature instead of Tc,HBT. The parametersSRK and
V* are fitted for each pure compound using pure component vapor pressure data (or vapor pressure correlation
equation like Antoine) and saturated liquid density data, respectively. For mixtures, the models use
pseudocomponent method with suitable mixing rules requiring pure component critical pressure also. No binary
interaction parameters are used. When the compressed liquid density is calculated, one should also have the pure
component vapor pressure available.
In our project, we have started to determine these parameters for a collection of pure compounds. The
determination of a large collection of pure-component parameters will probably point out problems in the HT
model and thus give us hints where the model can be improved. The parameter collection will be available at the
web-site (note small and capital letters):

8 References
[1] Aalto, M., Keskinen, K.I., Aittamaa, J., Liukkonen, S., Proceedings, The 4th Asian Thermophysical
Properties Conference (ATPC95), September 5-8, 1995, Tokyo, Japan.
[2] Aalto, M., Keskinen, K.I., Aittamaa, J., Liukkonen, S., Fluid Phase Equilibria 114 (1996) 1-19.
[3] Aalto, M., Keskinen, K.I., Aittamaa, J., Liukkonen, S., Fluid Phase Equilibria 114 (1996) 21-35.
[4] Aalto, M., Fluid Phase Equilibria 141 (1997) 1-11.
[5] Aalto, M., Keskinen, K.I., Proceedings, The 5th Asian Thermophysical Properties Conference (ATPC 98),
August 30 - September 2, 1998, Kim, M.S., Ro, S.T. (eds.), Seoul, Korea, 1998, 179-182.
[6] Aalto, M., Keskinen, K.I., Fluid Phase Equilibria 166 (1999) 183-205.
[7] Chang, C.-H., Zhao, X., Fluid Phase Equilibria 58 (1990) 231-238.
[8] Daubert, T.E., Danner, R.P., Sibel, H.M., Stebbins, C.C., Physical and Thermodynamic Properties of Pure
Chemicals: Data Compilation, Taylor&Francis, Washington, D.C., 1997.
[9] Hankinson, R.W., Thomson, G.H., AIChE J. 25 (1979) 653-663.
[10] Iglesias-Silva, G.A., Hall, K.R., Fluid Phase Equilibria 131 (1997) 97-105.
[11] Nasrifar, Kh., Ayatollahi, Sh., Mosfeghian, M., Fluid Phase Equilibria 168 (2000) 149-163.
[12] Nasrifar, Kh., Mosfeghian, M., Fluid Phase Equilibria 158-160 (1999) 437-445.
[13] Pokki, J.-P., Aalto, M., Keskinen, K.I., Remarks on computing the density of dense fluid, Paper presented in
the 9th International Conference of Properties and Phase Equilibria for Product and Process Design, PPEPPD
2001, May 20-25, 2001, Kurashiki, Japan, to be published.
[14] Poling, B.E., Prausnitz, J.M., O'Connell, J.P., The Properties of Gases and Liquids, 5th edition, McGrawHill, New York, 2000.
[15] Reid, R.C., Prausnitz, J.M., Poling, B.E., The Properties of Gases and Liquids, 4th edition, McGraw-Hill,
New York, 1987.
[16] Spencer, C.F., Danner, R.P., J.Chem.Eng.Data 17 (1972) 236-241.
[17] Thomson, G.H., Brobst, K.R., Hankinson, R.W., AIChE J. 28 (1982) 671-676.