## Development and critical evaluation of group contribution methods for the estimation of critical properties, liquid vapour pressure and liquid viscosity of organic compounds.

##### Abstract

Critical properties, liquid vapour pressures and liquid viscosities are important thermophysical properties required for the design, simulation and optimisation of chemical plants. Unfortunately, experimental data for these properties are in most cases not available. Synthesis of sufficiently pure material and measurements of these data are expensive and time consuming. In many cases, the chemicals degrade or are hazardous to handle which makes experimental measurements difficult or impossible. Consequently, estimation methods are of great value to engineers. In this work, new group contribution methods have been developed for the estimation
of critical properties, liquid vapour pressures and liquid viscosities of non-electrolyte organic compounds. The methods are based on the previous work of Nannoolal (2004) & Nannoolal et al. (2004) with minor modifications of structural group definitions. Critical properties, viz. critical temperature, critical pressure and critical volume, are of
great practical importance as they must be known in order to use correlations based on the law of corresponding states. However, there is a lack of critical property data in literature as these data are difficult or in many cases impossible to measure. Critical property data are usually only available for smaller molecules of sufficient thermal stability. The proposed group contribution method for the estimation of critical properties reported an average absolute deviation of 4.3 K (0.74%), 100 kPa (2.96%) and 6.4 cm3.mol1 (1.79%) for a set of 588 critical temperatures, 486 critical pressures and 348 critical volumes stored in the Dortmund Data Bank (DDB (2006)), respectively. These results were the lowest deviations obtained when compared to ten well known estimation methods from literature. In addition, the method showed a wider range of applicability and the lowest probability of prediction failure and leads to physically realistic extrapolation when applied to a test set of components not included in the training set. For the estimation of the critical temperature using the new method, knowledge about
the normal boiling point is required. If there is no information on the latter property, then the previous group contribution estimation method can be employed for estimation. Because of their great importance in chemical engineering, liquid vapour pressures have received much attention in literature. There is currently an abundance of experimental data for vapour pressures, especially for smaller molecules, but data are scarce or of low quality for larger and more complex molecules of low volatility. The estimation of liquid vapour pressures from molecular structure has met with very limited success. This is partly due to the high quality predictions required for vapour pressures for use in the design of for example distillation columns. This work presents a new technique for the estimation of liquid vapour pressures by developing a two-parameter equation where separate parameters model the absolute value and slope while at the same time the equation is able to approximate the nonlinearity of the curve. The fixed point or absolute value chosen was the normal boiling point for which a large amount of experimental data is available. A group contribution estimation of the slope was then developed which showed nearly no probability of
prediction failure (high deviation). Employing experimental normal boiling points in the method, an absolute relative deviation of 6.2% in pressure for 1663 components or 68835 (68670 from DDB and 165 from Beilstein) data points was obtained. This result is in comparable accuracy or slightly higher in deviation than correlative models such as the Antoine and DIPPR equations (direct correlations). A test of the predictive capability by employing data that were not used in the training set also showed similar results. Estimations are possible up to the inflection point or a reduced normal boiling temperature of ±1.2. If there is no information about the experimental normal boiling point, two options are recommended to obtain this value. The first and more reliable is back-calculation using the known boiling point at other pressures and the estimated slope of the vapour pressure equation. Results in this case are similar to cases where experimental normal boiling points were used. The second possibility is to estimate the normal boiling point using the method developed previously. In this case, an absolute relative deviation of 27.0% in pressure is obtained. The saturated liquid viscosity is an important transport property that is required for
many engineering applications. For this property, experimental data are limited to mostly simple and more common components and, even for these components the data often cover only a small temperature range. There have been many different approaches to estimate liquid viscosities of organic compounds. However, correlative and empirical methods are often the only or preferred means to obtain liquid viscosities. The technique used for the estimation of the liquid viscosity is similar to that in case of liquid vapour pressures, i.e. a two-parameter equation models the absolute value, slope and the non-linearity of the curve. As there was no convenient reference point at a standard viscosity available to model the absolute value (viscosity reference temperature), an algorithm was developed to calculate this temperature which was
chosen at a viscosity of 1.3 cP. This work then presents a group contribution estimation of the slope and using calculated or adjusted reference temperatures, an absolute relative deviation of 3.4% in viscosity for 829 components or 12861 data points stored in the DDB was obtained. This result is in comparable accuracy or slightly higher in deviation than correlative models such as the Andrade and Vogel equations (direct correlations). The estimation method has an upper temperature limit which is similar to the limit in case of liquid vapour pressures. If no data are available for a viscosity close to 1.3 cP then, as in case of the vapour pressure estimation method, the temperature can be back calculated from data at other viscosity values. Alternately, the viscosity reference temperature can be estimated by a group contribution method developed in this work. This method reported an average absolute deviation of 7.1 K (2.5%) for 813 components. In case both the slope and absolute value were estimated for the liquid viscosity curve, an average absolute
deviation of 15.3 % in viscosity for 813 components or 12139 data points stored in the DDB was obtained. The new method was shown to be far more accurate than other group contribution methods and at the same time has a wider range of applicability and lower probability of prediction failure. For the group contribution predictions, only the molecular structure of the compound is used. Structural groups were defined in a standardized form and fragmentation of the molecular structures was performed by an automatic procedure to eliminate any
arbitrary assumptions. To enable comparison, chemical family definitions have been developed that allow one to automatically classify new components and thus inform the user about the expected reliability of the different methods for a component of
interest. Chemical family definitions are based on the kind and frequency of the different structural groups in the molecule.