Development of an improved group contribution method for the prediction of vapour pressures of organic compounds

Abstract

Vapour pressure is an important property in the chemical and engineering industries. There are therefore many models available for the modelling of vapour pressure and some of the popular approaches are reviewed in this work. Most of the more accurate methods require critical property data and most if not all require vapour pressure data in order to regress the model parameters. It is for this reason that the objective of this work was to develop a model whose parameters can be predicted from the molecular structure (via group contribution) or are simple to acquire via measurement or estimation (which in this case is the normal boiling point). The model developed is an extension of the original method that was developed by Nannoolal et al. The method is based on the extensive Dortmund Data Bank (DDB), which contains over 180 000 vapour pressure points (for both solid and liquid vapour pressure as of 2007). The group parameters were calculated using a training set of 113 888 data points for 2332 compounds. Structural groups were defined to be as general as possible and fragmentation of the molecular structures was performed by an automatic procedure to eliminate any arbitrary assumptions. As with the method of Nannoolal the model only requires knowledge about the molecular structure and the normal boiling point in order to generate a vapour pressure curve. In the absence of experimental data it is possible to predict the normal boiling point, for example, by a method developed by Nannoolal et al. The relative mean deviation (RMD) in vapour pressure was found to be 5.0 % (2332 compounds and 113 888 data points) which compares very well with the method of Nannoolal et al. (6.6 % for 2207 compounds and 111 757 data points). To ensure the model was not simply fitted to the training set a test set of liquid vapour pressure, heat of vaporization and solid vapour pressure data was used to evaluate its performance. The percentage error for the test set was 7.1 % for 2979 data points (157 compounds). This error is artificially high as the test data contained a fair amount of less reliable data. For the heat of vaporization at 298.15 K (which is related to vapour pressure via the Clausius-Clapeyron equation) the RMD was 3.5 % for 718 compounds and in the case of solid vapour pressures the RMD error was 21.1 % for 4080 data points (152 compounds). Thus the method was shown to be applicable to data that was not contained in the training set.