Prediction of the flash points of organic mixtures by QSPR models
Résumé
Promoted by REACH regulation and by the development of computational capabilities, QSPR are increasingly used for the prediction of physico-chemical properties of pure compounds. If substances used in industrial processes are mainly (complex) mixtures, only few QSPR models are devoted to mixtures. So, QSPR models for mixtures would represent very useful tools at R&D level to help the development of new chemical compositions, the selection of substances in a new industrial process or the estimation of hazards (involved by mixtures) in process safety issues. Flash point of organic mixtures characterizes flammability hazards of liquids and is a key safety issue in the risk assessment of industrial processes. In particular, flammability can be dramatically changed upon concentrations in flammable compounds following, in many cases, non-linear trends. For this reason, the prediction of mixtures flash point represents a great challenge. Most of existing QSPR models for the prediction of flash points dedicated to pure compounds require the knowledge of the boiling point. To address mixtures, mixing rules are commonly used but they need knowledge of the flash point of each pure component. To access reliable prediction of the flash point for mixtures of flammable liquids, a robust mixing rule was combined with flash points predicted using QSPR models for the pure compounds that reached reliable predictions with a MAE of 4.4K evaluated on 284 experimental data of binary mixtures. Another developed approach was to derive new QSPR models directly dedicated to mixtures based on descriptors adapted for the case of mixtures.