Combining mixing rules with QSPR models for pure chemicals to predict the flash points of binary organic liquid mixtures

Abstract

Flash point is a key property of liquids to evaluate the safety of industrial processes. Mixing rules are commonly used to calculate the flash point of liquid mixtures, but they need knowledge of the ones of pure compounds. Theoretical methods notably based on quantitative structure property relationships (QSPR) already exist to predict flash points of pure compounds. So, in this paper, direct combination of these two types of approaches was investigated to achieve predictions even when the flash points of pure compounds were unknown. Three relevant mixing rules and four QSPR models, based on simple constitutional descriptors, were considered. Based on a data set of 284 experimental data of binary mixtures extracted from literature, two reliable combinations were highlighted. The most accurate one reached an error in prediction of only 2.9°C but needed knowledge of the boiling point and Antoine's coefficients of each component of the mixture. A new full-predictive method was in particular proposed with also a low error in prediction (4.4°C), requiring only knowledge of the molecular structure of each pure compound and molar fraction of the mixture. Errors in each predictive method keep quite reasonable against expected accuracies of direct measurements of flash point of binary mixtures.