Molecular interactions consideration using Hansen solubility parameters in a multilayer perceptron artificial neural network for flash point prediction of organic liquid mixtures

The flash point temperature, or simply the flash point (FP), is the most significant thermophysical property of organic components and must be calculated precisely to handle them safely. In this work, a single hidden layer multilayer perceptron artificial neural network (MLPANN) was developed using only the Hansen solubility parameters (HSPs) of organic solvents and their concentrations in the mixture to be as straightforward as possible. This is the first time that HSPs have been used to calculate the FP of components. The potential of the proposed ANN in calculating the FP of organic liquid mixtures was thoroughly examined using numerous complex ternary mixtures, including methanol-2,2,4-trimethylpentane-toluene, methanol-decane-acetone, 1-butanol-acetic acid-ethylbenzene, 2-pentanol-acetic acid-ethylbenzene, 1-butanol-acetic acid-propyl butyrate, 2-pentanol-acetic acid-propyl butyrate, isopropyl alcohol-ethanol-octane, and 2-butanol-ethanol-octane. The unusual minimum and maximum flash point behaviors, resulting from significant differences in intermolecular forces among the components in the mixtures, were properly represented by the ANN-based model.

Furthermore, based on the obtained results, the coefficient of determination (R²) for the training, validation, and testing samples was 0.999, 0.998, and 0.972, respectively, and 0.997 for all data. Additionally, the root-mean-square deviation (RMSD) for the training, validation, and testing samples was 0.34 °C, 0.81 °C, and 1.8 °C, respectively, and 0.83 °C for all data.