The use of neural network modeling for the estimation of the Hansen solubility parameters of polymer films from contact angle measurements

Abstract

A neural network model is proposed to predict the Hansen Solubility parameter (HSP) of a polymer / solid surface via simple contact angle measuring using only three different types of liquids (solvents) with known physical properties and HSP. A set of 70 data points of contact angle measurements on five different polymer films using fourteen different liquids (solvents) were used to train the neural network model. The trained model predicts the HSP distance between the different solvents and the solid surfaces. The predicted distance was further used to estimate the total and individual HSP parameters of the tested solid surface. The type of solvents/ liquid used in the sessile drop measurements used in the contact angle measurements can affect the accuracy of the estimation of the HSP even using the same trained network. It was found that different structure of the neural network model can affect the robustness of the results. Therefore, the neural network was trained further for better accuracy. The two accuracy indicators were used in this work are the R2 (Pearson Product-Moment Correlation Coefficient) and the root mean square error (RMSE). A network with three hidden layers and 20 neurons each were found to overcome the sensitivity in the type of liquid/ solvent selected in the contact angle measurements. Nevertheless, it is recommended to use the following specific solvents in the sessile drop measurements on any type of solid film, for the estimation of HSP: ethylene glycol, formamide, and dimethylformamide. The inputs to the robust neural network are the contact angle of each liquid on the tested film, the surface tension and the viscosity of the liquid used.