Machine-Learning Approach in Prediction of the Wettability of a Surface Textured with Microscale Pillars

Tuning the wettability of a flat surface by introducing an array of microscale pillars finds wide applications, especially in engineering a superhydrophobic surface. The wettability of such a pillared surface is quantified by the contact angle (CA) of a water droplet. It is desired to know the CA prior to construction of pillars, in order to obviate the trial-and-errors in experimenting with many different topographies. Given an accurate theoretical prediction of CA has been elusive, we propose a convolutional neural network (CNN) model of CA for a surface patterned with rectangular or cylindrical pillars. By employing a three-dimensional descriptor of the surface topography, the present CNN model can predict experimental CAs within errors comparable to the uncertainties in measuring CAs.