Prediction of the Lotus Effect on Solid Surfaces by Machine Learning

Superhydrophobic surfaces with the “lotus effect” have wide applications in daily life and industry, such as self-cleaning, anti-freezing, and anti-corrosion. However, it is difficult to reliably predict whether a designed superhydrophobic surface has the “lotus effect” by traditional theoretical models due to complex surface topographies. Here, a reliable machine learning (ML) model to accurately predict the “lotus effect” of solid surfaces by designing a set of descriptors about nano-scale roughness and micro-scale topographies in addition to the surface hydrophobic modification is demonstrated. Geometrical and mathematical descriptors combined with gray level cooccurrence matrices (GLCM) offer a feasible solution to the puzzle of accurate descriptions of complex topographies. Furthermore, the “black box” is opened by feature importance and Shapley-additive-explanations (SHAP) analysis to extract waterdrop adhesion trends on superhydrophobic surfaces. The accurate prediction on as-fabricated superhydrophobic surfaces strongly affirms the extensionality of the ML model. This approach can be easily generalized to screen solid surfaces with other properties.