# Investigation of Droplet Spreading and Rebound Dynamics on Superhydrophobic Surfaces Using Machine Learning

## TL;DR

This paper uses machine learning to study how droplets spread and bounce on superhydrophobic surfaces, revealing new insights into the factors that influence these behaviors.

## Contribution

The novel contribution is the use of machine learning to analyze droplet impact dynamics and derive empirical correlations that outperform existing models.

## Key findings

- Droplet impact velocity is the dominant factor in spreading and rebound dynamics.
- Droplet spreading is independent of surface microtopography but rebound efficiency is influenced by the unwetted area fraction.
- New empirical correlations for maximum spreading coefficient and rebound efficiency were developed and shown to outperform existing models.

## Abstract

The spreading and rebound of impacting droplets on superhydrophobic interfaces is a complex phenomenon governed by the interconnected contributions of surface, fluid and environmental factors. In this work, we employed a collection of 1498 water–glycerin droplet impact experiments on monolayer-functionalized laser-structured aluminum samples to train, validate and optimize a machine learning regression model. To elucidate the role of each influential parameter, we analyzed the model-predicted individual parameter contributions on key descriptors of the phenomenon, such as contact time, maximum spreading coefficient and rebound efficiency. Our results confirm the dominant contribution of droplet impact velocity while highlighting that the droplet spreading phase appears to be independent of surface microtopography, i.e., the depth and width of laser-made features. Interestingly, once the rebound transitions to the retraction stage, the importance of the unwetted area fraction is heightened, manifesting in higher rebound efficiency on samples with smaller distances between laser-fabricated microchannels. Finally, we exploited the trained models to develop empirical correlations for predicting the maximum spreading coefficient and rebound efficiency, both of which strongly outperform the currently published models. This work can aid future studies that aim to bridge the gap between the observed macroscale surface-droplet interactions and the microscale properties of the interface or the thermophysical properties of the fluid.