The jumping direction is an essential characteristic of jumping droplets, but it is poorly understood and uncontrollable at present. In this work, we present a method to control the jumping direction by surface structures, where the jumping direction is controlled by changing the inclination angle of the structure. The underlying mechanism is analyzed experimentally, with numerical simulations, and using a theoretical model developed to relate the jumping direction and the inclination angle for a few cases with a specific distribution. Because random droplet distributions are more common on actual condensation surfaces, a more comprehensive prediction model was developed based on a convolution neural network (CNN) to predict the jumping direction for more general cases. The input to the CNN is an image of droplets with various distribution features, which are detected by the neural network and used to predict the jumping angle. SHapley Additive exPlanations methods were then used to analyze the feature importance and to give human-understandable insights from the prediction model. This work offers avenues for improving cooling rates, anti-icing/freezing characteristics, and self-cleaning attributes and contributes to a better understanding of the jumping direction.
Keywords: coalescence-induced droplet jumping; convolution networks; jumping direction; machining learning; superhydrophobic surfaces.