Data-Efficient Automatic Shaping of Liquid Droplets on an Air-Ferrofluid Interface with Bayesian Optimization

TL;DR

This paper presents a data-efficient, real-time method for shaping nonmagnetic liquid droplets into diverse forms at an air-ferrofluid interface using Bayesian optimization, surpassing previous limitations in shape complexity and adaptability.

Contribution

It introduces the first real-time, automatic shaping technique for nonmagnetic droplets into complex shapes using magnetic fields and Bayesian optimization.

Findings

Achieved shape errors as low as 0.81 mm.

Successfully shaped droplets into triangles, rectangles, and letter-like patterns.

Demonstrated real-time, adaptive control of droplet shapes.

Abstract

Manipulating the shape of a liquid droplet is essential for a wide range of applications in medicine and industry. However, existing methods are typically limited to generating simple shapes, such as ellipses, or rely on predefined templates. Although recent approaches have demonstrated more complex geometries, they remain constrained by limited adaptability and lack of real-time control. Here, we introduce a data-efficient method that enables real-time, programmable shaping of nonmagnetic liquid droplets into diverse target forms at the air-ferrofluid interface using Bayesian optimization. The droplet can adopt either convex or concave shapes depending on the actuation of the surrounding electromagnets. Bayesian optimization determines the optimal magnetic flux density for shaping the liquid droplet into a desired target shape. Our method enables automatic shaping into various triangular and rectangular shapes with a maximum shape error of 0.81 mm, as well as into letter-like patterns. To the best of our knowledge, this is the first demonstration of real-time, automatic shaping of nonmagnetic liquid droplets into desired target shapes using magnetic fields or other external energy fields.