Autonomous Bayesian Optimization-Based Control System for Droplet Generation

Abstract

Droplet generation has been utilized in various applications, including drug delivery, the fabrication of functional particles, and material synthesis. Achieving the goals of these applications requires droplet generation of a desired size. Microfluidic droplet generation offers precise control of droplet dimensions. However, the flow rates of the droplet and medium phases depend on the channel configuration and properties of the working fluids, and their optimization is a labor-intensive and time-consuming process. To overcome these limitations, an autonomous Bayesian optimization (BO)-based control system for droplet generation (ABCD) is developed. In ABCD, BO is employed to inform decision-making and refine the experimental conditions. Additionally, computer vision techniques, including image processing and convolutional neural networks, are utilized to analyze the experimental results and provide a dataset for use in decision-making. The ABCD identified the optimal flow rates to achieve desired droplet sizes and generation frequencies via precise and efficient searches, regardless of the droplet generation target, working fluids, channel geometry, and droplet morphology, within 15 iterations on average. It is anticipated that this system will contribute to the acceleration of research utilizing droplet-based microfluidic systems while also extending microfluidic process automation in various industrial applications.