(365b) Convolutional Neural Network Augmented Soft-Sensor for Autonomous Microfluidic Production

In this study, we introduce an AI-driven soft sensor feedback control system designed for the production of bubbles, droplets, and hydrogel rods in microfluidic environments. Precise control of microfluidic bubbles and droplets is crucial in fields such as pharmaceuticals, chemical synthesis, and DNA sequencing. However, achieving consistent size, shape, and functionality in these microscale entities is challenging because unpredictable factors that can disrupt microfluidic processes, thus requiring constant user-input to maintain setpoints such as size, shape, and frequency of generation.

To address this, we propose a soft-sensor approach that utilizes an object detection convolutional neural network (CNN) for both image classification and feature extraction. This method enables the detection of flow regimes and the assessment of bubble and droplet size, shape, and uniformity. The soft sensor, integrated with a controller, demonstrates effective setpoint tracking, disturbance rejection, and long-term stability.

Our research presents a robust methodology for precise and automated microfluidic control using soft-sensor and AI-driven feedback, applicable to a wide range of applications.