Mixing characterization of binary-coalesced droplets in microchannels using deep neural network

Area of Science:

• Microfluidics
• Computer Vision
• Machine Learning

Background:

• Real-time object identification and classification are crucial for microfluidic applications, particularly in droplet microfluidics.
• Monitoring droplet behavior, such as merging and mixing, is essential for process control and analysis.

Purpose of the Study:

• To apply convolutional neural networks (CNNs) for real-time detection and classification of merged microdroplets.
• To categorize merged droplets based on their mixing extent (low, intermediate, high).
• To evaluate the performance of CNNs in diverse experimental conditions.

Main Methods:

• Droplet generation in polydimethylmethylacrylate (PMMA) microfluidic devices using flow-focusing and cross-flow configurations.
• Image acquisition using a CCD camera attached to a microscope.
• Deployment of real-time object detection networks like You Only Look Once (YOLO) and Single Shot Multibox Detector (SSD).
• Creation and manual labeling of a custom dataset for training deep neural networks.

Main Results:

• Trained CNN models demonstrated high accuracy and precision in detecting and classifying merged microdroplets.
• The models successfully categorized droplets into low, intermediate, and high mixing levels.
• Performance remained robust across various ambient conditions, droplet characteristics, and fluid combinations.

Conclusions:

• CNN-based schemes are efficient for real-time localization and classification of coalesced binary droplets.
• The developed method is effective regardless of experimental variations, offering a reliable tool for microfluidic analysis.