Tracking droplets in soft granular flows with deep learning techniques

Area of Science:

• Fluid Dynamics
• Computational Science
• Artificial Intelligence

Background:

• Analyzing complex fluid flows with many moving objects, such as emulsions and soft crystals, is challenging.
• Traditional methods for tracking droplets in simulations are often labor-intensive and slow.

Purpose of the Study:

• To develop an accurate and efficient deep learning-based method for analyzing fluid dynamic simulations.
• To track moving droplets in complex multi-core emulsions and soft flowing crystals.
• To enable low-cost, high-speed analysis of systems with numerous moving objects.

Main Methods:

• Combined You Only Look Once (YOLO) for object recognition and DeepSORT for object tracking algorithms.
• Trained the YOLO network using synthetically prepared data to recognize droplets.
• Applied the trained YOLO + DeepSORT procedure to digital images from fluid dynamic simulations.

Main Results:

• The YOLO + DeepSORT procedure achieved high accuracy in tracking droplet trajectories in fluid simulations.
• Low error levels were observed when comparing inferred trajectories to independently computed ground truth.
• The developed application analyzes data at speeds exceeding typical digital camera acquisition rates (30 fps) on desktop GPUs.

Conclusions:

• The YOLO + DeepSORT method provides a low-cost, practical tool for studying systems with many moving objects.
• This approach facilitates the automatic extraction of equations of motion for many-body soft flowing systems.
• The study demonstrates the potential of deep learning for advancing fluid dynamics simulation analysis.