DROPLET GENERATION DYNAMICS IN MICROFLUIDIC SYSTEMS: A COMPARATIVE STUDY OF MACHINE LEARNING APPROACHES FOR NON-NEWTONIAN FLUIDS
The formation of droplets in non-Newtonian multiphase systems is fundamental in industries such as pharmaceuticals, biotechnology, food processing, and materials science. This research examines droplet formation in two microfluidic systems: a co-flow system with Xanthan Gum (XG) and mineral oil and a flow-focusing system with Hyaluronic Acid (HA), mineral oil, and surfactants. The study analyzes how fluid properties, flow conditions, and system geometry affect droplet size, shape, and frequency.
A key aspect is the shear-thinning behavior of non-Newtonian fluids, which decreases viscosity at higher shear rates. Experimental findings indicate that lower Capillary numbers (Ca) lead to larger droplets due to the dominance of viscous forces. In the flow-focusing system, surfactants significantly influence interfacial tension and droplet stability.
To enhance the analysis, machine learning models such as Neural Networks (NN), Random Forest (RF), and Bayesian Machine Scientist (BMS) were used. While NN and RF provided insights, the BMS model achieved the highest accuracy (R² = 0.98 for droplet length and 0.97 for height), delivering interpretable equations linking flow rates, viscosity, and droplet size.
These findings contribute to optimizing microfluidic systems in biomedical applications and pharmaceutical production. Future studies could explore other non-Newtonian fluids and microfluidic geometries for improved control. This research demonstrates that combining experiments with machine learning enables precise control over droplet-based technologies, offering innovative solutions for industrial and biomedical applications.
Download:
