A Physically Consistent AI Framework for Predictive Design of Microfluidic Double Emulsions Toward Precision Liposome Fabrication

Microfluidic double emulsions offer a powerful route for templated liposome fabrication; however, precise and reproducible control over droplet architecture remains challenging due to the highly nonlinear coupling between flow dynamics and interfacial phenomena. Here, we present a physics-consistent, data-driven framework that combines artificial neural networks (ANNs) with genetic algorithms (GAs) to enable predictive inverse design and target-driven optimization of double-emulsion generation. By systematically varying lipid concentration, surfactant concentration, and flow-rate parameters within a flow-focusing microfluidic platform, the ANN model learns the underlying relationships governing droplet formation with near-unity predictive accuracy. Coupling the trained model with a GA enables precise control of droplet morphology, demonstrated here by the generation of droplets with a targeted 30 µm core diameter and 5 µm shell thickness, resulting in liposomes with diameters of approximately 30 µm after solvent extraction. Experimental results closely matched the model predictions for both the selected processing conditions and the resulting droplet properties, confirming the reliability of the proposed framework. The flow-related parameters had the greatest impact on droplet diameter and production frequency. This study demonstrates a practical, interpretable AI-guided optimization approach for complex microfluidic platforms, while offering a scalable route for the fabrication of liposomes and other structured soft-material systems.