Dynamic Analysis of Membrane Emulsification: An In Situ Observation and Calculation with Deep Learning Model

Membrane emulsification, an energy-efficient technology, faces application challenges such as membrane parameters optimization (e.g., pore size–porosity synergy) and unclear dynamic mechanisms. Existing studies lack in situ observation of formation processes, hindering correlation between pore structures and emulsion performance. In this study, a novel in situ observation device was developed, which couples a visible membrane module with a high-speed microscope system. Additionally, a deep learning model was embedded to realize the first visualization of the dynamic droplet formation process under demanding conditions. The trained model achieved a precision of 0.945, a recall of 0.998, and a mAP@0.5 of 0.989, enabling fully automated extraction of microdroplet size and size distribution. Using this device, we demonstrated that decreasing the pore size or increasing the dispersed phase flow rate induces droplet formation transition from dripping to jetting, with the Weber number (Wed) quantifying this critical condition (Wed = 16 ± 12%). Pore spacing significantly affects droplet formation dynamics in the dripping mode but exerts minimal impact on those in the jetting mode. This study provides theoretical and technical insights for optimizing the design and processes of membrane emulsification.