Fig. 6: Trained machine learning models and custom search algorithms enable design automation of SE and DE droplet generation.
From: Design automation of microfluidic single and double emulsion droplets with machine learning
a Design automation of SE droplet generation. (i.) DAFD 3.0 takes user-specified diameter, rate, fluid properties, and optional constraints as inputs and returns the necessary geometry and flow rates required to generate the desired droplets. (ii.) DAFD-predicted and measured droplet diameters after specifying desired SE diameters of 25, 30, and 35 μm for an unseen fluid combination (left) and representative images of generated droplets (right). Measured droplets differed from specified droplets by a MAE of 2.3 μm (MAPE of 7.9%). Bars indicate average measured diameter and error bars' total length represents two standard deviations in diameter across 10 measured droplets. b Design automation of DE droplet generation. (i.) DAFD 3.0 also converts user-specified DE inner and outer diameters to the necessary geometries and flow rates required to generate them. (ii.) DAFD-predicted generation rates as a function of middle, inner, and outer flow rates are used to predict generation rate differences (GRDs) between FF1 and FF2 to identify likely stable (GRD <5%) and unstable (GRD >5%, gray shaded areas) regimes. (iii.) Comparison between observed and DAFD-specified DE inner (blue) and outer (orange) diameters for an unseen fluid combination and 9 different flow rates (left); images show representative DE droplets generated under each condition (right). For stable droplets, observed inner and outer diameters differed from those specified by an MAE of 2.7 μm (MAPE of 6.3%). Experiments were carried out once. Two sets of droplets generated 5 min apart were analyzed to report the mean diameter. Scale bars represent 50 μm. Source data are provided as a Source Data file.
