Machine learning enables design automation of microfluidic flow-focusing droplet generation

Droplet-based microfluidic devices promise low-cost, high-throughput platforms for life science applications, yet adoption is hindered by the lack of predictive understanding of droplet generation, costly fabrication, and unreliable simulations. The authors present DAFD, a web-based tool that uses machine learning to predict and automate the design of flow-focusing droplet generators, achieving mean absolute errors under 10 µm for droplet diameter and 20 Hz for generation rate. The tool delivers user-specified performance within 4.2% (diameter) and 11.5% (rate) and is designed to be extensible via community-contributed data without requiring extensive ML expertise or large datasets, thereby reducing design iterations and facilitating broader use of microfluidics in the life sciences.

Ali Lashkaripour, Christopher Rodriguez, Noushin Mehdipour, Rizki Mardian, David McIntyre, Luis Ortiz, Joshua Campbell, Douglas Densmore