A physics-based and data-driven approach to regime mapping ultra-low interfacial tension microfluidic flows

Droplet microfluidics has several important biomedical applications, such as cell encapsulation and drug delivery. Successful commercial implementation requires a high degree of bio-compatibility. Commonly used water–oil systems are toxic to cells, hindering successful translation toward these applications. One non-toxic alternative for microfluidic systems is an Aqueous Two-Phase System (ATPS) made from dissolving poly(ethylene glycol) and dextran in water. However, ATPS’s ultra-low interfacial tension makes experimental exploration of its flow regimes challenging. Typical microfluidic flow regime mapping approaches rely heavily on qualitative observations of droplet formation constructed from a limited sample of experimental or simulation data points, resulting in coarsely interpolated regime boundaries. In this study, we develop a Computational Fluid Dynamics model to study ultra-low interfacial tension microfluidic flows. Using the Volume of Fluid multiphase solver, a total of 130 simulations are conducted, sweeping across a wide range of capillary numbers and flow rate ratios. Using a statistically informed mapping approach, simulation results are statistically analyzed and classified into five key regimes: squeezing, dripping, transitional, jetting, and threading. Droplet size distributions are found to highlight and distinguish transitional and jetting regimes. Machine Learning algorithms are trained on simulation data and predict regimes from a synthetic dataset containing 62 500 points, producing a densely populated map with refined regime boundary predictions, showcasing a method to construct high-resolution mapping for microfluidic flow regimes with a limited number of simulations or experiments.