U-Net-Based Surrogate Model for Evaluation of Microfluidic Channels

Microfluidics have shown great promise in multiple applications, especially biomedical diagnostics and separations. While the flow properties of these microfluidic devices can be solved by numerical methods such as computational fluent dynamics (CFD), process mesh generation setting up a solver requires some domain familiarity, while more intuitive commercial programs StarCCM expensive. Hence, this work, we demonstrated use U-Net convolutional neural network surrogate model for predicting velocity pressure fields that would result particular set filter designs. The is fast, easy to set-up used predict assess across new designs interest via input geometry-encoding matrix. In addition, demonstrate same methodology also train based on data, propose an alternative algorithms calculating measurements from particle image velocimetry measurements. Critically, both prediction test errors less than 1%, suggesting indeed viable method.