Geometry-free learning predicts inertial lift forces in...

Why does this matter? Inertial microfluidic platforms promise cheap, rapid handling of particles and cells, yet their design hinges on accurately forecasting the forces that steer objects through tiny channels. Traditionally, engineers rely on computational fluid dynamics, a process that can be both time‑consuming and geometry‑specific. Recent attempts to inject machine learning into the workflow have cut run times dramatically, but each new channel shape—whether rectangular, triangular or otherwise—has demanded its own dedicated model, pushing the training burden onto researchers.

Here’s the thing: a new study sidesteps that requirement entirely. By feeding a neural network a novel set of parameters that omit any explicit description of the channel cross‑section, the authors achieve performance on familiar geometries that matches earlier approaches, while also handling previously unseen designs with far less degradation. The model slots into existing particle‑tracing software, reproducing migration patterns reported in the literature across a range of layouts. In short, the work suggests a path toward more flexible, reusable predictive tools for microfluidic engineering.