Autonomous in silico optimization framework for high-performance micromixers

Effective mixing at the microscale is essential for lab-on-a-chip systems, yet designing micromixers that achieve both high mixing efficiency and low pressure drop remains challenging and resource-intensive. Here, we introduce an autonomous in silico framework for designing obstacle-based micromixers through integrating 3D geometry generation, computational fluid dynamics (CFD) simulations and a multi-objective artificial intelligence optimization algorithm within a fully automated close-loop workflow. A constraint-aware NSGA-II variant is used, incorporating a repair operator to ensure design feasibility. Experimentally validated against hyperspectral imaging-based mixing characterization and pressure-drop measurements, the framework eliminates manual trial-and-error workload and alleviates researchers from the tedious tasks of navigating across 3D modeling, CFD simulations, and optimization algorithms, reducing optimization time by 48% compared to the conventional simulation-assisted approach. By autonomously screening hundreds of designs, it identifies Pareto-optimal micromixers and generates an extensive database that supports inverse design and reveals the mixing structure-performance relationship, facilitating the establishment of general design guidelines. The framework is generally applicable to a wide range of passive micromixers. • An autonomous in silico framework reduces micromixer optimization time by 48%. • A hyperspectral-assisted microfluidic platform is developed for experimental validation. • The resulting dataset enables inverse design and derivation of design guidelines. • Stretching interfaces and reducing vortices improve obstacle micromixer performance at low Re .