Improved Pillar Array Design Using Bayesian Optimization

In recent years, machine learning has been increasingly applied across various scientific fields, including microfluidics. This study presents a novel method for enhancing the performance of microfluidic systems for liquid chromatography. In particular, Bayesian optimization is combined with computational fluid dynamics and Brenner’s homogenization theory to compute the axial liquid dispersion in periodically repeating structures. The proposed approach was used to optimize the shape and arrangement of perfectly ordered pillar arrays, with the aim of minimizing their minimal chromatographic separation impedance E_min. This metric depends on both the degree of dispersion (quantified using an axial dispersion coefficient or plate height) and the permeability of the structure, which are in general contradicting demands. Limiting the optimization to pillar arrays with the same external porosity (ε = 50%), it has been found that, compared to the most commonly used pillar array in the literature (lattice angle α = 60^◦, circular pillars), E_min can be reduced by 25% by adjusting the lattice angle to 79^◦ for circular pillars, and that E_min can be reduced by 35% by adjusting the lattice angle to 50^◦ for elliptical pillars with a width-to-length ratio of 1.710.