Acoustic-based microdevice for biomedicine applications: Comprehensive parametric analysis, predictive surrogate modeling, and machine learning-based optimization

Acoustofluidic micromixers are critical tools for biomedical applications due to their ability to achieve efficient, non-contact mixing at microscales. In this study, an acoustic-based micromixer incorporating sharp triangular edges is comprehensively modeled, analyzed, and optimized in an application-based manner. The process begins with parametric geometry design of the micromixer, incorporating five dimensionless parameters: edge height (h*), edge angle (α*), spacing between edges (d*), vertical offset (ϕ), and tip curvature radius (rc*). A detailed finite-element model is developed in COMSOL Multiphysics, integrating fluid flow, acoustics, and species transport under acoustic streaming in the frequency domain based on perturbation theory. A central composite design-based response surface methodology is then employed to develop a surrogate model for the mixing index (MI) and pressure drop (ΔP). A total of 152 simulations are conducted to build statistically significant models (R2 = 0.9590, F = 33.25 for MI), and the results are validated against the experimental literature. Subsequently, a Multi-Objective Particle Swarm Optimization algorithm is applied to identify Pareto-optimal designs that balance mixing efficiency with pressure losses. Key findings include a maximum mixing index exceeding 96%, with pressure drops in optimal solutions maintained between 1 and 4 Pa. Increasing both the applied voltage and the edge height each yielded over 70% improvement in mixing. The robust, validated framework demonstrates strong potential for the intelligent design of bulk acoustic-based devices in clinical and pharmaceutical micro-systems.