Intelligent micromixer design via a synergistic framework: concave-star geometry exploration with prior-data-fitted neural networksCoal Mining Research Institute, China Coal Technology and Engineering Group Co. Ltd., Beijing, China
E-mail: yj-liu19@tsinghua.org.cn Received: 23 April 2026 Accepted: 1 June 2026 Abstract: The design optimization of passive micromixers for efficient fluid homogenization at low Reynolds numbers remains a multifaceted challenge, balancing geometric complexity, mixing performance, and manufacturing constraints. This work introduces a novel T-shaped passive micromixer incorporating a series of four concave-star units to induce chaotic advection. A comprehensive parametric study is conducted via two-dimensional (2D) finite element analysis (FEA) under the shallow-channel approximation, systematically varying the star tip radius (R), wall thickness (t), and concave angle (θ) across 9261 design points to establish a high-fidelity dataset. The results demonstrate that the concave-star geometry effectively disrupts laminar flow through flow splitting, acceleration, and secondary vortex generation, achieving a mixing index exceeding 0.99 at the outlet for optimal configurations. Furthermore, an integrated FEA and machine learning (FEA-ML) framework is proposed to rapidly predict performance. The Prior-Data Fitted Network (TabPFN) model achieves exceptional prediction accuracy (R2 > 0.9998), significantly outperforming conventional algorithms like Random Forest and XGBoost. This study provides not only an effective micromixer design but also a robust data-driven framework that accelerates the exploration and optimization of complex microfluidic geometries. Keywords: Micromixers; Computational fluid dynamics (CFD); Numerical simulations; Microfluidic Full paper is available at www.springerlink.com. DOI: 10.1007/s11696-026-05155-z
Chemical Papers 80 (11) 12903–12920 (2026) |
Monday, November 09, 2026 
|
|||
© 2026 Chemical Papers |
|
|||
