May 18 – 23, 2026
Europe/Rome timezone

Machine Learning-Based Design Optimization of a Microfluidic Micromixer for Enhanced Mixing Performance

May 20, 2026, 6:00 PM
20m
Oral Computational and data-driven approaches in microfluidics Wednesday 20/05, 14 - 19; Room 35

Description

Efficient mixing in microfluidic systems at low Reynolds numbers remains a major challenge due to dominant laminar flow and slow diffusive transport. Micromixers play an important role in many biomedical applications such as lab-on-chip platforms, biochemical assays, drug delivery etc. Conventional micromixer design usually depends on parametric sweeps based numerical optimization, that requires high computational cost and often fails to fully explore complex design spaces. Recent advances in machine learning (ML) offer new approaches to improve design optimization, reduce unnecessary simulations, and identify effective, non-intuitive geometric configurations that enhance mixing performance under practical constraints.
In this study, a ML-based framework is proposed to optimize the micromixer design using Bayesian Optimization (BO) and Constrained-Bayesian Optimization (CBO) approaches combined with Gaussian Processes (GP). For this purpose, python is integrated with COMSOL Multiphysics to create an automated simulation framework that allows the system to create geometry, meshing, simulation, data extraction and dataset generation. Important geometrical configurations of micromixer design such as rectangular shape obstacle size, channel length, channel width, spacing between the obstacle and angles are used as design variables. Mixing index and pressure drop are selected as the main objective and design constraints respectively.
BO is then employed to suggest new design candidates that provide the best balance between finding new designs and improving promising ones. Moreover, CBO approach ensures satisfaction of pressure drop limits. The expected outcomes provide significant improvement in micromixer design with enhanced mixing performance and pressure drop compared to baseline study. In addition, the proposed ML framework is expected to identify optimal flow perturbation mechanism that stimulates chaotic advection and flow mixing at low Reynolds numbers. Overall, the proposed ML framework combined with multiphysics simulation provides valuable insights to enhance mixing performance and accelerate scalable microfluidic device development.

Author

Muhammad Waqas (Faculty of Mechanical Engineering and Design, Kaunas University of Technology, Kaunas, Lithuania)

Co-authors

Prof. Arvydas Palevicius (Faculty of Mechanical Engineering and Design, Kaunas University of Technology, Kaunas, Lithuania) Dr Ivana Kundacina (University of Novi Sad, BioSense Institute, Novi Sad, Serbia)

Presentation materials

There are no materials yet.