Surrogate-based optimization with adaptive sampling for microfluidic concentration gradient generator design.

Area of Science:

• Microfluidics
• Computational Engineering
• Optimization Methods

Background:

• Designing microfluidic concentration gradient generators (μCGGs) is crucial for applications requiring precise concentration gradients (CGs).
• Traditional design methods may lack efficiency and robustness in achieving specific CG profiles.

Purpose of the Study:

• To develop and validate a surrogate-based optimization (SBO) method with adaptive sampling for designing μCGGs.
• To identify optimal surrogate model components and adaptive sampling strategies for enhanced design accuracy and efficiency.

Main Methods:

• Utilized a physics-based component model (PBCM) for data generation.
• Employed Kriging surrogate models with various regression and correlation combinations.
• Investigated different adaptive sampling (infill) techniques, including the lower bound (LB) strategy.
• Performed comparative analysis against random sampling and gradient-based optimization.

Main Results:

• The combination of first-order polynomial regression and Gaussian correlation models yielded the most accurate Kriging surrogate model.
• The lower bound (LB) infill strategy proved most efficient for global optimum search.
• Optimized μCGG designs achieved concentration gradients with less than 12% discrepancy.
• SBO with adaptive sampling demonstrated superior efficiency, accuracy, and robustness compared to random sampling and gradient-based methods.

Conclusions:

• Surrogate-based optimization with adaptive sampling is a highly effective approach for designing μCGGs.
• The developed method enables precise control over concentration gradients, meeting design specifications efficiently.
• This SBO strategy offers a robust alternative for global optimum seeking in microfluidic device design.