Machine Learning and Genetic Algorithms in the Microfluidic Design of Highly Efficient Micromixers

Microscale reactions are often used to determine the reactivity and byproducts of combinatorial reactions between diverse chemicals. Despite significant work in designing microfluidic mixers, there are still questions regarding achieving high mixing efficiency and low-pressure drop at low Reynolds numbers. Microchannels with complex shapes have the potential to give high mixing efficiency at low speeds, but their design and construction are difficult, and channel clogging due to channel shape complexity restricts their practical usage. In this study, we developed an approach to create a highly effective micromixer using simple geometrical concepts and barriers. We explored the optimal geometries of microchannels and obstacles by employing machine learning in conjunction with the Reynolds number (Re) to obtain maximum mixing efficiency and minimum pressure drop. Approximately 1000 different micromixer designs were simulated and analyzed to train the machine learning model, focusing on the mixing index and pressure drop. We then utilized a genetic algorithm to optimize key parameters such as the height of the obstacles, the dent angle of the obstacles, the angular offset between two barriers of the obstacles, the radius of curvature of the micromixer, and the Reynolds number. The optimization revealed that a 190 mu m obstacle height, 67 degrees dent angle, 1102 mu m curvature radius, and 53 degrees angular distance between two obstacles resulted in the maximum mixing efficiency and lowest pressure drop at low Reynolds numbers. To validate the proposed design, we fabricated the micromixer with standard soft lithography techniques.