Machine learning-aided design optimization of a mechanical micromixer

In real-life mechanical engineering applications, it is often complex to achieve an optimal multi-objective design, because of the costs related to fabrication and test of different prototypes. For this reason, the use of computational tools is a recommended practice. In this work, the design of an efficient mixing mechanical device composed of a rectangular pillar confined in a microchannel is aided by machine learning techniques (addressed as machine learning-aided design optimization, MLADO, proposed in this work). A random forest classifier is trained to predict which geometric configuration may lead to vortex shedding. Later, a multi-objective optimization problem is investigated, which consists of minimizing the required pumping power and maximizing the mixing efficiency under some design constrains. If extra training data are desired for surrogates, the random forest classifier can be used to predict whether it is worthy or not to simulate the new configuration, avoiding to run irrelevant computational intensive cases and accelerating the data-driven process. The resulting optimal designs from using the NSGA-II genetic algorithm on the surrogates are simulated, and their performance is shown. The optimal geometric configurations, even for very unfavorable mixing conditions and a medium-low Reynolds number of 200, give a maximum mixing efficiency of around 50% at very low pumping power cost in a short channel, outperforming existing devices in the literature. The MLADO framework followed in this work can be easily extendable and automated for other similar design processes in mechanical engineering at any scale, by including shape parameterization strategies.