Model-Free Control for Drop-on-Demand Droplet Generation Using Reinforcement Learning

Mikhail Vulf, Anton Bolychev, Dmitry Kolomenskiy, Pavel Osinenko

Abstract: Recently we have developed a Drop-on-Demand droplet generator which allows to generate coarse suspension droplets in a wide range of sizes (from 0.75 to 4.40 mm), However, there is a common problem that it is difficult for different liquids to generate one drop of the desired size. To solve this problem, an auto-calibration and dynamic control box is being developed. In the present work, we simulate the dynamics of a hydraulic part of the droplet generator by an experimentally verified model. This simulation is used as a black box for model-free control developed by reinforcement learning approach. The obtained results are consistent with previous droplet generation experiments. The proposed model-free control method can be used for automatic parameter adjustment for generating a single drop of different liquids.