Feedback machine learning control of 3D falling liquid film
Please login to view abstract download link
The control of liquid film instabilities is pivotal in many industrial applications. Common approached to handle this kind of problem encompass reduced order models and stability analysis. However, these techniques suffers from model uncertainties, and they are hampered by the intrinsic limitations of the simplified models. Machine learning emerged as a viable alternative, capable of treating the control problem as a black box (model-free) optimization problem. In this work, we apply various machine learning based control methods to the stabilization of 3D waves in a liquid film dragged by a moving substrate. The liquid film flow is modelled via a reduced order model that extends the Kapitza-Skhadov model, and the control performances are measured in terms of wave attenuation. The preliminary results show a considerable decrease of the 3D perturbation, with few observations of the liquid film height.
