Design and process/measurement for immersed element control in a reconfigurable vertically falling soap film

Abstract

Reinforcement learning has proven successful at harnessing the passive dynamics of underactuated systems to achieve least energy solutions. However, coupled fluid-structural models are too computationally intensive for in-the-loop control in viscous flow regimes. My vertically falling soap film will provide a reconfigurable experimental environment for machine learning controllers. The real-time position and velocity data will be collected with a High Speed Video system, illuminated by a Low Pressure Sodium Lamp. Approximating lines of interference within the soap film to known pressure variations, controllers will shape downstream flow to desired conditions. Though accurate measurement still eludes those without Laser Doppler Velocimetry, order of magnitude Reynolds numbers can be estimated to describe the regime of controller inquiry.