Machine Learning-based optimization and PIV analysis of the active control of a cylinder wake via synthetic jets

The aim of the present paper is the investigation of the Synthetic Jet (SJ) based control of a cylinder wake through Linear Genetic Programming (LGP) technique and the flow field via Particle Image Velocimetry (PIV)technique. Machine Learning is a branch of Artificial Intelligence aimed at extracting knowledge and experience from big volumes of data and it is the art of building models from data using optimization and regression algorithms. A SJ is an actuator used mainly for flow control and heat transfer performances. In this work, a loudspeaker attached to an hollow cylinder is used as SJ actuator device and the optimization procedure regards the input signal sent to the device. A gradient-enriched machine learning control, know as gMLC algorithm, is used as optimization tool. A preliminary phase of analysis of the ML algorithm, in which an optimal control law is found, is conducted. The latter allows to obtain a very complex control law which is able to give a percentage drag reduction of 7.6 % with respect to the natural case and this reduction is found to be better than the one obtained by a reference sinusoidal signal characterized by a fundamental frequency of 44 Hz. After the ML analysis, an investigation via Particle Image Velocimetry is performed with the aim of of obtaining a comparison between the natural case, i.e. the uncontrolled configuration, and two different controlled cases: the optimal waveshape obtained via gMLC algorithm and the previous-mentioned sinusoidal waveshape.