Influence of Machine Learning-based active flow control on the turbulent statistics of the flow over a circular cylinder

Aug 4, 2026

The paper investigates how machine learning (ML) can reduce aerodynamic drag of a circular cylinder in cross flow by actively controlling its wake using a synthetic jet. It frames wake control of bluff bodies in terms of passive and active strategies, emphasizing synthetic jets for cost, size, and reliability. It extends prior work by optimizing the input signal waveshape through Linear Genetic Programming, avoiding parametric assumptions. Particle Image Velocimetry then measures velocity fields and characterizes mean flow and turbulent statistics.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages1

Reading Time3 min

AccessFree to download

Motivation and Background

Passive vs. active flow control

Synthetic jets for wake control

ML-Based Waveshape Optimization

Linear Genetic Programming

Drag reduction objective

Experimental Measurements and Results

Particle Image Velocimetry

Mean flow and turbulent statistics

Influence of Machine Learning-based active flow control on the turbulent statistics of the flow over a circular cylinder A. Scala∗, C.S. Greco∗ , G.Paolillo∗ , T. Astarita∗ and G. Cardone∗ The aim of the present paper is to investigate the capabilities of Machine Learning (ML) to reduce the aerodynamic drag of a circular cylinder in cross flow, by actively controlling its wake with a synthetic jet. The control of the wake behind bluff bodies has a great relevance in several engineering applications not only for the purpose of drag reduction, but also for the suppression of vortex shedding. Many flow control strategies have been adopted in previous researches, which can be categorized into two main groups: passive and active flow control techniques. The latter group includes SJ actuators which have been proven an efficient flow control technique thanks to their advantageous features, such as reduced size and weight, improved manufacturability, low cost and high reliability. Previous works 1 2 have analysed the performance of a synthetic jet-based control of a cylinder wake varying essentially two control parameters, the momentum coefficient and the dimensionless frequency. On the other side, few attempts have been made to evaluate the effects of the input signal waveshape 3 . In these works, the signal waveshape was defined in a parametric way. In the present study, Machine Learning, in the form of Linear Genetic Programming, is used to overcome the limitations inherent to the assumption of a parametric waveshape and to find the optimal waveshape of the input signal for the drag reduction of the cylinder body. Once obtained the optimal waveform, Particle Image Velocimetry is used to obtain instantaneous two-dimensional velocity fields measurements in a plane containing the synthetic jet slot and to characterize the mean flow quantities and turbulent statistics of the phenomenon, as reported in Figure 1 . Figure 1: Left: Distribution of the costs during the ML optimization process; Middle: timeaveraged streamwise velocity field for the ML-based control law; Right: Turbulent Kinetic Energy map for the ML-based control law. ∗ Dep. Industrial Engineering, University of Naples ”Federico II”, 80125 Naples, Italy 1 L. H. Feng and Wang,J. Fluid Mech. 662,232-259 (2010) . 2 Carlo Salvatore Greco et al. J. Fluid Mech. 901,(2020) . 3 Li Hao Feng et al. J. Fluids Struct. 26 900-917,(2010) .

FAQs

What is the main goal of the study?

Reduce the aerodynamic drag of a circular cylinder in cross flow by actively controlling the wake using a synthetic jet guided by machine learning.

Why is synthetic jet control considered effective?

It enables efficient active wake control with practical advantages such as reduced size and weight, improved manufacturability, low cost, and high reliability.

How does the paper improve on previous waveshape-related approaches?

It uses Linear Genetic Programming to determine the optimal input signal waveshape without relying on a predefined parametric waveshape form.

Aug 4, 2026

Upload by Lucas Martin