Closed-loop separation control using machine learning

Abstract

We present the first closed-loop separation control experiment using a novel, model-free strategy based on genetic programming, which we call 'machine learning control'. The goal is to reduce the recirculation zone of backward-facing step flow at Re h = 1350 manipulated by a slotted jet and optically sensed by online particle image velocimetry. The feedback control law is optimized with respect to a cost functional based on the recirculation area and a penalization of the actuation. This optimization is performed employing genetic programming. After 12 generations comprised of 500 individuals, the algorithm converges to a feedback law which reduces the recirculation zone by 80 %. This machine learning control is benchmarked against the best periodic forcing which excites Kelvin-Helmholtz vortices. The machine learning control yields a new actuation mechanism resonating with the low-frequency flapping mode instability. This feedback control performs similarly to periodic forcing at the design condition but outperforms periodic forcing when the Reynolds number is varied by a factor two. The current study indicates that machine learning control can effectively explore and optimize new feedback actuation mechanisms in numerous experimental applications.

Drag reduction of a car model by linear genetic programming control

“…Sensor-based feedback has been shown to outperform periodic forcing for drag reduction of a D-shaped body (Pastoor et al 2008). A very general method for sensor-based feedback is provided by genetic programming control (GPC) (Gautier et al 2015;Debien et al 2016;Parezanović et al 2016). Yet, the advantages of filtering out noise has hardly been explored in GPC ).…”

“…The regression problem is solved with symbolic genetic programming using the plant (experiment) to evolve the control law. This model-free control strategy can detect and exploit nonlinear actuation mechanisms in an unsupervised manner as evidenced in several shear flow control experiments (Gautier et al 2015;Debien et al 2016;Parezanović et al 2016).…”

“…The flow is monitored with pressure sensors distributed at the rear side. We apply a model-free control strategy building on Dracopoulos & Kent (1997) and Gautier et al (2015). The optimized control laws comprise periodic forcing, multi-frequency forcing and sensor-based feedback including also time-history information feedback and combination thereof.…”

Stochastic modelling and feedback control of bistability in a turbulent bluff body wake

“…While some flow control methods such as extremum seeking (Henning et al 2008) and machine learning control (Gautier et al 2015) are model free, many more rely on having a mathematical model, and often a linear model, of the fluid system. For example, black-box linear models have often been used for convectively unstable flows such as the backward-facing step (Barbagallo et Modelling and feedback control of bistability in a turbulent bluff body wake 729 flows behave like a linear amplifier, thus allowing linear models to be obtained for various input-output behaviours.…”

Multi-Input Genetic Algorithm for Experimental Optimization of the Reattachment Downstream of a Backward-Facing-Step with Surface Plasma Actuator

“…In the present study, because genetic algorithm only requires the evaluation of the cost function regarding a set of control parameters, such an evolutionary algorithm is used in conjunction with experimental measurements in a feedback loop process. Here, the objective is not to define a function transfer (control law) between the sensors, and thus local or global flow conditions, and the actuator as it has been recently done in [33,34]. The present contribution interests in finding the best operating conditions over a large parameter space by coupling for the first time a numerical optimizer and tests conducted in an experimental wind-tunnel.…”