arXiv subjects

Julien Dandois

Publications and source records attributed to Julien Dandois.

4 recordsLinked to original sources

arXiv2022

Reinforcement-learning-based actuator selection method for active flow control

This paper addresses the issue of actuator selection for active flow control by proposing a novel method built on top of a reinforcement learning agent. Starting from a pre-trained agent using numerous actuators, the algorithm estimates the impact of a potential actuator removal on the value function, indicating the agent's performance. It is applied to two test cases, the one-dimensional Kuramoto-Sivashinsky equation and a laminar bi-dimensional flow around an airfoil at Re=1000 for different angles of attack ranging from 12 to 20 degrees, to demonstrate its capabilities and limits. The proposed actuator-sparsification method relies on a sequential elimination of the least relevant action components, starting from a fully developed layout. The relevancy of each component is evaluated using metrics based on the value function. Results show that, while still being limited by this intrinsic elimination paradigm (i.e. the sequential elimination), actuator patterns and obtained policies demonstrate relevant performances and allow to draw an accurate approximation of the Pareto front of performances versus actuator budget.

physics.flu-dyn↗

arXiv2021

Uncertainty Quantification Applied to the Propagation of a Transonic Wind Tunnel Inflow Inhomogeneities

The uncertainty associated with the experimental inflow in a wind tunnel affects the prediction of the flow of interest by numerical simulations. We evaluate this impact using uncertainty quantification. A method is developed and applied to the simulation of the drag generated by the flow past a cylinder installed in the transonic S3Ch ONERA mid-scale facility. The inflow uncertainty results from the imperfect knowledge and variability of the flow in the settling chamber. It is taken into account via the inlet boundary condition in the numerical companion setup and evaluated experimentally by measuring the inflow using a hot-wire rake. The propagation of the input uncertainties is carried \alert{out} through a two-dimensional RANS model of the experiment. A polynomial surrogate model is developed to infer the uncertainty associated with the drag of the cylinder. Following observations of Gaussian inputs, the parameters of the stochastic model are constructed in two ways, first through a projection approach, based on the Gauss-Hermite quadrature rule, and then using a sparsity based regression approach, based on compressed sensing. The latter drastically reduces the number of deterministic numerical simulations. The drag is most influenced by the central part of the inflow but the overall uncertainty remains low.

physics.flu-dyn↗

arXiv2021

Identification and reconstruction of high-frequency broadband fluctuations in a turbulent flow exhibiting low-frequency deterministic motions

The turbulent flow around a squared-section cylinder at $Re=22000$ exhibits both a low-frequency Vortex-Shedding (VS), a common feature of all bluff-body flows, and high-frequency Kelvin-Helmholtz (KH) structures evolving on the shear layers formed at the top and bottom side of the cylinder. The VS motion triggers a periodic movement of these shear layers, inducing oscillations of the strength of their velocity gradients. The high-frequency KH structures respond to these low-frequency oscillations of the base-flow gradients in a quasi-steady manner. In this paper we will propose a general framework to capture and reconstruct stochastic high-frequency fluctuations (the KH structures) evolving on top of a low-frequency deterministic motion (the periodic VS). Based on a generic triple decomposition, that allows separation of the stochastic component from the deterministic one, we propose a Phase-Conditioned Localized Spectral Proper Orthogonal Decomposition (PCL-SPOD), which isolates high-frequency spatio-temporal structures within the stochastic component at different phases of the VS motion, with a short-time Fourier transform. Finally, we compare these structures to those obtained with a Phase-Conditioned Localized Resolvent formalism (PCL-Resolvent), which consists in a Resolvent analysis around instantaneous snapshots of the periodic VS.

physics.flu-dyn↗

arXiv2020

Robust flow control and optimal sensor placement using deep reinforcement learning

This paper focuses on a drag-reducing control strategy on a 2D-simulated laminar flow past a cylinder. Deep reinforcement learning algorithms have been implemented to discover efficient control schemes, using two synthetic jets located on the cylinder's poles as actuators and pressure sensors in the wake of the cylinder as feedback observation. The present work focuses on the efficiency and robustness of the identified control strategy and introduces a novel algorithm (S-PPO-CMA) to optimise the sensor layout. An energy-efficient control strategy reducing drag by 18.4% at Reynolds number 120 is obtained. This control policy is shown to be robust both to the Reynolds number in the range [100,216] and to measurement noise, enduring signal to noise ratios as low as 0.2 with negligible impact on performance. Along with a systematic study on sensor number and location, the proposed sparsity-seeking algorithm has achieved a successful optimisation to a reduced 5-sensor layout while keeping state-of-the-art performance. These results highlight the interesting possibilities of reinforcement learning for active flow control and pave the way to efficient, robust and practical implementations of these control techniques in experimental or industrial systems.

physics.flu-dyn↗