Active Control of Turbulent Airfoil Flows Using Adjoint-Based Deep Learning

<p><span>Large-eddy simulation (LES) is an often used technique in the geosciences to simulate turbulent oceanic and atmospheric flows. In LES, the effects of the unresolved turbulence scales on the resolved scales (via the Reynolds stress tensor) have to be parameterized with subgrid models. These subgrid models usually require strong assumptions about the relationship between the resolved flow fields and the Reynolds stress tensor, which are often violated in reality and potentially hamper their accuracy.</span></p><p><span>In this study, using the finite-difference computational fluid dynamics code MicroHH (v2.0) and turbulent channel flow as a test case (friction Reynolds number Re<sub>τ</sub> 590), we incorporated and tested a newly emerging subgrid modelling approach that does not require those assumptions. Instead, it relies on neural networks that are highly non-linear and flexible. Similar to currently used subgrid models, we designed our neural networks such that they can be applied locally in the grid domain: at each grid point the neural networks receive as an input the locally resolved flow fields (u,v,w), rather than the full flow fields. As an output, the neural networks give the Reynolds stress tensor at the considered grid point. This local application integrates well with our simulation code, and is necessary to run our code in parallel within distributed memory systems.</span></p><p><span>To allow our neural networks to learn the relationship between the specified input and output, we created a training dataset that contains ~10.000.000 samples of corresponding inputs and outputs. We derived those samples directly from high-resolution 3D direct numerical simulation (DNS) snapshots of turbulent flow fields. Since the DNS explicitly resolves all the relevant turbulence scales, by downsampling the DNS we were able to derive both the Reynolds stress tensor and the corresponding lower-resolution flow fields typical for LES. In this calculation, we took into account both the discretization and interpolation errors introduced by the finite staggered LES grid. Subsequently, using these samples we optimized the parameters of the neural networks to minimize the difference between the predicted and the ‘true’ output derived from DNS.</span></p><p><span>After that, we tested the performance of our neural networks in two different ways:</span></p><ol><li><span>A priori or offline testing, where we used a withheld part of the training dataset (10%) to test the capability of the neural networks to correctly predict the Reynolds stress tensor for data not used to optimize its parameters. We found that the neural networks were, in general, well able to predict the correct values. </span></li> <li><span>A posteriori or online testing, where we incorporated our neural networks directly into our LES. To keep the total involved computational effort feasible, we strongly enhanced the prediction speed of the neural network by relying on highly optimized matrix-vector libraries. The full successful integration of the neural networks within LES remains challenging though, mainly because the neural networks tend to introduce numerical instability into the LES. We are currently investigating ways to minimize this instability, while maintaining the high accuracy in the a priori test and the high prediction speed.</span></li> </ol>

Laminar airfoil has extensive application prospect in the civil aviation area as its low drag characteristic. However, the laminar flow is too sensitive to be disturbed and then laminar-turbulent transition location will move forward, which results in a significant increase in drag. Therefore, it is meaningful to obtain the aerodynamic characteristics of laminar airfoil both under natural transition and disturbed transition conditions. This paper focuses on the aerodynamic characteristics of a laminar airfoil named NACA65(1)412. Firstly, numerical simulations of natural transition of the airfoil flow at various angles of attack were carried out using γ-Reθ model. Secondly, forced transition simulations were taken to imitate the disturbed airfoil flow using k-ω SST model. Results show that, the transition location has a great influence on the aerodynamic characteristics of laminar airfoil, especially on the drag. For the natural transition, as the angle of attack increasing, the natural transition location on the upper surface moves upstream while the one on the lower surface moves the opposite. The upper-surface transition location moves upstream rapidly after the 4° angle of attack. For the forced transition, the drag increases approximately linearly with the transition location for both the upper and lower surface, but the upper-surface behaves much more significant. The lift changes little with the movement of transition locations.

An experimental investigation was made to determine the variation of shock shape and extent of subsonic region behind the shock wave with angle of attack for a 70° cone at various Mach numbers. The main interest was centered on those Mach numbers which produced detached shock waves or for which the possibility of detachment at angle of attack existed. The tests were conducted in the GALCIT 2.5 Supersonic Wind Tunnel at angles of attack of 0°, 3°, 6°, 9°, 12°, and 15° and Mach numbers of 1.438, 1.544, 1.584, 1.857, 1.986, and 3.01. It was found that, with increasing angle of attack and constant Mach number, the subsonic region behind the shock wave increased in the lower portion and decreased in the upper portion. With increasing Mach number the subsonic region decreased for all angles of attack. Interaction between the upper and lower portions of the shock wave affects the subsonic region behind the wave, suppressing its appearance on the upper surface, retarding its disappearance from the lower surface.

Different noise generation mechanisms of a controlled diffusion aerofoil and their dependence on Mach number

In this study, Reynolds-averaged Navier-Stokes simulation (RANS) of the uniform blowing and suction (UB/US) control on a Clark-Y airfoil was performed aiming at improving an airfoil performance by friction drag reduction. First, the control effect when only the uniform blowing control or uniform suction control is applied on the airfoil surface was investigated by changing the control locations. The blowing or suction velocity was 0.14% of the free-stream velocity and the blowing/suction area was set at four different locations from the leading edge to the trailing edge on both the upper and lower surfaces. The Reynolds number based on the chord length is 1.5 × 106. The angle of attack is set to 0°. It was found that friction drag is decreased/increased by single UB/US control. It was also found that the lift-to-drag ratio improved with UB on the lower surface or US on the upper surface, and decreased with UB on the upper surface or US on the lower surface. In the combined control of UB and US, the blowing and suction velocity was 0.14% or 0.26% of the free-stream velocity and the locations of blowing/suction and flow conditions were the same as those in the cases with either UB or US. It seemed that the lift-to-drag ratio was improved by the combined control of UB on the lower surface and US on the upper surface. In particular, the lift-to-drag ratio was most improved by US on the lower rear surface and UB on the upper rear surface.

The Large-Eddy Simulation technique of compressible flows and the effect of compressibility on mixing layers are the main subjects of this thesis. Direct Numerical Simulations (DNS) and Large-Eddy Simulations (LES) of the temporal compressible mixing layer at various Mach and Reynolds numbers have been conducted to investigate these subjects. With respect to the LES technique, Large-Eddy Simulations have been performed at convective Mach numbers 0.2, 0.6 and 1.2 and the results have been compared with filtered DNS-data. It appeared that the dynamic subgrid-models lead to relatively accurate results compared to the other models tested. The dynamic approach turned out to yield acceptable results too in LES of a mixing layer that currently cannot be simulated using DNS. Care has to be taken to ensure that the numerical errors in LES are sufficiently small. It was found that these errors are usually sufficiently small if the filter width equals twice the grid-spacing. In addition to modelling the turbulent stress tensor, compressible LES formally requires the modelling of the subgrid-terms in the energy equation, which do not occur in incompressible LES. However, the compressible Large- Eddy Simulations demonstrated that the turbulent stress tensor is the dominant subgrid-term, even at convective Mach number 1.2. This important subgrid-term was also investigated from a theoretical point of view and realizability conditions for this tensor were derived. Regarding compressibility effects in the mixing layer, shock-waves were found in the three-dimensional DNS at convective Mach number 1.2. Furthermore, we have investigated the cause of the mixing layer growth rate reduction with increasing compressibility, using four DNS-databases covering the range of convective Mach numbers from 0.2 to 1.2. It was found that the growth rate reduction cannot be explained by the dilatational terms, but rather by the reduced pressure fluctuations, leading to reduced pressure strain and turbulent production terms.

This study examines the aerodynamic behavior of the symmetric National Advisory Committee for Aeronautics (NACA) 0018 airfoil at a chord-based Reynolds number of 30,000, utilizing Direct Numerical Simulation (DNS) as the primary tool, supplemented by selected Large Eddy Simulation (LES) cases. DNS covers a full sweep of angles of attack from 0° to 10°, while LES was conducted at α = 4° and 8° to verify consistency between the two approaches. Despite the limited LES dataset-restricted by computational cost, the results showed nearly identical aerodynamic loads and surface pressure distributions, thereby reinforcing confidence in the DNS database. To further assess robustness, two additional DNS cases were performed for the NACA 0012 airfoil at α = 3° and 6° under identical grid resolution and flow conditions. These simulations demonstrated very good agreement with published experiments, extending validation beyond a single geometry. In both DNS and LES, clean inflow conditions were imposed, ensuring that the domain inlet did not introduce resolved turbulence. The results captured laminar boundary-layer separation on the suction side without reattachment, producing bluff-body-like wake dynamics dominated by periodic vortex shedding, while classical Kelvin–Helmholtz instabilities were not significant. Computed aerodynamic coefficients, including time-averaged lift, showed good agreement with measurements. The novelty of this work lies in providing the first DNS dataset for a thick symmetric airfoil (NACA 0018) at Re = 30,000, further validated through NACA 0012 comparisons. The findings establish a benchmark for future investigations of unsteady aerodynamics, particularly dynamic stall in pitching motions, and are directly relevant for the design of low-Reynolds-number devices such as micro air vehicles, unmanned aerial vehicles, and vertical-axis wind turbines.

Roughness Based Crossflow Transition Control for a Swept Wing Airfoil Design

Tonal airborne noise of aerofoils appears in a limited range of moderate Reynolds numbers and angles of attack. In these specific conditions, the aerofoil is characterised by a large region of laminar flow over the aerodynamic surface, typically resulting in two-dimensional laminar instabilities in the boundary layer, generating one or more acoustic tones. The numerical simulation of such phenomenon requires, beside an accurate prediction of the unsteady flow field, a proper modelling of the laminar to turbulent transition of the boundary layer, which generally imposes the use of highly CPU demanding approaches such as large eddy simulation (LES) or direct numerical simulation (DNS). This paper aims at presenting the results of numerical experiments for evaluating the capability of capturing the tonal airborne noise by using an advanced, yet low computationally demanding, unsteady Reynolds-averaged Navier-Stokes (URANS) turbulence model augmented with a transitional model to account for the laminar to turbulent transition. This approach, coupled with the Ffowcs Williams and Hawkings (FW-H) acoustic analogy, is adopted for predicting the far-field acoustic sound pressure of a NACA 0012 aerofoil with Reynolds number ranging from $0.39 · 10^6$ to $1.09 · 10^6$. The results show a main tone located approximately at 1.6–1.8 kHz for a Reynolds number equal to $0.62 · 10^6$, increasing to 2.4 kHz at Reynolds number equal to $0.85 · 10^6$ and 3.4 kHz at $1.09 · 10^6$, while no main tones are observed at $0.39 · 10^6$. The computed spectra confirm that the acoustic emission of the aerofoil is dominated by tonal structures and that the frequency of the main tone depends on the Reynolds number consistently with the ladder-like tonal structure suggested by Paterson et al. Moreover, in specific conditions, the acoustic spectra exhibit a multi-tonal structure visible in narrowband spectra, in line with the findings of Arbey and Bataille. The presented results demonstrate the capability of the numerical model of predicting the physics of the tonal airborne noise generation.

The shock-induced separation easily occurred on the upper surface of supercritical airfoil at transonic speeds, which would change the aerodynamic characteristics. The problem of the shock-induced separation was not solved completely for the complicated phenomena and flow mechanism. In this paper, the influencing factors of shock-induced separation for supercritical airfoil CH was analyzed at transonic speeds. The Navier-Stokes equations were solved, in order to investigate influence of different attack angles, Mach numbers and Reynolds numbers. The computation attack angles of CH airfoil varied from 0oto 7o, Reynolds numbers varied from 5×106to 50×106per airfoil chord while Mach number varied from 0.74 to 0.82. It was shown that the shock-induced separation was affected by attack angles, Mach numbers and Reynolds numbers, but the influence tendency and areas were quite different. The shock wave location and intensity were affected by the three factors, and the boundary layer thickness was mainly affected by Reynolds number, while the separation structure was mainly determined by the attack angle and Mach number.

Supercritical wings are widely used in large aircraft due to their excellent transonic performance, but their aerodynamic characteristics are highly sensitive to Reynolds number. To systematically study the influence of Reynolds number on the aerodynamic characteristics of a supercritical wing, cryogenic high Reynolds number pressure measurement tests were conducted in the European Transonic Wind Tunnel (ETW). A 1:17.87 scale wing-body combination model of a typical supercritical wing was employed. The Reynolds number was increased via the pressure increase and cooling technique, covering a test Reynolds number range from 2.3 × 106 to 3.5 × 107. Model deformation effects were isolated to obtain pressure data reflecting pure Reynolds number effects. The variation patterns of pressure distribution, lift characteristics, and pitching moment characteristics with Reynolds number were analyzed. The results indicate that, at lower speeds (Ma = 0.4 and 0.6), the supercritical wing is less affected by Reynolds number; the upper surface is more significantly influenced by Reynolds number than the lower surface; the Reynolds number effect primarily manifests in the transonic regime by delaying the onset position of the shock wave on the upper wing surface, thereby affecting aerodynamic force characteristics; several aerodynamic characteristic parameters such as ΔCL, α0, and Cm exhibit a linear relationship with the logarithm of Reynolds number. Experimental results obtained at low Reynolds numbers cannot be directly extrapolated to actual flight conditions, necessitating the consideration of Reynolds number effect in the aerodynamic design optimization of large aircraft.

PurposeThe purpose of this paper is the examination of fluid flow around NACA0012 airfoil, with the aim of the numerical validation between the experimental results in the wind tunnel and the Lattice Boltzmann method (LBM) analysis, for the medium Reynolds number (Re = 191,000). The LBM–large Eddy simulation (LES) method described in this paper opens up opportunities for faster computational fluid dynamics (CFD) analysis, because of the LBM scalability on high performance computing architectures, more specifically general purpose graphics processing units (GPGPUs), pertaining at the same time the high resolution LES approach.Design/methodology/approachProcess starts with data collection in open-circuit wind tunnel experiment. Furthermore, the pressure coefficient, as a comparative variable, has been used with varying angle of attack (2°, 4°, 6° and 8°) for both experiment and LBM analysis. To numerically reproduce the experimental results, the LBM coupled with the LES turbulence model, the generalized wall function (GWF) and the cumulant collision operator with D3Q27 velocity set has been used. Also, a mesh independence study has been provided to ensure result congruence.FindingsThe proposed LBM methodology is capable of highly accurate predictions when compared with experimental data. Besides, the special significance of this work is the possibility of experimental and CFD comparison for the same domain dimensions.Originality/valueConsidering the quality of results, root-mean-square error (RMSE) shows good correlations both for airfoil’s upper and lower surface. More precisely, maximal RMSE for the upper surface is 0.105, whereas 0.089 for the lower surface, regarding all angles of attack.

A turbulent jet emanating from an unconfined, premixed burner is investigated by using large eddy simulation (LES) and direct numerical simulation (DNS) as well as experimentally by means of optical (OH* chemiluminescence), acoustic (microphone), and laser-optical measurement techniques (Particle Image Velocimetry). Comparison of the results obtained through experiments, LES, and DNS indicate a reasonable agreement. In order to analyze the impact of mesh refinement on the resolved flame properties and acoustic radiations, computational grids with varying resolutions are used for the LES. As large coherent flow motion exists in the considered flow case, due to an over-predicted diffusion the flame length calculated with LES is underestimated. On the other hand, DNS exhibits a similar intensity distribution for OH* as the experiment and, hence, the flame length is predicted accurately by DNS. The emitted noise spectrum has a tonal shape with peaks at the burner’s resonance frequency for the non-reacting flow which changes to broadband noise and, in general, is raised in amplitude for reacting flows. In addition, it is shown that an increase in Reynolds number, preheat temperature, or a decrease in equivalence ratio close to stoichiometric ratios yields more noise being emanated from the burner. The latter indicates the fact that direct combustion noise is linked to interactions of turbulent fluctuations with the flame front. When using an equivalence ratio closer to stoichiometric ratio, a thinner reaction zone is expected and the intrinsic interaction between the flame and turbulent flow is more pronounced.

A new class of power generation devices that experiences increased losses due to bulk flow separation in segments of their expected in-flight regime is emerging. As such, active flow control becomes increasingly relevant to mitigate these losses and reclaim the entire flight envelope. This study explores the effect of flow injection on transonic flows experiencing bulk separation. Reynolds-averaged Navier–Stokes simulations of a 3D wall-mounted hump at low Reynolds numbers are conducted to assess the response of transonic bulk separation to flow injection. Unsteady simulations are performed to understand the differences between slot and discrete port injection and determine optimum forcing frequencies. Discrete ports require higher pressures to overcome the momentum deficit associated with the smaller injection area relative to the width of the domain. Steady and unsteady injections are found viable strategies for mitigating the extent (or appearance) of bulk separation. Experiments are conducted with discrete injection for a range of Mach and Reynolds numbers. The response of the bulk separation to said injection is evaluated by analyzing both local pressure measurements and schlieren imaging. The study shows that the required pressure of injection is strongly correlated to the length scale of the uncontrolled separation. With large-eddy simulations, the flow separation and frequency content within the separated region can be reasonably predicted. This study aims to take further steps to establish guidelines for applying flow control to the emerging class of power generation devices experiencing losses from bulk separation.

Even for the stationary airfoils, due to the boundary and shear layer interactions of upper and lower surface of the airfoils, alternating vortex patterns form and the flow becomes time dependent. In the current study, the unsteady behavior of the flow around a symmetric airfoil is considered as incidence angle increases. The flow patterns are presented for wide range of angles of attack values. The vortex pattern generated is analyzed numerically for different angles of attack at Re=1000 around NACA 0012 airfoil. At this Reynolds number, the flow is laminar and boundary layers are quite thick. Flow separation and unsteady vortex shedding is observed even at low angles of attack. For NACA 0012 airfoil, the unsteady vortex pattern is observed at about 8° angle of attack for Re=1000. Spectral analysis is performed for angles of attack ranging from 0° to 90°. It is presented that amplitude spectrum of lift coefficient (C1) start to shows a peak at 8° for NACA 0012 and the aerodynamic forces presents oscillatory behavior afterwards. The effect of angle of attack to wake pattern and instantaneous and mean aerodynamic coefficients are discussed. The time-averaged streamlines, pressure and skin friction coefficients are analyzed to observe the vortex formation and separation from the airfoil upper surface at this low Reynolds number.