Hostname: page-component-7b7567c8c-2n9c9 Total loading time: 0 Render date: 2026-11-16T08:28:57.550Z Has data issue: false hasContentIssue false

Second-order sensitivity of parallel shear flows and optimal spanwise-periodic flow modifications

Published online by Cambridge University Press:  08 October 2015

E. Boujo*
Affiliation:
LFMI, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, SwitzerlandCAPS, ETH Zurich, CH-8092 Zurich, Switzerland
A. Fani
Affiliation:
LFMI, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland
F. Gallaire
Affiliation:
LFMI, École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland
*
†Email address for correspondence: edouard.boujo@gmail.com

Abstract

The question of optimal spanwise-periodic modification for the stabilisation of spanwise-invariant flows is addressed. A second-order sensitivity analysis is conducted for the linear temporal stability of parallel flows $U_{0}$ subject to small-amplitude spanwise-periodic modification ${\it\epsilon}U_{1},{\it\epsilon}\ll 1$ . It is known that spanwise-periodic flow modifications have a quadratic effect on stability properties, i.e. the first-order eigenvalue variation is zero, hence the need for a second-order analysis. A second-order sensitivity operator is computed from a one-dimensional calculation, which allows one to predict how eigenvalues are affected by any flow modification $U_{1}$ , without actually solving for modified eigenvalues and eigenmodes. Comparisons with full two-dimensional stability calculations in a plane channel flow and in a mixing layer show excellent agreement. Next, optimisation is performed on the second-order sensitivity operator: for each eigenmode streamwise wavenumber ${\it\alpha}_{0}$ and base flow modification spanwise wavenumber ${\it\beta}$ , the most stabilising/destabilising profiles $U_{1}$ are computed, together with lower/upper bounds for the variation in leading eigenvalue. These bounds increase like ${\it\beta}^{-2}$ as ${\it\beta}$ goes to zero, thus yielding a large stabilising potential. However, three-dimensional modes with wavenumbers ${\it\beta}_{0}=\pm {\it\beta}$ , $\pm {\it\beta}/2$ are destabilised, and therefore larger control wavenumbers should be preferred. The most stabilising $U_{1}$ optimised for the most unstable streamwise wavenumber ${\it\alpha}_{0,max}$ has a stabilising effect on modes with other ${\it\alpha}_{0}$ values too. Finally, the potential of transient growth to amplify perturbations and stabilise the flow is assessed with a combined optimisation. Assuming a separation of time scales between the fast unstable mode and the slow transient evolution of the optimal perturbations, combined optimal perturbations that achieve the best balance between transient linear amplification and stabilisation of the nominal shear flow are determined. In the mixing layer with ${\it\beta}\leqslant 1.5$ , these combined optimal perturbations appear similar to transient growth-only optimal perturbations, and achieve a more efficient overall stabilisation than optimal spanwise-periodic and spanwise-invariant modifications computed for stabilisation only. These results are consistent with the efficiency of streak-based control strategies.

Information

Type
Papers
Copyright
© 2015 Cambridge University Press 
Figure 0

Figure 1.Sketch of the base flow configuration: a 1D parallel flow $U_{0}(y)$ is modified with a small-amplitude 2D spanwise-periodic flow ${\it\epsilon}U_{1}(y)\cos ({\it\beta}z)$.

Figure 1

Figure 2.Variation of the leading growth rate for optimally destabilising flow modification (superscript $d$, dark lines, red online) and optimally stabilising flow modification (superscript $s$, light lines, green online), as predicted by sensitivity analysis. (a) Optimisation for the leading eigenmode of wavenumbers $({\it\alpha}_{0},{\it\beta}_{0})=(1,0)$. Suboptimals are also shown as thin lines. (b) Optimal growth rate variation for ${\it\alpha}_{0}=0.9,1.0,1.1$ and ${\it\beta}_{0}=0$.

Figure 2

Figure 3.Optimal (branch I) and first suboptimal (branch II) destabilising (dark, red online) and stabilising (light, green online) base flow modification for the most unstable mode $({\it\alpha}_{0},{\it\beta}_{0})=({\it\alpha}_{0,max},{\it\beta}_{0,max})=(1,0)$, at control wavenumbers ${\it\beta}=0.4$ (a) and ${\it\beta}=3$ (b). The long ticks on the left show the positions of the critical layers at $y=\pm 0.86$.

Figure 3

Figure 4.Optimal eigenvalue variations diverge like ${\it\beta}^{-2}$ as the control wavenumber decreases (same data as figure 2 in logarithmic scale, ${\it\alpha}_{0}=0.9,1.0,1.1$,${\it\beta}_{0}=0$).

Figure 4

Figure 5.Effect of optimal 2D flow modification (spanwise-periodic, control wavenumber ${\it\beta}=1$) on the growth rate of the leading eigenmode $({\it\alpha}_{0},{\it\beta}_{0})=(1,0)$. The lines show the 1D sensitivity prediction and the symbols show the full 2D stability calculations.

Figure 5

Figure 6.Effect of the most stabilising spanwise-periodic flow modification ${\it\epsilon}U_{1}(y)\cos ({\it\beta}z)$ on leading 2D and 3D eigenvalues at ${\it\alpha}_{0}=1$, with flow modification wavenumber (a) ${\it\beta}=0.4$ and (b) ${\it\beta}=1.6$, and $U_{1}$ optimised for the most unstable eigenmode $({\it\alpha}_{0},{\it\beta}_{0})=(1,0)$. The 3D eigenpair with ${\it\beta}_{0}=\pm {\it\beta}/2$ undergoes a splitting, and one of the two modes becomes unstable. This issue is avoided by choosing larger ${\it\beta}$ values, since the 3D eigenmodes affected by splitting are then very stable. In (a), ${\it\epsilon}=(0.18,0.35,0.53,0.71)\times 10^{-3}$, and in (b), ${\it\epsilon}=(0.18,\ldots ,0.71,0.88,1.06)\times 10^{-3}$.

Figure 6

Figure 7.Effect of optimal spanwise-periodic flow modification ${\it\epsilon}U_{1}(y)\!\,\cos ({\it\beta}z)$ on the leading growth rate with $U_{1}$ optimised for $({\it\alpha}_{0},{\it\beta}_{0})=({\it\alpha}_{0,max},{\it\beta}_{0,max})=(1,0)$, at control wavenumber ${\it\beta}=1$ (a) and ${\it\beta}=3$ (b). Branch I$^{d}$ is the most destabilising one and branch I$^{s}$ is the most stabilising one; ${\it\epsilon}=0.001$.

Figure 7

Figure 8.Upper and lower bounds on ${\it\lambda}_{2}$, i.e. maximal destabilising (dark, red online) and stabilising (light, green online) effect on the leading growth rate, as predicted by sensitivity analysis. (a) For ${\it\alpha}_{0}=0.5$, crossing of branches I and II, optimal at large and small ${\it\beta}$ respectively. The circles are calculations for the full stability problem. (b) Upper and lower bounds for ${\it\alpha}_{0}=0.3\ldots 0.9$. (c,d) Variation with ${\it\alpha}_{0}$ of the local maxima and minima shown as triangles in (a), and corresponding spanwise wavenumbers.

Figure 8

Figure 9.(a) Most destabilising (dark, red online) and stabilising (light, green online) $U_{1}$ for ${\it\alpha}_{0}={\it\alpha}_{0,max}=0.45$. (b) Total modified flow $U_{0}(y)+{\it\epsilon}U_{1}(y)\cos ({\it\beta}z)$ at $z=0$ and $z={\rm\pi}/{\it\beta}$ for ${\it\epsilon}=0.2$.

Figure 9

Figure 10.Effect of optimal 2D (spanwise-periodic) flow modifications at ${\it\beta}=0.8$ on the leading growth rate at ${\it\alpha}_{0}={\it\alpha}_{0,max}=0.45$. The lines show the sensitivity prediction and the symbols show the full stability calculations.

Figure 10

Figure 11.Effect of optimal spanwise-periodic flow modification ${\it\epsilon}U_{1}(y)\cos ({\it\beta}z)$ on the leading growth rate with $U_{1}$ optimised for ${\it\alpha}_{0}={\it\alpha}_{0,max}=0.45$ and ${\it\beta}=0.8$ (a), ${\it\beta}=1.5$ (b), ${\it\beta}=3.0$ (c). Branch I$^{d}$ is the most destabilising one and branch I$^{s}$ is the most stabilising one; ${\it\epsilon}=0.1$.

Figure 11

Figure 12.(a,b) Optimal and (c,d) first suboptimal perturbations for transient growth only. The arrows show the initial $(v,w)$ components (streamwise vortices) and the contours show the $u$ component after optimal amplification (streamwise streaks, low speed in dark, high speed in light). (e,f) Growth versus time for optimal and first suboptimal perturbations. (a,c,e) Plane channel flow, $\mathit{Re}=6000$, $({\it\alpha},{\it\beta})=(0,2)$; (b,d,f) mixing layer, $\mathit{Re}=100$, $({\it\alpha},{\it\beta})=(0,0.35)$.

Figure 12

Figure 13.Optimal combined transient growth and eigenvalue variation in the mixing layer at $\mathit{Re}=100$. The thick lines show the optimal combined effects for stabilisation and destabilisation; the triangles indicate the maximal stabilising effect over all times. The dashed lines (respectively thin dash-dotted lines) show the optimal (respectively first suboptimal) perturbations for transient growth only; the circles show the corresponding eigenvalue variation at optimal amplification time. Transient growth at (a) $({\it\alpha}=0,{\it\beta}=1)$ and (b) $({\it\alpha}=0,{\it\beta}=2)$; effect on the leading eigenvalue $({\it\alpha}_{0},{\it\beta}_{0})=(0.45,0)$.

Figure 13

Figure 14.The largest stabilising effect ${\it\lambda}_{2r}^{c}$ (thick line) obtained from the combined optimisation of transient growth and eigenvalue variation versus the control wavenumber ${\it\beta}$. Any other perturbation falls above the thick line, i.e. outside the grey region. The dashed line (respectively thin dash-dotted line) shows the stabilising effect of transient growth-only optimal (respectively first suboptimal) perturbations. Mixing layer, $\mathit{Re}=100$, ${\it\alpha}_{0}={\it\alpha}_{0,max}=0.45$.

Figure 14

Figure 15.(a) Growth rate variation for combined optimal perturbations as a function of amplitude ${\it\epsilon}$, compared with 1D and 2D optimal flow modifications for pure stabilisation. (b) Corresponding profiles of streamwise velocity: 1D optimal $U_{1}^{1D}$, 2D optimal $U_{1}^{2D}$ and combined optimal after amplification $u_{opt}^{c}(T)$. The optimal perturbation for pure transient growth $u_{opt}(T)$ is also shown as a dashed line superimposed onto the combined optimal, but the two curves are indistinguishable. Mixing layer, ${\it\beta}=0.8$, eigenmode $({\it\alpha}_{0},{\it\beta}_{0})=(0.45,0)$, $\mathit{Re}=100$.