Hostname: page-component-7b7567c8c-tnkh4 Total loading time: 0 Render date: 2026-11-16T04:53:49.814Z Has data issue: false hasContentIssue false

Optimal white-noise stochastic forcing for linear models of turbulent channel flow

Published online by Cambridge University Press:  24 April 2023

Jacob J. Holford*
Affiliation:
Department of Aeronautics, Imperial College London, London SW7 2AZ, UK
Myoungkyu Lee
Affiliation:
Department of Aerospace Engineering and Mechanics, The University of Alabama, Tuscaloosa, AL 35487, USA
Yongyun Hwang
Affiliation:
Department of Aeronautics, Imperial College London, London SW7 2AZ, UK
*
†Email address for correspondence: jh5315@imperial.ac.uk

Abstract

In the present study an optimisation problem is formulated to determine the forcing of an eddy-viscosity-based linearised Navier–Stokes model in channel flow at $Re_\tau \approx 5200$ ($Re_\tau$ is the friction Reynolds number), where the forcing is white-in-time and spatially decorrelated. The objective functional is prescribed such that the forcing drives a response to best match a set of velocity spectra from direct numerical simulation (DNS), as well as remaining sufficiently smooth. Strong quantitative agreement is obtained between the velocity spectra from the linear model with optimal forcing and from DNS, but only qualitative agreement between the Reynolds shear stress co-spectra from the model and DNS. The forcing spectra exhibit a level of self-similarity, associated with the primary peak in the velocity spectra, but they also reveal a non-negligible amount of energy spent in phenomenologically mimicking the non-self-similar part of the velocity spectra associated with energy cascade. By exploiting linearity, the effect of the individual forcing components is assessed and the contributions from the Orr mechanism and the lift-up effect are also identified. Finally, the effect of the strength of the eddy viscosity on the optimisation performance is investigated. The inclusion of the eddy viscosity diffusion operator is shown to be essential in modelling of the near-wall features, while still allowing the forcing of the self-similar primary peak. In particular, reducing the strength of the eddy viscosity results in a considerable increase in the near-wall forcing of wall-parallel components.

Information

Type
JFM Papers
Copyright
© The Author(s), 2023. Published by Cambridge University Press
Figure 0

Figure 1.Premultiplied two-dimensional spectra for $k_zh = 14$ (shaded), 30 (solid), 50 (dashed) ($\lambda _z/h\approx 0.45, 0.21, 0.13$; $\lambda _z^+\approx 2300, 1100, 650$) of (a) streamwise, (b) wall-normal and (c) spanwise velocities and (d) Reynolds shear stress. Here the spectra are normalised and the selected contour levels are $0.20$ and $0.80$ of the maximum. The horizontal and vertical pairs of dashed lines in (a–c) are for $y/\lambda _z = 0.1$, $\lambda _x/\lambda _z = 10$ and $y/\lambda _z = 0.5$, $\lambda _x/\lambda _z = 2$ respectively.

Figure 1

Figure 2.Premultiplied two-dimensional spectra in self-similar coordinates for $k_z = 14/h$, $k_z^+ = 0.0027$ ($\lambda _z = 0.45h$, $\lambda _z^+ = 2333$) of (a) streamwise, (b) wall-normal and (c) spanwise velocities and (d) Reynolds stress. Here the shaded contours indicate DNS spectra and the solid contours are optimisation results from (2.19). The contour levels are separated by 0.20 times the maximum value for each spectrum.

Figure 2

Figure 3.Premultiplied two-dimensional spectra in self-similar coordinates for $k_zh = 14$, $k_z^+ = 0.0027$ ($\lambda _z = 0.45h$, $\lambda _z^+ = 2333$) of (a) streamwise, (b) wall-normal and (c) spanwise forcing components and (d) the total forcing spectra. The contour levels are separated by 0.1 times the maximum value for each spectrum. The horizontal dashed lines show $y/\lambda _z = 0.5$ and the vertical dashed lines show $\lambda _x/\lambda _z = 1$.

Figure 3

Figure 4.Premultiplied two-dimensional spectra in self-similar coordinates for $k_z = 14/h$, $k_z^+ = 0.0027$ ($\lambda _z = 0.45h$, $\lambda _z^+ = 2333$) of (a–c) streamwise, (d–f) wall-normal and (g–i) spanwise velocities and (j–l) Reynolds stress with the forcing intensity (a,d,g,j) $\boldsymbol \varOmega _u$, (b,e,h,k) $\boldsymbol \varOmega _v$ and (c,f,i,l) $\boldsymbol \varOmega _w$. Here the contour levels are separated by 0.10 times the maximum value of the spectra with total forcing intensity $\boldsymbol \varOmega$.

Figure 4

Figure 5.Premultiplied two-dimensional spectra in self-similar coordinates: (a,d,g,j) $k_zh = 6$ ($\lambda _z/h \approx 1.05$); (b,e,h,k) $k_z h= 30$ ($\lambda _z/h \approx 0.21$ or $\lambda _z^+ \approx 1100$); (c,f,i,l) $k_z^+ \approx 0.0064$ ($\lambda _z^+ \approx 100$). (a–c) Streamwise velocity spectra, (d–f) wall-normal velocity spectra, (g–i) spanwise velocity spectra and (j–l) Reynolds shear stress co-spectra. The shaded contours indicate DNS spectra and the solid contours are optimisation results from (2.19).The contour levels are separated by 0.20 times the maximum value of the spectra with forcing intensity $\boldsymbol \varOmega$ for comparison.

Figure 5

Table 1.The optimisation errors defined in (4.1)–(4.3) for the different spanwise wavelengths. Regularisation weighting $\gamma$ and number of sine polynomials used $N_s$ are reported together.

Figure 6

Figure 6.Premultiplied two-dimensional spectra in self-similar coordinates for the (a) streamwise, (b) wall-normal and (c) spanwise forcing components and (d) total forcing. Here the spanwise length scales are $k_zh = 14$ (shaded), $30$ (solid), $50$ (dashed). The contour levels are chosen to be $0.2$ and $0.8$ times the maximum value.

Figure 7

Figure 7.Premultiplied two-dimensional spectra in self-similar coordinates for $k_zh = 14$, $k_z^+ = 0.0027$ ($\lambda _z = 0.45h$, $\lambda _z^+ = 2333$) of (a) streamwise, (b) spanwise, (c) wall-normal and (d) energy inner-product weighted wall-normal solenoidal forcing spectra. The contour levels are separated by 0.1 times the maximum value for each spectrum. The horizontal dashed lines show $y/\lambda _z = 0.5$ and the vertical dashed lines show $\lambda _x/\lambda _z = 1$.

Figure 8

Figure 8.Premultiplied two-dimensional solenoidal forcing spectra in self-similar coordinates for (a) the wall-normal component and (b) the wall-normal component with respect to the energetic inner product. The spanwise length scales are $k_zh = 14$ (shaded), $30$ (solid), $50$ (dashed). The contour levels are chosen to be $0.2$ and $0.8$ times the maximum value.

Figure 9

Figure 9.Premultiplied spectra for the effective lift-up forcing mechanism, $k_z/k^2 (\mathcal {D}U)\varPhi _{\hat {v}\hat {v}}^{\infty }$, in self-similar coordinates. The spanwise length scales are $k_zh = 14$ (shaded), $30$ (solid), $50$ (dashed). The contour levels are chosen to be spaced out by 0.2 times the maximum value.

Figure 10

Figure 10.Premultiplied two-dimensional spectra in self-similar coordinates for $k_zh = 14$, $k_z^+ = 0.0027$ ($\lambda _z = 0.45h$, $\lambda _z^+ = 2333$) associated with (a) $k_x^2/k^4 R_{(\mathcal {D}\hat {v})(\mathcal {D}\hat {v})}$, (b) $k_z^2/k^4 R_{\hat {\eta }\hat {\eta }}$, (c) $k_x^2/k^4 R_{\hat {\eta }\hat {\eta }}$ and (d) $k_z^2/k^4 R_{(\mathcal {D}\hat {v})(\mathcal {D}\hat {v})}$. The contour levels are chosen to be 0.05 and 0.1 times the maximum value of the streamwise velocity spectra in (a), and separated by 0.1 times the maximum value of the spanwise velocity spectra in (b–d).

Figure 11

Figure 11.The variation of the errors in velocity spectra. (a) The relative error between normalised spectra using the $Q$-norm defined in (4.2) and (b) the ratio of the $Q$-norms between the linear model velocity spectra and the DNS spectra for the streamwise (solid), wall-normal (dashed) and spanwise (dash-dotted) velocity spectra and Reynolds shear stress co-spectra (dotted) defined in (4.3).

Figure 12

Figure 12.Premultiplied two-dimensional spectra for the solenoidal (a,c,e) streamwise and (b,d,f) wall-normal forcing components in self-similar coordinates at $k_z = 30/h$ ($\lambda _z = 0.20h$; $\lambda _z^+ = 1100$) for (a,b) $\alpha = 0$, (c,d) $\alpha =0.1$ and (e,f) $\alpha =0.4$. The contour levels are separated by 0.10 times the maximum value for each spectrum.

Figure 13

Figure 13.Premultiplied two-dimensional spectra for the (a,d,g) streamwise and (b,e,h) wall-normal velocities and (c,f,i) Reynolds shear stress co-spectra in self-similar coordinates at $k_z = 30/h$ ($\lambda _z = 0.20h$; $\lambda _z^+ = 1100$) for (a–c) $\alpha = 0$, (d–f) $\alpha =0.1$ and (g–i) $\alpha =0.4$. The contour levels are separated by 0.10 times the maximum value for each spectrum.

Figure 14

Figure 14.The effect of the regularisation term on the relative error for $k_z = 30/h$, $k_z^+ = 0.0058$ ($\lambda _z \approx 0.21h$, $\lambda _z^+ \approx 1100$). The relative error from (4.1) $\tilde {\epsilon }_s$ is shown for the streamwise (solid), wall-normal (dashed); spanwise (dash-dotted) and Reynolds shear stress (dotted) spectra.

Figure 15

Figure 15.Premultiplied two-dimensional spectra in self-similar coordinates for $k_zh = 14$ ($\lambda _z = 0.45h$, $\lambda _z^+ = 2333$) associated with the contribution of $- k_xk_z/k^4R_{(\mathcal {D}\hat {v})\hat {\eta }}^{\infty } -k_xk_z/k^4R_{\hat {\eta } (\mathcal {D}\hat {v})}^\infty$. Here low-level contributions are visualised, separated by $6\times 10^{-4}$ times the maximum value of the total streamwise velocity spectra ranging up to 0.06 times the maximum value of the total streamwise velocity spectra.