Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: The sensitivity of the Orr–Sommerfeld equation due to base flow deviations is investigated by means of
Available online 17 June 2017 a Monte Carlo-type perturbation strategy and a Chebyshev collocation method. In particular, the sensi-
tivities of the peak growth rate and frequency, both of which are of importance in airfoil applications,
Keywords:
are investigated in the low Reynolds number regime. A separated boundary layer with a nominal shape
Transition
Flow instability
factor of H = 5.9 was perturbed for both velocity and wall-normal position deviations. Wide bands of
eigenvalue spectra are obtained due to both types of perturbations. The standard deviation of the peak
growth rate and frequency due to both perturbations does not exhibit a pronounced Reynolds number de-
pendence. To broaden the results, six boundary layers were investigated with shape factors ranging from
H = 5.9 − 22. Perturbations resulting in a standard deviation of 1% of the nominal shape factor were ap-
plied. It is found that sensitivities of both the peak growth rate and frequency are more pronounced at
lower shape factors, with a decrease in sensitivity with increasing shape factor. This result suggests that,
at low Reynolds numbers, boundary layers with larger separated regions are less sensitive to base flow
perturbations.
© 2017 Elsevier Inc. All rights reserved.
v ( x, t ) = V ( y ) + v ( x, t ) , p(x, t ) = P (x ) + p (x, t ), (1) The solution of the Orr–Sommerfeld equation dictates the sta-
where V and P are the mean or base flow velocity and pressure, bility of a base flow, U. Eq. (3) has been used to analyze a number
respectively, and v and p are the small perturbations. The pertur- of flows and has provided fairly accurate predictions (Yarusevych
bations are assumed of the form, et al., 2006; Boutilier and Yarusevych, 2012; Nishioka et al., 1990;
Reed et al., 1996; Orszag, 1971).
v (x, y, t ) = vˆ (y )ei(αx−ωt ) , (2) More sophisticated analyses have been performed in recent
years, such as BiGlobal and TriGlobal stability analysis (Theofilis,
∗
Corresponding author. 2003, 2011; Zhang and Samtaney, 2016). These analyses require
E-mail addresses: paul.ziade@ucalgary.ca, paul.ziade@mail.utoronto.ca (P. Ziadé). the accurate computation of the entire flow field by either direct
http://dx.doi.org/10.1016/j.ijheatfluidflow.2017.06.005
0142-727X/© 2017 Elsevier Inc. All rights reserved.
P. Ziadé, P.E. Sullivan / International Journal of Heat and Fluid Flow 67 (2017) 122–130 123
numerical simulation (DNS) or high-resolution large-eddy simula- The transformed equations were solved at the Gauss–Lobatto
tion (LES) and allow for the stability computation in two and three points,
dimensions. Both TriGlobal and BiGlobal analysis lead to very large
πj
eigenvalue problems and are computationally expensive. Further- η j = cos , j = 0, 1, . . . N, (5)
more, these methods cannot be used with experimentally-obtained N
base flow data, which does not describe the entire flow field, is
where N is the number of points in the Chebyshev domain. The
prone to data scatter, and does not have adequate resolution.
vertical velocity fluctuation in (3) is now expressed as a Chebyshev
Several papers have reported the sensitivity of this one-
polynomial series:
dimensional linear stability analysis. Bottaro et al. (2003) have
examined the effect of base flow variation using a variational
N
be destabilizing. The most perturbation-sensitive eigenvalues for The applicable boundary conditions in Chebyshev space are Schmid
plane Poiseuille flow have been identified by Reddy et al. (1993) as and Henningson (2001),
those nearest to the intersection of the three branches of the Orr–
Sommerfeld spectrum. The nonnormality of the Orr–Sommerfeld
N
N
N
an Tn (1 ) = 0; an Tn (−1 ) = 0; an Tn (1 ) = 0;
operator has been identified as being a key contributor to the pro-
n=1 n=1 n=1
nounced eigenvalue sensitivity (Reddy et al., 1993; Schmid et al.,
1993; Trefethen and Embree, 2005). Non-normal matrices and op-
N
an Tn (−1 ) = 0, (7)
erators are those that do not commute with their adjoint; that is,
n=1
AA∗ = A∗ A, where A∗ is the conjugate transpose. These operators
have nonorthogonal eigenfunctions. The analysis of the sensitiv- which corresponds to vˆ (±1 ) = vˆ (±1 ) = 0. The above Chebyshev
ity of these types of operators has been performed using matrix expansions with appropriate boundary conditions were substi-
perturbation techniques and the -pseudospectrum (Trefethen and tuted into the Orr–Sommerfeld equation. A fourth order polyno-
Embree, 2005). A detailed discussion regarding the nonnormal- mial eigenvalue problem with matrix coefficients, Cj , and α as the
ity of the Orr–Sommerfeld operator can be found in Reddy et al. parameter is obtained (Bridges and Morris, 1984; Morris, 1992):
(1993) and Schmid et al. (1993). More recently, Boutilier and Yaru-
D4 ( α ) a = 0 , (8)
sevych (2013) examined the effect of the method of analysis on
the eigenvalue sensitivity. In this latter study, it was shown that where
the typical data scatter in experimental measurements as well as
the curve fitting approach must be considered when performing D4 ( α ) = C 0 α 4 + C 1 α 3 + C 2 α 2 + C 3 α + C 4 . (9)
linear stability analysis. This can be reformulated using the companion matrix method to
Linear stability analysis using the Orr–Sommerfeld equation re- yield a complex generalized eigenvalue problem (Bridges and Mor-
mains an important tool in aerodynamics. This present work was ris, 1984):
motivated by a recent study comparing the LES computations of ⎛⎡ ⎤ ⎡ ⎤⎞
a low-Reynolds number airfoil with hot-wire measurements con- −C1 −C2 −C3 −C4 C0 0 0 0
⎜⎢ I 0 0 0 ⎥ ⎢0 I 0 0⎥⎟
ducted in a low-turbulence recirculating wind tunnel at the Uni-
⎝⎣ 0 − α⎣
versity of Toronto. Despite good agreement between the compu- I 0 0 ⎦ 0 0 I 0⎦⎠
tations and experiment, minor deviations in base flow resulted in 0 0 I 0 0 0 0 I
⎧ 3 ⎫
⎨ α 2 a⎪
large discrepancies in predicted disturbance growth rates using the ⎪ ⎬
Orr–Sommerfeld equation. This paper is an attempt to quantify the α a
× = 0.
⎩ αa ⎪
effects of base flow variation on the resulting growth rates pre-
⎪ ⎭
dicted by linear stability analysis, with particular attention to air- a
foil applications. The quantification of the sensitivity due to inflex-
ion point location variation, overall level of velocity deviation, and The eigenvalues, α , were computed using the QR algorithm. The
boundary layer shape factor is sought. In addition, this paper seeks results presented in this paper were computed using 150 Cheby-
to determine whether the sensitivity to base flow variation has any shev polynomials. It was found that for a given level of base flow
dependence on the Reynolds number. variation and all other parameters remaining fixed, the relative dif-
ference in growth rate in going from N = 150 to N = 200 polyno-
mials is less than 1 percent. Excellent accuracy using relatively few
Chebyshev polynomials has been reported by several researchers
2. Orr–Sommerfeld equation solution methodology (Khorrami, 1991; Khorrami and Malik, 1993; Trefethen et al., 1999).
The Orr–Sommerfeld equation (Eq. (3)) was solved using a 3. Base velocity profile
Chebyshev collocation method. It has been shown by Orszag
(1971) as well as others (Biringen and Danabasoglu, 1988; Lee and The separated shear layer profile used in this study as the base
Reynolds, 1967; Morris, 1992) that accurate and efficient solutions flow is that over a NACA 0025 airfoil at a chord Reynolds number
to the Orr–Sommerfeld equation can be obtained using Chebyshev of Rec = 10 0, 0 0 0 and an angle-of-attack of 5° obtained by hot-wire
expansions. The semi-infinite domain of the shear layer profile, y measurement. This profile was taken at x = 0.35c, where c is the
∈ [0, ∞), was mapped onto the Chebyshev polynomial domain, chord length, and is just post-separation. This boundary layer has a
η ∈ [−1, 1], by an algebraic mapping function, f(η). Algebraic map- nominal shape factor of H = 5.9. The average step size in the wall-
pings have been shown to be superior to logarithmic mappings normal direction is 0.13mm. It has been shown by Boutilier and
(Schmid and Henningson, 2001; Boyd, 2001; Malik, 1990). The nth Yarusevych (2013) that analytical velocity profiles provide better
Chebyshev polynomial of the first kind is defined as: stability results than using discrete velocity points in the stability
analysis. The curve fit by Dovgal et al. (1994) (Eq. (10)) was fairly
Tn (η ) = cos(n cos−1 η ). (4) insensitive to data scatter (Boutilier and Yarusevych, 2013) and was
124 P. Ziadé, P.E. Sullivan / International Journal of Heat and Fluid Flow 67 (2017) 122–130
a given frequency coincide with the upper and lower curves of for each realization. The standard deviation of the nondimen-
the envelope, respectively. Fig. 5 shows the eigenvalue envelope sional peak growth rate at the three levels of refinement is pre-
for σU = 1.25% at a Reynolds number of Rec = 10 0, 0 0 0, computed sented in Fig. 6. The resulting standard deviations are herein
with 300 realizations. denoted by std to avoid confusion with the imposed level of
Of particular interest is the variability of peak growth rates con- scatter, σ . There is a pronounced variability in predicted peak
tained within the envelope. The peak growth rate was recorded growth rates in going from Ns = 150 to Ns = 300 realizations;
126 P. Ziadé, P.E. Sullivan / International Journal of Heat and Fluid Flow 67 (2017) 122–130
Fig. 7. Growth rate spectra for a) Rec = 10 0,0 0 0, b) Rec = 50 0,0 0 0, c) Rec = 1,0 0 0,0 0 0 due to velocity profile perturbations.
Fig. 8. Standard deviation of resulting maximum growth rates due to velocity per-
turbations.
Fig. 10. Growth rate spectra for a) Rec =10 0,0 0 0, b) Rec = 50 0,0 0 0, c) Rec = 1,0 0 0,0 0 0 due to wall-normal position perturbations.
Fig. 11. Standard deviation of resulting maximum growth rates due to wall-normal
position perturbations.
Fig. 12. Standard deviation of peak disturbance frequency due to wall-normal po-
sition perturbations.
Fig. 14. Standard deviation of nondimensional growth rates at various shape factors Fig. 15. Standard deviation of nondimensional peak frequency at various shape fac-
and Rec = 10 0,0 0 0, 50 0,0 0 0 and 1,0 0 0,0 0 0. tors and Rec = 10 0,0 0 0, 50 0,0 0 0 and 1,0 0 0,0 0 0.
Boutilier, M.S.H., Yarusevych, S., 2012. Separated shear layer transition over an airfoil
at a low reynolds number. Phys. Fluids 24 (8). doi:10.1063/1.4744989.
Boutilier, M.S.H., Yarusevych, S., 2013. Sensitivity of linear stability analysis of mea-
sured separated shear layers. Eur. J. Mech. B/Fluids 37, 129–142. doi:10.1016/j.
euromechflu.2012.09.003.
Boyd, J.P., 2001. Chebyshev and Fourier Spectral Methods, second ed. Dover Publica-
tions Inc.
Bridges, T.J., Morris, P.J., 1984. Differential eigenvalue problems in which the
parameter appears nonlinearly. J. Comput. Phys. 55, 437–460. doi:10.1016/
0 021-9991(84)90 032-9.
Dovgal, A.V., Kozlov, V.V., Michalke, A., 1994. Laminar boundary layer separation:
instability and associated phenomena. Prog. Aerosp. Sci. 30 (1), 61–94. doi:10.
1016/0376-0421(94)90 0 03-5.
Kawall, J.G., Shokr, M., Keffer, J.F., 1983. A digital technique for the simultaneous
measurement of streamwise and lateral velocities in turbulent flows. J. Fluid
Mech. 133, 83–112. doi:10.1017/S0 0221120830 01809.
Khorrami, M.R., 1991. A Chebyshev spectral collocation method using a staggered
grid for the stability of cylindrical flows. Int. J. Numer. Methods Fluids 12 (July
1989), 825–833. doi:10.1002/fld.1650120903.
Khorrami, M.R., Malik, M.R., 1993. Efficient computation of spatial eigenvalues for
hydrodynamic stability analysis. J. Comput. Phys. 104 (1), 267–272. doi:10.1006/
jcph.1993.1026.
Lee, L.H., Reynolds, W.C., 1967. On the approximate and numerical solution of orr-
Fig. 16. Second derivative of the boundary layer. sommerfeld problems. Q. J. Mech. Appl. Math. 20 (1), 1–22. doi:10.1093/qjmam/
20.1.1.
Malik, M.R., 1990. Numerical methods for hypersonic boundary layer stability. J.
Comput. Phys. 86 (2), 376–413. doi:10.1016/0021-9991(90)90106-B.
Boundary layer profiles of different shape factors ranging from Morris, P.J., 1992. The eigenvalue spectrum of the Rayleigh equation for a plane
H = 5.9 to H = 22 were perturbed with first standard deviations shear layer. Int. J. Numer. Methods Fluids 15, 1407–1415.
Nishioka, M., Asai, M., Yoshida, S., 1990. Control of flow separation by acoustic ex-
of 1% of their nominal shape factor. For both the peak growth
citation. AIAA J. 28 (11), 1909–1915. doi:10.2514/3.10498.
rate and frequency, the boundary layers with smaller shape fac- Orszag, S.A., 1971. Accurate solution of the Orr–Sommerfeld stability equation. J.
tors (i.e., fuller boundary layers) were much more sensitive to per- Fluid Mech. 50 (04), 689. doi:10.1017/S0022112071002842.
turbations. This interesting result merits further exploration. No Reddy, S.C., Schmid, P.J., Henningson, D.S., 1993. Pseudospectra of the Orr-Sommer-
feld operator. SIAM J. Appl. Math. 53 (1), 15–47.
Reynolds number dependence was noted in the standard deviation Reed, H.L., Saric, W.S., Arnal, D., 1996. Linear stability theory applied to boundary
of these variables. This finding shows that, for the cases considered layers. Annu. Rev. Fluid Mech. 28 (1), 389–428. doi:10.1146/annurev.fl.28.010196.
in this paper, boundary layers further along the route to separation 002133.
Schmid, P.J., Henningson, D.S., 2001. Stability and Transition in Shear Flows.
are less sensitive to perturbations. Springer-Verlag.
Schmid, P.J., Henningson, D.S., Khorrami, M.R., Malik, M.R., 1993. A study of eigen-
Acknowledgments value sensitivity for hydrodynamic stability operators. Theor. Comput. Fluid Dyn.
4, 227–240. doi:10.10 07/BF0 0417929.
Simpson, R.L., Chew, Y.-T., Shivaprasad, B.G., 1981. The structure of a separating tur-
The authors gratefully acknowledge the support of the Natural bulent boundary layer. Part 1. Mean flow and Reynolds stresses. J. Fluid Mech.
Sciences and Engineering Research Council (NSERC), the Southern 113, 23–51. doi:10.1017/S002211208100339X.
Theofilis, V., 2003. Advances in global linear instability analysis of nonparallel
Ontario Smart Computing Innovation Platform (SOSCIP) and the and three-dimensional flows. Prog. Aerosp. Sci. 39 (4), 249–315. doi:10.1016/
Ontario Graduate Scholarship (OGS) program. Computations were S0376-0421(02)0 0 030-1.
performed on the GPC and BGQ supercomputers at the SciNet HPC Theofilis, V., 2011. Global linear instability. Annu. Rev. Fluid Mech. 43 (1), 319–352.
doi:10.1146/annurev- fluid- 122109- 160705.
Consortium which includes support from the Canada Foundation
Trefethen, A.E., Trefethen, L.N., Schmid, P.J., 1999. Spectra and pseudospectra for
for Innovation under the auspices of Compute Canada; the Gov- pipe poiseuille flow. Comput. Methods Appl. Mech. Eng. 175 (3–4), 413–420.
ernment of Ontario; Ontario Research Fund - Research Excellence; doi:10.1016/S0 045-7825(98)0 0364-8.
and the University of Toronto. Trefethen, L.N., Embree, M., 2005. Spectra and Pseudospectra. Princeton University
Press.
Yarusevych, S., Kawall, J.G., Sullivan, P.E., 2008. Separated-shear-layer development
References on an airfoil at low Reynolds numbers. AIAA J. 46 (12), 3060–3069.
Yarusevych, S., Sullivan, P.E., Kawall, J.G., 2006. Coherent structures in an airfoil
Biringen, S., Danabasoglu, G., 1988. Solution of the Orr-Sommerfeld Equation for boundary layer and wake at low Reynolds numbers. Phys. Fluids 18 (4).
the Blausius Boundary-Layer Documentation of Program ORRBL and a Test Case. Zhang, W., Samtaney, R., 2016. BiGlobal linear stability analysis on low-Re flow past
NASA Contractor Report 4169 (August). an airfoil at high angle of attack. Phys. Fluids 28 (4), 044105. doi:10.1063/1.
Bottaro, A., Corbett, P., Luchini, P., 2003. The effect of base flow variation on flow 4945005.
stability. J. Fluid. Mech. 476, 293–302. doi:10.1017/S0 0221120 020 0318X.