Aerosp. Sci. Technol. 5 (2001) 317328 2001 ditions scientiques et mdicales Elsevier SAS.

All rights reserved S1270-9638(01)01110-5/FLA

A numerical and experimental study of a transitional separation bubble

Carl P. Hggmark 1 , C. Hildings 2 , Dan S. Henningson 3
Department of Mechanics, KTH, Osquars Backe 18, 100 44 Stockholm, Sweden Received 28 February 2000; revised and accepted 2 May 2001

Abstract

A combined numerical and experimental study of a two-dimensional transitional separation bubble due to an adverse pressure gradient is reported. The experiments have been performed in the MTL wind tunnel with a contoured wall imposing an adverse pressure gradient on the ow over a at plate. The separated shear-layer is highly unstable and transition to turbulence occurs in the ow. The experimental separation bubble ow is modelled numerically using two-dimensional direct numerical simulations (DNS). Prescribing free stream boundary conditions in the wall normal velocity the experimental bubble is reproduced. The development of articially forced two-dimensional instability waves is investigated and good agreement is found between experiments, simulations and linear stability theory (LST). The performance of several engineering transition prediction methods applied on the present separation bubble is presented and compared. Methods based on simplications of the en -method yield predictions in accordance with experiments and DNS. 2001 ditions scientiques et mdicales Elsevier SAS transitional separation bubble / two-dimensional instability wave

1. Introduction Separation on a smooth surface is a consequence of the nite adverse pressure gradient a boundary-layer can sustain. If a separated shear-layer reattaches downstream, a separation bubble is formed. Separation bubbles can be divided into three main types laminar, transitional and turbulent depending on the status of the boundary layer at separation and reattachment. The laminar separation bubble has both laminar separation and reattachment. For the transitional separation bubble only separation is laminar, while reattachment is turbulent, whereas a turbulent separation bubble arises in a turbulent boundarylayer. The transitional separation bubble is often denoted laminar in earlier aeronautical literature. Here however, the above denitions are adopted in order to distinguish

between separation bubbles with laminar and turbulent reattachment, respectively. Separation bubbles are common in several engineering applications, e.g. airfoils, turbine blades and curved surfaces in general. Their presence has often undesirable effects, as in low Reynolds number applications like human powered aircraft, Drela [9]. One effect can be a dramatic change in pressure drag. Moreover, the highly unstable nature of the bubble leads to the development of a turbulent boundary-layer at lower Reynolds numbers, resulting in higher drag. In other situations the bubble may cause a thicker turbulent boundary layer than would prevail if transition had occurred upstream of the bubble. This is due to the rapid growth of the momentum thickness in the latter part of a transitional separation bubble. One problem associated with separation bubbles that has gained much attention is of computational character.

E-mail addresses: chaggmar@volvocous.com (C.P. Hggmark), henning@mech.kth.se (D.S. Henningson).

1 Present address: Volvo Car Corporation, Aerodynamics, Dept 92 464, PV T3, 405 31 Gothenburg, Sweden. 2 Present address: HiQ Data, Stockholm, Sweden. 3 Correspondence and reprints.

318

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

To accurately predict the performance of airfoils the transition location in the separation bubble has to be known, see for instance Cebeci [4]. Some applications, such as high-lift airfoils, are especially sensitive to problems with transition prediction. For such systems the airfoil often consists of several elements like slats and aps, where the Reynolds number for some parts may be quite low in spite of a large chord Reynolds number. Transition in a leading edge separation bubble on a slat may change the characteristics of the main airfoil considerably, as it experiences the turbulent wake of the slat. In such cases simple transition prediction methods used in older airfoil analysis codes, are not sufcient (Davis and Carter [8]). There is therefore a need for better understanding of transition and the onset of ow unsteadiness in separation bubbles in order to improve prediction models. Numerical work in this direction can be found in Gruber and Bestek [14], Bestek et al. [2], Maucher et al. [26], Theolis et al. [34] and in Tatineni and Zhong [33]. As a background a brief overview of some of the existing models and methods used in transition prediction in separation bubbles is given below. In Hildings [17] several existing methods in the literature are described in more detail. 1.1. Engineering transition prediction methods for separation bubbles Most transition models for transitional separation bubbles are presented as a part of a program designed to calculate the ow and performance characteristics of an airfoil. A correctly predicted transition point in the bubble and the following evolution of turbulence is one of the most important factors, as recognized by for instance Vatsa and Carter [39], Drela and Giles [10] and Cebeci et al. [6]. To a rst approximation, transition in the separation bubble can be assumed to occur at the point of separation (Eppler [13], Kusunose and Cao [21], Vatsa and Carter [39]). Several other methods explicitly predict the distance between the separation and transition point, among them von Doenhoff [40], Horton [18], OMeara and Mueller [28] and Roberts [31]. In the eld of transition prediction the en -method is widely used. One of the developers of the en -method, van Ingen, has extended the method to separation bubble ows (van Ingen [3537] and van Ingen and Boermans [38]). To be a valuable engineering tool for transition prediction a method has to be computationally fast. To achieve this, most methods using the en -method are based on stability calculations of families of known velocity proles. These results are then included in a simple criterion for transition. Thus the model given does not depend on the actual ow but rather a simplied ow, believed to approximate the real ow. For separation bubbles the family of velocity proles most often used is that of Stewartson [32]. This is a second set of solutions to the

FalknerSkan equation with reverse ow. Similar methods have been devised by Cebeci et al. [6] and Cebeci and Chen [5] and for use in airfoil codes by Drela [9]. 1.2. Present study In this study we use wind tunnel experiments and two-dimensional direct numerical simulations to study a transitional separation bubble ow. The development of two-dimensional wave disturbances is investigated with the main emphasis put on a detailed modelling of the experiment in the computations enabling detailed comparisons. Furthermore, the performance of existing transition prediction methods for separation bubbles is evaluated and discussed. 2. Experimental and numerical technique 2.1. Description of the experiment The experimental part of the work was carried out in the MTL wind tunnel at KTH, Stockholm. The MTL wind tunnel features a low noise and disturbance level, T u = 0.02%, qualities which are important for transition experiments in general and for experiments on separation bubbles in particular. A curved wall insert, mounted at the upper wall inside the test section, establish a steady adverse pressure gradient on a at plate, positioned opposite to the curved wall. A sketch of the experimental set-up is shown in gure 1. In order to avoid separation at the curved wall suction is applied at the leeward side through a porous section of the wall. Hereby, the ow over the at plate will separate due to the adverse pressure gradient and reattach to the surface further downstream. The line of separation is steady and highly two-dimensional. Figure 2 shows a smoke visualization of the separation bubble in the experiment. The free stream velocity at the test section inlet was held constant at 7 m/s in the experiments and served as a reference velocity for the numerical computations. A Cartesian coordinate system is used with its origin at the leading edge centerline and with axes x, y and z, corresponding to the streamwise, wall normal and spanwise directions, respectively. The velocity components in

Figure 1. Schematic picture of the experimental set-up in the MTL wind tunnel. The dimensions of the test section cross section are 1.2 0.8 m2 .

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

319

Figure 2. Smoke visualization picture of the separation bubble in the experiments. Top streamwise ruler labelled in mm from the leading edge. 20 mm marks indicated on right hand side spanwise ruler. Separation and reattachment approximately at 700 and 900 mm, respectively.

these directions are u, v and w, respectively. An overline denotes dimensional quantities and capital letters denote mean quantities. 2.1.1. Measurement technique The instantaneous ow velocity was determined with hot-wire anemometry. A single hot-wire probe was used for measurements of u in the separation bubble and in the separated shear layer as well as in the boundary layer upstream and downstream of the bubble. Measurements of both the streamwise and wall normal velocity components, u and v were made with an X-wire probe in a region further out from the wall, conned by the lines x = 0 mm, x = 1800 mm, y = 7 mm and y = 55 mm. The number of measurement gridpoints in this region was 47 by 25 in the streamwise and wall

normal direction, respectively. The streamwise spacing between two measurement points, x, was x = 50, 25, 20 or 10 mm, with the denser spacing at the reattachment region. The measurement grid for the X-wire measurements is marked in gure 3. These data do not give information on the bubble itself but provides boundary conditions for the numerical simulation. The single wire measurements reached from x = 0 mm to x = 1000 mm, one measurement series typically contained 40 wall normal velocity proles each with 25 measurement points, with x = 10 mm in the rearward part of the separation bubble. All the reported hot wire measurements were made at the centerline plane, z = 0. The hot wires were calibrated in a position upstream of the curved wall against a Pitot tube. For the single wire calibration a modied Kings law was tted to the calibration data, (Johansson and Alfredsson [19]). An angle calibration was carried out for the X-wire probe, with a variation of the angle of attack of 24 in steps of 2 and two fth order polynomial surfaces were tted to the calibration data. The accuracy of the calibration was within 1% for all calibration points except at velocities below 1 m/s where it increased up to 5%. The hot-wire probes were attached to a computer controlled ve-axis traversing mechanism which enabled automatic traversing during measurements and angle calibration of the X-wire. Controlled two-dimensional wave disturbances could be generated in the boundary layer at x = 189 mm by suction and blowing through a slit in the plate. For this purpose loudspeakers and power ampliers were used. Flexible hoses transmitted the forced perturbations from the loudspeakers to a number of individual pipes mounted below the slit in the plate. A detailed description of the wave disturbance generation technique is found in Elofsson [12]. In this study low-amplitude two-dimensional single-frequency waves were generated. The forcing fre-

Figure 3. Measurement grid for X-wire measurements of U and V . The separation bubble and the boundary layer edge, , are marked with solid lines. The location of the computational domain and fringe region is marked with solid boxes for the case with a boxlength of 1000.

320

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

quency f of the waves was chosen in the frequency band of the most amplied disturbances of the natural separation bubble ow at unforced conditions, Hggmark et al. [15]. The frequency in the experiments was f = 95 Hz. The amplitude was measured by the rms value, ltered around the forcing frequency, of the uctuating streamwise velocity component. The disturbance generation, the data aquisition and the traversing equipment were all controlled by the same computer. 2.2. The numerical code and modelling of the experiment Two-dimensional calculations were performed with a modied version of the spectral code of FFA/KTH, developed by Lundbladh et al. [2224]. It solves the NavierStokes equations in the velocity-vorticity formulation using Fourier series in the spanwise and streamwise directions and Chebyshev series in the normal direction. The algorithm is similar to that of Kim et al. [20]. Time advancement is carried out with a semi-implicit scheme. The advective terms are discretized using a third order four stage RungeKutta scheme, while the diffusive terms use a second order implicit CrankNicholson scheme. The time-step is regulated at 90% of the spectral CFL limit. The modications made for the twodimensional simulations in this investigation consisted in new boundary conditions that improved the possibilities to prescribe the streamwise free-stream velocity. A Cartesian coordinate system is used in the computations which is translated a distance x 0 = 400 mm in relation to the experimental coordinate system. The velocity components in the streamwise, x, and wall-normal direction, y, are denoted u and v, respectively. The ow variables and governing equations are expressed in non dimensional form by using a length scale 0 and a velocity scale U0 , corresponding to the boundary layer displacement thickness and the streamwise edge velocity in the experiment at the origin of the computational coordinate system, x = x0 , together with a Reynolds number Re = U0 0 /. Thus, x= u= x x0 , 0 u , U0
0

Table I. The different numerical parameters used for the calculations. Box length 1600 1200 1000 1000 nx 2048 3072 2048 3072 Resolution ny 97 97 97 97

where f is the dimensional forcing frequency, f = 95 Hz. The amplitude of the disturbance, u, was prescribed to match the corresponding value in the experiment. To efciently generate harmonic wave disturbances their wavelength has to be known. It was calculated from r , using x = 2/r , which in turn was computed using linear stability theory on the boundary-layer prole at the point of generation. From the measurements urms was known which, for sinuous disturbances, corresponds to u/ 2, where u is the amplitude of the disturbance. The ow at the end of the computational box is returned to the specied inow boundary condition by the fringe region technique (Lundbladh et al. [24]). The fringe is a volume force applied in a part of the computational box which is added to the physical region. The NavierStokes equations, together with the fringe formulation can be written, Du = NS(u) (x)(u u ), Dt (3)

y= v . U0

y , 0 (1)

where u = (u, v) is the velocity eld solved for. u denotes the solution forced to in the fringe region and is the strength of the fringe, which is zero in the computational domain. For an analysis of the fringe region technique see Nordstrm et al. [27]. Calculating ows involving transitional separation bubbles requires the best possible damping of the fringe region as a consequence of the rapid amplication of disturbances in the ow. In table I the different numerical parameters used for the results presented are shown. The location of the computational domain and the fringe in relation to the experiment is shown in gure 3. 2.3. Boundary conditions Inow and outow conditions are periodic, which are enforced by the fringe technique. The inow boundary condition in the computations was determined by tting the free stream velocity distribution in the favourable pressure gradient region in the experiment to a Falkner and Skan similarity solution, U = Cx m , where the virtual origin of the similarity solution can be determined from knowledge of the displacement thickness calculated from

v=

= 1.185 mm, U0 = 8.15 m/s In the computations, and Re = 644. Controlled wave disturbances were introduced in the computations by an oscillating volume force. The amplitude and frequency in the computations were scaled to match the experimental values, using F = /Re,
= 2f 0 /U0 ,

(2)

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

321

the experiment. A value of m = 0.0909, equivalent of a 30 wedge, gave good agreement with the experimental free stream velocity distribution and was therefore chosen to prescribe the inlet velocity prole corresponding to the forced solution in the fringe, u . At the wall of the at plate no-slip boundary conditions are used. Several free stream boundary conditions were investigated. It was found that by using the boundary condition, proposed by Coleman and Spalart [7], prescribing the normal velocity and zero spanwise disturbance vorticity, it was easier to obtain a separation bubble in agreement with the experiment. For the two-dimensional cases reported here these boundary conditions read Du vx = DU Vx , v = V. (4)

2.4. Code verication for boundary layer ows with separation The numerical code has been extensively used in treating transitional boundary layer and channel ows, (see, e.g., Henningson et al. [16] and Berlin et al. [1]). In the present study the code was modied in order to deal with separated ows and it was veried against the DNS-study of a separation bubble ow in Rist and Maucher [30]. In the work of Rist and Maucher a varying free-stream velocity is prescribed over a at plate. A reference length L = 0.05 m is introduced, which is used to scale all variables and to dene a global Reynolds number Re = U0 L/ = 105 (U0 = 30 m/s, = 1.5 105 m2 /s, L = 0.05 m). The computational box dened starts at x 0 = 0.37L and a decrease of the free-stream velocity of 9% is prescribed using a fth order polynomial from x 1 = 0.71L to x 2 = 2.43L. The spectral code uses a scaling based on the displacement thickness at the inow of the computational domain. The step in free-stream velocity would therefore start at x = (x 1 x 0 )/0 and end at x = (x 2 x 0 )/0 . Using
0 = 1.72

A second boundary condition used is the Neumann condition, which can be described by, Du = DU, Dv = DV , (5)

where U and V are the base ow velocity components. A third boundary condition prescribes the ow to asymptotically approach the solution of the inviscid linearized equations. This boundary condition has proved to be effective as it allows the use of a lower computational box when TS-waves are generated. This becomes Du + u = DU + U, Dv + v = DV + V , (6) where is the streamwise wavenumber. A fourth boundary condition, used less frequently, prescribes the streamwise free stream velocity as well as vanishing spanwise disturbance vorticity at the upper boundary, i.e. u = U, u v = DU Vx . y x (7)

x 0 U0

(8)

A problem with boundary conditions prescribing the normal velocity is that it is very difcult to know what kind of in- or outow will take place in the fringe part of the upper boundary. This was usually handled by calculating a bubble of similar size using the boundary condition prescribing the streamwise velocity. The resulting v at the upper boundary of the fringe was then used in the boundary condition prescribing the normal velocity. In the physical part of the computational domain new values for v can then be prescribed, e.g. using the measured values. Due to the streamwise periodicity, it is necessary that the total mass ow through the upper boundary is zero, i.e. the integral of v at the upper boundary with respect to x must be zero. To ensure this, v needed to be slightly adjusted in the physical part of the domain when measured values of v was used as boundary conditions.

one nds that Re0 = 1.72 x 0 Re/L = 331 and that the step will start at (x 1 x 0 )Re/(Re0 L) and end at L), i.e. x1 = 103 and x2 = 622. The (x 2 x 0 )Re/(Re0 height of the computational box was 180 . The numerical parameters used in this test simulation are not contained in table I. For these computations the initial condition was a Blasius boundary layer. In the beginning of the computation the streamwise free-stream velocity of Rist and Maucher is instantaneously introduced. This gradually changes the ow eld, until a laminar separation bubble is formed. This bubble continues to grow until it becomes larger than the bubble reported by Rist and Maucher [30]. As the bubble attains its maximum size it starts to shed vortices. The development of the separation bubble is very sensitive to parameter variations, such as size and shape of the prescribed free-stream velocity. For example, if this step is governed not by a fth order polynomial, but instead by a third order polynomial, no bubble appears at all. This is a result of the lower gradient in the third order polynomial. The evolution of the bubble is also inuenced by the boundary condition used at the upper boundary. If one uses the boundary condition with vanishing derivatives at y = ymax , equation (5), a quicker growth of the bubble is found than with the use of the condition with vanishing vorticity and prescribed streamwise freestream velocity at the upper boundary, equation (7). The

322

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

Figure 4. Growth rates for most amplied wave length. : Rist and Maucher [30], : computed with present code, : computed with linear stability theory.

asymptotic boundary condition, equation (6), is not possible to use from the start of a calculation as one needs to know the correct base ow for it to work. It is however possible to impose this condition at a later time, after the bubble is established. The growth of the bubble is then stabilized. By changing boundary condition in this manner it was possible to achieve a stable bubble very similar to the one reported by Rist and Maucher. As the last part of these computations, low-amplitude wave disturbances were introduced upstream of the bubble by an oscillating volume force. In gure 4 the growth rate for the frequency found to be the most unstable by Rist and Maucher is plotted together with the ndings of this investigation. Also shown are results from linear stability calculations using an Orr Sommerfeld solver with mean elds extracted from the computations. The good agreement shows that the bubble of Rist and Maucher is well reproduced. It should be noted that Rist and Maucher also used the asymptotic boundary conditions in their calculation, which according to Rist (private communication) also stabilized their bubble. Thus, this points to the extreme care one needs to take in order to match all parameters when comparisons between investigations of separated ows are made. 3. Results 3.1. Mean and disturbance ow elds 3.1.1. Velocity distribution above the bubble The streamwise and wall normal mean ow variations in the ow outside the separation bubble and the separated shear layer are shown in gure 5. These distributions were measured with an X-wire probe at a constant height, y = 15.9 mm, above the plate, corresponding to one row of the gridpoints shown in gure 3. The

Figure 5. Streamwise () and wall normal () experimental mean velocity distributions outside the boundary layer edge. The dashed lines represent the smoothed prescribed distributions used as free stream boundary conditions in the computations.

U -component reaches a maximum value shortly downstream of x = 0, in order to level out after x = 300, in a region known as the pressure plateau. Further downstream a strong velocity decrease occurs associated with reattachment. The V -component, the lower curve in gure 5, increases beyond separation and exhibit a sharp minimum in the vicinity of reattachment. Based on measurements of U and V at various wall normal positions in the outer ow, free stream boundary conditions could be prescribed in the computations. Smooth velocity distributions used in the computations are shown in gure 5. There are some variations in velocity between neighbouring positions in the measurements; these are not well suited to use directly in the expressions for the boundary conditions. To smooth the distribution of U an adjusted curve was calculated using a least square t. A typical result of this treatment is shown in the top graph in gure 5. For the measured data of V slight changes of certain values were made individually. Most of the altered points are at the end or just downstream of the separation bubble, as is seen in gure 5, bottom graph. 3.1.2. Experimental ndings In gure 6, displaying contours of U , the mean ow structure at the centerline in the experimental separation bubble is shown. The natural separation bubble ow, top graph of gure 6, exhibits a low velocity region, the actual separation bubble, bounded by a separating and reattaching shear layer. The lower graph in gure 6 shows the same contour levels for the disturbed bubble, i.e. with controlled forcing of low-amplitude instability waves present. In the disturbed separation bubble a clear change in mean ow structure is observed, in terms of

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

323

gure 7, showing contours of the overall urms in the ow. The locations of the separating and reattaching shear layer are indicated for reference by two U -velocity contours at the levels of 0.2 and 0.8, marked with dashed lines. A strong exponential growth of urms is found in the ow with a maximum in the separated shear layer. The r.m.s. uctuations reach a maximum downstream of the bubble at a level of 17%. 3.1.3. Computational results The rst attempts to reproduce the experimental bubbles were done by prescribing the measured streamwise velocity at the upper boundary, which resulted in a bubble placed downstream of the experimental bubble. Prescribing the normal velocity instead, which was done using the procedure described at the end of section 2.3, was found to give much better agreement. However, the size of the bubble computed this way was smaller than the one measured without added articial disturbances. The measured bubble with disturbances, however, are much closer to the computational bubble, while the size of the computational bubble with and without disturbances is approximately unchanged. It was possible to obtain the larger undisturbed experimental bubble in the computations by articially multiplying the measured values of V used in the boundary conditions with a factor of 1.4. In this manner the details of both the large and the small experimental bubbles could be reproduced in the calculations. Thus, the boundary condition prescribing V xes the size of the bubble in a much stronger manner than is done by the corresponding boundary conditions in the experiment. We will return to this question in the next section. It was found that a stable computational bubble was possible to obtain by repeatedly computing the time averaged bubble and to use this eld as the initial condition in a new calculation. This iterative procedure leads to very small background disturbances. As a result it is possible to introduce controlled disturbances in an accurate way. The bubble obtained in the computations with the measured V multiplied by 1.4 will be denoted large while the one obtained by the measured values directly will be denoted small. 3.2. Development of controlled wave disturbances In the experiments a disturbance of 95 Hz was found in the frequency range of the most amplied disturbances. A disturbance with the corresponding frequency and amplitude was introduced in the computations. This was done at the streamwise position where the measured disturbance showed the smallest amplitude, i.e. x = 500 mm or x = 84 in the computational box. Disturbances were generated with v prescribed directly from measurements. In gure 8 an undisturbed eld is shown together with

Figure 6. Mean streamwise velocity contours in the experiment at natural, unforced conditions (top graph) and at forced conditions (bottom graph). Contour levels of U0 : 10, 20, 30, 40, 50, 60, 70, 80 and 90%.

Figure 7. Contours of overall streamwise r.m.s. velocity uctuations in the experiment at unforced conditions. Contour spacing in logarithmic scale. The dashed lines represent U -velocity contours at levels 20 and 80%.

a reduction in both length and height of the separated region. The region in the vicinity of the separation point is however largely unaffected. This mean ow effect has been documented before in separated ows, see Dovgal et al. [11], Rist and Maucher [30] and Hggmark et al. [15]. The unforced experimental separation bubble is inherently unstable, with growing natural wave disturbances in the separated shear layer, which eventually lead to turbulent breakdown, gure 2. A quantitative picture of the growth of disturbances at natural conditions is given in

324

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

Figure 9. Amplitude curves for controlled instability waves in the separation bubble ow. : experiment, : DNS with the same disturbance level as measurements, - - -: linear disturbance from DNS and - - -: linear stability theory. For the linear disturbance and LST the levels have been normalized to the same level as the measured one.

Figure 8. Contours of streamwise velocity with and without forced disturbances. Top gure, time averaged stable bubble resulting from prescribing a normal velocity distribution adjusted for best t with the measured displacement thickness. Middle gure, instantaneous picture of the same bubble. Bottom gure, time averaged disturbed bubble.

a disturbed and a time-averaged disturbed eld. It is apparent that the computational bubble gets shorter due to the introduced disturbance, but that its thickness remains approximately the same. The amplitude evolution of the introduced disturbances is compared with the experiment in gure 9. In the computations of the small bubble it was possible to study a disturbance that remained linear through the whole bubble. As is seen the agreement with the experiment

is excellent. On the other hand, the amplitude curve obtained in the computations with the same disturbance level as in the measurements starts to deviate from the experiment at about x = 850. This is most likely due to 3D effects which are further discussed in section 3.4. The agreement in initial amplication of all curves in gure 9 is explained by the change in size of the experimental bubble, associated with the introduction of disturbances. As pointed out previously when a disturbance is introduced, the bubble in the experiment shrinks to roughly the same size as the smaller computed bubble where the normal velocity is prescribed directly from measurements. In the computations a much smaller change takes place for the time-averaged disturbed bubble. This is shown in gure 10. Comparisons between the disturbed larger bubble and the experimental bubble are made in gure 11. A case when the asymptotic boundary condition, equation (6), is used for the upper boundary is also included. In this case small disturbances are eventually sufciently magnied to give vortex shedding. With shedding this boundary condition permits a global change of size of the bubble closely matching that obtained in the experiment. It is apparent that the asymptotic boundary condition, equation (6), gives a good description of the change in the bubble characteristics when disturbances are introduced, while prescribing v xes the size of the bubble. In all cases the agreement with local linear stability theory is excellent. Non-parallel effects are therefore not considerable, presumably due to the short wavelength of the instability wave compared to the change of the mean ow. Figure 12 compares the wave amplitude distribution in the simulations and in the experiment at x = 720 mm, x = 800 mm and x = 880 mm, i.e. downstream of sepa-

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

325

Figure 10. Displacement thickness with and without forcing of 2D-waves in experiment and computations. Experiments, +: unforced bubble, : with harmonic forcing. DNS, : unforced bubble, - - -: forced bubble. In the DNS, free stream boundary conditions in V were prescribed from measurements.

Figure 12. Wall normal wave amplitude proles at three streamwise positions. : DNS with the same disturbance level as in the experiment, : experiment. f = 95 Hz.

Figure 13. Predicted transition locations in the separation bubble by several engineering methods, prescribing n = 12 at transition. Displacement, (+), and momentum thickness, (), from experiment. Solid and dashed curves from DNS.

Figure 11. as function of x. : computed with v set for best agreement with undisturbed measurements, - - -: computed time averaged bubble when a disturbance is introduced, : disturbed measured bubble, - - -: computed time-averaged bubble when the asymptotic boundary condition is used.

3.3. Evaluation of transition prediction methods Several transition prediction methods used in engineering applications were applied to the present oweld, including both simple empirical correlations but also more sophisticated methods based on the en -method. The result of the evaluation is shown in gure 13, where the vertical lines indicate the predicted transition location, x tr . The correlations by Horton [18] and Roberts [31] predict transition at x tr = 768 and 839 mm, respectively, that is upstream of the maximum displacement of the bubble. Experimentally, the location of the maximum bubble displacement, x max , has been chosen as a criterion for the transition point, see Brendel and Mueller [3] and Ripley and Pauley [29]. In the present ow, x max = 863 mm, which is close to the position x = 867 mm where the two amplication curves from DNS in g-

ration, in the middle of the bubble and after reattachment. The wave amplitude in the experiment is measured by the r.m.s. of the uctuating streamwise velocity of the disturbance. The agreement between DNS and experiment is good at the two rst x-positions, with a double peak developing in the amplitude prole in the region near the wall. At x = 880 mm however the near wall maximum is located further out from the wall in the experimental amplitude prole.

326

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

ure 9 start to deviate, due to nonlinear development in the case when the inital wave amplitude is matched to the experiment. Figure 13 shows that the point of maximum displacement corresponds to the position where a sharp rise in the experimental momentum thickness occurs, caused by a rapid skin friction increase due to transition. The method of van Ingen [37] predicts x tr = 855 mm, close to the value of x max = 863 mm, whereas the method of Drela and Giles [10] and the shape factor formulation of the method of Cebeci et al. [6], predict x tr = 933 and 968 mm, respectively. In the comparison between the different methods transition is assumed to occur when the n-factor reached the value of n = 12, suggested by Mack [25] for a wind tunnel free stream turbulence level of 0.02%. However, adopting the denition of the point of transition as above, x tr = 863 mm, and applying linear stability theory on velocity proles from DNS, matching the experimental proles, a critical n-factor of n = 13 is obtained. On these premises the van Ingen method predicts x tr = 872 mm and Drela and Giles x tr = 1144 mm. The method of Cebeci et al. gives the same prediction since the n-factor does not explicitly enter in the prediction formulation; calibration of the method is based on the assumption of a critical n-factor equal to 9. For the bubble presented here it is clear that the method of van Ingen gives the best prediction of the transition point. In general, from the close correspondence of the linear stability results and the computed and measured values of the disturbance growth, it is clear that a method based on the linear exponential growth (en -method) is preferred to the older correlation type methods discussed earlier. 3.4. Some remarks on three-dimensional ow development in the separation bubble The present study considers a two-dimensional separated ow. The numerical computations presented are all two-dimensional and therefore they do not take into account the ow development in the spanwise direction. The experiment has been carefully designed to provide a two-dimensional separation bubble and to avoid effects of a three-dimensional mean ow. The greater part of the ow eld in the experiment is found to be highly twodimensional, including the attached adverse pressure gradient boundary layer upstream of separation, the steady separation line and the separated shear layer. A strong exponential growth of two-dimensional wave disturbances is found to occur in the latter region. This development in the separation bubble is fully captured by the twodimensional computations. Nevertheless, the experiment is three-dimensional and the development in the spanwise direction can be and has been studied. Flow visualization recordings, cf. gure 2, show that a rapid breakdown of 2D-instability waves

occurs in the reattachment region resulting in a threedimensional, unsteady ow development downstream of the steady 2D-shear layer. This nonlinear disturbance development cannot be studied using two-dimensional numerical simulations. However, for the purposes in this article two-dimensional computations sufce, since it is only at a very late stage that any differences are seen to occur.

4. Summary and conclusions A transitional separation bubble ow was studied computationally and experimentally using hot-wire anemometry and direct numerical simulations. Low-amplitude two-dimensional instability waves experience a rapid growth in the separated shear layer, leading to transition in the ow. Good agreement is found between experiment and DNS in this respect. Modelling the experimental bubble numerically it was found that prescribing free stream boundary conditions in the streamwise velocity is not sufcient in order to reproduce the streamwise location and size of the separation bubble. Rather, boundary conditions in the wall normal velocity are needed. In both experiments and DNS the introduced wave disturbances result in a global change in the mean ow. In the computations, however, the size and behaviour of this change are closely related to the type of boundary conditions used. A boundary condition xing the normal velocity at the upper boundary xes the height of the bubble while a boundary condition prescribing the solution to asymptotically approach the linear inviscid solution captures the experimental behaviour qualitatively. Furhermore, calculating ows involving transitional separation bubbles requires the best possible damping of disturbances in the fringe region. Application of several transition prediction methods for separation bubbles shows that the en -method predicts transition in accordance with experiment. The critical n-factor is higher than what is generally found for attached boundary layer ows. Of the models based on simplications of the en -method, the method of van Ingen gives best agreement with the results found here. Also, the empirical correlation of Roberts predicts transition close to the experimental location. The latter model avoides the prescription of the ambiguous n-factor.

Acknowledgements This work was supported by TFR, Swedish Research Council for Engineering Sciences and NFFP (Swedish National Aeronautics Program) which is gratefully acknowledged.

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

327

References
[1] Berlin S., Lundbladh A., Henningson D.S., Spatial simulations of oblique transition in a boundary layer, Phys. Fluids 6 (1994) 19491951. [2] Bestek H., Gruber K., Fasel H., Self-excited unsteadiness of laminar separation bubbles by natural transition, The prediction and exploitation of separated ow, P. Roy. Aeronaut. Soc. (1989) 14.114.16. [3] Brendel M., Mueller T.J., Boundary-layer measurements on an airfoil at low Reynolds numbers, J. Aircraft 25 (1988) 612617. [4] Cebeci T., Calculation of multielement airfoils and wings at high lift, AGARD-CP-515, 1992. [5] Cebeci T., Chen H.H., Prediction of transition on airfoils with separation bubbles, swept wings, and bodies of revolution at incidence, in: Cebeci T. (Ed.), Numerical and Physical Aspects of Aerodynamic Flows IV, SpringerVerlag, 1990. [6] Cebeci T., Wang G.S., Chang K.C., Choi J., Recent developments in the calculation of ow over low Reynolds number airfoils, in: Aerodynamics at Low Reynolds Numbers 104 106 , Vol. II, The Royal Aeronautical Society, London, 1986, pp. 15.115.13. [7] Coleman G.N., Spalart P.R., Direct numerical simulation of a small separation bubble, in: Speziale C.G., Launder B.E. (Eds.), Near-Wall Turbulent Flows, Elsevier Science, 1993. [8] Davis R.L., Carter J.E., Analysis of airfoil transitional separation bubbles, Nasa-CR-3791, 1984. [9] Drela M., Low-Reynolds-number airfoil design for the M.I.T. Daedalus prototype: A case study, J. Aircraft 25 (1988) 724732. [10] Drela M., Giles M.B., Viscous-inviscid analysis of transonic and low Reynolds number airfoils, AIAA J. 25 (1987) 13471355. [11] Dovgal A.V., Kozlov V.V., Michalke A., Laminar boundary layer separation: Instability and associated phenomena, Prog. Aerospace Sci. 30 (1994) 6194. [12] Elofsson P., Experiments on oblique transition in wall bounded shear ows, Ph.D. Thesis, TRITA-MEK, Tech. Rep. 1998:5, Dept. Mechanics, KTH, 1998. [13] Eppler R., Recent developments in boundary layer computation, in: Aerodynamics at Low Reynolds Numbers 104 106 , Vol. II, The Royal Aeronautical Society, London, 1986, pp. 12.112.18. [14] Gruber K., Bestek H., Fasel H., Interaction between a TollmienSchlichting Wave and a Laminar Separation Bubble, AIAA Paper 87-1256, 1987. [15] Hggmark C.P., Bakchinov A.A., Alfredsson P.H., Experiments on a two-dimensional laminar separation bubble, Philos. T. Roy. Soc. 358 (2000) 31933205. [16] Henningson D.S., Lundbladh A., Johansson A.V., A mechanism for bypass transition from localized disturbances in wall-bounded shear ows, J. Fluid Mech. 250 (1993) 169207. [17] Hildings C., Simulation of laminar and transitional separation bubbles, Lic. Thesis, TRITA-MEK, Tech. Rep. 1997:19, Dept. Mechanics, KTH, 1997.

[18] Horton H.P., A semi-empirical theory for the growth and bursting of laminar separation bubbles, Aeronautical Research Council, Current Paper 1073, 1969. [19] Johansson A.V., Alfredsson P.H., On the structure of turbulent channel ow, J. Fluid Mech. 122 (1982) 295 314. [20] Kim J., Moin P., Moser R., Turbulence statistics in fully developed channel ow, J. Fluid Mech. 177 (1987) 133 166. [21] Kusunose K., Cao H.V., Prediction of transition location for a 2-D NavierStokes solver for multi-element airfoil congurations, AIAA Paper, 94-2376, 1994. [22] Lundbladh A., Henningsn D.S., Johansson A.V., An efcient spectral integration method for the solution of the NavierStokes equations, FFA-TN 1992-28, 1992. [23] Lundbladh A., Schmidt P.J., Berlin S., Henningson D.S., Simulation of bypass transition for spatially evolving disturbances, AGARD-CP-551, 1994. [24] Lundbladh A., Berlin S., Skote M., Hildings C., Choi J., Kim J., Henningson D.S., An efcient spectral method for simulation of incompressible ow over a at plate, TRITA-MEK Tech. Rep. 1999:11, Dept. Mechanics, KTH, 1999. [25] Mack L.M., Transition prediction and linear stability theory, AGARD-CP-224, 1977, pp. 1.11.22. [26] Maucher U., Rist U., Wagner S., Secondary disturbance amplication and transition in laminar separation bubbles, IUTAM Symp. on Laminar-Turbulent Transition, Sep. 1317, 1999, Sedona Arizona, USA, 1999. [27] Nordstrm J., Nordin N., Henningson D.S., The fringe region technique used in the direct numerical simulation of the incompressible NavierStokes equations, SIAM J. Sci. Comp. 20 (1999) 13651393. [28] OMeara M.M., Mueller T.J., Laminar separation bubble characteristis on an airfoil at low Reynolds numbers, AIAA J. 25 (1987) 10331041. [29] Ripley M.D., Pauley L.L., The unsteady structure of twodimensional steady laminar separation, Phys. Fluids A 5 (1993) 30993106. [30] Rist U., Maucher U., Direct numerical simulation of 2-D and 3-D instability waves in a laminar separation bubble, in: Application of Direct and Large Eddy Simulation to Transition and Turbulence, AGARD CP-551, Chania, Cretel, 1994. [31] Roberts W.B., Calculation of laminar separation bubbles and their effect on airfoil performance, AIAA J. 18 (1980) 2531. [32] Stewartson K., Further solutions of the FalknerSkan equation, P. Camb. Phil. Soc. 50 (1954) 454465. [33] Tatineni M., Zhong X., Numerical and Theoretical Studies of the Instability of Low-Reynolds-Number Separation Bubbles, AIAA Paper 2000-2308, 2000. [34] Theolis V., Hein S., Dallmann U., On the origins of unsteadiness and three-dimensionality in a laminar separation bubble, Philos. T. Roy. Soc. 358 (2000) 32293246. [35] van Ingen J.L., On the calculation of laminar separation bubbles in two-dimensional incompressible ow, AGARD-CP-168, 1975. [36] van Ingen J.L., Transition, pressure gradient, suction, separation and stability theory, AGARD-CP-224, 1977.

328

C.P. Hggmark et al. / Aerosp. Sci. Technol. 5 (2001) 317328

[37] van Ingen J.L., Research on laminar separation bubbles at Delft University of Technology, in: Kozlov V.V., Dovgal A.V. (Eds.), Separated Flows and Jets, Proc. IUTAMSymp., Springer-Verlag, Berlin, 1991, pp. 537556. [38] van Ingen J.L., Boermans L.M.M., Aerodynamics at low Reynolds numbers: a review of theoretical and experimental research at Delft University of Technology, in: Aerodynamics at Low Reynolds Numbers 104 106 ,

Vol. I, The Royal Aeronautical Society, London, 1986, pp. 1.11.40. [39] Vatsa V.N., Carter J.E., Analysis of airfoil leading-edge separation bubbles, AIAA J. 22 (1984) 16971704. [40] von Doenhoff A.E., A preliminary investigation of boundary-layer transition along a at plate with adverse pressure gradient, NACA-TN 639, 1938.