Sei sulla pagina 1di 15

Exp Fluids DOI 10.

1007/s00348-011-1071-9

RESEARCH ARTICLE

Stochastic estimation of ow near the trailing edge of a NACA0012 airfoil


Ana Garcia-Sagrado Tom Hynes

Received: 4 February 2010 / Revised: 21 December 2010 / Accepted: 10 March 2011 Springer-Verlag 2011

Abstract A stochastic estimation technique has been applied to simultaneously acquired data of velocity and surface pressure as a tool to identify the sources of wallpressure uctuations. The measurements have been done on a NACA0012 airfoil at a Reynolds number of Rec = 2 9 105, based on the chord of the airfoil, where a separated laminar boundary layer was present. By performing simultaneous measurements of the surface pressure uctuations and of the velocity eld in the boundary layer and wake of the airfoil, the wall-pressure sources near the trailing edge (TE) have been studied. The mechanisms and ow structures associated with the generation of the surface pressure have been investigated. The quasiinstantaneous velocity eld resulting from the application of the technique has led to a picture of the evolution in time of the convecting surface pressure generating ow structures and revealed information about the sources of the wall-pressure uctuations, their nature and variability. These sources are closely related to those of the radiated noise from the TE of an airfoil and to the vibration issues encountered in ship hulls for example. The NACA0012 airfoil had a 30 cm chord and aspect ratio of 1.

A. Garcia-Sagrado T. Hynes Whittle Laboratory, Department of Engineering, University of Cambridge, 1 JJ Thomson Avenue, Cambridge CD3 0DY, UK Present Address: A. Garcia-Sagrado (&) Applied Modelling and Computation Group, Department of Earth Science and Engineering, Royal School of Mines, Imperial College London, Prince Consort Road, London SW7 2BP, UK e-mail: a.garcia-sagrado@imperial.ac.uk

List of symbols C Chord of the airfoil (m) f Frequency (Hz) h Trailing edge thickness (m) kx Streamwise wavenumber (1/m) p0 , p0 w Fluctuating surface pressure (Pa) qL Linear wall-pressure uctuating sources, 0 qL du ov (1/s2) dy ox qL/r Weighted linear wall-pressure uctuating sources, accounting for the distance to the wall (1/m s2) r Distance between the wall-pressure source and the point of observation on the surface of the airfoil (m) S Saddle point u0 , v 0 Streamwise and normal velocity uctuations (m/s) u0 s , v 0 s Stochastically estimated streamwise and normal velocity uctuations (m/s) u0 i,s Stochastically estimated velocity uctuation (m/s) u0 i Velocity uctuation (m/s) U1 Freestream velocity (m/s) Ue Velocity at the edge of the boundary layer (m/s) Uc Convection velocity (m/s) urms Root mean square of velocity uctuations (m/s) V Vortex x Streamwise distance from the airfoil leading edge (m) x0 Streamwise distance from the airfoil trailing edge (m) y Normal distance from the wall or from the airfoil extended centre line (m) z Lateral distance from the airfoil mid-span (m)

123

Exp Fluids

Rec d d* s U Upi pj

Uup , Uvp

q LSE MLSE MSLSE MQSE QSE TE

Reynolds based on the chord of the airfoil Boundary layer thickness (m) Boundary layer displacement thickness (m) Time delay (s) Surface pressure power spectral density (Pa2/Hz) Cross-spectra between surface pressure uctuations from microphones pi and pj (Pa2/Hz) Cross-spectra between streamwise and normal components of velocity and surface pressure [(m/s)Pa/Hz] Density (kg/m3) Linear stochastic estimation Multi-point linear stochastic estimation Multi-point spectral linear stochastic estimation Multi-point quadratic stochastic estimation Quadratic stochastic estimation Trailing edge of the airfoil

1 Introduction Understanding the relationship between the surface pressure and velocity elds is essential to provide further insight into the mechanisms responsible for ow-induced noise and vibration and devise solutions to control the ow eld and minimise their undesirable effects. Due to the impossibility to measure the pressure uctuations inside the ow because of the lack of non-intrusive techniques, measurements of the pressure uctuating eld have been conned to the wall. Wall-pressure measurements have been done by researchers with the aim to improve the understanding of the ow structures through their wall-pressure manifestation (Blake 1970, 1975; Brooks and Hodgson 1981; Daoud 2004; Hudy et al. 2003; Gravante et al. 1998; Roger and Moreau 2004; Wark et al. 1998). Furthermore, wall-pressure measurements have been performed in the literature (Brooks and Hodgson 1981; Roger and Moreau 2004; Rozenberg et al. 2006, 2007; Yu and Joshi 1979), in order to obtain surface pressure statistics upstream of the trailing edge (TE) of an airfoil, which are intimately related to the statistics of airfoil self-noise. A few researchers have performed simultaneous measurements of wall pressure and velocity such as Blake (1975), Daoud (2004) and Goody (1999). Blake (1975) reported simultaneous measurements of uctuating surface pressures on trailing edges of at struts and of the uctuating velocities in the wakes that generated those pressures. The aim was to investigate the inuence of trailing edge shapes on vortex induced vibrations of turbine blades, by analysing the pressure eld and the relationship of the

uctuating surface pressures with the near wake region of each trailing edge. Daoud (2004) investigated the pressure and velocity elds downstream of a separating/reattaching ow region on a splitter at plate with fence. As Daoud (2004) reported, separating/reattaching ows contain highly energetic structures which generate large wallpressure uctuations that are a direct representation of the excitation forces produced by the turbulent ow on the surface. If the excitation happens at the frequencies of the surfaces resonant modes, considerable vibrations and noise are generated. Daoud (2004) emphasised the importance of a better understanding of the wall pressure characteristics to control such unwanted effects. Furthermore, by measuring the velocity eld at the same time, the turbulent ow activity above the surface responsible for the wall-pressure generation can be investigated. Goody (1999) measured surface pressure and velocities over a wing-body junction and a 6:1 prolate spheroid in order to investigate three dimensional boundary layers. The correlation between surface pressure and velocity was analysed. However, measurements were only done with a single pressure transducer instead of using multiple transducers spatially separated and hence, the analysis of the pressure eld was limited. In this paper, a stochastic estimation technique has been applied to the simultaneous measurements of instantaneous pressure and velocity near the TE of a NACA0012 airfoil, downstream of a separated laminar boundary layer. From the cross-correlations between the simultaneously measured signals, regions of high correlation levels have been associated with the location of the main pressure generating ow structures (Garcia-Sagrado and Hynes 2011). However, the stochastic estimation technique has permitted a deeper investigation into the ow structures associated with the surface pressure generation. First used by Adrian (1977, 1979), this technique estimates the pseudo-instantaneous velocity eld from its wall-pressure signature. Evolution in time of the estimated velocity eld provides a picture of the convecting wall pressure generating ow structures and allows information about the variability and nature of the sources of the wall-pressure uctuations.

2 Experimental setup 2.1 Airfoil model and experimental setup Figure 1 depicts the experimental set-up where the airfoil investigated has been placed at the exit (open jet conguration) of an open-circuit blower type wind tunnel, supported by two perspex side walls. The symmetric airfoil employed in the investigation was a NACA0012 airfoil with a chord of 300 mm and an aspect

123

Exp Fluids

Fig. 1 Simplifying sketch of the exit of the wind tunnel and perspex side walls holding the airfoil

ratio of 1 that can be seen in Fig. 2. It consisted of three parts of which the middle one, with a span of 200 mm, was hollow. This part is also made of three different pieces allowing the airfoil to be completely dismantled and providing access to the interior in order to place the instrumentation. The two lateral parts are solid and have a span of 50 mm each. Measurements of wall-pressure uctuations are often carried out in acoustically quiet wind tunnels or are treated afterwards by applying noise cancellation techniques (Agarwal and Simpson 1989; Helal et al. 1989; Naguib et al. 1996), in order to minimise the low-frequency contamination on the surface pressure signals by the facility background noise. In the present case, measurements were done in the most quiet wind tunnel available with the internal walls of the tunnel covered with a foam material (5 cm thickness) that reduced the background noise of the facility by up to 20 dB from 100 Hz onwards. This allowed the measurement of the pressure signature of interest without the need to apply a noise cancellation technique. Noise cancellation techniques (Naguib et al. 1996) were applied to the rst set of data, and since it was found that there was no further improvement by the application of the noise

cancellation methods, these have not been applied to the data. Furthermore, a series of at plate measurements with embedded microphones were carried out to ensure that the background noise of the wind tunnel with the foam material was lower than the aerodynamic wall-pressure signature. The exit of the tunnel has a rectangular cross section of 0.38 m 9 0.59 m (after the foam). The freestream turbulence intensity of the wind tunnel is 0.4% allowing the investigation of the ow around a NACA0012 in a smooth non-turbulent inow. The results presented in this paper correspond to measurements done at a Reynolds number of Rec = 2 9 105 based on the chord of the airfoil and a freestream velocity of 10 m/s. The TE thickness h was 1.6 mm and h/d* [ 0.3, d*, being the boundary layer displacement thickness at the edge. As indicated by Blake (1986), this corresponds to a blunt TE, where vortex shedding is then normally observed from the TE. Note that the NACA0012 airfoil was truncated near the TE in order to have a thickness at the TE of 1.6 mm and hence, a blunt TE. This resulted in a nal chord C of 297 mm. 2.2 Model instrumentation and outline of surface microphone array The unsteady surface pressure measurements were performed with microphones FG-3329-P07 from Knowles Electronics that are 2.5 mm diameter omnidirectional electret condenser microphones with a circular sensing area of 0.79 mm. These microphones have been embedded in the airfoil under a pin hole of 0.4 mm diameter in order to minimise attenuation effects at high frequencies due to the nite size of the microphones. A microphone under a pin hole arrangement is depicted in Fig. 3a. The geometrical dimensions of the pin hole conguration (diameter, length of pin hole) were selected such that the resonant frequency associated with the arrangement (similar to a Helmholtz resonator) was greater than 20 kHz which is outside the range of interest. This was done by performing tests on a at plate with microphones under different pin hole geometrical congurations. The dimensions of the microphone and nal pin hole conguration are indicated in Fig. 3c.

Fig. 2 NACA0012 model used for the experiments: a NACA0012 assembled, b middle hollow part and lateral solid parts and c middle hollow part made of three different pieces, one of them, an exchangeable TE

123

Exp Fluids Fig. 3 a Microphone under a pin hole conguration, b remote microphone arrangement and c dimensions of microphone and pin hole conguration

Additional microphones were located within the airfoil in order to be able to measure the surface pressure uctuations very close to the TE. These microphones were located inside the airfoil in a remote microphone arrangement, since due to the reduced space close to the TE, they could not be placed directly beneath the pin hole. They were linked to the pin holes on the surface by plastic tubes of 0.4 mm internal diameter that run along inner passages and were continued innitely to avoid reections from standing waves. The plastic tubes passed through the interior of small boxes, each one containing a microphone on one of their lateral sides. The boxes and microphones were all inside the airfoil. The pressure uctuations were felt by the microphones through a small hole in the wall of the plastic tube and in the wall of the box in contact with the microphone. A sketch illustrating this remote microphone arrangement can be seen in Fig. 3b. With this method, it was possible to measure the pressure uctuations as close to the edge as 2 mm (1 % of the chord), less than 1d*TE. The surface microphone array is depicted in Fig. 4. The positions of the microphone pin holes on the surface1 are summarised in Table 1. Note that the black dots represent the location of the pin holes but do not correspond to their actual size (0.4 mm). 2.3 Measurement techniques As mentioned earlier, the unsteady pressure measurements were performed with FG-3329-P07 microphones from
1

Fig. 4 Surface microphone array. The system of coordinates indicated has its origin at the TE and at the airfoil mid-span

The position of the pin holes correspond to the microphone location for those microphones in the pin hole conguration; microphones in the remote microphone arrangement are placed further inside the airfoil.

Knowles Electronics. The sensitivity of the FG-3329-P07 microphone was provided by the manufacturer to be about 22.4 mV/Pa (45 Pa/V) in the at region of the microphone response (from 70 to 10,000 Hz approximately). From the calibrations performed in the laboratory, the sensitivity of the FG-3329-P07 microphones employed varied approximately between 20.2 and 23.5 mV/Pa in the at region. The microphones were powered by two units with 8 channels each containing the circuits needed for each microphone (manufactured by the Electronics Development Group at the Engineering Department of the University of Cambridge). The microphone signals were amplied and low-pass ltered (at 30 kHz) before being connected to the BNC-2090 connector panel. The sampling frequency was 65,536 Hz, and a total of 223 = 8,388,608 samples were acquired. A tube with a length of 110 mm was used for the calibration of the microphones. The output from a white noise

123

Exp Fluids Table 1 Positions of microphones in the airfoil model (chord C = 297 mm) xa/C 0.99 0.98 0.97 0.92 0.90 0.88 0.86 0.83 0.80 0.76 0.70 0.61 0.42 0.16 0.99 0.98 0.83

Microphone number 1c, 25c 2c, 26c, 27c 3c 4, 15, 16, 17, 18, 19, 24 5 6 7 8 9 10

Distance from TEb, x0 (mm) 2.0 5.0 10.0 25.0 30.0 35.0 40.0 50.0 60.0 70.0 89.2 117.2 173.2 250.0 2.0 5.0 50.0

Distance from mid-span, z (mm) 0, 2 0, 4, 10 0 0, 5, 10, 15, 35, 65, -35 0 0 0 0 0 0 0 0 0 0 0 0 0

Chordwise distance from the LE to the microphone location (to the pin hole)

11 12 13 14 1psc 2psc 8ps

Chordwise distance from the TE to the microphone location (to the pin hole) Microphones placed using the remote microphone arrangement sketched in Fig. 3b

signal generator was connected to a loudspeaker that was placed at one end of the tube. At the other end, in a cap closing the tube, the FG-3329-P07 microphone and an ENDEVCO 8507C-1 pressure transducer were placed equidistant from the centre of the circular cap. The FG3329-P07 microphone was placed under a pin hole conguration with the same dimensions as the one on the airfoil. The ENDEVCO transducer with a diameter of 2.5 mm was placed ush mounted. The ENDEVCO known calibration (slope of the linear Pa vs. V calibration) is constant with frequency for the range of frequencies of interest, i.e., up to 20 kHz. First of all, the performance of the tube-calibrator was assessed by calibrating another ENDEVCO pressure transducer whose calibration had been previously obtained using a Druck DPI520 pressure indicator as a pressure source. In this case, both ENDEVCO transducers were placed ush mounted on a cap at the end of the tube. Two tubes with two different diameters, 10 and 12.5 mm, were used for the calibration of the microphones. According to acoustic theory (Fahy and Gardonio 2007), plane wave propagation will occur in the ducts (tube) until ka & 1.84, where a is the radius of the tube and k = x/c is the wavenumber (c the speed of sound). Hence, the mentioned tubes provided a calibration of the microphones up to f & 20 kHz when using the tube of 10 mm diameter and f & 16 kHz with the tube of 12.5 mm diameter. In order to calibrate the microphones connected to pin holes very close to the TE, and hence, placed inside the airfoil using the remote arrangement, the tube of 12.5 mm diameter was employed. In this case, the calibration tube

was left open at the opposite end of the loudspeaker and it was placed upside down covering the microphone to be calibrated and an ENDEVCO ush mounted on a small at plate connected during the calibration to the TE of the airfoil. This way, the microphones under the remote arrangement could be calibrated in situ and their frequency response (amplitude and phase versus frequency) could be obtained. The attenuation and possible resonances induced by the plastic tube used to connect the microphone with the pin hole on the surface were accounted for by this in situ calibration (calibration method described next). Examples of the calibration of two of these remote microphones are given in Fig. 5. Note the deviation of the frequency response of the microphone under the remote arrangement from the frequency response of the same microphone under the pin hole conguration. The measured signal was rst of all converted into the frequency domain by means of calculating its Fourier transform, resulting in an amplitude (Volts) and a phase versus frequency. The microphone frequency response obtained from the calibration was then applied to the measured amplitude and phase for each frequency. Then the inverse Fourier transform was performed resulting in a signal with pressure units. Note that the units of the microphone frequency response amplitude are dB (20 log10 (A/Aref), where Aref is 1V/0.1Pa (1/0.1 V/Pa) and A is the amplitude of the calibration in V/Pa (1/A Pa/V)). The method employed in the calibration of the FG-3329P07 microphones is based on the calibration procedure from Mish (2001). The calibration consists of taking two different measurements (a and b) as it has been sketched in

123

Exp Fluids

Fig. 5 Examples of the frequency response (amplitude and phase) from two of the microphones connected to pin holes very close to the TE and thus located in the airfoil using the remote microphone conguration

Fig. 6 Method employed in the calibration of the FG-3329-P07 microphones (Following Mish 2001)

Fig. 6. In the rst one, (a), the output signal from the white noise source is measured at the same time as the output signal from the ENDEVCO (with the BNC-2090 board). Since the calibration of the ENDECVO is known, the output signal from the speaker can be calculated. Once the output from the speaker is known, since the input to the speaker has also been measured at the same time, the speaker response can be calculated. In the second measurement, (b), the output signal from the white noise source is measured at the same time as the output from the FG-3329-P07 microphone. The frequency response of the system formed by the speaker and microphone can be calculated and since from measurement a), the speaker response was obtained, the FG-3329-P07 microphone response can be calculated. The above-described calibration method provided the frequency response, i.e., amplitude and phase calibration of each individual microphone used in the measurements of the surface pressure uctuations over the frequency range of interest (20 Hz to 20 kHz). Note that these FG-3329P07 microphones clip if the signal is [30 Pa ([124 dB). It was checked that the surface pressure uctuations on the airfoil under the ow conditions investigated were lower than that limit.

A total of 223 = 8,388,608 samples were recorded in order to characterise the wall-pressure eld. The uncertainty in the surface pressure spectra was mainly due to the statistical convergence error, which is inversely proportional to the number of records used. In order to reduce this error, the spectra was calculated as the average of the spectra of individual data records obtained from dividing the pressure time series into a sequence of records. The total number of records used was 1,024 resulting in an p uncertainty of about 3% (1= Nr ; Nr being the number of records). The velocity measurements have been performed with constant temperature hot-wire anemometry. The hot-wire instrumentation included a fully integrated constant temperature anemometer with built-in signal conditioning. The AC and DC components were logged separately using different pre-amplier gains in order to ensure the adequate resolution of the uctuating AC component. This component was obtained by band-pass ltering the signal between 1 Hz and 30 kHz. The hot-wire signal was then reconstructed by adding the DC level to the zero-mean ltered AC output and thereby enhancing the signal resolution. The cross-wire probe that was used to simultaneously measure the normal and streamwise components of velocity was a subminiature boundary layer probe, especially designed by Dantec to given specications. The length of the wires was 0.7 mm with a diameter of 2.5 lm. From specications, the angle between each wire and the horizontal axis is 45 and the angle between the wires is 90. This subminiature cross-wire allowed measurements as close to the surface of the airfoil as 0.3 mm, which is approximately 3% of dTE or 15% of d . TE Boundary layer and turbulence intensity proles from the cross-wire probe were rstly compared with those

123

Exp Fluids

measured by a single hot-wire. This was a Dantec type P15 boundary layer probe with a tungsten element of 1.2 mm length and 5 lm diameter set to an overheat ratio of 1.8. The hot-wire probe was calibrated in the freestream region of the test section by logging the voltage from the probe together with the total and static pressures at approximately the same position. These pressures and hence the freestream velocity were measured with a pitotstatic probe that was placed at the same height as the hotwire in the test section and at a different spanwise location. A best-t calibration for Kings law (E2 - A = BUn) was then found, where A, B and n are calibration constants and A = E2 is the zero ow output voltage. The effect of air o temperature drift was taken into account with the correction described by Bearman (1971). The surface of the airfoil was found by using a resistance circuit that measured the electrical contact between the probe and the surface. In order to take into account the proximity of the surface and its effect on the cooling of the wire, a traverse with no ow was performed to nd out the Cox (1957) correction to be applied to the measured data. A yaw calibration was performed in a calibration tunnel to determine the relationship between the effective cooling velocity for each wire and the streamwise and cross-stream velocity components u and v and to conrm the angle between the two wires of the cross-wire conguration. The calibration was carried out by changing the yaw angle of the cross-wire probe over a certain range in a known velocity ow (magnitude and direction) and recording the output of the wires at every angle. In a single hot-wire probe, the wire is perpendicular to the ow direction experiencing the most cooling inuence and resulting in the maximum output voltage for a given velocity. The angle between the velocity vector and the normal to the wire in a plane that contains both, wire and velocity, is zero. If there is an angle between the ow direction and the normal to the wire, the wire mainly registers the cooling effect of the component normal to the wire and very small amount of cooling results from the parallel component. The velocity corresponding to the net cooling inuence is the effective cooling velocity, Ue. The form of the yaw-response function used has been that proposed by Champagne et al. (1967) and Champagne and Sleicher (1967), that was previously introduced by Hinze (1959). This function is expressed below as: Ue UF 2 90 a 1=2 1 U cos2 90 a k2 sin2 90 a where U is the ow velocity magnitude and a is the angle between the velocity direction and the wire. Therefore, 90 - a is the angle between the velocity direction and the normal to the wire, i.e., the yaw angle. The parameter k is a

constant that is determined from the yaw calibration, and k2sin2(90 - a) is associated with the cooling inuence of the velocity component parallel to the wire. A logging frequency of 65,536 Hz, equal to the sampling frequency of the microphones, was used. For the mean and rms proles, a total of 219 = 524,288 samples for the AC signal and of 217 = 131,072 samples for the DC signal were acquired. When acquiring simultaneously velocity and surface pressure data, 220 = 1,048,576 samples were acquired at each channel of the data acquisition board. In order to calculate the cross-correlation between velocity and pressure, a total of 256 records were used. The velocity spectra were calculated by averaging 512 records, resulting in an uncertainty due to statistical convergence error of about 4%. For the cross-spectra between velocity and surface pressure, a total of 2,048 records were employed and the uncertainty is approximately 2%. The velocity with the cross-wire probe was measured at approximately 80 wall-normal (y) locations. These locations went from y = 0.3 mm to 40 mm with increments of approximately 0.05 mm (0.007d) up to 0.1d, 0.1 mm (0.014d) up to 0.2d, 0.5 mm (0.071d) up to d, 1 mm (0.140d) up to 1.5d and 5 mm (0.710d) up to at least 2d. For verication of the cross-wire calibration and measurement procedure, the boundary layer and turbulence intensity proles at several positions were compared with those measured with a single hot-wire. A good agreement between the two measurements was observed with a maximum error of 3% of U1 for the boundary layer prole and of 1% of U1 for the turbulence intensity prole. The plots have not been included for brevity. To check for a possible interference caused by the presence of the cross-wire probe on the wall-pressure measurements, the microphones output (autospectra) was compared with the cross-wire probe at different y locations (not included for conciseness). The results with the crosswire located further away from the wall was considered as the no interference case. The spectra corresponding to the cross-wire position closer to the wall showed good agreement with the other spectra, indicating that the microphone measurements were not affected by the presence of the probe in the ow. The velocity measurements were taken in the mid-span region of the airfoil after conrming that the same velocity results were obtained at different spanwise locations within that region.

3 Principle of stochastic estimation technique In Garcia-Sagrado and Hynes (2011), a cross-correlation analysis between the velocity in the wake and in the boundary layer of the airfoil and the surface pressure

123

Exp Fluids

uctuations in the region near the TE shed light on the identication of ow structures associated with the surface pressure uctuations. However, that identication was based on time-averaged statistical information which does not disclose in detail the nature and variability of the mechanisms of the wall-pressure generation. In order to improve the understanding and provide additional information about the relationship between the ow eld and the surface pressure, a stochastic estimation of the velocity eld associated with wall-pressure events was carried out. With this technique, important information can be gathered about the instantaneous dynamics and spatio-temporal structure of the various events and their evolution (Guezennec 1989). The quasi-instantaneous velocity at different normal positions over the surface of the airfoil, i.e., at the locations of the cross-wire traverse, has been estimated from the wall-pressure uctuations as if the velocity at all the different positions over the wall had been measured simultaneously and at the same time as the pressure signals from the microphones on the surface. Evolution in time of the estimated velocity provides a picture of the quasi-instantaneous ow structures passing by the streamwise location of the cross-wire and affords further information about the wall-pressure generating mechanisms. The turbulent velocity eld, u0i;s ro Dr; t s, has been estimated from a known surface pressure signature or event, p0w ro ; t, where ro xo ; 0; zo , with components xo, 0 and zo, in the streamwise, wall-normal and spanwise direction, respectively, is the location of the pressure event, subscript i refers to the velocity component and s denotes stochastic estimation. The estimated mean-removed velocity u0i;s ro Dr; t s is obtained from a Taylor series expansion in terms of the mean-removed surface pressure event p0w ro ; t. u0i;s ro Dr; t s Ai ro Dr; sp0w ro ; t Bi ro Dr; sp02 ro ; t w 2

investigation, a multi-point estimation has been used since, as proved by different authors such as Bonnet et al. (1998) and Daoud (2004), it provides generally a more accurate estimation of the dominant ow structures than single-point estimations. 3.1 Single-point LSE The single-point LSE provides a linear estimation of the conditional velocity eld from the surface pressure of a single point of observation: u0i;s ro Dr; t s Ai;lin ro Dr; sp0w ro ; t 3

In order to determine the linear estimation coefcients, Ai,lin, the long-time mean squared error between the measured velocity and its estimate, ei ro Dr, is minimised, leading to: Ai;lin u0i ro Dr; t sp0w ro ; t 4

p20 ro ; t w u0i ro Dr; t sp0w ro ; t 2 p0w;rms ro ; t

Therefore, Ai,lin is equal to the ratio between the crosscorrelation of u0 i and p0 w, at the time delay s between the estimate and the pressure event, and the square of the rootmean-square of the surface pressure. The equations to obtain the coefcients for the singlepoint QSE can be found in Daoud (2004) or Naguib et al. (2001) 3.2 Multi-point LSE (MLSE) The multi-point LSE is based on a weighted linear combination of surface pressure events at multiple positions. The relationship between the estimated velocity and the surface pressure would be in this case (the surface pressure p0 w has been called p0 to simplify the equations): u0i;s Ai;1 p01 Ai;2 p02 Ai;3 p03 Ai;4 p04 Ai;5 p05 Ai;6 p06 Ai;7 p07

where Ai ro Dr; s and Bi ro Dr; s are the estimation coefcients for the linear and quadratic terms, respectively. The objective of the stochastic estimation technique is to obtain these coefcients Ai and Bi in order to be able to calculate the velocity from Eq. 2. When only the linear term in the equation is employed, the estimation is known as Linear Stochastic Estimation (LSE), whereas when the rst two terms are included in the estimation, it is called Quadratic Stochastic Estimation (QSE). Both can be implemented using a single point of observation (data from a single microphone or pressure transducer on the surface of the airfoil) or multiple points simultaneously, leading to what is called multi-point stochastic estimation, linear (MLSE) or quadratic (MQSE). In the current

using seven observation points on the surface of the airfoil, corresponding to the seven streamwise microphones closer to the TE. In this case, Ai;1 ;Ai;2 ...Ai;7 are the unknown coefcients, p0 1 is the microphone closer to the TE at 2 mm upstream (x/C = 0.99) and p0 7 is the seventh microphone upstream of the TE in the streamwise direction, at 40 mm from the edge (x/C = 0.86). As in the single-point LSE, the coefcients are determined by minimising the mean square error between the estimate and the measured velocity leading to:

123

Exp Fluids

3 2 0 0 p1 p1 Au;1 6 Au;2 7 6 p0 p0 7 6 2 1 6 6 Au;3 7 6 p0 p0 76 3 1 6 6 . 7 6 . . 5 6 . 4 . 4 . Au;7 p07 p01 2

p01 p02 p01 p03 p02 p02 p02 p03 p03 p02 p03 p03 . . . . . . 0 p0 p0 p0 p7 2 7 3

31 2 3 . . . p01 p07 u0 p0 7 6 0 17 . . . p02 p07 7 6 u p02 7 7 . . . p03 p07 7 6 u0 p03 7 7 7 6 . 7 6 . 7 . . 5 4 . 5 . . . . . . . p07 p07 u0 p07

Similar expression for the normal component of the velocity v0 . The inverse matrix on the right-hand side of Eq. 6 contains data from the two-point correlation between the data of the seven streamwise microphones closer to the TE. The second matrix on the right-hand side represents the crosscorrelation between the velocity and the surface pressure from all the microphones used in the MLSE. Note that the cross-correlation values must not be normalised (i.e. they must not be the cross-correlation coefcient values).

3.3 Multi-point spectral LSE (MSLSE) Multi-point Spectral LSE (MSLSE), which is an extension to the classical stochastic estimation technique described above, was applied in this case (Tinney et al. 2006; Ewing and Citriniti 1997). The coefcients for the calculation of the stochastically estimated mean-removed velocity are in this case a function of the frequency f and relate the Fourier transforms of the estimated velocities with the Fourier transforms of the surface pressure events: u0fi;s f Afi;1 f p0f 1 f Afi;2 f p0f 2 f Afi;7 f p0f 7 f

As Tinney et al. (2006) summarised, the advantage of the Multi-point Spectral Linear Stochastic Estimation (MSLSE) is that for elds where the estimated condition is separated in time (typically by a convection velocity: s Dx=Uconv ), the time lag s between the unconditional source, p0 j(t - s), and the conditional estimator, u0 i(t), is embedded in the spectral estimator. It is important to remark that in the cases where the unconditional terms are marginally correlated within themselves, the quality of results expected from both the MLSE and the MSLSE will be limited. This is particularly the case in situations such as the present study where surface pressure is used to estimate the velocity. There is a high level of ltering that occurs between the pressure eld and the velocity eld, that is, the pressure eld is driven by the most compact/coherent ow structures. Therefore, a decent correlation between velocity and pressure as well as between the pressure eld itself is necessary if good results are to be obtained from stochastic estimation techniques. The level of the correlation marks the success or failure of the technique in identifying the dynamic ow structure associated with the surface pressure generation.

4 Results The laminar boundary layer over the surface of the airfoil present at Rec = 2 9 105, separated at approximately x/C = 0.65 (x & 193 mm) and reattached a short distance upstream of the TE, at about x/C = 0.97 (x & 288 mm). The results shown correspond to a cross-wire location at 5 mm upstream of the TE of the airfoil (x/C = 0.98, x = 292 mm), which is immediately downstream of the reattachment point. The cross-wire was traversed normally to the surface of the airfoil. The surface pressure signals used for the multi-point stochastic estimation of the velocity over the surface at that position correspond to the signals from the seven streamwise microphones closer to the TE, covering a region of 40 mm upstream of the TE, between x/C = 0.86 to x/C = 0.99. The results from the multi-point spectral LSE will be shown. Figure 7 illustrates the MSLSE mean-removed velocity vector eld together with the pressure time series at x/C = 0.97, from the microphone immediately upstream of the location of the cross-wire at x/C = 0.98. The stochastically estimated velocity vector has been plotted for two consecutive time windows. The uctuating surface pressure has been normalised by p0 rms. The horizontal axis represents the normalised time. Figure 8 represents the stochastically estimated velocity vector viewed in a frame of reference moving with 0:6U 1 . For this, the mean velocity has been added to the uctuating components of the velocity u0 and v0 and then, the convection velocity

Similarly as in the previous cases, the coefcients are determined by minimising the mean squared error between the estimate and the actual measured velocity, leading to a similar expression to Eq. 6 to be solved for each frequency but instead of with cross-correlations between the different signals, with the cross-spectral density between signals (Up0i p0j , Uu0 p0 and Uv0 p0 ). The solution from these equations leads in this case, to an array of complex spectral estimation coefcients A,j(f). 3 2 0 0 Up1 p1 Afu;1 6 Afu;2 7 6 Up02 p01 7 6 6 6 Afu;3 7 6 Up03 p01 76 6 6 . 7 6 . 4 . 5 4 . . . 2 Afu;7 Up07 p01 Up01 p02 Up02 p02 Up03 p02 . . . Up07 p02 Up01 p03 Up02 p03 Up03 p03 . . . Up07 p03 ... ... ... . . . ... 3 2 3 Up01 p07 1 Uu0 p01 Up02 p07 7 6 Uu0 p02 7 7 6 7 Up03 p07 7 6 Uu0 p03 7 7 6 7 . 7 6 . 7 . 5 4 . 5 . . Up07 p07 Uu0 p07 8 Similar expression for the normal component of the velocity v0 .

123

Exp Fluids

(a)
y/*
rms

5 4 3 2 1 0 5

p/p

10

20

30

40

50

60

Ue/*

(b)
y/* p/prms

5 4 3 2 1 0 5

5 60

70

80

90

100

110

120

Ue/*
Fig. 7 Surface pressure time series p0 at x/C = 0.97 and stochastically estimated mean-removed velocity vectors at x/C = 0.98 (5 mm upstream of the TE) using MSLSE for two consecutive time windows Fig. 8 Surface pressure time series p0 at x/C = 0.97 and stochastically estimated velocity vectors at x/C = 0.98 (5 mm upstream of the TE) using MSLSE for two consecutive time windows, viewed in a frame of reference moving with 0:6U 1

0:6U 1 has been subtracted from the horizontal velocity component. The saddle points and vortices associated with positive and negative pressure peaks, respectively, have been indicated in the gure with letter S and letter V. They are well identied when viewed in a frame of reference moving with a similar velocity to that of the vortex cores (Fiedler 1988; Vernet et al. 1999). In turbulent boundary layer cases, the convection velocity can be calculated from the wall-pressure wavenumber-frequency (kx-f) spectra that displays an inclined ridge where most of the uctuating energy is concentrated. A line representing the peak locus of the spectrum ridge passes through the origin and its slope (f/kx) indicates the convection velocity of the dominant wall-pressure disturbances. In the current case, where the laminar boundary layer separates, due to the dominant periodic disturbances, no spectrum ridge could be observed in the kx-f spectra. On the contrary, most of the energy was concentrated around

the main tonal peaks. No convection velocity could hence be extracted in this way for this case. To investigate the effect of the convection velocity, the stochastically estimated velocity vectors were viewed in different frames of reference, using different convection velocities. It was observed how varying the velocity of the frame of reference affected the location of the vortical structures and saddle points. With lower constant convection velocity, the saddles and vortex cores appeared closer to the wall whereas with higher values they moved further away. However, as Adrian et al. (2000) observed, the fact that the vortices are recognisable with the three convection velocities indicates that it is not necessary to remove an eddys exact translational velocity, sometimes, a windowaveraged streamwise velocity works well identifying a majority of vortices in a given ow. In the current case, 0:6U 1 seems to be a good averaged value. It will be seen later, that with this convection velocity, when plotting the

123

Exp Fluids

velocity vectors superimposed onto the spanwise vorticity, the centres of the vortices coincide with regions of high vorticity. Note that following Daoud (2004), the time s is dened as s = -(t - t0) with t \ t0, t0 being an arbitrary time offset. It is possible to display different time windows of the data by selecting different values of t0. With this operation, the normalised s values have been folded, that is, they increase in the time-backward direction. Hence, by reversing the time, the progress of information with increasing time is from right to left. The motivation for this is that it allows to view the vector eld in a frame of reference where increasing s may be considered as increasing downstream distance by using Taylors hypothesis of frozen turbulence. The stochastically estimated velocity eld viewed in a frame of reference moving with the velocity of the vortex cores can be used to depict the spatial velocity eld considering Taylors hypothesis of frozen turbulence (Taylor 1938). This hypothesis which in its more rigorous form is only applicable to statistically stationary, homogeneous ows, affords an approximation to obtain spatial information from temporal data. Following a similar analysis as Daoud (2004), it can be argued that the applicability of the hypothesis holds fairly well if as Taylor (1938) showed, urms/Uc ( 1, Uc being the convection velocity of the main ow structures. Furthermore, the hypothesis is appropriate over a time window T that is much smaller than an eddy turnover time, l/urms, where l is the characteristic length scale of the dominant eddies. The latter means that there is hardly any time for the eddy to evolve and change state during T. Therefore, T should be much smaller than l/urms. The dominant length scale l of the structures associated with the wall-pressure generation was estimated from the wall-pressure wavenumber-frequency (kxf) spectra, which has not been incorporated in the paper for brevity. If kxpeak is the wavenumber associated with the peak in the spectra, then l & 1/kxpeak. In this case, kxpeak & 45 and l & 0.022 m. Examining the velocity (turbulent intensity, urms =U 1 ) in the region upstream of the TE covered by the seven microphones used in the estimation, a value of urms of about 0:1U 1 is selected (the turbulent intensity proles were presented in GarciaSagrado and Hynes (2011). Consequently, considering that U e % U 1 , the structural features can be considered as spatial structures over a time window TUe/d* ( (l/ 0.1Ue)Ue/d*, hence, TUe/d* ( l/0.1d*. For an estimated d* of 1.82 mm at this location (Garcia-Sagrado and Hynes 2011), TUe/d* ( 120. Thus, over a time window TUe/d* & 12, the Taylors hypothesis can be considered to hold reasonably well and the observed structures can be regarded as spatial structures. This time window has

been indicated as a reference in the velocity vector plots of Fig. 8. In both Figs. 7 and 8, the periodic nature of the structures is clearly observed. The vortex cores seem to be at a distance from the wall equal to y/d* & 2, and the saddles or stagnation points result from the vortex interactions. As the vortex cores convect past a point of observation on the wall, they produce negative pressure and the saddle points generate positive pressure values. In addition to the main vortices and saddle points, there seems to be other small vortices below the saddle points that are rotating in counterclockwise direction, in contrast to the principal clockwise vortices. The time between two vortices (or two saddle points) is sUe/d* & 23, which is equivalent to a normalised frequency fd*/Ue of 0.044. These observations are in agreement with the results from the cross-correlations between velocity and pressure presented in Garcia-Sagrado and Hynes (2011) Furthermore, these results corroborate that in this case, the surface pressure generation is dominated by the passage of the periodic structures. The periodicity of the structures can be further conrmed in Fig. 9, where the surface pressure power spectral density referenced to po = 20 lPa has been represented for streamwise microphones covering a region between x/C = 0.61 and x/C = 0.99. A main sharp peak can be seen at a frequency close to 200 Hz at all microphones.2 This peak is related to the periodic vortical structures originated in the separated shear layer and is associated with the instabilities from the laminar boundary layer, the so-called TollmienSchilichting (TS) instability waves that become amplied along the separated shear layer. These disturbances interact with the KelvinHelmholtz (KH) instabilities from the separated shear layer that could be related to the second peak (&230 Hz). The peaks in the spectra seem to be part of a complex mechanism where the different instabilities interact. The peak close to 400 Hz corresponds to the rst harmonic of the fundamental peak. In order to conrm that the convection velocity selected to plot the velocity vectors in Fig. 8 is the velocity of the vortex cores and that the circulatory patterns observed in the vectors are indeed vortices, the vorticity which is a frame of reference independent measure is plotted next. Note that there can also be regions of high vorticity without necessarily the existence of a vortex. Nonetheless, it is generally expected that vortex cores will be associated with regions of high vorticity. Figure 10 shows the spanwise vorticity, xz;s oxs oys , that has been calculated from the stochastically estimated
2

ov0

ou0

With the value of the displacement thickness d* at the location of the cross-wire of approximately 2 mm and the velocity Ue of 10 m/s, the normalised frequency fd*/Ue is 0.04.

123

Exp Fluids

cores identied in Fig. 8 coincide reasonably well with the localised regions of high vorticity; hence, this conrms the existence of vortices whose cores move with a velocity equal or similar to 0:6U 1 , which was the velocity of the frame of reference selected to visualise the vortices. Poissons equation governs the pressure uctuations for incompressible turbulent ows: 1 2 o2 r p Ui Uj q oxi oxj 9

Fig. 9 Surface pressure power spectral density referenced to po = 20 lPa at different streamwise locations

where p is the pressure, q the uid density and Ui is the total velocity. By applying Reynolds decomposition and for a two-dimensional ow such as the particular ow under investigation, with only one important mean shear component, that is homogeneous in the spanwise and streamwise direction (the latter being an approximation for boundary layer ows), the previous equation can be written as: 1 2 0 du ov0 r p 2 u0i;j u0j;i q dy ox 10

velocity eld by means of a central-nite-difference scheme. Note that when calculating the term the temporal variation is converted into spatial variation by con0 0 ot sidering ov ov ox and invoking Taylors hypothesis in ox ot combination with the local mean velocity. The estimated velocity vectors of Fig. 8 have been plotted superimposed onto the contours of vorticity. The vortices have been represented by a circular arrow indicating the direction of the rotation and the saddle points by a small circle. The vortex
ov0s ox ,

where u is the mean streamwise velocity, the prime denotes the mean-removed or uctuating quantities. The second term on the right-hand side of the equation has been expressed using tensor notation. Equation 10 indicates that the surface pressure uctuations are a function of the ow sources in the turbulent ow eld. The rst term on the right-hand side of the

Fig. 10 Spanwise vorticity and associated stochastically estimated velocity vectors using MSLSE at x/C = 0.98 (5 mm upstream of the TE) at a particular time window, viewed in a frame of reference moving with 0:6U 1

123

Exp Fluids

(a)
y/*

5 4 3 2 1 0 5

0.2 0 -0.2

q *2/U2
L e

-5

20

40

60

80

100

120

Ue/*

(b)
y/*

5 4 3 2 1 0 5

0.2

-0.2

qL/r * /Ue

-5 0 20 40 60 80 100 120

Ue/*
Fig. 11 Surface pressure time series p0 at x/C = 0.97 and pseudoinstantaneous linear pressure source, qL, at x/C = 0.98 (5 mm upstream of the TE). a Contours of qL and b contours of qL weighted, taking into account the distance to the point of observation of the pressure

equation represents the turbulencemean shear interaction. It is linear with respect to the velocity uctuations and is sometimes called the rapid term because it responds faster to changes in the mean ow. The second term on the righthand side represents the turbulence interaction with itself. It is non-linear with respect to the velocity uctuations and is sometimes called the slow term because it responds to changes in the mean ow only after the mean ow alters the turbulence. Often, the two source terms are also referred to as the mean-turbulent and the turbulent-turbulent sources, respectively. These have been expressed below as q(x,y,z,t) that represents the spatial distribution of the strength of the ow pressure sources at time t. qx; y; z; t 2 du ov0 u0i;j u0j;i dy ox 11

du/dy in comparison with the turbulent velocity gradients of the non-linear term. An order of magnitude analysis similar to the one done by Daoud (2004) can also be used to conrm this. Figure 11a illustrates the time evolution of the y distribution of the linear pressure sources qL that have been normalised by U2/d*2. It has been calculated from the e stochastically estimated or quasi-instantaneous velocity eld. It can be observed that negative pressure peaks coincide with negative pressure sources and positive peaks are associated with positive pressure sources. The sources seem to be distributed across the boundary layer (d & 3.8d*) and there seems to be good coincidence between the negative pressure sources and the location of the strong clockwise vortices observed in the ow structure of Fig. 8. The positive pressure sources appear upstream and downstream of these vortices, occupying the same region of the boundary layer as the negative sources. This scenario conrms that these dominant clockwise vortices are responsible for the pressure generation in this particular case. The term qv0 /qx from qL is very strong in the region near the cores of the main vortices and switches between negative and positive values creating the negative and positive pressure sources illustrated in Fig. 11a. The strength of qv0 /qx decreases away from the vortex cores, however, near the wall, the values of the sources associated with the wall-pressure generation are still very high. This is due to the term du/dy from qL, i.e., the mean velocity gradient, that increases substantially near the wall. On account of the 1/r decay present in the integral solution to the Poisson equation, the source term has been weighted as such and the results are presented in Fig. 11b. The results from Fig. 11b seem to indicate that from the point of view of the pressure, the strongest sources are near the wall, although since the pressure is the result of the integration over the ow volume, weaker sources that extend over larger volumes may be equally important as stronger, more localised sources (Daoud 2004). Hence, it can be concluded that in the present case, the wall-pressure sources are distributed across a signicant part of the boundary layer.

p/prms

p/prms

5 Conclusions The complex mechanisms present in the generation of surface pressure uctuations by a separated laminar boundary layer convecting past an airfoil have been investigated by performing simultaneous measurements of the surface pressure uctuations and of the velocity eld in the boundary layer and wake of a NACA0012 airfoil, especially in the region close to the trailing edge. These detailed measurements have provided further understanding about

Most investigations of surface pressure uctuations in wall-bounded ows consider only the linear term, and the simplication is normally justied by the large value of

123

Exp Fluids

the ow structures responsible for the wall-pressure generation. These ow structures comprise the sources of the wall-pressure eld, which are closely related to the sources of the so-called airfoil self-noise. Improving the understanding between the velocity eld and the surface pressure uctuations is also very important for the ow-induced vibration problems found in ship hulls or submarine periscopes for example. The stochastic estimation technique presented in this paper provided a picture of the quasi-instantaneous (velocity vectors) ow structures convecting past the location of a cross-wire and revealed information about the variability and evolution of the sources of the wall-pressure generation. From the stochastic estimation technique, due to the dominance of the large-scale vortices from the separated shear layer, the pseudo-instantaneous linear pressure 0 sources (qL 2 du ov ) were found to be distributed across dy ox the boundary layer.
Acknowledgments This work was funded by the UK Department of Trade and Industry under the MSTTAR DARP programme. The MSTTAR DARP was supported by Rolls-Royce plc and BAE Systems. The authors are grateful for fruitful discussions with Ahmed Naguib and his students Mohamed Daoud and Laura Hudy from Michigan State University and with Charles Tinney from University of Poitiers (currently at the University of Texas at Austin) and Lawrence Ukeiley from Florida State University. Special thanks for their support to Phil Joseph, Neil Sandham, Tze Pei Chong and Richard Sandberg from the University of Southampton and John Coupland and Philip Woods from Rolls-Royce and BAE Systems, respectively. The authors wish to thank Gavin Ross for his outstanding technical help with the airfoil model instrumentation and the experimental facility. Ana Garcia-Sagrado also gratefully acknowledges the nancial support from the Zonta International Amelia Earhart Fellowship.

References
Adrian R (1977) On the role of conditional averages in turbulence theory. In: Proceedings of the 4th biennal symposium on turbulence in liquids. Princeton, NJ Adrian R (1979) Conditional eddies in isotropic turbulence. Phys Fluids 22:20652070 Adrian R, Meinhart C, Tomkins C (2000) Vortex organisation in the outer region of the turbulent boundary layer. J Fluid Mech 422:154 Agarwal NK, Simpson RL (1989) A new technique for obtaining the turbulent pressure spectrum from the surface pressure spectrum. J Sound Vib 135((2):346350 Bearman PW (1971) Correction for the effect of ambient temperature drift on Hotwire measurements in incompressible ow. DISA Inf 11:2530 Blake W (1970) Turbulent boundary-layer wall-pressure uctuations on smooth and rough walls. J Fluid Mech 44:637660 Blake WK (1975) A statistical description of pressure and velocity elds at the trailing edges of a at structure. Technical report, David W. Taylor Naval Ship R&D Center Report No. 4241 Blake W (1986) Mechanics of ow induced sound and vibration, Volume I (General Concepts and Elementary Sources) and

Volume II (Complex Flow-Structure Interactions). Applied Mathematics and Mechanics Volume 17-I and Volume 17-II Bonnet J, Delville J, Glauser M, Antonia R, Bisset D, Cole D, Fiedler H, Garem J, Hilberg D, Jeong J, Kevlahan N, Ukeiley L, Vincendeau E (1998) Collaborative testing of eddy structure identication methods in free turbulent shear ows. Exp Fluids 25:197 Brooks TF, Hodgson TH (1981) Trailing edge noise prediction from measured surface pressures. J Sound Vib 78:69117 Champagne FH, Sleicher CA (1967) Application of multi-point correlation techniques to aerodynamic ows. Am Inst Aeronaut Astronaut 28:177182 Champagne FH, Sleicher CA, Wehrmann OH (1967) Turbulence measurements with inclined hot-wires. Part. 1 Heat transfer experiments with inclined hot-wire. J Fluid Mech 28:153175 Cox RN (1957) Wall neighborhood measurements in turbulent boundary layers using a hot-wire anemometer. A.R.C. Report 19191 Daoud MI (2004) Stochastic estimation of the ow structure downstream of a separating/reattaching ow region using wallpressure array measurements, Ph.D. thesis, Michigan State University Ewing D, Citriniti J (1997) Examination of a LSE/POD complementary technique using single and multi-time information in the axisymmetric shear layer. In: Sorensen, Hopnger, Aubry (eds) Proceedings of the IUTAM symposium on simulation and identication of organised structures in ows. Kluwer, Lyngby, pp 375384 Fahy FJ, Gardonio P (2007) Sound and structural vibration: radiation, transmission and response. Elsevier, Amsterdam Fiedler HE (1988) Coherent structures in turbulent ows. Prog Aerospace Sci 25:231269 Garcia-Sagrado A, Hynes T (2011) Wall pressure sources near an airfoil trailing edge under separated laminar boundary layers. AIAA J (submitted) Goody MC (1999) An experimental investigation of pressure uctuations in three-dimensional turbulent boundary layers, Ph.D. thesis, Virginia Polytechnic Institute and State University, Blacksburg, VA Gravante SP, Naguib AM, Wark CE, Nagib HM (1998) Characterisation of the pressure uctuations under a fully developed turbulent boundary layer. AIAA J 36(10) Guezennec Y (1989) Stochastic estimation of coherent structures in turbulent boundary layers. Phys Fluids A 1:10541060 Helal HM, Casarella MJ, Farabee T (1989) An application of noise cancellation techniques to the measurement of wall pressure uctuations in a wind tunnel. Am Soc Mech Eng 22:1422 Hinze JO (1959) Turbulence: an introduction to its mechanism and theory. McGraw-Hill Series in Mechanical Engineering, New York Hudy L, Naguib A, Humphreys W (2003) Wall-pressure-array measurements beneath a separating/reattahing ow region. Phys Fluids 15:706717 Mish PF (2001) Mean loading and turbulence scale effects on the surface pressure uctuations occurring on a NACA0015 airfoil immersed in grid generated turbulence, Master thesis, Virginia Polytechnic Institute and State University, Blacksburg, VA Naguib AM, Gravante SP, Wark C (1996) Extraction of turbulent wall-pressure time-series using an optimal ltering scheme. Exp Fluids 22:1422 Naguib A, Wark C, Juckenhfel O (2001) Stochastic estimation and ow sources associated with surface pressure events in a turbulent boundary layer. Phys Fluids 13:26112626 Roger M, Moreau S (2004) Broadband self-noise from loaded fan blades. AIAA J 42(3) Rozenberg Y, Roger M, Moreau S (2006) Effect of blade design at equal loading on broadband noise. In: Proceedings of the 12th

123

Exp Fluids AIAA/CEAS aeroacoustics conference, Cambridge, Massachusetts. AIAA Paper No. 2006-2563 Rozenberg Y, Roger M, Guedel A, Moreau S (2007) Rotating blade self noise: experimental validation of analytical models. In: Proceedings of the 13th AIAA/CEAS aeroacoustics conference, Rome. AIAA Paper No. 2007-3709 Taylor G (1938) The spectrum of turbulence. Proc R Soc Lond Ser A 164:476490 Tinney C, Coiffet F, Delville J, Hall A, Jordan P, Glauser M (2006) On spectral linear stochastic estimation. Exp Fluids 41:763775 Vernet A, Kopp GA, Ferre JA, Giralt F (1999) Three-dimensional structure and momentum transfer in a turbulent cylinder wake. J Fluid Mech 394:303337 Wark C, Naguib A, Ojeda W, Juckehoefel O (1998) On the relationship between the wall pressure and velocity eld in a turbulent boundary layer. Am Inst Aeronaut Astronaut, AIAA Paper No. 98-2642 Yu JC, Joshi MC (1979) On sound radiation from the trailing edge of an isolated airfoil in a uniform ow. AIAA Paper 79-0603

123

Potrebbero piacerti anche