Effect of a Simulated Glaze Ice Shape on the Aerodynamic Performance of a Rectangular Wing

Abdi Khodadoust University of Illinois at Urbana-Champaign Urbana, IL

AlAA 17th Ground Testing Conference July 6-8, 1992 1 Huntsville, AL

For permission to copy or republish, contact the American Institute of Aeronautics and Astronautics 370 L'Enfant Promenade, S.W., Washington, D.C. 20024

AIAA-92-4042 EFFECT OF A SIMULATED GLAZE ICE SHAPE ON

THE AERODYNAMIC PERFORMANCE OF A RECTANGULAR WING

AWi Khodadoust'

University of Illinois at Urbana-Champaign

ABSTRACT

The effect of a simulated glaze Ice accretion on the flow field of a three-dimensional wing is studied experimentally. The model used for these tests was a semi-span wing of effective aspect ratio five, mounted from the sidewall of the UlUC subsonic wind tunnel. The modd has a NACA 0012 airfoil section on a rectangular, untwisted planform with interchangeable leading edges to allow for testing both the baseline and the iced wing geometry. A four-beam two-color fiber-optic laser Doppler velocimeter (LDV) was used to map the flow field along three spanwise cuts on the model. Results of the LDV measurements on the upper surface of the 0012 finite wing model with and without simulated glaze ice accretion are presented for a = 0, 4, and 8 degrees at Reynolds number of 1.5 million. Measurements on the centerline of the clean model compared favorably with theory and centerline measurements on the iced model compared well with measurements on a similar 2-D model. The iced model results indicate that the flow has the largest separation bubble at the model midspan with the smallest separation bubble occurring near the root and the wing tip. Data on the model with simulated ice compare well with the 3- D computations and earlier 3-0 measurements.

I. INTRODUCTION

Aircraft and rotorcraft often collect ice on aerodynamic surfaces when encountering clouds of super-coded water droplets in flight. These accretions, if allowed to accumulate, result in serious aerodynamic penalties. The adverse effects of leading-edge ice formation on the aerodynamic characteristics of fixed wing aircraft are well known1. Understanding the aerodynamic penalties due to ice accretion on both lifting and non-lifting surfaces is important since many

components are not ice protected. The initial cost, cost of maintenance and weight penalty associated with Ice protection systems makes their use practical only on the most critical components.

Most icing experiments, where aerodynamic measurements have been made, have only dealt with twodimensional aircraft components. The experimental work of Bragg et. a ~ . ' ~ , and the corresponding computational research of ~otapczul? , ~ e b e c f , and ~anka?, have focused on a 2-D NACA 0012 airfoil with a simulated glaze ice accretion. Only the most recent work, Bragg et aL8-11 and Sankar et. have begun to investigate the flow field about a wing with simulated glaze ice accretion. ~ r a g d measured the surface pressures on a straight aspect ratio 5 wing with a NACA 0012 section and the simulated ice shape of Ref. 2 - 4. Kwon12 compared Navier- Stokes calculations with these data and showed good results except near the root where the sidewall boundary conditions differed. sankat3 and Kwon14 modeled the tunnel sidewall and improved the prediction near the root. Khodadous? and ~ r a ~ g ' " ' ~ extended the 3-D wing pressure and force balance measurements to include the effect of wing sweep.

This paper presents the results of the LDV measurements on the 001 2 finite wing model with and without simulated glaze ice accretion. Boundary-layer surveys on the centerline of the clean model are compared with theory. Data on the model with simulated ice will be presented next. LDV data is presented which maps the separation bubble at several spanwise stations. These data are compared with the 3-0 computationd4 and earlier 3-D mea~urementse"~ .

II. EXPERIMENTAL PROCEDURE

Model and Wind Tunnel The 3-D model used for this test is a

1 Graduate Research Assistant, Department of Aeronautical and Astronautical Engineering, Member AI AA.

Copyright * 1992 by Abdi Khodadoust. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. 1

semispan wing with a chord of 15 inches and a span of 37.25 inches, Fig. 1. A NACA 0012 airfoil section was used on this unswept wing. This airfoil section was chosen since tt is the same one used in the earlier 2-D and zero-sweep 3-D tests. The ice accretion used is a simulation of that measured on a NACA 0012 airfoil in the NASA Icing Research Tunnel, Fig. 2. The icing conditions were a free-stream velocity of 130 mph, angle of attack of four degrees, icing time of f i e minutes, vdume median diameter droplet of 20 microns, LWC = 2.1 and a temperature of 1gF.

The model is constructed primarily of fiberglass laid in a mdd and later filled with a foam core. Aluminum is used to reinforce the joints. The modd was constructed with a removable leading edge, extending to 0.15 chord. A clean leading edge and a simulated ice accretion leading edge were built for this study. The model also has removable root and tip sections to allow the study of swept wings, however this feature was not used in the present study. The root section is steel with a fiberglass skin. A steel spar of rectangular cross section was welded to this section and extends out of the tunnel to support the assembly.

The model is equipped with surface static pressure taps. The taps are located in 5 major rows plus a row on the tip section. The centerline row of taps has 80 taps in the no-ice configuration and 83 in the iced configuration. The other four rows on the main element have 40 and 41 taps in the no-ice and iced configurations, respectively. Including the 21 taps on the wing tip section, the model has a total of 261 taps in the no-ice configuration and 268 taps in the iced configuration. Pressure measurements have been made in previous experimepts using f i e Scanivalves.

These tests were conducted in the subsonic wind tunnel at the University of Illinois at Urbana- Champaign. The tunnel is of conventional design with approximately a three-by-four foot test section, eight feet in length. The tunnel operates at speeds from zero to 165 miles per hour at Reynolds numbers of up to 1.5 x 106 per foot. The tunnel is of open return type and uses four turbulence screens and honeycomb in the settling chamber to reduce tunnel turbulence to below 0.1 percent. Fig. 3 shows the model coordinate system with respect to the tunnel, or free stream, coordinate system. Earlier 2-D and 3-D experimental data were acquired in a similar, three-by-five foot tunnel, at The Ohio State University.

Laser Dowwler Velocimeter The laser Doppler velocimeter (LDV) system

used in the current investigation provided direct measurement of instantaneous velocity at any spatial location without intrusion into the flow field. The two-component velocity measurements were made using a four-Watt Argon-Ion laser together with a commercially available (TSI) four- beam twocdor fiber-optic LDV system, Fig. 4. The TSI Colorburst system, which consisted of a beam splitter, Bragg cell, cdor separator and beam alignment, separated the green component (514.5 nm) and the primary blue component (488.0 nm) of the Argon-Ion laser beam and guided them into input couplers. The cdor separated and frequency shifted beams from input couplers were transmitted via single-mode polarization preserving optical fibers to a 1200 mm transmitting lens. A 2.6:l beam expander was used to increase the input beam diameter and the effective scattered light collection aperture. This resulted in a smaller measuring vdume and better signal quality.

The green beam pair was aligned to measure the streamwise velocity component, while the blue beam pair measured the normal component. The measuring volume formed at the intersection of each beam pair had a diameter (e2) of 126 p m and a spanwise extent of 2.3 mm for the green beams, and a diameter of 1 19 p m and a spanwise extent of 2.2 mm for the blue beams. For this optical configuration, a total of 26 fringes were produced in the measurement vdume. A summary of measurement vdume dimensions for the optical setup used in this study is given in TaMe 1.

Various seeding particles, such as polystyrene latex spheres, mineral oil smoke and propylene glycd were available for these tests. Three factors were considered in choosing proper seed material: first, the closed operating environment of the UlUC tunnel excludes the use of chemicals not suitable for human intake; second, seed particles should be large enough to provide adequate signal-to-noise (SNR) ratio; and third, the seed particles should follow the flow fluctuations with high fidelity. Based on the above considerations, propylene glycol solution in water was chosen for use with a commercially available (TSl SixJet) atomizer. Propylene glycol produces harmless polydisperse particles with a median diameter of approximately 0.9 p m15. This size particle produces adequate SNR and will follow the flow fluctuations up to 7.0 kHz with better than 99 percent fidelity.

The four-beam twocolor system was designed to operate in back-scatter configuration. This means that the scattered light from particles

crossing the measuring volume was collected by a 350 mm lens on the same side of the tunnel and the same axis as the transmitting optics. The collected light was transmitted through a multl- mode fiber-optic cable to a cdor separator. After the green scattered light was separated from the Mue scattered light, each was sent to a photomultiplier. The photomultiplier signals were downmixed, then processed using TSI 19808 counter processors. Counter filters were set to pass frequencies between 30 MHz and 100 KHz. A total of 8 fringe crossings counted were required to validate a Doppler burst, and a 1 percent comparison was used in the single- measurement-per-burst mode. The counter outputs were sent to an AT&T 6386WGS PC by a TSI 1998M master interface. Depending on the measurement location and data rate, a total number of samples ranging from 4096 In coincidence mode to 1024 in random mode were used for flow calculations. The outputs were processed and various flow statistical quantities were calculated using the TSI FIND Ver. 2.5 LDV data acquisition and display package on the PC.

In keeping with convention, the boundary- layer profiles were developed by traversing the LDV measurement volume in a direction normal to the airfoil contour. For this purpose, a mapping routine was written and linked to the LDV three- axis traverse driver program. The routine allows the operator to choose any scan length and increment combination normal to the airfoil surface, normal to the chord, or normal to the free stream. The commercially available (VELMEX) traverse, oriented to the tunnel axis system was used to move the measuring volume in a two-foot cube to within 0.001 inch resolution. All boundary-layer sweeps, unless otherwise noted, were made normal to the airfoil upper surface.

Errors and Statistical Uncertainty The prima sources of bias and uncertainty

in the were due to cross-beam angle measurement, clock synchronization, and diverging fringes. The cross-beam angle measurement is affected by the accuracy of the focusing lens and the spacing between the illuminating beams prior to entry into the beam expansion module. The error due to clock synchronization stems from the & 1 clock coygt ambiguity inherent to counter type processors . The problem with diverging fringes occurs when the two illuminating beams do not cross to form a measurement volume at their focal points and as a result, a measurement inaccuracy occurdg. This proMem can be circumvented by passing the

laser beam through a beam collimator on the transmitting optics train. The beam collimator in the optical setup used for the current study, consisting of a positive and negative lens, placed the beam crossing point and the focused-beam waist in the same location, thus assuring that the fringes were parallel. A summary of bias and random errors are given in Table 2. The total error in the system due to these components amounted to a minimum of 0.65 percent and a maximum of 0.85 percent.

The high-speed counter used for velocity measurements during this study operates on Individual bursts that are produced by particles passing through the measurement volume. These random individual realizations are collected into an ensemble from which the flow mean velocity and higher statistical moments are calculated. Since a finite number of samples are used to calculate these quantities, statistical procedures can be used to estimate the number of samples which would be required to determine mean velocity and turbulence intensity within an acceptable error level for a given confidence interval and an assumed population distribution17.

Fig. 5 shows the uncertainty in the measured mean value versus sample size for f i e turbulence intensity levels and a 95 percent confidence interval from a normal or Gaussian distribution. The result of a similar analysis for the uncertainty in standard deviation is shown in Fig. 6. Here, for a normally distributed parent population, the uncertainty in standard deviation is only a function of the confidence level and the sample size.

Based on the curves shown in Figs. 5 and 6, at a 95 percent confidence limit the statistical uncertainty was less than one percent at a turbulence level of 20 percent when 2048 samples were taken. The uncertainty was three percent at a 50 percent turbulence level and six percent at a 100 percent turbulence level when 1024 samples were acquired. The statistical uncertainty in standard deviation was 4.3 percent for a sample size of 1024 and 3.1 percent for a sample size of 2048 at the 95 percent confidence limit. The number of samples collected in the flow field under investigation depended on the data rate and the turbulence intensity. Near the wing surface, in the shear layer, or in the recirculation zone aft of the ice horn, the data rate decreased substantially. Only 1024, and whenever possible, 2048 data samples were taken at these locations. An attempt was made to collect more data points at these locations, but the amount of time required to obtain this ensemble proved to be unrealistic (on the order of 30 minutes per measurement location) in a boundary layer survey consisting of at least six such locations. In the

free stream, the data rate was very high, but the recorded turbulence intensities were no more than two percent, therefore only 1024 data samples would have sufficed in order to achieve a high degree of accuracy. When the data rate was not too low, 4096 data samples were collected. Thls included the region near the stagnation point on both the clean and the iced wing leading edges. Altogether, close to 10 million samples were cdlected during the entire study.

A possible source of error associated with counter processors is called velocity bias. Velocity bias occurs since, in a uniformly seeded flow, more particles pass through the measurement vdume at higher velocities than at lower velocities. As a result, the arithmetic mean of the individual realizations is biased towards the higher velocitiede. In general, velocity bias is ne ligible when the Reynolds shear stress is lo$'. Velocity bias becomes significant when the Reynolds stresses are large. This occurs in the shear layer where the turbulence intensities are high. Several correction methods, discussed by the panel chaired by ~dwardg', were applied to a subset of the LDV measurements. The subset was chosen so that the highest turbulence intensities in the flow (-35 percent) were represented. The velocity profile at a = 8 0 , x/c=0.02, y/b=0.470 is shown in Fig. 7. The uncorrected profile is compared with the profile corrected with residence time weightindl. The velocity profile was also corrected with an inverse velocity method which takes into account the effect of frequency shiftind2. Both correction methods showed little change in velocity profiles compared with the uncorrected profile.

All bias correction procedures recommended by the Edwards panel assume ideal behavior, which is violated in the recirculation region aft of the ice h o d 3 . One of the asswmptions is that seeding should be spatially uniform. Another is that each detected particle must produce only one valid Doppler burst. Not only is the uniform seeding assumption violated in the recirculating region of the flow, but also frequency shifting may cause a detected particle to produce more than one valid Doppler burst. As a result, the burst time reported by the counter may not be representative of the true residence time. The corrected results shown in Fig. 7 seem to have lower velocity values, which is the correct trend. In light of the violation of the fundamental assumptions underlying these corrections, however, it is not clear whether these correction methods yield corrected numbers or just different numbers. Since the corrections reported by these methods seem to cause little change in the velocity profiles, it was decided to present the

data without velocity bias correction. Two other possible sources of error are fringe

bias and velocity gradient bias. Briefly stated, fringe bias may exist because a particle whose velocity vector is perpendicular to the fringes in the measurement vdume has the highest detection probability while a particle whose velocity vector penetrates the measurement vdume at shallow angles to the fringes has the lowest detection probability. This measurement dependence on the velocity direction is called fringe bias or directional ambiguity. Frequency shifting, In the amount at least twice the maximum anticipated flow velocity, removes the fringe bias p r o ~ e d 4 . The #-MHz frequency shift imposed on the Illuminating beams produced a fringe velocity of 190 m/s. This is more than twice the maximum anticipated velocities in the flow field, therefore fringe bias effects were negligible.

Gradient bias occurs due to the finite size of the measurement vo lum~?~. If large velocity gradients exist within the measurement vdume, false turbulence fluctuations as well as mean velocity will be recorded by the LDV instrument. In the semispan wing flow-field under study, the maximum velocity gradients in the mean velocity occur in the middle of the shear layer near the model surface. Gradients as steep as 36 m/s per millimeter existed in the 28 m/s shear layer. For the green beam, which has a measurement volume diameter of 126 microns, this yields 0.25 percent error for the mean velocity and 0.22 percent for the turbulence intensity due to spatial resdution. These values represent the 'worst cases' in the flow field under study. Less than f i e percent of the measurement locations have such high mean velocity gradients, therefore, the spatial resolution error was negligible in the flow field.

Ill. RESULTS AND DISCUSSION

The presence of the ice shape drastically alters the flow field around the wing. This can perhaps be seen most effectively by comparing the pressure distribution around the clean model with the pressure distribution around the iced mod&. Figure 8 shows the effect of ice on the measured pressure distributions at a wing angle of attack of six degrees. The simulated ice causes a reduction in lift coefficient (0.461 to 0.431). Also note that the spike in the pressure distribution associated with a clean subsonic airfoil has been replaced by constant pressure regions leveling at -1.4 on the upper surface and - 0.25 on the lower surface. These constant pressure plateaus are caused by the separated

flow aft of the upper and lower surface ice horns, respectively. The LDV-measured velocity profiles on the clean and iced model will be presented next.

Clean Wina Data Profiles taken with the laser Doppler

velocimeter (LDV) on the clean wing are compared with preliminary inviscid flow-field computations since CFD data were not available in a suitable format for one-to-one comparison at the time of this writing. The measurements were taken at y/b = 0.470 on the model. All LDV surveys on the clean model, starting at 0.050 inches from the model surface, were taken normal to the wing surface. The computational scheme is a 2-D panel method which models the NACA 0012 airfoil plus the wind tunnel upper and lower wall?. This computational model, written originally for wind tunnel droplet trajectory computations, was modified to calculate the velocity at any given LDV traverse coordinate.

Profiles for the streamwise component of the velocity at a = 0 are shown in Fig. 9. Comparison between measured and computed profiles show very good agreement near the wing section leading edge, gradually deteriorating as the trailing edge is approached. Note that at all chordwise stations, near the surface, comparison shows little agreement between measurement and computation. This is to be expected since the computational model is based on inviscid flow and therefore does not model viscous effects near the wing surface. Further away from the surface, comparison shows very good agreement up to x/c=0.30. Near x/c=0.30, and further downstream, the agreement between measurement and computation becomes less favorable, differing by as much as three percent near the wing section trailing edge.

Also shown here is the comparison between LDV measurements and a computational model based on Theodorsen's methd7 which, unlike the computational model mentioned above, does not take into account the effect of wind tunnel walls. The agreement between measurement and this computational model was quite good up to x/c = 0.60 (not shown). Further downstream, however, the computational model which included the wall effects clearly was in better agreement with LDV measurement. Further comparisons at a = 4' and a = g are made only with the panel code with tunnel walls modelled.

Comparison between computation and measurement for a =4' is shown in Fig. 10. At x/c = 0.02, agreement is excellent away from the model surface. Near the wing surface, the two results differ by as much as 11 percent. Very

good agreement between computation and measurement is seen further downstream at x/c = 0.30. The difference between the measured and computed profiles at all x/c stations (some not shown) is less than four percent, except in the region very near the model surface. As mentioned previously, the region close to the model surface can not be modelled properly with the computational model used for this preliminary comparison.

Further downstream, LDV measurements agree less favorably with the panel method results. In addition to threedimensional effects on the wing, this can also be attributed to the fact that boundary-layer growth on the model results in a new geometry which includes the boundary- layer displacement thickness plus the wing cross section. Since the computational scheme does not use a boundary-layer model, this new 'inviscid shape' is not accounted for in the computations.

LDV measurements of the streamwise velocity component at a =€? are compared to computation in Fig. 11. Overall, measurement shows good agreement with lnviscid computational results. Here, there is excellent agreement between computation and measurement at x/c=0.70 while the computed and measured profiles differ by as much as three percent at x/c = 0.02.

Iced Wina Data A comparison of the boundary layer profiles

on the 3-D model centerline with those made on a similar but 2-0 model may show the two- dimensional behavior of the 3-D wing centerline. No LDV measurements on the 2-D model have been made. Split hot-film measurements, however, were obtained on the 2-D model in the Ohio State University tunnef4. In order to compare the split hot-film-measured profiles from the OSU tunnel with the LDV-measured profiles in the UIUC tunnel, the proper 3-0 wing angle of attack had to be determined. Doing so would ensure that the flow-field at the wing centerline would behave in the same manner as the flow- field on the 2-0 model in the OSU tunnel.

The split hot-film measurements at OSU were taken at a '4" in a tunnel similar to the one in use at UIUC. The measured pressure distribution for this 2-D model was compared to the measured pressure distribution at y/b=0.497 for the 3-0 model, also tested in the OSU tunnd. The information obtained from this comparison was used to determine a suitable angle of attack for the 3-D model in the UlUC tunnel, which has a different test-section dimension.

Assuming that the spanload distribution of the model in the UIUC tunnel was the same as that in

the OSU tunnel, it was determined that at a = 4.?, the 3-D model centerline would have the same pressure distribution as the 2-0 model at a =$, where the split hot-film profiles were obtained in the OSU tunnel.

In Fig. 12, LDV measurements obtained at y/b=0.470 on the iced 3-D wing are compared with split hot-film measurements made on the similar 2-0 model in the OSU wind tunnel. The hot-film profiles were obtained by traversing a split hot-film probe normal to the free stream. For these comparisons, the LDV profiles were also obtained by traversing the measurement vdume normal to the free stream.

Results for x/c=0.02 show very good agreement in the reverse-flow region and through the shear layer near the wing surface. The LDV- measured edge velocity differs from the split hot- film-measured edge velocity by as much as ten percent. Both measurements, however, seem to approach the same nondimensional free-stream velocity.

The trends described above are aiso seen further downstream at x/c=0.04, 0.08 and 0.10. Keep in mind that, although care was taken in choosing a suitable angle of attack for comparison of LDV results with the split hot-film results, the threedimensional features of the flow- field on the 3-0 wing, near the tip and root, can not be avoided. For these comparisons, a =4.? was chosen based on an estimate of how closely the pressure distribution on the centerline of the 3-D wing and that on the 2-0 model in the OSU tunnd would match. Earlier pressure measurements on this model indicated that at a =$, the separation bubble extended well beyond ten percent chord in the streamwise direction. The edge velocity was & /U = 1.463 3% for the hot-film-measured profiles shown in Fig. 12. The edge velocw measured by the LDV instrument was y /U = 1.614 1 % for the LDV-measured profiles shown in Fig. 12. This means that, based on the edge velocity, both sensors indicated a constant- pressure region with a 27 percent difference in the level of pressure plateaus. The pressure distribution, and therefore the C, and a,,, at this spanwise location on the 3-D model in the UlUC tunnel may not be quite the same as that in the OSU tunnel. Overall however, the flow field and aerodynamic performance of the 2-0 airfoil compares very well to the centerline of the 3-D model.

Previous measurements on the similar 2-D model indicated a large recirculating region of fluid aft of the ice horn.?. On the 3-D model under study, a similar region has been identified. This region is bounded by the model surface on

the bottom and a shear layer on the top. The time record of the fluctuating velocity (ul=instantaneous velocity - mean velocity) at x/c = 0.10 for a = 4' on the model centerline is shown in Fig. 13 for several scan locations normal to the iced wing contour. At z/c=0.0033 (0.050 inches) above the model surface, the u' values were centered about a small bandwidth of fluctuations with sporadic peaks of large positive velocity. The recorded turbulence intensity, based on the free-stream veiocity, was 15.4 percent. The intermittency factor, r, , is defined by ~ i m ~ s o r ? as the fraction of the time the flow moves downstream in a turbulent separating boundary layer. 7, =0.053 at this location meant that the flow was moving upstream, against the free-stream direction, most of time. The spikes of large positive velocities indicate the passage of large scale structures within the shear layer. The structures within the shear layer have a higher velocity than the flow very near the wall. As the measurement vdume is traversed further into the middle of the shear layer, where the highest turbulence intensities were recorded, the u' fluctuations became centered about a wide velocity band with seemingly equal distributions of positive and negative velocities. The maximum recorded turbulence intensity was 32.5 percent occurring at z/c=0.0133. 7, was 0.544 for this station, indicating almost equal number of negative and positive velocity occurrences. As the LDV measurement vdume exited the shear layer toward the edge of the boundary layer, the measured turbulence intensity dropped to 12.9 percent at z/c =O.O267. The u' fluctuations appear to become centered about a narrow bandwidth with sporadic large negative peaks. These peaks correspond to the large scale structures moving downstream in the shear layer. The velocity of these structures is smaller than that in the free stream, hence the peaks in slower velocity. 7, was 1.0 for z/c=0.02 and points further out, indicating that the flow was moving downstream 100 percent of the time.

The points noted above can be generalized to the entire reverse flow region aft of the ice horn. Typical turbulence intensity levels in the shear layer were around 30 to 35 percent based on the free stream velocity. This is 10-1 5 percent higher than typical values for plane mixing layers. The high turbulence level, similar to that found behind a backward facing step, is believed to be largely due to a low frequency vertical or flapping motion of the reattaching shear layep.

Streamwise velocity contours on the upper surface of the iced wing are shown in Fig. 14 for a = 0,4, and g . At a =d' , the reverse flow region was measured to x/c=0.07, 0.085, and 0.07 near

the root, mid-section, and the wing tip respectively. The tunnel wall-root interaction near the model root and the tip Induced vortex flow near the wing tip were the probable causes for the shorter extent of the reverse flow region compared with that in the model mid-section. At a =$, the reverse flow was seen to extend to x/c = 0.135 near both the model root and the mid- section. Near the wing tip, at y/b=0.819, this region extended only to x/c =0.125. At a =€? the model flow field was reported to be very unsteadf*" . The streamwise velocity contours for a = g , also shown in Fig. 14, show the extent of the reverse flow to have reached x/c=0.36 near the modd root and x/c=0.24 near the wing tip. The reverse flow region was measured up to x/c =0.60 in the streamwise direction on the wing centerline (not shown).

The reattachment lines from flow visualization perhaps best show the extent of reverse flow on the wing upper surface. In Fig. 15, experimental and computational surface oil flow visualization results are shown for the iced wing ata =g. The computational results are from simulated oil flow generated by tracing the trajectories of massless particles in the Navier-Stokes flow field. The experiment was carried out at a chord Reynolds number of 1.2 million while the computation was performed at 1.5 million. Both pictures show the largest separated flow region near the model mid- section with the smaller separation occurring near the model tip and root. Note that the reattachment line can not be characterized as a well defined line, mainly due to the unsteady flow in the reattachment zone.

instrument makes It a less-than-ideal instrument for investigating 3-D flow fields.

Inspection of the model centerline flow field indicated that a large region of reverse flow existed aft of the ice horn on the iced model. At a =@, this region extended to 7 percent chord. At a = 4", the bubble grew to more than 12 percent chord. At a = g , the time-averaged separation bubble was measured well beyond 50 percent chord. Earlier experiment!$' have shown the flow field on the iced model to be very unsteady at thls angle of attack. Experimental and computational flow visualization support these findings. More detailed analysis of the LDV data, including analysis of vertical component velocities, may provide a better understanding of this complex flow field.

The flow in the vicinity of the ice shape contained many of the features of the flow over a backward facing step. Typical recorded turbulence intensity levels were within the established range for a backward facing step. Also similar to the backward facing step, the magnitude of reverse flows measured in the recirculation zone aft of the upper surface ice horns were typically over 30 percent of the free- stream velocity.

At a =$, little spanwise effect was observed near the wing root and the centerline. Close to the wing tip, however, the effect of the wing tip- induced vortex flow was seen to shorten the length of the separation bubble. The effect of the wing tip and wing root-wall interaction was more pronounced at a=€?, as observed from experimental and computational flow visualization.

IV. SUMMARY ACKNOWLEDGEMENT

Results from LDV measurements were presented for the upper surface of a rectangular semi-span wing with and without simulated glaze ice. Preliminary comparison of LDV measurements with inviscid theory show the measured profiles to be in good agreement. More detailed corn rison with CFD results of Kwon and Sankatpaare planned for the near future. Comparison between the profiles measured with a split hot-film probe and profiles measured with the LDV instrument show very good agreement near the model surface. Near the boundary layer edge, however, agreement is less favorable between the two measurement techniques. It must be kept in mind that LDV is a non-intrusive measurement technique, while the presence of the hot-film probe does alter the flow field. While this may be suitable for 2-D type flows, the intrusive nature of the hot-film

This work was supp~rted in part by a grant from NASA Lewis Research Center. The author wishes to thank Dr. M. G. Potapczuk of NASA Lewis for his support of this research.

REFERENCES

1. Preston, G. M. and Blackman, C. C., 'Effects of Ice Formations on Airplane Performance in Level Cruising Flight,' NACA TN-1598, May 1948.

2. Bragg, M. B. and Coirier, W. J., "Aerodynamic Measurements of an Airfoil with Simulated Glaze Ice,' AIM-86-0484, paper presented at the 24th Aerospace Sciences Meeting, Reno, Nevada, Jan. 6-9. 1986.

3. Bragg, M. B. and Spring, S. A.,"An Experimental Study of the Flow Field about an Airfoil with Glaze Ice,' AiAA-874100, paper presented at the 25th Aerospace Sciences Meeting, Reno, Nevada, January 12-1 5, 1987.

4. Bragg, M. B. and Khodadoust, A., "Experimental Measurements In a Large Separation Bubble Due to a Simulated Glaze Ice Accretion,' AIM-884116, paper presented at the 26th Aerospace Sciences Meeting, Reno, Nevada, January 1 1-1 4, 1988.

5. Potapczuk, M. G., 'Navier-Stokes Computations for a NACA 0012 Airfoil with Leading Edge Ice,' AlAA Paper No. 874101, presented at the 25th Aerospace Sciences Meeting, Reno, Nevada, Jan. 12-15, 1987.

6. Cebeci, T., "Effects of Environmentally Imposed Roughness on Airfoil Performance,' NASA CR 179639. June 1987.

7. Sankar, L. N., Wu, J. C. and Kwon, 0. J., "Development of Two- and Three-Dimensional Navier-Stokes Solvers for Aircraft lcing Studies," presented at the Annual Airfoil Performance-in- lcing Workshop, NASA Lewis Research Center, Cleveland, Ohio, July 25, 1988.

8. Bragg, M. B. and Khodadoust, A., 'Effect of Simulated Glaze Ice On a Rectangular Wing," AiAA-894750, paper presented at the 27th Aerospace Sciences Meeting, Reno, Nevada, January 9-1 2, 1989.

9. Khodadoust, A. and Bragg, M. B., "Measured Aerodynamic Performance of a Swept Wing With a Simulated Glaze Ice Accretion," AlAA Paper 90- 0490, 1990.

10. Bragg, M., Khodadoust, A., Soltani, R., Wells, S. and Kerho, M., 'Effect of Simulated Ice Accretion on the Aerodynamics of a Swept Wing," AIM-914442, paper presented at the 29th Aerospace Sciences Meeting, Reno, Nevada, Jan. 7-10, 1991.

11. Bragg, M. B., Khodadoust, A., Kerho, M., 'Aerodynamics of a Finite Wing With Simulated !ce,' paper presented at the 5th Symposium on Computational and Physical Aspects of Aerodynamic Flows, Cebeci, T., ed., Long Beach, California, January 1992.

12. Kwon, 0 . and Sankar, L, "Numerical Study of the Effects of lcing on Finite Wing Aerodynamics,' AIM-904757, paper presented at the 28th

Aerospace Sciences Meeting, Reno, Nevada, January 8-1 1, 1990.

13. Sankar, L. and Kwon, L., "Numerical Studies of the Effects of lcing on Fixed and Rotary Wing Aircraft Aerodynamics,' presentation at the Airfoil- in-Icing Workshop, Nasa Lewis Research Center, Sept. 1990.

14. Kwon, 0. J. and Sankar, L. N., "Numerical Investigation of Performance Degradation of Wings and Rotors Due to Icing," AIM-924412, paper presented at the 30th Aerospace Sciences Meeting, Reno, Nevada, January 6-9, 1992.

15. Instruction Manual, 'Model 9306 6Jet Atomizer,' TSI Incorporated, June 1987

16. Young, Warren H., Meyers, James F., Hepner, Timothy E.,' Laser Velocimeter Systems Analysis Applied to a Flow Survey Above a Stalled Wing," NASA TN 0-8408, 1977.

17. Coleman, H. W. and Steele Jr., W. G., Experimentation and Uncertainty Analysis for Engineers, John Wiley and Sons, New York, 1989.

18. Durst, F., Melling, A., Whitelaw, J. H.,Principles and Practice of Laser-Doppler Anemometry, Academic Press, 1981.

19. Hanson, S., "Broadening of the Measured Frequency Spectrum in a Differential Laser Anemometer Due to Interference Plane Gradients,' J. Phys. D: Appl. Phys. 6(2), Jan. 1973, pp. 164-1 71.

20. Meyers, J. F., March 1992, Priiate Communication.

21. Edwards, R. V., 'Report on the Special Pand on Statistical Bias Problems in Laser Anemometry," Transactions of ASME: Journal of Fluids Engineering, Vd 109, June 1987, pp. 89- 93.

22. Meyers, J. F., and Clemmons J. I., Jr, "Processing Laser Velocimeter High-speed Burst Counter Data," Laser Velocimetery & Particle Sizing, Thompson, H. D. and Stevenson, W.~H., Editors, Hemisphere Publishing Corp., Washington, 1979, pp.300-313.

23. Adriin, R. J., Feb. 1992, Private Communication.

24. Whiffen, M. C., 'Pdar Response of an LV Measurement Vdume,' Proc. of the Minnesota Symposium on Laser Anemometry, Bloomington, Minnesota, October 1975, pp. 589-590.

25. Karpuk, M. E., and Tiederrnan, W. G. Jr., 'Effect of Finite-Size Probe Vdume Upon Laser Doppler Anemometer Measurements,' AlAA Journal, 14(8), August 1976, pp. 1099-1 105

26. Wells, S., L and Bragg, M. B., 'A Computational Method for Calculating Droplet Trajectories Including the Effects of Wind Tunnel Walls,' AIM-92-0642, paper presented at the 30th Aerospace Sciences Meeting, Reno, Nevada, Jan. 6-9, 1992.

27. Bragg, M. B., 'Rime Ice Accretion and Its Effect on Airfoil Performance,' Ph.D. Dissertation, The Ohio State University, Columbus, Ohio, 1981.

28. Bragg, M. B. ,Khodadoust, A. and Spring, S. A., 'Experimental Measurements in a Large Leading-Edge Separation Bubble Due to a Simulated Airfoil Ice Accretion,' AlAA Journal, to be published in July 1992.

29. Simpson, R. I. ' Interpreting Laser and Hot- Film Anemometer Signals in a Separating Boundary Layer,' AlAA Journal, vd. 14, no. 1, pp. 124-1 26, January 1976.

30. Johnston, J. P. and Eaton, J. K., 'A Review of Research on Subsonic Turbulent Flow Reattachment,' AlAA Journal, vd. 19, no. 9, pp. 1093-1099, September 1981.

span = 37.25" chord = 1 5 . 0 Eflect~ve Aspect Ratio = 4.96 Unlw~sted. Untapcrcd. Straight P l ~ n r o r m N A C A 0012 Cross Section

N A C A 00 12 Icing Conditions a - 4 ' U,= 130 m p h

7 - 20pm LWC - 2.1 g / m 3

5 min

- Measured Shape

- - - - - - Sirnulaled S h a p e

Fig. 2 Simulated Glaze Ice Accretion.

S~de V~ew

Table 1 Measurrnwnt Vdume StatiDba I

--

Table 2. Systemat~c Errors

Bl w

3 1

4.512

11086

2208

02907

25

M V Stat6

Half-Angle(d0~ )

Frmge S p a c ~ n ~ Ir rn)

Bameter (r rn)

Lmgth (rnrn)

volume ( m d )

Number of Fr~nges

( Model P~ io r Locarion: J c 4 . 3 3 3 . y/c=0, z l c 4

\ I

&eon

3 1

4 7%

126 16

2 329

0 5407

26

Clock Synchron~zatlon

Cwerg~ng Frmges

Total System Error

Total Effecltw Error

Top V ~ e u

9 Fig. 3 Tunnel and Model Coordinate System

+O25

4 13

+ 0 77 4 53

z 02s

20 12

20.37

065% to 0.85%

U-COMPONENT PROFILE x / c = o 0 2

O6 I l l CORREC110 0 0 5 : LANGLEY WEIGHT FUNCTION

RESIDENCE TIME WEIGHING

I Col l~ rn~ lor 6 F~ber Couplcr 11 Rccc~vmg Fiber Opt~c 2 Opi~cal F~lter 7 Flber Optlc Bundle 12 Rccclvlng Collmator 3 Bcam Guldc 8 SO rnrn Flber Probe 13 b c h r o ~ c Mlnur 4 Color Burst 9 Beam Expander I 4 Pho~cnnulopl~cr (green) 5 Transrn~ii~nf 10 Transrnmlng Lens I 5 Pho~omulupl~cr (blue)

F ~ k r Oylc Cables

Fig. 4 Layout of LDV Setup and Electronics Fig. 7 Velocity Bias Corrected Data

LEVEL OF COIdFIDENCE = 0 35

NACA O G 1 1

- 6 d.8

P. - 1 . 5 . 106

Fig. 5 Uncertainty in Mean Velocity

I I STATISlICAL UNCERTAINTY i

I N STANDARD DEVIATION

i - 857. COtIrlDENCE - - - 90Z CONFIDENCE

- - - - - - 95Z CONFIDENCE - - s9? CObJrlDFtlCC

. \ , .

Fig. 8 Surface Pressure Distribution From Wing Midspan With and Without Simulated Glaze Ice.

0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5000

NUMBER OF VELOCITY SAMPLES

Fig. 6 Uncertainty in Standard Deviation

LDVDATA o THEODORSEN

00 o PANEL CODE WlTH

00 WALLS MODELLED O D 0 m 0 l

0 0 00 m 0 I 0 0

0 1 0 a

o a

LDV DATA o THEODORSEN

0 o PANELCODE WITH 0 . WALLS MODELLED D.

w e m m e 00.

001

0 a

LDV DATA o THEODORSEN 0 PANEL CODE WlTH

0 l WALLS UOOELLEO 0 .

w . m e Do.

0 0

o m 0 a

LDVDATA o n o THEODORSEN

0 0 . o PANEL CODE WITH

0 0 WALLS MODELLED

0 0 . 0 0 8

0 0

0 0 . 0 0 .

Fig. 9 Comparison Between 2-D computation and LDV Measurements of the Streamwise Component of Velocity on the Clean Rectangular Wing at a = oO, Re= 1 SXI 06, y/b= 0.470.

LDV MEASUREMENT 0 PANEL CODE WlTH

WALLS MODELLED

8 LDV MEASUREMENT o PANEL CODE WlTH

WALLS MODELLED

0 50 r LDV MEASUREMENT

0 8 o PANEL CODE WITH

0. WALLS MODELLED

LDV MEASUREMEIUT o PANEL CODE WlTH

0 . WALLS U ~ D E L L E D 0 .

0 .

Fig. 10 Comparison Between 2-D Computation and LDV Measurements of the Streamwise Component of Velocity on the Clean Rectangular Wing at a = 8 , Re = 1 SXI 0 6 , y/b = 0.470.

LDV MEASUREMENT 0 PANEL CODE WITH

WALLS MODELLED

X/C = 0.50 a, = 8"

LDV MEASUREMENT o PANEL CODE WITH

WALLS MODELLED

LDV MEASUREMENT . 0 0 PANEL CODE WITH

0 WALLS MODELLED . 0 . 0 . 0

n n l 0 .

LDV MUSCIREMENT o PANEL CODE WITH

WALLS M O D E L L E ~

Fig. 11 Comparison Between 2-D Computation and LDV Measurements of the Streamwise Component of Velocity on the Clean Rectangular Wing at a =€?, Re= 1 .5~ ld5 , y/b=0.470.

LDV MEASUREMENT. 3-D 0 LDV MEASUREMENT. 3-D o SPLIT HOT-FILM MEASUREMENT, 2-0 o SPLIT HOT-FILM MEASUREMENT, 2 - D

Fig. 12 Comparison Between Split Hot-Film Measurements of Refs. 2-4 and LDV Measurements of the Streamwise Component of Velocity on the Rectangular Wing with Simulated Glaze Ice Accretion at a =4.7', ~ e = 1 . 5 ~ 1 $ , y/b=0.470.

O 'O

0 08

0 0 7 .

5 0 0 5 .

0 0 3

0 0 2

' X/C = 0.08 O'O'

. l LDV MEASUREMENT. 3 -D . 0 08 0 SPLIT HOT-FILM MEASUREMENT, 2-D 0

0 0 0 7 .

< 0 0 5 .

Pi - 0 ooO .a 0 0 3 . 0 0 O .

o C O O ~ O O O - 0 0 2 .

X/C = 0 10

. l LDV MEASUREMENT. 3 - D 0 o SPLIT HOT-FILM MEASUREMENT. 2 - 0 . .

0 a

.o',oO

0 0 0

*Qob, 8"0°

@'

- 0 4 0 - 0 1 0 0 2 0 0 5 0 0 8 3 1 1 0 1 4 0 170 - 0 40 -0 :0 0 20 0 50 0 8: 1 1C 1 4 3 ; 'C

u / u LI / Y

I " " " " " " " " 1

Fig. 13 Time History of the Fluctuating Velocity on the Rectangular Wing with Simulated Glaze Ice Accretion at x/c=0.10, Re = 1 SXI Cf , y/b = 0.470.

Fig. 14 See Page 16.

NOTES