Overview and Summary of The Second AIAA High Lift Prediction Workshop (Invited)
Overview and Summary of The Second AIAA High Lift Prediction Workshop (Invited)
Overview and Summary of The Second AIAA High Lift Prediction Workshop (Invited)
R=20140011924 2020-01-31T16:43:32+00:00Z
Christopher L. Rumsey∗
NASA Langley Research Center, Hampton, VA 23681
and
Jeffrey P. Slotnick†
The Boeing Company, Huntington Beach, CA 92647
The second AIAA CFD High-Lift Prediction Workshop was held in San Diego, California, in June 2013.
The goals of the workshop continued in the tradition of the first high-lift workshop: to assess the numer-
ical prediction capability of current-generation computational fluid dynamics (CFD) technology for swept,
medium/high-aspect-ratio wings in landing/takeoff (high-lift) configurations. This workshop analyzed the flow
over the DLR-F11 model in landing configuration at two different Reynolds numbers. Twenty-six participants
submitted a total of 48 data sets of CFD results. A variety of grid systems (both structured and unstructured)
were used. Trends due to grid density and Reynolds number were analyzed, and effects of support brackets
were also included. This paper analyzes the combined results from all workshop participants. Comparisons
with experimental data are made. A statistical summary of the CFD results is also included.
Nomenclature
AR aspect ratio
alpha, α angle of attack, deg
B wing span
CD drag coefficient
CDp idealized drag coefficient = CD − CL2 /(πAR)
CL lift coefficient
CL,max maximum lift coefficient
CM pitching moment coefficient
Cf surface skin friction coefficient
Cf,x x-component of Cf
Cp surface pressure coefficient
Cv coefficient of variation = σ̂/μ̂
c mean aerodynamic chord
h average grid spacing measure
K confidence interval coverage factor
LB lattice-Boltzmann
M Mach number
N number of grid points or grid cells
Re Reynolds number
Rec Re based on c length scale
u velocity component in the x-direction
x, y, z Cartesian coordinate directions
Δ difference between high and low Re results
μ̂ median of sorted data
σ̂ standard deviation
∗ Senior Research Scientist, Computational AeroSciences Branch, Mail Stop 128, Fellow AIAA.
† Boeing Technical Fellow, Computational Sciences, Senior Member AIAA.
1 of 37
I. Introduction
The prediction of high-lift aerodynamic flows is a challenging but important aspect in the analysis of aircraft
configurations. Today’s computational fluid dynamics (CFD) codes have proven to be reliable and consistent for
simplified, cruise-type configurations, but the complexities inherent in high-lift configurations add a significant degree
of uncertainty. The complexities in high-lift flows include wakes in pressure gradients, wake/boundary-layer merging,
streamline curvature, separated flow, possible unsteady flow, wing-tip vortical flow, and laminar/turbulent transition
regions on each wing element (slat, main wing, flap).1
To advance the state of the art in predicting high-lift flows, an open international workshop series was estab-
lished with the following long-term objectives: (1) assess the numerical prediction capability (mesh, numerics, tur-
bulence modeling, high-performance computing requirements, etc.) of current-generation CFD technology for swept,
medium/high-aspect ratio wings in landing/take-off (high lift) configurations, (2) develop practical modeling guide-
lines for CFD prediction of high lift flow fields, (3) advance the understanding of high lift flow physics to enable
development of more accurate prediction methods and tools, (4) enhance CFD prediction capability for practical high
lift aerodynamic design and optimization, (5) provide an impartial forum for evaluating the effectiveness of existing
computer codes and modeling techniques, and (6) identify areas needing additional research and development.
Previously, the first high lift prediction workshop (HiLiftPW-1) was held in June 2010 in Chicago, Illinois. The
workshop focused on the three-element NASA Trapezoidal Wing configuration2, 3 at Rec = 4.3 × 106 ; the workshop
overview can be found in Slotnick et al.,4 and the summary is given in Rumsey et al.5 The overall conclusions were
as follows. CFD tended to under-predict lift, drag, and the magnitude of the pitching moment (pitching moment
was negative) compared with experiment, and there was significantly more spread among the CFD solutions near
stall. Despite the general under-prediction in lift, many participants were able to predict CL,max and the angle of
attack at which it occurred reasonably well. The trends with grid refinement were generally in the correct direction
(approaching experiment), but it was difficult to draw firm conclusions because CFD grid convergence studies were
only done without support brackets. Including support brackets resulted in lower predicted lift. Transition modeling,
included by only a few participants, appeared to be very important for this configuration, tending to increase predicted
lift. An unstructured tetrahedral grid solution was found to exhibit greater grid sensitivity than the solution on the
same grid with its boundary-layer tetrahedra merged into prisms. The flow field near the wing tip was very difficult
to predict. Nearly all entries seriously under-predicted the suction near the wing-tip upper surface, and neglecting
viscous cross-derivative terms yielded even worse predictions near the wing tip than full Navier-Stokes. Statistically,
the deltas predicted by CFD between two different configurations were somewhat low at α = 13◦ , and significantly
high at α = 28◦ . Subsequently, many of the HiLiftPW-1 participants wrote papers highlighting their workshop
results,6–19 and additional papers were produced for a later special session focused on the NASA Trapezoidal Wing at
an AIAA meeting in June 2012 in New Orleans, Louisiana.20–32
The second high lift prediction workshop (HiLiftPW-2) was held in June 2013 in San Diego, California.a The
configuration selected for analysis was the DLR-F11 three-element wing-body model.33 This configuration was pre-
viously used in the EUROLIFT research program.34 The DLR-F11 was of interest because it was more representative
of a realistic aircraft configuration than the NASA Trapezoidal Wing. Also, because the DLR-F11 was tested at both
low and high Reynolds numbers, a major focus of HiLiftPW-2 was the ability of CFD to capture the Reynolds number
effects.
This paper summarizes the HiLiftPW-2 workshop data, including several corrections and additional submissions
by some participants soon after the workshop. A total of 26 groups submitted 48 entries. (This was slightly more than
HiLiftPW-1, which had 21 groups and 39 entries.) In addition to describing the overall collective results in comparison
to experimental data, an attempt is made in this paper to identify trends as well as outliers. In the interest of brevity,
only representative results are shown in some cases.
The paper is organized as follows. First, the geometry and experimental data are briefly described, followed by a
summary of the test cases and grid systems. A summary of the entries is then given. The results section includes a
turbulence model verification test that was an optional exercise for the workshop participants, intended to help identify
possible turbulence model implementation issues. Then, for the DLR-F11, the main results include force, pitching
moment, grid convergence, iterative convergence, velocity, and surface pressure and skin friction coefficients. A
statistical analysis is conducted in which scatter limits are calculated. As part of this statistical analysis, the ability to
predict the differences between low and high Reynolds numbers is looked at in greater detail. Finally, conclusions are
drawn.
a http://hiliftpw.larc.nasa.gov, accessed 11/13/2013.
2 of 37
II. Geometry and Experimental Data
The DLR-F11 model—a generic semi-span, three element high lift wing configuration with a body pod—is shown
in Fig. 1. It is intended to be representative of a commercial wide-body twin-jet high lift configuration, but the
workshop focused on a simplified full-span flap geometry. The untwisted swept wing had a semi-span (B/2) of 1.4 m
and an aspect ratio (AR) of 9.353. The mean aerodynamic chord (c) was 0.34709 m. The high lift system consisted of
a leading edge slat and a trailing edge Fowler flap, both full-span for the cases considered in this workshop. Transition
was fixed on the body pod, but not on any of the wing elements. The slat was attached to the main wing using seven
slat tracks. The flaps were mounted with five flap tracks and a fixing of the inner flap edge at the fuselage. The flap
tracks were covered by flap track fairings. The high lift devices directly intersected with the fuselage. Due to the
small size of the slat tracks, it was necessary to attach pressure tube bundles along the outside of the tracks for their
routing into the slat element. Additional geometry and instrumentation details can be found in Rudnik et al.33 Note
that throughout this paper, the slat tracks and flap track fairings are often collectively referred to as “support brackets”
or “brackets.”
The model was tested both at low Reynolds numbers in the low-speed wind tunnel Airbus-Deutschland (B-LSWT),
as well as at high Reynolds numbers in the European Transonic Windtunnel (ETW). The wing was equipped with
487 pressure taps in ten chordwise pressure sections. In addition to forces, moments, and surface pressures, the
low Reynolds number test included flow field velocity measurements using particle image velocimetry (PIV) and oil
flow visualization. Other measurements not directly considered here included hot film, infared, and tuft videos. At
Rec = 1.35 × 106 , CL,max occurred near α = 19◦ . Oil flow visualization at low Reynolds number suggested that
at stall there were large isolated areas of separated flow over the main wing behind two of the slat tracks, spreading
out laterally towards the trailing edge. In stall, the largest area of separated flow over the main wing was near mid-
span. Close to the flap trailing edge, separation trends were visible even at low angles of attack. The high Reynolds
number test included forces, moments, and surface pressures. Tuft videos were made, but are not considered here. At
Rec = 15.1 × 106 , CL,max occurred near α = 20 − 21◦ . Additional details about the experimental data, including
short- and medium-term repeatability as well as facility effects, can be found in Rudnik et al.33
3 of 37
• Case 2b (required) - high Reynolds number conditions;
• α = 0◦ , 7◦ , 12◦ , 16◦ , 18.5◦ , 20◦ , 21◦ , 22.4◦ ;
• run fully turbulent
V. Summary of Entries
A summary of the entries to the HiLiftPW-2 is given in Tables 2 and 3. It includes an identifier for each submission,
the code used, the code type (structured or unstructured), the grid system, the turbulence model, and other relevant
information. All submissions were assigned an entry number, which consists of a primary number in front of the
decimal point that identifies the group or person submitting the data, and a secondary sequence after the decimal point
that identifies variations (for example, different grids, codes, or turbulence models). There were 26 individuals/groups
who submitted 48 entries. The primary numbers are not sequential because there were three groups who dropped out
prior to the workshop. Note that the structured codes that used grid system A employed point-match blocking, whereas
those that used grid system E employed overset.
For turbulence models, most participants used the Spalart-Allmaras (SA) model35 or variant. Five of the variants
are rotation/curvature (RC),36 rotation (R),37 Edwards,38 SALSA,39 and quadratic constitutive relation (QCR).40 The
so-called “noft2” variant simply ignores the ft2 term in the turbulence model equation, which is believed to make very
4 of 37
Table 1. Summary of supplied grids for Case 1 (number of grid points/cells given in millions)
little difference.b Other models utilized by participants were SST,41 SST-V,42 K-e-Rt,43 EARSM,44 and Wilcox1988.45
There were also several non-traditional (not Reynolds-averaged Navier-Stokes) methods. The UNICORN code, a gen-
eral Galerkin adaptive finite element method, used an implicit subgrid scale (SGS) model by numerical stabilization.
The PowerFLOW code, a lattice-Boltzmann (LB) method, used a proprietary very large eddy simulation (VLES) wall
model. The XFlow lattice-Boltzmann code used a proprietary wall-modeled local-eddy simulation (WMLES) model.
For transition, an amplification factor transport (AFT) model,46 the SST-GRET model,47 and a proprietary transition
method (in PowerFLOW) were used.
VI. Results
A. Turbulence Modeling Verification
An optional verification exercise (Case 4) was included in HiLiftPW-2 in part because a similar exercise had previously
proved to be very useful for identifying discrepancies in turbulence modeling implementations at the fifth drag predic-
tion workshop (DPW-5).48 Here, a simple 2-D bump case was used. At the website http://turbmodels.larc.nasa.gov,
several different independent CFD codes (some of which have been previously verified) using the SA, SST, SST-V,
and Wilcox2006 models have been shown to produce nearly identical results for this case as the grid is refined. Thus,
the website has established high confidence in a grid-converged result for each of these models for this case. Any
other code should expect to approach the same answer as the grid is refined, when using the same turbulence model
and boundary conditions. A different solution indicates either an incorrect implementation of the model, an incor-
rect procedure or boundary condition, or a bug in the code. For example, in DPW-5, it was learned that inaccurate
computation of the minimum distance function could lead to significant errors in SA results.
There were ten Case 4 submissions in HiLiftPW-2. Eight of these were for the SA model (or variant). Figure 3
shows the SA and SA variant results. In Fig. 3(a), the Cf over a portion of the bump is shown. Three of the results
were indistinguishable: 002.1, 002.2, and 013.1. (Note that 002.1 and 002.2 corresponded to results already posted on
the website http://turbmodels.larc.nasa.gov.) Of the other results, 007.1, 007.2, and 011.1 were slightly different, and
the model variants 011.2 (SA-Edwards) and 011.3 (SA-salsa) were significantly different.
Comparing like-models only (SA) in the grid convergence study of Fig. 3(b), one can see that indeed 002.1, 002.2,
and 013.1 were all approaching the same Cf at location x = 0.6322 as the grid was refined (h2 → 0). Submission
007.1 was only slightly low, but 007.2 was 2% low and 011.1 was about 3% high in this particular metric. Comparing
Cf at other locations yielded similar results. This study indicates that there are potential problems or inconsistencies
in the implementation of the SA model in 007.2 and 011.1 that should be investigated.
The other two Case 4 submissions were 010.1 (K-e-Rt) and 018 (SST). The former currently has no other code to
compare against for Case 4, and the latter used a different turbulence model for its Cases 1 and 4. Therefore, these
results are not shown here.
b See http://turbmodels.larc.nasa.gov, accessed 11/13/2013.
5 of 37
Table 2. Summary of entries
6 of 37
Table 3. Summary of entries (continued)
of CL , CD , or CM is plotted as a function of N −2/3 , which in 3-D is proportional to h2 . Finer grid results are on the
left of the plots and coarser grid results are on the right. An infinitely fine grid would be plotted as N −2/3 = 0. The
corresponding experimental value is shown as a circle symbol on the y-axis of each plot for reference. If all codes
and implementations were completely consistent and if all turbulence models were the same, then the band of results
should tighten, approaching a single result as N −2/3 → 0. This did not occur here. Overall, the scatter range remained
more or less the same across the different grid levels. One trend was evident at the higher angles of attack (Fig. 6):
the block-structured and overset results tended to be fairly tightly clustered, whereas the unstructured results showed
a larger variation (scatter). Although not shown, results at α = 21◦ were similar to those in Fig. 6, whereas results at
α = 22.4◦ showed large scatter for all CFD results.
7 of 37
entire lift curve. Figure 7 also highlights the effects of Reynolds number. Ignoring a few outlier computations, at both
Reynolds numbers much of the linear portion of the polar CL was under-predicted, CD was over-predicted, and the
magnitude of CM was over-predicted. Collectively, the CFD showed a greater spread near maximum lift, with some
under-predicting and some over-predicting CL,max and the angle at which it occurred. In Fig. 8, the idealized profile
drag coefficient CDp is shown. Displaying CDp instead of CD helps to expand the scale of the plot and see the drag
variation better.
Only a few participants investigated the effects of including transition. Lift, idealized profile drag, and pitching
moment coefficients are shown in Fig. 9. (Note that 003 and 009 used Config 4 whereas 012 used Config 5.) The
results were mixed. Recall that the 009.2 results were outliers for non-transitional results, so it is not clear whether
the trends displayed here are representative. For 003, including transition had little influence on the lift and idealized
drag below CL,max , and tended to only slightly lower the pitching moment magnitude in general. However, for 012
transition had more significant effects. In particular, lift was increased over the entire lift curve. This trend of transition
causing higher lift is consistent with conclusions from NASA Trapezoidal Wing investigations.7, 18, 20, 22, 28
There were only three participants who included computations for Case 3 (the configuration with pressure tube
bundles alongside the slat brackets): 012, 021, and 026. Results for CL are shown in Fig. 10. Participant 012 only
computed Case 3, so the influence of the pressure tube bundles cannot be isolated. However, both 021 and 026
computed on both Config 4 (Case 2) and Config 5 (Case 3). In Figs. 10(b) and (c) one can see that the pressure
tube bundles had little influence except near stall, where they caused loss of lift. This trend was consistent with an
experimental investigation.33 Although it is not clear which configuration yielded better agreement with experiment,
certainly the potential influence of the pressure tube bundles on stall behavior for this model cannot be ignored.
C. Iterative Convergence
One of the conclusions from HiLiftPW-1 was that iterative convergence information should be collected from future
workshop participants. This collection was done for HiLiftPW-2 for Case 1, although not all participants provided
the requested information (see Tables 2 and 3). All who did comply provided force and moment histories; many
also included various residual convergence histories. Some representative results are plotted in Figs. 11 and 12 for
α = 7◦ . In Fig. 11, results are shown from those entries for which the final CL still appeared to be changing by the
last reported iteration. Iterations are normalized so that 0 represents the first iteration reported and 1 represents the
last. Entry 004.1, 005.1, and 005.2 each showed a changing CL with negative slope at the last iteration, while 011.1,
011.3, and 022 each showed significant oscillatory behavior (note that 022 was run time-accurately). As shown in
Fig. 11(b), the 004.1, 005.1, and 005.2 entries indicated converging density equation residuals (all residuals have been
normalized so that they start at 1). Both 011.1 and 011.3 displayed “stalled” residual behavior, and 022 did not supply
residual information.
Figure 12 shows representative results from several entries for which the final CL appeared to be better converged
(relatively flat or unchanging). Interestingly, most of the residuals for these apparently better-converged results either
“stalled” or leveled off. Three of the representative results drove the residuals down about 5 or 6 orders of magnitude
prior to leveling off, while two exhibited weaker convergence over the iteration range reported. Although not shown,
representative iterative results for other angles of attack were similar to those shown here. These results indicate that
the HiLiftPW-2 high lift cases were generally not easy to converge for many CFD solvers. They also offer a caution
when analyzing the workshop data: insufficient iterative convergence may account for some of the differences between
different CFD codes.
D. Velocity Profiles
Velocity data were collected for Cases 1, 2, and 3, although not all participants provided the requested information.
In this section, we divide a sampling of the results into two parts. First, results are shown for Case 1 (CFD only -
no brackets), for which CFD results at α = 7◦ are compared to each other for consistency. (Because Case 1 did not
involve brackets, experimental data are not included.) Second, results are shown for Cases 2a and 3a, for which CFD
results at α = 7◦ and 18.5◦ are compared to low Reynolds number experiment.
Figures 13 and 14 show CFD Case 1 results at α = 7◦ at all eleven stations. As depicted in Fig. 13(a), the stations
were labeled in the experiment according to their spanwise row location (1 for y = 246.386 mm, 2 for y = 979.596
mm, and 3 for y = 1223.999 mm), then by area (B for near the slat trailing edge, C for forward part of the main wing,
D for aft part of the main wing, and E for the flap). The final number designation stands for the line number extracted
within the area. In the figures, several outliers have been highlighted in blue. In some cases these results may have
been different due to errors made by the participants when extracting the velocity profiles. Ignoring the outliers, the
8 of 37
results in red show reasonably consistent velocity profiles at many of the locations, but this consistency deteriorated at
the outboard two stations on the rear part of the main wing and on the flap. The results at location 3E2 showed a very
large variation.
Inspecting the plots in Figs. 13 and 14, one can see that some of the results showed distinct, relatively deep wake
profiles and some did not. This difference was primarily a function of the grid employed. As an example, Fig. 15 shows
results at location 3E1, divided among different grids. Structured results on medium grids A and E are highlighted in
bold blue in Fig. 15(a), unstructured results on medium grids A and C are highlighted in bold red in Fig. 15(b), and
unstructured results on medium grids B and D are highlighted in bold red in Fig. 15(c). The B and D grids yielded
significantly smeared (under-resolved) wake profiles from the upstream elements compared to the other grids. Even
use of the fine B and D grids (shown in Fig. 15(d)) did not improve this aspect of the results at all. Although not
shown, results at other locations showed similar trends.
Figure 16 shows low Reynolds number results at three selected stations and two angles of attack. The CFD results
include both Case 2a (brackets included) and Case 3a (brackets and pressure tube bundles included). Aside from a few
outliers, the CFD results were reasonably consistent with each other at α = 7◦ , but less so at α = 18.5◦ . CFD results
tended to be shifted from experiment, predicting velocities too low. Only two participants computed Case 3a; there
were no notable consistent trends exhibited at these locations by adding the pressure tube bundles.
9 of 37
F. Statistical Analysis
In the statistical analysis of this section, the basic method used by Morrison49 was employed. For any quantity of
interest (CL for example), the scatter limits were established as μ̂ ± K σ̂, where μ̂ is the median √ of the sorted data,
σ̂ is the standard deviation, and K is a confidence interval coverage factor, taken to be K = 3. The coefficient of
variation, defined as Cv = σ̂/μ̂, provides a measure to compare the variation of populations with different medians.
This same statistical methodology was also used in the analysis of HiLiftPW-1.5 A statistical outlier (a point residing
outside of the scatter limits) is by definition different from the results that lie within the scatter limits. In terms of CFD
solutions, being an outlier means that some aspect of the CFD solution was significantly different from the collective,
and its cause(s) should be investigated. In the plots to follow, any sample lying well outside of the limits is noted in
the figure. However, because the scatter limits depend on the choice of K, the fact that a particular point lies outside
may or may not be significant.
Figure 20 shows results from the statistical analysis of Case 1 (no brackets, high Re) at α = 7◦ . The top row is
CL , middle row is CD , and bottom row is CM . The left column shows coarse grid results, middle column is medium
grid results, and right column is fine grid results. In all cases, (as is desirable) the scatter range and Cv both decreased
as grid was refined. However, the biggest difference occurred between coarse and medium grids, and very little change
occurred between the medium and fine grids. The few outliers produced lift values that were too low compared to the
collective. Median lift increased with grid refinement while pitching moment decreased. Drag changed very little.
Figure 21 shows results from the statistical analysis of Case 1 at α = 16◦ . In this case, scatter range and Cv again
decreased for CL with grid refinement, but for drag they remained fixed on the medium and fine grids and for pitching
moment the scatter range increased slightly on the fine grid. The outliers again produced lift values that were too low
compared to the collective. Median lift increased with grid refinement while pitching moment decreased and drag
increased only slightly.
Figure 22 summarizes the coefficients of variation for Case 1 at α = 7◦ and 16◦ and compares them with the Cv
values from HiLiftPW-1. Cv levels for lift and drag were somewhat similar between the two workshops, although
the current HiLiftPW-2 results decreased less (flattened more) between the medium and fine grids (i.e., the fine grid
Cv levels for HiLiftPW-2 tended to be higher than the fine grid Cv levels for HiLiftPW-1). |Cv | levels for pitching
moment were higher than for HiLiftPW-1, and also flattened more than for HiLiftPW-1 between the medium and fine
grids. In other words, for HiLiftPW-2, going from medium to fine grids overall did not tend to bring the various CFD
results much closer together, as would generally be expected. The reason for this is not known.
In Fig. 23, an attempt was made to seek trends for Case 1 by turbulence model, by grid type, and by grid. Specific
trends should manifest themselves by showing a particular bias compared to the median. However, few significant
trends stood out among the results with a sufficiently large population; most symbols appeared to be fairly randomly
distributed. One possible trend was that structured-overset results (corresponding to grid E) for α = 16◦ tended to
give higher than average lift. Another noteworthy point is that at both angles of attack the two SA results giving the
lowest lift were 007.2 and 011.1. Recalling the verification exercise from Fig. 3(b), these particular implementations
of SA were suspect.
Figures 24 through 28 display statistical analyses of forces and pitching moments for Cases 2a and b (which
included brackets) at α = 0◦ , 7◦ , 16◦ , 20◦ , and 21◦ , respectively. Experimental values (yellow circles) are also
displayed in the figures for comparison. For both Reynolds numbers at α = 0◦ and 7◦ , the median values of lift,
drag, and pitching moment were somewhat different compared to experiment, but the CFD scatter bands contained
the experiment. At α = 16◦ , the median CFD value of CL agreed almost perfectly with experiment at both Reynolds
numbers, but drag was significantly over-predicted and pitching moment was significantly under-predicted. At α =
20◦ , the median CFD value of CL was again excellent compared with experiment at both Reynolds numbers. Drag
was predicted well at the low Reynolds number condition, but again over-predicted at high Reynolds number. Moment
was significantly under-predicted at both Reynolds numbers. At α = 21◦ , the median CFD value of CL was only
slightly high compared to experiment (the CFD scatter band contained the experiment). Drag was under-predicted
at the low Reynolds number condition and over-predicted at high Reynolds number. Moment was again significantly
under-predicted.
Referring back to Fig. 7, recall that the entries exhibited more spread near stall (with some entries over-predicting
CL,max and some under-predicting it). Interestingly, the median value from the CFD codes happened to come out
relatively accurately in predicting the lift at the highest angles of attack. However, this may have been luck. At
high Reynolds number the median CL of the CFD was still increasing at α = 21◦ , whereas in the experiment it
was starting to decrease. Furthermore, the large spread of CFD behavior near CL,max makes it doubtful that the
CFD codes collectively and consistently captured the physical mechanism(s) responsible for stall. The reasons for the
mis-predictions in drag and pitching moment over much of the polars above α = 7◦ is not known.
10 of 37
To further explore how well the median values from the CFD codes captured Reynolds number trends, delta values
of CL , CD , and CM are plotted in Fig. 29. For example, ΔCL = CL,hiRe − CL,loRe . Overall, the shapes of the
delta curves as a function of angle of attack were qualitatively captured, but ΔCL at low α was over-predicted, and
the magnitude of the negative ΔCD at high α was under-predicted.
VII. Conclusions
The entries from the second AIAA High-Lift Prediction Workshop (HiLiftPW-2) have been summarized, both by
looking at collective results compared with experiment and through statistical analyses. A large number of surface
pressures, skin frictions, and velocity profiles were collected, but only a small portion of these have been analyzed to
date, and only a few representative plots have been shown here.
An optional turbulence model verification case was included in this workshop, and computed by a few of the
participants. For those who used the SA model, there was near-perfect consistency among three entries, one other
entry was very close, and two exhibited notable differences. These two differences suggest potential problems or
inconsistencies in implementation, which should be investigated by the owners of the codes in question.
For the DLR-F11 case in general, the importance of including slat and flap brackets when comparing with experi-
ment was established. From previous experimental oil flow, it appeared that two slat tracks were influential in causing
large wedge-shaped regions of separated flow on the main element near stall. Without brackets, the CFD could not
possibly capture this type of stall mechanism, and indeed most entries without brackets tended to predict increasing
CL well past the nominal stall angle. There was also some evidence from the few participants who ran Case 3 that
pressure tube bundles included alongside the slat tracks on the wind tunnel model had an influence on the flow field
near CL,max .
In terms of consistency, the CFD exhibited spreads in results that tended to remain about the same when going
from the medium to the fine grid levels. In other words, the CFD scatter did not decrease much past a certain grid
refinement level; the reasons for this are not known. Scatter was larger at the angles of attack near stall, as expected. A
small part of the CFD inconsistencies may have been due to poor or insufficient iterative convergence, as participants
had some difficulty fully converging many of the cases. Looking at details such as pressure coefficients and velocity
profiles, the inconsistencies between CFD results tended to be larger on the flap as well as at the outboard stations of
the wing. Some of the unstructured grids were shown to be poor for capturing the wakes of upstream elements.
Most of the participants used the Spalart-Allmaras turbulence model or variant. Nonetheless, the variation among
entries using this model appeared to be as great as differences due to the use of other turbulence models. In other
words, no clear trends with turbulence modeling were exhibited in the results. A few participants investigated the use
of transition, but unlike results for the NASA Trapezoidal Wing of the first high lift workshop, results were mixed and
no clear trends stood out.
When including the slat and flap brackets, the median CFD results tended to be slightly low predicting lift at
α = 7◦ , accurate at α = 16◦ and 20◦ , then somewhat high at α = 21◦ . However, the scatter range for CFD’s CL
results (approximately 0.2 at low angles of attack and typically higher in the range of 0.3−0.6 at high angles of attack)
always included the experimental values. This was generally not the case for drag and pitching moment, both of which
were consistently mis-predicted at angles of attack above α = 7◦ . Reynolds number trends were only qualitatively
captured by the CFD.
HiLiftPW-2 continued in the tradition of the first workshop in the series, by again bringing together many re-
searchers from around the world with a variety of CFD codes to analyze a specific high lift configuration. This new
configuration offered different challenges from the configuration of the first workshop. The CFD results again lacked
consistency to the levels desired, but much was learned by quantifying the differences and by taking steps to attempt to
isolate possible causes. For example, including verification testing as well as iterative convergence information were
both very useful aspects that could serve as prerequisites for future workshops. Workshops like this are very valuable
for the aeronautical engineering community, and should continue to be pursued.
Acknowledgments
The authors wish to express their gratitude to HiLiftPW-2 committee members Dave Levy and Mark Chaffin of
Cessna for their help preparing plots and grid analysis for the workshop talks. Thanks also go to Stefan Melber
of DLR, Leonel Serrano, Raul Mendoza, Karuna Rajagopa, Neal Harrison, Yoram Yadlin, and John Vassberg of
Boeing, and committee members Carolyn Woeber of Pointwise, Mike Long of University of Wyoming, Mark Chaffin
of Cessna, and Tony Sclafani of Boeing for creating the grids. The authors further acknowledge committee members
11 of 37
Ralf Rudnik and Kerstin Huber of DLR for pulling together and organizing the DLR-F11 data. Finally, the authors
thank all committee members mentioned above plus Thomas Wayman of Gulfstream, Tom Pulliam of NASA Ames,
Dimitri Mavriplis of University of Wyoming, and Judi Hannon of NASA Langley for their continued support with the
planning and execution of the high lift workshops and special sessions.
References
1 Rumsey, C. L. and Ying, S. X., “Prediction of High Lift: Review of Present CFD Capability,” Progress in Aerospace Sciences, Vol. 38, 2002,
pp. 145–180.
2 Johnson, P. L., Jones, K. M., and Madson, M. D., “Experimental Investigation of a Simplified 3D High Lift Configuration in Support of CFD
the 14- by 22-Foot Subsonic Tunnel (Invited),” AIAA Paper 2012-0706, January 2012.
4 Slotnick, J. P., Hannon, J. A., and Chaffin, M., “Overview of the First AIAA CFD High Lift Prediction Workshop (Invited),” AIAA Paper
January 2011.
8 Wiart, L. and Meunier, M., “Computational Assessment of the HiLiftPW-1 Trap-Wing Model Using the elsA CFD Software,” AIAA Paper
the First High Lift Prediction Workshop,” AIAA Paper 2011-866, January 2011.
10 Eliasson, P., Peng, S.-H., and Hanifi, A., “Improving the Prediction for the NASA High-Lift Trap Wing Model,” AIAA Paper 2011-867,
January 2011.
11 Reyes, D. A., Girimaji, S. S., Pandya, M., and Abdol-Hamid, K. S., “Computations of High-Lift Wing-Body Configuration on Unstructured
January 2011.
16 Antunes, A. P., Azevedo, J. L. F., and da Silva, R. G., “Numerical Simulations of Turbulent Flow over a High-Lift Configuration,” AIAA
June 2012.
28 Sclafani, A. J., Slotnick, J. P., Vassberg, J. C., and Pulliam, T. H., “Extended OVERFLOW Analysis of the NASA Trap Wing Wind Tunnel
12 of 37
30 Lopez, O. D., Ochoa, N., Mahecha, J. C., Leguizamon, S., Escobar, J. A., Ramirez, S., Jimenez, R. A., Giraldo, A. M., and Silva, C.,
“Numerical Simulation of NASA Trap-Wing Model as a Colombian Contribution to the High-Lift Prediction Workshop,” AIAA Paper 2012-2921,
June 2012.
31 Fabiano, E., Fares, E., and Nolting, S., “Unsteady Flow Simulation of High-Lift stall Hysteresis using a Lattice Boltzmann Approach,”
Wing with a Moving Flap Configuration,” AIAA Paper 2012-2923, June 2012.
33 Rudnik, R., Huber, K., and Melber-Wilkending, S., “EUROLIFT Test Case Description for the 2nd High Lift Prediction Workshop,” AIAA
5–21.
36 Shur, M. L., Strelets, M. K., Travin, A. K., and Spalart, P. R., “Turbulence Modeling in Rotating and Curved Channels: Assessing the
Spalart-Shur Correction,” AIAA Journal, Vol. 38, No. 5, 2000, pp. 784–792.
37 Dacles-Mariani, J., Zilliac, G. G., Chow, J. S., and Bradshaw, P., “Numerical/Experimental Study of a Wingtip Vortex in the Near Field,”
1598–1605.
42 Menter, F. R., “Improved Two-Equation k-omega Turbulence Models for Aerodynamic Flows,” NASA TM 103975, October 1992.
43 Goldberg, U., Peroomian, O., Batten, P., and Chakravarthy, S., “The k-e-Rt Turbulence Closure,” Engineering Applications of Fluid Me-
Codes,” AIAA Journal, Vol. 47, No. 12, 2009, pp. 2894–2906.
48 Levy, D., Laflin, K., Tinoco, E., Vassberg, J., Mani, M., Rider, B., Rumsey, C., Wahls, R., Morrison, J., Broderson, O., Crippa, S., Mavriplis,
D., and Murayama, M., “Summary of Data from the Fifth AIAA CFD Drag Prediction Workshop,” AIAA Paper 2013-0046, January 2013.
49 Morrison, J. H., “Statistical Analysis of CFD Solutions from the Fourth AIAA Drag Prediction Workshop,” AIAA Paper 2010-4673, June-
July 2010.
13 of 37
Figure 1. Photograph of DLR-F11 high lift model in the B-LWST wind tunnel.
(a) Cf on finest grid (detail) (b) Grid convergence study of Cf at x = 0.6322 (SA only)
14 of 37
(a) CL , medium grids (b) CL , fine grids
Figure 4. Case 1 (no brackets, high Re) forces and pitching moments (experiment shown for reference).
15 of 37
(a) CL , α = 7◦ (b) CL , α = 16◦
Figure 5. Case 1 (no brackets, high Re) grid convergence behavior at α = 7◦ and 16◦ (experiment shown for reference).
16 of 37
(a) CL , α = 18.5◦ (b) CL , α = 20◦
Figure 6. Case 1 (no brackets, high Re) grid convergence behavior at α = 18.5◦ and 20◦ (experiment shown for reference).
17 of 37
(a) CL , Rec = 1.35 × 106 (b) CL , Rec = 15.1 × 106
Figure 7. Case 2 (medium grids with with brackets) forces and pitching moments.
18 of 37
(a) Rec = 1.35 × 106 (b) Rec = 15.1 × 106
(c) CM
Figure 9. Effects of including transition on forces and pitching moments, Rec = 1.35 × 106 (medium grids with brackets).
19 of 37
(a) Participant 012 (b) Participant 021
Figure 10. Effect of including pressure tube bundles on CL (medium grids with brackets).
20 of 37
(a) Lift coefficient (b) Normalized density equation residual
Figure 11. Case 1 (no brackets, high Re) α = 7◦ iterative convergence information for entries with CL still changing at final iteration.
Figure 12. Case 1 (no brackets, high Re) α = 7◦ representative iterative convergence information for entries with CL relatively flat at
final iteration (all entries not shown).
21 of 37
(a) Sketch showing locations of velocity profiles (b) Location 1B1
Figure 13. Case 1 (no brackets, high Re) α = 7◦ velocity profiles from CFD results on medium grids (part 1); obvious outliers denoted by
blue lines.
22 of 37
(a) Location 2B2 (b) Location 2D1
Figure 14. Case 1 (no brackets, high Re) α = 7◦ velocity profiles from CFD results on medium grids (part 2); obvious outliers denoted by
blue lines.
23 of 37
(a) All medium grid results, with structured results on medium A and (b) All medium grid results, with unstructured results on medium A
E grids highlighted in bold blue and C grids highlighted in bold red
(c) All medium grid results, with unstructured results on medium B (d) Medium and fine unstructured B and D grid results only
and D grids highlighted in bold red
Figure 15. Case 1 (no brackets, high Re) α = 7◦ velocity profiles at location 3E1, highlighted by grid type (outliers removed).
24 of 37
(a) Location 1C1 at α = 7◦ (b) Location 1C1 at α = 18.5◦
Figure 16. Case 2a (with brackets, low Re) and 3a (with brackets and bundles, low Re) velocity profiles at selected locations using medium
grids.
25 of 37
(a) Flap at 2y/B = 0.15, no brackets (b) Flap at 2y/B = 0.15, including brackets
(c) Slat at 2y/B = 0.89, no brackets (d) Slat at 2y/B = 0.89, including brackets
(e) Flap at 2y/B = 0.96, no brackets (f) Flap at 2y/B = 0.96, including brackets
26 of 37
(a) Slat at 2y/B = 0.54 (b) Slat at 2y/B = 0.89
Figure 18. Pressure coefficients at two span stations for α = 20◦ (with brackets included).
27 of 37
(a) Flap at 2y/B = 0.15, α = 7◦ (b) Flap at 2y/B = 0.15, α = 16◦
Figure 19. The x-component of skin friction coefficients on the flap at α = 7◦ and 16◦ (with brackets included).
28 of 37
(a) CL , coarse grid (b) CL , medium grid (c) CL , fine grid
Figure 20. Statistical analysis of Case 1 (no brackets, high Re), α = 7◦ ; solid blue line is median of data, dashed blue lines are scatter
limits.
29 of 37
(a) CL , coarse grid (b) CL , medium grid (c) CL , fine grid
Figure 21. Statistical analysis of Case 1 (no brackets, high Re), α = 16◦ ; solid blue line is median of data, dashed blue lines are scatter
limits.
30 of 37
(a) Lift coefficient (b) Drag coefficient (c) Moment coefficient
(a) α = 7◦ , separated by turbulence (b) α = 7◦ , separated by grid type (c) α = 7◦ , separated by grid
model
(d) α = 16◦ , separated by turbulence (e) α = 16◦ , separated by grid type (f) α = 16◦ , separated by grid
model
Figure 23. Statistical analysis of Case 1 (no brackets, high Re) CL on medium grids, separated by types.
31 of 37
(a) CL at Rec = 1.35 × 106 (b) CL at Rec = 15.1 × 106
Figure 24. Statistical analysis of Cases 2a and b (includes brackets), α = 0◦ on medium grids (yellow circle at right represents experiment).
32 of 37
(a) CL at Rec = 1.35 × 106 (b) CL at Rec = 15.1 × 106
Figure 25. Statistical analysis of Cases 2a and b (includes brackets), α = 7◦ on medium grids (yellow circle at right represents experiment).
33 of 37
(a) CL at Rec = 1.35 × 106 (b) CL at Rec = 15.1 × 106
Figure 26. Statistical analysis of Cases 2a and b (includes brackets), α = 16◦ on medium grids (yellow circle at right represents experi-
ment).
34 of 37
(a) CL at Rec = 1.35 × 106 (b) CL at Rec = 15.1 × 106
Figure 27. Statistical analysis of Cases 2a and b (includes brackets), α = 20◦ on medium grids (yellow circle at right represents experi-
ment).
35 of 37
(a) CL at Rec = 1.35 × 106 (b) CL at Rec = 15.1 × 106
Figure 28. Statistical analysis of Cases 2a and b (includes brackets), α = 21◦ on medium grids (yellow circle at right represents experi-
ment).
36 of 37
(a) Lift coefficient (b) Drag coefficient
Figure 29. Deltas due to Reynolds number for Cases 2a and b (includes brackets), comparing median CFD results with experiment.
37 of 37