Correlating CFD Simulation With Wind Tunnel Test

for the Full-Scale UH-60A Airloads Rotor

Ethan Romander Thomas R. Norman I-Chung Chang
[email protected] [email protected] [email protected]
Flight Vehicle Research and Technology Division
NASA Ames Research Center
Moffett Field, CA

Abstract
Data from the recent UH-60A Airloads Test in the National Full-Scale Aerodynamics
Complex 40- by 80- Foot Wind Tunnel at NASA Ames Research Center are presented and
compared to predictions computed by a loosely coupled Computational Fluid Dynamics
(CFD)/Comprehensive analysis. Primary calculations model the rotor in free-air, but initial
calculations are presented including a model of the tunnel test section. The conditions stud-
ied include a speed sweep at constant lift up to an advance ratio of 0.4 and a thrust sweep
at constant speed into deep stall. Predictions show reasonable agreement with measurement
for integrated performance indicators such as power and propulsive but occasionally deviate
significantly. Detailed analysis of sectional airloads reveals good correlation in overall trends
for normal force and pitching moment but pitching moment mean often differs. Chord force is
frequently plagued by mean shifts and an overprediction of drag on the advancing side. Loca-
tions of significant aerodynamic phenomena are predicted accurately although the magnitude
of individual events is often missed.

Notation M2 cc sectional chord force coefficient,

d f c /dr
1/2ρa2∞ c
d f n /dr
M2 cn sectional normal force coefficient, 1/2ρa 2 c
a∞ freestream speed of sound ∞
dm/dr
c local chord length M2 cm sectional pitching coefficient, 1/2ρa2 c2
∞
fc force parallel to local chord line Mtip tip Mach number
fn force perpendicular to local chord line L rotor lift
m moment about local quarter chord N number of blades, 4
A total rotor disk area, πR2 =2262 ft2 P total rotor power
CL rotor lift coefficient, L
2
R rotor radius, 26.83 ft
ρ(ΩR) A
P T rotor thrust
CP rotor power coefficient,
ρ(ΩR)3 A V∞ freestream velocity
CPi induced power coefficient X rotor propulsive force
CPo profile power coefficient αc wall corrected shaft angle
CT rotor thrust coefficient, T αs geometric shaft angle
ρ(ΩR)2 A V∞
X μ advance ratio, ΩR
CX rotor propulsive force coefficient,
ρ(ΩR)2 A ψ rotor azimuth, deg
Presented at the American Helicopter Society 67th Annual Fo- ρ freestream density
·c
σ πR = 0.0826
rotor solidity, N
rum; Virginia Beach, VA; May 3–5, 2011. This is a work of the
U.S. Government and is not subject to copyright protection in Ω rotor angular velocity
the United States.

1
 at matching conditions from the previous full-scale
flight test and small-scale DNW wind tunnel test to
assess rotor and wind tunnel scaling issues. Finally,
unique slowed-rotor simulations were performed
at reduced RPM (40% and 65%), achieving advance
ratios up to 1.0. A complete enumeration of data
acquired during the test is available in Ref. 1.
Unfortunately, not all test data is perfect and no
simulation is exact. The goal of this work is to pro-
vide an initial correlation between measured data
and a state of the art simulation. This correlation
is intended to help discover flaws in experimental
technique while at the same time identifying op-
portunities to enhance rotorcraft simulation tech-
Figure 1 – UH-60A Airloads Rotor in 40- by 80-
nology.
Foot Wind Tunnel.

Introduction Methodology
The complexity of rotorcraft aeromechanics is
Testing was successfully completed in May
not easily modeled. The analytical results pre-
2010 on a full-scale UH-60A rotor system in the
sented herein were obtained using two separate
USAF’s National Full-Scale Aerodynamics Com-
codes—each a specialist in a particular aspect of
plex (NFAC) 40- by 80-Foot Wind Tunnel [1].
rotorcraft simulation—loosely joined for this pur-
The primary objective of this NASA/Army test
pose. This section will describe the two codes and
program was to acquire a comprehensive set of
how they work together to maximize simulation
validation-quality measurements on a full-scale
accuracy and efficiency.
pressure-instrumented rotor system at conditions
that challenge the most sophisticated modeling
and simulation tools. A secondary objective was CAMRAD II
to meet one of the original goals of the UH-60A
CAMRAD II belongs to a family of software
Airloads program: to provide data to evaluate the
known as “Comprehensive Codes” for the analysis
similarity, or lack thereof, of measurements be-
of rotorcraft. These analyses incorporate a myriad
tween small-scale wind tunnel [2], full-scale wind
of models to simulate the different aeromechani-
tunnel (current test), and full-scale flight test [3].
cal subsystems of rotorcraft. CAMRAD II brings
The test hardware included the same rotor
together a multibody dynamics model, a nonlin-
blades used during the flight test. Figure 1 shows
ear finite elements structural model, and an aero-
these blades installed on the NFAC Large Rotor
dynamics model based on lifting line theory [4].
Test Apparatus (LRTA) in the wind tunnel test
CAMRAD II has seen a great deal of use in the
section. Key measurements included rotor per-
simulation of the UH-60 aircraft in a variety of
formance, blade loads, blade pressures, blade dis-
flight conditions [5–9].
placements, and rotor wake measurements using
The CAMRAD II structural dynamics model for
large-field Particle Image Velocimetry (PIV) and
the UH-60A has been decades in development by
Retro-reflective Background Oriented Schlieren
NASA and the U.S. Army. The specific model used
(RBOS).
here was refined by Yeo et al. in 2004 [10]. The
Data were acquired over a wide range of test
model simulated the rotor using 7 1-D structural
conditions, including speed sweeps at 1-g sim-
beam elements and 20 aerodynamic panels. The ro-
ulated flight conditions and parametric thrust
tor was trimmed using Newton-Raphson iteration
sweeps (up to and including stall) at various com-
on collective and cyclic to meet specified trim tar-
binations of shaft angles and forward speed. These
gets.
conditions included airspeeds up to 175 kt and
thrusts up to 32,000 lb. Data were also acquired

2
OVERFLOW 2 illustrated in Fig. 2. The baseline near-body grid
system (the hub and all four blades) contained
All Navier-Stokes CFD analysis presented herein
approximately 10.7 million points.
was performed using OVERFLOW 2 version 2.2b
Two different off-body grid systems provided
[11]. OVERFLOW 2 is an overset, structured-mesh
two different simulation scenarios:
flow solver developed at NASA. For two decades
The first system used a series of ever larger
the OVERFLOW solver has served to analyze a
Cartesian grids to create shells expanding outward
variety of rotorcraft under a wide range of flight
from the near-body grid set. The grid point spacing
conditions [12]. OVERFLOW 2 offers a wide vari-
within each shell is twice that of the shell imme-
ety of numerical schemes, turbulence models, and
diately preceding it. The finest off-body grid had
boundary conditions. For the present study, OVER-
a spacing equivalent to 0.1 tip-chord lengths in
FLOW 2 was run with 4th order central differencing
all three directions. Seven such shells created a
and 4th order artificial dissipation in space. Time
cubic computational domain spanning ten rotor
marching was performed using a 2nd order dual
radii in every direction. Domain boundaries in
timestepping scheme. Turbulence was modeled
this scenario were set to a freestream characteristic
near blade surfaces using the Spalart-Almaras one-
condition thereby simulating a rotor operating in
equation model with rotational corrections. The
free air. Flow in the off-body grids was treated as
turbulence model was deactivated in regions one
inviscid and the turbulence model was deactivated.
chord length or further from the rotor blades to
This off-body grid set consisted of 15.5 million grid
reduce numerical dissipation of the wake. Blade
points and is depicted in Fig. 3.
surfaces were modeled as viscous, adiabatic walls;
The second off-body grid system used a sin-
outer boundaries were modeled using a character-
gle grid to envelop the near-body grid-set forming
istic condition imposing freestream quantities.
a computational domain that mimicked the size
OVERFLOW 2 computes the flowfield by dis-
and shape of the wind tunnel test section. This
cretizing the Navier-Stokes equations on a series of
grid made no attempt to model the contraction or
overset, structured grids. Grids modeling the ro-
expansion sections of the tunnel but rather main-
tor blades were body-fitted and curvilinear. These
tained the test section profile for 4.6 rotor radii
grids, often called near-body grids, extended ap-
upstream and downstream. The rotor plane was
proximately one chord length from the blade sur-
located 20.4 feet above the simulated tunnel floor.
face. The near-body grids were nested within one
The inlet end of the tunnel grid employed the
or more grids, called off-body grids, which filled
same freestream characteristic boundary condition
the space between the rotor and the boundary of
as the free air system, and the outlet plane used
the computational domain. The OVERFLOW 2
a boundary condition that ensured conservation
model included a notional hub, but the LRTA and
of mass within the tunnel grid. Viscous terms of
wind tunnel struts were not modeled. All grids
the Navier-Stokes equations were discarded for
exchanged flow information in regions of overlap.
this grid. This paradigm was initially proposed by
The amount of this overlap was sufficient to sup-
Chang et al. [16] This off-body grid contained 4.2
port full 4th order accuracy at the boundaries.
million grid points and is pictured in Fig. 4.
The grid surface is the latest definition derived
In addition, a refined version of the free-air grid
from the as-built CAD model. Notable differences
system was prepared to ascertain grid convergence
between this surface definition and that of prior in-
of the simulation. This grid system reduced grid
vestigations [14,15] are an outboard shift of the trim
point spacing in both the near-body and off-body,
tab by approximately 4 inches, a small bump on the
approximately doubling the number of grid points
upper surface near the blade grip, a slight thinning
to 50.7 million.
of the airfoil very near the tip, and a blunt trailing
Unless otherwise indicated, all predicted data
edge across the whole span.
presented herein were computed using the baseline
The near-body grid representing the bulk of
near-body grid and the baseline free-air off-body
each rotor blade had an “O” topology with 157
grid set. This near-body and off-body combination
points wrapping around the blade chordwise, 145
totals 26.2 million grid points. To reduce computa-
points along the blade span, and 75 points nor-
tion time CFD simulations were run using a hybrid
mal to the surface. The initial spacing at the blade
distributed/shared memory scheme with 160 Mes-
surface had a y+ value of 1. This grid system is

3
 Figure 2 – OVERFLOW 2 Near-Body Grid System. (Not every point shown.)

Figure 3 – OVERFLOW 2 Free-Air Off-Body Grid System.

Figure 4 – OVERFLOW 2 Wind Tunnel Off-Body Grid System. (Not every point shown.)

4
sage Passing Interface (distributed memory) ranks analysis, this generally occurred after 24 coupling
and eight OpenMP threads (shared memory) per iterations. Since OVERFLOW 2 was allowed to it-
MPI rank for a total of 1280 parallel tasks. OVER- erate for ¼ revolution between coupling exchanges,
FLOW 2 required approximately 31 minutes to ad- this equates to 6 full revolutions for the converged
vance the solution for this configuration ¼ of a ro- solution. A fully converged coupled solution re-
tor revolution using 1280 CPUs of an SGI Altix ICE quired approximately 17 hours to compute for the
computer. baseline grid on 1280 SGI Altix ICE processors.

Coupling Methodology Selected Test Conditions And Trim Ap-

CAMRAD II uses a lower-fidelity aerodynamics proach
model than that available in modern CFD codes, For the present investigation, two parametric
and most CFD codes lack the sophisticated Com- sweeps were selected for analysis. First, a speed
putational Structural Dynamics (CSD) and trim ca- sweep was selected to test simulation accuracy over
pabilities of comprehensive codes like CAMRAD II. a wide range of advance ratios. Second, a thrust
Coupling a CFD code (e.g. OVERFLOW 2) to a sweep was selected to test simulation accuracy un-
comprehensive code (e.g. CAMRAD II) marries the der conditions ranging from a lightly loaded rotor
strengths of the two approaches and produces the through deep stall.
highest-fidelity solution currently possible. For this The selected speed sweep ranged from μ = 0.15
study, coupling is achieved by alternate execution to 0.4 with a constant CL /σ=0.09 and a constant tip
of OVERFLOW 2 and CAMRAD II. At the end of Mach number of 0.65. During testing, lift, propul-
each code’s turn to execute, it passes data to the sive force, and hub moments were trimmed to
next code. The data passed from OVERFLOW 2 match nominal values at each flight condition by
to CAMRAD II is airload data integrated from its varying αs along with collective and cyclic pitch.
Navier-Stokes model of the UH-60 rotor. This air- In CFD calculations it is difficult to change the
load data is used to augment CAMRAD II’s inter- shaft angle once it has been set, which compli-
nal aerodynamics model (which consists of airfoil cates its use as a trim control in adjusting lift and
tables and a lower-order wake model). At the end propulsive force. Instead, these flight conditions
of its execution, CAMRAD II generates updated were matched in simulation by setting a constant
control positions and a description of how the blade αs as indicated by tunnel data while trimming CT
deforms elastically as it revolves around the shaft. and hub moments to match tunnel values using
These quantities are used to give OVERFLOW 2’s collective and cyclic pitch.
grids a realistic motion in response to the aero- The selected thrust sweep was conducted at
dynamic environment. This algorithm, called the μ =0.3 with Mtip = 0.625, αs = 0◦ , and the rotor
delta coupling technique, was pioneered by Tung et trimmed for minimum hub moments. The thrust
al. [13] and implemented in OVERFLOW by Pots- was varied by changes in collective pitch up to
dam et al. [14] Significantly improved airloads pre- stall. Absolute collective angles for a given thrust
diction capability has been demonstrated for the are generally different between simulation and test.
UH-60A rotor in steady level flight conditions us- To nullify this offset, a baseline case was selected
ing this loosely coupled approach [14, 15]. at CT /σ = 0.08 and the simulation was trimmed to
The CFD solution is advanced ¼ revolution dur- match the measured thrust. Collective deltas were
ing a coupling iteration because this allows each derived from test data relative to this baseline and
of the rotor’s four blades to sweep through a full then applied to the simulated baseline to produce
quadrant of the rotor disk. Taken in aggregate, the the remaining target points.
four blades thereby determine the airloads at every In all cases, free-air simulations used corrected
azimuth for every coupling iteration. shaft angles, αc , derived from the geometric shaft
Convergence of the coupling process was deter- angle, αs , by applying a Prandtl-Glauert wall cor-
mined by monitoring blade airloads for periodic- rection [1]. Simulations including tunnel walls used
ity. When the airloads did not vary significantly uncorrected shaft angle data.
from one coupling iteration to the next, the solu-
tion was judged to be converged. For the present

5
 Results
  
The 2010 UH-60A Airloads Test is a valuable
   !"#$
resource not only for the breadth of flight condi-

tions obtained but for the depth of detail available


in each measurement. Failure of analysis to pre-
dict bulk loads accurately can be diagnosed by ex-
ploring detailed measurements and how they com-


pare to quantities extracted from simulation results.
This approach will now be applied to the present

 
simulation methodology first for the selected speed


sweep and then for the selected thrust sweep.

Speed Sweep



Integrated Results
Figure 5 plots the value of the trimmed thrust


coefficient versus advance ratio for the range of







speeds simulated. Since thrust was a trim target, 


it was expected that the computed points would
lie very close to the measured data. Indeed this Figure 5 – Thrust Trim for Speed Sweep.
is the case for the data derived from CAMRAD II
output. However, the data computed by integrat-
dynamics, only its data will be presented in the
ing the pressure and viscous forces acting on the
remainder of this paper. The reader should bear in
CFD grid yield a somewhat higher thrust. The
mind that all results presented here are the result
average difference between the thrust reported by
of a trim solution with a slightly higher (∼2.5%)
CAMRAD II and OVERFLOW 2 is 2.5%. There are
than prescribed thrust.
a number of factors contributing to this difference.
Recall that the CFD calculations for the speed
First, small discrepancies in the local twist between
sweep were performed by matching the wall cor-
the OVERFLOW 2 model and the CAMRAD II
rected shaft angles from the test data rather than
model can alter the computed thrust by altering
trimming propulsive force. Figure 6a shows the
in-plane and out-of-plane contributions by nor-
difference between corrected and uncorrected shaft
mal and chord force. Second, the airloads passed
angle as a function of advance ratio. Figure 6b
from OVERFLOW 2 to CAMRAD II are only the
depicts the corresponding propulsive force at each
loads integrated from the main blade grid and ne-
speed. Propulsive force is underpredicted at low
glect any forces contributed by the tip and root
speed and overpredicted at high speed. A portion
caps. The coupling mechanism makes no effort to
of this prediction error at high speed is due to in-
fully conserve forces between OVERFLOW 2 and
accurate trim. The large shaft angles at high speed
CAMRAD II. Third, the CAMRAD II model used a
divert a substantial portion of the excess thrust due
low-resolution—just 20 aerodynamic panels span-
to trim error into propulsive force.
wise and 24 azimuthal stations—representation of
Figure 7 compares measured and predicted
the rotor disk. This low-fidelity representation
power. Figure 7a shows total power which is
doubtlessly fails to capture all of the nuances of the
underpredicted by approximately 2%. Figure 7b
spanwise loading on the blade. CAMRAD II is ca-
is formed by removing the parasite power from
pable of including many more aerodynamic panels
Fig. 7a leaving the sum of profile and induced
in its blade model, and doing so should henceforth
power, the two of which are inseparable in CFD
be considered standard practice. Improving force
simulation. From Fig. 7b it is clear that the sim-
conservation in the coupling process is an area that
ulation is having difficulty predicting profile and
requires future work.
induced power accurately at the high speed points.
Since the OVERFLOW 2 model contains the
This is masked in Fig. 7a by a fortuitous overpredic-
most complete representation of the rotor aero-

6
      
     





































      





 

 

   
(a) (b)

Figure 6 – Prescribed Shaft Angle and Resultant Propulsive Force for Speed Sweep.


  
  
















  







 






























 

 

 


 

 





 
 

(a) (b)

Figure 7 – Speed Sweep Power.

7
tion of propulsive force, and hence parasite power, mean shifts at r/R = 0.4 and 0.865. In addition, a
at these points. A more accurate trim would result small surge in normal force from ψ =120◦ to 180◦
in a small thrust reduction, causing a complemen- is present in the test data for all of the outboard
tary reduction in induced power, further degrading stations, but the analysis prefers a nearly continu-
the correlation throughout Fig. 7 slightly. ous decrease in normal force throughout the second
Also shown in Figs. 6 and 7 are predictions us- quadrant.
ing the simulated tunnel walls for a single condi- The pitching moment is generally mean shifted
tion (μ =0.15). The Cx data from this model has with a tendency toward overprediction. It should
been corrected using the same Prandtl-Glauert wall be noted that the pitching moment measured dur-
correction used for the test data and is included in ing the test is very sensitive to transducer perfor-
Fig. 6b. Predictions with the tunnel walls underpre- mance at the trailing edge, so a mean shift is likely
dict propulsive force and overpredict power at this to include significant measurement error. One com-
advance ratio to a greater extent than free air sim- mon exception to the trend of overpredicted mean
ulations. One possibility is that the inviscid walls occurs at r/R = 0.775. A possible cause is the trim
of the simulation necessitate a different wall correc- tab which extends from r/R =0.73 to 0.86. The trim
tion than the test measured data. tab on the tested blades was deflected upward by
approximately 2◦ which increases the local pitch-
Sectional Comparisons ing moment significantly compared to adjacent
stations. Although the trim tab is represented in
In Fig. 8 are presented the sectional loads— the CFD grid, it is not deflected, and the predicted
M2 cn , M2 cm , M2 cc —at several radial stations for the pitching moment mean changes little compared to
lowest speed in the speed sweep, μ = 0.15. All data r/ = 0.4 or 0.865. The moment pulses due to BVI
R
are presented with means included. This is at a are predicted accurately in phase but the peak-to-
slightly higher thrust than, but otherwise is similar peak amplitude is generally smaller compared to
to, the oft-studied c8513 flight condition from the measured data.
UH-60 flight test [3, 14]. Chord force displays a significant mean shift
Several predicted curves are plotted against the with a tendency toward underprediction. There is
test data. The first, in green, was computed us- an additional offset that further reduces the chord
ing the 26.2 million point baseline grid. The sec- force only on the advancing side. The sign conven-
ond was computed with the 50.7 million point fine tion for chord force is positive toward the leading
grid, in red. Although there are instances where edge so an underprediction represents an excess
the fine grid improves on the baseline solution (e.g. of drag. One contributor to this disparity between
M2 cc on the advancing side at r/R = 0.865), there are measurement and prediction is viscous force. The
also instances where the solution on the baseline test data is derived from integrated pressure mea-
grid is closer to the measured data (e.g. M2 cm in surements with no correction for viscous effects,
the second quadrant at r/R =0.92). Otherwise, these while the predicted data is the combination of inte-
two curves are very much in agreement with each grated pressure data and viscous drag at the airfoil
other. Because the baseline grid performed so well surface. Predicted airloads computed from pres-
for this low-speed flight condition—where accurate sure data alone are also presented in Fig. 8. Note
capture of the wake is most important—it was de- that the viscous forces make significant contribu-
cided to use the baseline grid for the remainder of tions solely to chord force, and that they make the
this study. largest contribution on the advancing side near the
In general, the trends in normal force are well tip. However, the viscous forces account for at best
matched. The normal force pulses due to blade vor- 20–30% of the difference between measurement
tex interaction (BVI) begin to appear at r/R =0.775 and prediction. Despite the mean shift, the trend
on both the advancing side (near ψ =70◦ ) and re- in chord force is well represented. Drag excursions
treating side (near ψ =280◦ ). These BVI events are coincident with the BVI events are captured by
visible in the CFD wake visualization of Fig. 9. The analysis, particularly on the retreating side where
magnitudes of the normal force pulses in Fig. 8 are the measured drag pulse is closely matched in
accurately captured by the simulation save for a magnitude, phase, and shape.
small underprediction on the advancing side BVI Figure 10 contains the same measured and
at r/R =0.865. The largest missed predictions are

8
 Prediction Prediction Prediction
Experiment
r/ = 0.4 (Baseline Grid) (Fine Grid) (Baseline, No Viscous)
R
r/ = 0.775
R
r/ = 0.865
R
r/ = 0.92
R

Figure 8 – Sectional Airloads at μ =0.15, CT /σ = 0.093.

9
 ment pulse due to supercritical flow appears on the
advancing side at ψ = 90◦ , but its magnitude is di-
minished in prediction. For both of the outboard
stations, the analysis underpredicts the magnitude
of the moment on the advancing side. Chord force
results are well correlated at the inboard station,
r/ = 0.4. At r/ = 0.775 the chord force trend on the
R R
advancing side is completely missed, but this may
be an effect of the difference in trim tab deflec-
tion. At the two outboard stations, chord force re-
mains underpredicted on the advancing side, but
the shape of the predicted chord force curve fol-
lows that of the test for the first two quadrants,
even replicating the two small peaks in the second
quadrant. Differences in the fourth quadrant are
consistent with excessive stall in the simulation.
Figure 12 presents a side-by-side comparison
of test and predicted normal force coefficient at
r/ = 0.92 for every speed in the sweep for which a
R
Figure 9 – Wake Visualization From CFD. Isosur- simulation was run. The test data exhibits vortex
face of Q-Criterion With Blades Colored induced loading at ψ = 70◦ and 280◦ for the lowest
By Pressure. advance ratios. At higher advance ratios, the in-
creasingly negative loading in the second quadrant
free-air predicted data from Fig. 8 but compares balances increased normal force over the rest of the
it to data predicted using the tunnel wall model. rotor disk. The second quadrant minimum starts at
The largest differences between the free-air model approximately ψ =135◦ and draws toward ψ =90◦
and the tunnel model appear on the advancing as advance ratio increases.
side, particularly near the ψ =90◦ BVI. This event Figure 13 makes the same comparison as Fig. 12
appears slightly attenuated compared to the free- but for moment coefficient. In the test data, vortex
air model in all three coefficients and at all of the effects are again evident at the lowest advance ra-
stations where it is to be observed. tios near ψ =70◦ and 280◦ . A large moment pulse
The low speed case is when the tunnel walls ex- consistent with supercritical flow develops near
ert their maximal effect on rotor performance. Yet ψ = 80◦ for the highest advance ratios. The moment
the free-air simulation compares well to the cor- minimum in the second quadrant migrates from
rected test and tunnel simulation data for both in- ψ = 135◦ at μ = 0.15, toward ψ = 90◦ at μ = 0.3, and
tegrated and sectional loads. This suggests that the then back to 135◦ by μ = 0.4. Lastly, the test data
simple Prandtl-Glauert correction for angle of at- appears to exhibit a weak stall event near ψ = 315◦
tack is reasonable. at μ =0.4.
Figure 11 presents the sectional airloads at the Observing the evolution of the advancing side
highest advance ratio in the speed sweep, μ =0.4. normal force trough in Fig. 12 reveals that both the
At the two outboard stations the analysis predicts test and analysis display the same monotonic re-
a dynamic stall cycle in the fourth quadrant. This duction in normal force with speed. The change in
is identifiable as dips in all three coefficients just minimum normal force from one speed to the next
after ψ =270◦ . Although there is some evidence of appears slightly larger for the prediction than in
stall in the test data (particularly in chord force), test. The simulation also predicts significant lift loss
the magnitude of the stall event was predicted to from stall in the fourth quadrant at the two highest
be much larger than was measured. The normal speeds. No corresponding feature is present in the
force trend is well represented, except for a ten- measured data. Despite the differences in moment
dency to overpredict the negative lift on the ad- coefficient magnitude between test and prediction
vancing side. Pitching moment predictions show in Fig. 13, the trend between speeds is very sim-
larger differences. From r/R = 0.775 outboard, a mo- ilar. The advancing side shock grows in strength

10
 Experiment Prediction (Free Air) Prediction (Tunnel)
r/ = 0.4
R
r/ = 0.775
R
r/ = 0.865
R
r/ = 0.92
R

Figure 10 – Sectional Airloads at μ =0.15, CT /σ = 0.093. Free Air Vs. Tunnel Model.

11
 Experiment Prediction
r/ = 0.4
R
r/ = 0.775
R
r/ = 0.865
R
r/ = 0.92
R

Figure 11 – Sectional Airloads at μ = 0.4, CT /σ = 0.093.

12
  




























Figure 12 – Normal Force Coefficient at r/R =0.92 for μ = 0.15, 0.2, 0.3, 0.37, 0.4. CT /σ =0.093.

 




































Figure 13 – Moment Coefficient at r/R = 0.92 for μ =0.15, 0.2, 0.3, 0.37, 0.4. CT /σ = 0.093.

13
and migrates toward ψ =90◦ . The moment trough and ends in an interesting double pulse at r/R = 0.92.
behind the shock also evolves very similarly across Although the phase of these pulses is well corre-
the speed range. Differences include an expansion lated, the peak-to-peak magnitude of the predicted
of the trough into the third quadrant for the pre- pulses is smaller. Moment recovery following the
dicted data, the prediction of a stall cycle at μ = 0.37 shocks also begins too early and progresses slowly
when none is present in measured data, and an compared to measurement for r/R = 0.865 and 0.92.
enhancement of the predicted stall cycle at μ =0.4 Chord force continues to be mean shifted outboard
compared to measurement. of r/R = 0.775, and the trend is completely mispre-
dicted at r/R = 0.775.
Thrust Sweep The sectional airloads presented in Fig. 16
are for the maximum collective condition where
Integrated Results CT /σ =0.1255. Comparisons at this condition must
be tempered with the knowledge that stall is by
Presented in Fig. 14 are the predicted perfor-
nature a chaotic phenomena and there exists con-
mance results for the selected thrust sweep. From
siderable variation in loading from one revolution
this figure it is clear that the analysis performs well
to another, even in simulation. Test data in Fig. 16
at predicting the maximum rotor thrust and the re-
are average loads sampled from 128 revolutions,
lationship between power and thrust throughout
but computing an ensemble average from CFD is
the sweep. Recall that the analysis was trimmed
impractical. The predicted data provided in Fig. 16
to match thrust at the CT /σ =0.08 point but that
are instead single-revolution snapshots.
in general CFD integrated thrust is slightly higher
The flow at the inboard station remains rela-
than specified. This leads to a small thrust surplus
tively benign and correlation is good. The most
throughout the linear region since all collective an-
significant deviations are a familiar mean shift in
gles are set relative to the baseline condition. At
pitching moment and a blunting of the chord force
high collectives the excess thrust vanishes because
peak in the second quadrant. Outboard, the load-
the rotor is stalled and appears incapable of pro-
ing is dominated by two severe stall cycles clearly
ducing more thrust. The analysis also exhibits re-
indicated near the forth quadrant. The analysis ap-
duced sensitivity to collective change evidenced by
pears to phase shift the stall cycles slightly, pre-
an approximately 7% reduction of slope in the lin-
dicting both events perhaps 15◦ earlier than mea-
ear region.
sured in test. Despite the phase shift, the magni-
Power is generally overpredicted, with differ-
tudes of the stall events are well predicted. One
ences up to 11% at the highest thrust conditions
glaring difference at this deeply stalled condition
(Fig. 14b). The fact that the predicted and measured
is a large first-quadrant excursion in all three coef-
points lie on very similar curves, however, suggests
ficients for the prediction. The similarity between
that if thrust was rigorously matched, power may
the frequency of this event and the torsional mode
be better predicted.
of the blade suggests that it is a torsional response
to the fourth quadrant stall cycles.
Sectional Comparisons Figure 17 compares sectional normal force at
r/ = 0.92 for all collective settings simulated. The
Sectional airloads for the minimum collective R

setting are presented in Fig. 15. At this collective, experiment observed an interesting evolution of
the measured thrust was CT /σ =0.02. The normal normal force in the second quadrant. The normal
force bucket in the second quadrant is narrower force increased with collective until CT /σ =0.12 af-
and shallower in prediction than in test. Normal ter which it began to decrease, forming a narrow
force is also slightly overpredicted for the majority trough at approximately ψ = 120◦ . The normal force
of the retreating side, except at the inboard station. reduction beginning at CT /σ =0.12 is necessary to
Moment coefficient mean is well predicted at this maintain roll balance across the rotor. Also visi-
condition—except at r/R =0.4—but there exist sig- ble in the test data is the development of two stall
nificant differences in trends between analysis and cycles, one at ψ =290◦ and the second at 340◦ .
test. For example, there is a shock-induced mo- Figure 18 presents moment coefficient at r/R =0.92
ment pulse that appears in the second quadrant for all collective settings simulated. The most
at r/R =0.775, migrates toward the first quadrant, prominent features in the test data are the two

14





















 
















   

  
   












 
  
(a) (b)

Figure 14 – Integrated Performance Results for Collective Sweep. μ = 0.3, Mtip = 0.625, αs = 0◦ .

moment stall events that develop in the fourth in magnitude and ultimately persists through all
quadrant. Both stall cycles appear abruptly at four quadrants at CT /σ =0.1255. Another differ-
CT /σ = 0.12. The first stall cycle quickly achieves ence is the second quadrant negative loading in
a stable minimum moment which slowly drifts Fig. 17 which is distorted in the predicted data.
toward ψ =270◦ . This is in contrast to the second The trough is diminished in magnitude and shifted
stall cycle where the moment decreases rapidly and toward the end of the second quadrant. This differ-
monotonically but remains fixed in azimuth. ence appears to be a consequence of the torsional
Changes in normal force from one collective oscillation persisting into the second quadrant.
setting to the next are similar between test and The effects of the torsional oscillation are also
prediction, with only a few notable differences. readily visible in the predicted data for Fig. 18.
Foremost are the advancing side oscillations in the The oscillation dominates the predicted data in
predicted data. This oscillation begins abruptly the first quadrant for the high thrust cases. The
at CT /σ =0.120 and increases in magnitude along oscillation may also be responsible for the missed
with collective. Examining the predicted moment prediction of the moment minimum appearing in
coefficient data in Fig. 18 reveals that there is a sin- the test data at ψ =135◦ . Examine the stall cycles
gle weak stall event at CT /σ = 0.10 which quickly in the fourth quadrant. The predicted responses
strengthens and spawns a second stall event by are more chaotic and appear to have higher fre-
CT /σ = 0.120. This suggests that the advancing quency content than the measured data. The first
side oscillation present in the predicted data is a cycle marches toward 270◦ in an orderly fashion
response to either the strengthening of the first according to test data, but the analysis predicts a
stall event, the formation of the second stall event, wider azimuthal range for the minima and high-
or both. Figure 19 plots the local pitch angle at frequency oscillations make it difficult to discern
r/ =0.92 for three of the highest collective settings. any orderly progression. The final location of the
R
At CT /σ =0.10 the rotor has a single stall cycle and first stall cycle—at CT /σ = 0.1255—is very similar
very little torsional oscillation is visible in Fig. 19. between test and prediction. Comparisons between
As thrust increases, the fourth quadrant stall cy- test and prediction for the second stall cycle in-
cles strengthen concurrently with a 5/rev torsional clude many of the same observations. The high
oscillation in Fig. 19. This 5/rev oscillation grows frequency content of the predicted data make it

15
 Experiment Prediction
r/ = 0.4
R
r/ = 0.775
R
r/ = 0.865
R
r/ = 0.92
R

Figure 15 – Sectional Airloads, CT /σ =0.02.

16
 Experiment Prediction
r/ = 0.4
R
r/ = 0.775
R
r/ = 0.865
R
r/ = 0.92
R

Figure 16 – Sectional Airloads, CT /σ =0.1255.

17
  

































Figure 17 – Normal Force Coefficient at r/R =0.92 for CT /σ =0.02, 0.04, 0.08, 0.10, 0.120, 0.123, 0.1247, 0.1253,
0.1255. (Measured values, predicted CT /σ is slightly higher.)

 
































Figure 18 – Moment Coefficient at r/R = 0.92 for CT /σ =0.02, 0.04, 0.08, 0.10, 0.120, 0.123, 0.1247, 0.1253,
0.1255. (Measured values, predicted CT /σ is slightly higher.)

18
 10 resentation of the aerodynamics, only loads from
CT/σ=0.1003 CFD have been used for the present work. Further
CT/σ=0.1203 work is required to improve the coupling mecha-
CT/σ=0.1255 nism and CAMRAD II model to minimize discrep-
5 ancies in loading between the two codes.
Local Pitch Angle, deg
0,1/rev Removed

For the speed sweep, the analysis displayed a

tendency to underpredict propulsive force at low
speed and overpredict it at high speed. This may
0 be partially attributable to inaccurate trim. The
analysis underpredicted total power by approxi-
mately 2%. Although total power appears to be
consistently predicted across the speed range, at
-5 high speed it is just a fortunate result of com-
pensating errors—overpredicted parasite power on
one side with underpredicted profile and induced
power on the other. When applied to the thrust
-10 sweep, the analysis was very accurate at predicting
0 90 180 270 360
Azimuth maximum thrust and at capturing the relationship
between thrust and power. Before the rotor stalled,
Figure 19 – Local Pitch Angle at r/R = 0.92, 2/rev and there was a small increase in thrust attributable to
Above. inaccurate trim in the simulation at the baseline
condition. The predicted thrust also exhibited a
reduced sensitivity to collective pitch resulting in a
difficult to accurately locate the minimum, but the
shallowing of the slope in the thrust vs. collective
majority seem clustered near 320◦ . The test shows
curve. Power was consistently overpredicted—as
a slightly later location for the second stall cycle,
much as 11% at the peak of stall. However, the vari-
around 330◦ .
ation of power with thrust is very similar between
simulation and test. This suggests that power may
Conclusions be better predicted if the simulation trim were more
accurate.
This paper has described the initial effort to Detailed sectional loading comparisons reveal a
correlate prediction with wind tunnel measure- number of differences between measurement and
ments from the recently completed UH-60A Air- prediction. In general, correlation of normal force
loads wind tunnel test. Differences identified in was good, with the prediction only occasionally de-
the present work represent opportunities to im- viating from test measurement. Comparisons of
prove prediction methodologies or identify defects pitching moment exhibited more variation. Pitch-
in measured data. ing moment predictions were frequently offset from
The simulation methodology involved a com- measurement. It was noted that these comparisons
prehensive rotorcraft simulation package, CAM- are hampered by the difficulty of accurately mea-
RAD II, augmented with the first-principles aero- suring the mean pitching moment in test. Notewor-
dynamics of a CFD simulation provided by OVER- thy flow features such as blade-vortex interaction,
FLOW 2. supercritical flow, and stall are accurately located
Two parametric sweeps from the wind tunnel in azimuth but the predicted magnitude of these
test were modeled: a speed sweep at constant lift events is often diminished. Chord force predictions
and a thrust sweep at constant speed. Simulation of bore the least resemblance to test data. These com-
both sweeps were complicated by a disagreement parisons were often characterized by mean shifts
between the two simulation codes on the integrated and a large overprediction of drag on the advanc-
trim targets. The largest effect of this disagreement ing side of the rotor. As much as 20–30% of this
is a discrepancy of 2.5% in the trimmed thrust be- difference is attributable to viscous forces included
tween the two codes, with the CFD being the higher in predictions but absent from test data. Correla-
of the two. Because they are a more complete rep- tion of all three sectional coefficients at r/R =0.775

19
may be improved by correctly modeling trim tab the effects of supercritical flow.
deflection.
Other observed trends included a tendency for
the analysis to overpredict the severity of stall, fre- References
quently predicting it at lower advance ratio and fur-
ther inboard on the blade than indicated by mea- [1] Norman, T.R., Shinoda, P., Peterson, R.L., and
surement. Individual stall events are not as well Datta, A., “Full-Scale Wind Tunnel Test of
defined in prediction as in test and their azimuthal the UH-60A Airloads Rotor”, American Heli-
locations do not progress smoothly as thrust in- copter Society 67th Annual Forum, May 2011.
creases. Also, for flows exhibiting a shock at or [2] Lorber, P.F., “Aerodynamic Results of a
near ψ =90◦ , the moment trough that formed in the Pressure-Instrumented Model Rotor Test at the
second quadrant just after the shock generally ex- DNW”, Journal of the American Helicopter Soci-
tended into the third quadrant according to simu- ety, Vol. 36, No. 4, October 1991.
lation, but the corresponding feature was confined
to the second quadrant in the test. [3] Kufeld, R.M., Balough, D.L., Cross, J.L., et al.,
One final observation was made only for the “Flight Testing of the UH-60A Airloads Air-
deeply stalled cases of the thrust sweep. The analy- craft”, American Helicopter Society 50th An-
sis predicted a large oscillation in all three sectional nual Forum, May 1994.
coefficients beginning in the first quadrant and per-
[4] Johnson, W., “Technology Drivers in the De-
sisting into the second. No such excursion was
velopment of CAMRAD II”, American Heli-
present in the measured data. This has been shown
copter Society Aeromechanics Specialist Meet-
to be a torsional response of the blade model to the
ing, San Francisco, CA, January 19-21, 1994.
formation of stall cycles in the fourth quadrant.
Modeling wind tunnel walls in the CFD simu- [5] Yeo, H., Bousman, W.G., and Johnson, W.,
lation made only small differences in the sectional “Performance Analysis of a Utility Helicopter
loads at μ = 0.15. The largest differences were on the with Standard and Advanced Rotor”, Journal
advancing side near the ψ =90◦ BVI. Larger differ- of the American Helicopter Society, Vol. 49, No. 3,
ences were observed in the integrated performance July 2004, pp. 250–270.
parameters, with the free-air model outperforming
the tunnel model. Further work is required to de- [6] Shinoda, P.M., Yeo, H., and Norman, T.R., “Ro-
termine if this is a trend or merely an isolated mis- tor Performance of a UH-60 Rotor System in
prediction. the NASA Ames 80- by 120-Foot Wind Tun-
The goal of this work was to provide an ini- nel”, Journal of the American Helicopter Society,
tial correlation with the newest data from the U.S. Vol. 49, No. 4, October 2004.
Army/NASA UH-60 test program. This dataset is
[7] Yeo, H., and Johnson, W., “Assessment of
of tremendous value to the rotorcraft community
Comprehensive Analysis Calculation of Air-
for the breadth of flight conditions sampled and
loads on Helicopter Rotors”, Journal of Aircraft,
the variety of detailed measurements made at each.
Vol. 42, No. 5, September–October 2005.
No significant anomalies were identified in the test
data studied, but the correlation effort revealed a [8] Yeo, H., and Johnson, W., “Prediction of Rotor
number of areas for improvement in the simulation Structural Loads with Comprehensive Analy-
technique. Development is required to improve sis”, Journal of the American Helicopter Society,
the coupling mechanics between CAMRAD II and Vol. 53, No. 2, April 2008.
OVERFLOW 2. Better force transfer between the
two is mandatory for improving trim accuracy and [9] Yeo, H., Romander, E., and Norman, T., “In-
accurately predicting integrated loads. Accurate vestigation of Rotor Performance and Loads of
power prediction remains a challenge for analysis. a UH-60A Individual Blade Control System”,
More detailed investigation is necessary to identify American Helicopter Society 66th Annual Fo-
the cause of the stall-related torsional oscillation rum, Phoenix, AZ, May 11–13, 2010.
in the blade model and to improve the capture of [10] Yeo, H., Bousman, W.G., and Johnson, W.,
sectional loading details such as dynamic stall and “Performance Analysis of a Utility Helicopter

20
 with Standard and Advanced Rotor”, Journal
of the American Helicopter Society, Vol. 49, No. 3,
July 2004, pp. 250–270.

[11] Buning, P.G., Gomez, R.J., and Scallion,

W.I., “CFD Approaches for Simulation of
Wing-Body Stage Separation”, AIAA-2004-
4838, AIAA 22nd Applied Aerodynamics Con-
ference, Providence, RI, August 16–19,2004.

[12] Potsdam, M., Strawn, R.C., and Meakin,

R., “Dynamic Rotorcraft Applications Using
Overset Grids”, 31st European Rotorcraft Fo-
rum, Florence, Italy, September 13–15, 2005.

[13] Tung, C., Caradonna, F.X., and Johnson, W.,

“The Prediction of Transonic Flows on an Ad-
vancing Rotor”, American Helicopter Society
40th Annual Forum, Arlington, VA, May 16–
18, 1984.

[14] Potsdam, M., Yeo, H., and Johnson, W., “Ro-

tor Airloads Prediction Using Loose Aerody-
namic/Structural Coupling”, Journal of Air-
craft, Vol. 43, No. 3, May–June 2006.

[15] Nygaard, T., Saberi, H., Ormiston, R.A.,

Strawn, R.C., and Potsdam, M., “CFD and
CSD Coupling Algorithms and Fluid Struc-
ture Interface for Rotorcraft Aeromechanics
in Steady and Transient Flight Conditions”,
American Helicopter Society 62nd Annual Fo-
rum, Phoenix, AZ, May 9–11, 2006.

[16] Chang, I-C., et al., “Airloads Prediction of a

UH-60A Rotor Inside The 40- by 80-Foot Wind
Tunnel”, American Helicopter Society Special-
ists’ Conference on Aeromechanics, San Fran-
cisco, CA, January 20–22, 2010.

21