You are on page 1of 99

Progress in Aerospace Sciences 37 (2001) 669767

Effects of ice accretions on aircraft aerodynamics


Frank T. Lyncha,*, Abdollah Khodadoustb
a

Lynch Aerodyn Consulting, 5370 Via Maria, Yorba Linda, CA 92886, USA (Former MDC and Boeing Technical Fellow)
b
Principal Engineer/Scientist, The Boeing Company, Huntington Beach, CA, USA

Abstract
This article is a systematic and comprehensive review, correlation, and assessment of test results available in the
public domain which address the aerodynamic performance and control degradations caused by various types of ice
accretions on the lifting surfaces of xed wing aircraft. To help put the various test results in perspective, overviews are
provided rst of the important factors and limitations involved in computational and experimental icing simulation
techniques, as well as key aerodynamic testing simulation variables and governing ow physics issues. Following these
are the actual reviews, assessments, and correlations of a large number of experimental measurements of various forms
of mostly simulated in-ight and ground ice accretions, augmented where appropriate by similar measurements for
other analogous forms of surface contamination and/or disruptions. In-ight icing categories reviewed include the
initial and inter-cycle ice accretions inherent in the use of de-icing systems which are of particular concern because of
widespread misconceptions about the thickness of such accretions which can be allowed before any serious
consequences occur, and the runback/ridge ice accretions typically associated with larger-than-normal water droplet
encounters which are of major concern because of the possible potential for catastrophic reductions in aerodynamic
effectiveness. The other in-ight ice accretion category considered includes the more familiar large rime and glaze ice
accretions, including ice shapes with rather grotesque features, where the concern is that, in spite of all the research
conducted to date, the upper limit of penalties possible has probably not been dened. Lastly, the effects of various
possible ground frost/ice accretions are considered. The concern with some of these is that for some types of
congurations, all of the normally available operating margins to stall at takeoff may be erased if these accretions are
not adequately removed prior to takeoff. Throughout this review, important voids in the available database are
highlighted, as are instances where previous lessons learned have tended to be overlooked. r 2002 Elsevier Science Ltd.
All rights reserved.

Contents
1. Introduction . . . . . . . . . . . . . . . . . .
2. Icing simulation techniques . . . . . . . . . .
2.1. Simulation fundamentals . . . . . . . . .
2.2. Overall simulation effectiveness . . . . . .
2.3. Scaling techniques . . . . . . . . . . . .
3. Aerodynamic simulation considerations . . . .
3.1. Single-element lifting-surface performance
3.2. Multi-element lifting-surface performance

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

*Corresponding author. Tel.: +1-714-693-8797; fax: +1-714-779-3541.


E-mail address: aerofrank@aol.com (F.T. Lynch).
0376-0421/01/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 3 7 6 - 0 4 2 1 ( 0 1 ) 0 0 0 1 8 - 5

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

671
673
674
678
679
680
680
681

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

670

3.3. Control surface effectiveness . . . . . . . . .


3.4. Drag characteristics . . . . . . . . . . . . . .
4. Effect of initial in-ight leading-edge ice accretions
4.1. Maximum lift reductions . . . . . . . . . . .
4.2. Stall angle reductions . . . . . . . . . . . . .
4.3. Drag penalties . . . . . . . . . . . . . . . .
4.4. Trailing-edge control surface characteristics .
5. Effect of runback and ridge ice accretions . . .
5.1. Maximum lift reductions . . . . . . . . . . .
5.2. Stall angle reductions . . . . . . . . . . . . .
5.3. Drag penalties . . . . . . . . . . . . . . . .
6. Effect of large in-ight ice accretions . . . . . . .
6.1. Maximum lift reductions . . . . . . . . . . .
6.2. Stall angle reductions . . . . . . . . . . . . .
6.3. Drag penalties . . . . . . . . . . . . . . . .
6.4. Trailing-edge control surface characteristics .
7. Effect of ground frost/ice accretions . . . . . . . .
7.1. Maximum lift reductions . . . . . . . . . . .
7.2. Stall angle reductions . . . . . . . . . . . . .
7.3. Drag penalties . . . . . . . . . . . . . . . .
8. Summary and conclusions . . . . . . . . . . . . .
9. Recommendations . . . . . . . . . . . . . . . . .
Acknowledgements . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . .
Nomenclature
List of Symbols
2-D
two dimensional
3-D
three dimensional
c
chord
1C
degrees centigrade
cd
2-D section drag coefcient
clmax
2-D section maximum lift coefcient
1F
degrees fahrenheit
F
ap
h=c
non-dimensional protuberance height
k=c
non-dimensional roughness height
L
lower surface
Mo
freestream Mach number
S
slat
t=c
thickness ratio
TT
total temperature
U
upper surface
V
velocity/speed
Vsig
1g stall speed
x=c
chord fraction
a
angle of attack (deg)
de
elevator deection angle (deg)
df
ap deection angle (deg)
ds
slat deection angle (deg)
y
glaze ice upper horn angle (deg)
Acronyms and abbreviations
An
Antonov

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

CFD
deg
DHC
ETW
FAA
GA
IBL
IRT
LE
LEWICE
LTPT
LWC
MAC
MD
MDC
min
mod.
MVD
NACA
NAE
NASA
NLF
NTF
RAF
RANS

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

683
684
684
685
695
698
703
704
705
709
709
709
719
735
738
749
750
751
755
758
759
761
762
763

computational uid dynamics


degrees
DeHavilland Canada
European Transonic Wind Tunnel
(US) Federal Aviation Administration
General Aviation
interactive boundary layer
(NASA Glenn) Icing Research Tunnel
leading edge
(NASA) Lewis Ice Accretion Code
(NASA Langley) Low Turbulence Pressure Tunnel
liquid water content (g/m3)
mean aerodynamic chord
McDonnell Douglas
McDonnell Douglas Corporation
minutes
modied
median volume diameter (mm)
National Advisory Committee for Aeronautics
National Aeronautical Establishment (of
Canada)
National Aeronautics and Space Admini
stration
Natural laminar ow
(NASA) National Transonic Facility
Royal Aircraft Factory
Reynolds Averaged Navier Stokes

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Ref.
RJ
RN, Rey. No.
S&C
SLD
TAI

Reference
Regional jet
Reynolds number
Stability and control
supercooled large droplets
thermal anti-ice

1. Introduction
Degradation of the aerodynamic effectiveness of
wings, tails, rotors, inlets, etc., due to contamination
of these surfaces in any form can have serious
consequences, particularly if not known and accounted
for. Typical contamination sources include damage
incurred during ground operations from foreign objects
thrown up from the ground, or from encounters with
ground support equipment, plus a number of in-ight
sources such as encounters with hail, birds, insects, etc.
However, the source which has undoubtably received
the most attention is ice accretions on these surfaces.
Numerous experimental results have shown that even
quite small ice accretions at critical locations can result
in substantial reductions in maximum lifting capability
and control surface effectiveness, control surface
anomalies, quite noticeable increases in drag, and, in
some cases, reduced engine performance and stability.
Decades of operational experiences have revealed
many situations in ight as well as on the ground when
ice can accrete on ice-protected aircraft. These
include:
*

Leaving selected areas or components (which are


susceptible to ice accretion) unprotected by ice
protection systems. This can be a result of either a
lack of aerodynamic criticality, and/or as a consequence of design optimization studies which consider
the cost and availability of resources needed for ice
protection versus other possible design solutions such
as increasing aerodynamic surface sizes, etc.
The initial, residual and inter-cycle ice accretions
inherent in the use of de-icing systems.
Delay in activating anti-icing systems (or any other
ice protection system).
Runback ice formations aft of areas protected by
anti-icing systems when water droplets are not
completely evaporated, either by design, or as a
consequence of a very intense icing encounter.
Ice accretions due to water droplet impingement aft
of leading-edge ice protection systems.
Upper-surface frost and/or ice accumulations caused
by exposure to adverse weather conditions on the
ground.
Frost or ice which forms on both upper and lower
surfaces of wings in proximity to the fuel tanks when
an aircraft is parked in high atmospheric humidity

TE
ted
teu
TIP
TU

671

trailing edge
(elevator) trailing-edge-down
(elevator) trailing-edge-up
(NASA/FAA) Tailplane Icing Program
Tupolov

conditions after fuel (which remains in the tank) has


been cold soaked in ight.
Much has also been learned over these many decades
about the many varied forms in which ice can accrete on
aircraft surfaces. Lessons gleaned from natural icing
encounters in ight and on the ground, from aircraft
ying behind tankers dispensing water droplets to
simulate icing cloud conditions, and from a great
number of tests conducted in icing tunnels, have taught
us that ice accretes in many locations, sizes, and shapes,
and with an assortment of ice surface roughnesses. The
size of in-ight leading-edge ice accretions are by-andlarge proportional to the liquid water content (LWC) of
the icing cloud, the velocity of the aircraft, and the
duration of the icing encounter. Conversely, the size
tends to be inversely proportional to the dimensions
(leading-edge radius, etc.) of the aircraft component
subjected to the icing encounter. Similarly, ice shapes
can vary anywhere from relatively small, nearly uniform-thickness buildups to a wide variety of large, very
irregular, so-called horned and lobster tail ice
shapes. These shapes are dependent again upon the
icing cloud conditions, duration of icing encounter, and
aircraft surface dimensions, but also very strongly
inuenced by the ambient temperature. Likewise, the
degree and type of resulting ice surface roughness/
irregularity are strongly dependent upon the size of the
water droplets being encountered, the ambient air
temperature, and the effectiveness of de-icing systems
in removing all the ice. Similarly, runback ice and
ground ice accretions vary widely.
With the very large variety of forms and sizes in which
ice can accrete on aircraft surfaces in real operational
conditions, the challenge facing researchers and aircraft
designers has been to establish an effective process for
dening the accretion process and physical characteristics of these ice accretions for any aircraft surface at
any ight or meteorological conditions, and to determine which are the most harmful accretions. Typically,
some ight testing in natural icing conditions is required
as part of the aircraft certication process for new
aircraft designs in order to demonstrate the effectiveness
of ice protection systems as well as overall aircraft
performance and handling characteristics. However,
such testing is not a practical approach for the bulk
of the effort involved in systematically assessing
the physical characteristics of ice accretions, for the

672

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

development and validation of ice protection systems, or


for conducting tests to measure the aerodynamic effects
of ice accretions. In addition to the major economic
and technical disadvantages inherent in the use of an
aerodynamic design process not centered around the use
of validated ice accretion simulation and scaling
techniques in conjunction with ground test facilities, a
design process requiring ight testing in natural icing
conditions would be plagued with seasonal limitations,
and the large uncertainty and risk involved in nding an
appropriate range of icing conditions. Such testing
would also not be sufciently general, nor would testing
in the accompanying atmospheric conditions be suitable
for quantifying aircraft performance effects. Studying
ice accretion characteristics by ight testing behind
tankers avoids some of these disadvantages, but also
introduces some important icing simulation issues of its
own due to evaporation effects in the subsaturated
conditions. Plume size, droplet size and spectra limitations are also icing tanker issues.
Hence, in order to realistically address the wide range
of real world in-ight ice accretion issues that must be
considered in the aircraft design process, much effort has
been focused for the past half century now on
developing and trying to validate analytical/empirical
simulation and scaling techniques. These techniques
have been used in conjunction with icing tunnel tests, for
both the development of ice protection systems, and to
dene the simulated in-ight ice shapes which are used
to assess a range of associated aerodynamic impacts
using conventional low-speed wind tunnels and ight
testing. And, interestingly enough, with the exception of
one very publicized fatal accident of an ATR-72 [1,2]
and a number of incidents thought to be associated with
encounters with the supercooled large droplets (SLD)
occuring with freezing rain and freezing drizzle (for
which there is still no formal certication requirement),
there does not appear to be any evidence that any other
accidents or incidents attributable to in-ight icing in
at least the last couple of decades or more were traceable
to inadequacies in these icing simulation and scaling
techniques utilized in the design and certication process
in spite of the many known limitations (i.e., inaccuracies) inherent in these techniques. Applications of these
methods do, however, tend to be rather time consuming,
and, there are some concerns that use of these methods
may at times result in ice protection system designs that
are not as energy efcient as they might be. As a
consequence, high priority has been given to the
development of essentially stand-alone, primarily computational (CFD), advanced simulation and scaling
methods which would ostensibly permit icing design
tasks to be accomplished more effectively and expeditiously in the future without compromising safety in any
way. However, some signicant CFD challenges still
remain to be overcome in this regard.

While the primary objective of this review is the


assessment of ice accretion effects on xed-wing aircraft
aerodynamic characteristics based on a survey of
available experimental results, some understanding of
the important factors and limitations involved in
computational and experimental icing simulation techniques is helpful when considering the utility of these
many measurements of the aerodynamic effects of
simulated ice accretions. Consequently, the next section
of this review provides somewhat of an overview of ice
accretion prediction fundamentals, corresponding experimental techniques, scaling rules, and the various
forms of ice accretions and critical parameters which
inuence them. More species on these methodologies
for those interested are provided in a 1998 review in this
journal by Kind et al. [3].
Following the discussion of icing simulation issues,
the remainder of this review is then focused on assessing,
correlating, and summarizing a range of measured
aerodynamic effects caused by various forms of simulated ice accretions obtained at a wide variety of test
conditions using numerous ground-test facilities, and,
where available, ight testing. To start this assessment,
some of the key aerodynamic testing simulation
variables and governing ow physics situations which
must be considered in order to realistically evaluate the
relative merits of various experimental measurements
are highlighted and discussed. Some crucial controlling
ow physics topics discussed include attachment line
conditions and related leading-edge ow conditions,
various airfoil/wing/tail stall mechanisms, and spanwise
variations in stall initiation on 3-D wings and empennages both with and without ice. The latter situation
clearly illustrates the importance of understanding the
prevailing ow physics for each ow situation being
studied in order to intelligently interpret test results for
assessing icing (and other) effects. A thorough understanding of testing simulation and ow physics strengths
and weaknesses is crucial for interpreting test results and
establishing their limitations, and a range of these
testing simulation and ow physics situations and
limitations are discussed for both single- and multielement airfoil/wing/tail geometries. Several important
practical considerations which must be taken into
account when planning test programs to assess ice
accretion effects, or when utilizing test results for specic
conguration applications, are also examined. These
include the typical lack of general applicability of test
results obtained with many specic 3-D geometries,
uncertainties involved in dening the most critical ice
buildups (including any residual accretions) for specic
applications, and potential unknowns in reliably dening just what ow mechanisms are controlling stall
characteristics for a wide range of practical aircraft
geometries. While the focus in this review is on xed
wing aircraft, and, in particular, on the lifting surfaces

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

(i.e., the wing and tail) of these vehicles, it is important


to note that many of the lessons learned can also provide
insights for icing situations with other applications such
as rotors, propellers, intakes, etc.
Ensuing sections then provide the actual review,
assessment, and correlation of a large number of
experimental measurements addressing the aerodynamic
effects of various forms of mostly simulated in-ight and
ground ice accretions, augmented where appropriate by
similar measurements for other analogous forms of
surface contamination and/or disruptions. For in-ight
icing, the various pertinent experimental results are rst
sorted or grouped by the type of contamination, starting
with the smallest accretions which would be of the order
of small roughness size (with a nominal roughnessheight-to-chord ratio o20  104), representing the
initial stages of an icing encounter or the required ice
buildup inherent in the use of de-icing systems. Runback
ice, or other (such as SLD) ice accretions located just aft
of a leading-edge ice protection system, are considered
next. Large rime and glaze ice accretions, including ice
shapes with rather grotesque features, are the third inight ice accretion category addressed. Following this,
the effects of surface roughness associated with hoar
frost accretions on wing upper surfaces that can be
accumulated on the ground are addressed, as well as
underwing frost associated with cold-soaked fuel. In
each of these four accretion categories, test results are
also grouped or sorted according to the complexity of
the test conguration whenever possible, following a
descending order of generality. For example, results
obtained with single-element airfoils, wings, tails, etc.,
are reviewed and assessed before results obtained with
more complex multi-element congurations, which are
more difcult yet to generalize. The effects of test
Reynolds number on the results will be illustrated
whenever possible by grouping and assessing relatively
low Reynolds number (o5  106) data separately from
corresponding higher Reynolds number wind-tunnel
and ight data more appropriate to larger (transport)
aircraft. This separation is necessary because even
though wind tunnel and ight studies carried out at
low Reynolds numbers may well be pertinent for the
ight regime occupied by some components (i.e.,
tailplanes, etc.) of general-aviation and smaller commuter-type aircraft, the use of such results by themselves
to establish performance trends for higher Reynolds
number applications can be very risky because of some
well-known low Reynolds number anomalies such as
laminar bubbles and transition variations which can
occur on the baseline un-iced geometry. Such aberrations also raise issues regarding the generality of such
results, even for other low Reynolds number applications. Consequently, concerns regarding the dangers of
relying on low Reynolds number data are brought up/
repeated a number of times throughout the ensuing

673

review because of the importance attached to this (often


overlooked) factor.
Incidentally, CFD methods have not been utilized in
this review to either correlate or expand the existing
experimental database on the aerodynamic effects of
various ice accretions. This is because even the most
advanced of these methods such as Reynolds Averaged
Navier Stokes (RANS) have not yet been demonstrated
to be reliable for this purpose, especially relative to
determining whether a ow is separated or not (even on
an uncontaminated surface). Claims of good agreement
between CFD and experimental results involving separation onset/progression characteristics typically involve post-test computations wherein a number of
adjustments (turbulence model, grid characteristics,
dissipation, constants, etc.) can be made to facilitate
the agreement. For example, having one turbulence
model work best for one ice shape and another one for
a different ice shape is not unusual. Also, obtaining
good predictions of global (integrated) forces without
agreement in pressure distributions (i.e., indicating that
the real ow physics are not being properly modeled)
has also been seen.

2. Icing simulation techniques


The range of in-ight ice accretion types for which
representative predictions and simulations are desired
varies from the very initial stages of the ice accretion
process up to the large, very irregular, and rough ice
accretions. Properly simulating the initial stages of the
ice accretion process is critical in the design and
validation of effective ice protection systems, as well as
in assessing the aerodynamic effects of relatively small
ice accretions such as those associated with delayed
activation of an ice protection system or inherent in the
use of de-icing systems where some ice thickness is
allowed/required to build up before it is (can be) broken
loose. For these conditions, parameters of most concern
are icing impingement limits, accumulation rates,
collection efciencies, and the severity of the ice surface
roughness. Approaching the other end of the spectrum,
a large variety of ice accretion types must be dealt with
including the ridges located somewhat aft of the
leading edge (LE) resulting from SLD encounters. And,
while rime and glaze ice are the most commonly known,
there are so many variations of these that a separate
vocabulary has been devised to describe the various
physical features [4,5]. Rime ice forms at temperatures
below about 101C, and at cloud LWCs at the lower
end of spectrum, where a high rate of heat loss from the
impinging droplets causes them to freeze upon impact.
These ice shapes normally have a single horn, tend to
conform somewhat to the surface geometry, and have a
relatively smooth surface. Glaze ice, on the other hand,

674

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

forms at temperatures between about 31C and freezing, and at higher LWCs, where there is insufcient heat
transfer to remove all of the latent heat of the impinging
water droplets, and so some unfrozen water remains.
Glaze ice accretions are normally larger, typically
develop pronounced double horns and/or lobster
tails, and have very rough surfaces, but there are a
myriad of variations possible over a range of air
temperatures, LWCs, and surface geometries. In between these rime and glaze ice accretions lies a wide
variety of mixed ice accretions. Obtaining representative predictions and/or simulations of these larger ice
shapes is clearly desired in order to enable realistic
assessments of the aerodynamic consequences of such
accretions. However, while accurate predictions and
simulations may be desired, it is important to put the
accuracy realistically achievable (now and in the
foreseeable future) with current simulation methodology
in proper perspective in light of some of the fundamental
limitations inherent in the methodology employed.
2.1. Simulation fundamentals
The objective of in-ight icing simulation is to
accurately represent the time-dependent ice accretion
process which occurs when an aircraft ies through a
cloud containing super-cooled water droplets, or encounters freezing rain or drizzle. It has been well
established that three fundamental factors must be
taken into account concurrently for realistic simulation.
They are the representation of the aerodynamic oweld
(potential and viscous) characteristics on and around the
body and growing ice accretion, the establishment of
the water droplet trajectories with subsequent impingement characteristics and limits, and, lastly, the thermodynamics of the freezing/ice growth process. In the
analytical simulation process, successive thin ice layers
are formed on the surface, and are followed by oweld,
droplet impingement, and ice growth recalculations. The
frame of reference adopted to date for addressing these
factors in analytical as well as ground-test (icing tunnel)
studies of the ice accretion process has been the standard
Galilean transformation of xed body with intended or
assumed uniform onset ow, and with the water droplets
at the freestream velocity prior to encountering the
aircraft oweld [3].
2.1.1. Flowfield determination
A close representation of the oweld around and on
the aerodynamic surface and growing ice shape, either
computationally or experimentally, serves two important functions. The off-body oweld is clearly important for proper simulation of water droplet
trajectories, while the (viscous) ow characteristics
on the surface are critical in establishing the convective
heat transfer levels and distributions which have a major

impact on ice accretion rates as well as design


requirements for ice protection systems. Since the
boundary-layer transition process, and the existence of
any separated ow regions, have a major inuence on
establishing local heat transfer rates, then any oweld
(or geometry) variables which are known to inuence
transition and separation such as Reynolds number and
freestream turbulence levels should be carefully considered and reproduced to the extent possible. Transition location is more critical in the early stages of ice
buildup, and for the design of ice protection systems,
while separation becomes more of an issue as ice shapes
become more convoluted with additional icing exposure.
Choice of a most appropriate oweld method for
this purpose is still being debated. Ideally, and
realistically, the method selected should be capable of
handling complex aircraft and ice geometries, and ow
situations, including large ow separations caused by
the existence of ice horns, etc. And, very importantly,
the method should also be able to account for the effects
of substantial (and often irregular) ice surface roughness. Candidate oweld prediction methods normally
suggested to provide these capabilities range from linear
panel methods with either integral boundary layer [6] or
interactive boundary layer (IBL) representation [7] up to
RANS [8]. While it might have been presumed that
(RANS) methods would be the obvious choice to
capture the necessary physics, particularly the separated
ow regions, this is clearly not yet the case. Reliable
predictions of separation onset and progression characteristics for a range of geometries and conditions
remains a most needed, but elusive, goal for RANS, and
that is the situation for smooth surfaces without high
local curvatures. The challenge would be even more
formidable for rough, irregular ice shapes with very high
local curvatures. Establishing suitable eld grids for the
irregular ice shapes has also been a formidable challenge
for the structured-grid Navier Stokes methods [9].
Important non-physical modications of these ice
shapes such as fairing out the deep cavities and
smoothing surface irregularities have been necessary in
order to avoid highly skewed and/or crossing grid lines.
Panel methods (with IBL), on the other hand, are
theoretically well suited for analysis of these complex
geometries because no eld grids are required, but they
are even more limited (than RANS) in their ability to
adequately address ows with any signicant regions of
separated ow or strong viscous/inviscid interactions.
Extensive, somewhat arbitrary smoothing of irregular
ice shapes is also necessary in order to obtain IBL
solutions for these geometries [7].
2.1.2. Droplet trajectories
Establishing representative water droplet trajectories
is the second step in the process for simulating ice
accretions which occur when an aircraft penetrates

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

a cloud containing supercooled water droplets, freezing


drizzle, or freezing rain. These trajectories are required
in order to determine the amount (i.e., total collection
efciency), distribution (local collection efciency), and
extent (impingement limits) of the water which strikes
the various aircraft surfaces (or developing ice shapes)
for a prescribed icing cloud content. The basic mathematical formulation for droplet trajectory calculations
in the presently used frame of reference was developed
over 50 years ago by Langmuir and Blodgett [10],
although various researchers have subsequently implemented a number of renements and enhancements
[1115]. In this process, the trajectory of a droplet is
determined by integrating a differential equation representing the force balance on the droplet involving
inertial, drag, buoyancy, and gravitational forces.
Trajectories for a series of droplets are determined
starting from an upstream (freestream) position until the
droplets either collide with or bypass the surfaces.
A fundamental assumption inherent in the use of this
process is that the proportion of water in the approaching oweld is low enough that the inuence of water
droplets on the oweld can be ignored [16,17].
Physical parameters of prime importance in droplet
trajectory simulation are the size(s) of the droplets, the
overall cloud content, and the size of the surface
(leading-edge radius) subject to collision with the
droplets. Smaller and/or more slender surfaces tend to
collect much more ice and have more aft impingement
limits because the droplets have less chance to be
deected by the (smaller) leading-edge oweld inuence. Droplet size is, however, the single most important
variable in droplet trajectory simulation, largely because
of the impact on the balance between droplet inertia
and drag forces. Size also has an inuence on whether
droplet terminal fall velocities become signicant
enough that they are no longer negligible. Droplet sizes
of interest for icing vary from 10 to 50 mm (diameter)
range routinely addressed to date in the certication
process at the low end, up to about 400 mm encountered
in freezing drizzle, and even larger in freezing rain.
Droplet sizes encountered in the latter two categories are
now often referred to as SLD [18]. At the smaller sizes
normally addressed in the certication process, the
terminal fall velocities are quite generally accepted as
being negligible [19]. At the low end, droplet drag forces
dominate the inertia forces, so these droplets tend to
closely follow the streamlines. However, as the droplet
size increases, the inertia forces (proportional to the
cube of the droplet diameter) become dominant over the
drag forces (proportional to the square of the diameter).
Under these circumstances, the droplets do not tend to
follow the streamlines, and are therefore much more
likely to impact the surface, with resultant greater
collection efciencies and often more downstream
impingement limits. Further, with the much larger

675

SLDs, terminal fall velocities which occur in natural


icing conditions might have a small impact. For
example, a 500 mm droplet would be expected to have
a fall velocity of about 2 m/s. This might be expected to
move upper-surface impingement limits even a little
further downstream.
Computational and icing tunnel simulations of the
overall cloud contents found in nature can only be
approximations, since the conditions encountered in the
natural icing environment are believed to be quite nonuniform and irregular. For example, precipitation
typically occurs when the cloud droplet population
becomes unstable, and some droplets grow at the
expense of others by collision and coalescence as a
consequence of having different fall velocities. For the
purpose of icing simulation, the cloud content is usually
dened by the LWC, and either a droplet diameter
distribution, or a single droplet size selected to be
representative, which is normally referred to as the
median volume diameter (MVD). One common distribution denition utilized to date is called the
Langmuir-D distribution [10]. Resultant collection
efciencies and impingement limits may be a little
different for these two approaches to modeling the
cloud contents, but the ice accretions predicted usually
end up being about the same.
The existing conventional (for wind tunnels and
computational studies) frame of reference used for
droplet trajectory (and icing) simulations has been
widely used and accepted for the past half century or
so. This widespread acceptance can be explained by the
absence of any glaring discrepancies between ice
accretions established using this framework and those
observed in-ight at corresponding conditions for
normal droplet sizes with single element wings, etc., that
were not thought to be explainable by some known
shortcoming in the simulation of the thermodynamics
of the freezing/ice growth process [20]. There have,
however, been some concerns raised more recently
regarding the adequacy of these techniques for SLD
conditions, and for some aspects of ice accretions
obtained with subscale model tests of multi-element
geometries in icing tunnels. For the SLD conditions, the
current concern is the need to account for loss of mass
because of drop break-up due to shear forces before
reaching the airframe, and then due to drop splashing
and bouncing. Gravity effects on SLD are also a concern
for icing wind tunnels. For multi-element high-lift
geometries, the concern has been with regard to some
of the ice accretions observed on the downstream
elements in icing tunnels. To illustrate this point, typical
ice accretions observed during a test of an advanced
high-lift system geometry [21,22] are depicted in Fig. 1.
While there have been a few (although not very many)
observations and reports [23] of ice accretions occurring
on ap LEs during ight in natural icing conditions,

676

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 1. Ice accretions on multi-element airfoil in icing tunnel [21,22].

there are no known observations or reports of ice


forming around the LE of the main element (aft of the
deected leading-edge device) as occurs in icing tunnels.
Initial speculation focused on the absence of any droplet
terminal fall velocity component in the simulation
process as the possible culprit, but subsequent studies
of droplet trajectories [24] have suggested that the
observed disconnect may also possibly be explained to
some degree by scale differences. Another possibility
is that aircraft incidence angles at realistic operating
conditions do not normally reach the levels where these
downstream accretions seem to be more prevalent
and/or pronounced in icing tunnel tests. Further

investigations to resolve this matter are certainly


warranted, including some determination as to just
how representative are the ice shapes formed on the ap
LEs in icing tunnels.
2.1.3. Thermodynamics/freezing process
For many years, it was assumed that once droplet
impingement characteristics were established, the actual
ice accretion process was governed almost exclusively
by the convective heat transfer characteristics at the
surface. The method widely used to simulate the
accretion process was initially formulated by Tribus
et al. [25] in the late 1940s, and later enhanced by

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Messinger [26]. The implementation of this method in


the American, Canadian and European ice accretion
prediction codes [3,6] involves a control volume
approach. In it, each segment of the surface is assigned
a control volume located on the surface and extending
beyond the boundary layer, and a mass and energy
balance is performed on each control volume using the
rst law of thermodynamics to determine the quantity of
ice accreting within it. Any water which does not freeze
in a control volume is assumed to run back and freeze in
a subsequent control volume. This methodology produced the popular concept of a freezing fraction,
which is the ratio of the liquid freezing within a control
volume to the total amount of liquid entering the control
volume. Unfortunately, even though this methodology is
typically adequate for rime ice predictions, it was shown
some time ago by Olsen and Walker [20] that this
freezing fraction concept/approach does not work well
for glaze ice accretions for a number of reasons. For
instance, the absence of any momentum balance results
in important water surface tension and external owshear effects on the runback droplets not being
accounted for [2729]. And, any imprecise knowledge
of local heat transfer characteristics and effects becomes
much more limiting in the simulation of these accretions.
The heat transfer/freezing process for glaze and mixed
ice accretions is complex, and is governed by many
factors in addition to the temperature difference between
the aircraft/ice surface and impinging water droplets.
For example, whether the boundary layer on the surface
is laminar, transitional, turbulent, or separated, has a
strong inuence on the heat transfer rates, as does the
existence of any water layer on the surface. The state of
the boundary layer is, in turn, strongly dependent upon
the Reynolds number, surface sweep, surface/ice geometry, surface roughness, and freestream turbulence/
background noise levels. And, for a particular boundary
layer state, heat transfer rates are a function of local skin
friction level, Prandtl number, surface velocity, etc.
Surface roughness also has a major impact on increasing
heat transfer rates with transitional and turbulent
boundary layers. While an accurate determination of
local heat transfer characteristics is not that important
for predicting representative rime ice accretions, it is if
consequential predictions of glaze (and mixed) ice
accretions are to be achieved. This requirement represents a formidable problem. Perhaps the toughest part
lies in being able to effectively account for both
roughness and surface tension effects, which are often
very closely intertwined. Contributing to the difculty in
understanding these complex ow phenomena, especially for larger aircraft, is the fact that most of
the experimental investigations carried out to explore
these phenomena have been conducted with small 2-D
single-element models at quite low Reynolds numbers
in existing icing tunnels (which often have elevated

677

turbulence levels caused by the water droplet dispensing


rig, etc.). Important factors in the freezing process,
especially in the early stages of glaze ice accretions, are
strongly inuenced by the particular surface viscous ow
features peculiar to these low Reynolds numbers on 2-D
single-element models, thereby causing the observed ice
accretion characteristics at these conditions to likely be
of limited value for assessing other geometries and
conditions (i.e., higher Reynolds numbers). For example, the signicant runs of laminar and transitional
boundary layers at the lower Reynolds numbers will
result in different accretion physics than would be
experienced on 3-D swept surfaces at higher Reynolds
numbers where the surface boundary layer may be all
(or nearly all) turbulent. Even 2-D test measurements
[30] have documented that modest increases in Reynolds
number result in higher heat transfer rates over and just
downstream of ice-type roughness. Unfortunately,
ground test facilities do not presently exist which would
permit the detailed study of ice accretion processes at the
noticeably higher Reynolds numbers and lower freestream turbulence levels representative of ight for
larger aircraft, although the use of so-called hybrid
models (discussed in Section 3.2) has enabled some quite
useful insights in this regard.
The surface conditions in the initial stages of the ice
accretion process are often described as closely resembling uniformly distributed roughness, with measured
roughness heights typically several times larger than the
boundary layer thickness close to the LE of an un-iced
surface [31]. However, rather than causing a quick
transition to a fully developed turbulent boundary layer
as was initially thought, detailed (2-D low Reynolds
number) experimental studies have shown that this
relatively large distributed roughness instead started a
transitional boundary layer which exists for a signicant
chordwise distance before attaining a fully developed
turbulent state [32]. Corresponding measurements have
revealed that much-higher-than-expected heat transfer
rates existed in this rough transitional boundary layer
region. Experimental studies of glaze and mixed ice
accretion characteristics have also shown, however, that
this stage with relatively uniform distributed roughness
does not last for very long. A rapid ensuing roughness
growth takes place which then leads to a very nonuniform surface roughness distribution which is difcult
to model effectively. This is a signicant concern since
good estimates of roughness size and characteristics are
needed as the rst step in establishing reliable heat
transfer coefcients.
As indicated earlier, water surface tension effects have
also been shown to play an important role in the
accretion process for glaze and mixed ice formations,
and surface roughness effects play a substantial role in
this aspect as well by impeding the movement of water
droplets along the surface. For rime ice accretions, it is

678

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

believed that there is little, if any, movement of surface


droplets, since freezing creates some rough ice which
greatly increases surface tension adhesion forces which,
in turn, prevent aerodynamic forces from moving the
droplets. On the other hand, studies of the glaze ice
accretion process by Olsen and Walker [20] (at low
Reynolds numbers) did reveal the presence of two
distinct regions in the early stages. One is a smooth
region near the stagnation line characterized by a thin
lm of water over a uniform layer of ice, while the
second is a region somewhat downstream where there is
an immediate change into a rough zone, and there is no
lm of water present. And, ice accretion rates are much
higher in this region due to the effects of roughness. It
was also observed that the boundary between the
smooth and rough regions propagates upstream with
increasing time, at colder temperatures, and with
increasing freestream velocity and LWC. Regarding
the details of the low Reynolds number glaze ice
accretion process, it was seen that as water droplets
impinge and wet the surface, they coalesce to form larger
beads, which, as a consequence of aerodynamic forces
acting upon them, run back slowly along the surface
until they either reach the point where the surface
tension forces balance the aerodynamic forces, or the
base of the water beads freezes, forming an ice substrate
which helps hold the water beads in place. The freezing
of these large droplets, greatly enhanced by the
attendant high heat transfer rates associated with the
large roughness, starts the formation of the characteristic glaze ice horns. Once the horns start to protrude,
they then catch an even greater number of droplets
relative to surrounding areas, and the horn growth rate
increases.
Unfortunately, even though the understanding gained
from these detailed studies of the glaze ice accretion
process on 2-D single-element models at quite low
Reynolds numbers does provide some important insights into what is important for effective analytical
modeling of these complex ow phenomena, this
existing database does not provide much of a basis to
guide the development and validation of future, primarilyFCFD modeling of glaze and mixed ice accretions for the wide range of surface geometries and
Reynolds numbers needed. For example, just for
starters, it would be expected that the aerodynamic
forces acting on the coalescing and coalesced water
droplets would be noticeably different if turbulent
attachment line conditions existed, rather than the
starting laminar boundary layer present in the existing
2-D low Reynolds number tests. And, as if the
computational modeling challenges for glaze and mixed
ice accretions were not already difcult enough, they are
even further complicated (at all Reynolds numbers) if
other known factors inuencing these accretions such as
droplet splash back effects, the shedding of surface

droplets (especially near freezing temperatures), and the


effects of unsteady ow phenomena involving vortical
ow structures, etc., are to be accounted for.
2.2. Overall simulation effectiveness
Clearly, the two issues/questions needing to be
addressed are what improvements to current icing
simulation capabilities are required (i.e., just how precise
or accurate do the droplet trajectory and simulated/
predicted ice shapes and roughness characteristics have
to be), and what improvements are realistically attainable. To start with, it is clear that modications to
permit realistic predictions of SLD impingement characteristics, including any terminal fall velocity component present, are denitely needed, and should be
possible. However, with regard to the need for other
improvements, a number of factors need to be considered. For one, it is important to recognize that icing
simulation methods at best can only provide approximate estimates of ice accretions, since it is really
impossible to accurately represent the non-uniform
and irregular conditions encountered in the natural
icing environment. For instance, just how important can
it be to go to great detail to represent very irregular
surface roughness characteristics on approximate ice
shapes when these roughness characteristics have been
observed on a very limited number of geometries in an
icing tunnel environment at relatively low Reynolds
numbers. Also, as indicated previously, other than one
exception thought to be associated with an SLD
encounter, there is no evidence that any other fatal
accidents in modern times attributable to in-ight icing
were caused by inadequacies in these commonly used
icing simulation (and scaling) techniques. This, together
with the existence of (high Reynolds number) test results
from years ago [33] which indicated that even very small
ice accretions can have a detrimental effect on maximum
lift capability nearly as large as that seen with large
horned ice shapes, has led to speculation that perhaps
any large glaze ice shape (and roughness) may provide a
representative order-of-magnitude degradation in maximum lifting capability for lifting surfaces (except,
perhaps, for SLD conditions). This possibility needs to
be explored since, if true, it leads to the conclusion that
additional improvements to the icing simulation methods (other than for SLD conditions) may not be needed,
at least from a safety perspective. At this point, there
does not appear to be a compelling justication for
pursuing improvements to current ice accretion predictions techniques (for the sake of science). This issue will
be revisited after the subsequent review of test results
addressing the aerodynamic effects of various simulated
ice accretions.
If it turns out (i.e., can be shown) that there is a real
need for pursuing further renements, etc. to these icing

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

simulation methods (in addition to the SLD capabilities), the technical community may be faced with the
reality that just wanting or needing these improvements does not make them possible (in the foreseeable
future). It is clear that the focus would have to be on
glaze ice shape and roughness predictions, since that
would be the area in most need of improvement.
However, based on the earlier discussion of the most
formidable CFD challenges involved in providing more
accurate glaze ice accretion predictions, it would not be
reasonable or prudent to presume that these advancements could be achieved any time soon.

2.3. Scaling techniques


As indicated earlier, ight testing in either natural
icing conditions or in simulated clouds produced by
tankers is not a viable approach for systematically
assessing the physical characteristics of a broad range of
ice accretions, or for the development and validation of
ice protection systems. And, since it is often not possible
to test full scale models of major aircraft components
such as wings or even tailplanes in icing tunnels except
for the very smallest of aircraft, a high priority has been
put on the development of scaling techniques in order to
enable the use of subscale models in existing icing tunnels
to study this wide range of icing issues. The objective for
scaling procedures is to permit the denition of similar
ice accretions, both in shape and surface roughness, on
geometrically similar aerodynamic surfaces under different ight and meteorological conditions, thereby negating test facility operational limits.
It should be understood without saying that the
starting point for any viable ice accretion scaling laws or
techniques must be the consideration of the same
fundamental factors essential for basic icing simulation,
namely the oweld, droplet trajectories and corresponding collection efciencies and impingement limits,
plus the thermodynamics of the freezing process. It then
follows that scaling techniques will tend to have similar
strengths and weaknesses. Matching of many of the
important nondimensional parameters which should be
duplicated for meaningful scaling, especially those
controlling droplet impingement characteristics, the
mass of water impinging on the surface, and the
thermodynamics of the freezing process, can be accomplished by maintaining the scale between the (scaled) test
object and droplet sizing/spacing (matching LWC), and
maintaining the same ambient temperatures. However,
the companion requirement to hold Mach number,
Reynolds number, and Weber number constant between
simulations is, unfortunately, not satised by such
relatively simple scaling. Even if the Mach number
requirement could be waived, specifying that both
Reynolds number and Weber number be held constant

679

requires different scale velocities, both higher than the


reference value when subscale models are used [34,35].
Most of the early scaling methods developed by a
number of different organizations worldwide [3641]
have typically employed similar approaches for scaling
droplet and accumulation parameters, but somewhat
different ones for selection of thermodynamic scaling
parameters, although usually based on Messingers
freezing fraction approach (with all associated limitations). As would be expected, these techniques produced
reasonably good scaling results for rime ice accretions,
but no one has ever claimed success for scaling glaze and
mixed ice accretions. Subsequent improvements to
include surface tension effects yielded some improvements, but still short of that hoped for, since water
droplet impact and lm dynamics were not addressed
[27]. Scaling methods wherein Weber number is held
constant (at the expense of Reynolds number and Mach
number simulation) have provided some encouraging
results for glaze and mixed ice accretions [42], albeit for
a very limited range of conditions at low Reynolds
numbers in an icing tunnel.
Since these scaling methods are sometimes employed,
or at least intended, to experimentally dene ice
accretions for subsequent vehicle performance assessments, it is important to understand the deciencies of
even the best of these current scaling techniques
[29,43,44]. Some important limitations naturally arise
due to the previously discussed inadequacies of the
current analytical simulation methods (for glaze ice) for
modeling the oweld and freezing/accretion process
upon which the scaling methods are based. Another very
important shortcoming is the lack of any validation
results at the higher Reynolds numbers of interest.
Obtaining similar ice accretions for two different low
Reynolds number conditions in an icing tunnel does not
automatically imply in any way that the resultant ice
shapes are accurately duplicating shapes which would be
produced in ight at a higher Reynolds number
reference condition. Other concerns with the relevancy
of existing scaling methods arise because these methods
are basically only applicable to 2-D single-element
geometries at this time, and therefore not obviously
directly applicable to swept and multi-element congurations, although some surprisingly good representations of rime and mixed ice accretions on swept wings in
ight have been obtained with these 2-D methods.
Further, the constant Weber number scaling method for
glaze ice accretions has some serious limitations because
of the noticeably higher freestream velocities required
when using subscale models in an icing tunnel. These
higher airspeeds generate higher total temperatures
which can, in turn, produce nonrepresentative accretion
characteristics when temperatures approach freezing.
Use of the aforementioned hybrid airfoil models with
closer-to-full-scale LEs, and redesigned aft sections, has

680

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

been successfully used in a number of cases to overcome


some of the aforementioned shortcomings [4550],
especially for holding and cruise congurations during
which extended exposure to icing occurs. However, the
use of this concept does have some limitations. For one,
the oweld around the LE of these designs at elevated
angles of attack can become non-representative because
of premature separation on the upper surface. Also, this
concept is not really applicable to high-lift geometries,
and these hybrid models are not useable for determining
the various performance effects of the ice accretions
formed. Consequently, considering the many deciencies
and limitations outlined, current scaling methods just
like the aforementioned icing simulation methods, do
not do a very good job of dening actual ice accretion
shapes for many practical 3-D geometries at ight
conditions, except perhaps for less disruptive rime ice
and some mixed ice accretions on single-element
surfaces. Similarly, however, it remains to be seen
whether these imperfect scaling techniques might actually be adequate from a safety perspective.

3. Aerodynamic simulation considerations


Detrimental aerodynamic effects caused by ice accretions which are of most concern to aircraft designers and
operators, as well as regulatory agencies, include:
*

Reductions in lifting capability of lifting surfaces,


such as wings and tail surfaces, especially loss of
control due to reductions in maximum lift capability
(i.e., stall margins) and associated pitching moment
changes.

*
*

Loss or rapid changes/reversals of control surface


effectiveness on wings, empennages, etc. (especially
for unpowered surfaces).
Noticeable drag increases.
Disruption of airow to engines.

For each of these, there are a number of key ow


physics characteristics and testing simulation features
which must be carefully evaluated if existing experimental results for ice accretion (and other forms of
contamination) effects are to be properly interpreted
and/or applied. These key factors and considerations are
reviewed in the following subsections for four important
situations.
3.1. Single-element lifting-surface performance
Establishing meaningful incremental lift effects due to
ice accretions on this class of geometries is typically
more dependent upon achieving/identifying representative aerodynamic lift levels and characteristics on the
baseline (un-iced) conguration than on the iced one,
particularly when using ground-test facilities. This is
because the maximum lift levels achieved by the un-iced
designs are usually quite sensitive to the leading-edge
ow physics, particularly the extent of laminar ow,
which in turn is a strong function of the particular airfoil
or wing design, the test Reynolds number, the test
facility ow quality, model installation effects, etc. This
sensitivity is well illustrated by test results obtained over
a range of Reynolds numbers in two pressurized wind
tunnels for a representative 2-D wing airfoil section
[23,51,52]. The measured maximum lift levels are
compared in Fig. 2, where it can be seen that the levels
obtained in the two tunnels differ noticeably at the
lowest Reynolds number, with the lower level being

Fig. 2. Maximum lift differences attributable to dissimilar tunnel ow quality.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

experienced in the NASA Langley Low Turbulence


Pressure Tunnel (LTPT). A comparison of the surface
pressure distributions in the leading-edge region revealed that a laminar bubble was present at the higher
angles of attack in the LTPT, thereby limiting the
maximum lift level attainable, but did not exist in the
NAE facility with its higher turbulence and noise levels.
It would therefore be expected that using the lower
Reynolds number LTPT measurements for this baseline
(un-iced) airfoil would yield a lower maximum lift
penalty for leading-edge ice accretions than would be
experienced in the NAE facility at this lower Reynolds
number, and would result in a signicant underestimation of the maximum lift penalty which would occur in
ight at higher Reynolds numbers where laminar
bubbles would not exist on the baseline. Hence, one
must be very careful about using low Reynolds number
test results to predict high Reynolds number consequences. A number of examples illustrating this
concern are presented in the following sections. However, low Reynolds number data on iced congurations
may be useful in indicating trends if interpreted with
care with full regard to the physics of the ow.
Two other important factors to be considered when
assessing 2-D airfoil measurements of maximum lift
penalties caused by ice accretions include being able to
differentiate between the different ow physics limiting
maximum lift on various airfoil types, and assuring that
wind-tunnel test measurements are not overly compromised by inadequate side-wall boundary-layer control. It
is well documented that some airfoil types experience
a leading-edge type of stall characterized by a rather
abrupt loss in lift beyond stall. Others, such as relatively
thin airfoils, have a leading-edge type of stall (at a much
reduced lift level) without the abrupt lift loss beyond
stall, while other designs tend to have a much-lessabrupt trailing-edge type of stall. And, of course, there
are variations and combinations of each. It should not
be a surprise if the effects of leading-edge ice accretions,
etc. on maximum lift levels were different for these
different types of stall characteristics, likely being more
severe for the congurations experiencing an abrupt
leading-edge stall. As a case in point, it has been shown
that the sustainable leading-edge peak suction pressure
coefcient is reduced dramatically with even the smallest
amount of leading-edge roughness [51,52]. Next, the
need for effective side-wall boundary-layer control to
maintain 2-D ow conditions for 2-D airfoil tests has
been well established [53]. Again, this control is likely
more important for attaining representative maximum
lift values for the baseline un-iced airfoil. In this case,
the boundary layer in the juncture ow region at the
sidewall, which is more susceptible to separation than
the 2-D-like boundary layer in the center of the airfoil,
can well dictate the maximum lift level achievable. On
the other hand, with leading-edge ice accretions, etc. on

681

the airfoil itself, the ow in the juncture region could


conceivably be relatively less critical, but that is pure
conjecture, and not to be counted upon.
The proper assessment of ice accretion effects on 3-D
wings, tails, winglets, etc. becomes somewhat more
complex because, in addition to all of the aforementioned factors critical in assessing 2-D airfoil effects
(with the exception of the sidewall-boundary-layer
control issue), it becomes necessary with 3-D congurations to determine and account for the spanwise
variation in stall initiation on the un-iced baseline
conguration, because that can be a critical factor in
establishing subsequent (maximum) lift and pitching
moment changes caused by the ice accretions. Well
designed wings, as well as other lifting surfaces, do not
normally have the stall occur simultaneously across the
whole span. Typically, a particular spanwise location is
selected to be critical for stall initiation to achieve
desired stall characteristics. As a result, regions of the
wing away from this area are less critical with regards to
stall initiation. Consequently, whereas ice accretions at
the critical spanwise location will cause an immediate
earlier stall, other locations along the span may possibly
accrue some small amount of ice before these sections
would become critical in initiating stall. It is important
to note that with relatively large wind-tunnel models,
tunnel wall effects can also modify those characteristics
[54]. Further, the installation of wing-mounted engines
(nacelles and pylons), fences, etc. can also signicantly
modify stall behavior. Hence, when planning tests or
assessing test results for the effects of ice accretion (and
other forms of contamination) on the lifting capability
of 3-D wings, tails, winglets, etc., it is essential that
potential spanwise differences in relative criticality for
stall initiation be taken into account. Typically, these
effects have not been considered or accounted for to
date in publications [55,56] reporting on the measured
effects of various disruptions on the maximum lift
capability of 3-D congurations. Therefore, these
previous ight-test results for burred rivets, insect
debris, chipped paint, etc., are not readily usable for
quantitative assessments. In addition, it is important to
note that ice accretion effects on the lifting characteristics of thin, highly swept wings suitable for supersonic
cruise represent a quite different set of ow physics and
consequences to deal with, although this situation is
probably not as critical because these congurations
typically do not experience the type of stall encountered
by subsonic aircraft.
3.2. Multi-element lifting-surface performance
Appropriate assessment of ice accretion effects on the
lift characteristics of multi-element high-lift geometries
is considerably more complex than for single-element
designs because of the wider range of geometries, ow

682

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

physics features, and ow conditions needing to be


addressed. To start with, it is necessary to consider both
takeoff and landing congurations/conditions in order
to address the effects of ground frost accumulations as
well as in-ight ice accretions. And, for each, it is
necessary to understand the relative criticality and
important interactions between the various elements
for any critical ight conditions such as maximum lift,
and lift at takeoff and approach attitudes. It is also
important to identify where (i.e., which elements, etc.)
ice can potentially accrete in each of these situations,
but, as discussed previously, there are still some
unresolved issues remaining with regard to identifying
realistic ice accretions on elements aft of the leadingedge element.
In order to put the complexities involved in assessing
ice accretion effects on multi-element high-lift geometries into perspective, it is instructive to examine the
aerodynamic characteristics of a representative 2-D
three-element (slat, main, and ap) geometry in the
landing approach conguration, and then consider how
the limiting ow physics for conditions of interest can
change with conguration modications. The baseline
geometry selected has been the subject of considerable
research [54,5763], both experimental and computational, and is well suited for this purpose. Both the slat
and ap are deected 301. This geometry, as well as the
measured element (and total) lift characteristics for a
range of angles of attack leading up to maximum lift
conditions in the NASA Langley LTPT, at a freestream
Mach number of 0.2, and a (stowed) chord Reynolds
number of 9  106, are shown in Fig. 3. Incidentally,

Fig. 3. Element lift distribution with multi-element high lift


airfoil.

these results were all obtained using a good state-of-theart sidewall boundary-layer control system. However,
there is no guarantee that all sidewall viscous inuences
have been avoided at critical conditions. It is important
to note that the requirement for effective sidewall
boundary-layer control is much more crucial for multielement geometries since the sidewall boundary layer
experiences the total adverse gradient from the three
elements in this case. At these testing conditions, and
with the basic slat and ap riggings employed, it can be
seen that the maximum lift attainable is limited by an
eventual reduction in lift on the main element. A close
examination of the surface pressure distributions reveals
that the lift reduction on the main element beyond
maximum lift is caused by an unloading of the aft part
of the main element, which, in turn, appears to be
caused by a rather dramatic unloading of the ap itself.
However, surface skin friction measurements show that
neither the ap nor main element surface boundary
layers are even close to separation at these conditions,
nor is there any ow separation on the slat upper
surface, even though the peak suction pressures near the
LE are quite high, corresponding to a local Mach
number of nearly 1.2. Instead, it appears, based on a
review of off-body ow eld measurements, that the
initial ap unloading is somehow brought about by the
very rapid spreading and merging of shear layers and
wakes (from the main element and slat) above the ap
(although there is also the possibility that some
separation in the sidewall juncture region may be
contributing to this limiting phenomena). In contrast,
at lower-angle-of-attack approach conditions (aB81),
wake spreading and merging are of much less consequence, but the ow on the ap upper surface is very
close to separation near the trailing edge (TE). The ow
on the upper surfaces of the slat and main element are
very noncritical at this condition.
With the preceding set of ow physics characteristics,
certain ice accretions are likely to have more serious
consequences than others. Reecting on maximum lift
conditions, any ice accretions which thicken the slat and/
or main element wakes are certainly expected to result in
some reductions in maximum lift. Also, any measurable
ice accretions on the slat are expected to result in an
earlier, leading-edge-type stall (similar to the thin-airfoil
type), since the peak velocities on the un-iced slat at
maximum lift are close to experimentally observed
limits. With the ap being so uncritical in terms of ow
separation at maximum lift conditions, any small ice
accretions on the LE of the ap, even if they existed at
ight conditions, do not pose that much of a threat.
However, at approach conditions where the ow on the
ap is close to separation, any such accretions could lead
to a loss in lift and ap buffeting. At the same time, any
ice accretions on the slat or main element at these
approach conditions are not nearly as critical. This

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

whole situation can, nevertheless, change quite signicantly with possible slat and ap rigging changes, or
even with a freestream Mach number change. For
instance, if the slat were deected further (i.e., overdeected), then the maximum lift attainable would not
be nearly as sensitive to initial ice accretions on the slat.
The opposite would be true if the slat were deected less
(i.e., under-deected). On the other hand, if the ap were
deected further, or if the negative overhang were
increased, then any ice accretions on the LE of the ap
would have a more adverse impact. The same would
apply at approach conditions with respect to the ap
rigging changes. Further, an increased freestream Mach
number at maximum lift conditions could result in even
more serious penalties for leading-edge ice accretions. At
takeoff conditions, where the slat and ap are normally
not deected as much as for approach, stall is normally
initiated by ow separation at the LE. Hence, any ice
accretions on the slat would be expected to have the
greatest adverse impact, although the magnitude of the
impact would be reduced if overspeed (i.e., using higherthan-the-minimum required speed) was being utilized.
Consequently, it is absolutely essential that the ow
physics controlling stall, or any other situation involving
potential ow separation, be carefully considered
when assessing the effects of various potential ice
accretions on the lifting characteristics of multi-element
congurations.
Analogous with the situation for single-element
lifting-surface congurations, the proper assessment of
ice accretion effects on the lifting characteristics of 3-D
multi-element high-lift systems becomes noticeably more
complex. Again, it is absolutely essential to determine
and account for where stall, or potential separation
on the ap at approach conditions, is initiated spanwise
on the un-iced baseline conguration. Further, because
of the much higher angles of attack encountered with
high-lift systems, any installation of wing-mounted
engines can have an even larger, often very dominant,
impact on stall initiation. In this latter case, it would
therefore be quite difcult to provide any very generally
applicable conclusions regarding ice accretion effects on
maximum lift characteristics for such congurations.

3.3. Control surface effectiveness


Ice accretions on wing and tail surfaces ahead of
trailing-edge control surfaces such as ailerons, elevators,
and rudders, will cause reductions in control surface
effectiveness, and can also lead to some undesired
anomalies (i.e., reversals, oating, etc.) with aerodynamically balanced control surfaces [64]. The severity of
any such losses in basic effectiveness or characteristics is
controlled by the following three quite-fundamental
uid dynamic properties:

683

Health of the approaching boundary layer at the


control surface hinge point on the un-iced baseline.
Severity of the required ow turning associated with
the required control surface deection.
Degree of degradation of approaching boundary
layer or viscous layer health brought about by any ice
accretions.

Several factors are important in establishing the


health of the boundary layer approaching the control
surface. These include the issue of suction versus
pressure side of the surface, proximity to stall, existence
of leading-edge devices, as well as Reynolds number.
With regard to the ability of the ow to withstand the
turning (i.e., additional adverse pressure gradient)
established by the required control surface deection,
the magnitude of the required deection is obviously of
paramount importance, and Reynolds number is also
important, but, in addition, there are some indications
that the maximum deection angle sustainable without
ow separation may, in some cases, be reduced by the
presence of any appreciable spanwise viscous ow
component. And, lastly, the degree of degradation of
the approaching boundary layer induced by ice accretions is a strong function of the size, location, shape, and
surface roughness characteristics of the ice, as well as
being somewhat dependent upon Reynolds number.
With the wide range of possibilities in each of the
aforementioned lists of inuential variables, there is a
very broad range of ice accretion effects on control
surface effectiveness and characteristics possible. At one
end of the spectrum there could be a small reduction in
effectiveness caused by an increase in boundary-layer
displacement thickness at the LE of the control surface.
Such a situation might well occur if the basic lifting
surface were operating at a relatively low, non-critical
angle of incidence well below stall conditions. It could
also occur if the ice accretion was relatively small, not
too rough, and located forward where the boundary
layer would have a chance to recover somewhat from the
disruption. Another possibility would be if the control
surface deection required was relatively low. However,
at the other end of the spectrum there could be a large
loss in effectiveness or rapid change in hinge moments
brought about by a massive ow separation ahead of the
control surface. This situation might be expected, for
example, if the basic lifting surface (suction side) were
close to stall, if the ice accretions were more substantial,
rough, and/or located such that the boundary layer was
close to separation as it approached the control surface,
and/or if a relatively large control surface deection was
commanded. And, in between these two extremes, and
with different combinations of these important variables, lie a vast range of possible adverse effects. So,
again it is absolutely essential that the prevailing ow
physics characteristics be carefully considered when

684

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

assessing the effects of ice accretions on control surface


effectiveness. Generalizations are not appropriate, except, perhaps, at both ends of the spectrum. The same
caution again applies, even more so if possible, when
assessing ice accretion effects on the potential abnormal
behavior of aerodynamically balanced control surfaces.
This situation is made even more complex by the many
variations in system design and logic employed for this
type of control surface.
3.4. Drag characteristics
Establishing meaningful incremental drag increments
due to various forms of ice accretions is likewise
dependent upon an appropriate evaluation of many of
the same or similar ow physics and testing simulation
considerations as required for the preceding aerodynamic characteristics. To start with, ow conditions on
the un-iced baseline must be understood and accounted
for regarding where boundary-layer transition occurs, as
well as any propensity toward boundary layer separation, since two potentially important sources of increased drag due to ice accretion are adverse transition
movements and the onset of separation. For singleelement wings and tail surfaces, the airfoil type,
thickness ratio, test Reynolds number, and test facility
ow quality are all important variables in this determination. Consideration of baseline characteristics is
particularly crucial when assessing drag penalties due
to relatively small, roughness-like ice accretions where
the difference between causing separation or not can
have a large impact on the magnitude of the incremental
drag penalty. Similarly, test Reynolds number, as well as
the size, degree of roughness, and location of the ice
accretion, are all logically important variables in
determining the incremental drag penalties for such ice
accretions. On the other hand, such careful consideration of the baseline ow situation, ice accretion details,
and test condition, may well not be as important when
dening drag penalties caused by the larger, more
irregular ice accretions. However, in this case, distinctions between rime and glaze ice accretions are certainly
worth noting. Although the principal source of drag
caused by ice accretions is increased prole drag, some
increases in skin friction due to transition movement
(on small aircraft) and increased roughness, and to liftdependent drag (from ice-induced changes to the spanload distribution) can arise.

4. Effect of initial in-ight leading-edge ice accretions


Many observations of the ice accretion process in
ight in natural icing conditions, as well as in icing
tunnels, have revealed that the initial ice accumulation
on the LE of aerodynamic surfaces closely resembles

uniformly distributed roughness in the form of small


hemispheres, with a disturbance height nominally equal
to the thickness of the ice buildup. This roughness
typically extends over the rst several percent of the
local wing or tail chord on both upper and lower
surfaces for airfoil thickness ratios of most interest. The
aerodynamic consequences of this initial roughness-like
buildup can be important for any aircraft design, but are
particularly critical for aircraft employing de-icing
systems, since an ice buildup of more than a millimeter
is typically necessary before it can be consistently
removed. However, it is not unusual to see recommendations that observable thicknesses (i.e., 14 112 in) of
ice be allowed to accumulate before operating pneumatic boot airframe ice protection systems. Small
buildups must also be addressed for any delayed
activation of an anti-icing system, and for unprotected
surfaces as well. Test results available since the early
1930s [65] have revealed that disturbances caused by
small (k=co20  104 ) surface protuberances in this
leading-edge region of single-element lifting surfaces are
more critical in terms of causing signicant aerodynamic
performance and control degradations than similarly
sized ones located slightly aft of the LE region.
Although we are not aware of any good quantitative
data available from full-scale aircraft ight testing for
leading-edge roughness effects which would permit an
assessment of these effects on congurations that have
turbulent attachment line conditions, a wide range of
wind-tunnel test results from many experimental studies
carried out over the past 50 years or so do exist to permit
a pretty good indication of some of the potential effects
of such initial ice accretions on important aerodynamic
characteristics for both single- and multi-element lifting
surfaces. It is of interest to note, however, that most all
of these data were obtained using sharp particles for
roughness rather than the (droplet) hemispheres believed
to occur in natural icing conditions. The roughness
density employed may also be less than that normally
occurring in natural icing conditions as well. Whether or
not these differences might be consequential or not
relative to other differences which may exist such as the
boundary layer state in the leading-edge region has not
been established, but it is assumed herein that they are
not critical factors considering the wide range of
roughness effects observed. Reductions in maximum
lifting capability and stall angle, as well as drag increases
due to such roughnesses, are used primarily herein to
illustrate the adverse aerodynamic effects of these initial
ice accretions. Some more qualitative assessments of
leading-edge roughness effects on trailing-edge control
surface effectiveness and/or characteristics are also
provided, although there is, unfortunately, very little
data available on these subjects upon which to base any
such assessments. Considering the large number of
general aviation and commuter aircraft ying around

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

with de-icing systems, this lack of data is certainly of


concern.
4.1. Maximum lift reductions
4.1.1. Single-element lifting surfaces
Even though quite a number of wind-tunnel test
investigations have been carried out over the past decade
or so specically to help quantify the effects of small
roughness-like leading-edge ice accretions on the aerodynamic performance and control characteristics of
single-element lifting surfaces, the most extensive set of
test results on this subject are those published by Abbott
and von Doenhoff [66] over 50 years ago. In this
experiment, leading-edge roughness effects were established for over 100 different airfoils at low speed
(freestream Mach number o0.2) and at a chord
Reynolds number of 6  106. A single nondimensional
(k=c) roughness height of about 4.6  104 was used
throughout, which would correspond to a 1 mm ice
(roughness) buildup on an airfoil/wing/tail chord of just
over 2 m. The roughness was applied using carborundum grains, and extended from the LE to 8% chord on
both upper and lower surfaces, a fortunately quiteadequate coverage for initial ice accretion modeling.
This test program was conducted in the early version of
the NASA Langley LTPT. Freestream ow quality was
adequate (for these purposes) at that time, although no
sidewall boundary layer control was used. This should
be kept in mind, but it is not as serious a limitation as it
might be if the test were for multi-element geometries.
The measured impact of this particular leading-edge
roughness application to this large group of airfoils is
quite instructive in that it clearly illustrates how the
response of a particular airfoil to the addition of
leading-edge roughness is strongly inuenced by the
ow physics controlling the stall process on each airfoil.
It can be seen from Fig. 4 that a very wide range of
(percentage) reductions (or increases) in maximum lift
capability can occur. Roughness actually improved or
had very little effect on the maximum lift level attainable
by congurations which had a basic (unroughened)
maximum lift (coefcient) capability of about 1.0 or less,
or a thickness ratio of o8%, i.e., congurations with a
very pronounced thin-airfoil-type leading-edge separation. However, for thickness ratios of 9% or higher, or
for geometries with a basic maximum lift capability
above about 1.1, there is a very denite reduction in the
maximum lift capability of anywhere from about 15% to
nearly 40%, which, not surprisingly, does not correlate
well with simple parameters such as either the basic
maximum lift capability or the airfoil thickness ratio
(or leading-edge radii).
While the foregoing set of data is quite revealing in
terms of establishing the broad range of leading-edge
roughness effects than can be incurred on 2-D airfoils at

685

basically incompressible conditions, and at a constant


Reynolds number and (nondimensional) roughness
height, it is also important to be aware of the additional
variations which may occur as a consequence of
changing Reynolds number, varying the nondimensional
roughness height, operating at conditions where compressibility effects can modify the results, and/or having
3-D geometries with nite aspect ratio such as swept
wings and tails. Fortunately, there are a variety of windtunnel test results available that allow in-depth insights
into some of these, such as the effects of varying test
Reynolds number and (nondimensional) roughness
height. For Reynolds number effects, it is instructive
to rst look at the (chord) Reynolds number range up to
about 57 million. That is because most of the wind
tunnels in the world are limited to this range. Very few
facilities have the capability to produce data at
signicantly higher Reynolds numbers more representative of ight conditions with large vehicles. In this lower
Reynolds number range, there are four good sources of
test results [33,51,67,68] which illustrate how signicantly the losses in maximum lift capability due to
leading-edge roughness can vary for a given 2-D
conguration over this range. Clearly, the most comprehensive set of test results addressing this issue are the
data from over a half century ago provided by Loftin
and Smith [67] where leading-edge roughness effects for
15 of the same airfoils used by Abbott and von
Doenhoff [66] were determined at low-speed conditions
(Mach number o0.15) at Reynolds numbers ranging
from 0.7 to 6 million using the same roughness
conguration (i.e., extent) and height (i.e.,
k=cE4:6  104 ) as used by Abbott and von Doenhoff.
For 12 of the 15 airfoils tested, there was a very
pronounced increase in the (percentage) maximum lift
penalty due to the roughness as the Reynolds number
was increased from 1 to 2 million and above. Results for
these 12, which had thickness ratios varying from 9% to
15%, are illustrated in Fig. 5 where it can be seen that
the average penalty increased from just over 10% at
1 million to 17% at 2 million, and then to over 25% at
6 million Reynolds number. This increase in indicated
penalty at the higher Reynolds number is, to a large
degree, attributable to an increase in the maximum lift
capability of the unroughened airfoils with increasing
Reynolds number in this range. However, there were,
not surprisingly, also some very noticeable increases in
the maximum lift capability with a number of the airfoils
in the roughened state as the Reynolds number was
increased, although there were also some where there
was little Reynolds number effect in this roughened
state.
The three other sets of 2-D test results which address
Reynolds number effects on the loss in maximum lift
capability due to LE roughness in the Reynolds number
range up to 57 million actually cover a wider Reynolds

686

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 4. Maximum lift penalty caused by leading-edge roughness [66].

number range, but attention is focused rst on this lower


range. All three sets of test results were obtained in the
NASA Langley LTPT. One was the data provided by
Ladson [68] for the 0012 airfoil, while the other two used
supercritical (i.e., aft-loaded) airfoils with thickness
ratios around 11%. Ref. [68] data for the 0012 airfoil
were obtained with the same roughness height (i.e.,
k=cE4:6  104 ) used previously in Refs. [66,67], but in
this case, the roughness only extended back to 5% chord
on both the upper and lower surfaces. Test results for
this conguration at lowspeed conditions (Mach number=0.15) are available at 2, 4, and 6 million Reynolds

number. Of the other two, one is the data presented by


Morgan et al. [33] with roughness (k=cE4:5  104 )
extending to 5% chord on both upper and lower
surfaces. In this case, test results are available for
3 and 7 million Reynolds number at 0.2 Mach number.
The other set of applicable test results are
those presented by Lynch and colleagues [51,52]
with roughness extending back to 6% and 7.5% chord
on the upper and lower surfaces, respectively. For this
case, two different roughness sizes (k=c 0:7  104 and
5.3  104) were investigated, and test results are
available at 0.2 Mach number for 2.5 and 5 million

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

687

Fig. 5. Reynolds number effects on maximum lift penalty [67].

Fig. 6. Reynolds number effects on maximum lift penalty [33,51,52,68].

Reynolds number. Incidentally, this is the same conguration discussed in the preceding section where a
laminar bubble exists at higher angles of attack at 2.5
million Reynolds number on the unroughened airfoil,
but not at 5 million. The results from all three of these
sources in terms of the percentage loss in maximum lift
caused by the leading-edge roughness in this lower
Reynolds number range are depicted in Fig. 6. In each

case, there is a very discernible reduction in the indicated


penalty at the lowest/lower test Reynolds numbers, with
the reduction being most pronounced, interestingly
enough, with the smallest roughness size. Clearly,
therefore, the preponderance of the data available
addressing Reynolds number effects in this lower
Reynolds number range on the indicated incremental
loss in maximum lift capability due to leading-edge

688

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 7. Reynolds number effects on maximum lift penalty [68,69,70].

roughness (i.e., initial ice accretions) show that test


results obtained at chord Reynolds numbers much
below 56 million are most likely not meaningful or
useful for higher Reynolds number applications (i.e.,
X56 million). In fact, such test results would most
likely be quite misleading in a nonconservative direction.
Incidentally, Ref. [67] test results for the 0012 airfoil
need to be kept rmly in mind later on in Section 6 when
assessing the usefulness of the results from a signicant
number of low Reynolds number investigations of the
incremental effects of larger ice accretions.
Having established that test results addressing leading-edge roughness effects on maximum lift capabilities
obtained at Reynolds numbers much less than 56
million are not appropriate for applications at Reynolds
numbers of 56 million (and above), the obvious
question remaining is just how high is high enough to
obtain representative results for the higher Reynolds
number applications. There are ve sets of test results
available (again 2-D) that address this issue, and the
indications are denitely mixed. There are the three
older sets of test results that indicate that 56 million
is denitely not high enough. These are the Ladson
results [68] for the 0012 airfoil, the Loftin and Bursnall
results [69] for a 9% thick airfoil (NACA 63-009), and
the Abbott and Turner results [70] for a 22% thick
airfoil (NACA 63(420)-422). While the same roughness
height (k=cE4:6  104 ) was used for the 9% and 12%
thick airfoil tests, three different roughness heights
ranging from k=cE0:6  104 to 3.1  104 were employed in Ref. [70] investigation with the 22% thick
airfoil. The results from these three sources in terms of
the percentage loss in maximum lift capabilities are

shown in Fig. 7. For the 9% thick airfoil, the penalty


indicated at 6 million Reynolds number is substantially
below that incurred at Reynolds numbers of 1525
million, indicating that 6 million is clearly not adequate
for this conguration. Incidentally, in this case, all of the
variation with Reynolds number is caused by changes in
the unroughened airfoil maximum lift level. There is
essentially no variation in the roughened airfoil maximum lift level over the range of test Reynolds numbers.
With the 12% thick airfoil, an increase in the penalty is
evident as the Reynolds number is raised from 6 to
12 million (which is as far as the data go). However, in
this case, all of the increase is caused by a (further)
reduction in the maximum lift level of the roughened
airfoil at the highest Reynolds number. Incidentally, the
penalty at 9 million Reynolds number may actually be
larger than shown, since it is not clear that the real
maximum lift value for the unroughened airfoil was
reached. For the 22% thick airfoil, the indicated penalty
at 6 million Reynolds number is also substantially below
those incurred at the higher Reynolds numbers, but,
rather than attening out (at 15 million) like the penalty
did for the 9% thick conguration, there is, in general,
a continuing increase in the penalty all the way up to
26 million, clearly indicating that 6 million Reynolds
number is not adequate for this conguration either.
However, on the other side, there are two sets of test
results [33,51] indicating that the penalties due to
leading-edge roughness obtained at 56 million Reynolds number can be representative for higher Reynolds
number applications. Those are the results for the
two supercritical-type airfoil congurations around
11% thick discussed in the preceding paragraph. The

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

689

Fig. 8. Reynolds number effects on maximum lift penalty [33,51,52].

indicated penalties for these congurations at chord


Reynolds numbers of 5 million and above are illustrated
in Fig. 8. For these cases, it can be seen that there is
essentially no change in the maximum lift penalty due to
leading-edge roughness as the Reynolds number is
increased above 56 million. There is also one other
piece of evidence that supports these results for airfoils
with thickness ratios around 11%. That is the data for
the unroughened 12% thick airfoil contained in Ref. [68]
where the variations in maximum lift level from 6 to
25 million Reynolds number for this conguration are
clearly less than seen with the corresponding 6%, 9%,
and 18% thick airfoils.
Overall, the trend is denitely for the adverse effects
of leading-edge roughness/initial ice accretions on
changes in maximum lift capabilities to be reduced at
lower Reynolds numbers, but dening just how low is
too low for meaningful applications of test results to
larger aircraft remains somewhat of an open issue. For
the purpose of this review, 5 million has been established
as the lower limit suitable for higher Reynolds
number applications in the thickness ratio range of
most interest, largely based on the Refs. [51,52] test
results which were for a relatively modern wing airfoil
section. However, it is clear that data obtained at this
Reynolds number could still be quite misleading for
some geometries in a very non-conservative way.
Next, with regard to the effects of variations in
nondimensional roughness heights at high Reynolds
numbers, the known related quantitative data for chord
Reynolds numbers of 5 million and above are displayed
in Fig. 9 as a function of roughness height, with the data
from Refs. [6669] represented by the range of results
observed for airfoil thickness ratios of 9% and above.

Also, shown is the often used Brumby curve [55,56]


which was originally intended to represent the effects of
these small leading-edge roughnesses, but, as can be
seen, it is obviously inadequate. From Fig. 9, it can also
be seen that the compilation of all the test results
obtained subsequent to Refs. [6669] does indicate the
anticipated general reduction in the magnitude of the
maximum lift loss with decreasing nondimensional
roughness height, but the reduction is reasonably
gradual. Therefore, even at the smallest roughness
height, which corresponds to about 0.1 mm (0.004 in)
with a 2 m chord, the minimum loss observed is about
10%, but the maximum loss does range up to nearly
30%, although most of the results do not exceed 20%. It
should be noted, however, that this data band is
established with a relatively limited set of test results for
just a few geometries, and hence may well not be as
broad as necessary to cover a comprehensive group of
possible or likely aerodynamic designs. As a case in
point, the established data band only takes in about
60% of the test results which made up the wider range
depicted for the results from Refs. [6669] at that
particular roughness height. Also, it is worth noting that
the Ref. [71] results, which are at the low end of the
band, were obtained with a much coarser roughness
spacing than all the others. Consequently, with the
potential for even greater (percentage) losses in maximum lift than indicated, especially at the smaller
roughness heights, it would be extremely risky at this
time to assume that the consequences of any very small
buildups of leading-edge ice (roughness) on singleelement lifting surfaces would be anything less than
the 20% to nearly 40% maximum implied by the data
band of Fig. 8 unless indicated by very representative,

690

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 9. Roughness size effects on maximum lift penalty at high Reynolds numbers.

high quality test results for the specic geometry under


consideration. In other words, it should never be
assumed that there is any such thing as just a little
bit of leading-edge ice buildup.
A corresponding collection of known test results for
leading-edge roughness effects at low Reynolds numbers
on lifting surfaces having thickness ratios of 9% and
above, and having an unroughened maximum lift
coefcent of at least 1.0, is summarized in Fig. 10. Test
(chord) Reynolds numbers for this set of data varied
from 1 to 3.5 million. In comparing this set of results to
the data band limits established for the high Reynolds
number case, it can be seen that some of the low
Reynolds number data do fall within the same band
or an extension of it. However, consistent
with the trends illustrated in Figs. 5 and 6 addressing
Reynolds number effects, most of the test results do fall
substantially below the high Reynolds number data
lower limit. And, the range of results from Ref. [67] at
low Reynolds numbers is at a substantially lower level
that the range of results seen in Fig. 9 from Refs. [6669]
at higher Reynolds numbers. The results do tend to
indicate (not surprisingly) that a greater nondimensional
roughness height is required at low Reynolds numbers
to offset problems with the un-iced baseline geometry in
order to simulate the (incremental) penalty that would
be incurred at higher Reynolds numbers. But, the
problem is deciding on how much greater is enough in
each case. And, anyone thinking that they can determine
this a priori is, unfortunately, just kidding themselves.
Hence, as indicated previously, relying on low Reynolds

number test results alone to be representative of higher


Reynolds number conditions can be extremely risky.
Unfortunately, there are really no applicable data
available to permit a realistic assessment of either
roughness size or Reynolds number inuences on the
effects of small leading-edge roughnesses on lifting
surfaces having thickness ratios o9% and/or having
unroughened maximum lift coefcients o1.0. The only
data available for congurations in this category, in
addition to the Ref. [66] results shown previously in
Fig. 4, are the Ref. [69] 2-D results for three 6% thick
airfoils at 6 million Reynolds number (with a
k=cE4:6  104 ), and the Refs. [77,78] 3-D low
Reynolds number (B1.8 million) test results (with a
k=cE7:7  104 ) for a swept (441 tailplane geometry
having a 9% thick dening airfoil section perpendicular
to the quarter-chord line. While the dening airfoil
section was 9% thick, the effective aerodynamic
thickness was clearly less, as the unroughened maximum
lift coefcient was below 0.8. In this case, as well as with
the three 6% thick airfoils in Ref. [69], adding small LE
roughness actually increased the maximum lift capability
a little (up to as much as 10%), consistent with the
Ref. [66] results for congurations in this category. One
possible clue provided by these results regarding likely
Reynolds number inuences on the effects of small
leading-edge roughnesses on some of these thinner
geometries are the Ref. [69] unroughened maximum lift
characteristics for the three 6% thick airfoils. With
each of these, the maximum lift level was essentially
unchanged as the Reynolds number was increased from

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

691

Fig. 10. Roughness size effects on maximum lift penalty at low Reynolds numbers [76].

3 to 6 to 9 million, but had started to increase by


15 million. By comparison, with the two 9% thick
airfoils evaluated, the unroughened maximum lift level
started to increase above 6 million Reynolds number
(as reected in the results shown previously in Fig. 7).
Hence, it might be anticipated that any signicant
variation in the maximum lift penalty (or gain) caused
by small LE roughnesses with increasing Reynolds
number on at least some of these thinner surfaces might
well be delayed until above 9 million. However, on the
other hand, some very recent test results obtained by
Papadakis et al. [80] over a range of Reynolds numbers
for an 8% thick full-scale tail with a 1.6 min ice shape
casting (i.e., larger roughness and more irregular shape
than the small roughnesses just addressed) provide a
quite different perspective. In this case, the ice accretion
led to a small (B2.5%) reduction in the maximum lift
capability at 1.36 million Reynolds number, but this
penalty grew to over 20% at 5.1 million Reynolds
number. And, contrary to the Ref. [69] results for the
unroughened 6% thick airfoils, the maximum lift level
for the clean conguration in this case was increased
about 15% as the Reynolds number was raised from 2.2
to 5.1 million. Also, contributing to the large growth in
the maximum lift penalty as the Reynolds number was
increased was a 6% reduction in the maximum lift level

with the ice shape installed. So, once again, we are


reminded of the dangers involved in generalizations,
and in trying to use low Reynolds number test results
to predict consequences at higher Reynolds number
conditions.
As we have seen, there are sufcient test data available
on leading-edge roughness effects to permit a general
assessment of some of the consequences of changing test
Reynolds number or varying the nondimensional ice
roughness height on single-element airfoils at largely
incompressible conditions, particularly for the thicker
(X9%) congurations. But, on the other hand, there is
only one set of test results [68] that provides some insight
as to how operating at conditions where compressibility
effects can inuence the maximum lift achievable would
modify the magnitude of any roughness penalties, and
there are no known (useful) data that would permit an
understanding of potential differences between all the
results obtained for 2-D airfoils versus what might be
expected on comparable 3-D swept geometries. With
regard to compressibility effects, the test results for the
0012 airfoil provided by Ladson [68] do indicate
(as would be anticipated) a substantial reduction in the
magnitude of the indicated penalties at a condition
where the performance of the unroughened baseline
airfoil (i.e., the maximum lift) is compromised/reduced

692

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

by such compressibility effects. In this particular case,


because of the signicant reduction in the maximum lift
capability of the baseline airfoil, the indicated penalty at
0.3 Mach number at 69 million Reynolds number was
only about 16% compared to around 27% at 0.15 Mach
number at the same Reynolds numbers. It should be
kept in mind, however, that these results are for an
airfoil design which has substantial upper-surface
leading-edge suction peaks at stall. Reductions of this
magnitude would likely not occur for other geometries
with lesser suction peaks (such as NLF, thicker designs,
etc.), although those designs would likely not have as
large penalties at incompressible conditions. Now, with
regard to 2-D versus 3-D, in addition to the absence of
data, there is no unanimity in the direction of the trends
suggested by various technical rationalizations. One of
the complexities involved in the 2-D versus 3-D dilemma
is that some run of laminar ow most likely existed on
the baseline (unroughened) geometry for all of the test
results available, while turbulent attachment line conditions would most likely exist on most lifting surfaces of
interest on larger aircraft. With the existing uncertainty
regarding direction of trends in both situations, aircraft
designers should again exercise caution when using the
foregoing test results to predict maximum lift penalties
when either of these situations exist.
4.1.2. Single-element lifting surfaces with trailing-edge
control surfaces
Understanding and coping with the adverse effects of
initial and inter-cycle leading-edge ice accretions on the
maximum lift capability of single element wings and

tails with trailing-edge control surfaces deected is


extremely important. Not only are there the loss in lift
and possible loss of control caused by any premature
stalling to contend with, but, in addition, there can also
be a loss of control due to a number of possible control
surface anomalies which can occur at these maximum
lift (stalling conditions). Loss or reversal of control
surface effectiveness is an issue for any control surface
concept. But, in addition, aerodynamically balanced
control surfaces are also susceptible to oating, oscillations, etc. at these maximum lift conditions with ensuing
loss of control a real possibility. Unfortunately, however, there are surprisingly little data available to guide
the assessment of maximum lift degradations caused
by initial and inter-cycle ice accretions with deected
trailing-edge control surfaces. The only known applicable data available are relatively low Reynolds number
test results for four horizontal tail designs with (and
without) deected elevators [71,77,78,81]. And, of these,
one of the congurations is the aforementioned 441
swept tailplane design from Refs. [77,78] which has thinairfoil-type stall characteristics, and only one [81] has
data for trailing-edge-down (ted) elevator deections
(i.e., opposing the lift) which are needed to address a
required pushover maneuver. The results for these four
geometries in terms of the percentage loss in maximum
lift caused by the initial or inter-cycle leading-edge ice
accretions are illustrated in Fig. 11 for the range of
elevator deections tested. With the exception of the one
conguration having thin-airfoil-type stall features, the
trend is clearly in the direction of having an increased
penalty (percentagewise) as the elevator deection goes

Fig. 11. Variations in maximum lift penalty observed with trailing-edge elevator deection.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

693

Fig. 12. Roughness size effects on maximum lift penalty for multi-element airfoil geometries at high Reynolds number.

from trailing-edge-up (i.e., aiding lift) towards tailingedge-down (i.e., opposing the lift). Incidentally, for the
two sets of test results reported on in Refs. [77,78], it was
acknowledged in Ref. [78] that only upward elevator
deections (negative de ) were tested, and that later it
was realized that the most critical cases are for aircraft
that need elevator down (xed tailplane) for trim at ap
extension. Interestingly, quite a range of roughness
characterizations are represented in the results shown in
Fig. 11. In Ref. [71], roughness extended back to 20%
chord on both upper and lower surfaces, whereas in
Refs. [77,78], the roughness was more localized to the
leading edge. On the other hand, the results from
Ref. [81] are for a representative inter-cycle ice accretion
remaining on the tail between pneumatic de-ice boot
operation after 15 min with the boot having been cycled
every 3 min up to 12 min. With regard to the test results
summarized in Ref. [81], some data were obtained at a
higher Reynolds number (i.e., a little above 4 million),
but the results are not complete enough to draw any
very quantitative conclusions regarding Reynolds number inuences on the maximum lift characteristics with
elevators deected. Clearly, this is another area where
more studies are warranted.
4.1.3. Multi-element lifting surfaces
Again, in contrast to the relatively large database
which exists for single-element geometries, much less
data is available to help quantify the effects of initial and
inter-cycle leading-edge ice (roughness) accretions on the
aerodynamic characteristics of the much more complex
(mechanical and aerodynamic) multi-element lifting
surfaces routinely employed during takeoff and landing.
In fact, there are only two somewhat limited sets of
relevant high Reynolds number test results [51,82], and
two limited sets of low Reynolds number results [74,83].
But, the little that is available, used in conjunction with
lessons learned with single-element geometries (i.e., there

is no single answer), does provide some very instructional insights.


Starting with the high Reynolds number results, both
are from 2-D tests conducted in the NASA Langley
LTPT utilizing an effective sidewall-boundary-layer
control system [53]. In the rst [51,52], a four-element
(slat, main, and two-segment ap) geometry with
leading- and trailing-edge device deections appropriate
for approach/landing conditions was employed, and the
primary roughness coverage applied to the slat was
based on a water droplet impingement analysis at
approach conditions. The effects of extending the
roughness to the entire upper surface of the slat were
also investigated. The second test [82], which utilized the
same basic model as the rst, was conducted with a
three-element geometry (slat, main, and single-segment
ap) representative of a takeoff conguration. Although
this particular test was primarily carried out to address
the impact of ground frost accumulations, one frost
conguration evaluated had just the entire upper surface
of the slat covered. And, since results from the rst test
indicated that there was no discernible difference in the
effect of roughness on maximum lift capability with this
roughness coverage relative to the primary uppersurface coverage on the slat, the results from this second
test are included in this particular assessment. Both sets
of test results are shown in Fig. 12. By comparing the
Refs. [51,52] results for the four-element geometry with
the corresponding Refs. [51,52] single-segment airfoil
results shown previously in Fig. 9, it can be seen that the
percentage losses in maximum lift capability experienced
for this approach/landing geometry (of from 7% to
11%) are about one third of those incurred with the
single-element airfoil. This ratio just happens to be the
inverse of the ratio of the (unroughned) maximum lift
capabilities. Interestingly, this same proportional relationship holds with the Ref. [82] results for the takeoff
geometry. What this relationship signies is that the

694

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767


R

Fig. 13. Roughness size effects on maximum lift penalty for multi-element airfoil geometries at low Reynolds number.

absolute loss in maximum lift (coefcient) capability


with these particular related geometries is about the
same for all of the particular rigging variations tested at
a given nondimensional roughness height. It will also be
subsequently shown that the (absolute) reduction in
angle-of-attack margin to stall is about the same in each
case as well.
Just based on this small sample of high Reynolds
number test results, the following important inferences
can be drawn:
*

Small leading-edge ice accretions would, in general,


be more serious at takeoff conditions (compared to
approach/landing), since the percentage loss in
maximum lift capability would be greater at the
same time that the normal operating speed margins
(percentagewise) to stall are smaller.
Congurations with hard LEs (i.e., no slat or
Krueger) would, in general, be more vulnerable than
congurations having leading-edge devices extended
at both takeoff and approach/landing conditions,
since the percentage loss in maximum lift would most
likely be greater.

However, having seen the large band of results possible


with a variety of single-element lifting surfaces, and
recognizing the likelihood of different limiting ow
physics situations with other multi-element congurations, it would denitely be expected that the magnitude
of percentage losses in maximum lift capability for other
multi-element congurations and applications could
well be noticeably different (higher or lower) than those
shown in Fig. 12. And, potential differences between
these 2-D test results and real world 3-D designs must

not be forgotten. So, again, aircraft designers need to be


extremely cautious when estimating potential penalties
for small leading-edge ice accretions on multi-element
designs.
Justication for this caution is underscored after
examining the results obtained from the two (2-D) low
Reynolds number investigations of leading-edge roughness effects on multi-element airfoil geometries that are
illustrated in Fig. 13. Each of the investigations explored
these roughness effects on congurations with singlesegment aps, and both with and without leading-edge
slats. One of the studies [83] addressed takeoff geometries, while the other [74] examined an approach/landing
geometry. The test Reynolds number range for these
tests was 2.62.7 million, and a sidewall boundary layer
control system was used with each. From the low
Reynolds number results shown in Fig. 13, together with
the related high Reynolds number results, it can be seen
that the leading-edge roughness penalty for the
Ref. [74] approach/landing geometry with slats is not
very different from the high Reynolds number approach/landing conguration results, and, not surprisingly, the maximum lift penalty due to roughness for the
approach/landing geometry without a LE device was
almost double the E10% penalty experienced when a
slat was used. However, on the other hand, the low
Reynolds number test results for the takeoff geometry
with slats extended do indicate maximum lift reductions
more than twice that expected based on the high
Reynolds number results for the takeoff geometry.
While some scatter band is certainly expected, deviations
this large are alarming, especially since the trend seen
with single-element airfoils was that low Reynolds

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

number tests typically produced lower maximum lift


penalties at a given nondimensional roughness height.
With regard to maximum lift penalties with takeoff
congurations not employing leading-edge devices, the
Ref. [82] low Reynolds number results do show the
sizable increase expected.
Overall, it remains somewhat of a question as to
whether these low Reynolds number test results showing
quite high maximum lift penalties due to small amounts
of LE roughness on multi-element takeoff-type geometries are meaningful and applicable. While the magnitude of the indicated penalties are clearly larger than
what might be anticipated based on the compilation of
measured penalties for single-element geometries, and
they cannot be reconciled with available high Reynolds
number results for takeoff geometries, it would be very
risky to assume that these results were not possibly
representative for some takeoff-type geometries, because
the magnitude of the maximum lift reductions are
equivalent to eliminating all of the typical minimum
takeoff speed margins (to stall). Hence, these results
should certainly provide added incentive to strictly
enforce the keep the wing (leading edge) clean
mandate, especially during takeoff.
4.2. Stall angle reductions
Reductions in the angle-of-attack at which stall occurs
caused by initial leading-edge ice accretions on wings are
of major signicance relative to the proper operation of
stall warning systems. Such reductions can either reduce

695

the margin from stall warning to actual stall to less than


desired and needed, or, worse yet, stall could occur
before the ight crew received any warning of an
impending stall.

4.2.1. Single-element lifting surfaces


As with the review of maximum lift reductions, the
most comprehensive set of test results on airfoil stall
angle reductions due to leading-edge roughness are
derived from the results published by Abbott and von
Doenhoff [66]. The compilation of the measured
reductions (or increases) in stall angle from this data
set are shown in Fig. 14 as a function of airfoil thickness
ratio, and vary from a maximum reduction of just over
61, to actual increases in stall angle of attack of slightly
over 31. The increases in stall angle at the lowest
thickness ratio (6%) are consistent with the improvements in maximum lift capability observed for these
designs. However, in the more commonly used thickness
ratio range from 9% up to perhaps as much as 15%,
aside from a very few exceptions, most of the congurations tested incurred a sizable reduction in stall angle,
ranging from about 11 up to as much as 561. At airfoil
thickness ratios above 15%, an even wider range of
results are seen. The overall broad data band observed
again serves as a reminder as to how the response of a
particular lifting surface to the addition of leading-edge
roughness is governed by the ow physics controlling
the stall process, and how this ow physics is altered by
the roughness. Clearly, many variations exist.

Fig. 14. Stall angle reductions caused by leading-edge roughness [66].

696

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 15. Roughness size effects on stall angle reductions at high Reynolds numbers.

Consistent with the changes observed in maximum lift


penalties with variations in nondimensional roughness
height, the stall angle reductions similarly change. But,
there is also another variable that comes into play in
establishing the magnitude of stall angle reductions for
practical aircraft wings and tails, and that is the lift curve
slope, which, in turn, is dependent upon the wing or tail
aspect ratio. The inuence of both of these variables can
be seen in the compilation of high Reynolds number test
measurements of stall angle reductions shown in Fig. 15
for geometries having thickness ratios ranging from 9%
to 15%. First, there is the variation with roughness
height for 2-D geometries other than Refs. [6669].
While the variation is substantial, it is less than the large
variations possible with just different airfoils. Then, with
regard to planform effects, it can be seen that the stall
angle reductions observed in the test of the 3-D tail
conguration are noticeably higher than the upper limit
of the Refs. [6669] results. It is important to note (see
Fig. 9) that this was not the case with the measured
maximum lift reductions. Hence, while the 2-D airfoil test
results are quite useful in showing the existence of a wide
range of possible reductions in stall angle due to small
leading-edge ice accretions, the magnitudes cannot be
directly used for real aircraft applications without rst
making appropriate adjustments for planform effects.
The corresponding collection of available and useable
low Reynolds number test results addressing stall angle

reductions on single-element lifting surfaces caused by


leading-edge roughnesses similar to initial and intercycle ice accretions are illustrated in Fig. 16. As can be
seen, this particular collection of test results is somewhat
of a mixed bag. On one hand, many of the 2-D low
Reynolds number test results do clearly fall below the
2-D high Reynolds number data band as would be
anticipated as a consequence of the generally lower
maximum lift penalties at these low Reynolds numbers.
However, on the other hand, there are also a few of the
2-D low Reynolds number test results which fall above
this band. And, while the test results for 3-D nite
aspect ratio congurations are by and large at the upper
limits of the low Reynolds number data, the expected
3-D effect (i.e., the consequences of reduced lift curve
slopes) is not as pronounced as seen at the higher
Reynolds numbers (in Fig. 15). All together, this
particular collection of low Reynolds number test results
are further evidence yet that any conclusions based on
low Reynolds number test results are most likely not
really very meaningful or useful for higher Reynolds
number applications.
4.2.2. Multi-element lifting surfaces
Very little test data exist at either high or low
Reynolds numbers to document stall angle reductions
caused by leading-edge roughness on the leading-edge
element of multi-element high-lift geometries, and the

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

697

Fig. 16. Roughness size effects on stall angle reductions at low Reynolds numbers.

Fig. 17. Roughness size effects on stall angle reductions for multi-element airfoil geometries.

little that there is was all obtained from 2-D testing.


These available results are presented in Fig. 17 where
they can be compared with the (high Reynolds number)
data bands established for single-element airfoils. It can
be seen that stall angle reductions observed for the
multi-element geometries, particularly at high Reynolds
numbers, tend to be somewhat larger than the corresponding single-element results for the same basic

(stowed) airfoil design. This trend would seem to be


logical since higher leading-edge velocities would be
experienced on a leading-edge device at maximum lift
conditions than exist on a single-element design [61]. It
would then seem to follow that the range of possible
reductions in angle-of-attack margin to stall that might
be incurred with multi-element airfoil high-lift designs
could be as large as that observed with single-element

698

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

designs. And, nally, it should also be remembered that


since all of the multi-element airfoil data was obtained
from 2-D testing, any absolute values extracted from
this limited database must be adjusted for planform (and
other conguration) effects before trying to apply them
to practical 3-D aircraft geometries.
4.3. Drag penalties
Increases in drag caused by initial leading-edge ice
accretions on wings and tails are of concern because they
can lead to potentially dangerous reductions in aircraft
climb and acceleration rates, aircraft range, and speed.
For example, accidents and incidents have occurred
when ight crews have used autopilots not augmented
by an autothrottle during exposure to icing, thereby
allowing the airspeed to bleed down to stall entry, often
without warning. In addition, knowledge of the potential magnitude of such drag increases will be critical in
determining whether aircraft performance monitoring,
as proposed in some quarters [84], can be a viable and
practical means of alerting ight crews to the presence of
these initial ice accretions. However, as will become
evident, there is a real scarcity of high-quality test data
available to enable a real credible documentation of
these drag increases since some of the (few) high
Reynolds number test programs investigating these
leading-edge roughness effects were focused strictly on
lifting characteristics.
4.3.1. Single-element lifting surfaces
As indicated earlier, many factors are involved in
determining the magnitude of drag penalties brought

about by the addition of small amounts of leading-edge


ice (roughness). In addition to the roughness drag itself,
adverse (i.e., forward) movement of the transition point
will contribute in cases where turbulent attachment line
conditions do not exist, as will excessive thickening of
the boundary layer or even separation onset over the
aft part of the lifting surface. And, of course, many
variations and combinations of these and other factors
can exist, resulting in a sizable range of possible
penalties, dependent upon the lifting surface geometry,
the prevailing ow conditions, test Reynolds number,
etc. Once again, the most comprehensive set of test
results available to demonstrate the wide range of drag
penalties possible is the early data provided by Abbott
and von Doenhoff [66]. However, when assessing these
results, it is important to keep in mind that the test
Reynolds number is (only) 6  106, and these are tests of
2-D airfoils. At this test Reynolds number (and tunnel
ow quality), transition movements due to the roughness are certainly involved. And, in considering 2-D
data, only changes to the basic 2-D parasite drag of the
lifting surface are involved. There are none of the
normal 3-D effects associated with realistic aircraft
lifting surfaces included.
In analyzing drag penalties from this and subsequent
data sets, two conditions are used. The rst is the
percentage increase in the minimum parasite drag level,
while the second is the percentage increase in drag at a
(higher) lift coefcient representative of ight at a speed
30% above the 1-g stall speed of the unroughened test
geometry. The compilation of percentage increases in
minimum parasite drag levels from the Abbott and
von Doenhoff data set [66] is shown in Fig. 18, where it

Fig. 18. Minimum parasite drag increase due to leading-edge roughness [66].

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

699

Fig. 19. Drag increase at 1:3 VS;1g due to leading-edge roughness [66].

can be seen that at this one particular test Reynolds


number and roughness height, increases in minimum
parasite drag level vary anywhere from about 50% up to
nearly 200%. The analogous drag penalties at the higher
lift-coefcient conditions, which are much more representative of realistic low speed ight conditions where
drag penalties due to ice accretion could be consequential, are displayed in Fig. 19. At this condition, the range
of drag penalties due to leading-edge roughness tend to
fall into a somewhat narrower (although still large)
band, varying from a little over 50% up to about 120%
for airfoil thickness ratios of 15% and less.
In addition to the relatively large and not-readily
correlated variations observed in the airfoil prole drag
penalties caused by small (i.e., k=cE4:6  104 ) initialice-accretion-like leading-edge roughness for quite a
variety of different airfoil geometries at 6 million
Reynolds number, there are, for the same small roughness, further signicant, quite diverse, and also notreadily correlated variations in these penalties as the
Reynolds number is both decreased and increased. The
largest source of data addressing these additional
variations are the Ref. [67] test results for 15 of the
airfoils used in Ref. [66] where these drag penalties were
determined at Reynolds numbers ranging from 0.7 to 6
million. Incidentally, of the 15 airfoils tested, other than
one 9% and one 18% thick designs, the thickness ratios
were either 12% or 15%. The measured increases in the
minimum parasite drag and the drag at the higher lift
condition are provided in Figs. 20 and 21, respectively,
for chord Reynolds numbers from 1 to 6 million. There
are some interesting differences between the trends seen

at the two conditions. Looking rst at the minimum


parasite drag penalties shown in Fig. 20, the range of
measured penalties for these 15 airfoil designs clearly
grows as the test Reynolds number is increased. And, it
can also be seen that although there are quite a few
congurations for which there is very little change in the
measured penalty between 2 and 6 million Reynolds
number, there are a couple where the penalty increases
noticeably as the Reynolds numbers is increased in this
range, and there are also some where the drag penalty
noticeably decreases. On the other hand, as can be seen
in Fig. 21, the range of drag penalties at the higher lift
condition clearly decreases with increasing Reynolds
number. And, also in contrast to the generally more
subdued changes seen between 2 and 6 million Reynolds
number with the minimum parasite drag penalties, there
are quite substantial variations in the drag penalties at
these condition for all except one of the test congurations as the Reynolds number is increased from 2 to
6 million. While the prevalent trend at this higher lift
conditions is for a substantial reduction in the drag
penalty as the Reynolds number is increased from 2 to
6 million, there were, however, four congurations,
including the 9% thick airfoil, where there was a very
noticeable increase in the drag penalty over this
Reynolds number range. What these results clearly
indicate is that test results obtained at chord Reynolds
numbers much below 56 million are likely to be very
misleading in terms of indicating likely results for higher
Reynolds number applications (X56 million).
The question remaining to be answered is whether
or not drag penalties caused by small leading-edge

700

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 20. Reynolds number effect on minimum parasite drag increase [67].

Fig. 21. Reynolds number effect on drag increase at 1:3 VS;1g [67].

roughness (i.e., initial ice accretions) at a test Reynolds


number of 56 million are appropriate for inferring the
drag penalties that would exist at even higher Reynolds
numbers. While there is only a very limited amount of
data available which addresses this issue, there are three
sets of test results which are quite revealing in this

regard. To start with, there are the results presented by


Ladson [68] for the 0012 airfoil, again using the same
roughness height (i.e., k=cE4:6  104 ) employed in
Refs. [66] and [67], with a test Reynolds number range
from 212 million. Next are the results presented by
Loftin and Bursnall [69] for a 9% thick airfoil, again

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

having the same roughness height (i.e., k=cE4:6  104 )


employed in [66] and [67], and where the test Reynolds
number was varied from 6 to 25 million. The third set of
useful results are those presented by Abbott and Turner
[70] for a 22% thick airfoil with three different small
roughness heights (k=cE0:6; 1.1, and 3.1  104) where
the test Reynolds number was varied from 6 to 26
million. Although this latter set of test results for a 22%
thick airfoil are not likely directly applicable to modern
aircraft (wings and/or tail) designs, the results do
provide an interesting link to the Ref. [67] results for
the 18% thick airfoil. And, even though the test results
obtained for the 9% thick airfoil [69] are also likely not
directly applicable to most modern wing designs, they
are most pertinent for tailplane icing concerns. These
three sets of test results, both in terms of the increases in
minimum parasite drag caused by the leading-edge
roughness applications, as well as the corresponding
drag penalties at the higher lift condition, are presented
in Figs. 22 and 23, respectively. It can be seen in Fig. 22
that the minimum parasite drag penalties for both the
0012 and 22% thick airfoils appear to have essentially
leveled out at 6 million Reynolds number, but that is not
the case for the 9% thick airfoil where the minimum
parasite drag penalty is reduced by more than half as the
Reynolds number is increased from 6 to 25 million. This
latter result was a bit unsuspected considering that the
Ref. [67] results for the 9% thick airfoil (see Fig. 20) had
indicated very little change in the penalty as the
Reynolds number was increased from 2 to 6 million.
Lastly, it is interesting to note that the drag penalty from
Ref. [68] for the 0012 airfoil actually falls below
the Ref. [66] data band at 6 million Reynolds number.

701

One possible explanation for this would be the


reduced extent of roughness coverage (i.e., 5% versus
8% chord).
Next, in examining the results for the higher lift
conditions as shown in Fig. 23, it can be seen that there
are some substantial and contrasting changes in the
indicated penalties due to leading-edge roughness for
these three airfoils as the Reynolds number is increased
above 6 million. Clearly standing out is the doubling of
the drag penalty for the 9% thick design at 15 million
Reynolds number. Curiously, this increased drag penalty has the appearance of being an extension of the
trend seen for the 9% thick airfoil in the Ref. [67] results
from 2 to 6 million Reynolds number (see Fig. 21).
Similarly, the overall reduction in the drag penalty
incurred with the 22% thick airfoil as the Reynolds
number is increased from 6 to 26 million also has the
appearance of being an extension of the trend seen from
Ref. [67] for the 18% thick design. However, on the
other hand, with the 12% thick airfoil, it can be seen
that the trend above 6 million Reynolds number is
obviously not an extension of the lower Reynolds
number test results, Hence, once again, the often-nonpredictability of these Reynolds number trends is clearly
illustrated. Consequently, until such time as additional
high Reynolds number data addressing these leadingedge roughness effects might become available, what
these three sets of test results, together with those from
Ref. [67], clearly indicate is that any use of test results
for assessing drag penalties caused by small initial ice
accretions obtained at Reynolds numbers much less than
full-scale ight values could very likely lead to erroneous
conclusions regarding ight characteristics.

Fig. 22. Reynolds number effects on minimum parasite drag penalty.

702

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 23. Reynolds number effects on drag increase at 1:3 VS;1g :

While it would seem to be quite obvious that the drag


penalties caused by leading-edge roughness would, in
almost all cases of interest, increase with increasing
roughness height, there is, unfortunately, clearly insufcient data available to make any kind of a meaningful
quantitative assessment of the magnitude of these
variations for a range of airfoil geometries. The only
known source of data directly addressing this issue are
the Ref. [70] test results for three 22% airfoil designs
where the roughness height (k=c) was varied from about
0.6  104 to 3.1  104. Test results were obtained at
26 million Reynolds number for all three airfoils, and
the Reynolds number was varied from 6 to 26 million for
one of them, the one just referred to in addressing
Reynolds number effects. The test results (i.e., drag
penalties) for this conguration with the three roughness
heights were shown in Figs. 22 and 23 where it can be
seen that there are only small increases in the minimum
parasite drag penalty over this limited range of (small)
roughness sizes, but the roughness size effects become
noticeably more pronounced at the higher lift condition,
even with the quite-small roughness heights utilized in
this test program. Incidentally, these same trends were
also evident in the results for the other two airfoils at
26 million Reynolds number. As indicated, the roughness sizes utilized in this investigation were quite small,
so there is no knowing what would have happened at
larger (nondimensional) roughness sizes of more practical interest, although the trend started at the higher lift
conditions is certainly ominous, at least for this airfoil
thickness. There are four other sets of test results

available [7274] for airfoils with thickness ratios


ranging from 12% to 15% where drag penalties
associated with larger (non-dimensional) roughness
sizes, varying from about 8  104 to over 50  104,
were assessed. Unfortunately, these are all low Reynolds
number (23.3 million) results, and only a single
roughness height was used for each. What these results
indicate are drag penalties near the lower limit of
the Ref. [67] band of results (for a k=cE4:6  104 and
Reynolds number of 6 million). In fact, the
result obtained with a 0012 airfoil [73] with the
largest roughness size (i.e., k=c > 50  104 ) actually
falls below the Ref. [67] band at the higher lift condition,
once again illustrating the danger of using low Reynolds
number test results for higher Reynolds number
applications.
4.3.2. Multi-element lifting surfaces
Fundamentally, it would be expected that the prole
drag penalties (percentagewise) caused by the existence
of leading-edge roughness on the leading-edge element
of multi-element high-lift geometries would be much less
than for single-element geometries. First, the basic drag
levels of the multi-element geometries are much higher
because of the coves, owing slots, etc. And, secondly,
the adverse effect of the roughness on the forward
element would likely be diminished somewhat by both
the presence of the downstream element. (i.e., pumping
up the forward element, and by having only a small
impact on the prole drag of the downstream elements)
Regrettably, however, there appears to be only one

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

known published source of test data which provides any


insight into the absolute magnitude of these drag
penalties on multi-element geometries, and, those results
somewhat unfortunately, are from a low Reynolds
number (2.6 million) investigation [74]. Be that as it
may, however, these low Reynolds number test results
do indicate that the percentage increases in drag for
multi-element (two- and three-element) geometries at
both minimum parasite drag conditions, and at speeds
30% above the 1-g stall speed of the unroughened
airfoil, are less than the single-element penalties by
factors of anywhere from 4.5 to at least 6.5, with the
reductions being greatest for the three-element geometries. While these reductions are of the general magnitude anticipated, it is impossible at this time to
accurately quantify what realistic increases in prole
drag might be like at high Reynolds number conditions.
4.4. Trailing-edge control surface characteristics
Changes in control surface effectiveness and characteristics caused by the roughnesses associated with
initial and inter-cycle ice accretions which can possibly
lead to aircraft control difculties clearly need to be
understood and accounted for. Regrettably, however,
there is very little data available to help enable this
understanding. The only known data presently available
are some elevator hinge moment measurements made
for three previously described (in relation to Fig. 11)
horizontal tail designs at relatively low Reynolds
numbers [77,78,81]. And, as previously indicated, only
one of these data sets [81] has results for the critical
trailing-edge-down elevator deections used for pushover maneuvers. Be that as it may, these results are
helpful in providing some good insights into possible
and/or likely control surface anomalies that might be
caused by small leading-edge roughnesses. However,
they also reinforce the concern that there are, in reality,
a wide range of possible effects depending on lifting
surface and control surface geometries and type, the
type of initial or inter-cycle ice accretion involved,
Reynolds number, etc.
Before addressing the lessons learned from these test
results, it is useful to dene the sign convention being
used. Herein, the rather standard convention adopted is
that trailing-edge-down elevator deections (i.e., opposing the download on the tail) are dened as being
positive, as are elevator hinge moments which tend to
rotate the elevator trailing-edge down. Conversely,
trailing-edge-up elevator deections (augmenting a
download) are dened as being negative, the same as
hinge moments which tend to rotate the elevator
trailing-edge up. Although this sign convention is, as
indicated, rather conventional, it is not universally used
as evidenced by the use of the opposite sign convention
for elevator hinge moments in Refs. [77,78].

703

After examining the three sets of available and


applicable test results addressing leading-edge roughness
effects on elevator hinge moment characteristics, there is
one inescapable conclusion. In the case of these three
geometries at a relatively low Reynolds number, it is
very clear that any really noteworthy changes are
typically tied very directly to noteworthy changes in
the tail stall characteristics brought about by these small
initial and/or inter-cycle ice accretions. And, the
converse is true as well, i.e., if the leading-edge roughness has little effect on the tail stall characteristics
(including the maximum lift level), then the elevator
hinge moments are impacted very little as well. This
latter situation is illustrated very well by the Refs. [77,78]
test results for the 441 swept tail design which, at the test
Reynolds number, had thin-airfoil-type stall characteristics in the unroughened state. In this case, the addition
of small (i.e. k=cE7:7  104 ) roughness to the LE had,
as seen previously in Fig. 11, almost no effect on the
maximum lift level (and stall characteristics). Consistent
with this, the elevator hinge moments changed very little
as well. Another lesson learned from these same test
results is that if the stall characteristics are quite mild,
i.e., no big drop-off in lift at the stall, then the elevator
hinge moments did not change dramatically either at the
stall. In this case, with elevator deections of 01 and
201, the hinge moments became more and more
positive as the tail incidence/download increased prior
to stall, but then just kind of attened out at and beyond
the stall.
A quite different set of stall characteristics and
resulting elevator hinge moment characteristics occurred
with the other two tailplane geometries examined, which
are both nominally 12% thick. These congurations are
the thickened 181 tailplane design from Refs. [77,78],
which was tested with and without the same small
leading-edge roughness (i.e., k=cE7:7  104 ) used
with the 441 swept tailplane, and the Twin Otter tail
geometry from Ref. [81], which was tested with and
without the somewhat larger and rougher inter-cycle ice
accretion. In both cases, and at all the elevator
deections evaluated (i.e., 01, 101, and 201 for the
181 tail, and 01, 101, 201, +101, and +14.21 for
the Twin Otter tail), the elevator hinge moments without
the leading-edge contamination became gradually more
positive (or less negative) as the tailplane incidence/
download was increased prior to stall (but not as steep
as with the 441 tail). Then, however, concurrent with the
stall, the rate of increase (toward more positive) becomes
noticeably greater. In fact, the increase is quite abrupt
with the Refs. [77,78] thickened 181 tailplane. This
abrupt change is consistent with the very abrupt/sharp
lift loss experienced by this conguration at stall. On the
other hand, the much less abrupt increase seen with
the Twin Otter elevator hinge moments goes with the
much tamer stall characteristics exhibited by this

704

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

conguration. Now, when the leading-edge contamination is added to both of these congurations, the
resulting elevator hinge moment characteristics look
much like those for the respective unroughened geometries except that the signicant increase in hinge
moments (in the positive direction) occurs at a lower
tail incidence/download as a consequence of the maximum lift penalty (i.e., lower stall angle) associated with
the addition of the leading-edge roughness, etc. Incidentally, this premature stall does result in an earlier change
in the sign of the elevator hinge moments from negative
to positive (i.e., control surface reversal) with positive
elevator deections.
Although these three sets of tailplane test results have
provided some good insights into some possible types
and degrees of changes to elevator hinge moment
characteristics that can be caused by initial, residual
and inter-cycle leading-edge ice accretions, two fundamental concerns still remain. One is that all three of the
experimental studies were carried out at Reynolds
numbers around 2 million, and we have previously seen
many instances where results obtained at this relatively
low Reynolds number are not representative for many,
more practical, higher Reynolds number applications.
Second, these three geometries only represent a small
sample of existing or likely tailplane (or wing) designs,
and hence other types and degrees of changes are likely.
As a case in point, the aforementioned very recent test
results obtained by Papadakis et al. [80] for an 8% thick
full-scale horizontal tail do show a different combination of characteristics. In the clean (unroughened) state,
some of the results look much like a thin lifting surface
design (like the 441 swept tailplane), with a maximum lift
coefcient below 1.0, and quite tame stall characteristics.
However, the elevator hinge moment characteristics at
5.1 million Reynolds number do not look like the
Refs. [77,78] 441 tailplane results at all. In fact, for
elevator deection ranging from 151 to +151, the
hinge moment variations more closely resemble something between the two nominally 12% thick designs just
addressed, with a very pronounced change in trend
toward being more positive (or less negative). However,
in this case, this change occurs a few degrees before
achieving maximum lift conditions rather than concurrent with it as was the case with the two 12% designs.
But, consistent with the test results for the two 12%
thick designs, when the 1.6 min ice shape casting was
added, this change in the slope of the elevator hinge
moments occurred at a 341 lower incidence angle than
occurred in the clean state, although it is not clear that
the stall angle was actually reduced that much.
So, one more time there is additional evidence available
to illustrate the dangers involved in generalizations,
and in trying to use low Reynolds number test results
to predict consequences at higher Reynolds number
conditions.

5. Effect of runback and ridge ice accretions


This category of ice accretions is known to occur in
either of two situations. First, typical runback ice
accretions occur when all of the impinging water
droplets are not evaporated by the heat produced by a
leading-edge anti-icing system, and the water which is
not evaporated runs back and subsequently freezes on
the cold, unheated surface just aft of the anti-icing
system coverage. This can occur either by design (to
minimize engine extraction losses) or as a consequence
of a very intense icing encounter. A wide variety of
runback-type ice accretion shapes, sizes, and roughnesses have been observed. In some cases, the water
which runs back freezes in the form of relatively smooth
rivulets, which are probably not overly detrimental to
the aerodynamic characteristics. On the other hand,
runback-type ice accretions can result in a spanwise
forward facing step (i.e., ridge) just aft of the ice
protection system coverage. These would be expected to
cause signicant degradations in the aerodynamic
characteristics, depending, of course, on their size, etc.
Accretions similar to those can also occur during icing
encounters involving larger-than-normal droplet sizes
which have more aft impingement limits.
The possibility of these ridge ice accretions
occurring due to droplet impingement aft of a de-icing
boot (or anti-icing system) has long been recognized.
Johnson in 1940 [85] pointed out that in certain cases,
ice has built up on the de-icer cap strips to such an
extent (that) no benet has been derived from the
operation of the de-icing boot. In fact, albeit at low
Reynolds number (1.6 million), his wind-tunnel test data
indicated greater maximum lift losses and drag increases
for these ridge ice accretions than occurred with the
full leading-edge ice shape (which, unfortunately, was
not described). These results were a cause for real
concern in those days as most de-icing boots at that time
only extended back to about 5% chord. Unfortunately,
however, very little attention was given to this potentially very serious problem area between then and the
time of the October 1994 American Eagle ATR-72
accident [1,2] where such accretions occurring in supercooled large droplet (SLD) conditions are thought to
have been the primary cause. The only known efforts
(somewhat) addressing this topic in this time frame were
the icing tunnel studies by Bowden in the 1950s [86], the
ight testing conducted in various environmental conditions with the University of Wyoming Beachcraft
Super King Air twin-turbo prop aircraft from 1978 to
1994 where some most noteworthy observations/results
from a number of large droplet icing encounters were
reported [8790], and the very limited wind-tunnel test
results reported by Welte et al. [91] in 1991. Sadly, in the
aftermath of this crash, more than half century after
Johnson [85] published his results, this lesson was

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

repeated. Ashenden et al. [92], in their low Reynolds


number wind tunnel tests of an airfoil with simulated ice
shapes, came to the same conclusion, i.e., that a freezing
drizzle ice shape consistent with de-icing boot operation
(i.e., ridge ice at 6.3% chord) resulted in a more
severe performance degradation than occurred without
de-icing boot operation.
Unfortunately, in spite of the potentially serious
consequences of such ridge ice accretions, there is a
real dearth of good quantitative data available for these
runback and ridge ice accretions. To start with, there
does not appear to be any really applicable quantitative
high Reynolds number data presently available for
single-element lifting surfaces, but, worse yet, there does
not appear to be any data at any Reynolds number for
this type of ice accretion on multi-element high-lift
geometries (for approach/landing). There are some low
Reynolds number test results available for singleelement lifting surfaces, but there are concerns that for
these runback and ridge ice accretions somewhat aft of
the LE, it would not be unrealistic to expect that there
could be some signicant differences in the consequences
between having laminar or transitional ow ahead of
these ridge ice shapes at low Reynolds number test
conditions versus having fully developed turbulent ow
(i.e., turbulent attachment line conditions) at high
Reynolds number ight conditions. Hence, the lack of
any good quantitative data at high Reynolds number
(ight) conditions for these runback and ridge ice
accretions represents a serious deciency in the presently
available database. This concern will become quite
apparent when some of the (large) magnitudes of the
aerodynamic penalties incurred with simulated ridge ice
shapes at low Reynolds numbers are seen.
5.1. Maximum lift reductions
5.1.1. Single-element lifting surfaces
There are several sets of low Reynolds number test
results available that provide some interesting insights
into the possibly quite-large aerodynamic consequences
of these spanwise runback and/or ridge ice accretions.
One of these is the previously mentioned data published
by Jacobs in 1932 [65], and the others are results
obtained by Lee and Bragg [9395], as well as Papadakis
and Gile-Loin [79], in the aftermath of the ATR-72
accident. Jacobs results were obtained with small
protuberances mounted at various chordwise locations
on a 0012 airfoil at 3.1 million Reynolds number,
whereas the Refs. [9395] results were obtained at 1.8
million Reynolds number with forward-facing quarterround shapes mounted at various chordwise locations
on four different airfoils with thickness ratios ranging
from 12% to 15%. One of these four airfoils was the
23012 conguration, selected ostensibly to be representative of (turboprop) commuter type aircraft (such as the

705

ATR-72, etc.). The other three are modied versions of


the 63A415 and 63A212 airfoils, and the NLF414
conguration. The Ref. [79] results were obtained with
four different protuberances at 2 million Reynolds
number using the same modied 63A212 airfoil design
(i.e., Twin Otter tail) investigated in Refs. [95]. There is
also a limited amount of data obtained at 1 million
Reynolds number. Jacobs results [65] were obtained
with protuberance heights (k=c) ranging from 4  104
to 125  104. The Refs. [93,94] results for the 23012
airfoil are for protuberance heights of 56  104,
83  104, and 139  104 but they also provided results
with a small roughness strip (k=c 14  104 ) for
comparative purposes. Results for the other three
airfoils used by Lee and Bragg [93,95] were obtained
at a single k=c 139  104 : Incidentally, for the Refs.
[93,95] results with simulated ridge ice accretions aft of
2% chord, a transition strip (k=c 7  104 ) was placed
ahead of the simulated ice shape at 2% chord. What this
strip/trip really simulates is uncertain. And, nally, with
the Ref. [79] results, protuberance heights (k=c) of
41  104, 107  104, and 115  104 were utilized.
The variation in indicated maximum lift penalty
caused by chordwise movements of small protuberances/roughnesses (i.e., k=cp20  104 ) for the 0012
airfoil in Ref. [65] and 14  104 for the 23012 airfoil in
Refs. [93,94] are illustrated in Fig. 24. At these small
protuberance/roughness heights, it can be seen that the
most critical location for these disturbances is indicated
to be very close to the LE. This trend for small
disturbance heights is consistent with the airfoil test
results presented by Morgan et al. [33] over a Reynolds
number range from 3 to 12 million where the
maximum lift penalty caused by a roughness strip
(k=cE4:5  104 ) extending form 3% to 5% chord
was only about 85% of that which occurred when the
roughness extended all the way to the LE. Incidentally,
for the record, even though test results were obtained at
12 million Reynolds number, turbulent attachment line
conditions would not have existed because of the (2-D)
conguration, low tunnel turbulence level, etc. What all
of these test results appear to indicate is, that for
quite small runback or ridge ice accretions (i.e.,
k=co20  104 ), the maximum lift penalties would be
less than those caused by initial leading-edge ice
accretions (roughness) of the same size.
Things change signicantly, however, as the size of the
protuberance or simulated ridge ice shape is increased,
both in terms of the magnitude of the maximum lift
penalty indicated, and the trend with chordwise location. Starting with the results for the 0012 and 23012
airfoils from Refs. [65] and [93,94] shown in Fig. 25 for
the somewhat larger (i.e., k=cX50  104 ) protuberances/simulated ridge ice sizes (believed to be plausible
for these types of accretions), some extremely large
maximum lift penalties are indicated, particularly for

706

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 24. Variation of maximum lift penalty with chordwise location of small disturbances.

Fig. 25. Variation of maximum lift penalty with size and chordwise location of simulated runback and ridge ice shapes.

simulated accretion locations somewhat aft of 5% chord


where runback and ridge ice accretions would be
expected to occur. Possible maximum lift reductions of
over 80% are indicated for the 23012 airfoil, more than
twice the maximum seen for the initial leading-edge ice
accretions addressed in the previous section. And, the
results for the 0012 airfoil are actually not that much

lower than those for the 23012 conguration either


where comparable results (in terms of disturbance height
and location) are available. Clearly, in this case, the
absence of data from Ref. [65] for disturbance locations
between 5% and 15% chord is a major limitation for
this application. However, be that as it may, we are
still left with two sets of potentially applicable

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

707

Fig. 26. Variation of maximum lift penalty with airfoil geometry and chordwise location of larger runback and ridge ice shapes.

low Reynolds number test results (for the 23012 and


0012 airfoils) which indicate that some potential runback and ridge ice accretions can lead to maximum lift
reductions of at least 70%, and possibly more than 80%.
This is downright scary. And, most interestingly, these
magnitudes are not that inconsistent with the 1982
incident experienced with the King Air aircraft [87]
where after encountering icing conditions which contained some large droplets, a pronounced buffet was
experienced at an indicated airspeed 40% above the
normal stall speed for this aircraft.
Penalties caused by the larger ridge ice simulations
(i.e., k=c 139  104 ) with the other three airfoil
geometries investigated by Lee and Bragg [93,95] at 1.8
million Reynolds number are also quite substantial, but
not as severe as seen with the 23012 and 0012 airfoil
geometries. The resulting penalties for these congurations are shown in Fig. 26 along with the corresponding
23012 test results. It can be seen that the reductions in
maximum lift capability are in the 5060% range in
contrast with the 80% plus range seen with the 23012
conguration. In searching for explanations for the
different characteristics, a number of possibilities exist.
To start with, from the information presented in Fig. 26,
it can be seen that the larger penalties are incurred by
congurations having the higher basic airfoil maximum
lift capability, and vice versa. It is also very likely that
some of differences in the penalties might be related to
differences in the levels of leading-edge suction pressures

on the clean airfoil, i.e., the higher the suction pressure


at maximum lift conditions, the greater the adverse
consequences of the simulated ridge ice accretions.
The opposite might be expected as well, and the NLF
airfoil (by design philosophy) and the thicker airfoil
could t in this category. Or, some of the differences
might be anomalies associated with the low test
Reynolds number. For example, it was seen earlier (in
Fig. 5) that there was little impact of increasing the test
Reynolds number from 1 to 6 million on the penalty
caused by leading-edge roughness on the 23012 airfoil.
However, on the other hand, for the NLF airfoil, which
has the lowest indicated penalties at this relatively low
Reynolds number, Addy and Chung [96] subsequently
demonstrated a 25% higher maximum lift level for this
same (clean) NLF airfoil at higher Reynolds numbers
(up to 10 million). So, it would be somewhat natural to
expect a higher penalty for this ridge ice simulation at
higher Reynolds numbers on this NLF airfoil.
Some additional instructive insights into factors which
can inuence the penalties incurred with runback/ridge
ice accretions (and simulations) are provided by the
Ref. [79] test results. For one, the results shown in
Fig. 27 help illustrate that there is another signicant
variable for these ridge ice accretions in addition to size,
location, lifting-surface geometry, Reynolds number,
etc., and that is the details of the aft geometry of the
forward-facing protuberance/simulations. This can be
seen from the differences in the resulting penalties

708

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 27. Variation in maximum lift penalty with geometry and chordwise location of runback and ridge ice shapes.

between the rst three protuberances which have very


similar heights. It is also interesting to note that the
corresponding Ref. [95] test results for this particular
airfoil obtained with the forward-facing quarter-round
protuberances do not indicate the greater penalty that
would be anticipated with the larger protuberance
height (i.e., k=c 139  104 versus 107115  104).
This could possibly be as a consequence of the
transition-xing technique employed in the Ref. [95]
results for all protuberance locations aft of 2% chord, or
it could just indicate the forward-facing quarter-round
simulation is not the most adverse possible simulation
for these runback/ridge-ice accretions. Unfortunately,
very little seems to be known (i.e., published) about the
actual ridge ice shapes encountered in ight other than
the King Air ight crew observations [90] that the
drizzle drops freeze as sharp feathers or glaze
nodules at or just beyond the de-icing equipment.
Another valuable insight provided by the Ref. [79] test
results has to do with the effect of varying test Reynolds
number on the indicated penalties. In this case, the
penalties indicated at 1 million Reynolds number for a
particular protuberance were signicantly lower than
those indicated at 2 million Reynolds number. For
example, with the larger protuberance heights, the
indicated penalties at 1 million Reynolds number were
as much as a third lower. But, with the smaller
protuberance height (i.e., k=c 41  104 ), some reductions were much more yet. However, probably the most
important lesson to be learned from the Ref. [79] test

results is that most of the maximum-lift-loss penalties


indicated for the larger runback/ridge ice accretion
simulations are substantially larger than those indicated
for/with a noticeably larger 45 min simulated glaze ice
accretion on this same airfoil at the same Reynolds
number.
As a consequence of the shear magnitude of the
potential penalties involved, as well as considering some
of the uncertainties remaining, some high Reynolds
number investigations of a range of potential runback/
ridge ice accretions are clearly warranted. Studies of
ways to ensure that these accretions do not occur, or are
avoided, would also certainly appear to be warranted.
Probably the only two ground test facilities suitable for
meaningful high Reynolds number experimental investigations of the effects of these runback/ridge ice
accretions on aircraft which are large enough to
have turbulent attachment line conditions in ight
are the NASA National Transonic Facility (NTF)
and the European Transonic Wind Tunnel (ETW).
Little is to be gained from further low Reynolds
number investigations of these potential effects
such as the experiment summarized in Ref. [97] which
was conducted at a (low) Reynolds number of 1.25
million with a quite large model of the 0012 airfoil in a
relatively small wind tunnel, and with sparsely located
small protuberances (i.e. k=c 35  104 ). And, realistically, any further 2-D tests at all, even at high
Reynolds numbers (especially with too-large models)
are destined to leave key questions unanswered because

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

of the leading-edge boundary-layer state simulation


issue.
5.2. Stall angle reductions
With the very large maximum lift reductions indicated
for these simulated runback/ridge ice accretions from
the available database, it follows naturally that there are
corresponding very large reductions in stall angle
indicated as well. For example, stall angle reductions
indicated for the 23012 airfoil in Refs. [93,94] with the
simulated ridge ice accretion placed at 10% chord range
from about 91 up to 131 as the height (k=c) of the ridge is
increased from 56  104 up to 139  104. As would be
anticipated, these reductions are about twice the
maximum seen from the 2-D results for initial leadingedge ice accretions. With the Ref. [79] test results for the
Twin Otter tail airfoil section, stall angle reductions of
up to 6 degrees are indicated. Although not as large as
the Refs. [93,94] test results for the 23012 airfoil, they are
consistent considering the reduced protuberance height
and less critical airfoil design. Again, these results clearly
point to the need for some new, well thought out studies
to assess their applicability because of the very large
magnitudes involved, and potentially very serious
consequences implied.
5.3. Drag penalties
Drag increases resulting from these simulated runback/ridge ice accretions are of obvious interest and
concern from a basic ight safety perspective, but also as
a possible means of alerting ight crews to the presence
of these ice accretions. Unfortunately, the data available
are not particularly well suited for assessing either of
these situations. To start with, there is always the
low Reynolds number concern. But, more than that,
with the Refs. [93,94] results there is the problem that,
with the very large maximum lift penalties incurred (with
the 23012 airfoil), the approaches used previously of
assessing increases in both the minimum parasite drag
level and the drag levels at a higher lift coefcient are not
possible. In reality, with these large maximum lift
reductions, the corresponding drag increases would
probably be of secondary importance. However, to give
some idea of the magnitude of the prole drag increases
incurred in this investigation, at a lift coefcient of 0.2,
which is about the maximum for which an assessment
can be made, the prole drag increases caused by the
simulated ridge ice accretions at 10% chord vary from
a factor of four for the smallest disturbance (i.e.,
k=c 56  104 ) to more than a tenfold increase for
the largest (i.e., k=c 139  104 ). In comparison, with
the Ref. [79] test results where some approximations
can be made because of the lower maximum-lift-loss
penalties, drag increases for the various simulations vary

709

anywhere from a factor of around two to nearly ten at


the selected higher lift condition. For Jacobs results
[65], the absence of any data for disturbance locations
between 5% and 15% chord, which is where such
runback/ridge ice accretions would be expected to occur,
remains a serious limitation for this application.
Incidentally, the King Air ight test results [8790] do
provide an additional perspective in terms of possible
total aircraft drag penalties which can be incurred as a
consequence of icing encounters with large droplets on
this size and type of aircraft. A number of cases were
reported having drag increases of a factor of over two,
up to a maximum of 2.3. However, there is no way of
breaking these increases down in terms of the contributions from the various aircraft components including all
of the unprotected areas. Consequently, probably the
only thing which can be said with any certainty at this
time is that a wide range of (large) drag increases is
anticipated with these runback/ridge ice accretions
depending upon the lifting surface geometry, ight
conditions, and runback/ridge ice geometries, locations,
severity, etc.

6. Effect of large in-ight ice accretions


Compared to the other categories of ice accretions
being addressed herein, far more effort for the past years
and decades has been focused on this category, both in
terms of developing methodology for predicting the
physical characteristics (i.e., size, shape, surface roughness, etc.) of these larger and often-more-irregularly
shaped ice accretions, and in trying to assess the
aerodynamic penalties associated with them. This
relatively very large emphasis on these larger ice
accretions, especially glaze ice accretions with horns,
etc., is quite natural and easy to understand if one makes
the thought-to-be obvious assumption that aerodynamic
penalties will be somehow proportional to the size and
grotesqueness of the ice accretion. However, how true
this is, especially relative to the runback/ridge ice
accretions, remains to be seen. One of the most
important objectives for a signicant portion of the
research which has gone on was to help dene critical
ice shape features, where critical ice shapes are dened
in Ref. [98] as those with ice accretion geometries and
features representative of that which can be produced
within the icing certication envelope that result in the
largest adverse effects on performance and handling
qualities over the applicable phases of ight of the
aircraft. Unfortunately, other than identifying glaze ice
accretions as the principal area to focus on, the authors
of Ref. [98] concluded (in late 2000) that existing
guidance material pertaining to the determination of
critical ice shapes was too general and did not provide a working denition of critical ice shape or

710

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

a description of engineering practices to be followed in


the determination of such shapes. Signicant additional
new research was suggested as a means of eventually
enabling these denitions to be made.
Although the results from many of the experimental
investigations conducted to date addressing the aerodynamic effects of these large in-ight ice accretions on
lifting surfaces are potentially applicable to both wings
and horizontal tails, a number of studies specically
focused on separately determining either wing or
horizontal tail characteristics have also been needed/
pursued as a consequence of some obvious as well as
some more subtle differences in geometries, operating
environment, requirements, etc. Four of the factors
which can lead to different ice accretions and consequences on horizontal tails compared to wings on any
given aircraft are the typically smaller surface size, the
corresponding reduced Reynolds number, the sometimes
nondimensionally thinner dening airfoil sections used
to ensure satisfactory high speed longitudinal control,
and the absence of a high-lift system. The difference in
physical size is important because it causes the (smaller)
tail to be a more efcient ice collector. And, any
reductions in nondimensional thickness ratio will further
enhance this. On the other hand, the reduced Reynolds
number, and possibly reduced thickness ratio, may in
some cases reduce the magnitude of some of the
aerodynamic penalties imposed by these large leadingedge ice accretions. However, probably the biggest
concern with tailplane icing is the potential loss of
longitudinal control which can result from premature
tailplane stall, elevator hinge moment anomalies, etc. A
number of studies have focussed strictly on this aspect
starting with the studies in the 1940s [99,100] addressing
uncontrollable pitching oscillations encountered on the
Vickers Viking aircraft with ice on the tailplane LE.
Other noteworthy studies specifically addressing tailplane icing phenomena include the work of the SwedishSoviet Working Group in the late 1970s [77,78], and the
results from the NASA/FAA Tailplane Icing Program
(TIP) started in the mid-1990s [101]. Similarly, there
have been a number of the experimental investigations
of the effects of these large ice accretions which are
much more pertinent to the wing characteristics. Clearly,
the studies of ice accretion effects on multi-element highlift airfoil congurations starting with the rst report
published by the Swedish-Soviet Working Group [74]
are in this category. Also, realistically, the results from
the numerous experimental studies focussed on the drag
penalties caused by these large ice accretions are much
more relevant for the wing (in terms of total aircraft
drag) because of the relatively much larger surface area.
Not surprisingly, with the much greater emphasis that
has been placed on studying the aerodynamic effects of
this category of larger ice accretions, there are more
varied sources of potentially applicable test results. Not

only are there a wide assortment of test results available


from conventional wind tunnels using simulated ice
accretions, but, there are also data available from ight
testing conducted in natural icing conditions, ight tests
carried out with simulated large ice accretions for both
research and certication programs, as well as from a lot
of testing in icing tunnels such as the Icing Research
Tunnel (IRT) at NASA Glenn. All of this additional
testing in ight and in icing tunnels has denitely
provided numerous quite unique, very insightful, and
useful perspectives on a number of the critical performance and ight control issues central to properly
assessing the range of aerodynamic effects which can be
caused by this category of larger in-ight ice accretions.
However, on the other hand, there are some inherent
technical shortcomings associated with each of these
additional test categories that signicantly limits the
amount of truly useful more-quantitative-type data that
can be extracted from the test results. Flight testing in
natural icing conditions provides some wonderful
examples of these pros and cons. One of the benets
derived from such testing has been in providing some
quite graphic examples (i.e., wake-up calls) illustrating just how severely these larger ice accretions can
impact aircraft performance and ight control characteristics. These include the following:
*

Johnson in 1940 [85] reported on a ight experience


with a Lockheed Electra wherein the aircraft was
iced to the extent that the indicated high speed with
all the power available to the pilot with carburetor
heat applied, was only 90 mph at 5000 ft. He also
reported that practically full power was required for
landing and only the pilots skill prevented serious
consequences in this particular case.
The aforementioned report by Morris in 1947 [99] of
uncontrollable pitching oscillations having occurred
during ight with the Vickers Viking aircraft when
there was ice on the tailplane LE.
In what appears to be the rst reported situation
where this type of testing was conducted with the
primary objective of determining somewhat quantitatively the effects of these larger ice accretions on
aircraft performance characteristics, Preston and
Blackman in 1948 [102] reported an 81% increase
in the parasite drag of a twin-engine airplane in level
cruising ight after 40 min in meteorological conditions which would at very most be considered
moderate icing by current standards.
Leckman in 1971 [103] reported a 34% increase in
stall speed having occurred on a Cessna Centurion
aircraft with just a 14 in rough ice accretion on the
wing LE. A signicant reduction in the rate of climb
with this ice accretion was also reported. Then,
following from the lessons learned from this testing
in natural icing conditions, Leckman pointed out

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

that elevator effectiveness in the landing are may


be impaired by ragged ice accumulations on the LE
of the horizontal stabilizer, and that resultant
airow disturbances at the LE could result in a
separated airow under the elevator and a corresponding loss in elevator effectiveness. He went on
to point out that these effects would be aggravated
by the increasing downwash angles associated with
larger ap deections, and concluded that as ice
accumulations become more severe, the wing ap
deections should be minimized as a precautions
against partial stabilizer stalling. This was 30 years
ago!
In the rst joint report published by the SwedishSoviet Working Group in 1977 [74], results from
ight testing the Antonov An-12 aircraft in natural
icing conditions are presented which indicated a 31%
reduction in the maximum lift capability, a 31
reduction in the stall angle, and a 50% reduction in
the lift-to-drag ratio, all caused by an ice deposit
shape on the LE of the wing having a maximum
thickness of 20 mm, which is described as only
moderate icing, since operational experiences show
that more severe icing conditions may take place in
practice. They also reported that the horizontal
ight speed of the TU-124 and TU-134 were reduced
by 4050 km/h with ice roughness of 68 mm formed
on the wing and tail units, and that the speed
decreased more sharply in the beginning of the icing
process. This led to the recommendations to never
ignore even light icing as it can be extremely
dangerous, and to always remember the main
general recommendation of timely activation of the
anti-icing system. They further pointed out that
for the approach icing case the test results indicated
a drag increase due to icing of such magnitude that in
the one-engine-out case a go-around may not be
possible within the limits specied in the Airworthiness requirements.
Other ight test results obtained with the An-12
[104], seemingly from about the same time period in
the early to mid 1970s, revealed a signicant
corruption of lateral controllability characteristics
at low airspeeds due to the ice accumulation on the
wing LE. Ice accreted on the LE of the wing caused
an aileron hinge-moment reversal and accompanying
snatching of the yoke when the trailing-edge-down
aileron deection reached only about one-third of the
full range at a condition (i.e., lift coefcient) far
from stall. And, this reversal occurred even earlier
(i.e., lower deection angles) at speeds closer to stall.
In the third joint report issued by the Swedish-Soviet
Working Group in 1985 [78], results were presented
from ight testing on the An-18, IL-18, An-24, and
Yak-40 aircraft at landing ap settings which showed
a denite lightening of (control) stick forces when ice

711

was allowed to build up on the LE of the horizontal


tail. In some of the cases, there was a relatively
sudden and sharp decrease and subsequent reversal
of the stick forces which occurred with very modest
ice accretions.
Results obtained from ight tests of the NASA Twin
Otter aircraft during 19831986 [105107] showed
total airplane drag penalties of up to 75%, and
increases in wing section drag up to 120%, after
ying through natural icing conditions.

While there are undoubtedly numerous additional


sources of data from ight testing in natural icing
conditions that are either not well publicized or are not
available (i.e., not in the public domain), the foregoing
chronology does appear to highlight most if not all of
the possible more adverse repercussions which can be
caused by these larger ice accretions.
Another very important benet derived from the ight
testing conducted in natural icing conditions is the
insight provided in terms of identifying the types of ice
accretions and some of the corresponding environmental
conditions in which they form that result in the most
serious degradations of performance and ight control
characteristics. These insights tend to fall into two
groups. First, there are the mostly very qualitative and
somewhat more obvious ones (today) provided by the
earlier (i.e., mostly pre-1980) ight testing. These include
Johnsons [85] observation that ice accreted at 401F
was so smooth that no detrimental effect could be
noticed in the airplane performance, and Leckmans
[103] conclusions that glaze (clear) type of ice formation is most adverse with respect to its effect on
aerodynamic drag and stall, and that double horn
glaze ice shapes make the drag contribution unusually
high. Other sources/insights that fall in this group
include the observations provided in the rst report by
the Swedish-Soviet Working Group [74] that in spite of
only half the thickness, the horn-shaped ice will be much
more detrimental to the aerodynamic characteristics
than the wedge-shaped ice, as well as those provided in
the third report [78] that the form of the ice on the
tailplane LE, and apparently its surface conditions
(smooth, rough, wavy) play an extremely important
role. Some of the test results provided in [78] indicated
that if horn shaped, the ice created a severe deterioration of the aircraft stability, but, when ice conformed to
the section shape, there was practically no inuence.
However, they also showed an example on another
aircraft where the wedge shape had a considerable
impact. As indicated, however, most of these insights/
observations should be (at least) somewhat self apparent. That is also true of some of the observations
provided in connection with the second group of later
(i.e., starting in the early 1980s) ight test results
obtained with the NASA Twin Otter aircraft. For

712

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

example, in Ref. [105], it is stated that glaze icing,


which generally occurs at total air temperatures just
below freezing, results in rough, irregular ice formations
with at or concave surfaces facing the airstream, and
that this type of icing causes the largest aircraft
performance penalties in terms of the increase in aircraft
drag. This is compared to rime icing which generally
occurs at lower total air temperatures below freezing,
and results in smoother, more pointed ice formations
facing the airstream, which cause much lower drag
penalties as compared to similar amounts of glaze ice.
But, in addition to these more qualitative ndings, the
Twin Otter test results in this second group do provide
some very instructive specics illustrating how some
relatively small differences in environmental conditions
can lead to some very signicant differences in the
effects on aircraft performance (i.e., total aircraft drag).
The summaries of the ight test results obtained with
the NASA Twin Otter [105107] contain three very
interesting sets of comparisons in terms of relating
aircraft drag increases to environmental conditions,
icing encounter durations, etc. (with the de-icing system
turned off). In the rst set [105], icing encounter specics
and the resulting drag increases are examined for two
glaze ice accretions and one rime ice accretion involving
three separate ights (in addition to the un-iced baseline
testing). Total temperatures for these encounters were
about 91C for the rime ice case, and about 31C and
21C for the glaze ice cases. This particular rime ice
encounter involved a relatively high LWC and a
relatively longer duration, making it a rather severe
rime ice encounter. And, even though one of the glaze
ice accretions was characterized on the basis of LWC
and encounter duration as being a relatively severe
(glaze) icing encounter, the amount of ice accreted
during the rime icing encounter was about twice that for
either of the glaze icing encounters. In spite of this
difference, the rime ice drag penalty was only 15%,
compared to 47% and 62% for the two glaze ice
accretions. Regarding the drag penalties for the two
glaze ice accretions, the larger penalty is associated with
the more severe icing encounter, but the magnitude of
the increases is nearly three times greater than might be
expected by just ratioing the accretion totals, indicating
a noticeably non-linear progression.
The second set of comparisons [106] is for what is
described as a severe glaze icing encounter, and two
mixed (i.e., mixed rime and glaze) icing encounters.
Incidentally, a stereo photography system was used to
document some of the in-ight ice shapes involved, but,
unfortunately, this documentation was not provided for
the glaze ice shape of (most) interest in this case. Total
temperatures for the three cases were 3.51C for the
glaze icing encounter, and 4.81C and 5.21C for the
two mixed icing encounters. As a consequence of
offsetting differences in LWC and icing encounter

durations, both of the mixed ice accretions had about


the same amount of ice accreted, and this was about
50% greater than with the glaze ice accretions. However,
despite this difference, the total airplane drag penalty
incurred with the glaze ice shape was double that
experienced with the two mixed ice accretions (i.e., 75%
versus 31% and 38%). The difference in measured wing
section drag penalties was even more dramatic yet. At a
representative condition, the wing section drag penalty
for the glaze ice of 120% is over six times greater than
that seen with one of the mixed ice accretions. Other
than the (less than 21C) total temperature difference, the
only other envisioned contributor to these very large
drag differences might conceivably be the higher
maximum droplet diameters present in the glaze icing
encounter (i.e., 41 mm versus 2730 mm). This possibility
remains an open issue.
Looking next at the third set of comparisons involving
Twin Otter ight test results [107], the comparison in
this case is between a moderate glaze icing condition and
what is described as a moderate-to-heavy mixed icing
condition. A 4.51C difference in temperature between
the two ights was what led to the formation of a glazetype ice accretion on one and a mixed-type ice accretion
on the other. With this difference in shapes, even though
the mixed icing encounter lasted 20% longer, and had a
LWC more than double that of the glaze icing
encounter, the glaze ice shape resulted in overall drag
increases 1520% higher than those incurred with the
mixed-type ice accretions. So, in all three of these
comparisons, a relatively small difference in ambient
temperature, when it resulted in the formation of a
glaze-type ice accretion was of paramount importance in
establishing appropriate drag penalties. There was also a
hint provided in the second set of comparisons that
larger droplet sizes (i.e., greater than 40 mm) might also
lead to ice shapes which cause even higher drag
penalties. Although this possibility would certainly seem
to be plausible, there just does not appear to be any
additional ight data available to really substantiate or
quantify this effect. As it is, the only other reported
results available for aircraft encounters with larger
droplet sizes [8790,108,109] are for situations where the
ice-protection systems were being operated during such
encounters. Hence, the insights provided relate primarily
to the effects of potential runback or ridge ice accretions
associated with these larger droplets, and/or the likely
larger inter-cycle ice accretions. Both sets of ight results
obtained with turboprop aircraft employing de-icing
systems, the aforementioned results obtained with the
Super King Air [8790] and the Ref. [108] results
obtained with an F-27, indicated (not surprisingly)
noticeably higher drag penalties associated with these
conditions. On the other hand, results obtained with a
Boeing 777 [109], which utilizes anti-icing on the wing,
indicated no noticeable effects on airplane performance

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

and handling qualities with the size of droplets


encountered, i.e., 5085 mm. Hence, in addition to
addressing the potentially very large adverse effects of
ridge-type ice accretions which might form in large
droplet encounters, additional attention also needs to be
focused on the potential consequences of the larger
inter-cycle ice accretions which would likely occur as
well with any de-icing concept.
Notwithstanding the valuable insights which have
been enabled by ight testing in natural icing conditions,
there are, as indicated previously, some inherent
limitations involved with this type of testing. Some
practical limitations were mentioned in Section 1 such as
cost, schedule, seasonal issues, and the risk involved
with nding appropriate (i.e., the most critical) icing
conditions. Clearly, the ight test results obtained with
the NASA Twin Otter aircraft [105107] showing that as
little as a 21C temperature difference can make a big
difference help put this risk in perspective. However, in
addition to these, there are some other very important
limitations as well which can be identied quite easily by
taking note of what is typically not reported from testing
conducted in natural icing conditions. Uppermost in this
grouping would be a good denition of the actual
accreted ice shape and surface conditions, a good
denition of the ice shape and surface conditions
existing subsequently when the actual performance and
ight control characteristics measurements are most
often made, and at least some reliable indication/
measurements of the increases in stall speed which
would be incurred. Typically, available descriptions of
the actual ice shapes accreted in natural icing conditions
and their surface condition/roughness characteristics are
based on a combination of visual observations and/or
photographs taken by the ight test crew, both of which
fall well short of providing a useful quantitative
description of the shape, let alone the surface condition/roughness characteristics. The only known exceptions to this (denitely inadequate) method of
dening the actual ice shapes were the utilization of
stereo photography during some of the Twin Otter ight
testing [106], and some reported cases such as those in
Ref. [78] when temperatures were cold enough that the
aircraft could be landed with the accreted ice shape
largely intact (i.e., not melted or shed) so as to permit
appropriate measurements, castings, etc. to be made.
However, even though the use of stereo photography did
enable a fairly good denition of some of the more 2-D
ice shape characteristics, there was still no adequate
description of surface conditions and small scale 3-D
features. Also, in this particular demonstration/application of the technology, denitions of the most critical
(i.e. glaze) ice shapes encountered were not obtained/
reported. And, with regard to making the necessary ice
shape, etc., measurements after landing, unfortunately
this only works with the much less bothersome rime ice

713

accretions which typically form at temperatures below


101C. For the more dangerous glaze ice accretions,
which form at temperatures just below freezing, these
accretions would likely be melted away in descending to
land (at typically warmer temperatures).
The analogous second typically missing ingredient
from reports of ight testing in natural icing conditions
is a good denition of the ice shape and surface
conditions existing when corresponding performance
and/or stability and control measurements/evaluations
are actually made. This requirement arises because very
often it is necessary to exit the natural icing conditions
before undertaking these tests, either for safety reasons,
and/or in order to nd suitable atmosphere conditions
for making meaningful (incremental) performance and
ight-control-related measurements/evaluations. However, in the time it takes to do this and prepare for the
desired testing, erosion/sublimation and even possibly
some shedding of the accreted ice can take place. There
have been a number of reports acknowledging such
changes having taken place. For example, in Ref. [78], it
was reported that some ice came off before starting
measurements. Occasional shedding of the ice accretion
in this time interval has also been acknowledged in other
reports as well, included cases involving loss of ice
caused by airloads and wing exure. Erosion or
sublimation of accreted ice shapes is probably much
more common, however, and although the geometric
changes may be more subtle, such changes can nevertheless potentially still lead to signicant differences in
their effects on aerodynamic characteristics. One common way in which this erosion or sublimation occurs is
associated with melting of the ice, and this can occur
either as a consequence of climbing above the icing
cloud to where sunlight can be an issue, or by
descending below the cloud to where warmer air exists.
Erosion caused by ice crystals has also been reported/
suggested [90]. One of the forms in which this erosion
has been observed to occur is a reduction in the surface
roughness. For example, it is reported in Ref. [106] that
conventional photographs taken during this time
indicate that some ice sublimation or erosion occurred,
and although the general ice shape remained, the
characteristic roughness was reduced. This is for a
situation when a little over 20 min was taken to exit the
icing cloud and take the performance measurements.
Although testing procedures have been adopted in some
cases [107] in an attempt to minimize this erosion, it is
hard to totally avoid. Hence, the typical result from
testing in natural icing conditions is a lack of a good
denition for both the actual accreted and the subsequent ice shapes and surface conditions, plus an
absence of any good data addressing what the consequences of the differences in ice shapes and surface
conditions might be. This is not an enviable state of
affairs, nor is it a good basis for determining the specics

714

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

for appropriate simulated ice shapes (and surface


conditions).
While all of the foregoing limitations discussed are
very important considerations when assessing the true
value of ight testing in natural icing conditions,
perhaps the most important shortcoming of this type
of testing has been, with only two known exceptions
[74,103], the inability to determine the stall speed
increases/penalties associated with the various ice
accretions formed in these conditions. Reasons for this,
primarily ight safety considerations, are most understandable. For example, oftentimes these icing encounters are experienced at relatively low altitudes where
adequate ground clearance for sensible stall testing is
clearly not available. Also, concerns regarding instrumentation status, visibility, etc., have been a factor, as
have been concerns with not really knowing what to
expect. One likely typical reaction to all these concerns
has been that the aircraft was not permitted to stall as
stated in Ref. [87]. Interestingly, in the two cases where
stall speed increases/penalties were reported, the indicated ice accretions were certainly not of the form that
would have been expected to be the most detrimental.
For the 34% increase in stall speed (i.e., 45% reduction
in the maximum lift achievable) reported in Ref. [103]
for the Cessna Centurion aircraft, the ice shape which
caused this was described as a 14 in rough ice accretions
on the wing LE, certainly not a large double-horned
glaze type ice accretion. Hence, the magnitude of this
particular penalty for such an ice accretion naturally
leads to some questions when subsequently examining
wind-tunnel measurements of maximum lift penalties
obtained with simulated ice shapes as to whether or not
the tunnel results are at all indicative of what can really
occur in ight with the thought-to-be much more critical
large glaze ice accretions. And, lastly, the extreme
reluctance of experienced ight test crews to stall aircraft
in natural icing conditions should certainly be an even
further incentive for others to avoid these potentially
very dangerous situations.
Turning next to ight testing conducted using
simulated large ice accretions, there are some different
pros and cons, but also some that are closely related
as well. Assuming that the simulations are appropriate
(and this is a major assumption), this type of testing
avoids a number of the limitations inherent with ight
testing in natural icing conditions, but it also introduces
some new concerns. On the positive side, ight testing
with simulated ice accretions avoids scheduling and
seasonal issues, concerns about having to nd the more
critical icing conditions, and not knowing what the ice
shape/surface conditions are for the measurements and
evaluations being made. Curiously, however, although
there have undoubtedly been many instances where
ight testing with simulated ice accretions has been a
part of the development and certication process for

specic aircraft addressing a variety of performance as


well as stability and control issues, the only published
results available are from investigations focused on
potential longitudinal stability and control problems
associated with tailplane icing. Such applications started
with studies conducted with the F-27 and F-28 in the
1970s [110], and include subsequent studies carried out
on a variety of Russian aircraft [78], as well as a number
of studies conducted utilizing the NASA Twin Otter
aircraft [107,111118]. Some of the lessons that have
been learned from these studies are that with these
congurations, almost any ice accretion on the LE of the
horizontal tail leads to some lightening of the stick
forces in a landing condition, and that the degradation is
related to the severity of the contamination. As an
example, results reported in Refs. [107,114,116] illustrated that the degradation got progressively worse in
going from an inter-cycle type ice accretion, to a failedboot type of accretion, and nally to a double-horned
glaze ice type accretion. A noteworthy aspect of the
results reported by the Swedish-Soviet Working Group
in Ref. [78] was that in two of three cases where the ight
results obtained with simulated ice shapes (referred to as
imitators in this case) could be compared to the
corresponding results obtained from testing in natural
icing conditions, the correlation was deemed unacceptable. This report should be discounted, however,
because in the one case when the correlation was judged
unacceptable and a comparison of the so-called
imitator geometry with the ice shape reported from
the natural icing encounter was provided, the imitator
was obviously not an adequate simulation. In fact,
whereas the natural icing shape was of the doublehorned type, the imitator did not have horns at all.
The consequence of this is that there is really no ight
data available to realistically assess the ability of
simulated ice shapes (no matter how dened) to reliably
produce the aerodynamic effects incurred with corresponding natural ice accretions, and there is no obvious
way to accomplish this validation either.
Similar to the situation existing with ight testing in
natural icing conditions, some of the important limitations associated with ight testing with simulated ice
accretions become apparent by taking note of what is
not reported. In this case, the most serious omissions are
the lack of any really quantitative-type data in terms of
maximum lift degradations, stall margin reductions, etc.
from any of the tailplane icing studies, and the absence
of any available data addressing aircraft/wing stall speed
or drag increases associated with large ice accretions.
Again, ight safety concerns would certainly seem to be
an important element in the absence of stall speed and
drag data, most notably serious concerns about taking
off with large horned ice shapes on the wing. Hence, this
category of testing has provided little additional
information to assist with the quantitative assessment

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

of the various aerodynamic effects caused by these larger


ice accretions other than to establish that the doublehorned glaze ice accretions were more detrimental from
a tailplane stall perspective than smaller less obtrusive
accretions with the tailplane designs on the aircraft tested.
In a very similar manner, numerous tests conducted in
icing tunnels have also enabled/provided a number of
most helpful insights into a number of factors critical in
establishing the magnitude of possible adverse aerodynamic effects/penalties caused by larger ice accretions.
However, once again, there are likewise some unique
and inherent limitations associated with this kind of
testing as well. Applicable results from a number of
these investigations are reported in Refs. [119132]. The
earliest of these appears to be the research conducted by
Gray and von Glahn in the 1950s and early 1960s [119
122] utilizing a number of different airfoils. Following
this, there appears to have been somewhat of a hiatus in
this type of research until the early 1980s when it started
again in earnest [123132]. Some of most important
insights gained from the results produced by all this
testing, as well as some of the important utilizations of
these results, include the following:
*

The results from such testing in the form of the actual


ice accretions observed have been the principal
source of guidance in dening the simulated ice
shapes used in subsequent ight tests and (conventional) wind tunnel testing to evaluate the various
aerodynamic effects of these large ice accretions
on both single-element and multi-element-high-lift
airfoils.
This testing has enabled assessments to be made of
the effectiveness of these intended-to-be simulations
of the actual icing tunnel accretions in duplicating the
drag penalties caused by the actual shapes. Unfortunately, in the only known (published) study addressing this issue [123] conducted nearly 20 years ago,
the results were not exactly encouraging. In this
particular study for a single-element airfoil, two rime
ice and two glaze ice accretions were addressed, and
both smooth and subsequently roughened (with an
abrasive) versions of each were evaluated. For one of
the rime ice accretions, the roughened simulation
produced results very similar to those seen with the
actual ice shape. In this case, results obtained with
the smooth version were not at all representative.
However, with the second rime ice simulation and the
two more critical glaze ice simulations, none of the
results produced by the simulations (either smooth or
roughened) were satisfactory at the lower angles of
attack where drag increases would be of consequence. Clearly, this issue is most denitely worthy of
more attention.
Test results obtained in icing tunnels, most specically the IRT at NASA Glenn, are the largest source

715

of (somewhat quantitative) data available addressing


the range of potential drag penalties that can be
incurred with these larger ice accretions. For
example, consistent with the previously discussed
results from the ight testing conducted with the
NASA Twin Otter aircraft [105107], results from a
number of investigations conducted in the IRT have
demonstrated that the ice shapes which cause the
largest drag penalties on single-element lifting surfaces are the glaze ice accretions formed at (total)
temperatures just below freezing conditions. Certainly amongst the rst to come to this determination
were Gray and von Glahn in 1953 [119] who reported
that glaze ice formations on the upper surface near
the LE of the airfoil caused large and rapid increases
in drag, especially at datum (total) air temperatures
approaching 321F and in the presence of high rates of
water catch. They compared these results with ice
formations at lower temperatures (rime ice) did not
appreciably increase the drag coefcient over the
initial (standard roughness) drag coefcient. These
results were obtained with a NACA 651-212 airfoil.
Similar results have been observed in a number of
subsequent investigations using a variety of different
airfoils [122,124,125,129132]. One of the more
graphic examples of this (localized) occurrence of
high drag penalties at temperatures just below
freezing was reported by Olsen et al. [125]. Their
results obtained with a NACA 0012 airfoil are
depicted in Fig. 28, where it can be seen that at the
higher speed condition, the rapid increase in drag
penalty as temperatures approach freezing conditions
coincides with the upper ice horn becoming noticeably more vertical. The same correlation also held
true at the lower speed condition. In addition, it is
signicant to note (from Fig. 28) that even though
the product of the airspeed, LWC, and the time
(duration) of the icing encounter was held constant
between these two cases, the ice accretion formed at
the higher speed condition resulted in noticeably
higher drag penalties. Although not a factor in the
results shown in Fig. 28 where the angle of attack was
held constant (at 41) throughout, higher speeds also
cause even more detrimental ice accretions because
ice is accreted at lower angles of attack where more
of the ice accretion occurs on the upper (suction)
surface.
Other noteworthy results/insights obtained from tests
conducted in the IRT regarding possible drag
penalties associated with these larger ice shapes are
that the penalties typically increase with increasing
LWC (as might be expected), and, adding credence to
the trends indicated by the Twin Otter ight test
results pertaining to droplet size effects, the available
data from the IRT [125,132] also indicate that these
penalties do increase as the droplet size increases,

716

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 28. Ice accretion temperature effect on parasite drag penalty for NACA 0012 Airfoil in IRT [125].

although the largest droplet sizes investigated (i.e.,


40 mm) were small compared to what we now
consider to be large droplets. Interestingly, Olsen
et al. [125] observed that these droplet size effects
appear to be greatest whenever the temperature effect
is greatest, i.e., with the most detrimental glaze ice
shapes. This observation should not be overlooked.
Although a large majority of the testing conducted to
date in the IRT has focused on determining the drag
penalties associated with a wide variety of large ice
accretions, the much more limited amount of data
pertaining to the corresponding effects on maximum
lift capabilities are nonetheless quite informative. For
example, results obtained for single-element airfoils
by Berkowitz et al for a Boeing 737-200 wing section
[126128], and by Addy [132] for a 14% thick GA
NLF airfoil [132], both demonstrate that the largest
reductions in maximum lift capability with these
types of airfoils are also caused by the glaze ice
accretions formed at temperatures just below freezing
conditions. One set of representative results from
Ref. [132] for the GA NLF airfoil are shown in
Fig. 29. However, in contrast with these results, other
test results reported in Ref. [132] for a 8.7% thick
business jet airfoil with a relatively low basic
maximum lift capability (B1.1) do not show as much
of a temperature effect (see Fig. 30), once again

illustrating the inadvisability of generalizing. Incidentally, while the trends indicated by the results
shown in Figs. 29 and 30 are thought to be
appropriate and meaningful, the absolute magnitude
of the penalties indicated may not be for a number of
reasons, not the least of which are doubts about the
appropriateness of the maximum lift levels indicated
for the clean (un-iced) congurations.
An additional noteworthy aspect of the Ref. [132]
results presented in Figs. 29 and 30 for both the GA
NLF and business jet airfoils is that the penalty
caused by the ice accretion in the rst 12 min
represents a signicant part of the penalty eventually
incurred after 22 min. This should not be surprising,
however, in light of the magnitude of the penalties
(seen previously) that are caused by initial leadingedge (roughness) ice accretions.
A further benet derived from testing in icing tunnels
is the number of very helpful and thought-provoking
insights provided relative to the effects of ice
accretions on the multi-element high-lift geometries
used on many aircraft types during takeoff and
landing. A good example of this is the data reported
by Potapczuk et al. [126128], illustrating types of ice
accretions, areas where they can form when testing in
an icing tunnel, and the corresponding effects of
these accretions on the aerodynamic characteristics

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

717

Fig. 29. Ice accretion temperature and duration effects on maximum lift penalty for GA NLF airfoil in IRT [132].

Fig. 30. Ice accretion temperature and duration effects on maximum lift penalty for business jet airfoil in IRT [132].

(lift, drag, and pitching moments) of these designs.


What these results revealed was that once again the
largest reduction in maximum lift level was caused by

ice accretions formed at temperatures just below


freezing conditions. However, in this case, the
temperature effect on drag penalties incurred was,

718

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

in general, not nearly as pronounced as that seen with


single-element airfoils. One interesting phenomena
encountered during this particular investigation was
a ow separation which occurred on the upper
surface of the ap over a wide range of angles of
attack with the ap deected 151 that was caused by
ice which formed around the LE of the ap. The
result was a signicant loss in lift and increase in
drag. It remains to be seen, however, as to whether
these results are truly meaningful or not. That is
because the real detrimental ice accretions on the ap
LEs were formed at incidence angles that are perhaps
too high to be really representative of likely ight
conditions (in icing situations).
In addition to the previously mentioned (in Section 2.1)
concern about the accuracy of simulation/scaling
techniques used when testing multi-element lifting surface geometries, there are some additional concerns
associated with testing in icing tunnels that also need to
be taken into account in order to put test results
obtained (or not obtained) from these facilities into
proper perspective. One large concern has been regarding tunnel wall interference effects contaminating test
results obtained with the relatively large models often
used (mostly in the past) in these facilities in order to
limit some of the uncertainties associated with scaling
techniques, and to achieve more meaningful Reynolds
numbers. Examples of this situation are the test results
from Refs. [119124] where model airfoil chords ranging
from 1.36 m up to over 2.4 m were used in the IRT,
which has a cross section of 1.82  2.74 m2. Clearly,
excessive wall interference would limit the usefullness of
these test results, especially at higher angles of attack.
In the next group of studies conducted in the IRT
addressing aerodynamic effects of these larger ice
accretions [125130], much smaller models with chord
lengths of about 0.46 m [126128] and 0.53 m
[125,129,130] were used. However, while the use of
these smaller models greatly alleviated concerns about
wall interference, the Reynolds number resulting from
the use of these smaller models in an atmospheric tunnel
such as the IRT is certainly lower than desired/needed.
And, if a higher tunnel velocity is used to raise the
Reynolds number, then compressibility effects contaminate the test results, especially for assessing maximum
lift penalties. The chord length of about 0.9 m (36 in)
used in the most recent test results reported by Addy
[131133] could be the best compromise, but even in this
case, concerns still remain with regard to wall effects and
compressibility effects at maximum lift conditions, and
Reynolds number effects at all conditions. Hence, it
appears that when testing for a range of aerodynamic
effects in an icing tunnel such as the IRT, there is a
choice of poisons involved if trying to extract
quantitative results, i.e., having the data contaminated

by either uncorrectable tunnel wall interference effects,


having an inadequate test Reynolds number, and/or by
having inappropriate compressibility effects. And, there
are always the quite-high freestream disturbance levels
which exist in these facilities due to the presence of the
spray rig, etc., in the circuit, as well as questions
regarding the accuracy of the scaling methods used.
Another limitation often mentioned relative to trying to
use test results from icing tunnels for reliable representation of maximum lift characteristics is the nonuniformity of the icing cloud over the span of the
model, and the corresponding limitations of force
balance measurements of lift, etc. This would be much
less of an issue for the measurement of section drag
characteristics.
At this point, the numerous lessons learned from
ight tests as well as tests in icing tunnels such as the
IRT addressing the aerodynamic effects caused by these
larger ice shapes, plus the insights gained from the
preceding reviews of the various effects caused by initial
in-ight leading-edge ice accretions as well as runback
and ridge ice accretions, have provided a solid foundation for effectively scrutinizing the total database
currently available for quantifying some of the more
critical consequences that can be caused by this category
of larger ice shapes, and for identifying important
deciencies in the current database needing attention.
Some of the more important lessons learned that should
be most applicable and helpful in this particular
assessment include the following:
*

Leading-edge ice accretions can have very different


effects on thinner versus thicker lifting surfaces,
particularly with regard to maximum lift characteristics/penalties.
Low Reynolds number test results can often be quite
misleading, mostly in a non-conservative manner,
usually due to Reynolds number effects on the uniced baseline geometry.
Uncorrectable tunnel wall effects and/or unrealistic
compressibility effects can also lead to erroneous
indications.
Roughness effects can be very signicant with many
geometries, usually in a harmful way, but sometimes
also in a helpful manner (for thinner geometries).
Larger protuberances (perhaps analogous to the
horns which occur with glaze ice accretions) tend to
be more detrimental to maximum lift characteristics
when they are located somewhat aft of the LE.
For many applications, the most detrimental large ice
accretions appear to be the glaze ice shapes which
form at temperatures just below freezing, and high
LWC, larger droplet sizes, and higher speeds tend to
accentuate this effect even further. In this situation,
small differences in temperature can lead to signicant differences in the consequences.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767


*

719

There does not yet appear to be a good understanding of just what constitutes an adequate or
effective simulated ice shape.
Ice accretions which form on some of the downstream elements of multi-element high-lift geometries
in an icing tunnel, especially on the LE of the main
element behind an extended slat or Krueger ap, may
well not be representative of what occurs during
ight in natural icing conditions.

6.1. Maximum lift reductions


6.1.1. Single-element lifting surfaces
Including three very recent additions, there are now
six known sources of potentially applicable high
Reynolds number data available to provide some
guidance for quantifying the reductions in maximum
lift capability that can be incurred by this broad class of
larger ice accretions. In chronological order, they are as
follows:
*

The previously mentioned ight test results for a


Cessna Centurion aircraft [103] wherein a 14 in rough
ice accretion on the wing LE caused a 34% increase
in stall speed (i.e., a 45% reduction in the maximum
lift level achievable).
The also-previously mentioned ight test results for
an An-12 aircraft [74] where an ice accretion
consistent with a moderate icing encounter led to
a 31% reduction in the achievable maximum lift
level.
Test results from the LTPT obtained over a range of
Reynolds numbers for a 11% thick airfoil with a
somewhat generic glaze-type ice accretion [33].
These results are particularly useful because there is a
direct comparison made with corresponding measurements obtained with simulated leading-edge frost
(i.e., roughness).
Test results obtained from the LTPT as part of the
NASA Modern Airfoils (ice accretion) program for
the previously discussed GA NLF 14% thick airfoil
over a range of Reynolds numbers and Mach
numbers with various simulated ice shapes [96,132].
These results also enable a comparison to be made
with results obtained from testing in the IRT with
ostensibly the same ice accretions.
Further test results obtained from the LTPT as part
of the NASA Modern Airfoils program, this time for
the previously discussed 8.7% thick business jet
airfoil with various simulated ice shapes [133].
Similarly, a comparison with supposedly corresponding results from the IRT can be made as well.
Results obtained from a test of a 8% thick full-scale
business jet T-tail model with a range of simulated ice
shapes conducted in the NASA 40  80 ft2 Wind
Tunnel [80].

Fig. 31. An-12 ight test ice accretion [74].

So, the available database consists of two ight test


results for not-the-most-critical ice accretions, two sets
of conventional wind tunnel test results for relatively
thin airfoils (i.e., 8% and 8.7% thick), and two sets for
thicker airfoils (11% and 14% thick). Looking rst at
the two available ight test measurements, these results
are helpful in terms of trying to establish a lower limit
for realistic maximum lift reductions that can be
incurred in ight with critical glaze ice accretions (on
thicker lifting surfaces). Certainly, the 45% reduction
experienced on the Cessna Centurion for just a 14 in
rough ice accretion on the wing LE [103] is most useful
in this regard. Similarly, the 31% reduction reported for
the An-12 [74] with the ice shape described (see Fig. 31)
is also pertinent, especially when the type and size of ice
accretion involved are considered. By comparing this
relatively small (less than an inch) ice accretion with the
ice shapes illustrated in Figs. 28 and 29, it can be seen
that the An-12 ice shape and size are indicative of an
accretion developed in a relatively short time duration at
quite low temperatures (i.e., well below freezing conditions). That being the case, the corresponding penalty
associated with a worst case (most critical) ice accretion
would be expected to be much larger than 31%. Hence,
considering these two available examples, it should not
be unrealistic to expect that the lower limit for maximum
lift reductions caused by the most critical ice accretions
on thicker wings at ight conditions would be (as a
minimum) close to 50%.
Considering next the two sets of high Reynolds
number test results from the LTPT obtained for thicker
airfoils using simulated large ice shapes, there are again
some very useful results/insights provided, but there are
also some uncertainties remaining relative to just how
well these particular results might represent the penalties
that would/could be experienced in ight with the most
critical of ice shapes. Starting with the investigation
involving the testing of a generic glaze ice simulation
on the 11% thick airfoil [33], the ice accretion
simulated was derived from IRT measurements for
a 0012 airfoil [125], but the orientation of this accretion

720

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 32. Maximum lift penalty for simulated glaze ice accretion [33].

on the 11% thick airfoil LE was modied somewhat.


The Ref. [125] ice accretion (and associated accretion
conditions) along with the simulation selected for the
LTPT test are shown in Fig. 32, together with the
associated test results. Also shown for comparative
purposes are the previously discussed penalties caused
by just LE roughness (k=c 4:5  104 ) on this same
airfoil. Looking rst at the Ref. [125] ice accretion,
although the LWC is signicant, the remaining parameters, notably the accretion temperature, duration/
time, speed, angle of attack, and droplet size, are not
quite what would be anticipated for a most critical ice
accretion condition. However, how all this is impacted
by the subsequent reorientation/rotation of this (simulated) ice shape on the 11% thick airfoil that moves the
upper horn more toward the upper surface of the airfoil
is a big unknown. Whether this reorientation/rotation
would simulate a reduced accretion angle of attack and/
or a (more critical) higher temperature (or something
else) is certainly debatable. There is also an issue as to
how well the simulated (i.e., smoothed) shape with an
extremely coarse grit added that was selected for this
study actually represents the actual ice accretion with its
3-D roughness. And, lastly, there is the ever-present
concern that inadequacies in the tunnel sidewall
boundary layer control capability could be masking
some characteristics (i.e., such as Reynolds number
effects).
Notwithstanding all of these open issues, the test
results obtained are still instructive in providing some
insights into the magnitude of penalties that can be
experienced with these larger ice shapes. As can be seen
in Fig. 32, the maximum lift penalties incurred in this

case with this particular ice accretion are only about 40%,
and do not vary much with Reynolds number over the
range tested. Relative to the corresponding penalties
incurred with the small leading-edge roughness on this
same airfoil, it can be seen that this 40% reduction is about
half again as much as that experienced with the roughness.
The second set of high Reynolds number test results
from the LTPT addressing the effects of simulated ice
accretions on thicker airfoils, this time for the 14% thick
GA NLF airfoil [96,132], are presented in Fig. 33.
Results are shown for both 6 and 22.5 min ice accretions,
and most importantly, results are shown for both
smoothed (2-D) ice accretion simulations and for
simulations having the actual (IRT measured) 3-D
roughness characteristics represented. Also included
are some comparable results from the IRT for the two
(3-D) ice accretions. Important lessons to be retained
from the test results depicted in Fig. 33 include the
following:
*

Maximum lift reductions of around 50% are


experienced for the 22.5 min ice accretion having
the 3-D roughness characteristics represented.
It is important to mimic the roughness characteristics
of these larger ice accretions when dening the
simulated ice shapes to be used for either ight
testing or testing in conventional wind tunnels.
Penalties indicated from the use of smoothed
simulations are noticeably lower than those obtained
with the 3-D roughness simulated. Unfortunately, it
is not known how well the typically used distributed
roughness would represent the actual 3-D roughness
effects.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

721

Fig. 33. Maximum lift penalty for actual and simulated glaze ice accretions on GA NLF airfoil [96,132].

In this case, indicated Reynolds number effects on


the maximum lift penalties are small over the range of
test Reynolds numbers covered. However, as indicated
previously in Section 5, the maximum lift level
attained in this case for the baseline un-iced airfoil
is 25% higher than that achieved at 1.8 million
Reynolds number for the same airfoil as reported in
Ref. [93]. Interestingly, however, the iced maximum
lift levels for similar shapes are remarkably similar.
The (percentage) maximum lift penalties indicated
from the IRT tests for the two (rough) ice accretions
are similar to those measured in the LTPT with the
3-D roughness simulated, even though the absolute
values of maximum lift in all cases are somewhat
different.

However, once again, it is necessary to consider some


of the limitations inherent with this set of LTPT test
results for the GA NLF airfoil in order to place the
foregoing observations in proper perspective relative to
just how representative the indicated penalties might be
of those that would be experienced in ight with the
most critical ice accretions. In this case, the concerns are
primarily with regard to the size of the model used in the
LTPT, and the ice accretion conditions utilized. Rather
than the more commonly accepted size having a 2224 in
chord (p0.6 m), such as used in the Ref. [33] investigation, a 36 in chord model was used in this instance,
causing the test results to be contaminated with
uncorrectable wall interference and nonrepresentative

compressibility effects. Perhaps the most obvious consequence of the excessive wall interference existing with
this size model is the very early (nonrepresentative)
separation present on the upper surface of the un-iced
(baseline) airfoil approaching the TE. Even prior to
reaching 41 angle of attack, this separation appears to
exist over the last 20% of the chord. This is certainly
indicative of a serious ow problem in the sidewall
juncture region. Hence, it is not likely that either the
indicated baseline or the iced airfoil maximum lift levels
and variations with Reynolds number are really
representative. How this would impact the magnitude
and variation of the indicated penalties is, unfortunately, not known.
With regard to the ice accretion conditions selected
for this study, the main areas of concern relative to not
having the most critical type of ice accretions are the
relatively low LWC, droplet size, and ight velocity.
The ambient temperature is also a little low as well.
Consequently, considering all of the concerns,
limitations, and uncertainties associated with the only
two available sets of high Reynolds number wind
tunnel results for thicker airfoils, it is not really possible
to reliably predict the maximum lift penalties likely
to be encountered in ight under the most critical
icing conditions, but they are probably in excess
of 50%. Clearly, some new well focused research is
needed to provide more insights into just how large
might these penalties be that could be encountered in
ight.

722

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

A somewhat analogous situation also seems to exist


relative to trying to use the two sets of high Reynolds
number wind tunnel test results available for large ice
accretions on thinner lifting surfaces to estimate
penalties likely to be encountered with these surfaces
under the most critical icing conditions. That is because
once again, in spite of the fact that there are some very
useful insights enabled by both sets of test results, there
are, unfortunately, aspects of each that tend to limit
their general applicability. Starting with some of the
results obtained and provided by Addy [133] from the
test of the 8.7% thick business jet airfoil in the LTPT,
representative maximum lift penalties for three different
ice accretions simulations are shown in Fig. 34. The
three ice shapes simulated were castings of 2 and
22.5 min accretions formed in the IRT at a total
temperature just below freezing (i.e., glaze icing conditions), and a casting of a 16.7 min rime/intermediatetype accretion formed at 111C (11.71F) in the IRT.
Corresponding results from the tests of this airfoil in the
IRT with the 2 and 16.7 min accretions are also provided
in Fig. 34 for comparative purposes. Important points to
be noted from these results are as follows:
*

There is a very signicant Reynolds number effect on


the indicated penalties, with the lower Reynolds
number results obtained at 3.5 million being well
below those obtained at the higher Reynolds
numbers. This effect, which is most noticeable with
the two shorter duration accretions, is due entirely to
changes on the baseline un-iced airfoil.

The largest penalty measured with this relatively thin


airfoil was just over 50% with the 22.5 min accretion
at a total temperature just below freezing.
Ambient temperature is a much more important
factor in establishing the magnitude of these penalties
than the duration of the icing encounter, as evidenced
by the similarity of the results for the 2 and 16.7 min
accretions, and the big difference between the 16.7
and 22.5 min accretions.
Maximum lift penalties (percentagewise) indicated
from the IRT tests for the two (smaller) ice accretions
tested there are again similar to those measured in the
LTPT, but, also once again, the absolute values of
maximum lift differ signicantly.

When trying to put these results into perspective,


however, it is necessary to keep in mind that these
results were also obtained with a 36 in chord model in
the LTPT, and, as a consequence, are most likely
somewhat contaminated with uncorrectable wall interference effects like the previously discussed results for
the GA NLF airfoil. The same is likely true of the IRT
test results as well. Hence, while perhaps the trends
indicated are helpful, the absolute values/levels are
almost certainly in error.
Results obtained from the test of the 8% thick fullscale business jet T-tail model [80] t in very well with
the business jet airfoil test results in a number of regards.
Three basic ice shapes were assessed in this program.
One was an ice-shape casting of a 1.6 min accretion from
the IRT, while the other two were LEWICE predictions

Fig. 34. Maximum lift penalty for actual and simulated ice accretions on business jet airfoil [133].

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

723

Fig. 35. Maximum lift penalty for simulated glaze ice accretions on business jet T-tail model [80].

for 9 and 22.5 min accretions at a total temperature


of 2.51C. Both smooth and rough (with a
k=cE24  104 ) versions of these two larger accretions
were tested during this program. Also included were
spoiler-type protuberances intended to approximate the
effect of the two larger ice accretions. Pertinent results
from this program are displayed in Fig. 35, from which
the following observations can be made:
*

Similar to the business jet airfoil test results, there is a


very signicant Reynolds number effect on the
indicated penalties. In this case, there is a very
noticeable drop off in the indicated penalties below
about 45 million. However, in this case, there are
variations with both the un-iced baseline and the ice
shapes with Reynolds number that contribute.
Similar to the results obtained with the GA NLF
airfoil [96,132], penalties indicated from the use of
smoothed simulations are denitely lower than those
obtained with roughness simulated.
The largest penalty observed with the (largest of the)
simulated ice shapes was a little less than 40%,
somewhat lower than that seen with the business jet
airfoil test results.
The attempt to approximate the effect of the 22.5 min
ice shape simulation with a spoiler-type protuberance
was partially successful in that it resulted in about
the right order-of-magnitude penalty, but not close
enough to be really useful.

Once again, however, there are aspects of this test


program which limit the usefulness and/or general
applicability of these particular results. For one, the
use of predicted ice shapes rather than actual measured
ice shapes for these glaze ice accretions raises doubts
because of the signicant differences observed in a
number of cases between predicted and actual ice shapes
such as illustrated in Ref. [131]. Also in this case, since
increased Reynolds numbers are accompanied by an
increased freestream Mach number, there is evidence
that compressibility effects, primarily on the baseline uniced tail, limit the maximum penalty observed at the
higher Reynolds numbers.
Thus, for one reason or another, whether it be not
having the thought-to-be most critical ice accretions, or
being concerned with the simulations utilized, and/or
having reservations about the test arrangement, conditions, etc., the foregoing six sources of high Reynolds
number test results available for these larger ice
accretions do not enable a meaningful determination to
be made of the likely upper limit of maximum lift
penalties that might be experienced at ight conditions
for any of the test geometries, let alone a wider range of
thick and thin lifting surfaces, other than the
previously stated estimate that they are probably in
excess of 50% for the thicker surfaces. Unfortunately,
the results from numerous (additional) low Reynolds
number test programs conducted over many years which
focused on this broad class of larger ice accretions

724

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

provide little in the way of additional insights into the


potential magnitude of the upper limit of these penalties
other than to perhaps reinforce the >50% estimate for
thicker surfaces. This can be seen by examining the
synopses of many of these test programs that are
provided in (generally) chronological order in Tables 1
and 2. In each case, the subject lifting surface, the
targeted ice accretion conditions, the simulated ice
shapes utilized, and the aerodynamic test conditions
and results (maximum lift penalties and stall angle
reductions) are documented if they were available/
published. Incidentally, all of the illustrations of the
ice shape simulations have the suction side of the lifting
surface shown as the upper surface whether the results
are for a wing or tail section. And, further, any place
there is an asterisk signies that there are some
extenuating circumstances involved that need to be
noted in order to place the indicated results in proper
perspective.
When examining the synopses/results provided in
Tables 1 and 2, there are a number of critical
observations to be made. To start with, in addition to
the ever present concern over the likelihood that such
low Reynolds number test results will not properly
simulate higher Reynolds number conditions, probably
the biggest concern with all the results shown is not
having the most critical ice accretion conditions
possible. Starting with the premise that the most critical
ice accretion conditions involve temperatures just below
freezing, high LWCs, maximum icing exposure durations/times, higher speeds (with corresponding lower
accretion angles of attack), and larger droplet sizes,
there just are not any results provided for these
combinations of conditions. Other signicant observations include the following:
*

Whenever a direct comparison can be made between


results obtained with both LEWICE-based glaze
ice simulations and IRT-based simulations, the
LEWICE simulations were not as detrimental. There
are two examples of this shown in Table 2. It is of
interest to note that in one of these cases [147], the
penalty incurred with the 45 min LEWICE-based
glaze ice simulation was only slightly more than that
observed with a much shorter duration IRT intercycle ice accretion.
Having essentially the same penalties for noticeably
different glaze ice simulations such as occurred with
the LEWICE-based and S&C glaze ice simulations
evaluated in Refs. [81,101] as well as the 22.5 and
45 min glaze ice simulations studied in Ref. [146], it
might be tempting to conclude that the details of the
glaze ice simulations are not that important. However, the Ref. [147] results tend to dispel that notion
indicating that there are some aspects of these glaze
ice simulations that are denitely more critical than

others. Incidentally, the asterisk shown by the results


for the rst ice accretions shown from Refs. [126128]
is because some sublimation of the ice accretion had
occurred before the aerodynamic test measurements
could be completed.
In contrast to some of the previously discussed high
Reynolds number test results [80,96,132] which
indicated that simulation of the surface roughness
characteristics of these larger ice accretions is
important (because they are detrimental), the results
from these low Reynolds number tests are certainly
mixed in that regard. There are cases where there
is no penalty indicated for adding roughness
[134,135,136,139141,146], cases where there is a
relatively small adverse effect indicated [143,146], and
even cases where adding roughness had a favorable
effect [136]. However, based on the conclusions
reached in Section 4 that low Reynolds number
indications of roughness effects are often not
realistic, the recommendation is to largely discount
these low Reynolds number results, and rely more on
the high Reynolds number indications. And, speaking of roughness, the asterisks shown with the Ref.
[146] results having a roughness with k=c 4  104
are a reminder that these results were obtained with
so-called loose grit rather than with sandpaper as
is more common.
There is a big difference indicated between the rime
ice results obtained in Refs. [134,135] and those in
Ref. [136], with a noticeable penalty for the former
and an improvement (i.e., increased maximum lift) in
the latter case. These differences occur with rather
subtle differences in the ice shapes, leading to the
suspicion that there may be some low Reynolds
number anomalies involved here.

In addition to this rather hodgepodge of low


Reynolds number test results obtained with a variety
of simulated ice shapes, there have been two somewhat
systematic low Reynolds number investigations conducted fairly recently using protuberances to simulate
glaze ice horns in an attempt to identify the most
detrimental features of these glaze ice accretions. One
was a study conducted by Papadakis et al. [149,150]
using spoiler-type devices on a NACA 0011 airfoil to
address horn angle, location and horn height effects.
Pertinent ndings from this study are summarized in
Fig. 36. In this case, two parameters are used to describe
the maximum lift penalties caused by these devices
since their use resulted in what is sometimes referred to
as a long bubble stall, where the lift curve slope
decreases very noticeably at some angle of attack, but
the lift still continues to increase somewhat beyond that.
Therefore, penalties based on both the maximum lift
level actually achieved as well as the lift level where the
lift curve slope decreases are used to describe the effects

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767


Table 1
Additional low Reynolds number test results addressing maximum lift penalties for larger ice shapes

725

726

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Table 2
Further low Reynolds number test results addressing maximum lift penalties for larger ice shapes [75,138,142,144,145]

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

727

Fig. 36. Maximum lift penalties for spoiler glaze ice simulations on NACA 0011 airfoil [149].

of these spoiler-type devices. It can be seen in Fig. 36


that based on the penalty utilizing the maximum lift
demonstrated, there is not a strong effect of the upper
horn angle except for the extreme case with the longer
horn in the vertical (901) position. However, if the
penalty based on the lift level where the lift curve slope
decreases very noticeably is used as being more
descriptive of the true adverse effect, then there is a
signicant inuence of upper horn angle, with larger
angles (i.e., more vertical displacement) being signicantly more harmful (as would be expected). In this case,
penalties well in excess of 50% are indicated, even for
quite moderate horn angles.
This strong inuence of upper horn angle also seems
to be apparent in the second set of results, those
obtained by Kim and Bragg [151,152], where thicker
protuberances were used to study the effects of upper
horn variations on the aerodynamic characteristics of an
NLF 0414 airfoil. Interestingly, in this case, with the
thicker protuberances and thicker airfoil, there were no
complications involved due to any long bubble stall in
dening the maximum lift penalties. Results from this
investigations are illustrated in Fig. 37, where it can be
seen that the strong inuence of upper horn angle on the
maximum lift penalties incurred seems apparent. However, in this case, this observation must be qualied
knowing that horn angle increases were also accompanied by some aft movement of the horn (as shown in
Fig. 27). Consequently, the trend shown in terms of
increased penalties with increasing horn angle is likely
a combination of effects. Nevertheless, with the very

modest aft movement involved, at least up to horn angles


of 401, it is likely that the predominant factor in the
increased penalties observed is the increased horn angle.
Now, with these results in hand, especially regarding
the adverse effect of increasing the upper horn angle,
and keeping in mind that having higher accretion
speeds, LWC, and droplet sizes, as well as temperatures
closer to freezing, will result in increased upper horn
angles (and sizes), it is instructional to go back and
reexamine some of the low Reynolds number results
summarized in Tables 1 and 2, as well as the previously
discussed high Reynolds number test results. For
example, in Tables 1 and 2, it can be seen that in most
of the cases where the largest maximum lift penalties
were experienced, such as the LEWICE-based glaze ice
shape used in Refs. [81,101], there is a very prominent
horn angle involved. Also, a reduced horn angle would
seem to explain why a lower penalty was seen with the
LEWICE-based simulation used in Ref. [147] compared
to the IRT-based glaze ice shape. Consequently, based
on all of the foregoing high and low Reynolds number
test results available at this time, it is quite apparent that
the maximum lift penalties likely to be encountered in
ight by thicker single-element lifting surfaces under the
most critical icing conditions are well in excess of 50%.
However, just how high they are, as well as whether they
are as large (or larger) than those encountered with the
potential runback/ridge ice accretions addressed in
Section 5, will remain unknown until such time as some
well conceived high Reynolds number test programs are
carried out for both types of ice accretions.

728

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 37. Maximum lift penalties for protuberance simulations of glaze ice accretions on GA NLF airfoil [151].

6.1.2. Single-element lifting surfaces with trailing-edge


control surfaces
It should be understood (without saying) that trying
to get a handle on the upper limit of maximum lift
penalties that could be incurred by single-element
wings and tails with trailing-edge control surfaces
deected at ight conditions is fraught with all the same
difculties preventing such an assessment being made
for congurations not having control surfaces. And, to
compound the situation even further, there are not that
many sources of applicable test results available either.
However, as was the case when examining these
characteristics for the effects of initial leading-edge ice
accretions (in Section 4), the trends indicated from the
results available do provide some helpful insights.
Looking rst at the test results available for thinner
lifting surfaces (i.e., those having un-iced maximum lift
coefcients o1.0 without control surfaces deected),
there are test results available for four congurations
(lifting surface/simulated ice accretion combinations),
including one set of high Reynolds number data for the
previously discussed 8% thick business jet T-tail
conguration [80]. These results, along with low
Reynolds number results for the same conguration
[153], and three other tail congurations [77,78,154,155],
are shown in Fig. 38. Incidentally, ice shape simulations
utilized in these investigations are illustrated in Table 1
for Refs. [77,78], and in Fig. 35 for Refs. [80,153].

Important items to be noted from the results summarized in Fig. 38 are as follows:
*

As has been noted before, there can be big differences


in the penalties incurred between congurations with
these thinner lifting surfaces (at low Reynolds
numbers). While both the high and low Reynolds
number test results for the business jet tail models
indicate penalties which become quite substantial,
especially at the critical elevator trailing-edge-down
conditions, the low Reynolds number results for the
other three indicate quite small penalties at the
conditions investigated.
Analogous to some of the results shown earlier in
Fig. 11 for initial leading-edge ice accretions, both
sets of test results for the business jet tail with the
22.5 min LEWICE glaze ice shape show that the
maximum lift penalty (percentagewise) increases
signicantly as the elevator deection moves toward
becoming more trailing-edge down. This occurs as a
consequence of a relatively constant incremental
maximum lift penalty becoming a larger percentage
of an un-iced maximum lift level that is decreasing as
the elevator is deected in this direction (opposing
the download).
An example of some of the difculties involved in
trying to interpret low Reynolds number test results
is provided by the two sets of low Reynolds number

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

729

Fig. 38. Variations in maximum lift penalty with trailing-edge elevator deection for thinner surfaces with large ice shapes.

results obtained for the business jet tail model


without the elevators deected. In this case, the
penalties indicated for the (rough) 22.5 min LEWICE
ice shape at a Reynolds number of 1.36  106 are
conspicuously different for the results from the test of
the full scale model [80] versus those from the test of
the 25% scale model, with the penalty indicated from
the test of the 25% scale model being about a third
larger. And, even more important, both sets of low
Reynolds number test results indicate penalties less
than those indicated at higher Reynolds numbers.
Turning next to the corresponding penalties incurred
by thicker lifting surfaces, the only known applicable
test results available appear to be low Reynolds number
results for an NLF-0414 airfoil [148] and three tail
geometries, namely the thickened LE design from Refs.
[77,78], and the Twin Otter [81] and F-28 [110]
horizontal tails. Results from these four investigations
are summarized in Fig. 39, while the simulated ice
accretions utilized in these investigations are all depicted
in Tables 1 and 2. Incidentally, the results shown for the
NLF-0414 airfoil in Fig. 39 are presented as though it
were a tail conguration in terms of the control surface
deection terminology utilized. It is also important to
note that many of the test results illustrated in Fig. 39
are agged. This signies that the actual penalties
could be higher than indicated because the angle of

attack range employed in the tests of the un-iced baseline


geometry was not large enough to ensure that the
maximum lift level had been attained. Finally, in
examining the test results summarized in Fig. 39, the
following points should be noted:

Some very large penalties, exceeding 80%, were


encountered with the LEWICE glaze ice shape
employed in the tests of the Twin Otter tail [81] at
the critical trailing-edge-down elevator deections.
And, as indicated, the actual penalties could be even
a littler larger.
When there were variations in test Reynolds number,
such as with the Twin Otter test results, the penalties
indicated at the higher Reynolds numbers were
generally larger. However, once most of the tail
lifting capability has been lost because of these ice
accretions, changes to the un-iced baseline maximum
lift level have little impact on the indicated (percentage) penalties.
At these low Reynolds numbers, the penalties
indicated at the trailing-edge-down elevator deections for the inter-cycle ice accretions are only about
half of those incurred with the larger glaze ice
accretions. Whether this relationship would hold at
higher ight Reynolds numbers is another question
however.

730

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 39. Variations in maximum lift penalty with trailing-edge elevator deection for thicker surfaces with large ice shapes.

Consequently, based on the limited applicable test


results available, indications are that maximum lift
penalties caused by these larger ice accretions on tails at
trailing-edge-down elevator deections (or on wings
with trailing-edge-up aileron deections) can become
quite large. Therefore, similar to the prevailing situation
with single-element lifting surfaces, some well conceived
high Reynolds number test programs utilizing the most
adverse ice accretions likely are needed to better
document just how large these penalties can be in ight.
And, perhaps even more important, existing efforts
aimed at educating pilots and ight crews of the dangers
involved in extending wing aps for landing if there is
any chance that ice has accreted on the tail must be
continued. Although this danger is and has been known
by many, as evidenced by the rather subtle quote of
the famous old Swedish pilot, Count von Rosen
referred to in Ref. [156] that there can not be such
an idiot as to use full ap if there is the slightest risk for
ice on the tailplane, it must be known and remembered
by all.
6.1.3. Multi-element lifting surfaces
Once again, in what has become an all too familiar
theme, there is very little really applicable data available

addressing the possible range of maximum lift penalties


which could be caused by large in-ight ice accretions on multi-element high-lift wings. However, be that
as it may, the available results, if viewed in the
framework of lessons already learned, do enable a
good understanding of important trends and relative
criticality for different geometries. Addressing rst
high-lift systems without any leading-edge devices, i.e.,
the so-called hard LE designs, what is available is
four sources of low Reynolds number test results.
The earliest of these were the results reported by
Johnson [85] in 1940 wherein the incremental loss in
maximum lift capability caused by the ice accretions
evaluated (see Table 1) was essentially the same with the
aps either extended or stowed. This resulted in the
maximum lift penalty being reduced from about 37%
with the aps stowed to 22% with them extended, which
is the inverse of the ratio of the un-iced maximum lift
coefcients. This is the same phenomenon seen earlier in
Section 4 with leading-edge roughness effects on multielement geometries. Not surprisingly, however, this
exact relationship does not always prevail as evidenced
by the results from the other three available (low
Reynolds number) data sources [74,135,148], but the
general trend of reduced (percentagewise) penalties with

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

731

Fig. 40. Variations in maximum lift penalty for high-lift congurations with Hard LEs.

the aps deected does prevail, especially at higher ap


settings appropriate for approach/landing conditions.
This can be seen by examining the results from these
other three sources illustrated in Fig. 40. Incidentally,
the ice accretion simulations employed in these investigations were portrayed in Tables 1 and 2. And, in all of
these cases, the use of simulated ice shapes was limited
to the main airfoil LE. Any ice accretions on the ap LE
were not considered. It is also interesting to note that the
results displayed in Fig. 40 from Ref. [148] are the same
as those used in Fig. 39. This acts as a reminder that the
test results for single-element lifting surfaces with
trailing-edge control surfaces shown in Figs. 38 and 39
indicating reduced penalties for negative elevator
deections (i.e., aiding the tail download) provide
further evidence of the trend of having reduced
(percentagewise) penalties with ap extension. However,
once again, analogous to the situation existing with
single-element lifting surfaces (including those with
trailing-edge control surfaces), additional appropriate
experimental studies are needed to determine just how
large these reduced penalties might be for the most
critical ice accretions possible at ight conditions. This
means that the testing must be done at higher, more
representative (of ight) Reynolds numbers as evidenced
by the fact that the penalties derived from the Ref. [148]
test results are most likely unrealistically low because
ow separation was present on the ap of the un-iced

baseline even at 51 ap deection at the low test


Reynolds number.
Looking next at the limited amount of applicable data
available addressing the effects of larger ice accretions
on high-lift systems having leading-edge devices, the
trend of reduced maximum lift penalties continues,
although there are again the seemingly ever present
concerns about probably not having the most critical ice
accretions and/or test conditions represented. Most of
the data available in this area comes from two low
Reynolds number investigations. These are the results
contained in the rst report of the Swedish-Soviet
Working Group [74] for a geometry with a vane ap
at a landing position, and the previously mentioned
results obtained by Potapczuk et al. [126128] in the
IRT with a Boeing 737-200 wing section/high-lift system
with ap settings appropriate for holding or initial
approach conditions. Illustrations of the high-lift
geometries and the simulated ice shapes employed
in these two investigations are shown in Figs. 41
and 42, respectively, while the ice accretions conditions and the test results are summarized in
Table 3.
In examining the earlier set of results [74], it can be
seen that the maximum lift penalties indicated for this
landing conguration are all less than 25%. Also evident
in these results are indications that the more critical
ice accretions are those occurring closest to freezing

732

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 41. Four-element high-lift geometry and ice imitators [74].

conditions, and those formed at higher angles of attack.


The indicated temperature effects are important as they
are clearly consistent with a number of other ndings,
but the indicated accretion angle-of-attack effects are
likely of little practical signicance. To start with, it can
be seen from Fig. 41 that these higher accretion attitudes
are the ones which result in the more detrimental ice
accretions that form on the LEs of the main element and
ap vane. However, the angles of attack involved
correspond to operating speeds very (too) close to stall.
In this case, more realistic accretion conditions corresponding to typical speed margins (to stall) would be
more like a couple of degrees or less. And, at these
attitudes, such detrimental accretions do not appear to
occur. That being the case, it would be concluded from
these results that ice accretions on the slat are
realistically the only ones to be concerned with.
Some very similar conclusions, trends, etc. are also
evident from the second set of low Reynolds number test
results [126128]. For one, even though the un-iced
maximum lift levels existing with the non-landing type
ap settings addressed in this test are well below that
seen in Ref. [74] (with a landing ap setting), still, the
highest penalties observed with the ice accretions
investigated are around 25% as well. Also, it can be
seen in Table 3 that once again the most detrimental
accretions (with everything else being equal) are those
occurring closest to freezing temperatures, and at the
higher angles of attack (which are probably not that
representative for holding conditions, etc.). Hence, once
again, and especially for the lower ap settings where the
slat LE is sealed against the main element upper surface,
and there are very small ap gaps, the quite obvious

conclusion is that ice accretions on the slat are of prime


importance.
Other than these two 2-D low Reynolds number sets
of test results, the only other known/available sources of
test results addressing the effects of these larger ice
accretions on multi-element high-lift systems with
leading-edge devices are from a very low Reynolds
number test of a complete 737 model at a landing ap
setting [157], and from a 2-D high Reynolds number test
of some relatively small simulated ice accretions on a
previously introduced (in Sections 2 and 3) three element
high-lift geometry, also in a landing conguration [158].
Results from the 737 model test indicate a maximum lift
loss of approximately 15% caused by a simulated ice
accretion on the wing LE. Unfortunately, however,
because of a lack of specics provided regarding the
icing simulation used, together with the quite low
Reynolds number for a really meaningful 3-D high-lift
test addressing maximum lift characteristics, it is
difcult to make use of this result. The 2-D high
Reynolds number test results do, however, provide an
interesting insight into Reynolds number effects on the
relative contributions of ice accretions on the slat versus
those on the downstream elements, even though the
icing simulations used are certainly not the most
detrimental possible (especially on the slat). Results
from this particular investigation are provided in
Fig. 43, where it can be seen that while there is very
little Reynolds number effect over the range investigated
(from 5 to 16 million) when the subject (rough) ice
accretions are simulated on all three surfaces, there is a
very denite Reynolds number inuence on the results
when only the main element and ap ice accretions are

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

733

Fig. 42. Five-element high-lift geometry and IRT ice accretions [126128].

modeled (representing the use of anti-icing on the slat).


The results shown also reinforce previous ndings
regarding the need to simulate the roughness characteristics of the ice accretions being modeled. This would be
especially critical on smaller, smoother shapes. And,
most importantly, the results again very clearly illustrate
the dominant criticality of any ice accretions allowed to
form on the slat versus those formed on downstream
elements. Further, the results would appear to indicate
that ice accretions on high-lift congurations with
leading-edge devices are not very important if the LE
device is effectively anti-iced. However, to put these
results into perspective for failure cases, or if the
leading-edge device is not anti-iced, the following need
to be considered:
*

The maximum lift penalty indicated by these tests of


around 10% with ice on the slat is not much more

than seen earlier (in Fig. 12) for an approach/landing


geometry at high Reynolds numbers with just
roughness on the slat.
By comparing the simulated ice accretion on the slat
used in this investigation with others observed in an
icing tunnel test with this same high-lift geometry (see
Fig. 1 in Section 2), it is clear that what was tested
here is certainly not the worst case possible.

Thus, based on the data available addressing maximum lift penalties caused by larger ice accretions on
high-lift systems having LE devices, the largest penalties
encountered to date are around 25% if ice is allowed to
form on the extended leading-edge device. This is
noticeably less than seen for high-lift geometries not
having leading-edge devices, and clearly much less than
incurred by single-element geometries. If ice is not
allowed to form on the leading-edge device, then any

734

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Table 3
Low Reynolds number test results addressing maximum lift penalties caused by Larger ice shapes on multi-element high lift
congurations with leading-edge devices

Fig. 43. High Reynolds number test results for ice accretions on multi-element airfoil [158].

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

penalties caused by (realistic) ice accretions on downstream elements would appear to be relatively minor.
However, because of well-founded concerns that the
most detrimental possible ice accretions were likely not
addressed in these investigations, some well conceived
new studies are needed to establish the upper limit of
these penalties, both with and without the LE device ice
protected. As part of these studies, attention needs to be
focused on establishing just what are the most critical
(and realistic) ice accretions which can form on the
downstream elements. Items needing to be addressed
include:
*

Establishing incidence angles consistent with realistic


ight operating conditions.
Considering higher LWCs, temperatures close to
freezing, and possibly longer exposure times.
Addressing a range of droplet sizes, especially larger
ones.
Assess the impact of variations in aircraft ight path
(i.e., descending for approach/landing, and ascending
for landing climb or takeoff climb).

It would be expected that current state-of-the-art


CFD capabilities for multi-element airfoil geometries
[159] would certainly be most helpful for a number of
these studies, especially for addressing droplet trajectory
issues.
6.2. Stall angle reductions
In reviewing the database available pertaining to the
stall angle reductions caused by a wide range of larger
ice accretions on a variety of aerodynamic lifting
surfaces, it is very important to keep in mind all the

735

concerns expressed regarding the limitations of the


available database for maximum lift penalties, because
all the same concerns obviously apply here as well.
These include likely not having the most critical ice
shape/accretion possible, not having very many good
quality high Reynolds number test results, having
inappropriate model/tunnel interactions, etc. However,
once again, while the database available may not be
appropriate for determining with any degree at certainty
the largest stall angle reductions possible, it is useful for
examining potential trends, especially for the iced
geometries.
6.2.1. Single-element lifting surfaces
Consistent with the very large range of maximum lift
penalties caused by these larger ice accretions, there is
likewise a broad range of stall angle changes indicated in
this category from the data available, including some
cases with simulated rime ice accretions where the ice
shape enables an increase in the stall angle of attack.
Starting with the existing database for 2-D airfoil
geometries, but not including test results where there
was an increase in the stall angle of attack because they
are obviously not relevant when trying to identify the
worst/most critical consequences, the available data are
presented in Fig. 44 as a function of the corresponding
incremental loss in maximum lift capability. It can be
seen that stall angle reductions up to (at least) 131 have
been observed, which is similar to the level encountered
to date with the simulated runback/ridge ice accretions
addressed in Section 5, and about twice that seen on 2-D
airfoils with the initial leading-edge-roughness-type ice
accretions reviewed in Section 4. While there is the
expected general correlation of the measured stall angle

Fig. 44. Stall angle reductions caused by larger ice accretions on 2-D single-element airfoils.

736

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

reductions with the incremental (not percentage) loss in


maximum lift (coefcient) capability, there are also some
signicant variations within this general trend as well.
For reference, also shown in Fig. 44 is a lift curve slope
of 0.1, which would correspond to the 2-D case in the
linear range (i.e., before the roundover approaching
stall). Having a preponderance of test results falling
above this reference line would, for one, seem to indicate
that there is typically more roundover in the lift curve
prior to stall with the un-iced baseline cases than there is
with the airfoils having simulated ice shapes. This is
certainly a characteristic which might well be more
prevalent in low Reynolds number test results, and/or in
2-D results obtained without any sidewall boundary
layer control. Another potential contributor to some of
the variations seen could be having uncorrectable wall
interference effects. In assessing the database summarized in Fig. 44, it is also important to note that there is
only one set of high Reynolds number test results
available/included [96,132], and these are believed to be
contaminated with uncorrectable wall interference effects caused by having a model too large for the test
facility. This lack of reliable high Reynolds number test
results is an important limitation in that low Reynolds
number test results are often not representative, both in
terms of maximum lift levels achieved, as well as with
regard to stall characteristics. With regard to maximum
lift levels achieved, it should also be noted in Fig. 44 that
the largest stall angle reductions displayed come from

test results [81,101] where there is no assurance that the


maximum lift level of the un-iced baseline airfoil was
reached. Hence, the actual stall angle reductions (at
these low Reynolds numbers) could well be larger than
indicated. However, the real question to be answered is
how large might these stall angle reductions be with the
most critical ice accretions at higher Reynolds numbers
more representative of ight, and without being
encumbered by uncorrectable wall interference (oor
and ceiling as well as sidewall viscous effects).
There is a similar scarcity of reliable results available
for assessing stall angle reductions incurred with larger
ice accretions on 3-D wings and tail. What is available is
shown in Fig. 45. By comparing these results with the
2-D results in Fig. 44, it can be seen that with this
collection of data, the stall angle reductions incurred
with the 3-D geometries follow the same general trend as
the 2-D results, but are, in general, not any larger than
those seen with the 2-D congurations. This is contrary
to what would normally be expected with the lower lift
curve slope typically associated with these nite aspect
ratio geometries. It is also contrary to what was seen
earlier (in Fig. 15) with the data available for initial
leading-edge ice accretions. The cause of this apparent
disparity for most of the results probably lies with the
low test Reynolds numbers involved since low Reynolds
number anomalies are often even more prevalent with 3D maximum lift characteristics. And, while there are two
sets of high Reynolds number test results included in

Fig. 45. Stall angle reductions caused by larger ice accretions on 3-D single-element wings and tails.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

737

Fig. 46. Stall angle reductions caused by larger ice accretions on tails with elevators deected.

Fig. 45, the one ight test data point for the An-12 [74],
and the test results for the 8% thick full-scale business
jet T-tail model [80], neither conguration is really that
appropriate for establishing more general trends. The
An-12 has a fairly thick wing, and the ice accretion
involved was certainly not at all representative of a
critical accretion, while the business jet tail model had
an un-iced maximum lift coefcient of only 0.94 at the
test Reynolds number. So, the well conceived high
Reynolds number testing recommended to better establish the upper limit of maximum lift penalties that can be
caused by the most detrimental large ice accretions is
likewise needed to establish the upper limit of corresponding stall angle reductions feasible at ight conditions, since that cannot be done now with data presently
available.
6.2.2. Single-element lifting surfaces with trailing-edge
control surfaces
The available and (somewhat) applicable test results
for these congurations with control surfaces deflected
for both 2-D and 3-D geometries are illustrated in
Fig. 46. Again, unfortunately, there is only one set of
high Reynolds number test results available, and these
are for the full-scale business jet tail model with the low
un-iced maximum lift level. Overall, the results appear to
be very similar to those displayed in Figs. 44 and 45 for
single-element geometries without such control surfaces
deected. And, there does not appear to be anything out
of the ordinary happening with the critical trailing-edgedown elevator deections. The only results indicating
that stall angle reductions might be noticeably greater
for these deections are the Ref. [153] very low Reynolds
number data. But, the Ref. [80] high Reynolds number
data for the same business jet tail geometry tend to
dispel that notion, and, in the process, provide yet

another example of the inadvisability of relying on low


Reynolds number test results. So, to overcome the
existing state of affairs where the available database for
these characteristics is predominantly from low Reynolds number testing, appropriate new high Reynolds
number testing of representative geometries with the
most critical ice accretions feasible is again needed.
6.2.3. Multi-element lifting surfaces
A very similar situation exists with the database
available for these geometries in that there are very few
reliable high Reynolds number test results, and,
typically, the ice accretions evaluated are not for the
most detrimental icing conditions possible. For example,
the only high Reynolds number test results available
[158] fall in this category of simulating less-than-themost-severe-possible ice accretions. However, again,
these results are useful in identifying likely trends. In
examining the existing database which is summarized in
Fig. 47, it can be seen that there tends to be more results
below the 0.1 lift curve slope than seen with the previous
two types of lifting surfaces. What is noteworthy is that
these data points below the reference slope are mostly
for congurations having LE devices (extended), and
those above the reference slope are mostly for hard
LE designs (similar to the previous lifting surfaces). This
indication that congurations having LE devices will
have proportionately signicantly smaller stall angle
reductions needs to be further explored (at high
Reynolds numbers with more critical ice accretions).
Another important observation to be made from the
data shown in Fig. 47 is with regard to the Ref. [158]
results representing a case with slat anti-icing activated.
In this case, even though there is a reduction in the
maximum lift level caused by having simulated ice
accretions on the LEs of the main element (behind the

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

738

Fig. 47. Stall angle reductions caused by larger ice accretions on multi-element high-lift congurations.

slat) and ap, there is not a reduction in stall angle, once


again indicating that the most consequential ice accretions would be those which could be allowed to form on
the slat.
6.3. Drag penalties
It is helpful when sorting through the large database
of test results available addressing the drag penalties
which can be caused by these larger ice accretions to
keep in mind some of the lessons already learned
regarding leading-edge geometry/size effects, what constitutes the most critical icing conditions in terms of
causing the largest penalties, concerns over the effectiveness of simulated-versus-actual ice accretions, the
importance of simulating roughness characteristics, as
well as issues with the performance testing conditions in
terms of Reynolds number, wall effects, etc. For
example, as pointed out previously, the most critical
ice accretions have been found to be those which are
formed at total temperatures just below freezing, and
additionally, have a high LWC, larger droplet sizes,
longer icing durations, and higher aircraft speeds and
lower angles of attack (while the ice is being accreted).
Hence, test results obtained at less critical conditions are
of no real use in establishing the upper limit of drag
penalties which might be encountered, nor are test
results presented without the associated accretion
conditions identied. Also, because of differences

reported in drag penalties measured with simulated ice


shapes compared to the penalties seen with actual icing
tunnel ice accretions, it is appropriate to be somewhat
skeptical of test results obtained using simulated ice
shapes other than those which are castings of actual ice
accretions. This would be particularly true for ice shape
simulations not having the roughness simulated (or
dened). And, further, errors or misinterpretations
likely to be introduced by the use of low Reynolds
number test results need to be factored into the
assessment, especially so when considering drag penalties expressed in terms of percentage increases above the
measured baseline un-iced drag levels. Caution also
needs to be exercised when assessing/interpreting icing
tunnel results obtained with small and/or thin lifting
surface models where the ice accretions are most likely
unrealistically large for representing more typical larger
and/or thicker (ight) geometries.
Another very important factor needing to be considered when assessing drag penalties caused by these larger
ice accretions is to recognize that drag penalties for a
given ice accretion will be different at different operating
conditions/speeds (i.e., lift coefcients). Unfortunately,
all too often, test results are presented at a xed or
different angles of attack, or as a function of angle of
attack, when, in reality, what is most needed is to be able
to assess drag penalties for various operating conditions
(i.e., reference speeds). Regrettably, this is not possible in many cases, either because of insufcient drag

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

739

Fig. 48. Variation in drag penalty with operating conditions for NACA 0012 airfoil with simulated IRT glaze ice shape [139141].

measurements, and/or not having a reliable basis for


establishing appropriate reference speeds (i.e., based on
the un-iced stall speeds). This is especially true of many
of the earlier studies in the IRT [119124] where,
because of the very large size of the models, stall speeds
were not established (and would not have been
representative even if measured).
An example of the variations in the incremental drag
penalties which can occur with a given large ice shape at
different operating conditions is provided in Fig. 48
using the 2-D test results published by Bragg et al. [139
141] for a 0012 airfoil with a glaze ice shape simulating
an IRT accretion which occurred at 41 angle of attack.
The ice shape and accretion conditions simulated were
previously documented in Table 2. Although these
particular test results were obtained at a relatively low
Reynolds number (1.5 million), they are nevertheless
quite useful in that the resulting maximum lift penalty
was substantial (54%), and the un-iced baseline prole
drag levels/variations are sensible (which is denitely not
always the case with the total available database). It can
be seen from Fig. 48 that there is a large variation in the
incremental drag penalty which occurs over the range of
possible operating conditions (i.e., speed margins to stall
with the un-iced baseline). The penalty increases quite
rapidly as operating speeds are reduced approaching the
(increased) stall speed associated with the ice accretion.
This is very important, especially if the ight speed has
been reduced for climbing, approach, etc. It can also
be seen from Fig. 48 that assessing drag penalties at
a constant angle of attack (accretion condition) such as

was done with the Ref. [125] test results shown


previously in Fig. 28 results in an under-prediction of
the incremental penalty which would occur at a constant
aircraft speed. These results also lead to the conclusion
that the most detrimental (largest) drag penalties are
those which will occur after ice has accreted at high(er)
speed/lower angle of attack conditions, and the aircraft
is then slowed for climb, approach, etc. In addition to
having larger ice accretions at higher speeds, etc., more
of the accreted ice forms on the upper surface of the
wing at lower angles of attack. Finally, the results shown
in Fig. 48 provide some basis for adopting the approach
that there are two components of the drag penalty, one
being a parasite-drag-type increment which stands out at
higher speeds/lower angle of attack conditions (and is
probably a strong function of the glaze ice horn height
and angle), and a second component which varies with
lift and becomes predominant at ight speeds approaching the iced-conguration stall speed.
It is also important to note that while there are a
number of (previously mentioned) ight test results
reported documenting drag increases experienced from
ight in natural icing conditions, these results are not
very helpful for assessing the upper limits of drag
penalties possible. To start with, results reported include
the contributions of (undocumented) ice accretions on
aircraft components in addition to the wing and/or
tail (i.e., landing gear, fuselage, engine cowling,
protuberances, etc.) But, much more important, it
would appear that, for obvious crew safety considerations, less-than-the-most-critical icing and operating

740

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

conditions were involved in all the reported cases. For


example, with the previously quoted case where the
Twin Otter experienced a 75% total aircraft drag
increase (with a glaze ice shape on the wing), the
corresponding wing section drag increase reported was
only 120%. This is almost an order of magnitude less
than the section drag increases seen in Fig. 28 with the
Ref. [125] test results, and those just seen at some
conditions in Fig. 48 with the Refs. [139141] results.
6.3.1. Single-element lifting surfaces
Development of a methodology for predicting the
drag penalties caused by large ice accretions on singleelement lifting surfaces has been a long standing
objective and focus of the icing research community.
One prime avenue pursued has been the development of
mostly empirical correlations relating measured drag
penalties to either accretion conditions/characteristics
and airfoil geometry, or the physical characteristics of
the ice accretions, most notably the (glaze ice) upper
horn height and inclination. Early on, Gray [122], using
data from tests of ve different airfoils in the IRT,
proposed a correlation based on using icing time, air
speed, temperature, LWC, impingement efciencies, as
well as airfoil chord, angle of attack, and leading-edge
radius of curvature. Incidentally, almost all of the test
results used in the development of this correlation had
drag penalties measured only at the ice accretion
angle(s) of attack. Two decades later, however, Olsen
et al. [125] demonstrated that Grays correlation did not
adequately predict the drag penalties measured for
various ice accretions on a NACA 0012 airfoil in the

IRT. In fact, the agreement was quite poor. Other


correlations, also based on ice accretion terms, such as
those developed in the early 1980s by Bragg [160], Miller
et al. [161], and Flemming and Lednicer [162], appeared
to overcome some of the shortcomings of Grays
correlation, but these still all had limited applicability.
More recently, efforts aimed at developing drag penalty
correlations have concentrated more on associating such
drag increases with the physical characteristics of the ice
accretion in the hope of obtaining a correlation that is
universally applicable. Toward this end, Cook [163]
demonstrated that measured ice shape maximum
height and angle may be correlated to measured drag
due to ice accretions. This nding is reected in the
recent correlation put forth by the 12A Working Group
[98] as illustrated in Fig. 49. In this case, the correlation
parameter selected represents the nondimensional height
of the ice horn (or protuberance) normal to an
undisturbed streamline. It can be seen from Fig. 49 that
the drag penalties for the ice accretion conditions
selected (i.e., ap3:2) generally increase with the correlation parameter utilized, but, there is a large amount of
scatter in the results, clearly indicating that some
fundamentals are denitely missing.
Unfortunately, however, even if the scatter in
correlations such as depicted in Fig. 49 could be greatly
reduced, such correlations are of very limited practical
value to aircraft developers and operators because, in
reality, as was illustrated previously in Fig. 48, there is
no single drag penalty for any particular ice accretion
even for any one airfoil geometry, yet alone for different
geometries, etc. This is an extremely fundamental point,

Fig. 49. 12-A working group correlation of airfoil drag increases due to ice accretions [98].

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

well known to aerodynamic designers of aircraft.


Interestingly, the authors of the rst joint report
published by the Swedish-Soviet Working Group back
in 1977 [74] were well aware of this fact. Much of the
relatively large database of IRT single point (i.e., xed
a) incremental drag measurements is also, unfortunately,
of very limited value for establishing realistic drag
penalties for a range of operating conditions. These
results have, however, been very useful in establishing
trends such as the pronounced peaking of drag
penalties at total temperatures just below freezing
conditions, illustrated by the Ref. [125] test results, as
well as identifying the detrimental effects of increased air
speed, droplet size, and LWC. And, the indications from
the limited number of (xed a) test results obtained at
angles of attack above the accretion condition that drag
penalties at these conditions are increased signicantly
above those at the accretion condition are also most
instructive. But, for some other trends indicated by these
single point drag increments, caution is suggested
before presuming they apply to a range of operating
conditions. An example of this would be the indication
from a number of test results [119,123,124,135] etc. that
the increases in drag penalties with accretion time tend
to be more linear in contrast to maximum lift penalties
where a signicant part of the eventual penalty occurs
very early in the accretion process. However, as will be
seen subsequently, this observation is misleading.
So, in order to place the drag penalties caused by large
glaze (and inter-cycle) accretions into an aircraft
operational framework rather than a research perspec-

741

tive, test results available where drag penalties for


mostly simulated ice accretions have been measured
over a range of lifting conditions, and, corresponding
maximum lift characteristics (and penalties) have also
been established, are used herein to assess drag penalties
at specic operating conditions (i.e., speeds relative to
the reference un-iced stall speeds). The (potentially)
applicable available database includes two primary sets
of high Reynolds number test results, and ten sources of
lower Reynolds number results. Except for one set of
low Reynolds number test results obtained with actual
ice accretions in the IRT [126128], all the others are
from tests in conventional wind tunnels using simulated
ice shapes.
High Reynolds number test results available are from
the LTPT test of the 14% thick GA NLF airfoil
[96,132], and the test of the 8% thick full-scale business
jet T-tail model in the NASA 40  80 ft2 Wind Tunnel
[80]. While tail drag increases due to icing per se are not
really a high priority issue (compared to the wing), the
latter set of test results, when interpreted as though they
were for a wing, are helpful in identifying drag penalties
which can be experienced on thinner (wing) congurations, and for demonstrating Reynolds number and
other effects.
Turning rst to the Refs. [96,132] tests results
obtained for the 14% thick GA NLF airfoil in the
LTPT, the indicated incremental drag penalties for the
four simulated glaze ice accretions evaluated are shown
in Fig. 50. These four are 6 and 22.5 min accretions
having the actual (IRT measured) 3-D roughness

Fig. 50. Drag penalties for simulated glaze ice accretions on GA NLF airfoil [96,132].

742

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

characteristics represented, and smoothed (2-D)


simulations of the 2 and 6 min accretions. Accretion
conditions and illustrations for these simulated ice
shapes were provided previously in Fig. 33 with the
corresponding maximum lift penalties. From the test
results illustrated in Fig. 50, the following observations
can be made:
*

At the higher speed conditions, the indicated drag


penalties are roughly proportional to the accretion
times.
There is a very pronounced growth of the drag
penalties at lower speed conditions approaching the
(increased) iced conguration stall speeds. At these
conditions, it appears as though the maximum
penalties likely (without stalling) will not be proportional to the accretion times, but they will occur at
different speeds with different ice shapes.
The Reynolds number effect on the indicated drag
penalties is rather inconsequential over the range
tested (i.e., from 4.6 to 10 million).
There is a very noticeable increase in the drag penalty
incurred with the larger 22.5 min accretion as the
Mach number is increased from 0.21 to 0.29.
Interestingly, though, there is actually a small
reduction in the indicated penalty for the 6 min
simulations at this same Mach number.

An instructive perspective is gained by examining


crossplots of the results shown in Fig. 50 at xed speed

conditions as a function of the corresponding maximum


lift penalties incurred with these ice simulations. Such a
presentation for the lower Mach number results (i.e.,
M0 p0:21) is provided in Fig. 51 where it can be seen
that the drag penalties at a given speed are clearly
related to the maximum lift penalties which, in turn, are
an indication of the proximity to stall. These results
would seemingly indicate that the incremental drag
penalty incurred by the 6 min simulated rough ice
accretion at a speed 30% higher than the un-iced
conguration stall speed would be about the same as the
penalty incurred by the 22.5 min accretion at a speed
50% above the un-iced stall speed. Hence, it is clearly
necessary to specify the operating conditions (i.e.,
proximity to stall) when dening drag penalties for
any ice accretion. A similar presentation comparing the
drag penalties observed at 0.29 Mach number to those
seen at M0 p0:21 is shown in Fig. 52. Here, the
signicant potential adverse impact of the drag increase
suffered by the larger ice accretion at 0.29 Mach number
on the magnitude of drag penalties possible (prior to
stall) is most obvious. Unfortunately, these data are the
only known set of airfoil test results where such
potential Mach number effects for a given ice shape
have been investigated. This is important to note
because caution must (again) be urged in making
quantitative evaluations using the results from this test
program (at any Mach number) because of the
previously discussed premature separation problem
present at very modest angles of attack with the (36 in

Fig. 51. Relationship of drag penalties and maximum lift losses with simulated glaze ice accretions on GA NLF airfoil [96,132].

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

743

Fig. 52. Compressibility effects on relationship of drag penalties and maximum lift losses for GA NLF airfoil [96,132].

Fig. 53. Drag penalties for simulated glaze ice accretions on business jet T-tail model [80].

chord) un-iced baseline conguration. This premature


separation caused by tunnel wall effects has a very
obvious (adverse) impact on the un-iced conguration
baseline drag polars. Because of this, it is feasible that
the indicated drag penalties (at most all conditions)
might well be lower than would be the case without these
wall effects.

Looking next at the other primary source of available


high Reynolds number test results addressing drag
penalties for larger (simulated) glaze ice accretions, the
indicated drag penalties from the Ref. [80] test of the
full-scale business jet tail model at a Reynolds number
of 5.1 million are shown in Fig. 53 as a function of speed
ratios relative to the un-iced baseline stall speed for the

744

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

three ice shapes addressed. The subject ice accretion


conditions being simulated, and illustrations of the two
larger accretions, were shown previously in Fig. 35 with
the corresponding maximum lift penalties. In Fig. 53,
results are presented for both smooth and roughened
versions of the two larger LEWICE-dened accretions
(9 and 22.5 min duration), as well as the roughened
version of the shorter duration IRT ice accretion. From
these results, it can be seen that the major impact
(increased drag penalty) of simulating roughness characteristics occurred at the lower speeds, primarily
reecting the increased maximum lift penalty. There is
little effect of roughness indicated at the higher speed
conditions. It can also be seen from Fig. 53 that, in
general, the drag penalties increase in a rather linear
manner with accretion duration after the initial 2 min
accretion. It should be noted also that the indicated drag
penalties incurred with this 3-D, relatively low aspect
ratio conguration, are in general about double those
seen with the Refs. [96,132] airfoil test results even
though the corresponding maximum lift penalties are
lower than the airfoil results. Clearly, some of this
increase may well be associated with the much thinner
design (i.e., 8% versus 14%), but it is also likely that
some part of the growth is due to non-viscous liftdependent drag increases. The signicant effect (i.e.,
reduction) of reducing the Reynolds number from 5.1 to
1.36 million on the indicated drag with this conguration for the two larger (roughened) ice accretions is
illustrated in Fig. 54. However, here again, the reduction
occurs mostly at lower speeds, and is, to a large degree,
associated with the reduced maximum lift penalties
incurred at this lower Reynolds number.

The crossplots of the drag penalties at xed speed


conditions as a function of the corresponding maximum
lift penalties measured for this business jet tail conguration provide an additional useful perspective regarding the potential inappropriateness of test results
obtained with non-roughened ice accretions and/or
those obtained at low Reynolds numbers, albeit these
results are for a quite thin lifting surface. In Fig. 55,
these crossplots are illustrated for the high Reynolds
number test results obtained with the three rough ice
accretions at speed ratios of 1.3, 1.5, and 1.8. In
addition, results are also provided in Fig. 55 at 1.3 for
the two smooth ice shapes tested at the same Reynolds
number, and for the two larger rough ice accretions at a
much lower Reynolds number. Similar comparisons at
speed ratios of 1.5 and 1.8 are shown in Fig. 56. At all
three conditions, it can be seen that even though the test
results at low Reynolds numbers as well as those for the
smooth ice simulations have reduced drag penalties, the
corresponding reductions in maximum lift penalties are
disproportionately larger, resulting in indicated drag
penalties which are too high relative to the maximum lift
penalties. It is worth noting, however, that the trend
regarding smooth simulated ice shapes was not apparent
in the Refs. [96,132] high Reynolds number test results
for the thicker GA NLF airfoil.
The remaining database of almost exclusively low
Reynolds number test results for 2-D airfoils is
presented in Figs. 5759 together with the Refs.
[96,132] high Reynolds number results for speed ratios
of 1.3, 1.5, and 1.8, respectively. Incidentally, all of these
low Reynolds number results are (believed to be) for
relatively low Mach numbers (i.e., o0.20), so, these

Fig. 54. Reynolds number effects on drag penalties for simulated glaze ice accretions onbusiness jet T-tail model [80].

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

745

Fig. 55. Relationship of drag penalties and maximum lift losses with simulated glaze ice accretions on business jet T-tail model [80].

Fig. 56. Reynolds number and roughness effects on relationship of drag penalties and maximum lift losses on business jet T-tail
model [80].

746

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 57. Relationship of drag penalties and maximum lift losses with simulated glaze ice accretions on various 2-D airfoils at speed
ratio=1.3.

Fig. 58. Relationship of drag penalties and maximum lift losses with simulator glaze ice accretions on various 2-D airfoils at speed
ratio=1.5.

additional results do not provide any new information


regarding the drag increases seen at higher Mach
numbers (i.e., 0.29) with the Refs. [96,132] test results.

In examining the results displayed in Figs. 5759,


there are probably three primary observations to be
made. For one, there is no indication from these

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

747

Fig. 59. Relationship of drag penalties and maximum lift losses with simulator glaze ice accretions on various 2-D airfoils at speed
ratio=1.8.

low Reynolds number data for thicker airfoils that test


results obtained with smooth versions of glaze ice
accretions yield disproportionately high drag penalties
such as seen with the thinner business jet tail model. The
second, and more important observation to be made is
that in almost all cases, the lower Reynolds number
results do yield higher drag penalties for a given
maximum lift penalty than seen with the Refs. [96,132]
higher Reynolds number results. In conjunction with
this, it is important to note that in all three of the low
Reynolds number data sets where results are available at
more than one Reynolds number for the same ice
shapes, the higher (but still relatively low) Reynolds
number results (shown as solid symbols) do generally
indicate reduced drag penalties for a given level of
maximum lift penalty, consistent with the business jet tail
results. The third observation of signicance is noting the
similarity of the indicated parasite drag penalties (i.e.,
approaching 0.10) just prior to stall at both the 1.3 and
1.5 speed ratios, even though the ice shapes and
maximum lift penalties are quite different. This again
demonstrates the importance of dening the operating
conditions being considered when attempting to identify
drag penalties caused by large glaze ice (or any other)
accretions. Incidentally, keeping in mind that the Refs.
[96,132] indicated drag penalties are likely somewhat
understated, a best-guess representation of the parasite

drag penalties likely caused by larger glaze ice accretions


on typical wings at Reynolds number representative of
ight conditions (but at incompressible conditions) based
on the current available database would be somewhere
between the two fairings indicated in Figs. 5759.
The foregoing review of available data addressing the
drag penalties caused by larger glaze ice accretions on
single-element lifting surfaces (primarily airfoils/wings)
has enabled a number of important observations, but, at
the same time, has made it very evident that there are
some major holes in the database needed to realistically
predict these drag penalties for the range of geometries
and potential operating conditions needed. Regarding
priorities, it is clear from a safety perspective that
avoiding stall in the iced conguration is top priority.
After that, it is important to avoid operating even close
to stall if large drag penalties are to be avoided.
However, for both of these, the previously suggested
well conceived high Reynolds number test programs
are needed to rst establish the upper limit of maximum
lift penalties possible with the so-called (most) critical ice
shapes. At the same time, believable (i.e., no more 36 in
chord models in the LTPT) drag penalties at high
Reynolds number conditions need to be determined for
the range of possible operating conditions for these
critical ice accretions. This would have to include
elevated Mach numbers, and determining penalties

748

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

likely on realistic 3-D nite aspect ratio designs where


drag sources in addition to just the parasite drag
contribution are involved.
6.3.2. Single-element lifting surfaces with trailing-edge
control surfaces
Drag penalties caused by large glaze ice accretions on
these geometries are of relatively minor signicance
considering the very short duration of such occurrences.
While there are (at least) a couple of sources of test
results available addressing this situation, namely, the
Ref. [80] results from the test of the full-scale business jet
T-tail, and the Ref. [81] results obtained with a section
from the Twin Otter tail, no analysis of these test results
is included herein because of the very low impact of any
such drag penalties. Again, avoiding (tail) stall is the top
priority.
6.3.3. Multi-element lifting surfaces
Even though the available database of test results
addressing drag penalties caused by larger ice accretions
on multi-element high-lift geometries is quite limited,
and not well suited for enabling good quantitative
estimates of the largest drag penalties possible, it is,
nonetheless, very useful for identifying some important
trends/tendencies regarding these penalties. There are
three sources of applicable test results, from Refs.
[74,126128,158], and most of the results are for
congurations having leading-edge slats. All were 2-D

tests, so, as usual, the penalties indicated are only the


parasite drag contribution. The indicated drag penalties
at 1.3 times the un-iced conguration stall speeds are
rst summarized in Fig. 60 for Reynolds numbers of
5 million and below. For clarity, test results for either
the IRT or simulated ice shapes involving noticeable or
meaningful accretions on the LE of the ap are
represented by solid symbols, and results for congurations not having leading-edge devices are denoted by
agged symbols. For the record, the ice shapes evaluated
in these investigations were shown previously in Table 1
and Fig. 41 for the Ref. [74] geometries, Fig. 42 for Refs.
[126128], and Fig. 43 for Ref. [158]. Upon reviewing
the indicated parasite drag penalties shown in Fig. 60,
there is one principal lesson to be learned, and that is,
without doubt, ow deterioration and/or separation on
the upper surface of the ap is the most important factor
involved in establishing the magnitude of drag penalties
incurred. First, for the cases where there was some
meaningful ice accretion on the ap LE, it can be seen
that the penalties indicated are about double most of
those indicated without such accretions. Second, for the
one data point from Ref. [74] which has a much higher
drag penalty indicated than the other results obtained
without ice on the ap LE, it must be noted that this
particular data point is for a conguration having a
single-segment ap deected 401. That being the case, it
is very likely, particularly at the low test Reynolds
number, that ow separation on the ap was imminent

Fig. 60. Drag penalties caused by larger ice accretions on multi-element high-lift airfoils at speed ratio=1.3.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

in the un-iced condition, and, hence, the ice accretion on


the LE of the main element would most certainly have
led to/triggered a large ow separation on the ap.
Consistent with this presumption, it can be seen from
Fig. 60 that the indicated drag penalty for this geometry
with the ap only deected 201 is in line with the rest of
the test results (not having ice on the ap LE).
Consequently, it seems quite apparent that drag
penalties incurred with larger ice accretions on multielement high-lift geometries will, in general, also be
strongly inuenced by what ow separation margins
exist on the ap of the un-iced conguration at the ight
conditions under consideration. Clearly, many possibilities exist.
Other issues needing to be considered relative to the
results shown in Fig. 60 are concerns that the ice
accretions involved are likely not the most critical
possible, plus the ever-present concern regarding the
adequacy of the test Reynolds numbers. However, it is
conceivable that these two factors may be somewhat
offsetting. Clearly, by denition, more critical ice
accretions would have greater drag penalties, but the
little data available illustrating Reynolds number effects
on drag penalties clearly indicates that these penalties
might well be reduced at higher Reynolds number more
representative of ight on larger vehicles. The data
indicating the possibility of reduced drag penalties at
higher Reynolds numbers are the Ref. [158] data shown
in Fig. 61 (for the not-too-large accretions considered).
It can be seen that for the three icing simulations

749

considered, there is an order-of-magnitude reduction in


the drag penalty as the Reynolds number is increased
from 5 up to 16 million. It remains to be seen, however,
what these Reynolds number effects would be with
larger, more critical ice accretions.
6.4. Trailing-edge control surface characteristics
With leading-edge roughness representative of initial
ice accretions, it was found that any really noteworthy
changes in elevator hinge moment characteristics were
tied very directly to noteworthy changes in the tail
maximum lift level and stall characteristics brought
about by these small ice accretions. And, conversely, it
was found that if these initial leading-edge ice accretions
had little effect on the tail maximum lift level and stall
characteristics, the elevator hinge moments were likewise
impacted very little. Not surprisingly, the same thing
happens with larger ice accretions, the only difference
being that when changes do occur, they happen at a
lower incidence angle consistent with the (generally)
somewhat larger losses in maximum lift and stall margin
that occur with these larger ice accretions.
The available database of test results where the effects
of larger ice accretions on elevator hinge moment
characteristics has been determined is predominantly
low Reynolds number, and comes from the same sources
which provided the material reviewed previously regarding leading-edge roughness effects on such characteristics. Results are available for three relatively thin tail

Fig. 61. Reynolds Number effect on drag penalty [158].

750

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

geometries and two thicker ones. Both sets of test results


for thicker geometries, the 181 tailplane with thick LE
from Refs. [77,78] and the 2-D section from the Twin
Otter tail from Ref. [81], are at low Reynolds numbers.
The same is true for the two thinner geometries
addressed in Refs. [77,78], the 441 tailplane and the 181
tailplane with normal LE. Unfortunately, the only high
Reynolds number data available are the Ref. [80] test
results for the 8% thick full-scale business jet tail.
With the thicker tail geometries, as was pointed out
previously, the elevator hinge moments without any
leading-edge ice become gradually more positive (or less
negative) as the tailplane incidence/download is increased prior to stall, but then, right around stall, the
rate of increase (toward more positive) becomes
noticeably greater. Then, when the larger ice shapes
are added, the S3 shape on the thickened 181 tail and the
45 min LEWICE and S&C shapes on the Twin Otter
tail section (see Tables 1 and 2 for details), the resulting
elevator hinge moment characteristics closely resemble
those for the respective baseline geometries except that
the noticeable increase in hinge moments in the positive
direction occurs at a much lower incidence angle
consistent with the stall angle reductions illustrated
earlier in Fig. 46. Interestingly, in the one set of test
results where trailing-edge-down elevator characteristics
were evaluated [81], the increase in hinge moments
occurring around stall at these deections is greater than
for trailing-edge-up conditions. However, in no case is
the change any more severe than seen with the un-iced
baseline, it just occurs earlier. Also, once again, the
results from these two investigations illustrate that the
magnitude of the change in elevator hinge moments
around stall is closely related to the baseline stall
characteristics, with the Refs. [77,78] results providing
a more abrupt change, consistent with a larger more
abrupt lift loss at stall with this geometry.
The elevator hinge moment characteristics measured
for the three thinner tail geometries with larger ice
accretions produce results somewhat different from
those seen with the thicker designs. For example, with
the 441 tailplane addressed in Refs. [77,78], the hinge
moments on the un-iced baseline prior to stall increase
(i.e., became more positive) at a greater rate than seen
with the thicker designs, but there is very little change in
slope seen around the maximum lift condition, consistent with a very gentle (i.e., unexciting) roundover of the
lift curve at these conditions. Going along with this,
when the R3 ice imitator (see Table 1) is added, there
is very little change in elevator hinge moment characteristics seen over the wide range of incidence angles
investigated. Pretty much the same thing holds true with
the results seen when the S2 ice imitator is added to
the 181 tailplane with normal LE. For both the un-iced
baseline conguration and the geometry with the ice
imitator added, there is no signicant change in slope/

trend seen around maximum lift conditions. In this case,


the steeping of the hinge moment rate of increase which
starts long before stall for both congurations seems to
happen a couple of degrees earlier with the ice
imitator installed. With the third set of test results
available for thinner tail geometries, the data provided
in Ref. [80] for the 8% thick business-jet tail geometry,
particularly at the high Reynolds number test condition
with the largest (rough) ice accretion evaluated (illustrated in Fig. 35), the hinge moments do have some
characteristics which start to resemble those seen with
the thicker designs (at lower Reynolds numbers), but
there are still some signicant differences remaining.
Probably the biggest difference remaining is that, in this
case, a rather abrupt increase (more positive) in hinge
moments occurs about 51 prior to stall with the un-iced
baseline rather than more coincident with it as occurs
with the thicker designs. The primary similarity with the
test results for thicker tail geometries is that the increase
which occurs with the simulated ice accretion, although
not quite as abrupt as seen with the un-iced baseline,
occurs about 561 earlier than on the un-iced baseline.
This is very different than what occurred with the ice
accretions evaluated in Refs. [77,78] on the two other
thinner tail geometries.
There is one other bit of related information available
on this subject which should be noted, and that is the
presentation of ight-measured stick forces obtained
with larger ice accretions on the tails of several Russian
transports (i.e., An-10, An-24, IL-18, IL-62, and Yak40) that is contained in Ref. [78]. In general, these results
illustrate the large and sudden changes that can occur,
particularly with ap extension, as well as a very
noticeable lightening of stick forces, and, for some, an
earlier sign reversal.
In concluding this section, what has again been
demonstrated herein is that there are a wide range of
possible effects depending on lifting surface geometries
and types, the severity of the ice accretions simulated,
test Reynolds numbers, etc. And, our biggest concerns
with the current database have to be that none of the test
results addressing elevator hinge moment characteristics
were obtained with the most critical ice accretions
possible, and there is no high Reynolds number data
whatsoever for thicker tail designs. These deciencies
need to be addressed if a proper understanding is to be
obtained of the most adverse possible types of control
surface problems which could be encountered in ight.

7. Effect of ground frost/ice accretions


Ground frost, which accumulates in a relatively rough
thin layer on the upper surface of wings (and horizontal
stabilizers) while the aircraft is parked, or during
subsequent taxi operations, can have a serious adverse

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

impact on aircraft aerodynamic characteristics and


operating margins if not removed before takeoff. These
adverse effects have been described by a number of
investigators such as Zierten and Hill [164], von Hengst
and Boer [165], Brumby [56], etc. In fact, as documented
in Refs. [56] and [165], a number of aircraft incidents/
accidents have been attributed to the failure to
adequately remove such so-called hoar frost accretions
from the wing upper surface. Similar accretions on the
upper surface of the tail are of much less consequence
since that is not the critical surface during takeoff.
Although the roughness characteristics of ground ice
accretions can vary signicantly depending on atmospheric conditions, etc., typical ground frost accretions
are often similar in roughness characteristics and
thickness (order of one millimeter) to the in-ight
leading-edge buildups necessary for removal by de-icing
systems. However, instead of just being conned to the
rst several percent of the wing chord on both upper and
lower surfaces as was the case with the in-ight
accretions, these ground frost accretions can cover the
entire exposed upper surface of the wing (slat, main
element, and ap). Some variations in this upper surface
coverage exist depending on whether or not the leadingedge devices (typically slats) and trailing-edge aps are
retracted during ground operation prior to takeoff in
these adverse weather conditions. If left extended, a
more extensive roughness coverage would exist. And, in
addition to aerodynamic performance effects, these
wing-upper-surface ice accretions are also of concern
for aircraft with aft-fuselage-mounted engines due to the
possibility of engine ingestion of this ice as it subsequently breaks loose. Another form of ice accretion
which can occur when an aircraft is sitting on the ground
is what is referred to as in-spar frost/ice, which most
often forms on the lower surface of wings in proximity
to fuel tanks containing cold soaked fuel, but it can
also form on the wing upper surface as well with some
congurations [166], and would have to be removed in
these cases. However, analogous to the situation with
upper surface frost on the horizontal tail, lower surface
ice accretions are of relatively minor consequence to the
aerodynamic characteristics of the wing at takeoff
conditions as shown by Bragg et al. [167]. While this
type of ice accretion could conceivably be an issue for
any aircraft that have wet tails, i.e., fuel in the tail, it is
felt that such an occurrence would be extremely rare.
Whereas there was a real scarcity of good quantitative
data available from full-scale aircraft ight testing to
help validate the aerodynamic effects of the other forms
of ice accretions considered up to here, there are,
fortunately, several sets of ight test results existing for
contemporary jet transports addressing the aerodynamic
consequences of upper-surface ground frost/ice accretions. The mere existence off these data is certainly a
clear indication of the widespread concern regarding the

751

consequences of failing to thoroughly remove these ice


accretions prior to initiating takeoff. Incidentally,
eliminating these ice accretions by applying highly
viscous ground anti-icing uids does not necessarily
remove all performance degradations. That is because
the rough residual from these uids that remains on
the wing during takeoff can itself, lead to performance
degradations, which are especially worrisome for smaller
aircraft having hard LEs.
7.1. Maximum lift reductions
Reductions in maximum lift capability are most
critical at takeoff conditions since speed margins to
stall are typically smaller at these conditions than at any
other time in the ight envelope.
7.1.1. Single-element lifting surfaces
Not surprisingly, considering that small leading-edge
disturbances are far more critical in terms of causing
reductions in maximum lift than those located aft of the
LE, maximum lift penalties for having the entire upper
surfaces of single-element lifting surfaces covered with
roughness simulating ground frost accretions are very
similar to those seen with just leading-edge roughness.
To illustrate this similarity, results from two high
Reynolds number wind tunnel tests [71,168] and an
MD-87 ight test program [169], all with the entire
airfoil or wing upper surface covered with roughness,
are presented in Fig. 62, together with the previously
established band of corresponding high Reynolds
number penalties observed for having just leading-edge
roughness (from Fig. 10). As can be seen, the penalties
with the entire upper surface covered with roughness are
well within the range established for just leading-edge
roughness effects. In fact, in one of these investigations
[71], penalties were measured for both types of roughness coverage, and the additional penalty due to having
the entire upper surface roughened was an order of
magnitude less than the basic penalty for just the
leading-edge roughness. Also conrming the non-criticality of aft-located roughness were further results from
the other wind tunnel investigation [168] which illustrated that having only the aft 50% of the airfoil upper
surface covered with roughness (k=cB104 ) resulted in
maximum lift penalties of only 1% or less. A very key
lesson provided by the rather unique set of ight test
results shown in Fig. 62 is that sizable maximum lift
penalties due to upper surface (leading-edge, etc.)
roughness do persist down to very low roughness
heights (i.e., k=c 2  105 ).
Following the persistent trend now anticipated,
available test results where the effect of upper surface
roughness on maximum lift penalties has been measured
over a range of Reynolds numbers again indicate that
low Reynolds number test results are not representative

752

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 62. Upper surface roughness (ground ice/frost) size effects on maximum lift penalty at high Reynolds numbers for single-element
lifting surfaces.

of higher Reynolds number conditions. It can be seen


from the available data summarized in Fig. 63 that the
indicated penalties at low Reynolds numbers are clearly
lower than those experienced at higher Reynolds
number conditions. That being the case, other available
low Reynolds number measurements of these upper
surface roughness effects such as those presented in
Ref. [165] at 0.9 million Reynolds number, Ref. [137] at
1.3 million, and those presented in Refs. [74,171] at
2.6 million, are of little value when needing to establish
the upper limit of penalties likely/possible at ight
conditions for a variety of aircraft types. For the record,
the Ref. [165] test results for the F27 do fall below the
lower limit of the band established at high Reynolds
numbers for just leading-edge roughness, and the
penalties indicated by the Refs. [74,172] test results for
the NACA 65A215 airfoil tend to be toward the lower
end of the band, as do the results from Ref. [137].
7.1.2. Multi-element lifting surfaces
The magnitude of reductions in maximum lift
capability caused by ground-frost-type accretions on
the upper surface of multi-element airfoils/wings at
takeoff-type conditions is well documented at high
Reynolds numbers for congurations employing leading-edge devices (primarily slats). Extensive ight test
results from Refs. [164,169] are available for four
contemporary jet transports (737, 757, 767 and MD87). In addition, there was also one 2-D high Reynolds
number wind tunnel investigation [82] which addressed
these reductions as well as just leading-edge roughness

effects, thereby permitting a direct comparison, as well


as a semispan model test of the Fokker F29 conguration [173,174] where upper surface frost/roughness
effects both with and without a slat were measured.
Incidentally, the upper surface frost (0.5 mm roughness)
coverage simulated during the 737, 757, and 767 ight
test programs assumed that the leading-edge device as
well as the ap were retracted when the frost was being
accreted, while the MD-87 ight test and the wind
tunnel test used a frost/roughness coverage that assumed
that these elements were extended during this time,
thereby resulting in a somewhat greater chordwise
extent of roughness near the LEs of the main element
and ap. This difference should be rather inconsequential, however, since these areas are not at all critical at
takeoff conditions as shown in Ref. [61].
These high Reynolds number test results for congurations with leading-edge devices extended are shown
in Fig. 64, along with the bands representing the range
of penalties observed at high Reynolds numbers for
single-element lifting surfaces for both leading-edge
roughness and having the entire upper surface covered
with roughness. In addition, the band obtained from low
Reynolds number wind tunnel tests of the 737, 757, and
767 with upper surface frost accretions is shown. Several
important conclusions can be reached based on the
results shown. First, consistent with the trends observed
with leading-edge roughness effects, percentage losses
in maximum lift incurred with upper-surface frost
accretions on multi-element lifting surfaces are
noticeable smaller than with single-element designs.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

753

Fig. 63. Reynolds number effects on maximum lift penalty caused by upper surface roughness (ground ice/frost) on single-element
airfoils [170].

Fig. 64. Maximum lift penalties caused by upper surface frost/roughness on multi-element high-lift congurations with leading-edge
devices at high Reynolds numbers.

754

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Next, percentage losses in maximum lift capability for


multi-element designs caused by upper surface frost
decrease with increasing ap deection (i.e., increasing
basic maximum lift). It is also very signicant to note
that the low Reynolds number wind tunnel test results
for the 737, 757, and 767 indicate that comparable
penalties for upper surface frost accretions do not occur
until an order of magnitude higher nondimensional
roughness height than happens at high Reynolds
number ight conditions, another indication that
extreme caution should be used when attempting to
directly apply low Reynolds number test results to
higher Reynolds number ight conditions. Another
important conclusion to note is that the reductions in
maximum lift do persist down to quite-small roughness
sizes, indicating that even very thin coatings of frost are
of concern. With regard to the test results for the F29, it
can be seen that they tend to fall in line with the wind
tunnel test results for the 737, 757, and 767, perhaps
indicating that an MAC Reynolds number of 5 million
may not have yielded results representative of ight in
this case.
It was indicated earlier that maximum lift penalties
at takeoff conditions are most critical because speed
margins to stall are typically the lowest at these
conditions. To illustrate the seriousness of any
reductions in maximum lift capability at takeoff
conditions, the maximum lift penalty which would
result in a stall at minimum takeoff speeds (i.e., 13%
higher than 1-g stall speeds) assuming no overspeed is
employed is highlighted in Fig. 64. It can be seen that if
overspeed is not employed on takeoffs where upper
surface frost is present (at the accumulations addressed),
then half or more of the stall margin can be eroded.
Even more of the margin would be eroded with larger
accumulations.
Lastly, although the data is not shown, the high
Reynolds number wind tunnel test results [82] indicate
that the additional penalty for having the entire upper
surface covered with frost/roughness is less than half the
penalty for just the LE roughness. Furthermore,
additional results from this same test program indicate
that if frost was just removed from the slat
upper surface, then the reduction in maximum
lift caused by the remainder of the frost was less
than a third of the penalty incurred when the
whole upper surface was rough, again indicating that
the focus in ground de-icing must start with the LE of
the wing.
Unfortunately, there is much less high Reynolds
number data available addressing upper surface frost/
roughness effects on the expected-to-be-more-critical
wing designs not having leading-edge devices (i.e.,
hard LEs). Unlike the situation for congurations
with leading-edge devices, there are no ight test results
available. What is available are results at (MAC)

Reynolds numbers around 5 million from semispan


model tests of the Fokker 100 and (aforementioned) F29
[173,174]. This is certainly not the most condenceinspiring high Reynolds number database, especially
considering the indications provided by the F29 test
results at this Reynolds number with a LE slat.
However, be that as it may, the results from these tests
plus the available and applicable database from tests at
lower Reynolds numbers are provided in Fig. 65.
Incidentally, when tests were conducted both with and
without leading-edge devices, both sets of results are
shown so as to illustrate the trend of higher (percentage)
penalties for hard leading-edge designs. In these cases,
test results with leading-edge devices are shown as solid
symbols. From the results presented in Fig. 65, there
are some important conclusions to be drawn. For
one, in both cases where congurations both with
and without leading-edge devices (slats) were evaluated,
the (percentage) maximum lift losses caused by having
the entire upper surface covered with frost/roughness
are clearly higher for the hard LE designs. It is also
quite apparent from all the data presented for hard
leading-edge designs that the level of maximum lift
losses incurred with these designs is higher than seen in
Fig. 64 for the designs with leading-edge devices.
Consequently, based on these test results, much more,
if not all, of the stall margin is eroded for hard
leading-edge designs by modest accumulations of upper
surface frost/roughness unless overspeed is employed.
Whether this situation might be even more critical or
require even more overspeed (to avoid) at ight
Reynolds numbers remains to be seen, and should be
investigated.
Somewhat similar to the results for congurations
having leading-edge slats, the limited (low Reynolds
number) database available for congurations having
hard LEs also indicate that the maximum lift
penalties are much reduced if just the forward part
of the airfoil is kept clean. In fact, the reductions
indicated are greater than the corresponding reductions seen (at low Reynolds numbers) in the same
investigations with congurations having leading-edge
slats. For example, the results presented for the NACA
652A215 airfoil (with ap) in Refs. [74,172] at a
Reynolds number of 2.6 million indicate that if the
forward 18% of the wing chord is kept clean, the
maximum lift penalties are only about 13214 of those
incurred with the whole upper surface roughened. When
only the forward 5% of the wing chord was kept clean,
the indicated reductions are somewhat less, but still
signicant. So, at least based on low Reynolds number
test results, it would seem clear that the absolute highest
priority in ground de-icing should be in ensuring that
the LEs of hard leading-edge designs are kept
clean. Accident statistics would also lead to the same
conclusion.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

755

Fig. 65. Comparison of maximum lift penalties caused by upper surface frost/roughness on multi-element high-lift geometries with and
without leading-edge devices.

7.2. Stall angle reductions


7.2.1. Single-element lifting surfaces
Since the maximum lift reductions caused by uppersurface ground frost/ice accretions fall in the same range
or band as those caused by the initial leading-edge ice
(roughness) accretions, it would be expected that the
associated stall angle reductions would be similar as
well. And, that is indeed the case, as can be seen from
the available high Reynolds number measurements of
stall angle reductions for upper-surface frost accretions
shown in Fig. 66. In examining the data, it is of interest
to note that the one set of test results for a 3-D
conguration, namely the MD-87 ight results, do not
stand out above the 2-D data as was the case for the
results with the 3-D tail with leading-edge roughness (see
Fig. 15). This difference can be explained, at least in
large part, by the signicantly higher aspect ratio of the
MD-87 wing. Hence, in terms of accounting for the
effects of lifting surface aspect ratio on stall angle
reductions, it is clearly more important for relatively low
aspect ratio geometries.
There are also a number of low Reynolds number test
results available addressing stall angle reductions caused
by upper surface ice/frost on single-element geometries
from most of the References mentioned in conjunction
with the low Reynolds number maximum lift losses

illustrated in Fig. 63. However, since the maximum lift


penalties measured at these low Reynolds numbers are
not believed to be representative of higher Reynolds
number ight conditions, neither would the indicated
stall angle reductions. Consequently, these results are
omitted herein. Although not shown, there is one most
noteworthy aspect of all the low Reynolds number
indications of stall angle reductions caused by upper
surface roughness, and that is the results for the one 3-D
geometry available [137], which had an aspect ratio of
just under ve, were at the upper end of the band of
indicated stall angle reductions. This result, although
conceivable a low Reynolds number anomaly in this
case, is consistent with the trend highlighted earlier at
high Reynolds numbers for LE roughness effects, and
would seem to support the notion that aspect ratio
effects are most important for relatively low aspect ratio
designs.
7.2.2. Multi-element lifting surfaces
As was the case with the maximum lift reductions
caused by ground-frost-type accretions on the upper
surface of multi-element airfoils/wings with leading-edge
devices at high Reynolds number takeoff-type conditions, a representative range of corresponding stall angle
reductions for these congurations is also well documented based on the four sets of ight test results on

756

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

Fig. 66. Upper surface roughness (ground ice/frost) size effects on stall angle reductions at high Reynolds numbers for single-element
lifting surfaces.

contemporary jet transports, and the one high Reynolds


number wind tunnel investigation of these effects. The
relatively narrow band of stall angle reductions determined from these ve sets of test results are presented in
Fig. 67 along with the band of corresponding high
Reynolds number test results (from Fig. 66) for singleelement airfoils. Similar to the results presented earlier
for stall angle reductions with initial in-ight leadingedge ice accretions, the reductions experienced with
multi-element designs having leading-edge devices tend
to be slightly larger than the corresponding singleelement designs. When examining the results shown
in Fig. 67, it is worth remembering that stall angle
reductions of three degrees were experienced in-ight
with the extremely small roughnesses evaluated on the
MD-87, and a reduction of over four degrees was
experienced on the 737 with just 0.5 mm roughness. This
certainly raises a concern over the possibility of stall
angle reductions noticeably higher than these on smaller
aircraft with similar relatively small frost (roughness)
accretions.
For congurations not having leading-edge devices,
the (all low Reynolds number) database available
shedding any light on stall angle reductions caused by
ground-frost-type accretions is very limited. In addition,
to make matters even worse, some of the (limited) results
available indicate reductions several times more than
those seen for congurations with leading-edge devices
at high Reynolds numbers. The results available for these

hard leading-edge designs are illustrated in Fig. 68,


and, as was done in Fig. 65 for the related maximum lift
penalties, where corresponding results are available with
leading-edge devices, these results are shown as solid
symbols. From the results shown, it would be difcult to
nd any prevalent trend in the results for hard LEs
versus those for congurations having leading-edge
devices, again illustrating that the higher (percentage)
maximum lift penalties for hard leading-edge
designs are primarily a consequence of the lower
basic (i.e., clean) maximum lift levels of these
designs. It can also be seen that in contrast to
the relatively narrow band of reductions indicated
at high Reynolds numbers for congurations with
leading-edge devices, the low Reynolds number
database for geometries both with and without
leading-edge devices portray a much broader range of
potential stall angle reductions caused by small
upper-surface ground frost/ice roughness accretions. In
fact, stall angle reductions observed in these low
Reynolds number tests range from o11 to over 101, an
alarmingly wide band, with relatively inconsequential
effects at the lower end, to potentially disastrous effects
at the upper end. Certainly, some well conceived high
Reynolds number testing is warranted, particularly for
the worst hard leading-edge conguration(s), to
document the reductions likely at ight conditions if
such frost/ice accretions are not removed prior to
takeoff.

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

757

Fig. 67. Upper surface roughness size effects on stall angle reductions for multi-element high-lift congurations with leading-edge
devices at high Reynolds numbers.

Note:
All configurations have upper surface roughness
0-100%.

Fig. 68. Comparison of stall angle reductions caused by upper surface frost/roughness on multi-element high-lift geometries with and
without leading-edge devices at low Reynolds numbers.

758

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

7.3. Drag penalties


Because of the much longer chordwise extent of the
roughness brought about by upper-surface ground frost/
ice accretions, it would seem reasonable to expect that
associated drag penalties on a given conguration would
be somewhat higher than those incurred with initial inight leading-edge type ice accretions having the same
roughness height. However, unfortunately, there does
not appear to be any data set available which would
permit direct substantiation of this expectation.
7.3.1. Single-element lifting surfaces
There appear to be only two (known) sources of
available test results which have any data, high or low
Reynolds number, potentially useable for assessing the
drag penalties caused by upper-surface ground frost/ice
accretions on single-element geometries. They are some
very low Reynolds number (i.e., 0.9 million) results for
the Fokker F27 [165], and some very limited higher
Reynolds number results for the NACA 0012 and RAF
34 airfoils from Ref. [168]. Unfortunately, neither one of
these sets of test results are very helpful. The results for
the F27 are denitely suspect because of the low test
Reynolds number, since, as was shown earlier in Figs. 20
and 21 when addressing Reynolds number effects on the
drag increases caused by initial (and residual) leadingedge ice/roughness accretions, penalties measured at
such a low Reynolds number are typically not representative of higher Reynolds number ight conditions,
nor can the direction of the Reynolds number trend be
predicted. As it is, the indicated drag penalties are right
at the bottom of the band of results for leading-edge
roughness effects shown in Figs. 18 and 19 (even though
the nondimensional roughness height employed is larger
in this case by a factor of three.
The test results for the two airfoils reported in
Ref. [168], albeit at higher Reynolds numbers, are also
of limited value because only minimum parasite drag
penalties are provided. There are no results given for the
more critical lower speed conditions. The minimum
parasite drag penalties in this case also fall toward the
bottom of the band shown in Fig. 18, although part of
the reason for this is likely the lower nondimensional
roughness height utilized in this case (i.e., 1.25  104
versus 4.6  104). Interestingly, in this case, the
variations in drag penalties with increasing Reynolds
numbers for the two airfoils have opposite trends (i.e.,
decreasing with the RAF 34, but increasing with the
0012). Consequently, with this very limited amount of
test data, it is not possible to make any really meaningful
quantitative assessments of the range of drag penalties
possible with these upper surface ground frost/ice
accretions, but it would be surprising if this range was
not at least as large as the very wide band documented
for the leading-edge type roughness accretions.

7.3.2. Multi-element lifting surfaces


Flight test results available for the 737, 757 and 767
from Ref. [164] together with test results from one 2-D
high Reynolds number wind tunnel test [82] provide a
very good indication of the range of absolute drag
penalties incurred with upper surface ground frost on
congurations having leading-edge devices. These ight
test results for a roughness height of 0.5 mm, and wind
tunnel results for a nondimensional roughness height
(k=c) of about 0.7  104, are shown in Fig. 69 at both
minimum drag conditions and at representative takeoff
V2 speeds. It can be seen that in all cases, as would be
expected because of the more adverse ow conditions on
the upper surface of the wing/airfoil at this condition,
the drag penalties at takeoff V2 speeds are always
signicantly higher than those incurred at minimum
drag conditions. At takeoff V2 speeds, the ightmeasured drag penalties vary from about 0.01 up to
over 0.04, compared to a range of from about 0.005 up
to about 0.012 at minimum drag conditions. It is
comforting to see that the 2-D high Reynolds number
wind-tunnel test results do appear to be consistent with
the ight measurements, especially when differences in
reference area are taken into account. Finally, it is of
interest to note that the drag penalties caused by these
relatively small upper surface frost/ice accretions are,
again as would be expected, much less than the penalties
caused by large in-ight glaze ice accretions.
With it already rather conclusively established that
the frost and ice accretions close to the LE of lifting
surfaces are clearly the most critical in terms of causing
reductions in the maximum lift capability and associated
stall angles, drag measurements made with an alternate
roughness coverage on the slat during the 737 ight test
program [164] give us a good indication that the same is
also true with regard to drag penalties at takeoff climb
conditions. For this alternate roughness coverage, the
simulated frost was partially removed from the slats to
evaluate the benets of the 737 ground mode slat
thermal anti-ice (TAI). The subsequent drag measurements obtained with this alternate coverage at the ap 1
and 5 riggings, which had suffered the largest absolute
drag penalties with the basic roughness coverage at
takeoff climb speeds, indicated that these drag penalties
were, on average, just about cut in half. This is yet
another set of (high quality) test results that provide
even more substantiation (if it were needed) that the
focus in ground de-icing must start with the LE of the
wing.
In contrast to the good, high-quality database
available for high-lift congurations with slats, there is
very little data of any kind available for assessing these
drag penalties for high-lift systems having hard LEs.
And, what little is available, such as the very low
Reynolds number F27 test results reported in Ref. [165],
are, by now, more than just suspect. So, as was the

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

759

Fig. 69. Drag penalties caused by upper surface ground frost/ice (roughness) accretions on multi-element takeoff geometries.

situation with single-element lifting surfaces, there is


really no good basis at this time for estimating the drag
penalties likely to be caused by ground-frost type
accretions on these congurations at critical takeoff
conditions. This is obviously another big hole in the
database that needs to be plugged.
Lastly, with regard to any drag penalties that would
be caused by wing-lower-surface in-spar frost accretions at takeoff conditions, available high Reynolds
number test results [82] indicate that these accretions are
of little practical consequence. With the same (nondimensional) roughness height, the drag penalty indicated for lower surface frost was about a factor of 50 less
than that observed with upper surface frost/roughness.
Even with a roughness height on the lower surface about
six times greater than that used on the upper surface, the
indicated drag penalty for the lower surface roughness
was about a factor of 15 less.

8. Summary and conclusions


The foregoing is a systematic and comprehensive
review, correlation, and assessment of wind-tunnel and
ight-test measurements available in the public domain

which address aerodynamic performance and control


degradations caused by various types of ice accretions
on the lifting surfaces of xed wing aircraft. The intent
has been to dene the range of possible consequences
which can occur at ight conditions, especially the worst
which could be encountered. A part of this assessment
has also been to identify critical voids in the available
database that need to be rectied. CFD techniques have
not been used herein to either augment or help correlate
existing test results because of known inherent limitations of current state-of-the-art techniques such as
RANS methods, etc., most specically with regard to
the inability to effectively predict separation onset
characteristics, which are at the core of establishing
the incremental effects of ice accretions.
Four types of ice accretions have been considered.
The rst category addressed are the initial small
accumulations that occur on the LE of aerodynamic
surfaces either before the activation of the ice protection
system and/or between de-icing system cycles. Next are
the runback and/or ridge ice accretions which can form
just aft of the ice protection system, especially during
large droplet icing encounters. Following these, the large
and often-more-irregularly shaped ice accretions which
form during longer encounters either as a consequence

760

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

of not having an ice protection system, and/or having a


failure of the ice protection system, are studied. The last
category considered are the ground frost/ice accretions
which can accumulate on the upper surface of wings
while the aircraft is parked, or during subsequent taxi
operations. In addressing the various consequences of
these different types of ice accretions, efforts have
focussed on assessing resulting maximum lift reductions,
the corresponding stall angle reductions, resulting drag
penalties, and trailing-edge control surface anomalies.
Based on this review, the four types of ice accretions
have been classied as follows:
*

Dangerous because of possibility of being underestimated and/or misunderstoodFClearly, heading


this category are the initial small leading-edge ice
(roughness) accretions and inter-cycle ice buildups
occurring on both wings and tails. Misconceptions
abound in terms of the thickness of this initial (or
inter-cycle) ice accretion which can be allowed before
any serious degradation of aerodynamic effectiveness
occurs. However, what this review and assessment
has again highlighted is that there are high Reynolds
number test results available which demonstrate that
even very small (i.e., 23 mils) buildups can result in
very noticeable reductions in maximum lift capability
and corresponding stall angles. Further, for the ice
thicknesses required for operation of state-of-the-art
de-icing systems, maximum lift reductions of up to
40% can occur for single-element surfaces.
Dangerous because of potential for catastrophic
reductions in aerodynamic effectivenessFLeading
the way in this class are the runback/ridge ice
accretions which can form just aft of a leading-edge
ice protection system due either to the impingement
of larger-than-normal water droplets in this area, or
by the runback and subsequent freezing of water
which is not evaporated by a leading-edge anti-icing
system. Initial (low Reynolds number) test results
indicate the possibility of maximum lift losses in
excess of 80% for some single-element geometries.
Clearly, more needs to be known about the
consequences of these types of ice accretions at full
scale ight conditions.
Dangerous because upper limits of potential aerodynamic consequences are not really definedFThis
category belongs to the larger glaze ice accretions
which, ironically, have been the focus of a majority of
studies conducted to date addressing the aerodynamic consequence of ice accretions. To date,
maximum lift losses approaching 60% for singleelement lifting surfaces have been indicated for these
accretions. However, it is clear that the maximum
possible reductions in aerodynamic effectiveness have
yet to be adequately dened. The penalties associated
with such accretions at ight-like Reynolds numbers,

with appropriate wind tunnel model installations,


and with simulations representing the most severe
icing environment in terms of ambient temperature,
droplet sizes, LWC, duration, and accretion altitude
and speed, have yet to be established.
Dangerous because of portion of flight operation
envelope involvedFThis is clearly the province for
ground frost/ice accretions that are not removed
prior to takeoff, since margins to stall are typically at
a minimum during takeoff compared to the rest of
the ight envelope. The data available indicate that
the penalties for these accretions are very similar to
those incurred with the initial in-ight leading-edge
and inter-cycle buildups. However, because of the
reduced margins at takeoff, these ground frost/ice
accretions can result in much if not all of the margin
being eroded/erased.

In reviewing the wide variety of test results available


for the foregoing types of ice accretions, it has been
shown that the aerodynamic consequences of these ice
accretions, in addition to being generally dependent
upon the ice accretion features, are also strongly
dependent on the geometry of the various baseline
lifting surfaces. It has been shown that a given type/size
of ice accretion can cause quite disparate effects on
different lifting surfaces. Examples of this include the
following:
*

Adverse ice accretion effects, especially those associated with stall characteristics, are much larger on
non-dimensionally thicker (single element) lifting
surfaces which have baseline maximum lift coefcient
capabilities in excess of 1.0. Corresponding ice
accretion effects on thinner surfaces with lower
maximum lift capabilities are often minimal.
Unfortunately, the superior performing thicker
surfaces are much more prevalent on xed wing civil
aircraft types of most interest.
Smaller, dimensionally thinner surfaces (with smaller
leading-edge radiuses) such as tails are more efcient
ice collectors than (dimensionally) thicker surfaces
such as wings. Consequently, (percentage) degradations on these smaller surfaces are often larger for the
same icing encounter. This can be an important issue
when addressing tail stall concerns.
Percentage penalties in terms of maximum lift
reductions and drag increases are typically roughly
inversely proportional to the number of surface
elements, (i.e., the baseline performance capabilities).
For example, congurations with leading-edge devices such as employed on most larger transport
aircraft are the least impacted. Conversely, congurations having hard LEs typical of most
smaller transports/aircraft typically incur noticeably
higher penalties. And, it follows that single-element

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

geometries are the worst. Accident statistics clearly


substantiate this order of criticality.
Testing conditions and ice accretions simulation issues
also emerged as critical factors in establishing the
appropriate magnitude of aerodynamic penalties caused
by the various ice accretions addressed. For example,
whenever test results were available for both simulated
shapes (computational or otherwise) and actual ice
accretion shapes, roughness, etc., more often than not,
the penalties were greater with the actual ice accretion.
With regard to testing issues, much evidence exists
to rmly establish that testing at a sufciently high
Reynolds number to adequately represent ight conditions is a critical requirement if the measured incremental aerodynamic penalties caused by the various
forms of ice accretions are to be representative/meaningful. Many low Reynolds number test results (i.e.,
usually below about 5 million) have produced penalties
noticeably lower than would occur at ight conditions,
certainly not a desired result. While Reynolds number
effects on the performance (and control) of the baseline
un-iced geometry predominate in many cases, especially
when addressing maximum lift characteristics, there are
also a number of examples of Reynolds number effects
on the iced geometry as well, particularly for smaller ice
accretions. Schemes aimed at extensively utilizing the
absolute values of low Reynolds number test results of
the maximum lift characteristics for larger ice shapes as
a consequence of the apparent general insensitivity of
these results to variations in Reynolds number could be
very risky. This is because it is not known as to whether
or not this insensitivity seen in 2-D test results could
have came about as a consequence of strong sidewall
boundary layer interaction effects controlling the stall
rather than the pure aerodynamics of the larger ice
shapes. Also, trying to use this approach with 3-D
geometries where different spanwise stations may be
critical at different Reynolds numbers, or where very
different ow physics may be controlling at high and
low Reynolds numbers, would present further major
challenges.
Other ndings of signicance from this review involve
ice accretion effects on control surface characteristics
and drag penalties. With regard to control surface
anomalies which can be critical in tail stall situations,
these anomalies (including reversals) are in most all
cases strictly a consequence of an earlier stall occurring
with the contaminated surface. There are no real
mysteries and/or subtleties involved. With regard to
drag penalties associated with the larger glaze ice
accretions, it is concluded based on the results presented
herein that long-standing efforts aimed at identifying a
single drag penalty based on parameters dening the
geometry of the ice shape are misguided. Not only is
there not a single penalty for a given ice shape on

761

a particular lifting surface, even if one did exist it would


be different for each surface geometry.
Lastly, after reviewing many test results and associated commentaries, a number of which have been
available for many decades, it is concluded that
important lessons often learned by one generation
regarding the various aerodynamic consequences of
various forms of ice accretions have not been adequately
passed on to, and/or accepted by, later generations,
leading to accidents which seemingly could have been
avoided if past lessons learned had been disseminated,
taken advantage of, and acted upon. Prime examples of
these failures include knowledge of tail icing consequences and procedures for minimizing the likelihood of
tail stall (and associated control surface anomalies), the
large effect of possible runback/ridge ice accretions, the
signicant adverse consequences of small initial and
inter-cycle ice accretions associated with the use of deicing systems, and the importance of having an adequate
test Reynolds number. In addition, lessons learned in
terms of what constitutes acceptable wind tunnel/model
installations for the determination of the consequences
of various ice accretions have often been ignored for any
number of reasons. Clearly (continuing) education
programs designed to ensure that important lessons
learned in the past are adequately disseminated, utilized,
and acted upon, are a must.

9. Recommendations
Presuming that the top level objective is to eliminate
aircraft accidents caused by icing which could have been
avoided if ight crews, operators, etc., had been alert to
the potentially serious consequences of various forms of
ice accretions, and hence taken appropriate precautionary actions, there are some important steps remaining
to be taken. Foremost amongst these should be the
following:
*

Establishment of continuing/enduring training programs by government research and regulatory


organizations to ensure that all designers, operators,
ight crews, etc., are actually aware of all lessons
which have been learned regarding the potentially
serious consequences of almost any possible ice
accretions, and know what actions are needed in
case such ice buildups do occur.
To be sure that the correct possible aerodynamic
consequences are adequately dened, there are
additional well conceived (primarily high Reynolds
number) experimental programs still needing to be
carried out.
Actions need to be taken to ensure accountability for
the sizeable adverse aerodynamic consequences of the

762

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

initial and inter-cycle ice accretions associated with


the use of de-icing systems.
Decisions need to be made regarding how to deal
with the consequences of possible large droplet icing
encounters.
A strong focus should be placed on developing new
certifiable ice detection systems/concepts that can
reliably alert ight crews as to the onset of any ice
accretions.

As indicated, there are a number of high priority, well


conceived wind-tunnel test programs still needed to
ensure that the upper limit of realistic aerodynamic
penalties which could be incurred by representative
lifting surfaces at ight conditions are properly quantied. These are as follows:
*

For the small initial ice accretions which will always


exist when de-icing systems are employed, or when
there is a delay in activating anti-icing systems, it
should be determined if penalties which would be
incurred with the densely distributed (droplet) hemispheres believed to occur in natural icing conditions
are much different from those indicated by the
existing (large) database based almost exclusively
on the use of less dense distributions of sharp
particles to represent these initial accretions (roughness). A representative range of droplet sizes should
be considered.
Meaningful high Reynolds number test results (i.e.,
without uncorrectable wall effects, etc.) are needed
for representative inter-cycle ice accretions in order
to properly establish the magnitude of aerodynamic
penalties inherent in the use of de-icing systems
(which are all too often ignored).
Since existing low Reynolds number test results
indicate potentially catastrophic aerodynamic penalties can result from the runback/ridge ice accretions
believed to be formed during large droplet icing
encounters, it is clearly imperative that both the
shapes, etc. of such ice accretions possible as well as
the magnitude of the resulting aerodynamic penalties
at ight-Reynolds-number-type conditions be determined. However, since properly representing the
state of the leading-edge boundary layer is most
certainly a critical requirement for establishing
believable aerodynamic consequences for these accretions, the aerodynamic testing required for many
vehicle types can realistically only be carried out in
large cryogenic test facilities such as the NTF and/or
ETW where leading-edge attachment line conditions
existing in ight on swept wings can be duplicated.
Although the existing available database of aerodynamic penalties indicated for large glaze ice
accretions has enabled many very useful insights into
the magnitude and form of such penalties, it is,

however, lacking in terms of identifying the likely


upper limit of penalties which could be encountered
in ight with the most critical ice accretions possible.
That is both because of usually not really representing the most critical ice accretion conditions possible,
and often not having unchallengable testing conditions, model installations, etc. Consequently, some
new meaningful high Reynolds number test results
are needed in order to better document the upper
limit of penalties possible with large glaze ice
accretions having the most critical accretion conditions in terms of droplet sizes, LWC, ambient
temperature, aircraft speed and altitude, duration of
encounter, etc. This is clearly an area where large
droplet effects need to be established.
Meaningful high Reynolds number test results are
also needed for takeoff geometries having hard
LEs in order to better establish (and illustrate) the
potentially catastrophic effects of not adequately
removing any ground frost/ice accretions with these
congurations.

For each of the foregoing categories of additional


tests needed, a sufcient number of lifting surface
geometries should be addressed so as to represent the
range of penalties possible considering both dimensional
as well as nondimensional differences with existing as
well as possible new designs.

Acknowledgements
The authors wish to acknowledge and express our
utmost appreciation for the very valuable contributions
of Gene Hill and Jim Riley from the FAA, Mike Bragg
from the University of Illinois, and our editor, Barry
Haines, in terms of the constant encouragement to
continue with this (long) review, supplying additional
new pertinent data and other information, and in
providing most helpful and insightful critiques of draft
versions of the paper throughout its preparation. We
would also like to recognize the signicant value added
to this review by Gene Addy from NASA Glenn in
providing yet-to-be published high Reynolds number
test results for a variety of ice accretion shapes.
Note to reader
While the authors have made every effort to
review and include herein all known available and
pertinent test results applicable to dening the important
consequences of various types of ice accretions on
aircraft aerodynamics, it is very possible because of our
language and other limitations that we may have missed
some. If there are additional pertinent data that are
available which would be useful in either modifying or

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

conrming positions established herein, researchers or


others knowledgeable of such information are encouraged to contact the authors. Then, if appropriate, an
addendum to this paper would be prepared reecting the
additional lessons learned from these new results.

References
[1] Anon, Aircraft Accident Report, ATR Model 72212,
Roselawn, Indiana, October 31, 1994, NTSB/AAR-96/01,
vol. 1, 1996.
[2] Marwitz J, Politovich M, Berstein B, Ralph F, Neiman P,
Ashenden R, Bresch J. Meterological conditions associated with the ATR72 aircraft accident near Roselawn,
Indiana on october 31, 1994. Bull Am Meteorol Soc
1997;78(1):4152.
[3] Kind RJ, Potapczuk MG, Feo A, Golia C, Shah AD.
Experimental and computational simulation of in-ight
icing phenomena. Prog Aerosp Sci 1998;34:257345.
[4] Hansman RJ, Turnock SR. Investigation of surface water
behavior during glaze ice accretion. AIAA J Aircr
1989;26(2):1407.
[5] Hansman RJ. Micro physical factors which inuence ice
accretion. Proceedings of the First Bombardier International Workshop on Aircraft Icing/Boundary-Layer
Stability and Transition, Montreal, Canada, 1993.
p. 86103.
[6] Wright WB. Users manual for the improved NASA lewis
ice accretion code LEWICE 1.6. NASA CR-198355,
1995.
[7] Cebeci T. Calculation of ow over iced airfoils. AIAA J
1989;27(7):85361.
[8] Chung J, Choo Y, Reehorst A, Potapczuck M, Slater J.
Navier-stokes analysis of the oweld characteristics of
an ice contaminated aircraft wing. AIAA Paper 99-0375,
1999.
[9] Joongkee C, Reehorst AL, Choo YK, Potapczuk MG.
Effect of airfoil ice shape smoothing on the aerodynamic
performance. AIAA Paper 98-3242, 1998.
[10] Langmuir I, Blodgett KB. A mathematical investigation
of water droplet trajectories. Army Airforce TR No.
5418, 1946.
[11] Bourgault Y, Habashi WG, Dompierre J, Boutanios Z,
DiBratolomeo W, An Eulerian approach to supercooled
droplets impingement calculations. AIAA Paper 97-0176,
1997.
[12] Lozowski EP, Oleskiw MM. Computer modeling of timedependent rime icing in the atmosphere. CRREL Report
83-2, USAF, 1983.
[13] Ruff GA, Berkowitz BM. Users manual for the NASA
lewis ice accretion code (LEWICE). NASA CR-185129,
1990.
[14] Cransdale JT, Gent RW. Ice accretion on aerofoils in
two-dimensional compressible owFa theoretical model.
RAE TR 82128, 1983.
[15] Brunet L. Conception et discussion dun modele de
formation du givre sur des obstacles varies. ONERA
Note Technique 1986-6, 1986.

763

[16] Brun EA. Icing problems, recommended solutions.


AGARDograph, 16, 1957. p. 11.
[17] Potapczuk M, Gent R, Guffond D. Review, validation
and extension of ice accretion prediction codes. AGARD,
AR-344, 1997. p. 5.15.13.
[18] Brahimi MT, Tran P, Chocron D, Tezok F, Paraschiviou
I. Effect of supercooled large droplets on ice accretion
characteristics. AIAA Paper 97-0306, 1997.
[19] Trunov OK. Icing of aircraft, the means of preventing it,
Translated from Russian, foreign technology division.
WPAFB, Ohio, 1967. p. 14.
[20] Olsen W, Walker E. Experimental evidence for modifying
the current physical model for ice accretion on aircraft
surfaces. NASA TM 87184, 1986.
[21] Shin J, Wilcox P, Chin V, Sheldon D. Icing tests on an
advanced two-dimensional high-lift multi-element airfoil.
AIAA Paper 94-1869, 1994.
[22] Miller D, Shin J, Sheldon D, Khodadoust A, Wilcox P,
Langahls T. Further investigation of icing effects on an
advanced high-lift multi-element airfoil. AIAA Paper 951880, 1995.
[23] Hill EG. private communication, February 2001.
[24] Addy HE. Private communication, July 2000.
[25] Tribus MV, Young GBW, Boelter LMK. Analysis of heat
transfer over a small cylinder in icing conditions on
Mount Washington. ASME Trans 1949;70:8716.
[26] Messinger BL. Equilibrium temperature of an unheated
icing surface as a function of airspeed. J Aero Sci
1953;20(1):2942.
[27] Bilanin AJ, Anderson DN. Ice accretion with varying
surface tension. AIAA Paper 95-0538, 1995.
[28] Anderson DN. Rime-, mixed and glaze ice evaluations of
three scaling laws. AIAA Paper 94-0718, 1994.
[29] Anderson DN. Methods for scaling icing test conditions.
AIAA Paper 95-0540, 1995.
[30] Bragg MB, Lee S, Henze CM. Heat transfer and
freestream turbulence measurements for improvement
of the ice accretion physical model. AIAA Paper 97-0053,
1997.
[31] Shin J. Characteristics of surface roughness associated
with leading-edge ice accretion. AIAA J Aircr
1996;33(2):31621.
[32] Kerho MF, Bragg MB. Airfoil boundary-layer development and transition with large leading-edge roughness.
AIAA J 1997;35(1):7584.
[33] Morgan HL, Ferris JC, McGhee RJ. A study of high-lift
airfoils at high Reynolds numbers in the Langley lowturbulence pressure tunnel. NASA TM-89125, 1987.
[34] Bilanin AJ. Proposed modications to ice accretion/icing
scaling theory. AIAA J Aircr 1991;28(6):3539.
[35] Anderson DN. Evaluation of constant-weber-number
scaling for icing tests. AIAA Paper 96-0636, 1996.
[36] Hauger HH, Englar KG. Analysis of model testing in an
icing wind tunnel. Report SM14933, Douglas Aircraft
Company, 1954.
[37] Sibley PJ, Smith Jr., RE. Model testing in an icing wind
tunnel. Report LR10981, Lockheed Aircraft Corporation, 1955.
[38] Dodson EO. Scale model analogy for icing tunnel testing.
Document D66-7976, Boeing Airplane Company, Transport Division, 1962.

764

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

[39] Jackson ET. Development study: the use of scale models


in an icing tunnel to determine the ice catch on a
prototype aircraft with particular reference to concorde,
SST/B75T/RMMcK/242, British Aircraft Corporation
Ltd, Filton Division, 1967.
[40] Armand C, Charpin F, Fasso G, Leclere G. Techniques
and facilities used at the Onera Modane centre for icing
tests. AGARDograph AR-127, 1978.
[41] Ruff GA. Analysis, verication of icing scaling equations.
AEDC TR-85-30, vol 1 (Rev), 1986.
[42] Bilanin AJ, Anderson DN. Ice accretion with varying
surface tension. AIAA Paper 95-0538, 1995.
[43] Anderson DN. Rime-, mixed and glaze ice evaluations of
three scaling laws. AIAA Paper 94-0718, 1994.
[44] Anderson DN, Ruff GA. Evaluation of methods to select
scale velocities in icing scaling testing. AIAA Paper 990244, 1999.
[45] von Glahn UH. Use of truncated apped airfoils for
impingement, icing tests of full-scale leading-edge sections. NACA RM E56E11, 1956.
[46] Saeed F, Selig S, Bragg MB. A hybrid airfoil design
method to simulate full-scale ice accretion throughout a
given Cc-range. AIAA Paper 97-0054, 1997.
[47] Saeed F, Selig MS, Bragg MB. Design of subscale airfoils
with full-scale leading edges for ice accretion testing.
AIAA J Aircr 1997;34(1):94100.
[48] Saeed F, Selig S, Bragg MB. Hybrid airfoil design
procedure validation for full-scale ice accretion simulation. AIAA J Aircr 1999;36(5):76976.
[49] Saeed F, Selig MS, Bragg MB. Hybrid airfoil design
method to simulate full-scale ice accretion throughout a
given a range. AIAA J Aircr 1998;35(2):2339.
[50] Tezok F, Brahimi MT, Paraschiviou I. Investigation of
the physical processes underlying the ice accretion
phenomena. AIAA Paper 98-0484, 1998.
[51] Lynch FT, Valarezo WO, McGhee RJ. The adverse
aerodynamic impact of very small leading-edge ice
(roughness) buildups on wings and tails. AGARD-CP496, Paper 12, 1991.
[52] Valarezo WO, Lynch FT, McGhee RJ. Aerodynamic
performance effects due to small leading-edge ice (roughness) on wings and tails. AIAA J Aircr 1993;30(6):80712.
[53] Paschal K, Goodman W, McGhee R, Walker B, Wilcox
PA. Evaluation of tunnel sidewall boundary-layer-control
systems for high-lift airfoil testing. AIAA Paper 91-3243,
1991.
[54] Lynch FT, Crites RC, Spaid FW. The crucial role of wall
interference, support interference, and ow eld measurements in the development of advanced aircraft congurations. AGARD CP-535, 1993.
[55] Brumby RE. The effect of wing ice contamination on
essential ight characteristics. SAE Aircraft Ground DeIcing Conference, 1988.
[56] Brumby RE. The effect of wing ice contamination on
essential ight characteristics. AGARD CP-496, Paper 2,
1991.
[57] Valarezo WO, Dominik CJ, McGhee RJ, Goodman WL.
High Reynolds number conguration development of a
high-lift airfoil. AGARD CP-515, 1992.
[58] Valarezo WO. Topicsin high-lift aerodynamics. AIAA
Paper 93-3136, 1993.

[59] Chin VD, Peters DW, Spaid FW, McGhee RJ. Floweld
measurements about a multi-element airfoil at high
Reynolds numbers. AIAA Paper 93-3137, 1993.
[60] Klausmeyer SM, Lin JC. An experimental investigation
of skin friction on a multi-element airfoil. AIAA Paper
94-1870, 1994.
[61] Lynch FT. subsonic transport high-lift technologyFreview of experimental studies. AIAA Overview of High
Lift Aerodynamics, 1995.
[62] Lynch FT, Potter RC, Spaid FW. Requirements for
effective high lift CFD. 20th ICAS, 1996.
[63] Ying SX, Spaid FW, McGinley CB, Rumsey CL.
Investigation of conuent boundary layers in high lift
ows. AIAA J Aircr 1999;36(3):55062.
[64] Dow Sr., JP. Roll upset in severe icing. Federal Aviation
AdministationFAircraft Certication Service, 1995.
[65] Jacobs EN. Airfoil section characteristics as affected by
protuberances. NACA Report 446, 1932.
[66] Abbott IH, von Doenhoff, Albert E, Stivers LS.
Summary of airfoil data. NACA Report 824, 1945.
[67] Loftin LK, Smith HA. Aerodynamic characteristics of 15
NACA airfoil sections at seven Reynolds numbers from
0.7 million to 9.0 million. NACA TN 1945, 1949.
[68] Ladson CL. Effects of independent variation of Mach,
Reynolds numbers on the low-speed aerodynamic characteristics of the NACA 0012 airfoil section. NASA TM
4074, 1988.
[69] Loftin Jr., LK, Bursnall WJ. The effects of variations in
Reynolds number between 3.0 million and 25.0 million
upon the aerodynamic characteristics of a number of
NACA 6-series airfoil sections. NACA Report 964, 1950.
[70] Abbott Jr., FT, Turner Jr., HR. The effects of roughness
at high Reynolds numbers on the lift and drag
characteristics of three thick airfoils. NACA ACR No.
L4H21, 1944 (Wartime Report No. L-46).
[71] Mavriplis F. Icing, Contamination of aircraft surfacesFindustrys concerns. Proceedings of the First Bombardier International Workshop on Aircraft Icing/
Boundary-Layer Stability and Transition, Montreal,
Canada, 1993. p. 515.
[72] Bragg MB, Gregorek GM. Environmentally induced
surface roughness on laminar ow airfoils: implications
for ight safety. AIAA Paper 89-2049, 1989.
[73] Gulick BG. Effects of simulated ice formation on the
aerodynamic characteristics of an airfoil. NACA WR
L-292, 1938.
[74] Ingleman-Sundberg M, Trunov OK, Ivaniko A. Methods
for prediction of the inuence of ice on aircraft
ying characteristics. Report JR-1, a joint report from
the Swedish-Soviet Working Group on Flight Safety,
1977.
[75] Bragg MB, Khodadoust A, Kerho M. Aerodynamics of a
nite wing with simulated ice. Computational and
Physical Aspects of Aerodynamic Flows, Long Beach,
CA, 1992.
[76] Ferguson S, Mullins Jr., B, Smith D, Korkan K. Fullscale empennage wind tunnel test to evaluate effects of
simulated ice on aerodynamic characteristics. AIAA
Paper 95-0451, 1995.
[77] Trunov OK, Ingleman-Sundberg M. Wind tunnel
investigation of hazardous tail stall due to icing. Report

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

[78]

[79]

[80]

[81]

[82]

[83]

[84]

[85]

[86]

[87]

[88]
[89]

[90]

[91]

[92]

[93]

[94]

[95]

JR-2, a joint report from the Swedish-Soviet Working


Group on Flight Safety, 1979.
Trunov OK, Ingleman-Sundberg M. On the problem of
horizontal tail stall due to ice. Report JR-3, a joint report
from the Swedish-Soviet Working Group on Flight
Safety, 6th Meeting, 1985.
Papadakis M, Gile-Lain BE. Aerodynamic performance
of a tail section with simulated ice shapes and roughness.
AIAA Paper 2001-0539, 2001.
Papadakis M, Yeong HW, Chandrasekharan R, Hinson
M, Ratvasky TP, Giriunas J. Experimental investigation
of simulated ice accretions on a full-scale T-tail. AIAA
Paper 2001-0090, 2001.
Hiltner D, McKee M, LaNoe R, Gregorek G. DHC-6
twin otter tailplane airfoil section testing in the Ohio state
university 7  10 wind tunnel. NASA CR-2000-20992
(2 vols), 2000.
Valarezo WO. Maximum lift degradation dueto wing
upper surface contamination. Proceedings of the First
Bombardier International Workshop on Aircraft Icing/
Boundary-Layer Stability and Transition, Montreal,
Canada, 1993. p. 10412.
Boer JN, Van Hengst J. Aerodynamic degradation due to
distributed roughness on high lift conguration. AIAA
Paper 93-0028, 1993.
Myers TT, Klyde DH, Magdelano RE. The dynamic
icing detection system (DIDS). AIAA Paper 2000-0364,
2000.
Johnson CL. Wing loading, icing, and associated aspects
of modern transport design. J Aero Sci 1940;8(2):
4355.
Bowden DT. Effect of pneumatic de-icers, ice formations
on aerodynamic characteristics of an airfoil. NACA TN
3564, 1956.
Cooper WA, Sand WR, Politovich MR, Veal DL. Effects
of icing on performance of a research airplane. AIAA J
Aircr 1984;21(9):70815.
Politovich MR. Aircraft icing caused by large supercooled droplets. J Appl Meteorol 1989;28(9):85668.
Politovich MR. Response of a research aircraft to icing,
evaluation of severity indices. AIAA J Aircraft
1996;33(2):2917.
Ashenden R, Marwitz J. Turboprop aircraft performance
response to various environmental conditions. AIAA J
Aircr 1997;34(3):27887.
Welte D, Wohlrath W, Seubert R, DeBartolomeo W,
Toogood RD. Preparation of the ice certication of the
Dornier 328 regional airliner by numerical simulation and
by ground test. AGARD CP-496, Paper 14, 1991.
Ashenden R, Lindberg W, Marwitz J, Hoxie B. Airfoil
performance degradation by supercooled cloud, drizzle,
and rain drop icing. AIAA J Aircr 1996;33(6):10406.
Lee S, Bragg MB. Effects of simulated-spanwise ice
shapes on airfoils: experimental investigation. AIAA
Paper 99-0092, 1999.
Lee S, Bragg MB. Experimental investigation of simulated large-droplet ice shapes on airfoil aerodynamics.
AIAA J Aircr 1999;36(5):84450.
Lee S, Bragg MB. Private communication, December
2000 (same data subsequently published in AIAA Paper
2001-2421, 2001).

765

[96] Addy H, Chung J. A wind tunnel study of icing on a


natural laminar ow airfoil. AIAA Paper 2000-0095,
2000.
[97] Calay RK, Hold AE, Mayman P, Lunn I. Experimental
simulation of runback ice. AIAA J Aircr 1997;34(2):
20612.
[98] Anon, Report of the 12A Working Group on Determination of Critical Ice Shapes for the Certication of
Aircraft. DOT/FAA/AR-00/37, 2000.
[99] Morris DE. Designing to avoid dangerous behavior of an
aircraft due to the effects on control hinge moments of ice
on the leading edge of the xed surface. RAE TN-Aero
1878, 1947.
[100] Worrall, Cole. Wind tunnel measurements of the effect of
ice formations on the hinge moment characteristics of the
Viking elevator. ARC 10691, 1947.
[101] Ratvasky TP, van Zante JF, Riley JT. NASA/FAA
tailplane icing program overview. AIAA Paper 99-0370,
1999.
[102] Preston GM, Blackman CC. Effects of ice formations on
airplane performance in level cruising ight. NACA TN
1598, 1948.
[103] Leckman PR. Qualication of light aircraft for ight in
icing conditions. SAE Paper 710394, 1971.
[104] Teymourazov R, Kofman V. The effect of ice accretion
on the wing and stabilizer on aircraft performance,
unidentied USSR Paper.
[105] Ranaudo RJ. Mikkelsen KL, McKnight RC, Perkins PJ.
Performance degradation of a typical twin engine
commuter type aircraft in measured natural icing conditions. AIAA Paper 84-0179 (also NASA TM 83564),
1984.
[106] Mikkelsen KL, McKnight RC, Ranaudo RJ, Perkins PJ.
Icing ight research: aerodynamic effects of ice and ice
shape documentation with stereo photography. AIAA
Paper 85-0468 (also NASA TM 86906), 1985.
[107] Ranaudo RJ, Mikkelsen KL, McKnight RC, Ide RF,
Reehorst AL, Jordan WC. Schinstock WC, Putz SJ. The
measurement of aircraft performance and stability and
control after ight through natural icing conditions.
AIAA Paper 86-9758, 1986.
[108] Brown AP. Analysis of the aerodynamic effects of
freezing drizzle inight icing on a turboprop aircraft.
AIAA Paper 99-3151, 1999.
[109] Cook DE. Unusual natural icing encounters during
boeing 777 ight tests. AIAA Paper 97-0304, 1997.
[110] Obert E. Low-speed stability, control characteristics of
transport aircraft with particular reference to tailplane
design. AGARD CP-160, Paper 10, 1975.
[111] Ranaudo RJ, Batterson JG, Reehorst AL, Bond TH,
OMara TM. Determination of longitudinal aerodynamic
derivatives using ight data from an icing research
aircraft. AIAA Paper 89-0754 (also NASA TM 101427),
1989.
[112] Ranaudo RJ, Batterson JG, Reehorst AL, Bond TH,
OMara TM. Effects of horizontal tail ice on longitudinal aerodynamic derivatives. AIAA J Aircr
1991;28(3):1939.
[113] Ratvasky TP, Ranaudo RJ. Icing effects on aircraft
stability and control determined from ight data. AIAA
Paper 93-0398 (also NASA TM 105977), 1993.

766

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

[114] Van Zante JF, Ratvasky TP. Investigation of dynamic


ight maneuvers with an iced tailplane. AIAA Paper
99-0371 (also NASA/TM-1999-208849), 1999.
[115] Ratvasky TP, Van Zante JF. In-ight aerodynamic
measurements of an iced horizontal tailplane. AIAA
Paper 99-0638 (also NASA/TM-1999-208902), 1999.
[116] Ratvasky TP, Van Zante JF, Sim A. NASA/FAA
tailplane icing program: ight test report. NASA/TP2000-209908 (also DOT/FAA/AR-99/85), March 2000.
[117] Ranaudo RJ, Ratvasky TP, Van Zante JF. Flying
qualities evaluation of a commuter aircraft with an ice
contaminated tailplane. SAE Paper 2000-01-1676 (also
NASA/TM-2000-210356), 2000.
[118] Bragg MB, Hutchinson T, Merret J, Oltman R,
Pokhariyal D. Effect of ice accretion on aircraft ight
dynamics. AIAA Paper 2000-0360, 2000.
[119] Gray VH, von Glahn UH. Effect of ice and frost
formations on drag of NACA 651-212 airfoil for various
modes of thermal ice protection. NACA TN 2962, 1953.
[120] Gray VH. Correlations among ice measurements, impingement rates, icing conditions and drag of a 65A004
airfoil. NACA TN 4151, 1958.
[121] Gray VH, von Glahn UH. Aerodynamic effects caused by
icing of an unswept NACA 65A-004 airfoil. NACA TN
4155, 1958.
[122] Gray VH. Prediction of aerodynamic penalties caused by
ice formation on various airfoils. NASA TN D-2166,
1964.
[123] Bragg MB, Gregorek GM, Shaw RJ. Wind tunnel
investigation of airfoil performance degradation due to
icing. AIAA Paper 82-0582, 1982.
[124] Shaw RJ, Sotos RG, Solano FR. An experimental study
of airfoil icing characteristics. NASA TM 82790, 1982.
[125] Olsen W, Shaw RJ, Newton J. Ice shapes and the
resulting drag increase for a NACA 0012 airfoil. NASA
TM 83556, 1984.
[126] Potapczuk MG, Berkowitz BM. An experimental investigation of multielement airfoil ice accretion and resulting
performance degradation. AIAA Paper 89-0752 (also
NASA TM 101441), 1989.
[127] Potapczuk MG, Berkowitz BM. Experimental investigation of multielement airfoil ice accretion and resulting
performance degradation. AIAA J Aircr 1990;27(8):679
91.
[128] Berkowitz BM, Potapczuk MG, Namdar BS, Langhals
TJ. Experimental ice shape and performance characteristics for a multielement airfoil in the NASA Lewis icing
research tunnel. NASA TM 105380, 1991.
[129] Shin J, Bond TH. Results of an icing test on a NACA
0012 airfoil in the NACA Lewis icing research tunnel.
AIAA Paper 92-0647 (also NASA TM 105374), 1992.
[130] Shin J, Bond TH. Experimental and computational ice
shapes and resutling drag increase for a NACA 0012
airfoil. NASA TM 105743, 1992.
[131] Addy HE, Potapczuk MG, Sheldon DW. Modern airfoil
ice accretions. AIAA Paper 97-0174 (also NASA TM
107423), 1997.
[132] Addy HE. Ice accretions, icing effects for modern airfoils.
NASA/TP-2000-210031 (also DOT/FAA/AR-99/89),
2000.
[133] Addy HE. Private communication, 2000.

[134] Bragg MB, Gregorek GM. Aerodynamic characteristics of airfoils with ice accretions. AIAA Paper 820282, 1982.
[135] Bragg MB, Gregorek GM, Lee JD. Airfoil aerodynamics
in icing conditions. AIAA J Aircr 1986;23(1):7681.
[136] Bragg MB, Zaguli RJ, Gregorek GM. Wind tunnel
evaluation of airfoil performance using simulated ice
shapes. NASA CR 167960, 1982.
[137] Wickens RH, Nguyen VD. Wind tunnel investigation of
wing-propeller model performance degradation due to
distributed upper surface roughness and leading edge
shape modication. AGARD CP-496, Paper 11, 1991.
[138] Bragg MB, Coirier WJ. Detailed measurements of the
ow eld in the vicinity of an airfoil with glaze ice. AIAA
Paper 85-0409, 1985.
[139] Bragg MB, Coirier WJ. Aerodynamic measurements of
an airfoil with simulated glaze ice. AIAA Paper 86-0484,
1986.
[140] Bragg MB. Experimental aerodynamic characteristics of
an NACA 0012 airfoil with simulated glaze ice. AIAA J
Aircr 1988;25(9):84954.
[141] Bragg MB, Khodadoust A. Experimental measurements
in a large separation bubble due to a simulated glaze ice
accretion. AIAA Paper 88-0116, 1988.
[142] Bragg MB, Khodadoust A. Effect of simulated glaze ice
on a rectangular wing. AIAA Paper 89-0750, 1989.
[143] Khodadoust A, Bragg MB, Kerho M, Wells S, Soltani
MR. Finite wing aerodynamics with simulated glaze ice.
AIAA Paper 92-0414, 1992.
[144] Khodadoust A, Bragg MB. Measured aerodynamic
performance of a swept wing with a simulated ice
accretion. AIAA Paper 90-0490, 1990.
[145] Bragg MB, Khodadoust A, Soltani MR, Wells S, Kerho
M. Effect of simulated ice accretion on the aerodynamics
of a swept wing. AIAA Paper 91-0442, 1991.
[146] Papadakis M, Gile Lain BE, Youssef GM, Ratvasky
TP. Aerodynamic scaling experiments with simulated
glaze ice accretions. AIAA Paper 2001-0833, 2001.
[147] Jackson DG, Bragg MB. Aerodynamic performance of
an NLF airfoil with simulated ice. AIAA Paper 99-0373,
1999.
[148] Gile Lain BE, Papadakis M. Experimental investigation
of simulated ice accretions on a natural laminar ow
airfoil. AIAA Paper 2001-0088, 2001.
[149] Papadakis M, Alansatan S, Seltman M. Experimental
study of simulated ice shapes on a NACA 0011 Airfoil.
AIAA Paper 99-0096, 1999.
[150] Papadakis M, Alansatan S, Wong S-C. Aerodynamic
characteristics of a symmetric NACA section with
simulated ice shapes. AIAA Paper 2000-0098, 2000.
[151] Kim HS, Bragg MB. Effect of leading-edge ice accretion
geometry on airfoil performance. AIAA Paper 99-3150,
1999.
[152] Lee S, Kim HS, Bragg MB. Investigation of factors that
inuence iced-airfoil aerodynamics. AIAA Paper 20000099, 2000.
[153] Papadakis M, Alansatan S, Yeong HW. Aerodynamic
performance of a T-tail with simulated ice accretions.
AIAA Paper 2000-0363, 2000.
[154] Laschka B, Jesse RE. Determination of ice shapes
and their effect on the aerodynamic characteristics

F.T. Lynch, A. Khodadoust / Progress in Aerospace Sciences 37 (2001) 669767

[155]

[156]

[157]

[158]

[159]
[160]
[161]

[162]
[163]

[164]

for the unprotected tail of the A300. ICAS Paper 74-42,


1974.
Laschka B, Jesse RE. Ice accretion and its effects on
aerodynamics of unprotected aircraft components.
AGARD AR-127, Paper 4, 1978.
Ingelman-Sundberg M. Tail plane stallFwhat Isthat.
FFAP-A-925 (aeronuatical research institute of Sweden),
1991.
Reehorst A, Potapczuk M, Ratvasky T, Gile Lain B.
Wind tunnel measured effects on a twin-engine short-haul
transport caused by simulated ice accretions. AIAA
Paper 96-0871, 1986.
Khodadoust A, Dominik C, Shin J, Miller D. Effect of inight ice accretion on the performance of a multi-element
airfoil. Proceedings of AHS/SAE International Icing
Symposium, 1995.
Thomas JL. (NASA Langley). Private Communication,
April 2001.
Bragg MB. Rime ice accretion, its effect on airfoil
performance. NASA CR-165599, 1982.
Miller TL, Korkan KD, Shaw RJ. Statistical study of an
airfoil glaze ice drag coefcient correlation. SAE Paper
830753, 1983.
Flemming RJ, Lednicer DA. High speed ice accretion on
rotorcraft airfoils. NASA CR-3910, 1985.
Cook DE. Relationships of ice shapes and drag to icing
condition dimensionless parameters. AIAA Paper 20000486, 2000.
Zierten TA, Hill EG. Effects of wing simulated ground
frost on aircraft performance. VKI Lecture Series, 1987.

767

[165] van Hengst J, Boer JN. The effect of hoar-frosted wings


on the Fokker 50 take-off characteristics. AGARD-CP496, Paper 13, 1991.
[166] Hill EG. Private communication, August 2001.
[167] Bragg MB, Heinrich DC, Valarezo WO, McGhee RJ.
Effect of underwing frost on a transport aircraft airfoil
at ight Reynolds number. AIAA J Aircr 1994;31(6):
13729.
[168] Jones R, Williams DH. The effect of surface roughness on
the characteristics of the aerofoils. NACA 0012 and RAF
34, ARC, R&M 1708, 1936.
[169] Morelli JP. Flight test wing surface roughness effects for
development of an airfoil surface contamination detection system. MDC Report 96K9377, 1996.
[170] Hooker RW. The aerodynamic characteristics of airfoils
as affected by surface roughness. NACA TN 457,
1933.
[171] Hoerner SF, Borst HV. Fluid dynamic lift. 1985, pp. 416
to 422.
[172] Ljungstrom BLG. Wind tunnel investigation of simulated
hoar frost on a 2-dimensional wing section with and
without high lift devices. FFA-AU-902, 1972.
[173] van Hengst J. Aircraft de-icing, future technical developmentsFan aircraft manufacturers point of view.
Wingtips (Fokker Aircraft) No. 26, 1993.
[174] van Hengst J, Gent R, Hammond D, Seubert R, Wagner
B. Ice accretion and its effects on aircraft. AGARD,
AR-344, 1997. p. 3.13.36.

You might also like