47th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Confere AIAA 2006-2188

1 - 4 May 2006, Newport, Rhode Island

Higher Order Statistical Signal Processing Studies on the Impact of Icing on Aircraft Stability and
Control Including Aeroelastic Effects

Claus W. Endruhn*, Ronald O. Stearman **, and

David B. Goldstein***
Department of Aerospace Engineering and Engineering Mechanics,
The University of Texas at Austin, TX, 78712

The impact of icing on aircraft stability and control, including aeroelastic effects, has been studied in wind
tunnel tests employing higher order statistical signal data processing along with theoretical model
formulations. Several wind tunnel tests and pilot flight experiences have shown that the presence of ice on the
tails and wings of General Aviation and small Transport category aircraft may trigger aeroelastic non-linear
(limit cycle) oscillations on the lifting surfaces that can cause a loss of stability and control of the aircraft.
Two different complete aircraft models of the same generic aircraft configuration were tested in the
University of Texas 5 ft. by 7 ft. subsonic wind tunnel. A rigid 1/10 scale powered sting balance aerodynamic
loads model was studied to measure aerodynamic stability derivatives and to conduct flow visualization
studies. A 1/10 scale powered radio controlled cable-mounted model was also studied to demonstrate possible
stability and control upsets due to icing. This cable model support is a scaled down version similar to the one
developed by the NASA Langley Research Center for the Transonic Dynamics Tunnel. The experiments on
the cable-mounted model captured time series signals from three accelerometers that measured instantaneous
changes about the aircraft yaw, pitch, and roll axes. In addition, time series signals from two pressure
transducers were measured to quantify influences of the unsteady flow effects due to ice induced separated
flows on the aircraft. Auto- and cross-bicoherence squared spectra along with linear coherency studies were
estimated from the captured time series data in order to quantify possible quadratic non-linear interactions
and energy transfers promoting the stability and control upsets. Another general class of model that was
experimentally tested in this 5 ft. by 7 ft. consisted of two different full scale aerodynamically sheltered horn
balance substructures found typically on the tail tips of general aviation and small transport category
turboprop aircraft. Time series signals were captured from several accelerometers and unsteady pressure
transducers located in different positions on the aerodynamic horn balances. Higher order statistical signal
data acquisition and processing was then employed to help validate, through the observed non-linear
interactions, the theoretically developed form of a Van der Pole type oscillator model predicting the observed
non-linear limit cycle motion of the horn.

*Graduate Student, Department of Aerospace and Engineering Mechanics, 210 East 24th Street, W.R. Woolrich
Laboratories University Station, C0600, Austin, Texas 78712-0235.

**Professor Aerospace Engineering and Engineering Mechanics, Bettie Margaret Smith Professor in Engineering,
Department of Aerospace and Engineering Mechanics, 210 East 24th Street, W.R. Woolrich Laboratories University
Station, C0600, Austin, Texas 78712-0235.

*** Professor Aerospace Engineering and Engineering Mechanics, Chevron Centennial Fellow in Engineering
#1,Department of Aerospace and Engineering Mechanics, 210 East 24th Street, W.R. Woolrich Laboratories
University Station, C0600, Austin, Texas 78712-0235.

1
American Institute of Aeronautics and Astronautics

Copyright © 2006 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
 Nomenclature

X[l] = calculated value of the digital Fourier transform of x[n]

Y[l] = calculated value of the digital Fourier transform of y[n]
x[n] = nth sampled value of a digital time series
y[n] = nth sampled value of a digital time series
N = total number of samples for a digital time series
F( f ) = Fourier Transform
Pˆ [A]
YX = ensemble averaged estimated discrete cross-power spectrum
M XX ( f ) = power-spectrum of a given signal
M = number of data realizations, i.e. data subsets
γˆ YX2 [A]
= discrete linear coherence spectrum
γ ( f j , fk )
2
xxx = auto-bicoherence
γ xxxx
2
( f q , f r , f s ) = auto-tricoherence
Sˆ [A , A ]
XXX 1 2 = ensemble averaged estimated discrete auto-bispectrum
bˆ 2 [A 1 , A 2 ] = discrete auto-bicoherence spectrum
SˆYXX [A 1 , A 2 ] = ensemble averaged estimated discrete cross-bispectrum
bˆ 2 [A , A ]
c 1 2 = discrete cross-bicoherence spectrum
R (t ) = Weibull Reliability Function
H (s) = Heaviside step Function

Fφ = instantaneous pivot torque

FW = force at the elastic center

I. Introduction

I ce formation on aircraft wings during flight has caused, according to the NTSB, more than 135 documented
aircraft accidents, and killed more than 171 people.1 When an aircraft flies into environments with meteorological
conditions that can cause ice formation, the aircraft’s ability to maintain flight will diminish rapidly with time.
Several studies throughout the years have shown that ice build up on the leading edge of a wing decreases lift and
causes wing stall at much lower angles of attack, but very few have investigated the aeroelastic effects of ice
accretions.2 Wind tunnel experiments conducted at the University of Texas at Austin have shown that ice buildup on
the leading edge of a wing significant can cause sub-critical-flutter events, also known as Limit Cycle Oscillations.3
Aircraft stability and control upsets have also been caused by icing and two such events are studied in the present
paper. See Bragg et al for other studies in this area.4

Ice buildups can form with several different shapes and characteristics during flight in icing
environments. The trajectories of the water droplets in the flow field around the wing are critical in the ice buildup
characteristics. The droplets that hit the leading edge of the airfoil directly tend to form rime ice accretions. Rime
ice, is an opaque granular deposit of ice formed by the rapid freezing of super-cooled water drops as they impinge

2
American Institute of Aeronautics and Astronautics
upon the exposed wing. Rime ice is composed essentially of discrete ice granules and has a density as low as 0.115–
0.170 ounces per cubic inch.5 Sometimes, the water droplets cover a small distance over the airfoil before touching
the surface, and form glaze ice accretions. Glaze ice is denser, harder, and more transparent than rime ice. Its density
may be as high as 0.46 or 0.52 ounces per cubic inch.5 Factors that favor glaze ice formation are large drop size,
rapid accretion, and slight super-cooling. Slightly higher temperatures, fast accretion rates, or very large droplets
may cause a delay in freezing causing a portion of the water to move downstream and creating two protuberances
facing the air stream just behind the centerline, sometimes denoted as ice horns, typical of the shape shown in Figure
1. In most cases it is common to find both rime and glaze ice on the leading edge of a wing during flight in an icing
environment, but the relative size of the horns or protrusions of glaze ice have a much larger effect on the flow,
making the rime ice effects less dominant. Figure 2 shows some examples of the rime and glace ice accretions.

The influence of in-flight icing on the aeroelastic response of an aircraft is not yet well understood except
possibly for the mass unbalance effect on flight controls due to ice accumulation. In the past, several studies by the
senior author were focused on the aeroelastic response of an aircraft wing due to the influence of in-flight warhead
ballistic damage. Some of the first work on this general subject was conducted in 1950 by Biot and Arnold.6 Their
findings showed that a reduction of flutter speed or divergence margins by 25% required nearly a 90% loss of
stiffness at a critical section within the wing, which implied that structural failure will occur earlier, due to a loss of
strength, before aeroelastic instabilities could happen due to a loss of stiffness. The impact of these and other
findings at that time discouraged serious investigations of this subject for the next 30 years.

Then in 1982 a study was published, guided by some of the early work of Ashley and Petre,7 demonstrating
that when an aerodynamic damage model was also added to the former warhead imposed structural damage model
of Biot and Arnold some significant aeroelastic limit cycle oscillations could be induced. Subsonic wind tunnel
studies were also carried out on a damage simulated aeroelastic stick and pod cantilevered wing model illustrating
that these limit cycle oscillations could be demonstrated experimentally. Thus, aeroelastic studies conducted on the
statistical fighter wing model proposed in the earlier work by Biot but also including an aerodynamic damage model
demonstrated the possibility of several limit cycle oscillations in a region formerly thought to be safe from flutter
and divergence. The results of these early findings are presented in Figure 3 and Figure 4. Later it became evident to
us that damage could also imply icing induced aerodynamic damage on lifting surfaces also promoting these limit
cycle oscillations at speeds sub-critical to classic flutter and divergence.8

Experiments conducted in the late 60’s at the NASA Glenn Research Center Icing Wind Tunnel on a twin
engine general aviation aircraft combined stabilizer and elevator horn with a sheltered horn balance have shown that
the glaze ice horns have a significant effect on the flow of air around an airfoil.9 Figure 5 documents a limit cycle
oscillation of the horn balance induced by icing nearly forty years ago. The proper icing simulation of these horns on
the wing and tail is required to provide an accurate reproduction of the overall effect of the ice on the airflow, over a
lifting surface.

Several ice simulation shapes have been tested on airfoils to determine the effects of the simulated ice
radius on the flight characteristics.10 Examples of possible shapes on potential airfoil leading edges are shown in
Figure 6. In our experiments, we employed standard ¼ round and T-shape cross section stringers that we cut and
attached to the model at the leading edge surface. Figure 7 illustrates the two variations of stringers employed for the
simulated ice shapes on our wind tunnel experiments.

II. Higher Order Spectra Analysis Tools

This section presents the fundamental signal processing theory essential for the time series analysis applied
in this study. This section begins with an introduction to the basics of direct Fourier Transform methods. The
objective of this section is mainly to introduce to the reader basic concepts. This section therefore, does not provide
detailed proofs, and further reading of the mentioned references is recommended for a better understanding of the
different available Higher Order Spectra Analysis tools.

3
American Institute of Aeronautics and Astronautics
 Most of the direct methods in Higher Order Spectra Tools employ the discrete Fourier Transform (DFT) as a
basis in their formulations. For a continuous function of one variable f(t), the Fourier Transform F(f) is defined as,11

∞
F ( f ) = ∫ f (t )e − j 2πft dt (1)
−∞

and the inverse transform as,

∞
f (t ) = ∫ F ( f )e j 2πft dt (2)
−∞

where j is the square root of -1 and e denotes the natural exponent

e jφ = cos φ + j sin φ (3)

In some occasions, F(f) and f(t) are also denoted as X(f) and x(t) respectively. Variable t is the time while f
is the frequency, and x(t) and X(f) form what is referred to as a Fourier Transform pair.
For the discrete formulation of the Fourier Transform consider a complex series x(k) with N samples of the form11
x 0, x1, x 2, x3, x 4, ...
where x is a complex number, and assume that the series outside the range 0, N-1 is extended N-periodic, that is, xk
= xk+N for all k. The forward discrete transform is defined as,11

N −1
1
X ( n) =
N
∑ x ( k )e
k =0
− jk 2πn / N
, for n = 0 … N -1. (4)

The inverse transform is defined as,11

N −1
x(n) = ∑ x(k )e jk 2πn / N , for n = 0 … N -1. (5)
k =0

The present study employs the classical auto and cross-power spectra to obtain the frequency content of
time series signals recorded during the wind tunnel experiments and to quantify the linear relationship between two
signals. To detect the non-linear interactions between different components of a signal x(t), the power spectrum fails
because phase relationships are not retained. Higher order spectra are therefore required to satisfactorily determine
the interactions.

Higher-order statistics are essential, since classical second-order statistics are not sufficient for
investigating nonlinearities in experimental data. For example, higher-order statistics have the ability to preserve
phase information in the form of a phase difference, the ability to detect phase coupling between different frequency
components, and insensitivity to Gaussian noise. 12

The linear coherence spectrum, also known as a normalized cross-power spectrum, was also employed in
this study. Non-linear spectral content is studied employing higher-order statistical tools, including the auto-
bispectrum and cross-bispectrum. Bispectra tools were fundamental to this study, since they preserve the phase
information, and permit the evaluation of phase-coupling and phase-coherence between frequency components
identified in the digital time series.13

Consider the experimentally gathered time series signals representing the input x(t), and the output, y(t).
Then, let X(f) and Y(f) represent the discrete Fourier transforms of the two random signals x(t) and y(t). The power
spectrum (auto-spectrum) measures the distribution of the mean square power of a signal with respect to its
frequency. The power spectrum can be estimated as

4
American Institute of Aeronautics and Astronautics
 [
M XX ( f ) = E X ( f ) X * ( f ) = E X ( f ) ] [ 2
] (6)

where * denotes a complex conjugate and E[ ] denotes an expected value, usually achieved in practice by averaging
over a large number of data segments.14

The linear coherency squared function is an estimate of the fraction of output power in y(t) and its linear
correlation to the input power x(t). At each frequency, the maximum value is present at those frequencies correlated
on a one to one frequency basis, between the input and output. In contrast the minimum value of linear coherency
squared is present at frequencies not one to one between the input and output at a given frequency. In other words,
coherence function is the fraction of the output power only linearly related to the input signal:14

Output power only due to the input

γ 2( f ) =
Total out power

The coherence function is mathematically described as:

Re 2 [S XY ( f )] [Re( F ( f )) ⋅ Re(G ( f )) + Im(F ( f )) ⋅ Im(G ( f ))] 2

Re(γ ) = =
[ ][ ]
2 (7)
S XX ( f ) ⋅ S YY ( f ) Re 2 ( F ( f )) + Im 2 ( F ( f )) ⋅ Re 2 (G ( f )) + Im 2 (G ( f ))
This function indicates the degree of linear dependence, on scale from zero to unity, between the output and
the input signals.

The auto-bicoherence and auto-tricoherence are given by,14

2
M xxx ( f j , f k )
γ xxx
2
( f j , fk ) = (8)
M xx ( f j ) M xx ( f k ) M xx ( f j + f k )

2
M xxxx ( f q , f r , f s )
γ 2
( fq , fr , fs ) = (9)
M xx ( f q ) M xx ( f r ) M xx ( f s ) M xx ( f q + f r + f s )
xxxx

Next consider two time series x(t) and y(t). The classical linear coherence spectrum,15

2
M xy ( f j , f k )
γ yx2 ( f ) = (10)
M xx ( f ) M yy ( f )

gives the degree of linear correlation, on a spectral basis. This may be extended to introduce the cross-
bicoherence γ yxx ( f j , f k ) , and the cross-tricoherence γ yxxx
2 2
( fq , fr , fs ) .
The cross-bicoherence γ yxx ( f j , f k ) is defined as,14
2

2
M yxx ( f j , f k )
γ 2
( f j , fk ) = . (11)
M yx ( f j ) M yx ( f k ) M yx ( f j + f k )
yxx

The cross-tricoherence γ yxxx

2
( fq , fr , fs ) is defined as,

5
American Institute of Aeronautics and Astronautics
 2
M yxxx ( f q , f r , f s )
γ yxxx
2
( fq , fr , fs ) = . (12)
M yx ( f q ) M yx ( f r ) M yx ( f s ) M yx ( f q + f r + f s )

The cross-bicoherence is a measure of the quadratic relationship between frequency components XT(j)XT(fk)
and YT(fj)XT(fk). The cross-tricoherence is a measure of the cubic relationship between frequency components
XT(fq)XT(fY)XT(fS) and YT(fq)YT(fY)YT(fS) and. The cross-bicoherence and cross-tricoherence are ideal tools for
non-linear system identification, and they can be used to detect quadratic and cubic frequency interactions
respectively. Because the bicoherence is just a normalized bispectrum, phase coherence between waves at two
frequencies f1 and f2 will generate a non-zero value in the bicoherence at (f1, f2).14

So far, the higher-order moment and spectra functions have been discussed in the context of continuous
functions, continuous spectra and continuous Fourier transforms. It is important to note that these functions will be
estimated using a digital computer. In this case, they would be estimated with discrete higher-order spectra and
discrete Fourier transforms (DFTs). The most general computational approach to deal with the discrete Fourier
transform is the Fast Fourier Transform (FFT), a well know and implemented algorithm that will not be discussed in
this report and if interested, the reader can find discussion of this in most of the digital signal processing literature.11

The higher-order spectra from the experimentally observed time series data, in our case the times series
obtained from the accelerometers and pressure transducers mounted on our wind tunnel models was estimated using
a non-parametric approach, the FFT approach.14 The steps underlying the FFT approach to calculate the bispectrum
are presented in Figure 9.

The D.C. component or signal mean is eliminated from the time series in order to obtain a better dynamic
range in the spectral computations. The D.C. component is eliminated by first computing the signal mean of the time
series, and then subtracting this from the time series data to obtain a signal with higher dynamic range for spectral
calculations. All of the higher order spectra calculations discussed in this study assume a signal with zero mean.

III. Pilot Reports of Cessna 208B Aircraft Stability and Control Upsets in Icing Flight Conditions

In an attempt to gain further insight into the Cessna 208 series aircraft icing accidents a review was made of
two types of aircraft upset experiences found in the NTSB records. The first upset discussed occurred in light rime
icing conditions at about 9,000 ft. MSL under straight and level low angle of attack cruise conditions, and is
described as a category I event. At least three of these upsets have been documented, where two have occurred to the
same pilot at different times and places. The category II upset reviewed occurred in a Cessna 208 type aircraft and
involves violent wing rocking modes of uncontrolled flight.

Several such category II upset events have been observed by both pilots and ground witnesses. It seems to
occur when the aircraft is at an angle of attack of approximately 6 degrees or more when either climbing or
descending. In one case it is known that the airspeed had dropped to about 102 knots in significant icing conditions
where a stall most likely occurred followed by repeated bank angle excursions of ± 40 degrees.

In another case this wing rock was observed by 10 ground witnesses as experienced by a professional pilot
with nearly 10,000 hours of flight experience. In addition, a second approximately 10,000 hour professional pilot
was having difficulty controlling the aircraft below his experienced 120 knot iced wing stall speed and his observed
iced 140 knot tail purposing speed which would cause the stalling of the horizontal tail and resulting the aircraft
uncontrolled pitch over. The ± 40 degrees observed wing rock during this flight experience was suggested by the
pilot to possibly be due to non-uniform shedding of ice on the wing. The present paper suggests an alternative wing
rock mechanism.

The category I and category II pilot reports are attached as part of this paper’s appendix.

6
American Institute of Aeronautics and Astronautics
IV. Weibull Reliability Analysis of the Cessna 208 Series Airplane for in-flight icing conditions

A National Transportation Safety Board (NTSB) “Safety Recommendation” to the Federal Aviation
Administration (FAA) on December 15th, 2004, expressed concerns about a possible systemic problem with the
Cessna 208 series airplanes’ design or with the airplanes’ flight characteristics into icing conditions. In fact, the
board in late 2003 initiated an in-depth assessment of several icing-related events it had on record.16 The FAA the
issued an “airworthiness concern sheet” on December 22nd, 2004 indicating that since 1996 the Cessna 208 series
aircraft averaged about two icing-related accidents per year in the United States only. Before 1996 it averaged
almost one per year. In a more recent time frame, a more current NTSB safety recommendation of January 17th,
2006, suggests to the FAA that this 208 series aircraft be restricted to allowed flight into only light instead of
moderate icing conditions.17 This followed after two fatal accidents occurred in known icing conditions in this series
aircraft in both Russia, November 2005, and in Canada, October 2005 with significantly experienced pilots on
board.

Historically, aircraft thought to have a safety problem when flying under certain environmental conditions
can be checked out by investigating the NTSB accident statistics for the specific environmental situation in question.
A life cycle reliability check for the flight performance over a certain time interval in the questioned environments
conditions will provide a statistically based reliability number that can clarify the safety issue. Such an approach was
used by the Boeing Company and Sikorsky during the early years of the UH-60 Blackhawk helicopter to study this
safety issue for the Army. In a similar manner several concerns were addressed for the Apache Helicopter.19 The
details of a such a life cycle Weibull reliability analysis for the Cessna 208B series aircraft is presented elsewhere by
Endruhn.20 This study,20 was based upon NTSB accident website statistics for this aircraft employing hours flown by
the aircraft up to the time of the icing accident.

The statistical results from this study are related to the failure of a mechanical system design based on the
analogy of people dying within the human life cycle. For example, some die in an infant mortality mode, while
others reach a useful life stage, but may be killed at random in an automobile or aircraft accident, while the majority
of us may reach a wear out stage dying finally of an older age. The Weibull reliability function,21

β
⎡ ⎛t −γ ⎞ ⎤
R (t ) = exp ⎢− ⎜⎜ ⎟⎟ ⎥ (13)
⎢⎣ ⎝ η ⎠ ⎥⎦

was system identified employing the NTSB accident data to clarify this issue.

The parameters β , η and γ are used to make statements regarding the reliability of the aircraft flying in
icing conditions. Figure 10 is a graphical representation of the human life cycle and represents the three possible life
failure modes (early failure, random failures in the useful life, and wear out failures).

For β <1, the distribution represents early failure and a decreasing failure rate with age. For β =1, the
Weibull distribution reduces exactly to the exponential distribution and can thus represent a constant failure rate.
For β >1, the curve represents an increasing failure rate with age. An analogy to these human failure modes is also
extended to the three possible life failure modes in the design life of a mechanical system. Failure in the burn-in
region can generally be attributed to a design not suitable to operate in the environmental conditions it was
supposedly designed for (e.g. a bad design) or one with manufacturing quality control problems.

Although a larger number of dataset points of flights into icing would be desirable to obtain better statistical
regularity, there are definite indications of some design problem with the aircraft in question when flying in an icing
environment. This can be seen in Figure 11 contrasting the obtained results of the Cessna 208 aircraft series with a
historically and statistically much safer C172 trainer. It can be observed, that in the case of the general aviation
trainer, the shape parameter beta converges to around the value of 1.5 for the larger number of data samples, while
the results obtained with the available data for the more sophisticated Cessna 208B clearly show a major drop in the
shape parameter β , toward the hazardous situation of early failures for a given sample size N. It is interesting to

7
American Institute of Aeronautics and Astronautics
note this significant drop to the early failure mode in spite of the fact that a more professional pilot is flying the
more advanced category aircraft than the student trainer. Furthermore the Cessna 208B aircrafts are FAA certified
for flight into known moderate icing conditions so this icing environmental effect should not be a factor in theory.

V. Horizontal Stabilizer and Elevator Horn Balance Geometry Investigated

The basic configuration of the horizontal stabilizer and elevator of the Cessna 208B series aircraft is depicted
in Figure 12. This study was undertaken to help clarify the pilot observed category I upset identified in the previous
section.

The elevator aerodynamic balance horn is identified as that portion of the elevator surface forward of the
elevator hinge line at the outboard tip of the elevator. The main role of the elevator horn aerodynamic balance is to
reduce the pilot’s required control force required to hold the elevator at a fixed angle of attack.22

When the elevator rotates about its hinge line, the elevator horn produces an aerodynamic lifting force and an
aerodynamic pitching moment.22 Since the center of lift for the horn is ahead of the elevator hinge line and the
center of lift for the remaining elevator remains behind the elevator hinge line, the total aerodynamic pitching
moment at the hinge point is subsequently reduced. Comparing an elevator with balance horn with an elevator
without a balance horn, it is easy to expect that elevator with the balance horn achieves higher net aerodynamic lift
with a lower net aerodynamic hinge moment at a given angle of attack. This net gain of aerodynamic lift with a
lower hinge moment is translated in a traditional (non-fly-by-wire) aircraft as a reduced control force exerted by the
pilot during a given maneuver.23 The aerodynamic lift (moment) due to the elevator balance horn, the remainder of
the elevator, and the total aerodynamic lift (moment) for the combined elevator and balance horn are denoted by
LE(ME), LB(MB), and LT(MT), respectively. A second important function of the elevator balance horn is its role in
B B

mass balancing the elevator surface. A balance weight is commonly placed in the balance horn ahead of the hinge
line. The purpose of the balance weight is to eliminate elevator control flutter. 24 The mass balance serves to
decouple the elevator rotational motion from that of the stabilizer. The selection of an appropriate weight and mass
distribution helps to eliminate control surface flutter.

In the case where the center of gravity of the elevator is ahead of the elevator hinge line, the elevator is
overbalanced.23 This mechanical overbalance causes the elevator to deflect against the stabilizer motion. If the
stabilizer deflects in a downward plunge the elevator will nose up. In the case that the stabilizer deflects pitch up, the
elevator nose will rotate nose down. This inertial coupling creates an aerodynamic force and moment that acts
opposite to the induced stabilizer motion and tends to stabilize dynamically the system. This is illustrated in Figure
13.

The correct design of the elevator horn aerodynamic horn balance is critical to the safety and integrity of any
general aviation aircraft. However, several recent pilot reports of the Caravan category aircraft under study indicated
a buzzing noise of the elevator horn balance during icing conditions flight that prompted a concern on the general
design of the aircraft aerodynamic horn balance. The pilots initially believed that vibration was perhaps coming
from the aircraft engine, but later on, reports indicate that the pilots actually observed the horn oscillating during
flight. The pilots were not always able to feel the vibration through the control system on their control sticks.

In a typical occurrence of control flutter, classical aeroelastic analysis would have recognized this problem,
however, our results below revealed that the oscillations are highly non-linear and do not correspond to a classical
example of the control flutter phenomenon. This is due to the disturbed wake created by the presence of ice on the
stabilizer tip in which the elevator balance horn lies. It can also be promoted by the shear layer caused by a stabilizer
sheltered horn balance. In addition, calculations made on the Cessna Grand Caravan elevator torsional stiffness with
an applied aerodynamic moment show a significant critical low value, that lies below the British Airworthiness
requirements.23 Since the system reveals at least non-linear aerodynamic characteristics, classical linear
aeroelasticity is not sufficient to predict this phenomenon. A non-linear shear layer model developed by Tate and
Stearman, based on physical insight and available experimental evidence, proved to be adequate to model this
phenomenon.23 It is interesting to note that a single flow model can be developed employing the Gaussian Field
shear layer flow field reviewed by Chang. 31 The following paragraphs describe the proposed model to explain the
limit cycle instabilities observed on the elevator balance horn.

8
American Institute of Aeronautics and Astronautics
 The first assumption is the use of a planar or flat plate model, which allows for the use of transient slender
wing theory in the aerodynamic modeling.23 The non-linear shear layer wake model describes how the aerodynamic
flow field decouples from the inboard flow field over the remainder of the elevator due to presence of a vortex trap
on the inboard side of the deflected horn.23 Employing this model, the lift and moment forces are calculated for two
flow regimes: a free stream condition for that of the elevator horn when deflected out of the wake region, and a
shear flow upstream for the condition of the elevator horn within the flow field.

The first modeling observation is existence of tip vortices generation by the elevator horn balance. These tip
vortices serve to isolate the flow field over the remainder of the tail surface. This isolation effect is shown in Figure
14. A several hundred degree of freedom finite element model was built of this elevator and horn system which was
clamped at the bell crank root chord. This model demonstrated the localization of the structural deformations to the
horn area as noted in the paper “A New Look at Galloping”.3

The next assumption is that the balance horn is reduced to a low aspect ratio wing which is known to have an
impulsive chord-wise pressure loading near the leading edge for the case of a straight non pointed leading edge
wing.25 This is analogous to the Kutta condition for high aspect ratio wings,23 and in this analysis there is no lift
downstream from the position of maximum cross section.

VI. Aeroelastic Non-linear Modeling of the Aerodynamic Horn Balance

For a non-linear flow model-shear layer, a two-dimensional typical section model subjected to a uniform time
dependent drag force across the span model is suggested.26 This drag may be harmonic or random in character and
may be generated by the formation of ice horns on the leading edge of the wing. The wind tunnel experiments with
ice on the leading edge of the wing showed similar non-limit cycle oscillations to the ones observed by Kim and
Stearman on damaged wings in wind tunnel studies. 25

A. Governing Equations of Motion for a Two-Dimensional Airfoil

A model for a two-dimensional airfoil wing, suspended by a torsional and a bending spring was developed
based on Tate’s previous model. 23

Consider a rigid, two-dimensional airfoil wing, suspended by a torsional spring and a bending spring as
shown in Figure 15.

This typical section is constrained by two-degrees of freedom by a mass-less pivoted bar that slides in a
bushing passing through the wing C.G. The bending of freedom w, is the distance along this bar from the pivot to
the C.G. The torsion degree of freedom is represented by a small angle, from which the bar and wing are rotated
clockwise to the zero angle of attack position ( φ = 0 ).

The shear layer is modeled after a description suggested by Chang,26 in which the downstream wake deficit
velocity profile is based on the Gaussian error function. Thus, the wake velocity field can be described as,

(
u ( z , t ) = u O 1 − εe −ηz
2
) (14)

where uO and ε are the free stream velocity and the velocity defect in the shear layer respectively.

The aerodynamic lift is given by the slender body theory as an integral of the running chordwise distributed lift,23

c/2 ∂L
L=∫ dx (15)
− c / 2 ∂x

9
American Institute of Aeronautics and Astronautics
 ∂L
where is the chord-wise running lift force, ρ is the density of the air, SO is the virtual fluid cross sectional
∂x
area.

The running lift for a small aspect ratio rectangular wing is given by

∂L D 2W
= − ρS O [H ( x + c / 2) − H ( x − c / 2)]
∂x Dt 2
− ρu ( z , t ) S O [δ ( x + c / 2 ) − δ ( x − c / 2)]
DW
(16)
DT

where H(s) is the Heaviside step function, δ (x) is the Dirac delta function, D(D) is the substantial derivative, and
the virtual area of the wing cross-section is defined as S(x)=SOH(x).

Recalling the wake sheltering hypothesis to points behind the leading edge, the resulting running lift is,23

∂L DW
= − ρU ( z , t ) S Oδ ( x + c / 2)
∂x Dt
⎡ ∂W ∂W ⎤
= − ρU ( z , t ) S O ⎢ + U ( z, t ) δ ( x + c / 2)
∂t ⎥⎦
(17)
⎣ ∂t

From Figure 15, W = w + elcφ and equation (17) becomes

∂L ⎡ ∂ (w + el cφ ) ∂ (w + el cφ ) ⎤
= − ρU ( z , t ) S O ⎢ + U ( z, t ) ⎥δ ( x + c / 2) (18)
∂x ⎣ ∂t ∂t ⎦
The dynamical equations of motion are derived from the section model found in Figure 15:

An instantaneous pivot torque,

Fφ = K φ φ (19)

and a force at the elastic center with magnitude

FW = KW ( w + e1cφ ) (20)

are applied on this typical section. By summing the moment about the pivot and the force along the linear spring,
and assuming a light viscous damping present on each degree of freedom of the system, the governing equations of
motion can be written as follows:

aO cq + C D cqφ = K W ( w + el cφ ) (21)
aO c (e + el )qφ + C D cqw = K φ φ
2
(22)

1
where q= ρU 2 ( z, t ) (23)
2
The drag coefficient is taken as a periodic time dependent force of the form

10
American Institute of Aeronautics and Astronautics
C D = C DO cos 2πf p t (24)

which in the limit f D → 0 becomes a static force coefficient CDO .

Solving for w in equation (21)

a 0 cq + C D cqφ
w= − el cφ (25)
KW
and substituting this result into equation (22) yields to,

a 0 c 2 q 2φ
a 0 c 2 eq + + a 0 el c 2 q − c 2 el C D qφ = K φ φ (26)
KW
e −ηz from the wake shear layer model,
2
By taking a Taylor series expansion on the term
−ηz 2 η 2z4 η3z6
e = 1 − ηz +
2
− + ... (27)
2 6
where z = (w + el cφ )
And inserting it into equation (14) and neglecting high order terms, the result is

⎛ εn 2 (w + el cφ )4 εn 3 (w + el cφ )6 ⎞
u ( z , t ) = U O ⎜1 − ε − εη (w + el cφ ) +
⎜ − ⎟
2
(28)
2 6 ⎟
⎝ ⎠
Expanding terms from equation (28) and inserting the result into equation (23) we obtain,

q=
1
2
(
ρU O [1 − ε − εη w 2 + 2cel φw + c 2 el2φ 2 )
+ εη 2 ( w 4 + 4cel φw 3 + 6c 2 el2φ 2 w 2 + 4c 3 el3φ 3 w + c 4 el4φ 4 )] (29)

From this result, one can clearly see that from inserting equation (29) into equation (26), there are high order
terms for w and φ . This can be seen in the appendix, in equation (27), where q was substituted into equation (26)
and all of the terms were expanded.

In the limiting case of very small bending deformations, equation (26) reduces to the Van der Pole oscillator
equation developed by Tate, implying relaxation-oscillations will occur under certain conditions. The form of this
equation is expressed as,

β "+δ (1 − εe −η β ) β '+ ⎡k0 − 2δ 1 − εe −η β

2

⎢⎣
( 2
) ⎤⎥⎦ β = 0
2
(30)

Where η is a non-dimensional constant of the geometry, and β is a non-dimensional parameter β (τ ) = φ (t ) .

A more familiar form of the equation results upon expanding the exponential function in a power series

e −η β = 1 − η β 2 + ...
2
(31)

Providing a simplified approximating form.

[
β "−δ (a − bβ 2 ) β '+ k0 − 2δ (a − bβ 2 ) β = 0
2
] (32)

11
American Institute of Aeronautics and Astronautics
 εη
where a = (ε − 1) and b = .
2!
The damping term of equation (32) possesses a classic form similar to that of Van der Pol’s equation.22 The
stiffness term possesses a cubic hardening term and a quintic softening term. The Van der Pole equation is a classic
example of an equation that describes limit cycle phenomenon.

VII. Experimental Setup and Calibration

A. Wind Tunnel
All the experiments conducted for this study took place in the 5’x 7’ Pickle Research Center subsonic wind
tunnel at the University of Texas at Austin. The tunnel is an atmospheric-intake, continuous flow type subsonic wind
tunnel powered by four 200 hp fans.27 The tunnel has adjustable blades and vanes for a variety of speeds. When the
four tunnel fans are on and at maximum power, the maximum airspeed is approximately 200 mph. The tunnel is
illustrated in Figure 16.

B. Wind Tunnel Models

This section describes the wind tunnel models employed for this study. The models consisted in a full scale
horizontal tail section of the Cessna 208B Grand Caravan as well as a 1/10 scale cable mounted model of the same
aircraft. This section presents a brief overview on the construction of the models as well as their characteristics.

1. Cessna 208B Grand Caravan Aerodynamic Horn Balance Model

The Cessna 208B Grand Caravan aircraft aerodynamic horn balance wind tunnel model is a full-scale
horizontal tail model segment. The exact measurements of the model were obtained by directly measuring a Cessna
208B Grand Caravan located in San Marcos, TX. Once the measurements were complete, a schematic of the aircraft
was made in AutoCAD and the construction of the model was performed by several aerospace engineering senior
design groups that worked on the model during the past year.

The materials for the model construction were craft wood, Styrofoam, metal spars and carbon fiber. The craft
wood was used as ribs and the Styrofoam was the filler material. The structural spars were made out of 1010 steel.
The exterior was covered with composite layer made of [902] carbon fiber cloth.

The torsional stiffness and support of the model’s elevator and aerodynamic horn balance is adjusted by
varying the thickness of the spring plate strips mounted in place to the cross flexure pivot blocks. These cross-
flexure pivots are then located on the shear web of the channel mounting bracket and held the aerodynamic horn
balance spar in place. The cross flexure pivots and channel mounting support are shown in Figure 17. There are
three spring strip plates that connect the elevator spar to the mounting brackets of the cross-flexure pivots. The
spring strip plates employed in this cross flexure pivot were made of stainless steel.

A torsional stiffness measurement was performed to obtain the model’s torsional stiffness and the obtained
stiffness is 100.3 ft-lbs/rad with a standard deviation of 2.25 ft-lbs/rad this is approximately an order of magnitude
less than the actual aircraft stiffness, which was measured to be approximately 2113 ft-lbs/rad with a standard
deviation of 128 ft-lbs/rad. It should also be noted that our model is not only of reduced stiffness, but has also
reduced density. The actual Cessna horn balance is made of metal while our model is mostly made of wooden cross
sections and Styrofoam filler. The characteristics of our model would imply that the observed oscillations will be
occurring at lower velocities that during actual flight conditions.

Two pressure transducers were attached on one side of the horn balance at the ¼ and ¾ chords of the horn
balance and two accelerometers were placed at the same chord lengths on top of the horn balance. In order to
minimize flow disturbance, the wires were run across the horn balance and along the elevator trailing edge before
being fed down through a hole at the base of the wind tunnel and out to the data acquisition system. This setup can
be seen in Figure 18.

12
American Institute of Aeronautics and Astronautics
2. Cessna 208B Grand Caravan Cable Mounted Wind Tunnel Model

One of the Cessna 208B Grand Caravan models is a 1/10 scale model and it was mounted in the wind tunnel
through a dynamic cable support system. This six degree of freedom dynamic cable support system was developed
by the NASA Langley Research Center in Virginia, and this concept has been applied with reasonable success to
obtain data from rigid and aeroelastic models ranging from a space shuttle orbiter to an F/A-18 E/F fighter.28 It was
anticipated that such a limited free flying model would demonstrate possible initial stability and control upsets
brought on by icing.

Our wind tunnel cable mounted model support system allows six degrees of freedom motion of the aircraft,
providing a reasonable accurate representation of the free flight behavior of the Cessna 208B series aircraft. The
cable mounting system consists in a pair of cables that pass trough the front and rear of the cable mounted model.
These cables are made out of stranded steel cable with a thin plastic coating and employed to hold the aircraft in a
free flight mode at the center of the tunnel, while preventing excessive large scale motions that would damage the
model. The front cable system allows for pitch and the rear cable system allows for roll and yaw motions. Each
cable loops passes through pulleys located within the model as well as the wind tunnel wall. The tension on the rear
cable is varied by adjustable weights and measured by a strain gage style load cell on the outside of the tunnel. A
viscous dashpot oil type damper is also employed on the weight platform holder.

The actual model weight, weight distribution (center of gravity) and inertial moments were dynamically
scaled the guidelines set forth by Hall.29 Table I shows the comparison between the geometry of the real Cessna
208B aircraft and the cable mounted wind tunnel model.

Figure 19 shows the configuration of the cable mounted support system in the wind tunnel. A set of three
accelerometers pointing in the x, y and z directions were located approximately 4 inches beneath the center of
gravity of the model.

Two pressure transducers as shown in Figure 20 were mounted on the sides of the aerodynamic horn balances
of the model. Five channels of sixty seconds time series data were taken in from all the sensors in our experiments at
a sampling frequency of 1 KHz.

Figure 21 shows the installation of the cable mounted model in the University of Texas Pickle subsonic wind
tunnel.

3. Cable Support System Fundamental Frequencies Calculations

A wind-off vibration study was conducted for the cable support and mounted Cessna 208B model to
determine the pitch, roll and yaw fundamental frequencies for that combined support and model system. The
oscillations were induced manually and a time series of the tri-axial accelerometers output were taken employing
our data acquisition system. The purpose of this experiment was to discard possible high order interactions of the
cable support frequencies with other structural and fluid frequencies inherent in this study. The primary frequencies
of the cable support system were approximately 3.4 Hz for the pitch motion, 4.2 Hz for the roll motion and 2.2 yaw
motions of the cables. The auto-powerspectra estimated from of these three accelerometers time series are shown in
Figure 22.

VIII. Wind Tunnel Results and Conclusion

A. Aerodynamic Horn Balance Wind Tunnel Tests and Results

Wind tunnel testing was conducted to substantiate the potential for localized flow disturbances due to ice
formations on the leading edge of the horizontal stabilizer and aerodynamic horn balance of the Cessna 208B Grand
Caravan to alter stability and to produce control upsets of the aircraft. This area of the aircraft is of concern since no
ice protection exists here. This section shows how non-classical aeroelastic responses of the aerodynamic horn
balance due to icing have been experimentally achieved and identified using bi-spectral analysis. It also

13
American Institute of Aeronautics and Astronautics
demonstrates the benefits of using this signal analysis tool along with experimental studies to evolve mathematical
models for these non-linear events.

Several wind tunnel tests were conducted in this study at the Pickle Research Center subsonic wind tunnel.
The objectives of these experiments were to observe the effects of different ice shapes and ice formations on the
aerodynamic horn balance model aeroelastic response of the Cessna 208B Grand Caravan.

1. Aerodynamic Horn Balance First Torsional and First Bending Mode Shapes

Wind tunnel wind-off experiments were conducted in this study employing structural impact tests performed
on the aerodynamic horn balance. The purpose of these cyclic impact tests was to determine the natural frequencies
of the structure and find the first bending and first torsion mode shapes of the elevator and horn balance structure.
These frequencies could be then employed in the data reduction analysis to discard or interpret some of the non-
linear interactions shown in the higher order spectral analysis. The tests consisted of impacting the structure about
50 to 60 times during a period of 60 seconds and sampling data at 1 KHz on both accelerometers. Then the time
series were analyzed using both linear and bispectral signal analysis.

The auto-powerspectrum from one accelerometer in the test is shown in Figure 23. From this auto-
powerspectrum estimate it was determined that the frequency peaks at approximately 1.5 Hz and 4 Hz are the
bending and torsion natural frequencies respectively of the structure.

2. Aerodynamic Horn Balance Tests without Ice Formations

The first set of experiments was conducted on the aerodynamic horn balance model without any type of
simulated ice. The purpose of these wind tunnel tests was to obtain a database on the response of the elevator horn
aerodynamic pressures and horn accelerations if a limit cycle oscillation was observed without ice. No such horn
limit cycle oscillations were observed up to an airspeed of 120 mph for this case of no ice. These tests were useful
even though no limit cycle oscillations were observed since it helped to order the importance of nonlinear physical
phenomena due to the free stream shear layer with the nonlinear parametric mechanism associated with a time
dependant drag term. In other words, the tests helped to characterize or order the importance of the nonlinear terms
in the equations of motion due to different physical mechanisms such as the free stream shear layer and/or the time
dependant drag term. These nonlinear ordering questions were resolved by examining the elevator horn pressure and
acceleration auto-power-spectra, auto-bispectra and auto-bicoherence data from the recorded time series to see if any
of the linear, along with higher order quadratic, cubic, quadric and quintic terms of φ from the proposed equation
(26) were present in the resulting polyspectra estimates. In summary, these tests were necessary in order to contrast
the stabilizer and elevator horn modeling under various air load conditions, such as shear layer and drag, for the
proper mathematical formulation of the problem. A contrast, for example, between the nonlinear free stream shear
layer effect and the nonlinear parametric effects due to the time dependant drag force can be resolved by the cross-
bicoherency squared function being present or absent in this higher order polyspectral function estimate. On the
other hand, the importance of the Van der Pole oscillator characteristic resulting from only the shear layer
nonlinearity could be identified physically in observed relaxation-oscillations related to control elevator pulsing and
eventually full scale limit cycle oscillations.

The tests performed on the ice free stabilizer and elevator horn were conducted at several speed settings of the
tunnel. The tests started at 25% setting of the tunnel with two fans on (approximately 30 mph at the wind tunnel test
section) and ranged up to 50% setting with the four fans on (approximately 120 mph at the wind tunnel test section).
Measurements were made with the two accelerometers and pressure transducers and the data were saved into time
series files in the laptop computer for later analysis. Also, the tests can be observed in the recorded videos from the
three cameras placed in different windows of the wind tunnel testing section

No significant limit cycle oscillations were observed on these no-icing tests, other than some signal noise and
possible wind tunnel structural vibrations noise at high frequencies (120 Hz bandwidth) shown if the auto-
powerspectrum plot of Figure 24.

14
American Institute of Aeronautics and Astronautics
 The auto-bicoherence of the same acceleration time series did not show any harmonic nonlinearities as one
could surmise from the linear spectra analysis of Figure 25. The auto-bicoherence plot of the acceleration in the case
where the aerodynamic horn balance was tested without any sort of ice formations is shown in Figure 26. This
implies no significant structural non-linearities are present and consequently no limit cycle oscillations are
occurring. Cross-bicoherency plots were also not considered since no limit cycle oscillations occurred.

3. Aerodynamic Horn Balance Tests with Ice Formations

The purpose of these tests was to characterize possible parametric and nonlinear aeroelastic response of the
aerodynamic horn balance under the influence of leading edge icing conditions. Figure 6 illustrates the various
physical icing conditions found on NASA icing wind tunnel tests on a similar stabilizer horizontal tail system.9 The
five minutes case with only horn ice was considered. A hundred and eight different icing test runs were performed,
resulting in a total of 432 individual acceleration and pressure time series data collections. Also, over twelve hours
of digital video camera were recorded using the three different camera positions discussed earlier. Different ice
configurations tests were performed on this study and are discussed later on in this section. The data acquired from
these wind tunnel experiments were analyzed using both classical linear spectral analysis and higher-order bispectral
analysis.

The following ice configurations shown in Figure 26 were tested with the Cessna 208B aerodynamic horn
balance geometry. A reduced gap was employed between the elevator horn and stabilizer shield as the result of a
personal communication that indicated that the elevator may have frozen to the stabilizer in the Cessna 208 series
aircraft in some icing conditions.

The results of the horn balance tunnel tests are presented for the icing geometry B shown in Figure 26. This
represents the simulation of the actual Cessna 208B horn to stabilizer full gap with ice restricted to horn simulations
slightly aft of the horn leading edge. Several other configurations such as geometry A were also studied but are only
reported by Endruhn.20 Since personal pilot communications have suggested

The accelerometer auto-powerspectrum for the limit cycle oscillation found for geometry B is shown in
Figure 27. This occurred for a velocity of approximately 102 mph. Without icing this horn was found to be stable
from limit cycle oscillations to the limit of our study, which was about 20 mph higher so the horn icing is seen to
have a significant destabilizing effect. A torsional response frequency of 9 Hz is seen both the accelerometers and
pressure transducers’ time series as shown also in Figure 28. Furthermore, nonlinearities consisting of second, third,
fourth and fifth harmonics of the fundamental pressure at 9 Hz are present indicating that nonlinear effects are in the
fluid model, but not the structural model. The auto-bicoherency function of the time series is shown in Figure 29 and
Figure 30 illustrating these aerodynamic nonlinear interactions stressing the importance of the shear layer model. A
squared cross-bicoherence estimate, shown in Figure 31, shows no significant features indicating no parametric
excitations are occurring and the nonlinearity due to the time dependent drag term can be neglected in this limit
cycle modeling.

B. Cessna 208B Cable Mounted Model Wind Tunnel Experiments

The results of the wind tunnel tests on the Cessna 208B cable mounted model are presented and interpreted in
this section. These Cessna 208B cable model experiments were performed in order to assist in the validation of the
hypothesis that ice formation on the unprotected elevator horn promotes instabilities that tend to drive the Cessna
208B aircraft into an unstable wing rock flight mode. In his research on airfoils as stated by Bragg30 “The most
significant unsteady flow-field effect on the iced-airfoil performance is a low-frequency flow phenomenon in the
order of 10 Hz that resulted in Strouhal numbers of 0.0048 – 0.0101. The low frequency oscillation produces large-
scale pressure fluctuations near separation at high angles of attack and elevated lift and moment fluctuations as low
as a pitch angle α of 5 degrees. The iced-airfoil flowfield exhibited a separation of varying thicknesses and
fluctuating reattachment, characteristics similar to those associated with the low frequency shear layer flapping and
bubble growth and decay of other separated and reattached flows.”30

15
American Institute of Aeronautics and Astronautics
 When icing simulations were placed on the Cessna 208B cable model horn, significant flow unsteadiness
were developed due to the ice shape, as suggested by Bragg. Figure shows flow visualizations illustrating the
unsteadiness of the separation bubble shear layer with the horn at an angle of attack.

During the cable model wind tunnel tests, when simulated icing horns were placed on the unprotected leading
edge of the aerodynamic horn balance a violent wing rock instability occurred when the aircraft was placed in an
approximate 6 degrees to 10 degrees angle of climb. The simulated icing forms employed were of the type shown in
Figure 20 while the aircraft attitude was approximately that shown in Figure 32. When the model was flown in a
lower angle of cruise of 4 to 5 degrees or less no wing rock occurred. It was also shown that if approximately the six
foot long drip plates over the rear passenger doors were extended back along the fuselage to about two feet forward
of the horizontal stabilizer leading edge this also stabilized the wing rock problem. The authors are of the opinion
that these drip plates are also acting as aerodynamic chines holding the trailing edge wing vortices at the wing root
fuselage junction in position keeping them below the horizontal tailplane. During a 6 to 10 degrees climb and above,
the cross flow Reynolds number is such that vortices would occur if these chines were not holding them in place.

Figure 34 illustrates very low Reynolds Number water tunnel test on a 1/32 scale model at about a 10 degrees
climb where the vortices have shifted above the horizontal tail plane into the vicinity of the vertical tail. No
aerodynamic chines where attached to such a small model. Figure 33 illustrates the idea of this vortex induced wing
rock when it was encountered by the E-6B aircraft operating near stall.22 The alternating vortex shedding running up
the vertical tail could easily force the aircraft into an unstable like Dutch Roll maneuver.

The higher order cross-bicoherency squared signal processing tool is employed in this study to verify the
possibility that this cross flow shedding event could be promoted by its phase lock-in to the significant lift and
moment forces created by the separated flow field over the aerodynamic horn balances due to leading edge horn
type icing. According to Bragg at about 10o of elevator angle of attack the Strouhal number was such that an
approximate 10 Hz unsteady low frequency lift and moment force develops. Figure 35 shows during an upset the
auto-powerspectrum of the pilot’s left aerodynamic horn pressure transducer. The highest spectral peak in our test
occurs at 10 Hz, which is the frequency anticipated for the Bragg Strouhal number of 0.0101, velocity of 60 ft/sec
and a characteristic ice and airfoil thickness of approximately 0.72 in. It is of interest to note that in the case of no
ice this spectral peak does not occur as shown in Figure 36 for the same climb angle when no upset occurred.

If one looks at the possibility of the fuselage cross flow shedding frequency, which is approximately at 1 Hz
locking in with the unsteady horn flow lift and moment frequencies, then a squared cross-bicoherency estimate
should show a significant energy exchange (i.e. phase lock in) between the 1 Hz and 10 Hz coordinate point on the
plot. This can be readily seen in Figure 37, Figure 38 and Figure 39 where a squared cross-bicoherency value
indicates a 55 to 65% frequency and phase correlation in each of the three degrees of freedom x, y and z of motion.
Once lock-in occurs one would expect model response to occur at approximately this 1 Hz shedding frequency. An
independent estimate of the model frequency responses during wing rock was measured from the video time counter
on the video movies. The results shown in Table II are taken from an eleven cycle video upset response sequence
which confirms the approximate 1 Hz.

Based upon this 1/10 scale cable model wind tunnel test the cross bicoherency squared function confirms the
hypothesis of cross-flow body vortex shedding at about 1 Hz locking in with the significantly unsteady lift, moment
and drag terms induced by the separated flow-field due to the ice formation.

16
American Institute of Aeronautics and Astronautics
 IX. Figures

Figure 1. Glaze ice horn accretions on leading edge of an airfoil5

Figure 2. Glaze, Mixed and Rime Ice accretions on the leading edge of an airfoil5

17
American Institute of Aeronautics and Astronautics
 Figure 3. Aeroelastic in-flight warhead ballistic damage simulated by employing both a
structural and aerodynamic model resulting in a classical Mathew-Hill equations of motion
allowing for Limit Cycle Oscillation involving combination and parametric resonant
instabilities8

Figure 4. Flutter and divergence boundaries as influenced by both aerodynamic and

structural damage modeling8

18
American Institute of Aeronautics and Astronautics
Figure 5. Close up of Extremely Long and Flexible Elevator Horn Balance tested in the
NASA Glenn Icing Tunnel9

Figure 6. Rime and Glaze (Clear) Ice shapes10

19
American Institute of Aeronautics and Astronautics
 Figure 7. Simulated Ice cross sectional shapes

Figure 8. Input/Output for a physical system with quadratic non-linearity

Figure 9. FFT approach for calculating the bispectrum

20
American Institute of Aeronautics and Astronautics
Figure 10. Operating Life showing the Early Failures region, the Useful Life period and the
Wear-Out Period21

Figure 11. Shape Parameter vs. Size of Dataset for General Aviation Trainer and
Transport Category Aircraft.

21
American Institute of Aeronautics and Astronautics
 Figure 12. Horizontal Tail of the Transport Category Aircraft under study

Figure 13.Aerodynamic Forces and Moments developed by inertial coupling on an

aerodynamic horn balance

Figure 14. Aerodynamic Isolation of the Aerodynamic Horn Balance due to tip vortices

22
American Institute of Aeronautics and Astronautics
Figure 15. Typical section of a two-dimensional aeroelastic wing model in a Gaussian Error
Function approximation wake velocity field

Figure 16. 5’ x 7’ Pickle Research Center subsonic wind tunnel exterior view

23
American Institute of Aeronautics and Astronautics
Figure 17. Mounting Bracket Close-up view and Setup Under the Subsonic Wind Tunnel

Figure 18. Sensor locations on the Cessna 208B Grand Caravan Horizontal Tail Model
with Aerodynamic Horn Balance

24
American Institute of Aeronautics and Astronautics
Table I. Cessna 208B Real Aircraft and Model Measurements comparison

Figure 19.Cable Mounted Model Configuration at the Pickle Low Speed Wind Tunnel

25
American Institute of Aeronautics and Astronautics
Figure 20. Pressure Transducer Installed on the Aerodynamic Horn Balance of the Cable
Mounted Model (Negative Image to Enhance Contrast)

Figure 21. Cessna 208B Cable Mounted Model installed in the Pickle Subsonic Wind
Tunnel

26
American Institute of Aeronautics and Astronautics
Figure 22.Auto-powerspectrum of the cable support fundamental frequencies taken from
X, Y, and Z directions accelerometers mounted on the transport category aircraft model

Figure 23.Auto-powerspectrum of the aerodynamic horn balance impact test employing

acceleration time series

27
American Institute of Aeronautics and Astronautics
Figure 24. Power-Spectrum of Acceleration Time Series in the Case without Ice formations

Figure 25.Auto-bicoherence of Acceleration Time Series in Case without Ice formations

28
American Institute of Aeronautics and Astronautics
Figure 26. Geometry A (Left) and Geometry B (Right) of Ice shapes on Aerodynamic Horn
Balance of the Cessna 208B

Figure 27. Auto-power Spectrum of Accelerometers (Torsion) of Geometry B – 102 mph

Figure 28. Auto-power Spectrum of Pressure Transducers (Torsion) of Geometry B – 102

mph
29
American Institute of Aeronautics and Astronautics
Figure 29. Auto-bicoherence of Pressure Transducers (Contour Plot) of Geometry B – 102
mph

Figure 30. Auto-bicoherence of Pressure Transducers (Isometric Plot) of Geometry B – 102

mph

30
American Institute of Aeronautics and Astronautics
Figure 31. Squared Cross-Bicoherence of Pressure and Acceleration Auto-bicoherence of
Pressure (Isometric Plot) of Geometry B – 102 mph

Figure 29. Flow Visualization of the Cessna 208B Aerodynamic Horn Balance with
Simulated Ice Shape

31
American Institute of Aeronautics and Astronautics
Figure 32.Cessna 208B Cable Model at an angle of attack of approximately 10 degrees

Figure 33. E-6B aircraft operating near stall.22

32
American Institute of Aeronautics and Astronautics
 Figure 34. Stagnation pressure profile behind Caravan without ice at AOA of 10 deg

Figure 35. Left pressure transducer time series auto-powerspectrum, ice case at AoA 10
degrees

33
American Institute of Aeronautics and Astronautics
 Figure 36. Left pressure transducer time series auto-powerspectrum, no ice case at
approximately AoA 10 degrees

Figure 37.Squared cross-bicoherence contour plot between right pressure transducer signal
and x-axis accelerometer, iced case at AoA 10 degrees

34
American Institute of Aeronautics and Astronautics
Figure 38. Squared cross-bicoherence contour plot between left pressure transducer signal
and y-axis accelerometer, iced case at AoA 10 degrees

Figure 39. Squared cross-bicoherence contour plot between right pressure transducer
signal and z-axis accelerometer, iced case at AoA 10 degrees

35
American Institute of Aeronautics and Astronautics
Table II. Wing Rock Oscillation Frequencies of the Cessna 208B Cable Model Upsets in
Wind Tunnel Experiments Estimated from Recorded Videos

36
American Institute of Aeronautics and Astronautics
 X. Appendix

c 2η 4 a0U O ε 2 ρ 2φw 3 2c 3η 4 a 0 elU O ε 2 ρ 2φ 2 w 7 7c 4η 4 a0 el U O ε 2 ρ 2φ 3 w 6

2 2 2

+ +
4KW KW KW
c 2η 3 a 0U O ε 2 ρ 2φw 6 14c 5η 4 a 0 el U O ε 2 ρ 2φ 4 w5 3c 3η 3 a0 elU O ε 2 ρ 2φ 2 w 5
2 3 2 2

− + −
2K W KW KW
35c 6η 4 a 0 el U O ε 2 ρ 2φ 5 w 4 15c 4η 3 a0 el U O ε 2 ρ 2φ 3 w 4 c 2η 2 a0 eU O ε 2 ρ 2φ 5 w 4
4 2 2 2 2

+ − +
2KW 2KW 2
c 2η 2 a0 elU O ερw 4 c 2η 2 a0U O ε 2 ρ 2φw 4 c 2η 2 a0U O ερ 2φw 4 c 2η 2 C D elU O ερφw 4
2 2

+ − + −
2KW 4KW 2KW 2
14c 7η 4 a0 el U O ε 2 ρ 2φ 6 w 6 10c 5η 3 a0 el U O ε 2 ρ 2φ 4 w3
5 2 3 2

+ − + 2c 3 eη 2 a0 elU O ερφw3
KW KW
c 3η 2 a 0 elU O ε 2 ρ 2φ 2 w3 2c 3η 2 a0 elU O ερ 2φ 2 w3
2 2

+ 2c η a 0 el U O ερφw − + − 2c 3η 2 C D el U O ερφ 2 w3
3 2 2 3 2

KW KW
7c 8η 4 a0 el U O ε 2 ρ 2φ 7 w 2 15c 6η 3 a0 el U O ε 2 ρ 2φ 5 w 2
6 2 4 2

+ − + 3c 4 eη 2 a0 el U O ερφ 2 w3
2

KW 2KW
3c 4η 2 a0 el U O ε 2 ρ 2φ 3 w 2 3c 4η 2 a0 el U O ερ 2φ 3 w 2
2 2 2 2

+ 3c 4η 2 a0 el U O ερφ 2 w 2 − +
3

2KW KW
c 2 eηa0U O ερw 2 c 2ηa0 elU O ερw 2 c 2ηa0U O ε 2 ρ 2φw 2
2

− 3c η C D el U O ερφ w − − +
4 2 3 3 2

2 2 2KW
c 2ηa0U O ερ 2φw 2 c 2ηC D elU O ερφw 2 c 6η 2 a0 el U O ε 2 ρ 2φ 5 2c 9η 4 a0 el U O ε 2 ρ 2φ 8 w
2 2 4 2 7 2

− + + +
2KW 2 2KW KW
3c 7η 3 a0 el U O ε 2 ρ 2φ 6 w c 4ηa0 el U O ερφ 2 c 4ηa0 el U O ερφ 2
5 2 2 3

− + 3c 4η 2 C D el U O ερφ 3 w 2 − −
3

KW 2 2
c 4ηa0 el U O ερ 2φ 3 c 4ηa0 el U O ερ 2φ 3 w c 4ηC D el U O ερφ 3 c 2 aO eU O ερ c 2 aO eU O ρ
2 2 2 2 3

+ − + − +
2KW 2KW 2 2 2
c 2 aO elU O ερ c 2 aO elU O ρ c 2 aOηU O ε 2 ρ 2φ c 2 aOU O ερ 2φ c 2 aOU O ρ 2φ c 2 C D elU O ερφ
2 2 2 2

− + + − + +
2 2 4KW 2KW 4KW KW
3c 2 C D el U O ε 2 ρφ c 6ηa0 el U O ε 2 ρ 2φ 5 c 6η 2 a0 el U O ερ 2φ 4 c 6 eη 2 a0 el U O ερφ 4
2 4 2 5 4

− + + + = Kφφ
2 4KW 2 2

(30)

37
American Institute of Aeronautics and Astronautics
Category I Pilot Report

38
American Institute of Aeronautics and Astronautics
Category II Pilot Report

39
American Institute of Aeronautics and Astronautics
 40
American Institute of Aeronautics and Astronautics
Acknowledgments

The authors would like to acknowledge the previous work done by former graduate students G.S Chen, T.R. Kim,
J.H. Chang, Ralph Tate, Eloy Gonzalo, Marcus Kruger, and Joshua Foxworth. In addition the help of senior design
students Aaron Snyder, Stephen Kwok, Janaina Standerwick, Randy and Ronald Lum, Clint Robertson, Marco
Romero, Bryan Tu, Travis Mercker, Robert Kondret, Romeo Pugao, Denver Tsui, Andres Hernandez, who helped
with some of the experiments at the University of Texas at Austin, and the signal processing expertise of Electrical
Engineering professor E. J. Powers was also helpful. Finally, appreciation to Arthur Alan Wolk is also expressed
for the financial support of this research.

References
1
. Potapczuk, Mark, “Simulating Ice”, 8/18/2005,
http://machinedesign.com/asp/viewSelectedArticle.asp?strArticleId=58948&strSite=MDSite&catId=2
2
. Addy, H.E., “Ice Accretions and Icing Effects for Modern Airfoils,” NASA/TP-2000-210031, April 2000.
3
.Kruger, Marcus, Endruhn, Claus, and Stearman, Ronald O. “A New Look at Galloping”,
th
47 AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Material Conference, 1-6 May 2005, Austin, TX.
4
. Bragg, M., Whalen, E. and Lee, S, “Characterizing the Effect of Ice on Aircraft Performance and Control from Flight
Data”, 40th AIAA Aerospace Sciences Meeting and Exhibit, 14-17 January 2002, Reno, NV.
5
. Addy, Harold E, Jr., Andy P. Broeren, and Michael B. Bragg, “Effect of Intercycle Ice Accretion on Airfoil Performance.”
AIAA 2002.
6
. Biot, M., Arnold, L. “Low speed flutter and its physical interpretation”, Journal of the Aeronautical Sciences, 15, pp. 232-
236, April 1948.
7
. Petre, A. and Ashley, H., “Drag Effects on Wing Flutter”, Journal of Aircraft, Vol 13, No. 10, October 1976, pp 755-763.
8
. Chen, G. S and Stearman, R. O., “A Damaged Induced Aeroelastic Failure Mode Involving Combination and Parametric
Resonant Instabilities of Lifting Surfaces”, AIAA, ASME, ASCE, AHS 23rd Structures, Structural Dynamics, and Materials
Conference, AIAA-82-o636-CP, May 10-12, 1982.
9
Wilson, Calvin, F. “Icing Tests of a Horizontal Tail Shielded Horn Balance on a Light Twin Engine Aircraft – PA 31”, Piper
Aircraft Corporation, Lock Haven, Pennsylvania, Report No. 1502, December 29, 1967. (Icing Tests Conducted in NASA Lewis
Tunnel at that time).
10
. Addy, H.E., “Ice Accretions and Icing Effects for Modern Airfoils,” NASA/TP-2000-210031, April 2000.
11
. Ingle, Vinay K, and Proakis, John G., Digital Signal Processing using Matlab, Brooks/Cole, Pacific Grove, CA 2000.
12
. Powers, Edwards, and Im, Sungbin, Higher-Order Statistical Signal Processing and its Applications, Longman-Wiley,
Melbourne, 1995.
13
. Kim, Y.C. and Powers, E.J., “Digital Bi-spectral Analysis and its applications to Nonlinear Wave Interactions”, IEEE
Trans. Plasma Sci. P5-7(2), pp 120-131, 1979.
14
. Powers, E. J., Hong, J. Y. and Ritz, C. P., “Reference Guide for Integrated Signal Analysis”, Austin, Texas: Integrated
Signal Processing, 1995.
15
. J.S. Bendat and A.G. Piersol. Random Data: Analysis and Measurement Procedures, Second Edition, Wiley, New York,
1986.
16
. National Transportation Safety Board, “Safety Recommendation” A-04-64 through -67, Washington D.C., December 15,
2004.

41
American Institute of Aeronautics and Astronautics
 17
. National Transportation Safety Board, “Safety Recommendation” A-06-01 through -03, Washington D.C., January 17,
2006.
18
. Federal Aviation Administration, “Airworthiness Concern Sheet”, Atlanta, GA, December 22, 2004.
19
. Von Achen, William, “The Apache Helicopter: an EMI case History”, Compliance Engineering, Fall 1991, pp. 11-17 and
111-112. (See section the Black Hawk)
20
. Endruhn, Claus, “Wind Tunnel Investigation on Potential Stability and Control Upsets of the Cessna 208B Grand Caravan
Aircraft when Flying in Icing Conditions”, Master’s Thesis Report, The University of Texas at Austin, Department of Aerospace
Engineering, May, 2006.
21
. Carter, A.D.S, Mechanical Reliability, Second Edition, John Wiley and Sons, New York, July 1997.
22
. Bertin, J.J. and Smith, M., Aerodynamics for Engineers, Prentice-Hall, Englewood Cliffs, New Jersey, 1979.
23
. Tate, Ralph, and R. O. Stearman: “The Aeroelastic Instability of an Elevator Balance Horn in a Shear Layer Wake Flow”,
SAE Paper 890516, General Aircraft Meeting and Exposition, Wichita Kansas April 16-19, 1985.
24
. Fung, Y.C., An introduction to the theory of Aeroelasticity, Dover, New York, 1969.
25
. Stearman, Ronald O., “A simplified Chordwise Divergence Analysis of an Aircraft Elevator Balance Horn”
(Aeroelasticity Class Notes – University of Texas), May 2004.
26
. Chang, J.-H and R. O. Stearman, Choi, D and Powers, E. J. “Identification of Aeroelastic Phenomenon employing
bispectral analysis techniques”, College of Engineering, The University of Texas at Austin.
27.
Westkaemper, John, “The University of Texas at Austin Aerospace Engineering Subsonic Wind Tunnel Facility”,
Facilities Pamphlet, 1992.
` 28
. Bennett, Robert M. et al, “Wind-Tunnel Technique for Determining Stability Derivatives from Cable-Mounted Models”,
Journal of Aircraft 1978 0021-8669 vol.15 no.5 (304-310).
29
. Hall, Stan, “Dynamic Modeling,” Sport Aviation, 30 July 1987, pp. 30-35
30
. Gurbacki, Holly, and Bragg, Michael, “Unsteady Flowfield about an Iced Airfoil”, 42nd Aerospace Sciences Meeting and
Exhibit, Reno, NV, January 5-8, 2004.

42
American Institute of Aeronautics and Astronautics