ANGLE OF ATTACK ESTIMATION USING AN INERTIAL REFERENCE PLATFORM

Joseph E. Zeis, Jr., Capt, U W *

6510 Test Wing
Edwards AFB, California
Heather H. Lambert**
NASA Ames Research Center
Dryden Flight Research Facility
Edwards, California
Robert A. Calico+
Air Force Institute of Technology
Wright-Patterson AFB, Ohio
Daniel Gleason, Maj, U W + +
Air Force Institute of Technology
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

Wright-Patterson AFEi, Ohio

Abstract

This paper presents the mathematical Tail contribution to lift

development and flight test results of an angle of
Lift of wing-body combination
attack estimation system based on inertial
navigation system inputs. The estimator uses these Mach number
inputs to determine the coefficient of lift Moment about y axis
required at any instant inflight. Angle of attack (pitching moment)
is then modeled through a regression analysis Zero-lift pitching moment
based on coefficient of lift, altitude and Mach.
Overall correlation of the estimator as tested was Load factor
Roll rate
generally within 0.5 degrees through 17 degrees
angle of attack on an F-15A aircraft. A robustness Pitch acceleration
analysis indicates that the system can be used Yaw rate
adequately in maneuvering flight. Wing area
True airspeed
Nomenclature Airmass velocity
Normal acceleration in Body axis velocity
body axis
Inertial velocity
c Mean aerodynamic chord of wing
Wind velocity
c, Wing-body coefficient of lift Distance from center of gravity
WB
Zero lift pitching moment to wing-body aerodynamic
nr' center
coefficient
Distance from center of gravity
g acceleration of gravity to horizontal tail aerodynamic
h A1 t itude center
Ix Moment of inertia about x axis Angle of attack
IY Moment of inertia about y axis Initial angle of attack guess
Iz Moment of inertia about z axis
Ixz Cross-product of inertia Pitch angle
L Total aircraft lift Roll angle
Heading angle
*Flight Test Weapon Systems Officer
Member, AIAA -
I. BACKGROUND
**Aerospace hgineer
Member, AIAA
Currently, angle of attack is measured by
+Professor of Aeronautical Engineering external probes mounted on the forward fuselage of
Associate Fellow, AIAA operational aircraft. These rotating mechanical
probes have slots which then align the probes to
++Instructor in Aeronautical Engineering the local airflow, thus providing an estimate of
Member, AIAA the true angle of attack of the wing-body
Ihis paper is declared a work of the U . S combination in the airmass. Flight testing, to a
Government and is not subject to copyrighi great degree, relies on pitot probe mounted angle
protection in the United States. of attack vanes. The mounting of probes far ahead
on the fuselage eliminates some of the error due
The material reported herein is based on the to local flow effects, but many problems still
author's thesis submitted in partial exist with external angle of attack (AOA) probes
fulfillment of the requirements for the Master and vanes. First, being away from the center of
of Science degree at the Air Force Institute of gravity of the aircraft, the rotating probes are
Technology, Wright-Patterson AFB, Ohio. subject to pitch, yaw, and roll motion-induced
 errors which vary with Mach number and maneuver, Previous A~proaches
increasing the difficulty of calibration
dramatically. Second, being mechanical pick-off Four basic approaches have been used to
devices, these probes have a definite time lag that estimate angle of attack. Freeman's method in 1973
hampers their use in flight controls and autopilot (3) utilized accelerometers only, with no inertial
applications. This time lag can produce extremely hardware. The additional aircraft dynamics were
undesirable characteristics in most flight then estimated by control surface deflection
controls. Third, several airplanes have been lost pickoff devices and basic aircraft equations of
due to failure of the angle of attack probe motion. An angle of attack processor then took
inflight. This has been due to the requirement for inputs of flight condition, accelerations, and
stall inhibitor systems which attempt to prevent surface position to estimate angle of attack for
the aircraft from exceeding a specified angle of the required maneuver. This method relied heavily
attack. The flight controls will respond to a high on aerodynamic modeling using stability derivatives
AOA with a pitch down maneuver to reduce the sensed and surface position to evaluate the maneuver being
extreme angle of attack. If both probes have performed and estimate an angle of attack required.
failed in flight, the aircraft could pitch down for
no reason, other than the false AOA derived from A second algorithm, requiring less extensive
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

the failed probe. For such a vital instrument on a aerodynamic modeling, is provided by Petrov and
high performace aircraft, single probe failure Studnev, et a1 (4). This method requires precise
could result in aircraft loss. Finally, the very modeling of the coefficients of lift and drag and
accuracy of the mechanical probe in determining accelerometers to model aero performance in gliding
angle of attack is hard to judge. It can be on the flight. The linearized equations are applicable
order of 1.5 to 2 times the actual value of alpha only for small alpha and beta (less than 10
for some aircraft.(l) degrees) and do not provide the angles throughout
the maneuvering envelope.
The realm of supersonic flight makes greater
demands on the use of probes and vanes to determine Perhaps the most detailed work has been
angle of attack. Increased performance dictates a accomplished by Olhausen, using INS outputs for
grpater requirement for knowledge of this angle YF-16 flight test (5). The method uses INS
)!owever, at extremely high speeds, penalties from accelerometer and velocity measurments, along with
drag and aerodynamic heating require a "cleaner" Euler angle measuments. Basic aircraft equations
method for angle of attack determination.without of motion were solved using appropriate order
~eference to external devices in the airflow. Runge-Kutta integration techniques. This method is
Indeed, the increased reliance on angle of attack effective, but like most, it requires the
information at these critical speeds and flight assumption that sideslip and vertical windage are
conditions places a strong requirement on sensor zero.
redundancy for critical flight control systems.
That requirement could be fulfilled by such an The fourth method, and the one used by
internal AOA estimation system. Thacker, is state-space estimation (1). Thacker
only used two states, B and q .In this model, 8
Angle of attack can be determined with a high represents the pitch angle and q is the pitch
degree of accuracy from inertial reference systems rate. These quantities are perturbation values
and central air data computers. The current about some nominal flight condition. The two
generation of ring laser gyro inertial navigation states were shown by Thacker to successfully
systems (INS) and inertial reference systems (IRS) determine angle of attack to approximately . 5
have accuracies in pitch, roll and heading angles degrees. Logically, a more accurate math model of
on the order of 5.6 x radians(.032 deg.) and the aircraft dynamics should result in more
angular rate measurment accuracies on the order of. accurate determinations of angle of attack.
7.5 x 10-4radians/second ( ,043 degrees/sec) (2) .
Using the three INS rates, accelerations, and Proposal
central air data computer (CADC) inputs, the angle
of attack can be estimated with a moment summation Angle of attack estimation has two inherently
and lift model. This lift can then be compared to opposing requirements, speed and accuracy. Speed
the current aircraft weight model and measured load of calculation is a critical requirement for flight
factor. An angle of attack estimation can then be control usage of alpha, as well as for pilot
generated based on the required load factor. information and three dimensional windage
derivation. Calculations must be minimized, with a
The speed of the digital INS, combined with judicious use of assumptions, while retaining
its accuracy and reliability provide the following accuracy to . 5 degrees. On the other hand, test
advantages. Primarily, if angle of attack can standards require an accurate knowledge of a.
indeed be estimated with a high degree of These requirements then drive the estimator design.
certainty, the speed of the INS/computer
calculations implies that the information can be Freeman's work with the alpha estimator
used in conjunction with advanced flight control provides a good basis for the inflight, real-time
systems as a feedback quantity. Being internal to portion of the alpha estimator problem.
the aircraft and extremely accurate, the INS will Incorporation of INS rates, angles and
eliminate local flow and Mach effects that must now accelerations can dramatically reduce the
be corrected in raw external probe data. Finally, requirement for extensive stability derivative
the reliability of the current INS systems, and modeling, and increase the accuracy and calculation
even higher reliability of the ring laser INS'S, speed.
adds sensor redundancy to probe derived AOA
systems.
 --
11. ANGLE OF ATTACK ESTIMATION THEORY empennage, and the resulting forces. While it is
possible to model downwash and analytically
determine tail lift, the equations quickly become
--
Basic Lift Eauation unmanagable in even slight maneuvering. In
addition, the algorithm would then vary
Angle of attack is directly related to the extensively with tail surface control design. The
coefficient of lift of the wing-body combination. simpler and more direct method is the calculation
If the total lift on the wing-body is obtained, of the lift contribution of the tail surfaces as
angle of attack may be subsequently computed. determined through aircraft moments.
Total lift is the sum of the lift of the wing-body
combination acting at the corresponding wing-body The moment action on the total aircraft can be
aerodynamic center, and the lift of the tail, described adequately by the zero-lift pitching
acting at the aerodynamic center of the empennage. moment, wing-body lift, and the moments due to the
tail lift. Since the lift is always perpendicular
to the relative wind, the resulting moments and
arms are shown in Figure 2.

-Ar&-
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

C.L. I -
7.

Fig 1 Basic Longitudinal Aerodynamic Forces

Denoting this lift of the wing-body combination as Fig 2 Moments on an Aircraft in Fligllr-
, the lift of the tail as L the total lift,
Lw E T'
L, is given by:
L = L + L (1) where m is the zero-lift pitching moment about the
WB T
quarter-chord and XT is the distance between the
But total lift can also be defined by:
center of gravity of the aircraft and the
aerodynamic center of the horizontal tail. Xm is
the corresponding distance between the center of
where n is load factor and W is total gravity of the aircraft and the wing-body
aircraft weight. Then: aerodynamic center.

The static margin will place the center of

gravity less than ten percent of the mean
Wing-body lift can be expressed in a standard aerodynamic chord away from the aerodynamic center.
non-dimensional coefficient C , where: A further assumption is that the lift of the
m wing-body combination acts at a single point called
the wing-body aerodynamic center. These two basic
assumptions are generally held to be true for
conventional aircraft throughout a large portion of
their flight regimes. In addition, centerline
thrust is assumed, resulting in no moments due to
thrust, as well as drag.
where p is atmospheric density at flight altitude,
V is true airspeed, and S is wing area. The term m as used here refers to the
Rearranging:
zero-lift pitching moment of the wing-body-tail
combination. As such, the conditions at zero lift
require that lift of the wing-body offset the lift
of the tail. Thus the terms L and LWB actually
so that in terms of load factor: refer to incremental lift from the zero-lift
condition. However, the zero-lift values of
wing-body and tail lift are negligible when
compared to total wing-body and tail lift in 1-g
flight. The simplification will therefore be made
that LT and Lm are total lift terms and not
Moment Summation incremental terms from the zero-lift condition. So
the summation of moments yields:
Lift on the tail must be modeled next. This
can be accomplished by applying an analagous
equation, but the effects of elevator deflection,
downwash, and surface position pickoffs, as well as
wake effects must be considered. These effects are
where X is the wing-body static margin as defined
difficult to account for due to the inaccuracies of WB
determining exact tail deflection, as well as the by the expression:
errors in aerodynamic modeling of the flow over the
 Considering the standard pitching moment and (15)
m
equation we obtain: C =
0

m 1 2 -
0 -pVSc

Solving for C yields the primary equation

WB
where q is pitch acceleration, p is roll rate and
r is yaw rate. These quantities are all for the estimator.
available directly from a USAF standard INS.
Ix,Iy, and IZ are respective moments of inertia
about the X, Y, and Z axes and Ixz is the
applicable cross product of inertia calculated in
the body axis.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

If we neglect the terms involving the product

of inertia times the rates-squared equation (8) It can then be noted that angle of attack is a
becomes : function of C
WB
Cm = qIY + pr(Ix - I,) (10)

This is a valid assumption due to the relative size

of the product of inertia. Now, only the moments
of inertia need to be modeled throughout the flight
regime. with M being Mach and h being altitude

So, combining equations (7) and (9) yields: This functional relationship can be developed
for each specific aircraft to be estimated. Mach
number and altitude are direct CADC outputs, while
can be derived form equation (16).
CL
WB

Now solving for L :

The change in coefficient of lift with a
change in a is assumed to be instantaneous, with no
associated dynamics. Thus the angle of attack
required for a specific coefficent of lift to
exist, as determined by aircraft accelerations and
moments, may be calculated at any point in time.

--
Load Factor Determination
This equation for the lift of the tail
provides several advantages. The algorithm does A required input to equation (16) to determine
not require extensive aerodynamic modeling of tail CL is load factor (n). The load factor is
effects. The theory is based entirely on the WB
effect of the developed tail lift on the aircraft defined as the ratio of the acceleration of the
pitch, roll and yaw motions. In that light, there aircraft normal to the flight path within the plane
is no downwash calculation error or surface of symmetry to the acceleration of gravity. INS
position indicator requirement. Indeed, the linear acceleration would then yield only
algorithm doesn't care if the horizontal tail acceleration normal to the aircraft within the
surface is a conventionally flapped elevator, a plane of symmetry. With zero angle of attack, the
full-flying stabilator with variable trim tabs, of relationship of a and n in level flight is:
Z
even a differential stabilator with roll control
mixing.

Equation (11) can now be introduced into

equation (6) with the result:
where a is normal acceleration to the aircraft
Z

and g is the acceleration of gravity at the

earth's surface, which is assumed to be a constant
for the flight conditions examined.
or: This equation can be reasonably corrected for
large angle of attack and flight path angle and the
existence of roll 4 by rotating the acceleration
vector to a position normal to the flight path.
Figure 3 depicts the difference in orientation of
the load factor and normal acceleration to the
flight .path.
where :
In straight and level flight, normal
acceleration is 0 , but load factor is 1. However,
 Solving for n in terms of body axis
accelerations and Euler angles yields:

-(cosa )a
G BZ
+ (sina )a
G Bx
n = + (COS~)(COS$) (23)
g

This load factor , derived from INS

accelerations and Euler angles, can become an input
Fig 3 Relationship of Load Factor and Normal to the primary estimator equation (16). aG' or the
Acceleration
approximate angle of attack, is the only value that
at a climb angle of 90 degrees, this load factor, needs to be calculated for use in the load factor
while normal acceleration is 0 , becomes 0 also. equation (23). A no-wind assumption will allow
The same is true in a bank of 90 degrees while quick compuiation of this rough guess.
maintaining a steady course. Normal acceleration
is 0, and load factor is also 0. This change from
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

normal acceleration to load factor, accounting for No -

- Wind Angle of Attack
large pitch and roll angles can be found by
rotating the body normal acceleration to flight A no-wind angle of attack can be immediately
path normal acceleration, a in the wind axis obtained from the INS. All inertial velocities are
Wz ' actually groundspeeds over the locally flat earth.
system. If we assume for this guess only that wind is
negligible as compared to groundspeed, the INS
However, it is immediately evident that prior groundspeeds will also be true airspeeds, and an
knowledge of angle of attack and sideslip is initial a can be derived from inertial velocities
required for this rotation. As this is only used
for a small correction factor to obtain an accurate in the X , Y, and Z directions
load factor, n , some simple approximations may be
used for a and p . For this case only, sideslip is
assumed 0. A no-wind angle of attack can be
quickly computed using direct INS data. While not
accurate for estimator purposes, this approximate u
can serve very well to correct body axis
accelerations to approximate flight path
accelerations.

Then, using the transformation matrix. normal

acceleration in the wind axis system is:

a
Wz
=- (sinu )a
G Bx
+ (cosaG )a Bz (19)

But normal acceleration in the wind axis may also These body velocities can then be evaluated to find
be written as: an approximate angle of attack using the body axis
relationship :
a - (component of g) = -ng (20)
Wz Wz

So to find the component of gravity in the z

direction in the wind axis, another transformation
matrix is used to rotate g from the vertical
reference frame to the wind axis. It should be
noted that wind Euler angles are used in this
transformation. However, for the purposes of this This no-wind guess of angle of attack is then
estimate of n , body axis Euler angles from the INS used only to make the small rotational correction
will be used. on normal acceleration to the load factor.

Then the component of g in the z direction in Flow Dia~rarn

wind axis is:
The overall components of the inflight
estimator act together to compute the no-wind guess
of angle of attack. At this point, all the
information necessary for estimation of actual
angle of attack is available from the INS, the
Substituting equations (19) and (21) into CADC, and the model of the current aircraft weight
equation (20) yields the following equation for and configuration. A CADC output is be used to
load factor determination. determine the values for the 5 stability
derivatives that must be modeled for varying flight
conditions. The angle of attack is then be
(cosa )a -(sine )a =-ng + gcos0cosq5 (22) computed by the estimator. This angle of attack
G Bz G Bx
and the INS ground speeds can then be used to find
a three-dimensional wind, given the current true
where a is the initial angle of attack guess, 0 airspeed. The flow diagram in figure 4 depicts
and 4 are pitch and roll angles. this action.
 stabilator action.

The NASA F-15A Highly Integrated Digital Engine

Control (HIDEC) was instrumented with a calibrated
yaw-angle of attack-pitot-static (YAPS) head. This
vane provided the baseline a information that the
inflight estimator was evaluated against. In
addition, the flight test boom also provided pitot
pressure for calculation of true airspeed within
the central air data computer (CADC). This
aircraft was also configured with production angle
Fig 4 Inflight Estimator Flow Diagram of attack probes which were used as a secondary
comparison of the angle of attack estimator.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

111. FLIGHT TEST The aircraft was fitted with a USAF standard
inertial navigation system. This system is a fully
Test Philosophv self-contained dead reckoning navigation system.
It continuously computes aircraft position by
The purpose of the flight test is to ensure double integration from a known starting point.
proper operation of the inflight angle of attack
Aircraft ground speed and attitude are interim
routine using actual flight data. This flight data
computations prior to position computation. This
collection is critical to ensure operation with
INS consists of three major components. The first
is the actual inertial measuring unit (IMU) which
noise-corrupted input data. The noisy nature of houses the gyros and accelerometers to sense
the actual INS measurements, coupled with possible aircraft motion. The second is the (IMU) mount
unknown or assumed effects, will test the ability which provides precision mounting and alignment of
of the algorithm to recover the necessary data, and the system to the aircraft body axis. The third
perform proper computations to calculate angle of component is the navigation control indicator which
attack. Unlike simulated data, however, no true interfaces the INS to the central computer of the
angle of attack is known. Measured angle of attack aircraft and also allows pilot control of the
from calibrated vanes, normal aircraft functions of the system. The INS was fully
instrumentation, and computer modeled performance instrumented.
will be correlated against the INS derived a. This
information will provide an acceptable measure of Weight was available through production fuel
the accuracy of the INS derived values against the sensors on board the aircraft which measure fuel
more conventional approaches in obtaining a. remaining in each tank to an accuracy of 200 lb.
The weight of the aircraft could then be easily
The initial flight test consisted of low calculated, knowing basic aircraft weight, serviced
performance longitudinal maneuvering flight data fluid weight, stores weight, and the changing fuel
tape analysis only. The reason for this is weight.
two-fold. First, it will demonstrate the
applicability of these methods for a determination Mach nunher and altitude signals were obtained
in large, transport type aircraft. As these from the air data computer, along with true
aircraft do not engage in high-g maneuvering or airspeed. Air density was calculated through the
extreme flight attitudes, the basic assumptions standard exponential atmosphere equation for input
should hold throughout the nominal flight regime. into the primary estimator equation. Overall, no
The applicability of these determination techniques signals were used which would not be obtainable
will be demonstrated for large aircraft in both the through current INS-equipped production aircraft
inflight and flight test analysis phases. instrumentation.
Secondly, nominal inflight accuracy should give an
indication of the proper formulation of the
estimator. The absence of high-g, coupled flight Data Reduction
conditions allows a straight forward evaluation of
the estimator. The robustness of the estimator in All inflight data was reduced using NASA real
maneuvering flight will be discussed later. This time and mainframe computers. Aircraft data
flight test was conducted in cooperation with NASA telemetry was retrieved from the computer for the
and in conjuction with a NASA propulsion test specific maneuvers required. This data consisted
flight. of time tagged values for all signals specified in
the estimator flow chart. The basic estimator
program, as implemented in FORTRAN 77, was altered
Test Aircraft to allow use in the time tagged seauential mode of
operation on the compuier system. This
Initial flight test was accomplished using a consideration was important in that the inflight
NASA F-15A aircraft manufactured by the McDonnell estimator was designed for real time operation, and
Douglas Corporation. This aircraft, S/N 10281, is the system provided that capability in reducing
an F-15A air-superiority fighter modified for flight data. Once the data was calculated
digital engine and control testing. It is a single sequentially, the results were *lotted and conred
s ' o t
seat aircraft powered by two Pratt and Whitney at #e s- ti& tag.
F-100 engines. Flight controls consist of twin
vertical stabilizers mounting a single rudder on
each. Lateral control is effected by ailerons on
the outboard wing surfaces, aided by split
stabilators, with pitch controlled by symmetrical
 Test Methods. Conditions, and Results There are two areas of interest in Figure 5 .
The most obvious is the initial portions of the a
The optimum flight test technique for stable data traces, where an approximately 1.3 degree
longitudinal flight at varying angle of attack was noise in estimator angle of attack is apparent.
determined to be the level acceleration. In The actual level acceleration maneuver does not
general, the aircraft was stabilized on conditions begin until 17 seconds into the data trace. This
in the slow speed regime with engines at the initial, high angle of attack regime is the climb
planned military or maximum power settings. This into the maneuver at slow flight. This slow flight
procedure required a climbing entry to the test is characterized by thrust set at test power, in
point. The aircraft was then allowed to accelerate this case military setting. The result is slow,
to its maximum speed while maintaining constant climbing flight in moderate buffet, with some
altitude and one-g flight. This required a internal vibration present. The difference between
constant reduction in angle of attack throughout a sources is reasonably constant during this entry
the level acceleration maneuver. into the acceleration. Initial skepticism of the
estimator would give more credence to the YAPS boom
Three level acceleration test points were a than the estimator. It should be realized that
planned to evaluate estimator angle of attack. there is no absolute source of angle of attack in
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

Additionally, Mach and altitude effects were this test. However, the a estimator would be
investigated using military and maximum susceptible to airframe buffet and vibration,
accelerations at three different altitudes. Level cluttering the normal load factor signal at these
acceleration test points are summarized in Table I. low speed conditions. It is of note, though, that
the a estimator does follow the peaks of the YAPS
Table I. boom a exactly, remaining within 1.5 degrees until
Level Acceleretion Test Points the initiation of the pushover at the beginning of
the level acceleration, occurring at 15 seconds
~txrPoinr UTITUDE POVEP MACH BAND into the trace. The YAPS boom appears to be the
I 10,- HIL 0.5 - 0.9 the most accurate source during this slow flight
a 10.000 MIL 0.4 - 0.9 phase.
3 40,OW MAX 0.5 - 1.5
Once the level acceleration has begun, the two
angles of attack remain within 0 . 5 degrees of each
other throughout the rest of the maneuver. For the
The results of the military power level first half of the acceleration, the inflight a
acceleration in test point 1 are depicted in Figure estimator is below the YAPS boom a. At 0 . 8 2 Mach,
5. This acceleration was from 0 . 5 to 0 . 9 Mach at the traces coincide, with estimated a becoming
2 0 , 0 0 0 FT. The data is presented as two angle of larger than boom a for the remainder of the trace.
attack traces. The first is AINF, or alpha They do stay generally within 0 . 5 degrees during
infinity, as derived from the YAPS boom. The this exchange. Overall, the two angle of attack
second is AWB, or alpha wing-body, representing the traces coincide well, with the exception of the
output of the inflight a estimator based on entry into the maneuver, during slow flight in
equation (17). Also accompanying the a traces are moderate buffet.
Mach and altitude data throughout the maneuver.
The next step in the buildup process to
evaluate the inflight a estimator is to introduce
abruptness into the estimation process, while still
F-~SA NSA ~ / 1 110aei restricting maneuvers to the longitudinal modes
Fuel: JP-4 Nominal C.O.
Mach 0 . 5 - 0 . 9 2 0 , 0 0 0 FT within the plane of gravity. The wings-level,
Flt 515 constant g pitch-up flight test technique was
2 . 1 MiliLary Power Level AccclersLion
considered the optimum for this phase of the flight
test. This technique required the aircraft to be
stabilized at a constant aim altitude and Mach
number. The aircraft was abruptly pitched to a
series of positive and negative constant g values,
much like a roller coaster. This technique was
accomplished within a standard 2 , 0 0 0 foot data
band. Three test points were evaluated during this
phase and are summarized in Table 11.

Table 11.
Pull-up Test Points
-1. 1
o 10. 20. 2 . 40. 5:. et. 72. CC. L:. :::

Figure 6 depicts the results of the abrupt

pitch evaluation. The upper graph shows
estimator/YAPS angle of attack correlation, while

D 10 rn
,
SO
-----
.a I r- r 6- s-
the lower graph shows corresponding normal load
factor. This figure includes all abrupt pitch test
7:.v
points. Of immediate note is the closeness with
which estimator a follows boom a below
Figure 5. Level Acceleration Results. approximately 16 degrees angle of attack.
 - F-1 SA NASA S 4 102.1 Table 111.
I i
rn1r.C
--------a hnl: P 4
n.=h d.; Nominal C.Q.
40,000 FT Wind-Up Turn Test Points

deviation during the sustained, high g portion of

the maneuver. Although the traces match in terms
of peak locations, they differ by almost a degree
at the sustained g point.

s. - -
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

3 c Wind Up Turn Hsneuver

-1. '
. ...
.*
-+ 1
0 as. .o m LOO. ..aI 10
.. 17s. ;::
.--
2 : L.. -.z
5
ZC. Zi. 4:.
-,.
5:.
.,,: .
C:. 7:. .:, $I, i::.

TICK

Figure 6. Pull-Up Test Results.

Negative g excursions match almostidentically. The

lack of high angle of attack modeling is the cause
of the deviation at the peak of the high g points,
as was noticed in the initial portions of the level
accelerations. However, the close correlation of
the separate angle of attack sources through rapid
changes in angle of attack and load factor do
support the basic concept of this form of a
estimator. It can indeed accurately recover angle
of attack with at least 0.5 degrees of precision in
upright, purely longitudinal motion.

The final stage of the quantitative flight

test evaluation of the angle of attack estimator is 0 10. Io. m. rg. .
TZ"Z
0:. 7:. e:. n: :::
examination of out-of-plane maneuvers. The purpose
of this phase of testing is to remove the gravity
vector from the longitudinal plane of the aircraft Figure 7. Wind-Up turn Results - 3 g.
and judge the effect of load factor and banked
flight of the estimator. The wind-up turn was Although the YAPS boom a does not provide an
judged to be the best flight test technique for absolute, true a , it should be the weighted
this flight test goal. In this maneuver, the preference. However, the boom a does show almost
aircraft is trimmed at a given Mach ,and altitude. 0.5 degrees worth of noise in its signal, while the
The aircraft is then steadily banked into a estimator is slighly smoother. The same result is
constant Mach turn while slowly increasing load true with the 5 g wind up turn. YAPS boom angle of
factor to the desired end point. This maneuver attack is consistently lower than estimator a at
takes place within a 2000 foot data band, as a the higher sustained g plateau, although its signal
descent is required to maintain constant Mach at a is much cleaner than the preceeding graph. Again,
trim power setting. Two test points were the peaks of each source match well, with no
identified for examination of the effect of g and noticable lag. Again, the lift model was not
bank angle on the estimator. The wind-up turn test corrected for load factor, and was a trimmed lift
points are defined in Table 111. curve. This could account for the deviations at
the sustained higher load factors, and indicate a
The wind up turn evaluation did uncover some requirement for a closer wing-body model of angle
angle of attack deviations in the estimator. of attack.
Figure 7 depicts the 3 g wind up turn results,
along with normal load factor achieved in the Another possibility is error in out of plane
maneuver. In this figure, there is some definite load factor calculation. This could be the result
 of actual accelerometer output as opposed to
theoretical accelerations about the center of
gravity. The result is an alteration in equation r-1%
!:=A: d?;4
IlUA wn loall
Nomlnml C.O.
(26) to replace the cosOcos~term with 1.0 as these 20.000 FT
FlL 515
angles are accounted for due to normal Robustness Uaneuver
accelerometer bias of 1 g. This bias is included
in all normal accelerometers to take into account
the gravitational pull of the earth. In straight
and level, unaccelerated flight, the normal
accelerometer reads 0.0 ft/sec2 acceleration of the
aircraft center of gravity. However, the aircraft
is indeed under 32.2 ft/sec2 or 1 g acceleration
due to the earth's pull.

Overall, the a estimator correlated to the

YAPS boom a well. Under most conditions, the
results were within the specified 0.5 degree
deviation. Where the deviations were greater than
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

that value, the estimator errors were explainable

and indicate a need to form a more precise C
WB
model than a linear regression on two independent
variables as was accomplished for this research. A
full, maneuvering flight demonstration will
indicate the degree to which the current model and
equations are adequate for high g, rolling flight
out of the longitudinal plane of motion.
Robustness

The purpose of this portion of the flight test

was demonstration only. The attempt was made to
devise a single flight test technique to quickly
and efficiently demonstrate any possible area of
weakness in the angle of attack estimator. In
other words, this portion of the investigation was
tq highlight areas to troubleshoot the estimator
algorithm or to point where future investigations
should be directed. Figure 8. Typical Robustness Results.

The robustness check was only accomplished at

one flight condition due to constrained flight test reasonably constant. However, as the roll to
time. The maneuver that was developed was inverted attitude begins at 13 seconds into the
therefore a dynamic one encompassing all expected trace, the curves come to within 0.2 degrees. The
problem areas such as loaded rolls and longitudinal 4 g split-S shows very close correlation, through
pulls out of the local horizontal plane. The recovery at 38 seconds. Of note during this phase
modified split-S maneuver was performed in of the maneuver is the noise within the a estimator
conjunction with a NASA propulsion test. signal. While no more than approximately 0.8
degrees, it disrupts an otherwise close match under
The overall robustness maneuver can be divided sustained g, inverted flight. Note that this
into four distinct segments. Initially, the situation did not occur with abrupt pitch
aircraft is flown in a true north heading. Once maneuvers. This deficiency, on the order of 0.3
established on conditions, a 30 degree banked. degrees per g always occurs to the high side.
climbing turn at 2 g's is begun. This is indeed a Again, a trimmed lift curve at 1 g was used as the
climbing turn, as a level 30 degree turn requires model. A higher order model of a as a function of
only 1.2 g's. Upon stabilization in this turn, the CL , altitude, Mach and load factor may provide
pilot then rolls inverted in the same direction as m
rolling into the 2 g turn initially. At this the key. However, correlation during these extreme
point, the pilot then begins a sustained 4 g pull maneuvers was quite acceptable, considering the
in a split-S maneuver, recovering in an upright, multiple changes in plane and velocity vector
wings le'vel attitude. The robustness maneuver thus during 40 seconds of robustness evaluations. The
evaluates a climbing, loaded turn, a pure roll out maximum difference was 2 degrees as observed in the
of the local horizontal plane, and a loaded pull Wind Up Turn test, and this occurred under
with the gravity vector constantly moving approximately 5.5 g's. In addition, it is
throughout the aircraft axis system. important to note that when in error under g, the a
estimator was always higher than YAPS boom angles
of attack.
The results of the typical robustness maneuver
are depicted in Figure 8. Pitch, roll and yaw
Two secondary aerodynamic effects were
rates are presented with the estimator and YAPS
observed which deserve note. At subsonic speeds,
boom a traces. The initial roll into the maneuver
with good estimator correlation to YAPS values of a
begins at 2 seconds into the trace. The initial
(within .2 degrees) , extension of the massive
difference between the higher estimator trace and
F-15A speedbrake caused an immediate jump to 0.5
the YAPS boom trace is approximately 0.7 degrees.
degrees difference between a sources. Two possible
As with the loaded rolls presented in the Wind Up explanations are readily apparent. First, the
Turn test, the difference between traces remains
 large speedbrake alters the aerodynamic capability. This limitation was based
characteristics of the wing, invalidating the a solely on regression algorithms available
WB
model developed for clean configurations only. The to formulate this estimator version. A
second aerodynamic effect was a 1.2 degree more powerful regression tool would allow
estimator to YAPS boom difference during air to air incorporation of flight regimes that were
refueling at 320 knots indicated airspeed. The not modeled by the current estimator.
effect here seems to be caused by tanker wake
effects on the F-15A local airflow, impacting both (2) The inflight a estimator was
the YAPS noseboom and the local aerodynamics as evaluated with flight test data, and
modeled by the a estimator. robustness examined during real-time
flight test. Several comments can be made
Overall, the inflight a estimator performance as a result of this test. The primary
was acceptable as qualitatively evaluated during result is that the concept is indeed
this flight and measured against YAPS boom values. feasable. With a basic multivariable,
linear regression modeling technique,
accurate angle of attack estimates can be
IV. CONCLUSIONS made in all attitudes to approximately 0.5
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

degrees, as demonstrated by the robustness

General maneuvers. Two areas of interest need to
be highlighted. Pitch acceleration was
In broad terms, the objectives of this thesis not available in the HIDEC configuration.
were met and the concept of angle of attack If it were, it could be expected to be a
estimators using standard inertial reference noisy signal due to the algorithms used in
platforms is highly feasible. The inflight a its calculation. A difference of pitch
estimator served to demonstrate that the concept of
an angle of attack estimator which is accurate to rates per time interval was used to
0.5 degrees is not only possible, but available for approximate 4 . This proved to be a
real time inflight use, with current generation satisfactory approximation of the term,
mechanical inertial navigation systems. Specific which was usually very close to 0.0. The
overall correction term in the primary
conclusions follow:
estimator equation was subsequently small
when compared to the other terms, and
(1) An inflight angle of attack estimator
could actually be neglected with only
was successfully developed for use. It is
of note that the self imposed requirement limited loss of accuracy. Secondly, the
rotation of the normal acceleration from
for minimum calulations, and hence maximum
computational speed, did not restrict the body to wind axes demonstrated the
difference between theoretical equations
accuracy of the estimator as tested.
Another goal under this objective was also and reality. The rotation equation itself
reached. All signals used in the inflight was based of c.g. accelerations of a and
BZ
a estimator were from standard INS or a . In straight and level unaccelerated
BX
onboard data sensors carried by almost all
operational military aircraft. In other flight in the wind axis system, these
words, apart from the data telemetry variables should be zero, whether the
systems, no special flight test aircraft is upright or inverted. However,
instrumentation was required for this a real accelerometer which is trimmed to
estimator. In addition, mechanical INS read 0 ft/sec2 a in upright
BZ .
platforms were used, allowing unaccelerated flight will read 64.4
incorporation of this estimator in current ft/sec2 in inverted unaccelerated flight.
generation aircraft as the need arises. The result is that the correction term in
Accuracy of the system was highly the rotation equation, cos@cosd, which
dependent on the modeling of the stability represents the component of the earth's
derivatives. The three moments of inertia gravity vector, as a correction to a
proved to be secondary effects, not perfect accelerometer, is taken into
requiring extensive mathematical models. account by the real accelerometer. It can
However, Cm was a critical factor and was be replaced simply by the constant one g
0 acceleration of gravity, as the rotational
used specifically to tune the system to correction is automatically applied in the
YAPS boom angle of attack at the beginning real accelerometer readings.
of flight test data evaluation. This step
in the test process points to an area of (3) The robustness maneuver demonstrated
limitation. An aircraft still must that the concept of a simple, efficient
undergo some flight testing with a YAPS angle of attack estimator was achievable.
boom prior to using an a estimator. The The estimator was accurate to within the
computer program must be "calibrated" to 0.5 degree desired specification with two
the aircraft, at least in the initial exceptions. First, under sustained higher
flight test stages, as the a estimator g loadings, the estimator accuracy was
requires historical data to model the degraded proportional to the loading.
lifting system. It is then obvious that This indicates that a g correction term
the most critical model must be the lift needs to be modeled in the a regression.
curve, with a as a function of Mach number Under high g, the estimator was always
and altitude. The most significant high, and this is the more favorable of
limitations of the estimator as formulated the possibilities. Use of the trim C
for this research was the lack of high a a
modeling and the lack of a supersonic curves could be the cause of this, and
 simple modeling under g of the wing-body
CL should suffice to correct the
Q
estimator back to predicted values.

(4) Finally, in the calculation of a

GUESS
a known singularity in the Euler angle
rotations was reached at 90 degrees pitch
angle. This situation could easily be
rectified by reverting to an earlier guess
of a , and holding that guess between the
80 to 90 degree pitch angle phases of
flight. A second solution would be a hold
register, allowing the previous estimated
a to become a for the next time
GUESS
segment. This would eliminate the need
for Euler angle rotations to find a in
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on December 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/6.1988-4351

GUESS
the first place, and seems to be the
better solution.

In conclusion, the area of angle of attack

estimation is an exciting, challenging arena,
encompassing many disciplines of aeronautical
engineering and statistical estimation. Its uses
are bounded only by imagination, and its
possibilities for incorporation into current
aircraft are limited only by desire.

BIBLIGRAPHY

l.Thacker, Thomas.
Use of State Estimation Calculate Angle
Of Attack Position Error From Flight Test
--
Data M.S. Thesis, GAE/M/85J-3,
Wright-Patterson AFB, Ohio: Air Force
Institute of Technology, Oct 85.

2.Department of the Air Force.

Specification for USAF
Standard Form, Fit and Function (F3) Medium
Accuracy Inertial Navigation Set/Unit SNU
84-1. Aeronautical Systems Division,
Wright-Patterson AFB, Ohio, 15 Oct 84.

3.Freeman, Duane.
Angle of Attack Computation Svstem AFFDL-
TR-73-89. Wright-Patterson AFB, Ohio:
Air Force Flight Dynamics Laboratory,
Oct 73.

4.Petrov, B. and R. Studnev et al.

Determinin~Angles of Attack and Sideslip
l?y Signals from Accelerometers
Installed On Board an Aircraft
FTD-ID(RS)T-1657-77. "Izvestiya- Vysshikh
-
Uchebnykh Zavedeniy, 'Priborostroyeniye',"
Vol 18, Nr 10, 1975.

5.0lhausen,J. "Use of a Navigation

Platform for Performance Instrumentation on
the YF-16". AIAA 13th Aerospace Sciences
Meeting. AIM-75-32. American Institute
of Aeronautics and Astronautics, Pasedena,
CA, Jan 75.