See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/272392904

Generic Global Aerodynamic Model for Aircraft

Article in Journal of Aircraft · January 2015

DOI: 10.2514/1.C032888

CITATIONS READS
51 1,521

2 authors:

Jared Grauer Eugene Morelli

NASA NASA
83 PUBLICATIONS 967 CITATIONS 89 PUBLICATIONS 3,534 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jared Grauer on 09 December 2018.

The user has requested enhancement of the downloaded file.

 A Generic Global Aerodynamic Model for Aircraft

Jared A. Grauer∗ and Eugene A. Morelli†

NASA Langley Research Center, Hampton, Virginia, 23681

Multivariate orthogonal function modeling was applied to wind tunnel databases for

eight different aircraft to identify a generic global aerodynamic model structure that could

be used for any of the aircraft. For each aircraft database and each nondimensional aero-

dynamic coefficient, global models were identified from multivariate polynomials in the

nondimensional states and controls, using an orthogonalization procedure. A predicted

squared-error criterion was used to automatically select the model terms. Modeling terms

selected in at least half of the analyses, which totaled 45 terms, were retained to form the

generic global aerodynamic (GGA) model structure. Least squares was used to estimate

the model parameters and associated uncertainty that best fit the GGA model struc-

ture to each database. The result was a single generic aerodynamic model structure that

could be used to accurately characterize the global aerodynamics for any of the eight air-

craft, simply by changing the values of the model parameters. Nonlinear flight simulations

were used to demonstrate that the GGA model accurately reproduces trim solutions, local

dynamic behavior, and global dynamic behavior under large-amplitude excitation. This

compact global aerodynamic model can decrease flight computer memory requirements for

implementing onboard fault detection or flight control systems, enable quick changes for

conceptual aircraft models, and provide smooth analytical functional representations of the

global aerodynamics for control and optimization applications. All information required

to construct global aerodynamic models for nonlinear simulations of the eight aircraft is

provided in this paper.

Nomenclature

Roman cov(.) covariance

b wing span [ft] c̄ mean aerodynamic chord [ft]

CD , CY , CL aerodynamic force coefficients h altitude [ft]

Cl , Cm , Cn aerodynamic moment coefficients p, q, r roll, pitch, and yaw rates [rad/s]

∗ Research Engineer, Dynamic Systems and Control Branch, MS 308, AIAA Member
† Research Engineer, Dynamic Systems and Control Branch, MS 308, AIAA Associate Fellow

1 of 23

American Institute of Aeronautics and Astronautics

q̄ dynamic pressure [lbf/ft2 ] δa , δe , δr aileron, elevator, rudder deflection [rad]

R2 coefficient of determination δt throttle position [0,1]

S wing reference area [ft2 ] σ2 variance

t time [s]

V airspeed [ft/s] Superscripts

˙ time derivative
T
Greek transpose

α angle of attack [rad] ˆ estimate

β sideslip angle [rad]

2 of 23

American Institute of Aeronautics and Astronautics

 I. Introduction

lobal aerodynamic modeling experience at the NASA Langley Research Center led the authors to

Gobserve that the nondimensional aerodynamic coefficients for numerous different aircraft had a similar
topology. In light of this, it was hypothesized that a single model structure could potentially describe the

aerodynamic coefficients for many different aircraft over a large portion of their flight envelopes. This work

is an investigation of that idea.

A successful generic model structure must meet several requirements for practicality. The model should

involve a relatively small number of parameters that could quickly be altered, for example to change from a

transport to a fighter type aircraft, or from a T-tail to a cruciform or conventional tail configuration. Despite

this compactness, the model should have sufficient complexity to accurately predict the dynamic response

of many different aircraft. The model should be global in the sense that it is valid over a large portion of

the flight envelope for each aircraft, in contrast to local perturbation models. The model should also be

formulated in a manner that enables a fundamental understanding of the functional dependencies.

Such a model structure would have numerous applications. This model structure could provide flight

simulator manufacturers with a straightforward and cost-effective method for complying with House Res-

olution No. 5900, which mandates that U.S. commercial airline pilots be trained in recovering from stalls

and upsets. Only a small list of aerodynamic parameter values would need to be loaded to change aircraft

or aerodynamic behavior, which would help with this new safety training requirement. This functionality

also supports the aircraft conceptual design process in that global aerodynamic models could rapidly be

generated based on geometry or configuration changes, akin to DATCOM,1 for performance analysis and

control design. An analytical formulation of the model using the aircraft states and controls in a Taylor

series expansion not only helps to gain insight, but also provides smooth gradient functions for optimization

routines. The relatively small number of parameters needed in such a model makes it small enough to store

in onboard flight software for model-based control design.

To determine such a model, key features in the aircraft aerodynamics need to be discerned. Approaches for

this have included the eigenmode analysis and the proper orthogonal decomposition, where the dependencies

are projected onto modal or principle components that may be ordered according to their importance.

These can then be truncated to form a reduced-order model that best describes the data in a least-squares

sense using the modeling terms retained. This method has been applied to many aerospace problems, such

as unsteady aerodynamics and aeroelasticity.2, 3 Similarly, singular value and eigenvalue decompositions

can be applied to databases of aerodynamic information. Researchers have used this technique to reduce

memory requirements for describing large wind tunnel databases,4, 5 as well as similar problems in other

fields.6, 7, 8 Other works have used basis functions, such as the Fourier series, Legendre series, and Chebyshev

3 of 23

American Institute of Aeronautics and Astronautics

polynomials.9

In this paper, multivariate orthgonal function (MOF) modeling10, 11, 12 was used to approximate large

wind tunnel databases for eight different aircraft, and to select the Global Generic Aerodynamics (GGA)

model. This process begins by orthogonalizing a large pool of candidate regressors based on a nonlinear

Taylor series expansion in the state and control variables. The regressors are then ordered according to

effectiveness in modeling the aerodynamic data, and a statistical metric is used to select the number of

modeling functions retained in the model. The least-squares estimator is then applied to determine model

parameters and uncertainty bounds. This method has been automated and is more efficient than iterative,

time-consuming methods for model structure determination such as step-wise regression.12 Modeling using

MOFs is similar to proper orthogonal decomposition methods in that the goal is to approximate the data

efficiently and in a least-squares sense using a small number of modeling terms. MOFs differ in that the model

truncation error is reflected in the estimated parameter Cramér-Rao bounds, and that there is a statistical

metric for selecting the number of modeling terms to retain. Additionally, because orthogonalization removes

only correlated portions of the regressors, the modeling terms retain a physical connection that is common

to all aircraft. Furthermore, the modeling terms represent a natural basis for describing the aerodynamics

that can result in fewer terms and more physical insight than using arbitrarily-chosen functions.

The paper is organized as follows. Section II presents the aerodynamic coefficients, ordinary least squares,

and model structure determination using MOFs. Section III describes the eight nonlinear flight simulations

used to identify models. Section IV presents the MOF models for the eight aircraft. Afterward, these models

are compared and the GGA model structure is presented. Model parameters are then estimated for the

aircraft and the GGA model is validated against the original wind-tunnel databases and nonlinear flight

simulations.

Computer programs for modeling with multivariate orthogonal functions, least-squares regression with

colored residuals, and the F-16C nonlinear simulation are all included in a MATLAB R toolbox called System

IDentification Programs for AirCraft (SIDPAC).12 This software was developed at NASA Langley Research

Center and is continually expanded and improved.

4 of 23

American Institute of Aeronautics and Astronautics

 II. Methods

II.A. Aerodynamic Coefficients

The aerodynamic coefficients can be computed from measurements of the body-frame applied forces X,

Y , Z and moments L, M , N as

    
 CD   − cos α 0 + sin α   X 
  1   
=
    
 CY 0 1 0 Y 
 q̄S 
 
  
    
CL + sin α 0 − cos α Z
    
 Cl   1/b 0  L 
0
  1   
= (1)
    
 Cm
 q̄S  0 1/c̄ 0 
  M 
  
    
Cn 0 0 1/b N

Drag and lift forces were used instead of body-frame longitudinal and heave forces because the aerodynamics

are natively written in the wind frame for most aerodynamic databases. This form is typically used when

measuring forces and moments during a wind tunnel test. If instead flight test data are available, standard

modeling assumptions can be used12, 13, 14 to compute these as

    
 CD   − cos α 0 − sin α   max − T 
  1   
 CY  =
    
  q̄S 
 0 1 0   may



    
CL + sin α 0 + cos α maz
    
 Cl   1/b 0 0   Ixx ṗ − Ixz (pq + ṙ) + (Izz − Iyy )qr 
  1   
= 2 2 (2)
    
  Iyy q̇ + (Ixx − Izz )pr + Ixz (p − r )
 Cm
 q̄S  0 1/c̄ 0 
  
 
    
Cn 0 0 1/b Izz ṙ − Ixz (ṗ − qr) + (Iyy − Ixx )pq

where m is the mass, Ixx , Iyy , Izz , Ixz are elements of the inertia tensor, ax , ay , az are linear accelerations,

and T is the thrust.

II.B. Ordinary Least Squares

Consider the model

z = y+ν

= Xθ + ν (3)

5 of 23

American Institute of Aeronautics and Astronautics

for N measurements in vector z, where y is the model output, X = [ x1 x2 ... xn ] is a matrix of n
model terms (also called regressors), θ is a vector of model parameters, and ν is the modeling error. The

least-squares cost function

1 T
J(θ) = (z − Xθ) (z − Xθ) (4)
2

is minimized by the solution

−1
θ̂ = XT X XT z (5)

The uncertainties of the estimated parameters are15, 16, 12

 
N N
−1 X X
−1
cov(θ̂) = XT X Rνν (i − j)xT (j) XT X


 x(i) (6)
i=1 j=1

where xT (i) is the ith row of X. The residuals are defined as

υ = z − ŷ

= z − Xθ̂ (7)

and the autocorrelation Rνν in Eq. (6) can be estimated as

N −k
1 X
R̂νν (k) = υ(i)υ(i + k) for k = 0, 1, 2, . . . , N − 1 (8)
N i=1

Simplified versions of Eq. (6) should not be used when determining GGA models because if Eq. (6) is not

used, model truncation leads to optimistic uncertainty estimates.16, 12

II.C. Model Structure Determination

The model structure, i.e. the list of model terms, must be known to use Eq. (5). There are several methods

available for the model selection process. In this work, a technique using Multivariate Orthogonal Functions

was used. An automated process having a statistical underpinning, this method has been successfully applied

to numerous practical problems,10, 11, 17, 18, 19, 12 and is briefly summarized here.

The process begins by selecting a matrix X of n candidate regressors, which can in general be nonlinear

functions of the explanatory variables. A modified Gram-Schmidt process is used to define orthogonalized

candidate regressors

p0 = 1

6 of 23

American Institute of Aeronautics and Astronautics

 j−1
X
pj = xj − γkj pk j = 1, 2, . . . , n (9)
k=0

where the first orthogonal function is selected as unity and then the remaining orthogonal functions are

recursively defined. During this process, the coefficients

pTk xj
γkj = k = 0, 1, . . . , j − 1 (10)
pTk pk

are defined, which populate the upper-triangular matrix

 
 1 γ01 γ02 ... γ0n 
 
 
 0 1 γ12 . . . γ1n 
 
 
G=
 0 0 1 . . . γ2n 
 (11)
 
 .. .. .. .. .. 

 . . . . . 

 
0 0 0 ... 1

The matrix G is a linear transform that relates the ordinary and orthogonalized regressors via

 
P= p0 p1 ... pn = XG−1 (12)

where pTi pj = 0 for i 6= j.

Substituting in the orthogonalized regressors, the least squares model Eq. (3) takes the form

z = Pa + ν (13)

and the parameter vector a can be computed analogously to Eq. (5). However, the orthogonalization process

diagonalizes the matrix PT P and decouples the least-squares problem so that the parameters become

âj = pTj z / pTj pj

j = 0, 1, 2, . . . , n (14)

which depend only on the measured output data and the j th orthogonal regressor, and do not depend on

the other model terms. Substituting Eq. (14) into Eq. (4), the least-squares cost function is

n
1 T 1 X T
2
pj z / pTj pj


J(â) = z z− (15)
2 2 j=0

where each term in the summation quantifies the cost reduction incurred when including the j th orthogonal

7 of 23

American Institute of Aeronautics and Astronautics

function in the model structure. The orthogonal functions can be ordered in terms of these values, which is

equivalent to ordering them in terms of their effectiveness in fitting the data.

A metric called the predicted squared error

1 T 2 n
PSE = (z − ŷ) (z − ŷ) + σmax (16)
N N

can be employed to determine how many ordered orthogonal functions should be retained in the final

model.20, 10, 12 The first term in the PSE is the mean squared fit error (MSFE), which monotonically de-

creases with each additional orthogonal modeling term and quantifies the error in the model fit. The second

term is the overfit penalty (OFP), which monotonically increases with each added model term and guards

against over-parameterizing the model and consequent poor prediction results. As each orthogonal function

is added to the model, the OFP part of the PSE is always increasing and the MSFE is always decreasing, so

there is always a single point at which the PSE is minimized. Choosing this point for selecting the number of
2
modeling terms results in a model with both low fit error and good prediction capability. The variance σmax

is the maximum model fit error variance, which is conservative in that it produces models with minimal

complexity,
N
2 1 X
σmax = [z(i) − z̄]2 (17)
N − 1 i=1

N
1 X
z̄ = z(i) (18)
N i=1

Once the orthogonal functions are selected using minimum PSE, the final regressor matrix can be as-

sembled using Eq. (12) to transform the selected orthogonal functions into original regressors, and ordinary

least squares can then be applied as described earlier. This procedure is automated using the mof .m code

contained in SIDPAC.12

III. Aircraft Nonlinear Flight Simulations

The eight aircraft considered were the A-7 Corsair II, F-4 Phantom II, F-16C Fighting Falcon, F-16XL,

Delta Dart F-106B, T-2 Generic Transport Model (GTM), DHC-6 Twin Otter, and X-31. The A-7, F-16C,

and F-16XL are fighters, whereas the F-4 is a fighter/bomber and the F-106B is an interceptor. The F-16XL

is an enhanced version of the F-16C with a larger cranked-arrow wing configuration. The X-31 is a research

aircraft used to study highly agile flight. The T-2 is a subscale model of a typical transport-type aircraft,

and the DHC-6 is a turboprop commuter aircraft. All of these aircraft represent conventional tube-and-wing

designs, and have various wing and tail configurations. Most are fighter-like aircraft.

Static and forced-oscillation wind tunnel tests using these aircraft were previously conducted. Aerody-

8 of 23

American Institute of Aeronautics and Astronautics

namic coefficient data were reported and later implemented as nonlinear flight simulations in MATLAB R .21, 12, 22

Table 1 lists the mass and geometry properties used in the simulation for each of these aircraft. The longitu-

dinal position of the aircraft center of mass xcm has been changed in some instances to make the simulations

flyable without feedback control. The simulations include routines for trimming the aircraft, generating

linear models from numerical finite-differences, computing aerodynamic coefficients from the aircraft states

and control surface deflections, and simulating the dynamic response of the aircraft to control inputs.

IV. Results

This section presents the GGA model structure determination, parameter estimation, and validation

using the eight aircraft mentioned in the previous section. The T-2 is highlighted here not only because the

wind tunnel testing was extensive and the simulation model is popular in the literature, but also because

the GGA model accurately predicted the aerodynamics of the T-2, even though it is a different aircraft type

than the majority of the aircraft used in the modeling process.

IV.A. Model Structure Determination

The aerodynamic databases were interrogated to obtain aerodynamic coefficient data over a large range

of conditions. Interrogations were performed separately for the longitudinal and lateral/direction aerody-

namics because some of the wind tunnel databases were already decoupled in this way, and typically the

interactions are small in normal flight regimes. A total of 3168 longitudinal and 9072 lateral/directional

cases were analyzed, using the ranges and resolutions of explanatory variables listed in Table 2. The ranges

selected remain within those used during the wind tunnel testing, and also remain within the normal op-

erating envelope.23 The resolutions selected are similar to what was used in the wind tunnel testing, to

avoid artificially decreasing the uncertainty on the parameter estimation results. To lower the number of

database interrogations, it was also assumed that the aircraft have lateral symmetry, e.g., so that an aileron

deflection on either wing causes the same magnitude of roll moment. During the longitudinal interrogation,

all lateral/directional states and deflections were set to zero, and vice-versa.

For each aircraft, the automated procedure described in Section II was used to determine model structures

for each aerodynamic coefficient in Eq. (1) and then to estimate model parameters and uncertainties for those

models. Explanatory variables for the longitudinal interrogation were α, q̃, and δe , whereas β, p̃, r̃, δa , and

9 of 23

American Institute of Aeronautics and Astronautics

δr were used for the lateral/directional interrogations, where

    
 p̃   b 0 0  p 
  1   
 q̃  = (19)
    
 2V  0 c̄ 0  q 
  
 
    
r̃ 0 0 b r

are the nondimensional body-axis angular rates. Explanatory variables were multiplied in all possible com-

binations up to fourth order, e.g., 1, α, q̃, δe , αδe , α2 q̃δe , α4 , etc., to form the candidate regressor pool.

Mach effects were not considered because the interrogations were restricted to subsonic speeds, and dynamic

pressure, geometry, and mass property effects are removed by using nondimensional aerodynamic coeffi-

cients. Thrust effects were not considered because they contribute only second-order interactions with the

aerodynamics. Any additional control surface deflections, such as flaps and canards, were set to zero for this

analysis.

Figure 1 shows interrogations of the T-2 roll moment coefficient. The data are not shown in any particular

order because the roll moment coefficient depends on many explanatory variables. Figure 2 illustrates

modeling this data using MOF. As the ordered orthogonal functions are retained in the model, the MSFE

drops dramatically and then asymptotes, while the OFP increases linearly. The PSE is minimized when

the first ten ordered orthogonal functions are used; however, only nine ordinary functions resulted from

decomposing the ten selected orthogonal functions. These functions, ordered in decreasing importance to

the model, are

 
X= 1 β δr p̃ δa β3 r̃ δr2 βδa

which produced a model with R2 = 0.9997 and συ = 0.0003, indicating an excellent fit to the data. Residuals

are plotted in Figure 3, which shows the small magnitude of the residuals.
0.02

0
roll moment coefficient, Cl

-0.02

-0.04

-0.06

-0.08
0 2000 4000 6000 8000 10000

database interrogation number

Figure 1. Wind-tunnel database interrogations of the T-2 roll moment coefficient

The aerodynamic coefficients were modeled for each aircraft using MOF. However, the identified model

structures were not the same for all aircraft. Model terms that were chosen in at least half of the analyses,

10 of 23

American Institute of Aeronautics and Astronautics

 8
PSE
MSFE
6 OFP

×10−7
4

0
0 5 10 15 20

ordered orthogonal functions included, n

Figure 2. Model structure determination for the T-2 roll moment coefficient using MOF

0.004

0.002
residual

-0.002

-0.004
0 2000 4000 6000 8000 10000

database interrogation number

Figure 3. MOF model residuals for the T-2 roll moment coefficient

11 of 23

American Institute of Aeronautics and Astronautics

i.e., for at least four of the eight aircraft, were retained to form the GGA model. For instance, Figure 4

displays all the important modeling terms for the roll moment coefficient, as well as the number of instances

in which they were selected. The roll rate and aileron deflection were selected for each of the eight aircraft,

which is expected because these two variables comprise the first-order roll mode approximation.13, 14, 12 The

sideslip angle, yaw rate, and rudder deflection were selected for seven out of the eight aircraft, which is not

surprising because these variables are important in the linearized lateral/directional modes of conventional

aircraft. The remainder of the modeling terms are nonlinear terms that appear in less than half of the

aircraft models. This disparity supports the decision to use at least half as a good criterion for selecting

which model terms are retained in the GGA model.

p̃
δa
β
r̃
δr
βδa
β2
β 2 δa
βδr
β3
δr2
1
δr3
p̃2
p̃3
r̃ 2

0 2 4 6 8
number of instances selected
Figure 4. Model terms selected for the roll moment coefficient

This process was applied to all of the aerodynamic coefficients, resulting in the following model structure

for the GGA model:

CD = θ1 + θ2 α + θ3 αq̃ + θ4 αδe + θ5 α2 + θ6 α2 q̃ + θ7 α2 δe + θ8 α3 + θ9 α3 q̃ + θ10 α4

CY = θ11 β + θ12 p̃ + θ13 r̃ + θ14 δa + θ15 δr

CL = θ16 + θ17 α + θ18 q̃ + θ19 δe + θ20 αq̃ + θ21 α2 + θ22 α3 + θ23 α4

Cl = θ24 β + θ25 p̃ + θ26 r̃ + θ27 δa + θ28 δr

Cm = θ29 + θ30 α + θ31 q̃ + θ32 δe + θ33 αq̃ + θ34 α2 q̃ + θ35 α2 δe + θ36 α3 q̃ + θ37 α3 δe + θ38 α4

Cn = θ39 β + θ40 p̃ + θ41 r̃ + θ42 δa + θ43 δr + θ44 β 2 + θ45 β 3 (20)

12 of 23

American Institute of Aeronautics and Astronautics

which uses 45 model terms. The expected bias terms are present in the longitudinal coefficients. Except

for the drag coefficient, all the linear terms are present in the aerodynamic coefficient models, with the

side force and roll moment coefficients being strictly linear. The drag, lift, and pitch moment coefficients

are the most nonlinear, having terms ranging up to α4 to model stall. The yaw moment coefficient is the

only lateral/directional coefficient having nonlinear terms, where the β 3 term is used to model asymmetric

variations at higher sideslip angles.

It should be pointed out here that the GGA model structure is currently biased towards agile, fighter-like

aircraft because more of those aircraft were used in the analysis. The GGA model structure might be slightly

different if more aircraft of different classifications, such as general aviation or transport, were included in

the analysis. However, it will be shown that the model structure in Eq. (20) still accurately predicts the

behavior of non-fighter aircraft, such as the T-2.

IV.B. Parameter Estimation

Ordinary least-squares parameter estimation described in Section II was used to estimate the model

parameters in the GGA model to best fit the aerodynamic database for each aircraft. Parameter estimates

and standard errors are reported for each aircraft in Tables 3 and 4. Together with the mass and geometry

properties provided in Table 1 and propulsion models, this information is sufficient to build nonlinear flight

simulations for up-and-away flight of each aircraft in this paper.

When comparing the MOF model for a particular aircraft to the GGA model, there are four possible

cases. The first case is that the models contain the same modeling terms, which results in the models being

the same. The second case is that the GGA model contains more terms than the MOF model for a particular

aircraft, which results in an over-parameterized model. In wind tunnel testing, the explanatory variables

are changed independently, and so the extra parameters in the GGA model will have very small values near

zero. The third case is when the GGA model contains fewer terms than the MOF model for a particular

aircraft. In this case, variations in the database cannot be completely captured by the GGA model, which

results in poorer fits to the data and larger error bounds for the GGA model parameters. The fourth case is

when the number of terms in the GGA model and the MOF model for a particular aircraft is the same, but

at least one term is different. While this shifts the modeling dependencies, it does not necessarily degrade

the fit or increase the error on estimated parameters; however, the predictive capability of the model could

deteriorate.

In general, the GGA model fit the aircraft aerodynamic databases well. The GGA model achieved R2

between 0.8676 and 1.0000 for all aircraft and all coefficients, indicating good modeling results. For example,

Figure 5 shows the fit of the GGA model to the database for the T-2 roll moment coefficient. The GGA

13 of 23

American Institute of Aeronautics and Astronautics

model has R2 = 0.9961 and συ = 0.0010, which indicates that the model fit the data well. Compared with

Figure 3, the residual is larger for the GGA model, because of the difference in the model structures. This

highlights one of the fundamental compromises with using reduced-order models: model simplicity is gained

at the expense of model accuracy.

0.004

0.002

residual
0

-0.002

-0.004
0 2000 4000 6000 8000 10000

database interrogation number

Figure 5. GGA model residuals for the T-2 roll moment coefficient

This GGA model can also be used for preliminary design and for testing control laws for a broad range of

aircraft. For instance, Figure 6 shows the parameter estimates and ± two standard deviation error bounds

of the model parameter θ30 , which is the traditional pitch moment stiffness derivative Cmα . All values lie

within the typical range24 −3 rad−1 to +1 rad−1 , and are consistent with results for other aircraft.25, 14 The

fighters have lower values because they are designed with lower static margins than the T-2 and DHC-6,

which are transport and commuter aircraft, respectively. The error bounds on the T-2 and DHC-6 estimates

are higher than the other aircraft because the GGA structure is more representative of a fighter-like aircraft

than a passenger aircraft, even though data-fitting and prediction cases matched well for these aircraft. For

conceptual design of an agile fighter-type aircraft, select a value of θ30 around −0.4; for a transport-type

aircraft, pick a value around −1.6. Control designers can use the uncertainty in the parameters to run

Monte-Carlo simulations for these aircraft. While these pitch moment stiffness terms are almost exactly

equal to those found by linearizing the models numerically, it should be noted that the longitudinal position

of the centers of gravity have been moved in the nonlinear simulations, per Table 1, so that normally unstable

aircraft can be flown without control laws by a pilot using the simulation.

IV.C. Validation

The GGA models were substituted into the nonlinear flight simulations in a quick and straightforward

manner, and dynamic responses were compared with those obtained using the original wind tunnel databases.

The T-2 simulation was trimmed for straight and level flight at h0 = 1200 ft altitude and α0 = 5.0 deg angle

of attack. A modified Newton-Raphson approach was used to determine the remaining state and control

settings to achieve trimmed flight.21, 12 Table 5 shows the results using the original aerodynamic database

14 of 23

American Institute of Aeronautics and Astronautics

 0.5

pitch stiffness, θ30

-0.5

-1

-1.5

-2

-2.5
A-7 F-4 F-16C F-16XL F-106B T-2 DHC-6 X-31

Figure 6. Identified pitch moment stiffness with two standard deviation error bounds

and the fitted GGA model. The velocity was matched exactly, while the throttle changed by 3% of full scale

and the elevator changed by 0.77 degrees. These values are very close, considering the large ranges of the

explanatory variables used in the modeling.

Table 6 shows the modal frequencies and damping ratios for the T-2 about these trim conditions using

both aerodynamic sources, computed using numerical central finite-difference approximations. The same

modes are present and the values reflect approximately the same modal behavior. Local linear models,

handling qualities, and modal parameters are very close between the database and GGA model. The small

differences in the results indicate that the GGA model is an excellent approximation of the wind-tunnel

aerodynamic database.

Time histories of the T-2 state and control variables are shown in Figure 7(a) for nonlinear simulation

using the wind-tunnel database and the GGA model. Starting values for these time histories are indistin-

guishable because the trim conditions were so close. Large-amplitude doublets having a 1 s pulse duration

were sequentially applied to the elevator, aileron, and rudder. Amplitudes were selected to excite nonlinear

motions over a large portion of the flight envelope, to pass through the stall regime, and to approximately

remain within the range of explanatory variables selected for the wind tunnel database interrogation. Fig-

ure 7(b) shows time histories of the aerodynamic coefficients. All traces in Figure 7 are very close and show

that the GGA model accurately characterizes the original wind-tunnel test data. A few traces are slightly

off in magnitude at low angles of attack, but there is no significant phase shift. Off-axis coupling was also

modeled. It should be noted here again that the T-2 is different than most of the other aircraft used to

determine the GGA model and yet predictions were very good. Many other aircraft, such as the F-16C, had

even closer predictions. Note that the original database interrogated for the T-2, consisting of over 150,000

data points (not including additional points for flaps, landing gear, and spoilers), has been reduced to only

45 parameters, representing a data compression factor greater than 3000.

15 of 23

American Institute of Aeronautics and Astronautics

 0.4
input [deg]

10
0.2

CD
0
0
-10
δe δa δr -0.2
200
V [ft/s]

0.2
150

CY
100 0
database GGA
20 -0.2
α [deg]

10 1.5
0 1
-10
CL 0.5
0
10 -0.5
β [deg]

0
0.02
-10
Cl

0
100
p [deg/s]

0 -0.02
-100 0.4
Cm

0
q [deg/s]

50
0
-50 -0.4
0.04
20
r [deg/s]

Cn

0 0
-20
-40 -0.04
0 5 10 15 0 5 10 15

t [s] t [s]
(a) state and control variable time histories (b) aerodynamic coefficient time histories
Figure 7. Large-amplitude dynamic excitation of the T-2

16 of 23

American Institute of Aeronautics and Astronautics

 V. Conclusions

This paper presented a procedure for determining a single generic global aerodynamic model structure

that is applicable to many different aircraft. This was accomplished by interrogating measured wind tunnel

aerodynamic databases for eight aircraft over a large range of aerodynamic angles, body-axis angular rates,

and control surface deflections. This data was then used with multivariate orthogonal function modeling to

determine nonlinear polynomial models for the aerodynamic coefficients. These models are both accurate and

simple. Using the modeling results obtained for the individual aircraft, a generic global aerodynamic (GGA)

model structure was created by selecting the model terms deemed important in at least half of the analyses.

Ordinary least squares was used to identify model parameters that best matched the GGA model structure

to each interrogated database, and the resulting models were substituted into nonlinear simulations of the

aircraft to validate modeling accuracy. Information necessary for building global nonlinear aerodynamic

models for flight simulation of all eight aircraft used in this work are presented in this paper.

A single, fixed aerodynamic model structure was identified for accurately approximating large aerody-

namic databases for eight different aircraft, including fighter, fighter/bomber, highly agile, commuter, and

transport types. It was demonstrated using the T-2 aircraft that by using this method, trim solutions were

accurately computed, local modal behavior was preserved, and large-amplitude state and aerodynamic co-

efficient time histories were accurately predicted. The T-2 represents a worst-case scenario in that it is a

transport-type aircraft and the GGA model was determined using mostly fighter-like aircraft. The original

wind tunnel database for the T-2 was compressed by three orders of magnitude using the GGA model.

Having a GGA model makes it very easy to perform simulations or analyses for different types of aircraft.

Simulations need only to change the values of 45 model parameters instead of large databases with many

aerodynamic data tables. Commercial airline pilots could be trained to recover from stalls or upsets on

a large number of fleet aircraft quickly. Conceptual designers can change a few parameters according to

historical trends, rules of thumb, or first principles to obtain dynamic flight simulations of new aircraft.

Control law designers can change parameters to efficiently check performance for a large range of aircraft.

Engineers can use the GGA model for prior knowledge of the nondimensional aerodynamic topology when

designing wind tunnel tests using modern design of experiments. Having compact functional representations

of the aerodynamics allows for efficient analytical derivation of derivatives for optimization applications.

This work could be extended by incorporating more aircraft, different aircraft configurations, and larger

ranges of the flight envelope into the analysis. Mach, thrust, and other effects of interest can easily be added

as explanatory variables to increase fidelity. Assumptions of longitudinal and lateral/directional decoupling

could be relaxed, at the cost of more computational resources, if that information is present in the databases.

In that case, cross terms, such as αβ, could model variations that might currently be attributed to other

17 of 23

American Institute of Aeronautics and Astronautics

high-order nonlinear functions. Other criteria could be used for selecting the GGA model terms, such as

weighting the number of instances of model terms selected by the aircraft classifications. Incorporating only

a specific class of aircraft in the analysis would provide the fidelity to identify differences between specific

aircraft configurations, such as T-tail versus conventional tail. Incorporating aircraft of all types would

facilitate switching from a transport to a general aviation aircraft, for instance, or describe intermediate

hybrid designs not yet physically realized.

VI. Acknowledgments

This research funded by the NASA Aviation Safety Program, Vehicle Systems Safety Technologies project,

and the Subsonic Fixed-Wing Project.

References
1 Williams, J., Vukelich, J., and Steven, R., “The USAF Stability and Control Digital DATCOM, Volume I User’s Manual,”
Tech. Rep. AFFDL–TR–79–3032, AFFDL, November 1979.
2 Dowell, E., “Eigenmode Analysis in Unsteady Aerodynamics: Reduced Order Models,” No. AIAA-95-1450-CP in Struc-
tures, Structural Dynamics, and Materials Conference, AIAA/ASME/ASCE/AHS/ASC, New Orleans, LA, April 1995, pp.
2545–2557.
3 Xie, D., Xu, M., and Dowell, E., “Proper Orthogonal Decomposition Reduced-Order Model for Nonlinear Aeroelastic
Oscillations,” AIAA Journal, Vol. 52, No. 2, February 2014, pp. 229–241.
4 Lorente, L., Vega, J., and Velazquez, A., “Generation of Aerodynamic Databases Using High-Order Singular Value
Decomposition,” Journal of Aircraft, Vol. 45, No. 5, September-October 2008, pp. 1779–1788.
5 Lorente, L., Vega, J., and Velazquez, A., “Compression of Aerodynamic Databases Using High-Order Singular Value
Decomposition,” Aerospace Science and Technology, Vol. 14, No. 3, May 2010.
6 Castillo-Negrete, D., Hirshman, S., Spong, D., and D’Azevedo, E., “Compression of Magnetohydrodynamic Simulation
Data Using Singular Value Decomposition,” Journal of Computational Physics, Vol. 222, 2007, pp. 265–286.
7 Miled, Z., Huian, L., Bukhres, O., Bem, M., Jones, R., and Oppelt, R., “Data Compression in a Pharmaceutical Drug
Candidate Database,” Informatica, Vol. 27, 2003, pp. 213–223.
8 Kanth, K., Agrawal, D., Abbadi, A., and Singh, A., “Dimensionality Reduction for Similarity Searching in Dynamic
Databases,” Computational Vision Image Underst, Vol. 75, 1999, pp. 59–72.
9 Leng, G., “Compression of Aircraft Aerodynamic Database Using Multivariate Chebyshev Polynomials,” Advances in
Engineering Software, Vol. 28, 1997, pp. 133–141.
10 Morelli, E., “Global Nonlinear Aerodynamic Modeling Using Multivariate Orthogonal Functions,” Journal of Aircraft,
Vol. 32, No. 2, March-April 1995, pp. 270–277.
11 Morelli, E., “Global Nonlinear Parametric Modeling with Application to F-16 Aerodynamics,” No. i–98010–2, American
Controls Conference, Philadelphia, PA, June 1998.
12 Klein, V. and Morelli, E., Aircraft System Identification: Theory and Practice, AIAA Education Series, AIAA, 2006.
13 McRuer, D., Ashkenas, I., and Graham, D., Aircraft Dynamics and Automatic Control, Princeton, 1973.

18 of 23

American Institute of Aeronautics and Astronautics

 14 Stevens, B. and Lewis, F., Aircraft Control and Simulation, Wiley, 2nd ed., 2003.
15 Morelli, E., “Determining the Accuracy of Maximum Likelihood Parameter Estimates with Colored Residuals,” Tech.
Rep. 194893, NASA, Hampton, VA, 1994.
16 Morelli, E. and Klein, V., “Accuracy of Aerodynamic Model Parameters Estimated from Flight Test Data,” Journal of
Guidance, Control, and Dynamics, Vol. 20, No. 1, January–February 1997, pp. 74–80.
17 Morelli, E. and DeLoach, R., “Wind Tunnel Database Development using Modern Experiment Design and Multivariate
Orthogonal Functions,” No. 2003-0653, 41st AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, January 2003.
18 Morelli, E. and Ward, D., “Automated Simulation Updates based on Flight Data,” No. 2007-6714, AIAA Atmospheric
Flight Mechanics Conference, Hilton Head, SC, August 2007.
19 Morelli, E., “Efficient Global Aerodynamic Modeling from Flight Data,” No. 2012-1050, 50th AIAA Aerospace Sciences
Meeting, Nashville, TN, January 2012.
20 Barron, A., Self-Organizing Methods in Modeling, chap. Predicted Squared Error: A Criteron for Automatic Model
Selection, Marcel Dekker, 1984.
21 Garza, F. and Morelli, E., “A Collection of Nonlinear Aircraft Simulations in MATLAB,” Tech. Rep. TM–2003–212145,
NASA, Hampton, VA, January 2003.
22 Murch, A., “A Flight Control System Architecture for the NASA AirSTAR Flight Test Infrastructure,” No. 2008–6990,
AIAA Guidance, Navigation, and Control Conference, Honolulu, HI, August 2008.
23 Wilborn, J. and Foster, J., “Defining Commercial Transport Loss-of-Control: A Quantitative Approach,” No. 2004–4811
in Atmospheric Flight Mechanics conference, AIAA, Providence, RI, August 2004.
24 Roskam, J., Airplane Flight Dynamics and Automatic Flight Controls, Roskam Aviation and Engineering Corporation,
Lawrence, KS, 1979.
25 Raymer, D., Aircraft Design: A Conceptual Approach, AIAA Education Series, AIAA, 3rd ed., 1999.

19 of 23

American Institute of Aeronautics and Astronautics

 Tables

Table 1. Aircraft simulation parameters

Aircraft Description Weight Ixx Iyy Izz Ixz S c̄ b xcm /c̄ xref /c̄
[lbf] [slug·ft2 ] [slug·ft2 ] [slug·ft2 ] [slug·ft2 ] [ft2 ] [ft] [ft] [-] [-]
A-7 fighter 22699 16970 65430 76130 4030 375 10.8 38.7 0.30 0.35
F-4 fighter/bomber 38924 24970 122190 139800 1175 530 16 38.67 0.29 0.29
F-16C fighter 20500 9496 55814 63100 982 300 11.32 30 0.25 0.35
F-16XL fighter 27867 18581 118803 135198 74 663 24.7 32.4 0.10 0.27
F-106B interceptor 29776 18634 177858 191236 5539 698 23.75 38.13 0.25 0.28
T-2 transport 49.6 1.327 4.254 5.454 0.120 5.902 0.915 6.849 0.25 0.25
DHC-6 commuter 10747 20922 24231 38425 1021 420 6.5 65 0.12 0.27
X-31 agility 16000 3553 50645 49367 156 226.3 12.35 22.83 0.30 0.51

Table 2. Ranges and resolution for aerodynamic database interrogation

Case Variable Minimum Maximum Resolution Unit

α −4 +30 2 deg
longitudinal q +0 +50 5 deg/s
δe −20 +10 2 deg
β +0 +20 4 deg
p +0 +100 20 deg/s
lateral/directional r +0 +50 10 deg/s
δa +0 +10 2 deg
δr +0 +30 5 deg

20 of 23

American Institute of Aeronautics and Astronautics

 Table 3. Generic aerodynamic force model parameters and standard errors

A-7 F-4 F-16C F-16XL F-106B T-2 DHC-6 X-31

i θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i )
1 +0.006 ± 0.000 +0.031 ± 0.000 +0.034 ± 0.000 +0.025 ± 0.000 +0.052 ± 0.000 +0.019 ± 0.000 +0.108 ± 0.000 +0.015 ± 0.000
2 +0.320 ± 0.001 +0.280 ± 0.021 −0.005 ± 0.005 −0.085 ± 0.004 −0.202 ± 0.010 −0.078 ± 0.017 +0.138 ± 0.074 −0.157 ± 0.004
3 −0.000 ± 0.505 −11.98 ± 14.99 +20.77 ± 1.817 +2.009 ± 1.503 −9.298 ± 1.022 −27.42 ± 653.7 −54.05 ± 329.8 +1.587 ± 2.062
4 +0.074 ± 0.040 +0.000 ± 0.009 +0.177 ± 0.048 +0.812 ± 0.006 +0.396 ± 0.031 +0.293 ± 0.001 +0.111 ± 0.011 +0.042 ± 0.017
5 −2.519 ± 0.151 −1.818 ± 1.742 +1.285 ± 0.689 −1.046 ± 0.233 −0.858 ± 1.565 +3.420 ± 2.856 +2.988 ± 5.922 +0.697 ± 0.496
6 +0.000 ± 19.59 +209.4 ± 454.6 −19.97 ± 63.87 +34.81 ± 52.33 −13.89 ± 39.58 +288.2 ± 24893 +302.1 ± 9566. −3.684 ± 63.02
7 +0.439 ± 0.188 +0.515 ± 0.046 +0.756 ± 0.178 +1.736 ± 0.024 +0.911 ± 0.117 −0.040 ± 0.003 +0.156 ± 0.052 +0.302 ± 0.087
8 +19.76 ± 1.766 +22.27 ± 18.83 +5.887 ± 8.770 +14.85 ± 2.163 +14.62 ± 20.01 +1.819 ± 36.23 −7.743 ± 62.51 +8.674 ± 6.222
9 −0.000 ± 41.63 −284.7 ± 892.1 +55.59 ± 129.6 −77.87 ± 108.2 +70.04 ± 82.49 −355.3 ± 52284 −218.8 ± 17467 +21.74 ± 120.9
10 −22.10 ± 1.967 −29.81 ± 21.04 −5.155 ± 9.954 −13.89 ± 2.472 −15.41 ± 22.68 −6.563 ± 40.83 +11.77 ± 69.67 −11.19 ± 7.077
11 −1.084 ± 0.003 −0.688 ± 0.000 −1.146 ± 0.000 +0.099 ± 0.000 −0.573 ± 0.000 −1.003 ± 0.000 −0.885 ± 0.000 −0.014 ± 0.019

21 of 23
12 +0.030 ± 0.000 +0.129 ± 0.000 −0.188 ± 0.000 −0.000 ± 0.000 −0.100 ± 0.000 +0.033 ± 0.001 −0.090 ± 0.000 −0.122 ± 0.034
13 +0.059 ± 0.001 +0.670 ± 0.000 +0.876 ± 0.000 −0.000 ± 0.000 +0.500 ± 0.000 +0.952 ± 0.003 +1.697 ± 0.000 +0.710 ± 0.149
14 +0.099 ± 0.001 +0.000 ± 0.000 +0.060 ± 0.000 +0.031 ± 0.000 −0.000 ± 0.000 −0.009 ± 0.000 −0.051 ± 0.000 +0.345 ± 0.038
15 +0.268 ± 0.000 +0.089 ± 0.000 +0.164 ± 0.000 +0.099 ± 0.000 +0.000 ± 0.000 +0.253 ± 0.000 +0.193 ± 0.000 +0.671 ± 0.010
16 −0.093 ± 0.000 +0.105 ± 0.001 +0.074 ± 0.000 −0.081 ± 0.000 −0.017 ± 0.001 +0.016 ± 0.000 +0.215 ± 0.004 −0.020 ± 0.000
17 +4.412 ± 0.002 +1.519 ± 0.043 +4.458 ± 0.004 +2.254 ± 0.013 +1.888 ± 0.047 +5.343 ± 0.054 +4.370 ± 0.129 +3.023 ± 0.019

American Institute of Aeronautics and Astronautics

18 −0.000 ± 0.001 +6.727 ± 0.451 +29.90 ± 0.193 +4.545 ± 0.010 −9.226 ± 0.402 +30.78 ± 10.06 +25.05 ± 29.09 +3.697 ± 0.033
19 +0.549 ± 0.000 +0.265 ± 0.000 +0.412 ± 0.000 +0.648 ± 0.000 +0.774 ± 0.001 +0.396 ± 0.000 +0.291 ± 0.000 +0.237 ± 0.001
20 +0.000 ± 0.024 +33.25 ± 7.323 −5.538 ± 3.206 −8.237 ± 0.159 +14.52 ± 6.694 +12.03 ± 165.5 +52.78 ± 271.3 +6.616 ± 0.542
21 +0.817 ± 0.277 +9.900 ± 6.887 −2.477 ± 0.476 −2.203 ± 2.363 −9.438 ± 6.519 +0.506 ± 8.930 +16.62 ± 13.72 −3.330 ± 3.204
22 −9.833 ± 3.643 −12.71 ± 96.96 −1.101 ± 6.625 +19.03 ± 33.08 +48.71 ± 92.40 −36.30 ± 122.4 −87.67 ± 178.4 +11.60 ± 44.53
23 +3.851 ± 4.336 −12.91 ± 114.8 +1.906 ± 7.816 −25.83 ± 39.45 −53.62 ± 109.1 +46.13 ± 146.1 +90.41 ± 212.0 −16.94 ± 53.07
 Table 4. Generic aerodynamic moment model parameters and standard errors

A-7 F-4 F-16C F-16XL F-106B T-2 DHC-6 X-31

i θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i ) θ̂i ± σ(θ̂i )
24 −0.054 ± 0.000 −0.034 ± 0.000 −0.071 ± 0.000 −0.005 ± 0.000 −0.000 ± 0.000 −0.109 ± 0.000 −0.112 ± 0.000 −0.002 ± 0.000
25 −0.313 ± 0.000 −0.236 ± 0.000 −0.445 ± 0.000 −0.202 ± 0.000 −0.300 ± 0.000 −0.366 ± 0.000 −0.413 ± 0.000 −0.395 ± 0.001
26 +0.031 ± 0.000 +0.025 ± 0.000 +0.058 ± 0.000 +0.058 ± 0.000 −0.000 ± 0.000 +0.061 ± 0.000 +0.191 ± 0.000 −0.021 ± 0.003
27 −0.137 ± 0.000 −0.035 ± 0.000 −0.143 ± 0.000 −0.093 ± 0.000 −0.147 ± 0.000 −0.079 ± 0.000 −0.206 ± 0.000 −0.048 ± 0.001
28 +0.004 ± 0.000 +0.013 ± 0.000 +0.023 ± 0.000 +0.015 ± 0.000 +0.009 ± 0.000 +0.021 ± 0.000 +0.116 ± 0.000 +0.101 ± 0.000
29 −0.023 ± 0.000 −0.013 ± 0.000 −0.024 ± 0.000 +0.009 ± 0.000 +0.020 ± 0.000 +0.182 ± 0.000 +0.057 ± 0.001 +0.028 ± 0.000
30 −0.810 ± 0.000 −0.254 ± 0.000 −0.288 ± 0.000 +0.088 ± 0.000 −0.267 ± 0.000 −1.782 ± 0.014 −1.419 ± 0.023 −0.870 ± 0.000
31 −7.033 ± 0.004 −2.916 ± 0.000 −8.267 ± 0.006 −0.272 ± 0.000 −0.497 ± 0.000 −44.34 ± 28.76 −27.95 ± 2.063 −2.352 ± 0.001
32 −1.032 ± 0.000 −0.403 ± 0.000 −0.563 ± 0.000 −0.174 ± 0.000 −0.378 ± 0.000 −1.785 ± 0.001 −1.626 ± 0.002 −0.166 ± 0.001
33 +0.502 ± 1.390 −3.955 ± 0.142 −5.513 ± 2.695 −4.315 ± 0.100 +0.313 ± 0.096 +374.0 ± 27268 +100.7 ± 854.0 −0.488 ± 0.168
34 +8.007 ± 50.13 −24.00 ± 5.442 +9.793 ± 101.5 +15.47 ± 3.693 −11.72 ± 3.649 −1748. ± 10962 −759.2 ± 31695 −9.691 ± 5.138

22 of 23
35 +1.215 ± 0.505 −0.270 ± 0.099 −1.057 ± 0.139 −0.365 ± 0.006 −0.644 ± 0.056 +2.439 ± 1.137 +7.664 ± 1.316 −0.064 ± 0.319
36 +17.15 ± 111.4 +55.32 ± 11.23 −2.018 ± 216.5 −18.25 ± 8.056 +19.60 ± 7.607 +1949. ± 22138 +1103. ± 66523 +1.262 ± 12.12
37 −1.278 ± 2.032 +1.479 ± 0.410 +1.897 ± 0.489 +0.848 ± 0.019 +1.443 ± 0.183 −0.038 ± 4.093 −8.121 ± 4.505 +0.361 ± 0.937
38 −1.969 ± 0.018 −0.448 ± 0.003 −0.094 ± 0.026 +0.581 ± 0.004 −0.048 ± 0.003 +0.803 ± 1.232 +2.468 ± 2.086 +1.795 ± 0.003
39 +0.102 ± 0.000 +0.142 ± 0.000 +0.234 ± 0.000 +0.102 ± 0.000 +0.152 ± 0.000 +0.183 ± 0.000 +0.088 ± 0.000 +0.406 ± 0.021
40 +0.060 ± 0.000 −0.006 ± 0.000 +0.056 ± 0.000 −0.007 ± 0.000 +0.002 ± 0.000 −0.022 ± 0.000 −0.043 ± 0.000 +0.205 ± 0.001

American Institute of Aeronautics and Astronautics

41 −0.294 ± 0.000 −0.358 ± 0.000 −0.418 ± 0.000 −0.282 ± 0.000 −0.308 ± 0.000 −0.405 ± 0.000 −0.426 ± 0.000 −0.875 ± 0.003
42 −0.020 ± 0.000 +0.001 ± 0.000 −0.034 ± 0.000 −0.014 ± 0.000 −0.090 ± 0.000 −0.009 ± 0.000 +0.023 ± 0.000 −0.128 ± 0.001
43 −0.121 ± 0.000 −0.053 ± 0.000 −0.085 ± 0.000 −0.046 ± 0.000 −0.044 ± 0.000 −0.129 ± 0.000 −0.087 ± 0.000 −0.223 ± 0.000
44 +0.557 ± 0.000 −0.000 ± 0.000 +0.372 ± 0.014 +0.086 ± 0.000 −0.000 ± 0.000 +0.184 ± 0.011 +0.337 ± 0.009 −2.575 ± 1.251
45 −0.923 ± 0.001 +0.337 ± 0.000 −0.725 ± 0.052 −0.140 ± 0.001 +0.000 ± 0.000 −0.377 ± 0.037 −0.766 ± 0.027 +4.329 ± 4.580
 Table 5. Comparison of T-2 trim settings (h0 = 1200 ft, α0 = 5.0 deg)

Variable Database GGA Unit

V0 125 125 ft/s
δt 0.15 0.12 –
δe 1.00 1.77 deg

Table 6. Comparison of T-2 modal parameters

Mode Frequency [rad/s] Damping Ratio

Database GGA Database GGA
spiral 0.05 0.09 – –
phugoid 0.32 0.34 0.05 0.05
roll subsidence 5.28 5.44 – –
dutch roll 5.89 5.27 0.15 0.18
short period 6.64 6.49 0.46 0.35

23 of 23

American Institute of Aeronautics and Astronautics

View publication stats