JOURNAL OF AIRCRAFT

Vol. 51, No. 2, March–April 2014

Framework for Flight Loads Analysis of Trajectory-Based

Manoeuvres with Pilot Models

Mudassir Lone,∗ Chi Kin Lai,† Alastair Cooke,‡ and James Whidborne‡
Cranfield University, Cranfield, England MK43 0AL, United Kingdom
DOI: 10.2514/1.C032376
A framework is presented to allow the investigation of interactions between manual control and flight control
systems for large flexible aircraft. It allows the fast simulation of particular flight test manoeuvres designed to expose
unwanted interactions between the pilot and aircraft flight dynamics. It combines a trajectory generation scheme and
an aeroelastic aircraft model together with pilot models to conduct flight loads analysis relevant to trajectory-based
flight test manoeuvres. The ability of the framework to assess the impact of variations in pilot control characteristics
on flight loads is illustrated through the simulation of two manoeuvres. The first required the pilot to align the aircraft
from an offset position in minimum time and the second required avoiding an obstacle by following a four-
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

dimensional trajectory. Pilot models representative of pilot control behavior were derived by applying a parameter
identification scheme to data gathered through a series of manual control experiments. The pilot models were coupled
with a nonlinear flexible aircraft model to demonstrate the effects on pilot–vehicle system frequency response due to
changes in task and intersubject differences. Worst case wing loads were then obtained and compared to aircraft loads
envelopes.

Nomenclature Wx = route deviation weighting matrix

CD , CL = drag and lift coefficients, respectively x = state vector
CD0 = zero-lift drag coefficient xrou = state vector of intended route
D = aerodynamic drag, N Yp = compensatory equalization
f = frequency, rad∕s Yf = feed-forward equalization
fD = forcing function for training tasks α = angle of attack, deg
Gnm = neuromuscular transfer function β = angle of sideslip, deg
g = gravitational acceleration, m · s−2 γa = air mass flight path angle, deg
K = induced drag factor δe = experimental stick deflection
Ke = pilot model error compensation δm = pilot model stick deflection
Ke_ = pilot model error rate compensation θ = pitch attitude, deg
Kf = pilot model feed-forward compensation μ = bank angle, deg
L = aerodynamic lift, N ζn = neuromuscular damping
m = aircraft mass, kg ρ = air density, kg · m−3
nz = normal load factor, g τL
 = pilot model lag
p = position vector, m τI
 = pilot model lead
pA , pB = position vectors defining the intended route, m τn = neuromuscular lag
pobst = obstacle position vector, m τ = pilot model delay, s
pN , pE , pD = north, east, and down positions, respectively Φe = frequency spectrum experimental stick deflection
q = pitch rate, deg ∕s Φm = frequency spectrum pilot model stick deflection
Robst , Rcirc = radius of spherical obstacle and circular tunnel, χa = air-mass heading, deg
respectively, m ωn = neuromuscular natural frequency
S = aircraft reference wing area, m2
T = engine thrust, N
T max = maximum available thrust, N I. Introduction
T
= throttle setting
T
Va
=
=
target set
true airspeed, m∕s
A IRCRAFT manufacturers must prove that flight loads
undergone by an airframe remain within certain limits to
satisfy airworthiness requirements [1]. These cover loads during
v = velocity vector, m∕s controlled manoeuvres executed by the pilot/autopilot, when the
Wt = manoeuvre time weighting aircraft is subjected to atmospheric disturbances and under certain
clearly defined failure conditions. Traditionally, these areas have
Presented as Paper 2012-4854 at the AIAA Modeling and Simulation
been analyzed separately with assumed pilot control inputs where
Technologies Conference, Minneapolis, MN, 13–16 August 2012; received 9 necessary [2,3]. However, a number of recent incidents [4] and
April 2013; revision received 13 August 2013; accepted for publication 15 accidents have highlighted a growing need to predict airframe loads
August 2013; published online 26 February 2014. Copyright © 2013 by the arising from unwanted interactions between 1) manual control,
American Institute of Aeronautics and Astronautics, Inc. All rights reserved. 2) sophisticated nonlinear flight control systems (FCS), and
Copies of this paper may be made for personal or internal use, on condition 3) atmospheric disturbances for large flexible airframes. Along with
that the copier pay the $10.00 per-copy fee to the Copyright Clearance Center, progress in human factors and pilot training, this has also led to the
Inc., 222 Rosewood Drive, Danvers, MA 01923; include the code 1542-3868/ development of more targeted flight test techniques [5]. However,
14 and $10.00 in correspondence with the CCC. the risks and cost associated with flight testing have also resulted in
*Research Fellow/Lecturer, Dynamics, Simulation and Control Group,
Department of Aerospace Engineering (Corresponding Author). the need to develop cost-effective fast simulation tools [6,7] where
†
Ph.D. Student, Dynamics, Simulation and Control Group, Department of flight incidents or critical flight test manoeuvres can be reproduced.
Aerospace Engineering. The effects of unwanted interactions, specifically on airframe
‡ loads, have not yet been studied through pilot-model-in-the-loop
Senior Lecturer, Dynamics, Simulation and Control Group, Department of
Aerospace Engineering. simulation. Such a simulation environment requires models of the
637
 638 LONE ET AL.

flight dynamics exhibited by flexible aircraft (including the FCS), A. Aircraft Dynamics Model
pilot control characteristics, and atmospheric disturbances. These The trajectory generation process employs a three-degree-of-
aspects are usually studied separately through highly complex loads freedom (3-DOF) point-mass model to evaluate the nominal
models and piloted flight simulation trials [3,8]. Modelling quantities related to aircraft performances [14]. For tactical
techniques for fast simulation of the flexible aircraft have matured manoeuvres undertaken over a short period, it can be assumed that:
considerably [7,9] and suitable models are now available for research 1) Fuel expenditure is negligible, and so the aircraft mass remains
[7,10]. On the other hand, pilot models have seen extensive use only constant.
in handling qualities studies [11] and the terms pilot modelling, 2) Atmospheric disturbances are handled by the inner guidance
aircraft–pilot coupling (APC), and/or pilot-induced oscillations are loop and, therefore, do not require explicit consideration. Therefore,
rarely seen in isolation from each other. Moreover, simulation of the relative-to-the-air-mass velocity is equal to the inertial velocity.
flight test manoeuvres used to evaluate the tendency of the pilot– 3) Only coordinated flight is considered, and so the side-slip angle
vehicle system (PVS) to unwanted interactions, such as those used is assumed to be zero.
by Lee [5] and suggested by Mitchell and Klyde [12] to expose 4) Angle of attack is assumed to be small, and so the nominal thrust
adverse APC, require the pilot to execute a predefined trajectory. is assumed to be along with the relative-to-the-airmass velocity.
Therefore, a complete simulation environment also requires a The equations of motion for a generic commercial aircraft over a
trajectory generation scheme capable of generating three- or four- flat Earth can be written as [15]
dimensional trajectories representing tasks that may vary from
precision offset landings to low altitude lateral S-turns. p_ N  V a cos γ a cos χ a (1a)
This paper presents a simulation framework that combines a
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

trajectory generation scheme [13] and an aeroelastic aircraft model

[7] together with pilot models to conduct flight loads analysis p_ E  V a cos γ a sin χ a (1b)
relevant to trajectory-based flight test manoeuvres. Figure 1 shows an
analytical scheme within which such a simulation framework can be
deployed. The following are its main aspects:
1) Task scenario assignment is effectively the stage where the flight p_ D  −V a sin γ a (1c)
test manoeuvre is specified by supplying a series of objectives and
constraints to the trajectory generator. The pilot’s objectives and his/
her role in the overall task is also considered here. T − D − mg sin γ a
V_ a  (1d)
2) A Monte Carlo simulation environment is needed due to m
the stochastic nature of the scenario itself because it can involve
variations in pilot behavior and atmospheric disturbances. Flight test
or flight simulator based trials only allow data collection from a L sin μ
χ_ a  (1e)
certain number of scenarios. This approach allows a wide range mV a cos γ a
of combinations within the PVS to be tested to provide a statistical
flight load.
3) Airframe loads analysis is where loads data generated from the L cos μ − mg cos γ a
simulations are evaluated, post processed, and compared with flight γ_ a  (1f)
mV a
loads envelopes derived using traditional methods [1].
In this paper, the required components are integrated together to where pN , pE , and pD are the center of mass’s coordinates in the flat-
illustrate this analytical scheme through the simulation of two tasks. Earth North-East-Down axes system, V a is the true airspeed (equal to
The paper is structured such that Secs. II and III describe the the inertial speed when there is no wind), γ a is the air-mass flight-path
trajectory generation scheme and manual control experiments, angle, χ a is the air-mass heading angle (equal to the aircraft heading
respectively. The derivation of the pilot model and details of the when there is no side-slip), μ is the bank angle (equal to roll angle
parameter identification process are given in Sec. IV. Modelling of when angle of attack and side-slip are both zero), m is the mass, g is
intersubject differences in a robust control context is briefly studied in the acceleration due to gravity, and T, L, and D are the thrust, lift, and
Sec. V before details of pilot-model-in-the-loop simulations are
drag respectively.
presented in Sec. VI. Finally, conclusions from this study and the
The aerodynamics and propulsion forces are aircraft type
focus of future work are discussed in Sec. VII.
dependent; they can be evaluated with the following equations:

1 1

max ;
T  TT L  nz mg  ρV 2a SCL ; D  ρV 2a SCD
II. Reference Trajectory Generation 2 2
As shown in Fig. 1, tactical trajectories satisfying the given (2)
objectives and constraints are generated to serve as reference to the
subsequent systems. This is done by solving an optimization problem where T
∈ 0; 1 is the nominal throttle setting, T max is the maximum
based on the models of aircraft performance and operational available thrust, ρ is the air density, S is the aircraft wing area, nz is the
environment. The formulation of the optimization problem along normal load factor, and CL and CD are the lift and drag coefficients,
with the models used are described in this section. respectively.

Fig. 1 Framework for flight load analysis with pilot model.

 LONE ET AL. 639

In particular, the maximum available thrust and the drag coefficient obstacle must be greater than the sum of the spherical obstacle’s
are known functions of aircraft specific parameters. The functions radius Robst and the circular tunnel’s radius Rcirc :
and the specific parameters for most of the operating aircraft types
can be found in the Base of Aircraft Data [16]. For instance, the drag kp − pobst k ≥ Robst  Rcirc (8)
coefficient is related to the lift coefficient by the parabolic drag polar:
where pobst is the obstacle position and p represents the position
CD  CD0  KC2L (3) vector obtained from the state vector; that is,

p
x   pN pE pD T (9)
where CD0 is the zero-lift drag coefficient and K is the induced drag
factor.
The state-space formulation of the preceding point-mass model is 3. Boundary Conditions
obtained by substituting Eqs. (2) and (3) into Eq. (1): At the initial time t0 when the trajectory generator is engaged, the
current values of the state variables are used as initial values:
x_  f
x; u; x   pN pE pD Va χa γ a T ;
u   T
CL μ T (4) x
t0   x0   pN0 pE0 pD0 V a;0 γ a;0 χ a;0 T (10)

where x is the state variable vector, u is the control variable vector, The final time tf for the tactical manoeuvre is left open.
and f
x; u is the resulting nonlinear vector-value function of the The final state xf is constrained to ensure that the aircraft is:
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

aircraft system. 1) back onto the intended trajectory, assumed here to be a straight
line defined by points A and B, with level and straight flight; that is,
B. Constraint Models
∃λ ∈ 0; 1; p
tf   pA  λ
pB − pA ; and
1. Manoeuvre Limitations
To assure that the required aircraft performance is achievable, the χ a
tf   χ AB ; γ a
tf   γ AB (11a)
control variables are constrained to their corresponding upper and
lower bounds:
μ
tf   0; γ_ a
tf   0; χ_ a
tf   0 (11b)
T
min ≤ T
≤ T
max ; CLmin ≤ CL ≤ CLmax ; μmin ≤ μ ≤ μmax
2) separated from the obstacle with a safe distance (dsafe ):
(5)
kp
tf  − pobst k ≥ dsafe (12)
To take into account the flight envelope and the climbing
performance, the state variables are bounded as follows: 3) not moving toward the obstacle:

− hmax ≤ pD ≤ 0; V a;min ≤ V a ≤ V a;max ; Δp  p
tf  − pobst ; Δv  v
tf ; Δp · Δv ≥ 0 (13)

γ a;min ≤ γ a ≤ γ a;max (6)
Here, v is the velocity vector, obtained from V, γ, and χ; and Δp and
Δv are the relative position of and velocity to the obstacle,
Considering the thrust build-up/decay times, the rolling performance
respectively.
of a flight control system, and the structural limitation on load factor,
the rates of control variables are also constrained to their
corresponding upper and lower bounds: C. Optimal Control Problem
1. Problem Formulation
T_
min ≤ T_
≤ T_
max ; μ_ min ≤ μ_ ≤ μ_ max ; nz ≤ nzmax (7) In the overall avoidance trajectory, a cost function consisting of
two quantities is used to penalize for manoeuvre time and route
deviation:
These limitation values are aircraft specific and can be varying with Zt
the flight conditions. Table 1 shows an example of these values in the f

specified flight condition. Jx
t; u
t  fW t  x
t − xrou
tT W x x
t − xrou
tg dt
t0
(14)
2. Obstacle Avoidance
A spherical no-fly zone is used to model a tactical obstacle. The where W t ∈ R is the weighting on the manoeuvre time, W x ∈ is R6×6
radius of the sphere completely encloses the obstacle. A circular a positive semidefinite, symmetric weight matrix, and xrou is the
tunnel surrounding the nominal flight path trajectory is used to intended trajectory segment defined by pA and pB . A suitable
introduce robustness against uncertainties in trajectory tracking avoidance trajectory can, therefore, be generated by solving the
performance. As a result, the distance between the aircraft and the following optimisation problem:

Table 1 Parameter values for the aircraft model and trajectory generation process
Aircraft model m  290; 000 kg S  511 m2 CD0  0.02
K  0.073
Aircraft performance T max  800 kN T
min  0.25 T
max  1
CLmin  0.064 CLmax  1.085 nzmax  2.5g
μmax  −μmin  45° hmax  13; 000 m V a;min  130 m∕s
V a;max  240 m∕s γ a;min  −20° γ a;max  25°
T_
max  −T_
min  0.25 μ_ max  −_μmin  15° ∕s
Flight conditions M  0.6 Altitude  12; 000 ft ρ  0.8491 kg∕m3
Trajectory criteria Robst  300 m Rcirc  200 m dsafe  500 m
pN 0  pE0  0 m pD0  −3660 m V a;0  195 m∕s
γ a;0  0 deg χ a;0  0 deg
 640 LONE ET AL.

Problem 1: Find the admissible control trajectories u and state flight path trajectory, it is evident that the aircraft’s tunnel brushes
trajectories x to drive the aircraft system f from the initial state x0 to past the obstacle to minimize the route deviation. Task B, shown in
the terminal target set T while minimizing the performance index J: Fig. 3, is a reference trajectory without any obstacle. The weights
8
were set to penalize the manoeuvre time only. As can be seen from the
>
> minu
t;x
t Jx
t;u
t
state and control trajectories, the aircraft attempts to achieve a full
>
>
>
> throttle and maximum bank manoeuvre in order to minimize the
>
>
>
> _  f
x
t;u
t ∀ t ∈ t0 ;tf 
subject to : x
t manoeuvre time. This task is similar to that used in flight tests aiming
>
>
>
> to excite high gain pilot behavior and, thus, expose potential APC or
< x
t ∈ X X  fx ∈ R6 jEqs:
6 and
8g
P1 ≜ pilot-induced oscillation tendencies [5].
>
> u
t ∈ U U  fu ∈ R3 jEqs:
5 and
7g
>
>
>
>
>
> x
t0   x0
>
>
>
>
>
III. Experimental Setup for Manual Control Tests
>
: x
tf  ∈ T T  fx ∈ R6 jEqs:
11;
12 and
13g The block diagram for all manual control tests and experimental
setup are shown in Figs. 4 and 5, respectively. Preliminary tests were
conducted with a mixture of 14 young pilots and engineers with an
average logged flight experience of 10 h. The subjects were presented
2. Solution Approach with a sum-of-sines compensatory task (two minutes in duration) and
root-mean-square (rms) error was taken as a performance indicator. It
Numerous numerical methods have been developed to solve the
was found that the subjects stabilized their performance between five
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

general nonlinear optimal control problems [17–19]. In particular,

to seven runs [27]. However, the tasks were found to require a
direct methods that convert nonlinear optimal control problems to considerable amount of concentration. Therefore, fatigue was a major
nonlinear programming problems via parameterization and factor when deciding the number of training and experimental runs
discretization are found to be promising in many applications allowed per session for future tests.
requiring real-time results. The multiple shooting method [20], The final series of experiments involved a test pilot and two
pseudospectral method [21], and inverse dynamics method [22,23] experienced commercial aircraft pilots, each with over 15,000 h of
are some examples. To deal with the complex boundary conditions flight time. The pilots performed the tasks five times after completing
for the posed problem and to obtain a smooth solution, the param- five compensatory training tasks. Past trials have shown that subjects
eterization and discretization scheme described in an inverse stabilize their performance within the first five attempts [27] before
dynamics method is implemented here [23,24]. The converted non- succumbing to either fatigue or boredom. Fatigue effects were
linear programming problems are then solved with the default avoided by allowing a short break after each task.
differential evolution algorithm given by Price et al. [25]. From the subject’s perspective, the task is to ensure the flight
In this study, the set of static parameters corresponding to a large director cross-hairs maintain position above the aircraft wings level
four-engined transport aircraft [26] in steady level flight were used. symbol. Here the cross-hairs represent the difference between desired
The aircraft characteristic and the flight condition used are given in and actual attitudes, implying that the task ensures the pilot
Table 1. minimizes these errors and consequently provides the attitudes
The resulting trajectories were used to drive a dual mode flight required for trajectory tracking. The subject acts as a dual-axis
director that subjects followed during the manual control controller tracking demands in pitch and roll by focusing on only two
experiments. Figure 2 shows task A, which is a reference avoidance variables and corresponding first derivatives. Furthermore, the
trajectory with an obstacle appearing slightly left of the intended experimental setup ensures no other visual cues are presented,
route. The weights were set to penalize state deviation only. From the minimizing the number of control variables and the level of attention

a) Flight path trajectory b) State and control trajectory

Fig. 2 Task A: avoidance trajectory with pobst
5850; − 100; − 3660T , W t
0, W x
I6×6 , pA
0;0; − 3660 and pB
12000;0; − 3660.
 LONE ET AL. 641
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

a) Flight path trajectory b) State and control trajectory

Fig. 3 Task B: minimum-time trajectory with no obstacle, W t
1, W x
0, pA
0;500; − 3560 and pB
5000;500; − 3560.

Fig. 4 Block diagram of manual control experiments and simulations.

allocation. Together with an almost equal number of subjects and task modern large transport aircraft. The FCS gains were selected such
repetitions, this also limits the use of higher brain functions required that aircraft response lay well within the category 1 C-star boundary.
to develop strategies and identify patterns in the given tasks. Stick command gearing was used to ensure a full aft, and forward
stick deflections led to 2.5 and −1g normal load factors. A full
A. Simulation Model and Hardware lateral stick deflection commanded a 15 deg ∕s roll rate. The lateral
The experimental setup used for both tracking and compensatory CSAS design ensures the the aircraft’s lateral dynamics meet the
tasks is shown in Fig. 5. A linear time invariant model representative specifications given in MIL-F-8785C.
of a large four-engine transport aircraft in the climb/approach (Mach Aircraft attitudes were presented to the pilot via a 110 × 115 mm
0.6, 12,000 ft) flight phase was used [26]. This is a linearized version display also shown in Fig. 4. The attitude indicator scales were spaced
of the model described in Sec. VI. All simulations began with the such that a 5 deg pitch attitude equalled a 10 mm separation. All tests
aircraft in steady level flight. were conducted on a desktop computer (IBM T-Series laptop with
A C-star command and stability augmentation system (CSAS) 2.7 GHz and 4 Gb RAM) in a MATLAB/Simulink® environment
was designed to present the subjects with dynamics representative of with a nominal computational time delay of 13 ms. Subjects
 642 LONE ET AL.

Fig. 5 Setup for desktop based manual control experiments.

performed tasks by manipulating a Thrustmaster HOTAS® joystick. identification process was done off-line meant that computational
The inactive nature of such an inceptor allows the relationship cost was not an issue. The approach uses nonlinear constrained
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

between pilot command and stick deflection to be kept relatively optimization to find model parameter values. The following cost
simple. function was used:
ZT ZF
B. Experimental Tasks
J
t; f 
δe
t − δm
t2 dt  Λ
Φe
f − Φm
f2 df
The subjects were first trained via five two-minute compensatory 0 0
tasks. When performing these tasks, the flight director was switched (16)
off and the subjects tried to align the aircraft attitude indicator with
the horizon line. The sum-of-sines disturbance signal injected as where Λ is a factor necessary to ensure that equal weighting is given
shown in Fig. 4 is of the following form: for frequency and time domain data during the optimization process.

X
15
A. Compensatory Component
fD
t  Ak sin
ωk t  ϕk  (15)
k1
The bimodal model uses first-order equalizations along both
attitude error and error rate feedback paths:
Values for Ak, ωk , and ϕk given by Zaal et al. [28] were used in this

study. The phase ϕk for each sinusoid was randomized such that τ s1 τLq s  1 −τs
Y p
s  Ke Lθ  K e_ s e (17)
subjects could not perceive any internal coherence and consequently τIθ s  1 τIq s  1
deploy high-level brain functions.
Training was followed by two tasks where subjects were asked to The state equation without the delay component can be written in
follow the flight director cross-hairs. Task A simulated an obstacle matrix form as follows:
avoidance manoeuvre whereas task B simulated the execution of a






minimum-time trajectory as described earlier in Sec. II. x_ e Ae 0 xe Be 0 e
  (18)
x_ e_ 0 Ae_ xe_ 0 Be_ e_

IV. Pilot Model Derivation and Identification Any desired control action is finally implemented through the
Pilot models can provide considerable insight into intersubject neuromuscular system, modelled as
variations in manual control characteristics. Differences in parameter
values can highlight changes in subjective factors such as relative ω2n
Gnm
s  (19)
aggression and varying degrees in a subject’s understanding of
τn s  1
s2  2ζ n ωn s  ω2n 
vehicle dynamics. For compensatory tasks, these form the complete
pilot model, whereas for pursuit tasks, a feedforward path is added to The state equation for the preceding model can be written as
the structure, as shown in Fig. 6. No threshold has been included in
the feedforward loop because it is assumed the pilot has exact x_ n  An xn  Bn δ^ s (20)
knowledge of the flight director demands displayed on the primary
flight display (PFD). Reviews by Lone and Cooke [29] and Grant and where δ^ s is the sum of the compensatory and tracking components, as
Schroeder [30] provide more detailed descriptions and comparisons
shown in Fig. 6.
of the pilot models used here together with methods of application.
Pilot model parameters can be identified either in the frequency or
time domains. Zaal et al. [31] provide a brief discussion of the B. Tracking Component
advantages and disadvantages of methods in these domains. The The tracking model assumes the subject implements a feedforward
approach adopted here attempts to obtain pilot model parameters that loop with lead-lag compensation along with the compensatory
provide matches in both domains. The fact that the parameter dynamics discussed in Sec. IV.A. This feedforward equalization is
modelled as the following transfer function:
  
τLf s  1 τLLf s  1 −τs
Y f
s  K f e (21)
τIf s  1 τILf s  1

Similar to the compensatory component, the state equation without

the delay can be written as follows:

Fig. 6 Pilot model structure for compensatory and tracking tasks. x_ f  Af xf  Bf r (22)
 LONE ET AL. 643

It should be noted that this uses the same time delay as the commenting on any significant variations observed in pilot model
compensatory loop. Given the multi-axis nature of the tasks, the parameters. Here, the variance of the remnant signal is used as a
complete pilot model consists of two sets of compensatory and metric to not only measure the subject’s adherence to linear control
tracking components: one for pitch tracking and another for roll. As a behavior, but also as an indicator of the quality of the obtained match.
result, the pilot is modelled as a dual channel controller. As shown in Fig. 10a, for both tasks, this variance is an order of
magnitude lower than the typical range of stick deflection,
C. Results demonstrating a high level of agreement. It should be noted that such
Example results from the parameter identification process are levels of agreement are obtained primarily due to the experimental
shown in Figs. 7 and 8. These effectively summarize the level of setup and tasks, due to which the subjects are restricted to adopt only a
agreement between the model and experimental data in both time and limited number of feedback loops.
frequency domains for a single test. The remnant characteristics are Relative to task B, pilots can be seen to adopt similar gains and
also shown in the form of frequency content and distribution. equalizations in task A. This leads to little variation in rms error.
Results for pitch control are shown in Figs. 9a–11b. Data points Furthermore, comparing the values obtained for the gains, it can be
occurring two standard deviations beyond the first and third quartiles said that subjects heavily rely on rate feedback to accomplish the task
are defined as outliers. The figures show variations for a total of 30 at hand. The same comment can be made for the roll control pilot
tests, excluding training. However, the level of agreement between model. The resulting values for the pilot model lead parameter τI also
theoretical and experimental data must be discussed prior to show that subjects introduced nonminimum phase dynamics.
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

0.6 Model
Experimental

0.4

0.2

−0.2

−0.4

0 10 20 30 40 50 60 70
Time (s)
−3
1.2 x 10
0.045 Model Remnant Remnant
Experimental
0.04 1000
1
0.035
Power Density
Power Density

0.8 800
0.03
Count

0.025 0.6 600

0.02
0.015 0.4 400
0.01
0.2 200
0.005
0 0
−1 0 1 −2 −1 0 1 2
10 10 10 10 10 10 10 10 −0.3 −0.2 −0.1 0 0.1 0.2
Frequency (rad/s) Frequency (rad/s)
Fig. 7 Tracking model comparison with subject A longitudinal stick deflection (task A).

Model
Experimental
0.4

0.2

−0.2

−0.4

−0.6
0 10 20 30 40 50 60 70
Time (s)
−3
x 10
0.025 3
Remnant Remnant
Model
Experimental 2.5 1200
0.02
Power Density

1000
Power Density

2
0.015 800
Count

1.5
0.01 600
1
400
0.005
0.5 200

−1 0 1
0 −2 −1 0 1 2
0
10 10 10 10 10 10 10 10 −0.3 −0.2 −0.1 0 0.1 0.2 0.3
Frequency (rad/s) Frequency (rad/s)
Fig. 8 Tracking model comparison with subject A lateral stick deflection (task A).
 644 LONE ET AL.

0.05 60
maximum
non-outlier

0 15
outlier 40
3rd Quartile

−0.05 median
10 20

Lθ
e

τ
K

K
−0.1
0
5
−0.15 1st Quartile
−20 minimum
non-outlier
−0.2 0
A B A B A B

8 35 50
30
40
6 25
20 30
4
τLq

Iθ

τIq
15

τ
10 20
2
5
10
0
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

0
−5 0
A B A B A B
a) Roll

0.2 2.5 20

2 15
0.1
1.5
10

Lθ
Ke

0 1

τ
K

5
0.5
−0.1
0 0

−0.2 −0.5 −5
A B A B A B

2.5 12 15

2 10

8 10
1.5
Lq

τIθ

τIq

6
τ

1
4 5
0.5 2

0 0 0
A B A B A B

b) Pitch
Fig. 9 Variation in compensatory model parameters.
2 3 2 32 3
x_ f Af 0 0 0 0 xf
Larger rms errors in both pitch and roll, along with relatively 6 7 6 76 7
higher levels of nonminimum phase characteristics clearly implies 6 x_ e 7 6 0 Ae 0 0 0 7 6 7
6 7 6 76 xe 7
that subjects found task B to be more challenging. Comparing the 6 7 6 76 7
6 x_ e_ 7  6 0 0 Ae_ 0 0 7 6 7
relative variations in equalzation parameters between pitch and roll 6 7 6 76 xe_ 7
6 7 6 76 7
components (Figs. 9a and b) also shows the degree of attention 6 x_ d 7 6 Bd Cf Bd Ce Bd Ce_ Ad 0 7 6 7
4 5 4 54 xd 5
allocation for this task. Greater variation in pitch parameters confirms
that observations where subjects were found to focus on the more _xn Bn Cf Bn Ce Bn Ce_ Bn Cd An xn
demanding task of tracking roll commands before ensuring pitch 2 B
f 0 0 3
demands were met. 2 3
6 0
6 Be 0 7 7 r
6 76 7
6 0
6 0 Be_ 776 e7
V. Pilot Model for Robust Analysis 6 74 5
6 B D B D B D 7 e_
4 d f d e d e_ 5
Previously described pilot models were converted to state-space
form to study the effects of the observed variations in parameters via Bn Df Bn De Bn De_
parametric uncertainty. For each parameter, the median was taken as 2 3
xf
the nominal value. The first and third quartiles were used to define the
6 7
lower and upper uncertainty bounds, respectively. The necessary 6 xe 7
state-space form for the pilot models is obtained by combining 6 7
6 7
Eqs. (18), (22), and (20) with an approximation of the pilot δm   0 0 0 0 Cn 6 xe_ 7  Remnant (23)
6 7
introduced time delay according to the block diagram in Fig. 6. The 6x 7
4 d5
following model effectively represents the pilot model with both
tracking and compensatory components: xn
 LONE ET AL. 645

1 0.14 6

0.8 0.12 5

0.6

erms
Th

r
0.1 4

σ
e
0.4
0.08 3
0.2
0.06 2
A B A B A B

14 0.35 0.11
0.105
13 0.3
0.1

0.25 0.095
12
ωn

n
0.09

τ
0.2 0.085
11
0.08
0.15
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

10 0.075
0.1
A B A B A B
a) Pitch

1 0.25 18
16
0.8
0.2 14

0.6 12

rms
eTh

σr

0.15 10

e
0.4 8
0.1 6
0.2
4
0 0.05 2
A B A B A B

20 0.11
0.3 0.105
18
0.1
16 0.25 0.095
ωn

ζn

τn

0.09
14
0.2 0.085
12 0.08
0.15 0.075
10

A B A B A B
b) Roll
Fig. 10 Variation in other pilot model parameters.

Subscripts f, e, and e_ refer to the tracking, compensatory error, and bandwidths being adopted for attitude control. For task B, a subject
error rate components, respectively. The inherent time delay and may adopt a bandwidth that is an order of magnitude lower than
neuromuscular components are denoted by subscripts d and n, another subject. The attitude frequency response also shows a region
respectively. A second-order Padé approximation has been used to of zero steady-state error between 0.4 to 10 rad∕s for task A. It is also
include the time delay here. Thus, Eq. (23) is a nine-state pilot model amplified at low frequencies, highlighting the possibility of high gain
(two-state error compensation, two-state feedforward compensation, behavior from the pilot such as overcontrol. In contrast, the response
three-state neuromuscular model, and a two-state delay approxima- for task B shows little variation due to the parametric uncertainty until
tion) that consists of 15 uncertain parameters. 2 rad∕s, beyond which the main intersubject difference is the adopted
Such a formulation allows the manual control dynamics to be bandwidth. Furthermore, in Fig. 13b, structural dynamics can be seen
coupled with the linearized aircraft model, as shown in Fig. 12. Effects to be insignificant and confined to the higher regions of the manual
of intersubject differences on closed-loop PVS performance can control frequency range.
now be investigated in the frequency domain. These differences are The static nature of the manual control experiments; that is,
propagated along the system via a structured uncertainty matrix. The desktop based simulation, means that the pilot is unable to use
upper and lower bounds for attitude and acceleration responses to vestibular senses to obtain acceleration cues and hence close an
perceived error in pitch and roll given such uncertainty in pilot model acceleration feedback loop. The resulting PVS acceleration response
parameters are shown in Fig. 13. must be interpreted with this in mind. The key feature in Fig. 13b is
Figure 13a also highlights the shaping introduced by the subject the resulting resonance in acceleration response due to low-frequency
into the closed loop system as a result of change in objective or task. airframe structural modes. Although the levels of amplification may
The relatively less demanding task A leads to a smaller range of be exaggerated when compared to real flight, this highlights the
 646 LONE ET AL.

0.02 0.7 6

0.6 5
0.01
0.5 4
0

If
0.4 3

Kf

τ
−0.01 0.3 2

0.2 1
−0.02
0.1 0
A B A B A B

14 25 30

12 25
20
10 20
15
8 15

τLLf
ILf

Lf
τ
τ

10
6 10

4 5 5
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

2 0
0
0 −5
A B A B A B
a) Pitch

0.02 0.7 4

0.015 0.6
3
0.01 0.5
2

τIf
0.005 0.4
f

τ
K

0 0.3 1

−0.005 0.2
0
−0.01 0.1
A B A B A B

2.5 12
10 10
2
8
1.5 5 6
LLf
ILf

τLf
τ

1 4

0 2
0.5
0
0
−5 −2
A B A B A B
b) Roll
Fig. 11 Variation in pilot tracking model parameters.

susceptibility to APC as a consequence of low-frequency structural VI. Pilot-Model-in-the-Loop Simulation

modes. It should also be noted that a relatively high value of overall The linearized aircraft model used for manual control experiments
structural damping was used. Modern aircraft are known to have only provided insight into a few flight dynamic parameters. Here, the
damping much less than the 10% assumed here. pilot models derived from the experiments are coupled with the full
nonlinear aircraft model to study the effects on airframe loads and
other interesting quantities. Such a simulation environment allows
pilot model parameters to be varied such that scenarios where the
airframe is overloaded due to pilot control action can be investigated.
Structured uncertainty
Simulations were conducted where pilot model parameters were
systematically varied to obtain probabilities for worst case loading.

c A. Nonlinear Flexible Aircraft Model

reference
Pilot model
Flexible aircraft model A modified strip theory has been implemented for the calculation
of wing aerodynamic loading. Variations due to local changes in
aerofoil profiles are taken into account via a correction factor
K FCS proposed by DeYoung and Harper [32] and compressibility effects
Flight control system have been included via the Prandtl–Glauert correction factor. Indicial
unsteady aerodynamics for the lifting surfaces have been included as
Fig. 12 Block diagram for robust analysis. a sum of three-dimensional circulatory forces and two-dimensional
 LONE ET AL. 647

20 60
Task A
Task A
Task B Task B
40
0

Singular values (dB)

Singular values (dB)
20
−20
0
−40
−20
−60
−40

−80 −60

−100 −80 −2 −1 0 1 2
−2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/s)
a) Attitude response b) Acceleration response
Fig. 13 Boundaries for closed-loop singular values for the PVS.
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

noncirculatory forces, as proposed by Leishman and Nguyen [33]. mode. Further details are not presented because certain components
This approach, therefore, models the wing downwash on the of the GLA, MLA, and AMC systems are considered to be
horizontal tailplane and unsteady effects due to control surface proprietary.
deflections. Effects on the lifting surfaces due to the fuselage and vice
versa are modelled by modifying the inboard and outboard lift, as B. Loads Envelope Comparison
suggested in ESDU 94009 [34]. Forces and moments due to the A loads envelope for this study was derived by considering both
flexible fuselage have been modelled by effectively treating the manoeuver cases as well as continuous turbulence scenarios. The
fuselage as a slender body [35]. Contributions from the engine manoeuver loads envelope for the wing and empennage were
nacelles are modelled by treating them as annular aerofoils [36]. obtained via the simulation of the following traditional book-case
The structural model considers the airframe as a system composed unchecked maneuvers:
of 40 nodes (with six degrees of freedom) connected to each other via 1) Positive pitching manoeuver as specified in CS-25.331(c)(1).
beam elements. The beam elements have their corresponding This involves the stick being deflected fully aft until the load factor
stiffness matrices leading to local variations in airframe stiffness, and reaches either a maximum or 2.5g at the centre of gravity. Similarly,
a global structural damping of 3% has been assumed [7]. Limiting the the unchecked negative pitching manoeuver requires the achieve-
model to only the first 12 modes was found to capture all the required ment of either a minimum or −1.0g load factor. The stick is released
dynamics in the frequency range of interest. This range consists of the immediately after the relevant load factors are achieved and loadings
low-frequency phugoid mode and the upper limit of manual control beyond this point are not considered.
frequencies. The 12 modes include the first and second wing bending 2) Rolling manoeuvre as specified in CS25.349(a) where the
and torsional modes as well as horizontal tailplane (HTP) and aircraft is banked until the maximum allowable roll rate is achieved,
fuselage modes. Modal frequencies and shapes are presented in which in this case is limited to 15 deg ∕s by the FCS. The aircraft is
Table 2 and Fig. 14, respectively. then recovered to zero bank angle.
The aerodynamic and structural forces and moment contributions 3) Yaw manoeuver as specified in CS25.351 where the rudder is
are then coupled with the nonlinear aircraft equations of motion in a deflected to its maximum at the fastest rate of deployment to cause an
MATLAB/Simulink environment. The aircraft model used here was overswing in sideslip after which the rudder is returned to zero
developed at Cranfield University and further details can be found in deflection. The rudder actuator has been modelled such that the
work done by Andrews [7], Andrews and Cooke [26], Lone et al. [6], maximum rudder deflection and deployment rate are airspeed
and Lone and Cooke [27]. The same command and stability dependent.
augmentation system (CSAS) described earlier in Sec. III is In all cases, the aircraft starts in a trimmed level flight condition
implemented here. Gust load alleviation (GLA) and manoeuvre load before the manoeuver is executed. These were performed at 53 flight
alleviation (MLA) systems as described by Burris and Bender [37] points to provide a conservative internal loads envelope.
are also modelled here, along with an active mode control (AMC) The approach used here adheres to the methodology given in
system that simply considered the first symmetric wing bending CS25.341(b). Given the nonlinear FCS components and the
significant influence of unsteady aerodynamics as highlighted by
Andrews [7], time history analysis was found to be the only adequate
approach. The turbulence profiles used are characterized by rms
velocities of around 6 m∕s, which corresponds to severe turbulence
Table 2 Aircraft structural mode types and frequencies or stormy weather conditions. These were generated using the von
Kármán turbulence model with the turbulence scale set to 2500 ft.
Mode number Mode type Frequency, rad∕s
Limit loads for the wing and empennage were then obtained by
1 Wing (asymmetric) 6.9
following the requirements specified in CS-25 and the process
2 Wing (symmetric) 6.9
3 Wing  fuselage (symmetric) 7.5 described by Hoblit [2].
4 Wing  HTP  fuselage (asymmetric) 12.8 The simulations were conducted with the aircraft initially in
5 Wing  Fuselage (asymmetric) 14.7 trimmed level flight with the autothrottle and height hold autopilots
6 Wing  HTP  fuselage (asymmetric) 16.0 engaged. Each simulation was of 5 min duration as suggested
7 Wing (asymmetric) 16.2 by Hoblit [2] to ensure sufficient data were generated to obtain
8 Wing (symmetric) 16.2 statistically significant rms values together with an adequate
9 Wing  HTP  Fuselage (symmetric) 16.9 quantity of peak count data. The aircraft model was subjected to
10 Wing  HTP (asymmetric) 17.2
11 Wing (asymmetric) 18.4
translational turbulence in all three axes simultaneously, similar
12 Wing (Symmetric) 18.4 to the approach dubbed “three-dimensional gust analysis” at
Lockheed-Martin [2].
 648 LONE ET AL.
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

Fig. 14 Airframe flexible mode shapes.

2.5 2
Continuous turbulence envelope Task A
Task A
Task B
Task B
2 +2.5g Unchecked 0
Manoeuvre Envelope

1.5 −2
Normalised Bending Moment

Normalised Torsion

1 −4

−6
0.5

+2.5g Unchecked
−8 Manoeuvre Envelope
0

Continuous turbulence
−10
−0.5
envelope
-1.0g Unchecked
Manoeuvre Envelope
−12
−1 −30 −20 −10 0 10 20 30
−30 −20 −10 0 10 20 30 Wing span (m)
Wing span (m) Fig. 16 Envelope comparison of spanwise torsion for tasks A and B.
Fig. 15 Envelope comparison of spanwise wing bending moment for
tasks A and B.

VII. Conclusions
The maximum and minimum spanwise wing bending moment and A framework is presented to allow the investigation of interactions
torsion for each task can be compared with the manoeuvre and between manual control and flight control systems for large flexible
continuous turbulence envelopes in Figs. 15 and 16. The presented aircraft. It allows the fast simulation of particular flight test
loads are the worst case loads obtained via the pilot-model-in-the- manoeuvres designed to expose unwanted interactions between the
loop simulations. As expected by the load factor time histories pilot and aircraft flight dynamics. It combines a trajectory generation
required for each task (see Figs. 2 and 3), execution of task B, which scheme and an aeroelastic aircraft model together with pilot models
involves the highest load of approximately 1.5g, results in the highest to conduct flight loads analysis relevant to trajectory-based flight test
loads. However, worst case loadings for both tasks remain well within manoeuvres. The ability of the framework to assess the impact of
the envelope showing the trajectory generation process’ overall variations in pilot control characteristics on flight loads is illustrated
effectiveness and robustness to variations in pilot dynamics. through the simulation of two manoeuvres. The first (task A) required
 LONE ET AL. 649

the pilot to avoid an obstacle by following a four-dimensional [14] Hull, D. G., Fundamentals of Airplane Flight Mechanics, Springer–
trajectory. The second (task B) required the pilot to align the aircraft Verlag, New York, 2007.
from an offset position in minimum time. [15] Menon, P. K., Sweriduk, G. D., and Sridhar, B., “Optimal Strategies for
Classical pilot models representative of pilot control behavior were Free-Flight Air Traffic Conflict Resolution,” Journal of Guidance,
Control, and Dynamics, Vol. 22, No. 2, 1999, pp. 202–211.
parameterized using data from a series of manual control experiments doi:10.2514/2.4384
involving pilots following flight director cues generated by a [16] Nuic, A., User Manual for the Base of Aircraft Data (BADA) Revision 3.10.
guidance system. Variations in pilot model parameters showed that EEC Technical/Scientific 12/04/10-45, EUROCONTROL, Centre de Bois
pilots found task B to be the most challenging. These variations were des Bordes, Brétigny-sur-Orge, France, April 2012.
also used to derive a pilot model with structured uncertainty, which [17] Betts, J. T., “Survey of Numerical Methods for Trajectory
was coupled with a flexible aircraft model to demonstrate the affects Optimization,” Journal of Guidance, Control, and Dynamics, Vol. 21,
on PVS frequency response due to changes in task and intersubject No. 2, 1998, pp. 193–207.
differences. Spanwise wing bending moment and torsion undergone doi:10.2514/2.4231
during these manoeuvres were then compared with manoeuvre and [18] Rao, A. V., Benson, D. A., Darby, C. L., Patterson, M. A., Francolin, C.,
Sanders, I., and Huntington, G. T., “Algorithm 902: GPOPS, A
continuous turbulence loads envelopes to assess the effects of manual MATLAB Software for Solving Multiple-Phase Optimal Control
control on internal loads. Problems Using the Gauss Pseudospectral Method,” ACM Transactions
With regards to future work, fixed and/or moving base flight on Mathematical Software, Vol. 37, No. 2, 2010, pp. 22.1–22.39.
simulation tests with experienced pilots are needed for comparison and doi:10.1145/1731022
validation of the presented results. For this purpose, piloted simulation [19] Becerra, V. M., “Solving Complex Optimal Control Problems at No
trials in the large aircraft simulator at Cranfield University are Cost with PSOPT,” Proceedings of the IEEE Multi-Conference on
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376

envisioned. An in-depth analysis studying the relationships between Systems and Control, IEEE Paper 978-1-4244-5355-9, 2010, pp. 1391–
pilot model parameters is required to understand how individual pilots 1396.
generally adapt to different control tasks. Furthermore, the brief study [20] Patel, R. B., and Goulart, P. J., “Trajectory Generation for Aircraft
Avoidance Maneuvers Using Online Optimization,” Journal of
showing the application of parametric uncertainty to study effects of Guidance, Control, and Dynamics, Vol. 34, No. 1, 2011, pp. 218–230.
intersubject variations on PVS performance is being extended to doi:10.2514/1.49518
demonstrate the application of robust control tools (typically used to [21] Bollino, K. P., and Lewis, L. R., “Collision-Free Multi-UAV Optimal
assess multi-input–multi-output controllers) and pilot models, such as Path Planning and Cooperative Control for Tactical Applications,” AIAA
the optimal control pilot model in this context. Guidance, Navigation and Control Conference, AIAA Paper 2008-
7134, 2008.
[22] Milam, M. B., “Real-Time Optimal Trajectory Generation for
Constrained Dynamical Systems,” Ph.D. Thesis, California Inst. of
References Technology, California, 2003.
[1] EASA Certification Specifications for Large Aeroplanes CS-25, Book 1, [23] Yakimenko, O. A., “Direct Method for Rapid Prototyping of Near-
Airworthiness Code, European Aviation Safety Agency, Sept. 2007. Optimal Aircraft Trajectories,” Journal of Guidance, Control, and
[2] Hoblit, F. M., Gust Loads on Aircraft: Concepts and Applications, Dynamics, Vol. 23, No. 5, 2000, pp. 865–875.
AIAA Education Series, AIAA, Washington, D.C., 1988. doi:10.2514/2.4616
[3] Wright, J. R., and Cooper, J. E., Introduction to Aircraft Aeroelasticity [24] Lai, C. -K., and Whidborne, J., “Real-Time Trajectory Generation for
and Loads, Wiley, New York, 2007. Collision Avoidance with Obstacle Uncertainty,” AIAA Guidance,
[4] Lambregts, A. A., Nesemeier, G., Wilborn, J. E., and Newman, R. L., Navigation and Control Conference, AIAA Paper 2011-6598, 2011.
“Airplane Upsets: Old Problem, New Issues,” AIAA Modelling and [25] Price, K., Storn, R. M., and Lampinen, J. A., Differential Evolution: A
Simulation Technologies Conference and Exhibit, AIAA Paper 2008- Practical Approach to Global Optimization, Springer–Verlag, New
6867, 2008. York, 2005.
[5] Lee, B. P., “Recent Experience in Flight Testing for Pilot Induced [26] Andrews, S., and Cooke, A., “An Aeroelastic Flexible Wing Model for
Oscillations (PIO) on Transport Aircraft,” ICAS 2000 Congress, ICAS Aircraft Simulation,” 48th AIAA Aerospace Sciences Meeting Including
Paper 2000-6.9.1, 2000, pp. 691.1–691.8. the New Horizons Forum and Aerospace Exposition, AIAA Paper
[6] Lone, M. M., Rouyan, M. N., and Cooke, A. K., “A MATLAB/Simulink 2010-35, Jan. 2010.
Suite for the Real-Time Flight Simulation of Large Aircraft,” Cranfield [27] Lone, M. M., and Cooke, A. K., “Pilot-Model-in-the-Loop Simulation
College of Aeronautics, Tech. Rept. CU/COA-2012/01, Cranfield, Environment to Study Large Aircraft Dynamics,” Proceedings of the
England, U.K., 2012. Institution of Mechanical Engineers, Part G: Journal of Aerospace
[7] Andrews, S. P., “Modelling and Simulation of Flexible Aircraft: Engineering, Vol. 227, No. 3, March 2012, pp. 555–568.
Handling Qualities with Active Loads Control,” Ph.D. Thesis, Dept. of [28] Zaal, P. M. T., Pool, D. M., and van Paassen, M.M., “Multimodal Pilot
Aerospace Engineering, School of Engineering, Cranfield Univ., Control Behaviour in Combined Target-Following and Disturbance
Cranfield, England, U.K., 2011. Rejection Tasks,” Journal of Guidance, Control, and Dynamics, Vol. 32,
[8] Kwatny, H. G., Dongmo, J. E. T., Chang, B. C., Bajpai, G., Yasar, M., No. 5, 2009, pp. 1418–1428.
and Belcastro, C., “Nonlinear Analysis of Aircraft Loss of Control,” [29] Lone, M. M., and Cooke, A. K., “Review of Pilot Modelling
Journal of Guidance, Control, and Dynamics, Vol. 36, No. 1, 2013, Techniques,” 48th AIAA Aerospace Sciences Meeting Including the New
pp. 149–162. Horizons Forum and Aerospace Exposition, AIAA Paper 2010-297,
doi:10.2514/1.56948 Jan. 2010.
[9] Palacios, R., Murua, J., and Cook, R., “Structural and Aerodynamic [30] Grant, P., and Schroeder, J., “Modeling Pilot Control Behavior for
Models in Nonlinear Flight Dynamics of Very Flexible Aircraft,” AIAA Flight Simulator Design and Assessment,” AIAA Modeling and
Journal, Vol. 48, No. 11, 2010, pp. 2648–2659. Simulation Technologies Conference, AIAA Paper 2010-8356,
doi:10.2514/1.J050513 Aug. 2010.
[10] Waszak, M. R., and Davidson, J. B., “A Simulation Study of the Flight [31] Zaal, P. M. T., Pool, D. M., Chu, Q. P., van Paaseen, M. M., Mulder, M.,
Dynamics of Elastic Aircraft, Vol. 1: Experiment, Results and Analysis,” and Mulder, J. A., “Modeling Human Multimodal Perception and
NASA, Tech. Rept. CR-4102, 1987. Control Using Genetic Maximum Likelihood Estimation,” Journal
[11] Mitchell, D. G., Doman, D. B., Key, D. L., Klyde, D. H., Leggett, D. B., of Guidance, Control, and Dynamics, Vol. 32, No. 4, 2009, pp. 1089–
Moorhouse, D. J., Mason, D. H., Raney, D. L., and Schmidt, D. K., “The 1099.
Evolution, Revolution, and Challenges of Handling Qualities,” AIAA [32] DeYoung, J., and Harper, C. W., “Theoretical Symmetric Span Loading
Atmospheric Flight Mechanics Conference and Exhibit, AIAA Paper at Subsonic Speeds for Wings Having Arbitrary Plan Form,” NACA TR-
2003-5465, Aug. 2003. 921, 1948.
[12] Mitchell, D. G., and Klyde, D. H., “Testing for Pilot-Induced [33] Leishman, J. G., and Nguyen, K. Q., “State-Space Representation of
Oscillations,” AIAA Atmospheric Flight Mechanics Conference and Unsteady Airfoil Behaviour,” AIAA Journal, Vol. 28, No. 5, 1990,
Exhibit, AIAA Paper 2005-5811, Aug. 2005. pp. 836–844.
[13] Lai, C.-K., Lone, M., Thomas, P., Whidborne, J., and Cooke, A., “On- doi:10.2514/3.25127
Board Trajectory Generation for Collision Avoidance in Unmanned [34] ESDU 94009, Symmetric Steady Manaeuvre Loads on Rigid Aircraft of
Aerial Vehicles,” IEEE Aerospace Conference, IEEE Paper 1251, Classical Configuration at Subsonic Speeds, ESDU International,
March 2011, pp. 1–14. London, 1994.
 650 LONE ET AL.

[35] ESDU 89014, Normal Force, Pitching Moment and Side Force of [37] Burris, P. M., and Bender, M. A., “Aircraft Load Alleviation and Mode
Forebody-Cylinder Combinations for Angles of Attack up to 90 Degrees Stabilisation (LAMS), B-52 System Analysis, Synthesis, and Design,”
and Mach Numbers up to 5, ESDU International, London, 2004. U.S. Air Force Flight Dynamics Laboratory, AFFDL-TR-68-161,
[36] ESDU 77012, Aerodynamic Centre of Wing-Fuselage-Nacelle Wright-Patterson AFB, OH, 1969.
Combinations: Effect of Wing-Pylon Mounted Nacelles, ESDU
International, London, 1977.
Downloaded by CRANFIELD UNIVERSITY on April 11, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.C032376