You are on page 1of 17

JOURNAL OF AIRCRAFT

Vol. 54, No. 6, November–December 2017

Stability and Control of Tailless Aircraft Using Variable-Fidelity


Aerodynamic Analysis

Jangho Park∗ and Jae-Young Choi∗


Virginia Polytechnic Institute and State University, Blacksburg, Virginia 24061
Yeongmin Jo†
Korea Advanced Institute of Science and Technology, Daejeon 34141, Republic of Korea
and
Seongim Choi‡
Virginia Polytechnic Institute and State University, Blacksburg, Virginia 24061
DOI: 10.2514/1.C034052
The purpose of this study is to evaluate the stability and control characteristics of the tailless aircraft using a
hierarchy of variable-fidelity aerodynamics analysis methods, including unsteady linear potential flow, Euler, and
Reynolds-averaged Navier–Stokes solvers as low-, medium-, and high-fidelity analysis, respectively. Contributions of
the study are numerous. First, derivations of the equations of motion for the tailless aircraft and stability criteria using
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

system matrices in longitudinal and lateral motions were carried out. Second, an efficient time-spectral Reynolds-
averaged Navier–Stokes computational fluid dynamics method, which is based on the solution approximation of a
discrete Fourier series, and the steady form of a corresponding adjoint solution approach can compute dynamic
derivatives at computational cost considerably less than that of the conventional time-marching method without
deterioration in solution accuracy. The innovative control effector 101 configuration was chosen, and the validation of
the variable-fidelity computation results was made with wind tunnel data for forces, moments, and their derivative
values in static and dynamic motions. Finally, example flight conditions of trim, climb, and side slip motions were
analyzed by the Reynolds-averaged Navier–Stokes solver for the stability analysis. It is concluded that the weak
instabilities observed in some of the dynamic modes can be easily controlled by appropriate control inputs.

Nomenclature x, y, z = x, y, and z directional force


A = axial aerodynamic force on aircraft α = angle of attack
AR = aspect ratio β = sideslip angle
b = span length of the aircraft γ = heat capacity ratio
Ck = aerodynamic coefficient with respect to k δ = deflection angle
c:g = center of gravity ΔZ = distance between moment reference center and
D = drag force on aircraft centroid of lateral area
g = gravitational acceleration ζ = decay ratio
Ii;j = moment of inertia about i and j axes θ = pitching angle
L = lift force on aircraft Λ = eigenvalue matrix
LEMAC = chordwise leading edge location of MAC λ = sweep angle
lbody = body length of aircraft ϕ = rolling angle
l; m; n = x; y; and z directional moment of the aircraft ψ = yawing angle
M = Mach number ω = frequency
M = mass of aircraft
MAC = mean aerodynamic chord of the aircraft Subscripts
N = normal aerodynamic force on aircraft
lat = lateral motion
p, q, r = x, y, and z directional angular velocity of the aircraft
long = longitudinal motion
q∞ = dynamic pressure on freestream flow
trim = trim condition of the aircraft
R = perfect gas constant
Sref = reference area of aircraft
T = temperature I. Introduction
u, v, w = x, y, and z directional velocity of aircraft
T HE purpose of this study is to investigate how effective the
variable-fidelity (VF) aerodynamic analysis methods are in the
prediction of the stability and controllability characteristics of tailless
Presented as Paper 2016-2024 at the 54th AIAA Aerospace Sciences
Meeting, San Diego, CA, 4–8 January 2016; received 2 June 2016; revision aircraft. A hierarchy of a linear potential flow, Euler, and Reynolds-
received 13 February 2017; accepted for publication 25 February 2017; averaged Navier–Stokes (RANS) computation methods consist of
published online Open Access 19 May 2017. Copyright © 2017 by the low-, medium-, and high-fidelity analysis, respectively. Stability
American Institute of Aeronautics and Astronautics, Inc. All rights reserved. criteria are derived for static and dynamic stability of the tailless
All requests for copying and permission to reprint should be submitted to CCC aircraft. One of the biggest merits is a time-efficient computational
at www.copyright.com; employ the ISSN 0021-8669 (print) or 1533-3868 fluid dynamics (CFD) method and the steady form of the adjoint
(online) to initiate your request. See also AIAA Rights and Permissions method to calculate unsteady dynamic derivatives.
www.aiaa.org/randp. The tailless configuration in aircraft design is attractive with some
*Ph.D. Candidate, Department of Aerospace and Ocean Engineering.
Student Member AIAA. advantages being the reduction of drag, radar cross section, and

Ph.D. Candidate, Department of Aerospace Engineering. Student Member weight due to an absence of vertical and horizontal stabilizers.
AIAA. However, the stability and controllability characteristics of tailless
‡ aircraft have been questioned, as the moments on aircraft during
Assistant Professor, Department of Aerospace and Ocean Engineering,
460 Old Turner St. Senior Member AIAA. maneuvers without tail structures become greatly sensitive to the
2148
PARK ET AL. 2149

location of center of gravity, often causing instabilities in control. The Table 1 Geometric parameters of ICE 101
precise evaluation of the stability and controllability characteristics of configuration
the tailless aircraft is important and carried out through the system Parameter Values
matrix analysis that requires the computation of aerodynamic forces
Area, Sref 75.125 m2
and moments as well as their derivatives. Body length, lbody 13.145 m
Traditional approaches to compute the system matrix include Span length, b 11.43 m
flight test, wind tunnel experiment, semiempirical modeling, and Aspect ratio, AR 1.74
CFD [1]. The flight test provides most accurate data, but is very Sweep angle, Λ 65 deg
expensive and almost impossible to be integrated into an iterative LEMAC 8.763 m
design procedure. Less expensive than the flight test, the wind tunnel MAC 4.085 m
experiment is a practical and reliable choice. However, the issues of Moment center (X), xc:g 38% of MAC (7.415 m)
model scaling and the simulation of the unsteady dynamic motion of Moment center (Y), yc:g 0m
the aircraft are difficult to resolve in relation to manufacturing and Moment center (Z), zc:g WL 100 (0.357 m)
operation of the wind tunnel.
CFD methods have become more available for complicated flow
analysis with the advancement of the numerical schemes and the motions were constructed. The example flight conditions of trim,
availability of high-performance computing platforms. There have climbing without acceleration, and side slip of the ICE 101
been studies to investigate the ability of the CFD method for several configuration were selected to assess the characteristics of stability and
types of tailless aircraft. The NATO RTO AVT-161 Task Group is one controllability. When the stability without control inputs was analyzed,
of the study groups to use a computational method to accurately predict weak instabilities were observed in some of the dynamic motions,
both static and dynamic stabilities of air and sea vehicles. A stability including a phugoid mode during the trim and climb conditions and
and control configuration (SACCON) is developed as a tailless aircraft in the roll mode during the side slip, which can be easily controlled with
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

model. The results of CFD computation have been validated [2–4] the control effectors introduced in the ICE 101 configuration.
against the wind tunnel experiments [5,6]. The Lockheed Martin The organization of the paper is as follows. In Sec. II, the ICE 101
tactical aircraft systems–innovative control effector (LMTAS-ICE, or geometry is introduced with various types of control surfaces, and the
ICE hereupon; shown in Fig. 1) configuration is another concept for the equations of motion are derived along with the stability criteria in
study of the stability of tailless aircraft [7,8]. The ICE configuration static and dynamic motions. VF aerodynamic analysis methods are
was developed during the early 1990s. Various wind tunnel explained in Sec. III to compute aerodynamic loads and derivatives.
experiments were performed in the Air Force Research Laboratory Section IV shows the validation of the computation results against
(AFRL) facility with different types of control surfaces [9]. The main various wind tunnel test data with good agreements. Finally, stability
focus of the experimental study was to improve aircraft performance in and controllability characteristics of the ICE 101 configuration for
stability and control through innovative control surfaces. the flight conditions of trim, climbing, and slide slip conditions are
In the current study, the ICE 101 configuration was chosen and VF discussed in Sec. V, followed by conclusions in Sec. VI.
computation methods were employed to validate wind tunnel
experiments. Merits of the current study are several. A complete set of
the equations of motion and system matrix in both longitudinal and II. Equations of Motion and Stability Criteria for
lateral motions were derived along with the stability criteria. A Tailless Aircraft
VortexJE open-source linear potential flow solver was modified to
simulate evolution of the leading edge vortex in time at high angles of In this section, the geometry of tailless aircraft of the ICE 101
attack. For the Euler and RANS CFD simulations, steady or pseudo- configuration is introduced. Derivation of equations of motion and
steady computations were carried out for dynamic motions of the criteria for stability characteristics for tailless aircraft are as follows
aircraft through an efficient formulation of a time-spectral CFD for both static and dynamic motions.
method with computational cost saved by an order of magnitude
when compared with a traditional time-marching formulation. A A. ICE 101 Configuration
corresponding steady form of the adjoint method can be directly ICE configuration was chosen for our study as many wind tunnel
applied for unsteady dynamic motions, although a straight-forward data are available with various types of control surfaces [8–10]. Out
but expensive finite difference method (FDM) is also used. This fact of two control effectors, ICE 101 and ICE 201, which were
becomes significant when it comes to the design aspect of the tailless constructed during the early 1990s as a land-based and a carrier-based
aircraft that requires fast turnaround time during iterative process to baseline, respectively, the ICE 101 configuration (Fig. 1) was chosen
find an optimal control surface for the tailless aircraft. One of the key for our study and the geometric details are described in Table 1.
aspects of the current study is a trade-off in the VF analysis between Figure 1 shows the model used in the wind tunnel test in Air Force
solution accuracy and computational efficiency. Research Laboratory.
In this way, force and moment coefficients as well as their derivative Several types of control surface were considered for the ICE 101
values in static and dynamic motions were computed by VF analysis configuration during the AFRL wind tunnel test [8,9], and are shown
methods with respect to varying flow angles, angular rates, velocities, in Fig. 2. The wind tunnel experiments concluded that the most
and accelerations. The system matrices in both longitudinal and lateral effective types are all moving wingtip (AMT), differential leading

Fig. 1 ICE 101 model for wind tunnel test.


2150 PARK ET AL.

Fig. 2 Control surfaces of ICE 101 configuration.

edge flap (DLEF), and spoiler slot deflector (SSD). The details in by an angle from the x axis to the freestream velocity vector in the
control and stability characteristics will be discussed in Sec. III.D. x–y plane.
The AMT is a triangular moving control surface at aircraft wing 6) A thrust vector is assumed to be aligned with the x direction
tips. The AMT is used to assist lift force and roll and yaw moments as passing through the origin.
well as general lateral directional control. The advantages of the As the CFD analysis uses a coordinate system and flight angles
AMT are that it is easy to be installed on thin sections of the wing, is different from those for dynamics analysis, a coordinate system for
simpler to actuate, and can be made larger in size than the clamshell the aerodynamic analysis, shown in Fig. 3, is employed for the
type for better controllability. The SSD has advantages in lateral/ derivation of equations of motion. That is, instead of a traditional
directional control at high angles of attack in the transonic condition. North-East-Down coordinate system for the aircraft motion, a North-
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Compared with a conventional aileron, it is less sensitive to high- West-Up coordinate system of (X E , Y E , ZE ) is used as a body-fitted
speed flexibility. The DLEF also provides lateral and directional reference coordinate system to identify flight angles and their angular
control at high angles of attack by controlling the leading edge vortex velocity. Corresponding equations of motion for tailless aircraft are
flows. As a result, parameters that affect the geometry of the wing derived based on the Newton’s second law of conservation of linear
leading edge, including leading edge sweep, leading edge curvature and angular momentum.
of the wing, and the local angle of attack, are critical for the DLEF.
X  M
 u_ − vr  wq − q sin θ (1)
B. Equations of Motion for Stability Analysis
The equations of motion and flight stability criteria for tailless
aircraft are different from those of conventional aircraft with Y  M
 v_  ur − pw  g cos θ sin ϕ (2)
stabilizers. Aerodynamic forces and derivatives associated with tails
or body are different in magnitude, and the equations of motion must
be rederived based on [11–16]. The following are the assumptions for Z  M
 w_  vp − uq  g cos θ cos ϕ (3)
deriving the equations of motion for the tailless aircraft.
1) The geometry is assumed to be a single wing–body configuration.
2) All motions are assumed to be rigid, and aeroelasticity and mass l  I xx p_ − I xz _r  pq  qrI zz − I yy  (4)
are assumed constant.
3) All perturbations are assumed small in the linearization process.
4) A body-fixed coordinate system (X B , Y B , ZB ) with an origin m  I yy q_  I xz p2 − r2   prI xx − I zz  (5)
fixed at the center of mass is used as a reference coordinate system.
Normal force, axial force, and side force acting on the aircraft along
with the coordinate system are shown in Fig. 3. For aerodynamics n  I zz r_ − I xz p_  pqI yy − Ixx   I xz qr (6)
analysis, the origin is at the aerodynamic center.
5) An angle of attack is defined by an angle from the x axis to the Moreover, the relationships between flight angles and angular
freestream velocity vector in the x–z plane. Sideslip angle is defined velocities are defined as

Fig. 3 Coordinate system and sign convention for ICE configuration.


PARK ET AL. 2151

ϕ_  p  q sin ϕ tan θ  r cos ϕ tan θ (7) moment with respect to the sideslip angle should be negative. Pressure
differences produced in a yaw direction due to fuselage, sweep angle,
and dihedral angle are easily computed by the CFD computation.
θ_  q cos ϕ − r sin ϕ (8)
   
∂Cn ∂Cl
> 0; and <0 (14)
∂β trim ∂β trim
φ_  q sin ϕ  r cos ϕ sec θ (9)
The dynamic stability criteria are derived from the linearized
Axial force and thrust are aligned with the x axis, with side force equations of motion of Eqs. (10) and (11). First, Eqs. (10) and (11) can
with the y axis. Normal force is on the z axis, aligned with the direction be simplified as in Eq. (15). A and B are the system matrices to define
of gravity. Substituting force terms into Eqs. (1–9), the equations of the characteristics of the motion and stability of aircraft in longitudinal
motion for tailless aircraft are derived. With an assumption of a steady and lateral directions, respectively.
equilibrium state with small perturbations, which does not take into
account unsteady flow including gusts, the aircraft is affected only by x_  Ax; and y_  By (15)
flows induced from its motion. In this way, the angle of attack α is
equivalent to the pitching angle θ, and the side slip angle β is to the The decay ratios (ζlong , ζ lat ) and the decay frequencies (ωlong , ωlat )
yawing angle ψ. As a result, Eqs. (1–9) are linearized as follows: predict the characteristic modes by the eigenvalues of each matrix. The

2 38 9
8 9 η ∂C η ∂C η ∂C η ∂C
∂u  μu0 CA0 ∂w  μw0 CA0 ∂q − w0 ∂α_  g cos θ0 > > Δu >
A A A A
> u_ > >
>
> > > 6 7> >
< w_ = 6 ∂CN
η ∂C η ∂C η ∂C 7>
< Δw >
=
6η  μu0 CN0 ∂w  μw0 CN 0
N
∂q − w0
N N
 g cos θ 07
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

 6 ∂u ∂α_
7 (10)
>
> >
q_ > 6 ∂C
η ∂C η ∂C η ∂C 7>
> Δq >
>
>
: > ; 4 η ∂u  μu0 CA0 ∂w  μw0 Cm0 ∂q − w0 ∂α_  g cos θ0 5> >
A m m m
>
: Δα >
;
α_ 0 0 1 0

2 3
8 9 8 9
> v > η ∂C s
η ∂C
∂p  w0
s
η ∂C
∂r − u0
s
η ∂C
∂ϕ − g cos θ0
s
η ∂Cs
Δv >
>
>
_ >
> 6 ∂v ∂ψ 7>>
> >
>
>
> > 6
> 6 K η l  K η ∂Cn ∂Cn 7>> >
< p_ >
> ∂C
= 6 xz ∂v xx ∂v Kxz η ∂C ∂Cn
∂p  K xx η ∂p
l
K xz η ∂C ∂Cn
∂r  K xx η ∂r
l
Kxz η ∂C ∂Cn
∂ϕ  K xx η ∂ϕ
l
K xz η ∂C
∂ψ  K xx η ∂ψ 7
l
>
< Δp >
>
=
7
r_  6 ∂Cn 7
> > 6 K η ∂Cl
 K ∂Cn
zz η ∂v K xz η ∂C ∂Cn
∂p  K zz η ∂p
l
Kxz η ∂C ∂Cn
∂r  K zz η ∂r
l
Kxz η ∂C ∂Cn
∂ϕ  K zz η ∂ϕ
l
K xz η ∂ψ  K zz η ∂ψ 7> Δr >
∂Cl (11)
>
> >
> 6 xz ∂v 7> >
>
> ϕ>_
> 6 tan θ0 7>> >
> Δϕ >
>
>
: > ; 4
0 1 0 0 5>: >
;
ψ_ 0 1 sec θ0 0 0 Δψ

q∞ Sref decay ratio of each mode should be between 0 and 1, and the decay
η ;
M frequency should be minimized to satisfy the dynamic stability criteria.
q S c Their relationships are shown in Eqs. (16) and (17):
ηi;j  ∞ ref ;
I i;j
jλlong I − Aj  0; λ2long  2ζ long λlong  ω2long  0 (16)
ρS ρS c
μ  ref mμi;j  ref ;
M I i;j
I q S b jλlat I − Bj  0; λ2lat  2ζ lat λlat  ω2lat  0 (17)
K xx  xx ∞ ref2 ;
I xx I zz − I xz
In summary, the coefficients and derivatives values for the stability
I q S b
Kzz  zz ∞ ref 2 ; analysis through wind tunnel experiments or CFD computations are
I xx I zz − I xz the following:
I q S b 1) Aerodynamic force and moment coefficients.
K xz  xz ∞ ref2 (12) 2) Static derivatives: Derivatives of force and moment coefficients
I xx I zz − I xz
with respect to velocity (u, v, w) and angular state (ϕ, θ, ψ).
Equations for longitudinal motions are shown in Eq. (10) and 3) Steady dynamic derivatives: Derivatives of force and moment
those for lateral motions are in Eq. (11). It should also be noted that coefficients with respect to angular velocity (p, q, r).
the perturbations in longitudinal and lateral directions are 4) Unsteady dynamic derivatives: Derivatives of force and moment
independent and their variables are decoupled, which is valid for coefficients with respect to the time rate of change of the angle of
aircraft of symmetric shape under a steady equilibrium condition. attack and the side slip angle (α,
_ β).
_

C. Static and Dynamic Stability Criteria for Tailless Aircraft D. Motions with Control Input for Tailless Aircraft
Static stability includes longitudinal and lateral motions. For the A control input matrix system is added to the state equation of
longitudinal stability, pitching moment at the zero-lift condition Eq. (15) to analyze the stability characteristics of the tailless aircraft
should be positive, and its derivative with respect to the angle of with respect to control inputs. The system is derived from the linearized
attack in a trim condition should be negative. equations of motion with assumptions of small perturbations in angles
  and forces. The size of a control input vector and a system matrix
∂Cm corresponds to the number of control surface on the aircraft. If an ICE
Cm L0 > 0; and <0 (13)
∂α trim configuration has three main control surfaces of AMT, DLEF, and
SSD, the control input matrix system of longitudinal motion can be
For the lateral stability, a derivative of yawing moment with respect written as Eq. (18). The control input matrix system can be simplified
to the sideslip angle should be positive, and a derivative of rolling as C and added to Eq. (15) to be written as in Eq. (19):
2152 PARK ET AL.

2 3
η ∂δ∂CAMT
A
η ∂δ∂C A
η ∂δ∂CSSD
A
8 9
6 DLEF
7> ΔδAMT >
6 η ∂CN ∂CN 7> < >
=
6 ∂δAMT η ∂δ∂CDLEF
N
η ∂δSSD 7
6 7 ΔδDLEF (18)
6 ∂Cm m 7> >
4 η ∂δAMT η ∂δ∂CDLEF
m
η ∂δ∂CSSD 5>: Δδ >
;
SSD
0 0 0

x_  Ax  Cu (19)

III. Variable-Fidelity Aerodynamic Analysis for


Stability Derivatives
A hierarchy of aerodynamic analysis methods to compute the
aerodynamic coefficients of tailless aircraft needs to be assessed in
Fig. 5 Leading and trailing edge vortex models [18].
terms of accuracy and computational cost. An unsteady linear
potential flow, Euler, and RANS flow solvers consist of low-,
medium-, and high-fidelity analysis method in our study. Details on Cs  Cs − ΔCD0  cos α sin β (22)
the computation of forces, moments, and derivatives in both steady
and unsteady flow conditions are different depending on the choice of
fidelity. Aerodynamic derivative values for static, steady dynamic, ΔZ
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

and unsteady dynamic motions are calculated with either a finite Cl  Cl − ΔCD0  cos α sin β (23)
difference or an adjoint solution method. b

A. Linear Potential Flow and Reynolds-Averaged Navier–Stokes ΔZ


Solvers Cm  Cm  ΔCD0  cos α (24)
MAC
As a low-fidelity analysis method, a linear-potential flow equation is
solved with its governing equation, ∇ ⋅ v  0. An open-source panel
code of VortexJE [17] is modified for the current study, which is based As higher-fidelity analysis methods, both Euler and RANS
on a C++ programming language and provides explicit time equations are solved by an SU2 flow solver [22–25]. The flow solver
integration with wake models for both steady and unsteady flows. The uses advanced numerical schemes, including a Roe’s second- to
method assumes a doublet source in every panel and models the time- third-order accurate spatial discretization and the Venkatakrishnan
dependent unsteady trailing edge wake. The number of panels used for limiter. An Euler implicit method with an LU-SGS scheme is applied
the ICE 101 configuration is approximately 4500–6500 (Fig. 4). for time integration with better numerical stability, and the multigrid
Unlike a traditional aircraft, the tailless aircraft in delta-wing technique is used to accelerate the solution convergence.
shape has a strong leading edge vortex at a high angle of attack, For primal and adjoint solutions, all the iterative computations of
which has to be modeled for the accurate prediction of lift and drag. CFD are converged to 10−8 of L2 norm residual of density. Using
In this paper, an explicit, unsteady leading edge wake is modeled Pointwise software, an unstructured mesh is generated using a T-rex
based on the approach suggested by Katz and Plotkin [18]. The approach to generate boundary-layer cells. The grid with 4.64–5.10
method computes the strength of the doublet source at the leading million volume cells is used for the Euler computation (Fig. 7). For
edge panel by the Kutta condition. Figure 5 shows the concept of the RANS computation, 14.50–14.65 million volume cells are used.
leading edge vortex modeling, and Fig. 6 shows the time history of To compare the panel, Euler, and RANS analysis, pressure contour
unsteady leading edge wake model [19–21]. To take into account and force coefficients of the ICE 101 configuration are compared in
the flow viscosity, an empirical model is constructed based on the Figs. 8 and 9, and Table 2. The flow conditions are at a Mach number
experimental data at zero angle of attack, and correction models are of 0.3 and an angle of attack of 3.76 deg. All results show relatively
presented in Eqs. (21–24). good agreement. Normal force predictions are in excellent agreement
with difference less than a few counts. However, axial force shows
CN  CN  ΔCD0  sin α (20) some discrepancies in the panel and Euler solutions mostly due to
their limitations in predicting viscous flow. In terms of computational
cost, the panel method takes about 2 min for steady flow and 2 h for
CA  CA  ΔCD0  cos α (21) unsteady flows case using a single CPU for the development of
leading edge vortex wake. The CFD calculation takes about 3–7 h
using a total of 63 cores. For all computation, a CPU with AMD
Opteron processor 6376 with 3.30 GHz is used.
A key aspect is a trade-off between the solution accuracy and
computational cost among different computation methods. One of
the purposes of the current study is to investigate the solution
accuracy with lower-fidelity analysis methods. Although the
validation study shown in Sec. IV uses the panel, Euler, and RANS
solution methods, analysis of the stability characteristics of the ICE
101 configuration in our study is carried out with the RANS
computation.

B. Computation of Steady Static Stability Derivatives


Steady static derivatives consisting of the partial derivatives of
forces and moments with respect to velocity and angular states are

∂Cx ∂Cx ∂Cx ∂Cx ∂Cx ∂Cn


; ; ; ; ; (25)
Fig. 4 A panel topology for the potential flow solver. ∂u ∂v ∂w ∂ϕ ∂θ ∂ψ
PARK ET AL. 2153

Fig. 6 Time history of the leading edge vortex wake.


Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Fig. 7 Unstructured mesh topology for RANS calculation.

Both an FDM and an adjoint solution method are used to compute three flow calculations, whereas the adjoint method requires one flow
the derivative values. The FDM is straightforward and great for the calculation and one adjoint calculation, which takes similar computation
sensitivity of a large number of outputs with respect to relatively a time as for the flow simulation. The FDM takes about 1.5 time longer
small number of inputs. However, as the method is greatly dependent than the adjoint method. The adjoint solution method is employed for the
on the step size for accuracy due to cancellation errors, a parameter derivative calculation. However, the adjoint method is implemented only
study with respect to a varying step size was carried out separately. On in the Euler and RANS solutions, and the FDM is used for the potential
the other hand, the adjoint solution method provides an efficient way to flow solutions.
compute aerodynamic derivatives [24,25] with a large number of Calculation of static derivatives with respect to x-direction
inputs [O (103 –104 )]. velocity, u, can be rewritten in Eq. (26) through a chain rule using the
The system equations (10) and (11) have the similar number of relationship shown in Eq. (27), where x represents force or moment in
derivative outputs and inputs, and so both methods can be used. any direction. Derivatives with respect to Mach number, temperature,
However, the second-order accurate central difference method needs and angle of attack have to be computed a priori either by the adjoint
2154 PARK ET AL.

Fig. 8 Comparison of pressure contours (M  0.3, α  3.76 deg).


Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Fig. 9 Pressure coefficients along the wing cross sections (M  0.3, α  3.76 deg).

solution method or the FDM. Examples of static derivative of ICE angle. For the steady dynamic derivatives, angle of attack or side slip
101 configuration, normal force derivative of ∂CN ∕∂α at β  0, and angle remain constant during the motions.
rolling moment derivative ∂Cl ∕∂α at β  5 are compared in Fig. 10
among wind tunnel test, central difference FDM, and adjoint method. 1. Steady Dynamic Condition
As no information on the error margins or bound available with the Steady dynamic derivatives are those with respect to angular
experiments, predictions by RANS CFD computation are reasonably velocities of p, q, and r, which represent roll, pitch, and yaw rate,
good. respectively, whereas angles in longitudinal and lateral directions are
fixed at constant values. The steady dynamic derivatives of normal
p p p  force are defined as follows [26–30]:
dCx γRT cos α M γR cos α M γRT sin α −1
  p − (26)
du CxM 2 T CxT Cxα
∂CN ∂CN ∂CN
; ; (28)
∂p ∂q ∂r

p The computation of the derivative of any aerodynamic coefficient


u  M γRT cos α (27)
Cx with respect to pitching angular velocity q is calculated by
Eq. (29). Assuming that there are no changes in angle of attack
(Δα  0) and time rate of change of angle of attack (Δα_  0), ΔCx is
approximated by the derivative Cxq and the change in normalized
C. Computation of Dynamic Stability Derivatives angular velocity Δq.
 Then, the value of CXq is calculated by the
Dynamic derivatives resulted from both steady and unsteady relation shown in Eq. (29).
conditions, depending on the rates of angle of attack and side slip
ΔCx
ΔCx  Cxα Δα  Cxα Δα_  Cxq Δq
 · · · ≈Cxq Δq;
 Cxq ≈
Δq
(29)
Table 2 Comparison of force coefficients
(M  0.3, α  3.76 deg)  
qlbody V
Method CN CA q  q (30)
2V R
Panel 0.122787 −0.0005133
Euler 0.122184 −0.0044026
RANS 0.123002 0.0030484 where L represents the aircraft length. As there is no change in the
angle of attack (Δα  0 and Δα_  0), a rotation of the aircraft is
PARK ET AL. 2155

Fig. 10 Comparison of derivatives among wind tunnel data, and results from the FDM and the adjoint method (Mach  0.3).
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

needed, as shown in Fig. 11a, to impose a nonzero angular velocity of 2. Unsteady Dynamic Condition
Δq. For the CFD computation, a nonzero pitch rate of q, which Unsteady dynamic derivatives represent sensitivities of forces and
defines the rotation speed of the aircraft, can be simulated by the moments with respect to accelerations (u, _ v,
_ and w) _ as well as time
rotation of inflows around a center of rotation located away from the rate of change of the angle of attack and the side slip angle (α_ and β).
_
aircraft by a distance R, while the aircraft is fixed at a certain location. For example, the dynamic unsteady derivatives of the normal force
In this way, steady and uniform rotating inflow condition can be coefficient are [31].
imposed for the CFD calculation rather than rotating the aircraft
itself. Figure 11b shows the pressure contour of Euler calculation ∂CN ∂CN ∂CN ∂CN ∂CN
with a nondimensionalized pitch rate of 0.1. ; ; ; ; (35)
Derivatives with respect to yaw and roll rates can be derived in a ∂u_ ∂v_ ∂w_ ∂α_ ∂β_
similar way and formulated in Eqs. (31–34)
The CFD computation of the unsteady dynamic derivatives with
pb respect to the time rate of change of the angle of attack and the side
ΔCx  Cxα Δα  Cxα Δα_  Cxp Δp
 · · · ≈Cxp Δp;
 p  slip angle can be carried out by a forced oscillation motion. Unsteady
2V
(31) time-marching integration or a frequency domain-based, time-
spectral method [26] or a harmonic balance method (HBM), which
was developed by the current authors [32–34] based on the Fourier
series approximation of the flow solutions, can be used to effectively
ΔCx simulate flows under forced oscillation with nonzero time rate of
Cxp  (32)
Δp change of angle of attack and side slip angle. A brief summary of
mathematical formulation of the harmonic balance method is shown.
If we assume a state vector of W and a flux vector of F as a discrete
Fourier series as in Eqs. (37) and (38), then Eq. (36) can be written as
ΔCx  Cxα Δα  Cxα Δα_  Cxr Δr · · · ≈Cxr Δr (33)
Eq. (39). If we use an inverse Fourier transform to represent Eq. (39)
in a time domain, then it can be finally written as Eq. (41).

ΔCx rb ∂Wt
Cxr  ; r  (34) Rt   Ft  0 (36)
Δr 2V ∂t

Fig. 11 Pitching motions at a constant angle of attack and corresponding pressure contour.
2156 PARK ET AL.

X
N improvement in the computational efficiency by up to an order of
Wt  W^ 0  W^ cos ωnt  W^ sin ωnt or Wt magnitude when compared with that of the conventional time-
n1 marching CFD method.
X
kN∕2–1 The forced oscillation motion is defined by
 W^ k eikt (37)
k−N∕2 αt  α0  αA sinωt (42)

αt
_  q  ωαA  αA cosωt (43)
X
N
Ft  F^ 0  F^ cos ωnt  F^ sin ωnt or Ft
n1
where α0 is the mean angle of attack, αA the amplitude of angle of
attack, and ω the oscillation frequency. By replacing Δα and Δα_
X
kN∕2–1
terms in Eq. (29) by Eqs. (42) and (43), Eq. (29) is rewritten as
 F^ k eikt (38)
k−N∕2
∂Cx c ∂Cx c ∂Cx
ΔCx  Δα  α_  q (44)
∂α 2V ∂α_ 2V ∂q
where R is a residual with a state vector of W and a flux term of F.
Then,  
∂Cx cω ∂Cx ∂C
 α sinωt   α cosωt (45)
ikW^ k  F^ k  0 (39) ∂α 2V ∂α_ ∂q A

The motion of the forced oscillation is shown in Fig. 12 with


Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

−1
oscillation frequency of 3.24 rad∕s, α0 of 0 deg, and αA of 4.17 deg. A
ωDW ts  Fts  0; where D  T MT (40) hysteresis curve of lift coefficient and surface pressure contour at
maximum and minimum pitch angles are shown in Figs. 12b and 12c,
respectively.
∂W ts For yawing and rolling moment coefficients, the chord length c
 ωDW ts  Fts  0 (41)
∂τ should be replaced by the wing span length b.

The details of the spectral derivative matrix of D are omitted and ΔCx  αA C1 cosωt  αA C2 sinωt (46)
explained in [32–34]. A final form of the governing equations,
Eq. (41), can be solved by steady-state time integration methods for On the other hand, if the value of ΔCx is available by wind tunnel
the unsteady flow simulations. This can lead to considerable experiment or the CFD computation, it can be written in Eq. (46) with

Fig. 12 Motion of forced oscillation and corresponding aerodynamic loads.


PARK ET AL. 2157

first Fourier coefficients of C1 and C2 . The unsteady CFD calculation mesh of the model with the AMT surface deflected by 30 deg, and
of the time-spectral method is ideal in getting the values of C1 and C2 Fig. 14 shows the surface mesh of the model with the DLEF deflected
by a solution approximation of a discrete Fourier series, and the by 30 deg. Figure 15 shows the pressure contour of the ICE 101
unsteady dynamic derivative terms can be obtained by comparing configuration with the left-hand-side AMT surface deflected by
Eqs. (45) and (46), 30 deg at various angles of attack (−2 deg to 10 deg). Figures 16 and
  17 show pressure contours and the flow streamline with the DLEF
cω ∂Cx ∂Cx surface deflected by 30 deg. As shown in the figures, the deflection of
C1   (47)
2V ∂α_ ∂q the control surfaces affects flows and pressure distribution of the
aircraft significantly, and causes a large difference in the stability
characteristics of the aircraft and produces nonzero terms in the
∂Cx control matrix system.
C2  (48)
∂α Figure 18 shows the comparison of lift coefficient from wind
tunnel test with varying DLEF angles (0 and 30 deg). The DLEF
Because it is assumed that there is no unsteady irregular flow such tends to reduce lift force at low angles of attack. However, It increases
as a gust, the derivative with respect to the angular velocity and the lift force at high angles of attack and can delay a stall condition. In
derivative with respect to the angular velocity of angle of attack is the Fig. 18, the stall is delayed by about 5 deg. In addition, if the DLEF is
same (∂Cx ∕∂α_  ∂Cx ∕∂q). With known values of C1 and C2 , the deflected asymmetrically with different deflection angles between
values of ∂Cx ∕∂α and ∂Cx ∕∂α_ Cx ∕∂q  V∕cωC1  are easily left and right sides, it assists other control surfaces (AMT, for
calculated. example) to control yawing and rolling moments through the
stabilization of flows around the aircraft. The DLEF is most effective
D. Computation of Control Derivatives at operations with a high angle of attack.
The control derivatives represent sensitivities of aerodynamic
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

forces and moments with respect to the deflection of control surfaces. E. Numerical Accuracy of the Calculation
For example, when the ICE 101 configuration has three control In this paper, VF aerodynamic analysis methods are used for the
surfaces of AMT, DLEF, and SSD, then the control derivatives of the calculation of aerodynamic and control derivatives. Panel method is
normal force coefficients are used as the low-fidelity method, and SU2 code is used for Euler and
∂CN ∂CN ∂CN RANS methods, which are used as medium- and high-fidelity,
; ; and (49) respectively.
∂δAMT ∂δDLEF ∂δSSD For Euler and RANS methods, ROE’s second-order scheme is
used with the Venkatakrishnan limiter as the spatial discretization
Because of a small number of control inputs, the derivative values scheme. An unstructured grid is generated for both calculations.
can be easily calculated by an FDM. Figure 13 shows the surface Code verification of SU2 with a turbulence model is done by Palacios
et al. [35]. In [35], the second-order accuracy of the high-fidelity CFD
code is guaranteed. A grid convergence test has been performed with
several grids with various number of volume cells, and the calculation
is converged to 10−8 in the residual of continuity. Therefore, it is
assumed that the numerical error of CFD calculation is negligible.
For the panel method, VortexJE is used. The source doublet
distribution is set as constant during the calculation. Therefore, the
result is assumed to have zeroth-order accuracy. The code is validated
with XFLR5 in [17]. Grid convergence tests have been done with
several grids with a various number of volume cells.
For the calculation of derivatives, adjoint method and FDM have
been used. The adjoint method is implemented in SU2, and it follows
second-order accuracy. The validation of the code is performed in
[24,25]. The adjoint calculation is converged until the iterative error
of residual is less than 10−8 . Therefore, the numerical error of the
adjoint method is negligible. To minimize the spatial error in the
central difference method, a step size test has been performed.
Fig. 13 Surface mesh for Euler computation (with the left AMT Figure 19 shows the step size test result for the aerodynamic
deflected by 30 deg). derivatives with respect to the angle of attack.

Fig. 14 Surface mesh for Euler computation (with the DLEF deflected by 30 deg).
2158 PARK ET AL.
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Fig. 15 Pressure contours on the surface with the left-hand-side AMT deflected by 30 deg (Mach  0.033).

IV. Validation and Stability Analysis


First, reconstruction of the ICE 101 geometry is carried out for the
CFD computation. Validation of the numerical results is conducted
using existing wind tunnel data. System matrices for stability analysis
in trim, climbing, and side slip conditions are constructed by the CFD
computation results, and the eigenvalue method, introduced in
Sec. II, is applied to check the stability characteristics of the ICE 101
configuration.

A. Geometry Reconstruction of the ICE 101


For the CFD computation, a watertight solid or surface CAD
model of the ICE 101 configuration is required; however, only raw
data of the CAD model with a cloud of points with connectivities to
existing lines were available as shown in Fig. 20. Thus, the entire
geometry was reconstructed using polynomial regression to smooth
Fig. 16 Comparison of pressure contours of baseline model and the out lines from points and create surface patches. This inevitably
model with the DLEF deflected by 30 deg (Mach  0.033, causes some discrepancies in the geometries used for wind tunnel
AoA  20 deg). experiment and the CFD computation.

Fig. 17 Comparison of streamlines between the baseline and the one with DLEF deflected (Mach  0.033, AoA  20 deg.)
PARK ET AL. 2159

Fig. 20 Original ICE 101 model in an AutoCAD format.

B  AT A−1 AT Y (53)


Fig. 18 Lift coefficients with and without DLEF (from wind tunnel
experiments [8,9]).
With n number of polynomials corresponding to (xj , yj ) and slope
conditions at the end points of a given curve, a system of n  2 ×
For the reconstruction, points that join neighboring curves, defined p  1 linear equations is constructed as in Eq. (52). If we define a
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

as (xjf , yjf ) where j  1; 2; : : : ; n, are set as control points to create left-hand-side vector as Y, a vector of unknowns as B, and a matrix as
smooth curves with given coordinate values and the slopes. All other A, then the coefficients of the polynomials are obtained by the
points, defined as (xi , yi ), where i  1; 2; : : : ; m, are fitted through relation of B  AT A−1 AT Y. The order of the polynomial equation,
associated curves. The pth-order polynomial is written as p, is decided by minimizing the sum of errors in the approximate
values of yi , y10 , and yn0 . The final form of the ICE 101 configuration is
y  Px  Rxβ0  β1 x  β2 x2  · · · βp xp  (50) shown in Fig. 21. Compared with the wind tunnel data shown in
Fig. 1, the inlet shape was closed with smooth fairing surface, rather
than by a blunt blockage for the wind tunnel experiment, and the sting
on the nozzle end was removed for the CFD model.
m 
X m  x−x
Y 
kf
where Px  yjf ; B. Wind Tunnel Experiments
j1 k1;k≠j
xjf − xkf
The subsonic aerodynamic research laboratory (SARL) wind
Y
m tunnel facility, located in Wright–Patterson Air Force Base, Ohio, is a
Rx  x − xjf  (51) low-speed wind tunnel with a 7 0 × 10 0 test section. The experiments
j1 are performed with a 1∕18 th-scaled high-speed model with the flow

8 9 2 38 9
>
>
> y 1 − Px 1  >
>
> Rx1  Rx1 x1 ··· Rx1 xp1 >
>
> β 0 >
>
>
>
> >
> 6 Rx2 xp2 7>> >
>
>
> y − Px 
2 > > 6 Rx2  Rx2 x2 ··· 7>> β 1 >
>
>
<
2
>
= 6 7> >
.. 6 .. .. .. .. 7< β 2 =
. 6
6
. . . . 7
7> .. >. (52)
>
> y − Px  >
> 6 Rx  Rx n xn ··· Rxn xpn 7> >
>
> n n >
> 6 0 n p−1 7>
> >
>
>
> y 0 − P x1  >
0
> 4 R x  R 0 x x  Rx  ··· R x1 x1  Rx1 x1 5>
0 p
> β >
>
>
> > > p−1 >
: y 0 − P 0 x  > > >
1 1 1 1 1
; R 0 xn  R 0 xn xn  Rxn  ··· 0 p p−1 : β
R xn xn  Rxn xn p
;
n n

Fig. 19 Step size test for FDM.


2160 PARK ET AL.

Fig. 21 Reconstructed ICE 101 model.

condition summarized in Table 3 [9]. Force and moment coefficients angles of attack at zero side slip angle (β  0). Both the potential
and static derivatives as well as their steady dynamic derivatives were flow solutions and the RANS simulations are carried out with the
available for the comparison with the CFD computation result. results compared with the wind tunnel data. The potential flow
The steady and unsteady dynamic derivatives were available by solutions from the VortexJE are corrected for viscosity consideration
experiments carried out in an AFRL vertical wind tunnel, which was introduced in Eqs. (21–24). Normal forces along z directions are
built in 1940s for an aircraft free-flight spin test and has a 12-ft- shown in Fig. 22a and show good agreement, but contain slight
diameter section with multi-axis test (MAT) rig through which discrepancies compared with the wind tunnel data. This may be partly
measurement of rotary balance and forced oscillations is possible. because the geometry of ICE-101 configuration has been simplified
The flow conditions used for the validation with the CFD results are for the CFD calculation, explained in a previous section. In Fig. 22b,
summarized in Table 4 [8]. the axial force coefficients along the x direction are plotted. The
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Finally, the control derivatives are validated against data from RANS computation results show reasonably good agreements with
Bihrle Applied Research LAMP (large-amplitude multiple-purpose) wind tunnel data. Although the results from the panel computation
10-ft vertical wind tunnel [9], located in Neuberg Donau, Germany. contain considerable differences from experiments (not shown here),
The facility has a rotary balance equipment, and the flow conditions the results with the drag correction for viscosity shows good
are summarized in Table 5. agreement with the experimental data. However, it should be noted
that the cost to develop the drag correction model for the panel
C. Validation of Aerodynamic Forces and Moments computation involves at least one wind tunnel experiment at zero
angle of attack (α  0), and the accuracy of the panel method alone
Comparisons of the aerodynamic coefficients of normal and axial
without correction still is not reliable for the prediction of the stability
forces as well as pitching moment are shown in Fig. 22 with varying
characteristics.

Table 3 Flow conditions for the D. Validation of Stability and Control Derivatives
SARL wind tunnel test Steady static derivatives computed by the RANS and the panel
Parameter Value method are shown in Fig. 23 and compared with the wind tunnel data.
Mach number 0.3
Normal force derivatives with respect to varying angles of attack at
Dynamic pressure 5745.63 Pa β  0 deg (∂CN ∕∂αjβ0 ), and rolling moment derivatives with
Re number 6.23360 × 106 ∕m respect to varying angle of attack at the side slope angle of β  5 deg
Freestream pressure 92,100.5 Pa (∂Cl ∕∂αjβ5 ) are compared. In Fig. 23b, as the order of magnitude
Freestream density 1.10557 kg∕m3 of derivatives is very small, errors, either physical or numerical, can
Freestream temperature 287.371 K be easily reflected in the results. No error or calibration bounds were
Freestream 101.851 m∕s available for the wind tunnel data, and a rigorous analysis on the
accuracy of the CFD computation results against wind tunnel data is
difficult to make.
Steady dynamic derivatives of a pitching moment coefficient with
Table 4 Test flow conditions for 12-ft
a varying pitch rate through the rotating motion of the ICE 101 in the
vertical wind tunnel experiment in AFRL AFRL wind tunnel are compared in Fig. 24 with the results from the
potential flow solver and the RANS computation. Calibrated error
Parameter Value bounds in the wind tunnel data were available as shown in Fig. 24a.
Mach number 0.026 Computational results by the RANS solver are within the margin,
Dynamic pressure 47.8802 Pa showing reasonable agreements with experimental data. Pitching
Re number 5.74147 × 105 ∕m moment derivatives with respect to pitch angular rate are obtained by
Freestream pressure 97,516.8 Pa computing slopes from Fig. 24a and plotted in Fig. 24b.
Freestream density 1.18672 kg∕m3
Freestream temperature 286.26 K
Unsteady dynamic derivatives are computed by the time-spectral
Freestream velocity 8.8362 m∕s method with a forced oscillation motion. A total of 11 harmonics are
selected through a separate study on the convergence and accuracy
with respect to varying number of harmonics, and the results are
compared with wind tunnel experimental data in Fig. 25. The circle
Table 5 Test flow conditions for 10-ft symbols in black and diamond symbols on a solid line in red indicate
vertical wind tunnel experiment in AFRL the experimental data and the HBM results, respectively. Figure 25
shows good agreements in normal force coefficients. The pitching
Parameter Value moment results, however, contain considerable deviations at a higher
Mach number 0.033 pitching rate. One possible reason is the geometric difference of the
Dynamic pressure 72.8204 Pa test model between computation and experiment and that the
Re number 7.51312 × 105 ∕m discrepancy affects flow more dramatically at a higher pitch rate
Freestream pressure 93,956.2 Pa under an oscillation motion.
Freestream density 1.19301 kg∕m3
Freestream temperature 274.36 K
Finally, the control input derivatives are computed with the FDM
Freestream velocity 10.973 m∕s with a baseline shape of the DLEF as a reference. Effects of the
DLEF are shown in Fig. 26 for lift coefficient and pitching moment
PARK ET AL. 2161

Fig. 22 Coefficients of normal and axial forces as well as moment.


Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Fig. 23 Steady static derivatives with a Mach number of 0.3.

Fig. 24 Steady dynamic derivatives with respect to varying pitching rate.

about y axis when the DLEF in both sides are deflected by 30 deg. and without the DLEF is plotted in Fig. 26e. The trend of lowering
Wind tunnel experimental data are available from [9] with 1∕13 th the lift coefficient slightly at low angles of attack and increasing it at
model at the rotary balance wind tunnel facility. A direct higher angles of attack, shown in Fig. 18 from the experiment, is
comparison of the lift coefficient by the RANS computation with similarly shown here.
2162 PARK ET AL.

Fig. 25 Coefficients of normal force and pitching moment.


Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

Fig. 26 Coefficients of lift force and pitching moment with and without the DLEF (Mach  0.03).

Figure 27 shows the comparisons among different computation that the DLEF improves the aerodynamic and stability performance
methods. The potential flow solver cannot predict stall and flow of the aircraft in high angles of attack.
separation, and shows a monotonically changing slope through low
to high angles of attack. However, RANS results in a blue line show a
sudden jump in the slope around the stall angle. For a normal force V. Stability Analysis for ICE 101 Configuration
coefficient, the derivative value is negative before 20 deg, which The stability analysis of ICE-101 configuration on trim, climbing,
means that as the DLEF is deflected, the normal force on the aircraft and side slip conditions is performed using the CFD computation
decreases. However, if the angle of attack is higher than 20 deg, the results. As no feedback control is considered, stability characteristics
value suddenly changed positive, which means that as DLEF is for baseline configuration without control surface deflection are
deflected, the normal force on the aircraft increases. The same trend is analyzed. System matrices shown in Eqs. (10) and (11) are
found on pitching moment coefficient in Fig. 27c, which indicates constructed for the eigenvalue analysis method.
PARK ET AL. 2163

Fig. 27 Comparison of control derivatives of ICE configuration (with the DLEF deflected by 30 deg).

A. Trim Condition aerodynamic coefficients through the RANS simulation and the
The condition for trim is set to be at a Mach number of 0.3 and the adjoint solution method. The results are represented in Tables 9 and
angle of attack of 1.63 deg. The values of Cm L0 and ∂Cm ∕∂αtrim
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

10, and the values are substituted into the system matrix of Eq. (10). A
are computed by both the potential flow solver and the RANS corresponding damping ratio is 0.4981 with the damping frequency
computations, and shown in Table 6. Based on the stability criteria of 1.5362∕s for the short-period mode. The aircraft is stable, but when
introduced in Sec. II, the aircraft is stable. compared with the trim condition, the damping frequency increased
For the longitudinal dynamic stability, aerodynamic coefficients of considerably at the high angle of attack of 9.97 deg. This means that
CA , CN , and Cm , and their derivative terms are calculated by the the aircraft will vibrate faster for stability recovery. For the phugoid
RANS simulation, and the adjoint method, respectively, and shown mode, the aircraft still diverges with the damping frequency of
in Tables 7 and 8. Substituting the values in the system matrix on 0.0767∕s. This value also increases when compared with the trim
Eq. (10), the eigenvalues of the matrix are calculated for the stability condition, but the magnitude is still small enough for the aircraft to be
characteristics. For a short period mode, the aircraft has a damping under control for the stability with appropriate control inputs.
ratio of 0.4852 and a damping frequency of 0.4988∕s, which implies
a good stability. For a phugoid mode, the aircraft appears to be C. Side Slip Condition
unstable and diverges with a damping frequency 0.0098∕s. However, The side slip condition is set to be at a Mach number of 0.3, the angle
low damping frequency for the diverging phugoid mode indicates of attack of 3.76 deg , and the angle of side slip of 5 deg . Lateral and
that the aircraft can recover dynamic stability easily with any control directional dynamic stabilities are evaluated with Eq. (11). With the
inputs before it loses the stability entirely. derivatives values (Table 11) substituted in the system matrix,
eigenvalues are computed and the characteristics of Dutch roll, roll, and
B. Climb Condition spiral modes are determined. The Dutch roll mode showed stability with
The climbing condition without acceleration at a Mach number of a negative eigenvalue (−0.7487). The spiral mode showed the
0.3 and the angle of attack of 9.97 deg is studied. The longitudinal convergence of the motion with a small decay ratio of 0.1709 and a high
dynamic stability is evaluated by computing derivatives of decay frequency of 0.2199∕s, implying that the aircraft is stable with a
slight margin. However, for the roll mode, the aircraft diverged with a
decay frequency of 0.4803∕s, which requires the stability recovery
Table 6 Static stability for the trim
condition
Table 9 Coefficients for normal
Analysis method Panel RANS and axial forces as well as pitching
moment for the climb condition
Cm L0 0.000490843 0.000632694
∂Cm ∕∂αtrim −0.0003982 −0.02043265 CA CN Cm
−0.0184432 0.383961 −0.0103034

Table 7 Coefficients of axial and Table 10 Derivative coefficients for dynamic


normal forces as well as pitching stability for the climb condition
moment
u w α
CA CN Cm
CA −3.99949 × 10−8 −1.05116 × 10−6 −0.0034368
0.00247919 0.0419428 −0.00056505 CN 3.91461 × 10−7 −9.43969 × 10−8 0.0470619
Cm −1.76102 × 10−8 −1.04076 × 10−7 −0.0015929

Table 8 Dynamic stability derivatives for the Table 11 Derivatives for dynamic
trim condition stability for the side slip condition
u w α v β
CA −2.5764 × 10−8 −1.0230 × 10−8 −0.00018 Cl 5.00338 × 10−6 5.95652 × 10−4
CN −4.2661 × 10−7 2.57979 × 10−6 −0.038146 CS −3.71314 × 10−7 −7.40447 × 10−5
Cm −3.1138 × 10−9 −1.00129 × 10−7 −0.02043 Cn 3.20501 × 10−6 3.4841 × 10−4
2164 PARK ET AL.

during a maneuver through a feedback control. The tailless aircraft in [13] Kolk, W. R., Modern Flight Dynamics, Prentice-Hall, Englewood Cliffs,
side slip motion need additional control surfaces for the stability and NJ, 1961, Chaps. 1–7.
maneuverability, and control surfaces introduced in Fig. 2 can be used [14] Boiffier, J. L., The Dynamics of Flight: The Equations, Wiley, New
for the control. York, 1998.
[15] Park, J., Choi, J., Ocheltree, C. L., Jo, Y., and Choi, S., “Stability
Derivative Computation of Tailless Aircraft Using Variable-Fidelity
VI. Conclusions Aerodynamic Analysis,” AIAA Paper 2015-3387, June 2015.
[16] Park, J., Jo, Y., Choi, J., and Choi, S., “Stability Derivative Computation
In this paper, the equations of motion and stability criteria for of Tailless Aircraft Using Variable-Fidelity Aerodynamic Analysis for
tailless aircraft were derived. For the analysis of control surface Control Performance Analysis,” AIAA Paper 2016-2024, Jan. 2016.
effects, the control input system for the tailless aircraft was added to [17] Baayen, J. H., “Vortexje—An Open-Source Panel Method for Co-
the system of the equations of motion. For the stability and control Simulation,” ARXIV, March 2013, http://arxiv.org/abs/1210.6956v3.
analysis, force and moment coefficients and their derivatives (steady [18] Katz, J., and Plotkin, A., Low-Speed Aerodynamics, 2nd ed., Cambridge
static derivative, steady dynamic derivative, unsteady dynamic Univ. Press, 2000, Chap. 15.
derivative, and control input derivative) should be calculated in an [19] Park, M. A., Green, L. L., Montgomery, R. C., and Raney, D. L.,
efficient and accurate way. These values were calculated by the “Determination of Stability and Control Derivatives Using Computa-
variable-fidelity analysis methods of the linear potential, Euler, and tional Fluid Dynamics and Automatic Differentiation,” AIAA Paper
1999-3136, June 1999.
Reynolds-averaged Navier–Stokes (RANS) flow solvers. For the [20] Raney, D. L., Montgomery, R. C., Green, L. L., and Park, M. A., “Flight
calculation of derivatives, the adjoint method and the finite difference Control Using Distributed Shape-Change Effector Arrays,” AIAA
method are used, and the results were compared with wind tunnel Paper 2000-1560, April 2000.
data for their efficiency (computational cost), and accuracy. [21] Scott, M. A., Montgomery, R. C., and Weston, R. P., “Subsonic
As discussed in the validation of the computational fluid dynamics Maneuvering Effectiveness of High Performance Aircraft Which
result, in the trim condition, the panel method can provide estimations in Employ Quasi-Static Shape Change Devices,” NASA Langley Research
good agreement with other methods at low angles of attack. However, Center NASA/TM-1988-207570, Hampton, VA, March 1998.
Downloaded by 2.53.17.18 on July 3, 2022 | http://arc.aiaa.org | DOI: 10.2514/1.C034052

the result of the panel method showed some discrepancy compared with [22] Stanford University Unstructured (SU2), su2.stanford.edu.
the other methods as angles of attack increased due to the presence of the [23] Palacios, F., Colonno, M. R., Aranake, A. C., Campos, A., Copeland, S.
R., Economon, T. D., Lonkar, A. K., Lukaczyk, T. W., Taylor, T. W. R.,
flow separation, while RANS results were in good agreement along with and Alonso, J. J., “Stanford University Unstructured (SU2): An Open-
the experimental data. Also, considering the numerical accuracy of the Source Integrated Computational Environment for Multi-Physic
calculation, RANS calculation provides more reliable data compared Simulation and Design,” AIAA Paper 2013-0387, Jan. 2013.
with the panel method. With the operation of the control surfaces, both [24] Economon, T. D., Palacios, F., and Alonso, J. J., “An Unsteady
panel method and Euler method provide poor agreement compared to Continuous Adjoint Approach for Aerodynamic Design on Dynamic
the experimental data. The RANS method was the only trustworthy Meshes,” AIAA Paper 2014-2300, June 2014.
method, because it could predict the flow separation with better [25] Jameson, A., and Shankaran, S., “Continuous Adjoint Method for
accuracy. In this paper, RANS simulations were considered the high- Unstructured Grids,” AIAA Journal, Vol. 46, No. 5, May 2008,
fidelity tool that provided estimations similar to experimental data and pp. 1226–1239.
doi:10.2514/1.25362
used for the static and dynamic stability analysis of tailless aircraft. [26] Lee, H., Kim, B., and Lee, S., “Computational Study of the Stability
Stability analyses for flight conditions of trim, climb, and side slip Derivatives for the Standard Dynamic Model,” AIAA Paper 2013-2658,
were performed, and it was concluded that, for certain modes, the June 2013.
aircraft was unstable, especially in the side slip condition. However, [27] Ye, C., and Ma, D., “Aircraft Steady Dynamic Derivatives Calculation
this can be easily controlled by the control effectors that were Method,” Proceedings of 2012 International Conference on Modelling,
introduced to the innovative control effector 101 configuration. Identification and Control, Wuhan, China, June 2012, pp. 855–860.
[28] Zhang, L., Deng, X., and Zhang, H., “Reviews of Moving Grid
Generation Techniques and Numerical Methods for Unsteady Flow,”
References Advances in Mechanics, Vol. 40, No. 4, July 2010, pp. 424–447.
[1] Cummings, R. M., and Schutte, A., “An Integrated Computational/ [29] Da Ronch, A., Vallespin, D., Ghoreyshi, M., and Badcock, K. J.,
Experimental Approach to UCAV Stability & Control Estimation: “Evaluation of Dynamic Derivatives Using Computational Fluid
Overview of NATO RTO AVT-161,” AIAA Paper 2010-4392, June 2010. Dynamics,” AIAA Journal, Vol. 50, No. 2, Feb. 2012, pp. 470–484.
[2] Roy, J. L., and Morgand, S., “SACCON CFD Static and Dynamic doi:10.2514/1.J051304
Derivatives Using elsA,” AIAA Paper 2010-4562, June 2010. [30] Da Ronch, A., Ghoreyshi, M., Badcock, K. J., Gortz, S., Widhalm, M.,
[3] Frink, N. T., “Strategy for Dynamic CFD Simulations on SACCON Dwight, R. P., and Campobasso, M. S., “Linear Frequency Domain and
Configuration,” AIAA Paper 2010-4559, June 2010. Harmonic Balance Predictions of Dynamic Derivatives,” Journal of
[4] Tormalm, M., and Schmidt, S., “Computational Study of Static and Aircraft, Vol. 50, No. 3, 2013, pp. 694–707.
Dynamic Vortical Flow over the Delta Wing SACCON Configuration doi:10.2514/1.C031674
Using the FOI Flow Solver Edge,” AIAA Paper 2010-4561, June 2010. [31] Hall, K. C., Thomas, J. P., and Clark, W. S., “Computation of Unsteady
[5] Cummings, R. M., Schutte, A., and Hubner, A., “Overview of Stability Nonlinear Flows in Cascades Using a Harmonic Balance Technique,”
and Control Estimation Methods from NATO STO Task Group AVT- AIAA Journal, Vol. 40, No. 5, May 2002, pp. 879–886.
201,” AIAA Paper 2013-0968, Jan. 2013. doi:10.2514/2.1754
[6] Irving, J. P., Vicroy, D. D., Farcy, D., and Rizzi, A., “Development of an [32] Im, D., Choi, S., and Kwon, J., “Unsteady Rotor Flow Analysis Using a
Aerodynamic Simulation Model of a Generic Configuration for S & C Diagonally Implicit Harmonic Balance Method and Overset Mesh
Analysis,” AIAA Paper 2014-2393, June 2014. Topology,” International Journal of Computational Fluid Dynamics,
[7] Roetman, E. L., Northcraft, S. A., and Dawdy, J. R., “Innovative Control Vol. 29, No. 1, March 2015, pp. 82–99.
Effector (ICE),” WL-TR-96-3074, Wright-Patterson Airforce Base, OH, doi:10.1080/10618562.2015.1015525
1996. [33] Im, D., Choi, S., and McClure, J., “A Mapped Chebyshev Pseudospectral
[8] Dorsett, K. M., and Mehl, D. R., “Innovative Control Effector (ICE),” Method for Unsteady Flow Analysis,” AIAA Journal, Vol. 53, No. 12,
WL-TR-96-3043, Wright-Patterson Airforce Base, OH, , 1996. Dec. 2016, pp. 3805–3820.
[9] Dorsett, K. M., Rears, S. P., and Houlden, H. P., “Innovative Control doi:10.2514/1.J054081
Effectors (ICE) Phase II,” WL-TR-97-3059, Wright-Patterson Airforce [34] Im, D., Choi, S., and Park, S., “Numerical Analysis of Synthetic Jet
Base, OH, 1996. Flows Using a Diagonally Implicit Harmonic Balance Method with
[10] Gillard, W. J., “Innovative Control Effectors Dynamics Wind Tunnel Preconditioning,” Journal of Computers and Fluids, Vol. 147,
Test Report,” AFRL VA-WP-TR-1998-3043, Wright–Patterson Air April 2017, pp. 12–24.
Force Base, OH, 1998. [35] Palacios, F., Economon, T. D., Aranake, A. C., Copeland, S. R., Lonkar,
[11] Etkin, B., Dynamics of Flight: Stability and Control, Wiley, New York, A. K., Lukaczyk, T. W., Manosalvas, D. E., Naik, K. R., Padron, A. S.,
1959, Chaps. 2–5. Tracey, B., Variyar, A., and Alonso, J. J., “Stanford University
[12] Etkin, B., Dynamics of Atmospheric Flight, Wiley, New York, 1972, Unstructured (SU2): Open-Source Analysis and Design Technology for
Chaps. 5–9. Turbulent Flows,” AIAA Paper 2014-0243, Jan. 2014.

You might also like