Aeroelastic Formulation Based on

CFD Applications

John Kim
Boeing Commercial Aircraft
Seattle, WA

1
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

2
 Issue with Current CFD ROMs

“For highly complex problems where a large number of

structural modes are required, the computational cost of
generating modal ROM (reduced order model) responses
will increase and surpass the cost of standard simulations”

D.M. Schuster, D.D. Liu and L.J. Huttsell, Journal of

Aircraft, Vol. 40, No. 5, September-October 2003

3
Start
Full-order analysis
Finish

1st (tf =108 hrs) nth analysis (tf)

Reduced-order analysis

ROM construction (tr1 =1250 hrs) 1st nth analysis (tr = 5 min)

Start Finish
tr << tf BUT tr1 > tf
4
5
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

6
 Objective

 Based on time-accurate CFD codes, develop statically

nonlinear, dynamically linearized aerodynamic and
aeroelastic ROMs in state-space form for accurate and rapid
aeroelastic analysis such as flutter and dynamic loads
calculation at an affordable cost by avoiding the mode-by-
mode approach of the traditional reduction methods.

7
8
 New Approaches

 Use of a Single Input instead of Multiple Inputs:

The Single-Composite-Input/Eigensystem Realization
Algorithm (SCI/ERA) provides an effective and economic
tool to handle the multiple structural mode shapes during
system identification.
 Direct Calculation of Aeroelastic Roots in Time Domain:
The Discrete-Time Eigen Analsyis (DTEA) coupled with
Mode Tracking produces V-g plot in which aeroelastic
damping & frequency are calculated directly in time domain.
Use of the typical p-k flutter analysis is avoided.

9
 Previous Work on Affordable Model
Reduction
 Taehyoun Kim and John Bussoletti (2001 SDM Conf.)
Introduced the Frequency-Domain Karhunen-Loeve/Single-
Composite-Input (FDKL/SCI) procedure for application to
Boeing’s TRANAIR code.

“… (traditional methods) took a total of 110 wall-clock

execution hours while the proposed methods took only 37
hours on a typical workstation, reducing the computation by
66 %’’

T. Kim and J.E. Bussoletti, AIAA Paper 2001-1525

10
 (continued)

 Taehyoun Kim (SDM Conf. 2001)

Used the SCI in both time and frequency domains with the
Karhunen-Loeve procedure for application to Vortex Lattice
aerodynamic model.

“… The computational time required for advanced unsteady

CFD codes is almost prohibitive for the multiple response
calculations. The proposed reduction scheme may be very
useful for an efficient model reduction of such models’’

T. Kim, AIAA Paper 2001-1380

11
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

12
 Highlights of SCI/ERA Procedure

 Takes system response due to a simultaneous excitation of all

the inputs, performs singular-value-decomposition (SVD) on
the output data set, and generates (A, B, C, D) system
matrices of a minimum size.
 The total number of time samples needed is much less than
that required for the traditional ERA.
 Any arbitrary input signals can be used for the SCI provided
that they are statistically independent.

Reference:
“An Efficient Model Reduction Method for Linear Dynamic
Systems with Multiple Inputs”, T. Kim, AIAA Paper 2004-
2036

13
Boeing Proprietary

14
15
Boeing Proprietary

16
17
18
Input 1 Input 2 Input i Input Ni

Markov parameters for M steps

Pulse/ERA

A, B, C, D

Schematic of Pulse/ERA Process

19
Redundant Solutions Made by Serial Excitations

10

10 10

Total # of solutions produced = 10+10+10=30

Net # of solutions < 30
20
Boeing Proprietary

21
Boeing Proprietary

22
23
24
 SCI

SCI response for M steps

+ Markov parameters for

- the first 2 steps

SCI/ERA

A, B, C, D

Schematic of SCI/ERA Process

25
26
Boeing Proprietary

27
28
29
30
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

31
Note: CFD equations are described in non-dimensional
time, τ = Vt/b with ∆τ= fixed.

32
33
 Input Signals for Simultaneous Excitation

Random signals based SCI (RSCI):

• Statistically independent random signals. In general, not
recommended for CFD applications.
Filtered signals based SCI (FSCI):
• Filtered signals with a low frequency bandwidth.
Step signals based SCI (SSCI):
• A series of step signals sequentially delayed in time.
Pulse signals based SCI (PSCI):
• A series of pulse signals sequentially delayed in time.
Not as robust and accurate as SSCI.

34
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

35
Note: Structure equations are described in
real time, t = bτ/V. ∆t=b∆τ/V is
adjusted in the discretization.

36
 Coupling of CFD ROM with Structure

 When coupling CFD ROM (∆τ dependent) with structure (∆t

dependent), allow both air density & speed to be independent
gain parameters between the two models. That is, leave the V
dependency in the discretized structure.


Structure (V)
(FEM)

ρ&V

Aerodynamics
(CFD ROM)

37
Boeing Proprietary

38
39
 Limitations of Aeroelastic ROMs

At the fixed Mach number,

1. Cannot account for the changes in the free stream
Reynolds number, temperature and pressure.
2. Cannot account for the change in the static aeroelastic
deformation from one dynamic pressure to another.

Assume these effects are small or negligible.

40
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

41
Vortex lattice grids for rectangular semi-span

42
43
CPU Time of ERAs vs. no. of inputs

44
ROM size vs. no. of inputs

45
46
47
48
49
50
51
52
53
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

54
 CFL3D Example Model
 Twin-Engine-Transport-Flutter-Model (TETFM) modeled by
CFL3D and FEM has been studied for the SCI/ERA method.
 For simplicity, Wing-Pencil-Nacelle (WPN) aerodynamic
model was used.
 The WPN model has 700,000 cells and 30 blocks, 10
structural modes.
 Reference flow condition:
fluid = freon with purity 0.8 Re = 2.735E+06
M = 0.831 Static pressure = 380.60 psf
a = -2.05 degree Speed of sound = 499.00 fps
q = 148.732 psf V = 414.67 fps

55
The Twin-Engine-Transport-Flutter-Model (TETFM) in the Transonic Dynamic
Tunnel

56
The CFL3D Grids for WPN Model

57
First 4 Structural Mode Shapes for WPN Model

58
 CFL3D Reduction & Flutter Analysis: Steps

1. After the static solution has converged, prepare a SCI using FSCI or SSCI,
sample the unsteady CFL3D GAF solution for about 1,000 time steps.
2. Sample pulse responses for the 10 individual inputs for about 100 steps.
3. Execute SCI/ERA, get (A, B, C, D) matrices.
4. If necessary, perform the second reduction on (A, B, C, D) using
FDKL/SCI method.
5. Discretize the structure with ∆t=b∆τ/V (∆τ fixed, V varying), combine it
with the CFL3D ROM.
6. Execute DTEA/Mode Tracking to generate V-g plot. To best simulate the
WT test, change only the density at the fixed Mach number & the fixed
free stream speed.

NOTE: Max. amplitude in all of inputs is 1.E-03.

59
CFL3D-Based Flutter Process

60
 SCIs and No. of Time Steps
FSCI I. FSCI II. SSCI
318 Hz 159 Hz 1496 Hz
fc
No. of subit. 20 15 15

No. of time 995 1327 995

steps
2nd Reduction No No Yes

ROM size (60 x 60) (68 x 68) (107 x 107)

61
CFL3D CPU Time Required for ERAs

CPU Time Total No. of

(hrs) Time Steps
Pulse/ERA 336 9950

FSCI/ERA I. 79 1805

FSCI/ERA II. 79 2137

SSCI/ERA 64 1805

Saving Factor = 4 ~ 6 for 10 mode shapes

62
Aerodynamic Eigenvalues: FSCI/ERA I. (60 x 60) (ωc =318 Hz)

63
Aerodynamic Eigenvalues: FSCI/ERA II. (68 x 68) (ωc =159 Hz)

64
GAFs: Pulse/ERA vs FSCI/ERA (0 < τ < 38)

Note: Pulse/ERA ROM is (512x512)

65
GAFs: Pulse/ERA vs FSCI/ERA (0 < k < .075)

66
V-g Plot: Pulse/ERA vs FSCI/ERA

67
V-g Plot: Pulse/ERA vs FSCI/ERA

68
Aerodynamic Eigenvalues: SSCI/ERA (107 x 107)

69
GAFs: Pulse/ERA vs. SSCI/ERA (0 < τ < 38)

Note: Pulse/ERA ROM is (512x512)

70
GAFs: Pulse/ERA vs. SSCI/ERA (0 < k < .075)

71
V-g Plot: Pulse/ERA vs. SSCI/ERA

72
V-g Plot: Pulse/ERA vs. SSCI/ERA

73
Time History of Aeroelastic Responses due to IC in
Mode 1. @ q=120 psf

74
Time History of Aeroelastic Responses due to IC in
Mode 1. @ q=152 psf

75
 Table of Contents

1. Challenge with CFD Based Reduced-Order Modeling

2. Objective
3. System ID & Model Reduction Using Single-Composite-
Input/Eigensystem Realization Algorithm (SCI/ERA)
4. Application of SCI/ERA to CFD
5. Coupling of CFD ROM and Structure
6. Example: Rectangular Wing Modeled by Vortex Lattice
7. Example: Twin-Engine-Transport-Flutter-Model (TETFM)
Modeled by CFL3D and FEM
8. Concluding Remarks and Recommendations

76
 Conclusions
 Based on SCI/ERA, have established a new procedure to construct
CFD based reduced-order aerodynamic and aeroelastic models at
affordable cost.
 Compared to current methods, the new method can save computational
cost by a factor of 4-6 for the case of 10 structural modes. The saving
factor will increase as more modes are included.
 The new aeroelastic ROMs can predict flutter instabilities as well as
aeroelastic damping & frequency very accurately, as accurately as
those obtained from Pulse/ERA.
 For CFD applications sharp, rapidly changing input signals are not
recommended since they require many subiterations for convergence.
 Since FSCI is smooth and limited in its frequency bandwidth, it is
possible to generate aerodynamic ROM of a small size using FSCI
without additional reduction. It may also need a smaller number of
subiterations to converge.

77
 Recommendations
 The method should be verified using more structural modes and various
configurations.
 An option to use extra aerodynamic measurements beyond the GAFs must
be explored.
 More research is recommended to investigate different types of SCIs,
especially those with limited frequency bandwidth.
 It is desirable to optimize the number of time steps quantitatively for
completion of the process using a Rank Updating Scheme.

78