Title: Configuration Design of Air Breathing
1 2nd Progress Seminar
22nd August,2003
Configuration Design of Air Breathing Hypersonic
Vehicle using Numerical Optimization
J. Umakant Research Student ( External) Roll No.
01401701
CASDE Dept. of Aerospace Engineering IIT, Bombay
2 Summary of 1st Progress
Seminar September 2002
- Problem Formulation
- Conceived Overall Design process for
Air-Breathing Hypersonic configuration - Disciplinary Interactions
- Parameterization of vehicle and Analysis
Modules - Review of literature of Aerospace Vehicle Design
using MDO - 3 Ph D thesis
- - McQuade Development of CFD based GLA
factors for 2D scramjet vehicle - - Guinta VCRSM of HSCT Wing
- - Old Robust design of SSTO vehicle
- 12 Papers from 1990 onwards
- - Bowcutt (1999) MDO Hypersonic Vehicle
Optimization - - Design synthesis tools for Launch Vehicles
- - Papers related to approximation strategies
- Fore-body optimization using engineering method
with FFSQP optimizer - - two design variables ( fore-body compression
angles) - objective function ma / Cd subject to
constraints on Mintk , L/D , h/l
3 2nd Progress Seminar
I Hypersonic Technology Demonstrator
Vehicle(HSTDV) - Mission -
Vehicle Background II HSTDV Configuration
- Problem Statement -
Parameterization and Trade-Offs -
Engineering Methods for Analysis III
Optimization and Results IV Potential
Improvements in Aerodynamic Prediction code
4HYPERSONIC TECHNOLOGY DEMONSTRATOR
TO DEMONSTRATE AUTONOMOUS SUSTAINED FLIGHT AT
HYPERSONIC SPEED
SCRAMJET TEST
ALTITUDE 32.5 km MACH NO. 6.5 TEST
DURATION 400 s
SCRAMJET MACH NO. 5.5
RAMJET
DUAL MODE TEST
ALTITUDE 20 km MACH NO. 4.5
1 m DIA
5 HSTDV Vehicle Discipline
Interactions
- Integrated Engine and Airframe
- Entire undersurface of the airframe forms part
of the engine - Fore-Body
- pre-compressed air to the intake , aerodynamic
characteristics, volume - After-Body
- thrust, stability characteristics , after-body
volume
Optimizer
XD
f , g
Aerodynamic heating, Detailed Modeling of Intake,
Combustor, Nozzle , Trajectory Optimization
6MDO - Implementation Issues
- Mathematical modeling and Computational Expense
- low fidelity methods computationally cheap
but not sufficiently accurate - high fidelity methods highly accurate but
computationally prohibitive
- Optimization Procedures
- problem formulation
- algorithms for global optimization
- Organizational Complexity
- disciplinary expertise is distributed across
the organization, not available - centrally
- difficulty in data exchange
7Broad Strategy for HSTDV Design using MDO
I Overall Vehicle Design using Engineering
Methods ( low fidelity ) - Sizing,
Aerodynamics, Propulsion and Performance
- Identify important design variables
- Build a multidisciplinary analysis tool
- Calibration factors -
Numerical optimization
II Methods to create Approximate Models for
High Fidelity Analysis - Design and
Analysis of Computer Experiments - Data
fusion ( low fidelity high fidelity )
III Global Optimization Strategies using DACE
suggested in Statistical literature.
IV Methods to take into account uncertainties
in approximate model
8Parameterization of HSTDV Body
W_fac
t_fac
XD ?1, ?2, ?3 , ?n_pl , ?wc , wfac_pl,
tfac_pl,,Hcruise
Wing AR0.6, b 1.6m, ?0.4 Tail AR2.3
, b 1.4m , ?0.4 Airframe thickness t 50mm
Lmid 2.5 m
9Fore-body Parameterization 3 Ramp configuration
10After-body Parameterization
?noz
Max pnoz Sin(?noz) s.t 0.2 pnoz/pne
Lab
Mne 1.5 pne 1.1 atm Lab 1.5m 1-D P-M
relations to estimate pnoz ?noz
17
1.0
pnoz/pne
Pnoz labSin ?noz
0.0
0
40
?noz (deg.)
?noz (deg.)
11Trade-Offs
Parameter Potential Trade
?n_plan higher body width ( volume) vs higher skin friction drag
?1, ?2, ?3 higher f/b height (volume) vs lower intake Mach No., lower Pressure recovery, higher CL CD
lmid higher volume vs higher skin friction drag, higher weight
lab higher a/b volume vs lower nozzle angle ( propulsive force propulsive moment)
w_fac Higher lift, lower trim angle of attack vs higher drag, more space
?w_cant Stability vs higher wave drag
t_fac higher trim deflections vs lower trim deflection, stability
Hcruise higher drag , higher ma vs lower drag , lower ma
12Optimizer
XD
f , g
External Configuration Model
External Compression Model
Forebody length and height
Volume Body Discretization
Mass flow of Air Capture
Aero Model
Mass C.G.
Overall Aero Control data
Adjust Ballast
Thrust Model
Specific Impulse
No
Trim
Yes
Fuel flow rate Thrust Deliverable
Trim deflection , Drag Updated mass
Performance Model
13HYPERSONIC TECHNOLOGY DEMONSTRATOR
-CONFIGURATION
PAYLOAD 400 kg FUEL 250 kg TOTAL WEIGHT
1240 kg
14External Compression Model
M? , ? Input Variables ?1 , ?2 , ?3
Output Fore-body dimensions ( l1
,l2 ,l3 ,h1 ,h2 ,h3 ) Intake Entry
Conditions pintk , Mintk , ma
External Compression Model Oblique shock theory
15Assuming shock on lip condition
16Typical Results from External Compression model
First Ramp Angle 5 deg.
Intake Entry Mach No.
Total Pressure Recovery
Mass flow rate ( Kg/s)
Mass flow rate ( Kg/s)
Calibration factor Mass flow rate based on Euler
CFD calculations is about 30 lower as compared
to the estimate from low fidelity analysis Each
Euler CFD run on a 8node P-III cluster requires
15 hours
17External Configuration Model
Input Variables ?1 , ?2 , ?3 , ?n_plan , ?wc
, wfac_pl , tfac_pl,
Outputs Body discretization
(x,y,z) Wing Tail discretization Internal
Volume Overall Mass (TOGW) Centers of
gravity
External Compression Model (l1,l2,l3
,h1,h2,h3)
External Configuration Model
18External Configuration Model
Body Discretization
- Input Parameters
- Swid_ntip 0.1m
- Lmid 2.5m
- ?noz 20
- a/b 2.0
Lnoz
Lmid
x0
xfb1_stn
xfb2_stn
xfb3_stn
xmid_stn
xnoz_stn
a
b
w
b
h
19External Configuration Model
Internal volume
body_int vol fb1_v fb2_v fb3_v midbd_v
aftbd_v
Airframe Mass
Mass ?s Swet ?s 20 kg / m2
surface area density
bodyaf_m 1.2 (fb1_m fb2_m fb3_m midbd_m
aftbd_m)
20External Configuration Model
Airframe Center of gravity
Xc.g.
Xc.g.
bodyaf_xcg ( fb1_m fb1_xcg fb2_m fb2_xcg
fb3_m fb3_xcg
midbd_m midbd_xcg aftbd_m aftb_xcg) /
bodyaf_m
21External Configuration Model
equip_m nc_m bal_m obc_m ins_m tm_m
tank_m
wing_m baseline wing mass w_fac
TOGW bodyaf_m equip_m eng_m fuel_m
wing_m tail_m
tm_xcg xfb3_stn 0.25(xmid_stn-xfb3_stn)
tm_zcg 0.5zlh_mid
act_xcg xmid_stn act_zcg zlh_mid
to_xcg ( bodyaf_m bodyaf_xcg
equip_mequip_xcg eng_meng_xcg
fuel_m fuel_xcg wing_m wing_xcg
tail_mtail_xcg) / TOGW
22Aerodynamics Model
External Configuration Model Vehicle geometry
definition (x,y,z)
Aerodynamics Model Tangent Cone / Tangent Wedge
Method ( local surface inclination )
Overall CN , Cm , CA Control surface characteristi
cs
23Typical Aerodynamic Characteristics of HSTDV
M 6.5
CN
Cm_xcg
?
?
24L/D
ANGLE OF ATTACK (deg.)
25Fidelity of Analysis
- Calibration Factors (Scale factor)
- Use CFD computations to generate calibration
factors. - Valid within specified move limits
-
CN Xcp/d CA m a(Kg/s)
Tangent Cone Method 2.028 3.911 0.431 8.1
CFD (Euler) 1.657 3.507 0.342 5.59
Zeroth order scale factor 20 15
15 30
Higher order scale factors will be used in future
studies
26Trim Model
Basic Body, Wing Tail Aero characteristics Propu
lsive force moment
Evaluate Static stability
Exit with tail size and updated Mass and
C.G ?trim and ?trim and trim drag
Yes
Statically stable
Adjust Ballast weight
No
27Trim Model
?
M?
N
A
?
N Np W Cos ? T A W Sin ?
W
?
0.25T
0.75T
Np 0.25 T / tan ?noz
?noz
Np
Mp Np (to_xcg - x_noz) 0.25T (to_zcg - z_noz)
0.75T (to_zcg - z_noz)
Mtrim Maero_cg Mp_cg 0
28Thrust Model
External Compression Model ma ,
Mintk
Equivalence ratio 1
Thrust Model
Isp (M, H_cr)
Thrust deliverable
29Performance Analysis
Aerodynamics
Performance - 2DOF trajectory simulation
Range R
Propulsion
Sizing
30 Multi-disciplinary Design Optimization for HSTDV
Problem Statement
- Minimize f(XD) - (Range)/2000
- g1 MI / 4.0 1 lt 0
scramjet considerations - g2 ? / 20.0 1 lt 0
Aero, control and actuation - g3 L / 7.5 1 lt 0
- g4 H / 0.85 1 lt 0
sizing - g5 W / 0.85 1 lt 0
- g6 TOGW / 1325.0 1 lt 0
system - g7 AF / Th deliv 1 lt 0.
Aerodynamics Propulsion -
- ?
Side constraints 3 ? ?1 ? 10 1
? ?2 ? 10 1 ? ?3 ? 10
3 ? ? n_pl ? 6 0 ? ? wc ? 6
0.8 ? wfac_pl ? 1 0.8 ? tfac_pl ? 1.1
30 ? Hcr ? 35
Optimization variables XD ?1, ?2, ?3 , ?n_plan
, ?wc , wfac_pl,tfac_pl,,Hcruise
31Results
?1
?2
?n_pl
?3
Iteration number
Iteration number
32?w_c
?w_fac
?t_fac
H_cr
Iteration number
Iteration number
33 Cruise Range
g1 MI / 4.0 1 lt 0
Iteration number
Iteration number
g2 ? / 20.0 1 lt 0
g3 L / 7.5 1 lt 0
Iteration number
Iteration number
34g4 H / 0.85 1 lt 0
g5 W / 0.85 1 lt 0
Iteration number
Iteration number
g6 TOGW / 1325.0 1 lt 0
g7 AF / Th deliv 1 lt 0.
Iteration number
Iteration number
35Comparison of Initial Configuration Outline with
Optimum configuration
Baseline Optimum
Body Outline
Tail planform
36XD Initial Design Setting Optimum Design Setting
?1 7.55 5.82
?2 3.88 3.64
?3 2.89 4.14
?n_plan 4.50 4.00
?wc 4.80 6.00
wfac_pl 0.80 0.80
tfac_pl 1.10 1.06
Hcr 31.25 31.65
- Optimum design with respect
- to initial design
- 4 increase in dry weight
- 15 increase in fuel volume
- 1.5 decrease in drag
- 17 increase in cruise range
- Physical constraints on
- Mintk , ? and TOGW are active
37Sensitivity of objective function with respect to
design variables at initial design point
?R / ? ?1 -66.02
?R / ? ?2 -89.09
?R / ? ?3 93.75
?R / ? ?n_plan 147.85
?R / ? ?wc 23.75
?R / ? wfac_pl 110.5
?R / ? tfac_pl -13.97
?R / ? Hcr 13.46
Optimum configuration has lower values for ?1 and
?2 as compared to initial design. Decreasing
these variables at the initial design point ,
results in a decrease in the objective function
ie, cruise range However, the inter-play
among the design variables has resulted in a net
improvement in objective function.
? ?R / ? Xi 109.69
38Fidelity of Analysis
- Physics Based Corrections
- Improve the accuracy of the Engg. Methods like
Tangent Cone - through correlation factors generated using CFD
- Globally valid
Equivalent Body for conical flow calculations
Actual Body
39Cone Body (semi-included angle 5 )
40Cone Body (semi-included angle 5 )
?Cp
Angle of attack (deg.)
Error ?Cp Cp (TCM) Cp (CFD) at ? 0
41Global Correction Factor
?Cp ?/Sin ?
?
Cp (corrected) Cp(TCM) - ?Cp
42Further course of Action
Focus on methods to include high fidelity
analysis
- Summary of methods adopted for Aerospace Vehicle
Design - Various strategies have been used to address the
issue of - computational burden associated with high
fidelity analysis - Parametric methods with RSM
- Global Local Approximation
- First Order Approximate Model Management
- Variable Complexity Response Surface Method
- Statistical Literature
- Design and Analysis of Computer Experiments
43Design and Analysis of Computer Experiments
Ref Sacks et.al. 10 Jones et.al. 11
Motivation Given function values Y at sampled
points x , one simple way to create response
surface is through linear regression
- In the above model, the errors are assumed
independent. This assumption - is justified for physical experiments.
- Computer experiments are however,
deterministic. - Lack of random error in computer experiments and
any lack of fit is entirely - due to collection of left out terms in x.
- In DACE model, ?(i) is interpreted as ?(x(i))
ie., errors are correlated.
44Further course of Action
- DACE modeling for ma , CN, Cm and CA
- Use data fusion ( low fidelity high fidelity)
validation - Optimization Strategies
- Robustness of design through error propagation
45Design and Analysis of Computer Experiments
The correlation is high if two points x (i) and x
(j) are close and low when the points are far
apart.
(Jones et.al.)
46Design and Analysis of Computer Experiments
DACE Model
global effect
local effect
(Jones et.al.)
are parameters estimated by maximizing
the likelihood of the sample y ( y(1),, y(n)
)'
47Design and Analysis of Computer Experiments
RSM model using DACE modeling
ri(x) Corr ?(x), ?(x(i)) , i1,.n
RSM model using regression
48Design and Analysis of Computer Experiments
Illustration (Jones et.al.)
DACE response surface
Branin test function Contours
Quadratic surface fit
49Design and Analysis of Computer Experiments
Global Optimization for a 1-D function using DACE
model (Jones et.al.) Expected Improvement
Criteria for selecting additional sample points
50DACE fit for Pitching Moment Data Predictions are
at the sampled points itself
Cm
?
?
- -45 , -35 , -25 , -15 , -5 , 0 , 5 ,
15, 25 35 , 45 - ? 0 , 2 , 4, 6,8
51Mean Squared Error
MSE
?
?
52Iso-contours of fit surface
Iso-contour of actual function
?
?
? (deg)
? (deg)
53DACE fit for Pitching Moment Data Predictions are
at untried points
Cm
?
?
54Mean Squared Error
MSE
?
?
55Iso-contours of fit surface
Iso-contour of actual function
?
?
? (deg)
? (deg)
56Review of MDO for Aerospace Vehicles
- McQuade Ph.D thesis Univ. of Washington, 1991
- Aerodynamic optimization of a 2D scramjet
vehicle using CFD (Euler). - Fore-body and Nozzle were separately optimized
to maximize thrust.
- Engg. Models used Oblique Shock theory, 1D
Heat Addition, MOC - correction factors based on 2D CFD (Euler)
analysis (GLA)
57General Application of Global-Local Approximation
Approximate Problem formulation
Complete Optimization ( 1
iteration)
No
Convergence?
Yes
stop
58Review of MDO for Aerospace Vehicles
Afterbody Optimization
Objective maximize the net thrust Subject to
constraints on geometric parameters CFD , 1D
isentropic flow, MOC Taylor Series, GLA using
1D, GLA using MOC
59Review of MDO for Aerospace Vehicles
Afterbody Optimization
Results
Method ? (deg.) ?curve (deg.) Fnet CFD calls Relative Cost/step
Init Design 18.000 0.0050 18.05 - -
CFD 20.541 0.0032 19.71 22 1.0
1-D 26.000 0.0082 18.42 - -
MOC 26.000 0.0082 18.42 - -
Taylor 20.362 0.0031 19.71 7 0.0098
1D GLA 20.850 0.0035 19.71 7 0.0083
MOC GLA 20.563 0.0033 19.70 7 0.0109
60Review of MDO for Aerospace Vehicles
Fore-body Optimization
Objective maximize the net thrust Subject to
constraints on geometric parameters CFD ,
Oblique shock theory Taylor Series, GLA based on
Oblique shock
61Review of MDO for Aerospace Vehicles
MDO of Air Breathing Hypersonic Vehicle
Ref Bowcutt J.of Propulsion and Power ,
Nov-Dec,2001
Optimization of Vehicle Configuration for
performance (range) across a specified Mach No.
vs Altitude Trajectory
- Sizing
- Aerodynamics
- Stability Control
- Propulsion
- Trajectory
Optimization variables nose angle, engine
axial location,
engine cant, cowl length and
chine length
62Review of MDO for Aerospace Vehicles
Key changes in the Optimized vehicle configuration
- Engine location moved forward by 6 of vehicle
length - Engine cant reduced by 2 deg
- Engine cowl length reduced by 5 of vehicle
length - Chine length reduced by 80 of vehicle length
The optimized vehicle, flying the same M-q
trajectory as the baseline, achieved 46
greater air-breathing range 9 improvement in
effective specific impulse 13 reduction in trim
drag over the baseline configuration
63Review of MDO for Aerospace Vehicles
Aerodynamic Lessons
- Wind tunnel testing and CFD analysis was
performed on the - Optimized vehicle
- HABP like Engg. Codes overpredicted lower surface
pressures in - the aft region of the vehicle.
- Vehicle range reduced by 6 based on W/T
Aerodynamics. - Vehicle instability levels in terms of negative
static margin increased - resulting in reduction in max. flight dynamic
pressure at which the - vehicle could operate.
64Review of MDO for Aerospace Vehicles
Hall mark of MDO
Range sensitivities to the five vehicle design
parameters
Parameter Derivative
A variation that is detrimental by itself can be
beneficial when working in concert with many
coupled variations.
65Review of MDO for Aerospace Vehicles
- John Robert Olds , Ph.D. thesis NCSU, 1993
- Advanced Space Transportation Vehicle optimized
for minimum weight. - Taguchi methods was used to select initial
experimental arrays. Parametric - methods were used to determine the settings
for design variables which - minimized weight. The effect noise
variables on the objective function was - included to ensure a robust design .
- Central Composite Design was used for the
final design variables . - Quadratic Response surface was created using
RSM - Non-linear optimizer was used to optimize the
quadratic surface - Remarks
- Parametric methods are useful only for very
early design stages where the - number of design variable are very few.
- Initial problem size 8 design variables
- Final problem size 3 design variables
- Inclusion of design constraints in the frame
work is not easy.
66Review of MDO for Aerospace Vehicles
- Giunta , Ph.D thesis VPI SU , 1997
- HSCT configuration optimized for TOGW.
- Variable Complexity Response Surface Modeling.
- Low fidelity methods used to screen the original
design space. - Response Surfaces (polynomial based) using
medium fidelity analysis created - for the reduced design space.
- RSMs were used for function evaluations in the
optimizer. - Preliminary investigation on the use of Design
of Computer Experiments - (Kriging) for creating response surfaces was
also carried out.
- Remarks
- RSM help to smoothen out the numerical noise in
analysis methods. This ensures that
the gradient calculations (search directions )
are not affected. - Constraints from aerodynamics, propulsion,
stability, performance. - Methodology demonstrated for problem sizes of 5
to 20 design variables. - Curse of dimensionality limits the problem size.
- Further studies are needed to investigate the
capabilities of DACE modeling.
67Review of MDO for Aerospace Vehicles
- Summary
- Various strategies have been used to address the
issue of - computational burden associated with high
fidelity analysis - Parametric methods with RSM
- Global Local Approximation
- First Order Approximate Model Management
- Variable Complexity Response Surface Method
68Review of MDO for Aerospace Vehicles
-
Summary - For problems at complete vehicle level, RSM
based on linear regression has - been widely used to overcome the challenge of
computational cost. - Once the response surface is available, in most
of the cases, an optimizer - has been used to find the minimum of the
surface - Issues
- It may be difficult to predict the form of the
linear regression. - Restriction on the number of design variables
is a serious limitation. - Sample points to construct the RSM are chosen
based on DoE. These - may not necessarily be in the region of
interest. - Multiple starts are required in the
optimization, to verify if the solution - is not a local minima.
- Conceptual problem is reported on the use of
RSM based on linear - regression for computer simulation
experiments.