SUPERSONIC BUSINESS JET

Transcription

1 SUPERSONIC BUSINESS JET Design Space Exploration and Optimization Josiah VanderMey Hassan Bukhari MIT /ESD.775

2 Overview 2 Problem Formulation Motivation and Challenges Objectives and Constraints Model and Simulation Model Overview and Description Benchmarking and Validation Optimization Algorithms and tuning Post Optimality Analysis Multi-Objective and Tradeoff Analysis Conclusions and Recommendations

3 3 Problem Formulation Motivation and Challenges Motivation Large potential market for a Supersonic Business Jet 1,2 Fast transportation for executives who travel frequently and are able to afford more expensive transportation (20-50% reduction in travel time) 3 Business aircraft less sensitive to economic fluctuations Application outside of solely business executives MEDEVAC Airfreight Military Challenges High speed flight aerodynamics Very expensive aircraft to own and operate 1 Because of increasing environmental awareness, the focus for the design of this aircraft must include environmental concerns in addition to traditional performance and economic metrics. 4 Overland flight with minimal sonic boom Engine must meet noise and emissions standards

4 Environmental Constraints Performance Constraints 4 Problem Formulation Objectives and Constraints Objective Statement: Design a highly profitable supersonic business jet that complies with noise and performance regulations required to operate out of commercial airports Outputs from system model divided into constraints or objectives based on their potential impact on profits (objectives) or compliance with regulations (constraints) Type Variable Name Min Max Take-off Field Length (ft) TOFL 11,000 Landing Field Length (ft) LANDFL 11,000 Objective Name Approach Speed (kts) APPSPD 155 Take-off Gross Weight (lbs) TOGW Approach Angle of Attack (deg) AANGLA 12 Fuel Volume Ratio (available/required) FRATIO 1.0 Delta Sideline Noise SNOISE 10 Fuel Weight (lbs) Average Yeild per Revenue Passenger Mile ($/mi) Acquisition Cost (Million $) FUELWT DPRPM ACQCST Delta Flyover Noise FNOISE 10 Delta Approach Noise ANOISE 10

5 5 Model and Simulation Overview Inputs Wing and tail geometry Engine Parameters Outputs Objective and Constraints Each output is modeled using a Response Surface Equation (RSE) Linear and interaction terms only RSE = β 0 + n i=1 β i x i n 1 n + β ij x i x j i=1 j =i+1

6 6 Model and Simulation Overview Limitations/Features of RSE 6 Accuracy only guaranteed in a small trust region around sample points Unable to predict multiple extrema Assumes randomly distributed error (not usually the case in computer experiments)

7 Engine Variables Planform Geometry Variables Translation Variables 7 Model and Simulation Variables and Parameters Type Variable Name Min Max Wing Apex (ft) XWING Horizontal Tail Apex (ft) XHT Vertical Tail Apex (ft) XVT Leading Edge Kink X-Location X1LEK Leading Edge Tip X-Location X2LET Trailing Edge Tip X-Location X3TET Trailing Edge Kink X-Location X4TEK Trailing Edge X-Location X5TER Kink Y-Location Y1KIN Wing Area (ft 2 ) WGARE Horizontal Tail Area (ft 2 ) HTARE Veritical Tail Area (ft 2 ) VTARE Nozzle Thrust Coefficient CFG Turbine Inlet Temperature ( o R) TIT Bypass Ratio BPR Overall Pressure Ratio OPR Fan Inlet Mach Number FANMN Fan Pressure Ratio FPR Engine Throttle Ratio ETR Suppressor Area Ratio SAR Take-off Thrust Multiplier TOTM Thrust-to-Weight Ratio FNWTR Variable & Parameter Influence Normalized Main Effects XWING XHT XVT X1LEKN X2LETP X3TETP X4TEKN ACQCST*10^-2 X5TERT FUELWT*10^-6 Y1KINK DPRPM*10 WGAREA TOGW*10^-6 HTAREA VTAREA CFG TIT BPR OPR FANMN FPR ETR SAR TOTM FNWTR

8 8 Model and Simulation Benchmarking and Validation Very fast Run time on the order of 7e-6 sec Modeled geometry of several supersonic business jet designs Model design space is unique, even for SSBJ Much larger than most supersonic aircraft Sukhoi-Gulfstream S-21 Aérospatiale-BAC Concorde

9 Engine Variables Environ mental Performance Constraints Planform Geometry Variables Translation Variables 9 Model and Simulation Benchmarking and Validation Type Variable Min Max Sukhoi- Gulfstream S-21 Aérospatiale- BAC Concorde Wing Apex (ft) Horizontal Tail Apex (ft) Vertical Tail Apex (ft) Leading Edge Kink X-Location Leading Edge Tip X-Location Trailing Edge Tip X-Location Trailing Edge Kink X-Location Trailing Edge X-Location Kink Y-Location Wing Area (ft 2 ) Horizontal Tail Area (ft 2 ) Veritical Tail Area (ft 2 ) Nozzle Thrust Coefficient Turbine Inlet Temperature ( o R) Bypass Ratio Overall Pressure Ratio Fan Inlet Mach Number Fan Pressure Ratio Engine Throttle Ratio Suppressor Area Ratio Take-off Thrust Multiplier Thrust-to-Weight Ratio Type Objective S21 Model S1 Actual Concorde Concorde Model Actual Take-off Gross Weight (lbs) 512, , , ,000 Fuel Weight (lbs) 289,560 67,409 2,502, ,940 Average Yeild per Revenue Passenger Mile ($/mi) Acquisition Cost (Million $) Variable Min Max S21 Model S1 Actual Concorde Model Concorde Actual Take-off Field Length (ft) 10,500 19,358 6, ,360 11,778 Landing Field Length (ft) 11,000 12,503 6,496 14,024 Approach Speed (kts) Approach Angle of Attack (deg) Fuel Volume Ratio (available/required) Delta Sideline Noise Delta Flyover Noise Delta Approach Noise

10 Model and Simulation Benchmarking Overall Planform and Validation Baseline design at center of regression Baseline (9000, ) HT VT X-loc (ft) Baseline (433,788.5 kg)

11 Design Vector Input Objective Vector Const. Param. 11 Optimization Overview Multidisciplinary aspects of the model are masked by the RSE s Necessitates Single level optimization Sizing Model Structures Stability Cost/Revenue Aerodynamics Noise Propulsion Performance Single-level optimizer Tradespace Exploration (DOE)

12 12 Optimization DOE and Design Space Exploration DOE Latin Hypercubes 10,000 levels Only 3 feasible designs found Most designs excluded based on TOFL and ANOISE constraints TOGW DPRPM Feasible Designs FUELWT ACQCST Best Designs from all Samples

13 13 Optimization Gradient Based Used DOE designs to come up with ballpark objective scaling to form a multi-objective objective function SQP implemented in MATLAB Very efficient on smooth response surfaces Fast convergence Convergence tolerance set to 1e-6 on constraints and objective function Started from feasible designs as well as Best designs found in DOE Very fast convergence Each starting point converged to a different optimal solution Islands of feasibility in design space Gradient solver not a very good solution

14 Planform Geometry Variables Translati on 14 Optimization Heuristic - SA SA implemented in MATLAB Much more costly than gradient based optimization Better, more stable solutions Run from multiple starting locations Tuning Parameter Value Overall Planform T0 100 Cooling Schedule exponential dt 0.95 neq 5.00E+03 nfrozen 10 Optimal Design Geometry X-loc (ft) TOGW: 849,647 FUELWT: 447,648 DPRPM: ACQCST: J: HT VT Design perturbation 4 variables at a time Normal distribution (standard deviation equal to 1/3 the allowable range Reset to upper and lower bounds if exceeded Type Variable Value Min Max Wing Apex (ft) Horizontal Tail Apex (ft) Vertical Tail Apex (ft) Leading Edge Kink X-Location Leading Edge Tip X-Location Trailing Edge Tip X-Location Trailing Edge Kink X-Location Trailing Edge X-Location Kink Y-Location Wing Area (ft 2 ) Horizontal Tail Area (ft 2 ) Veritical Tail Area (ft 2 ) Optimal Design Geometry Active Constraints Shown in Red

15 15 Optimization Heuristic - SA Optimized Feasible 1 Feasible 2 Feasible

16 Y-loc (ft) Y-loc (ft) Optimization MOO Overall Planform Overall Planform Individual objective optimizations 90 Overall Planform Min TOGW HT VT TOGW: 825,974 X-loc (ft) FUELWT: 484,575 DPRPM: ACQCST: Overall Planform Max DPRPM HT VT TOGW: 932,407 X-loc (ft) FUELWT: 517,315 DPRPM: ACQCST: HT HT Min FUELWT X-loc (ft) TOGW: 832,765 FUELWT: 438,856 DPRPM: ACQCST: VT Min ACQCST X-loc (ft) TOGW: 826,734 FUELWT: 487,553 DPRPM: ACQCST: VT

17 Post Optimality Sensitivity 17 XWING XHT XVT X1LEK X2LET X3TET X4TEK X5TER Y1KIN WGARE HTARE VTARE CFG TIT BPR OPR FANMN FPR ETR SAR TOTM FNWTR TOGW FUELWT DPRPM ACQCST J Scaled Output Sensitivity at Optimal Design

18 Post Optimality Pareto Front and Trade-off Analysis 18 TOGW, FUELWT, and ACQCST are all mutually supportive The trades occur with DPRPM Pareto front: Using AWS approach

19 Conclusions and Recommendations 19 Fairly confident that we have found global optimal design for our weight selection Consistent heuristic convergence to the optimal design Need to get a better understanding of the customer wants Include additional performance metrics and constraints Stability Emissions Range Altitude Speed Refine model around optimal solution Limited domain of RSE Go back to high fidelity model Consider higher order model Re-evaluate constraints black box leads to a poor understanding of assumptions, parameters, etc. Include additional parameters (e.g. wing thickness)

20 References B. Chudoba & Al., What Price Supersonic Speed? An Applied Market Research Case Study Part 2, AIAA paper, AIAA , 45th AIAA Aerospace Sciences Meeting and Exhibit, Reno, C. Trautvetter, Aerion : A viable Market for SSBJ, Aviation International News, Vol. 37, No. 16, Deremaux, Y., Nicolas, P., Négrier, J., Herbin, E., and Ravachol, M., Environmental MDO and Uncertainty Hybrid Approach Applied to a Supersonic Business Jet, AIAA , Briceño, S.I., Buonanno, M.A., Fernández, I., and Mavris, D.N., A Parametric Exploration of Supersonic Business Jet Concepts Utilizing Response Surfaces, AIAA , Federal Aviation Administration (FAA), Federal Aviation Regulations (FAR), FAR Cox, S.E., Haftka, R.T., Baker, C.A., Grossman, B.G., Mason, W.H., and Watson, L.T., A Comparison of Global Optimization Methods for the Design of a High-speed Civil Transport, Journal of Global Optimization, Vol. 21, No. 4, Dec. 2001, pp B. Chudoba & Al., What Price Supersonic Speed? A Design Anatomy of Supersonic Transportation Part 1, AIAA paper, AIAA , 45th AIAA Aerospace Sciences Meeting and Exhibit, Reno, Chung, H.S., Alonso, J.J., Comparison of Approximation Models with Merit Functions for Design Optimization, AIAA , 200.

21 21 Backup

22 Post Optimality Scaling 22 Objective function was scaled to be O(1) Since the response surface does not have any second order terms, the diagonal of the Hessian is 0

23 MIT OpenCourseWare ESD.77 / Multidisciplinary System Design Optimization Spring 2010 For information about citing these materials or our Terms of Use, visit: