Evolutionary Algorithms Eleventh Australian International Aerospace Congress 13-17 March , Melbourne Convention Centre and Australian International Airshow 2005 at Avalon Airport Design November 15-19, 2004 L. F. Gonzlez, E. J. Whitney, K. Srinivas, K.C Wong The University of Sydney, Australia J. Priaux Dassault Aviation Pole Scientifique, INRIA Sophia Antipolis, OPALE project associate OUTLINE Introduction Unmanned Aerial Vehicle (UAV/UCAV) Design Requirements
The need and requirements for a Multidisciplinary Design Optimsation Framework in Aeronautics
OUTLINE UAVDESIGN REQUIREMENTS Use and development of UAV for military and civilian applications is rapidly increasing.
Similar to the manned aircraft the challenge is to develop trade-off studies of optimal configurations to produce a high performance aircraft that satisfy the mission requirements.
UAV systems are ever increasingly becoming important topics for aerospace research and industrial institutions.
There are difficulties in these new concepts because of the compromising nature of the missions to be performed, like high-- or medium--altitude surveillance, combat environments (UCAV) and many others.
Multi-missions highmedium-- altitude surveillance High Performance Complex trade- offs Optimization-Optimal Solution(S) Pareto optimal Surface of UAV, UAV MDO Complex Task - UAV -Example Multiple Goals Minimise-Maximise Purchase Price Aerodynamic Performance Takeoff weight Multiple Disciplines Structures Fight Controls Aero acoustics Propulsion Sensors Aero elasticity Aerodynamics Search Space Large Multimodal Non-Convex Discontinuous Post-Processing Visualization tools Multi-objective, trade- off WHY A FRAMEWORK FOR MDO? in-house/ commercial solvers- inaccessible modification Optimisation Multiple Disciplines Parallel Computing A software system to integrate and evaluate different complexities of MDO is required REQUIREMENTS FOR A MO-MDO FRAMEWORK Robust Optimisation methods (Global solutions, handle noise, complex functions, ease of integration of legacy codes CFD-FEA- black-boxes).
Problem formulation and execution (Automatic movement of data, parallel Processing heterogeneous computers).
Architectural design and information access (GUI, object oriented, no-overhead on optimization, easily extended, database- management, post-processing, visualization capabilities, fault tolerance mechanisms) Data Data GUI Aerofoil Design MSES, XFOIL NSC2ke Wing Design FLO22 CalculiX Aircraft Design FLOPS , ADA Nozzle Design HDASS Mathematical Test Functions GUI Design of Experiments
Optimisation EA Optimiser Gradient Based Optimiser Parallel Computing MPI PVM Analysis Modules RSM Kriging Post-Processor Propeller Design
Mesh generator MDO FRAMEWORK Traditional Gradient Based methods for MDO might fail if search space is:
Advanced Optimisation Tools: Evolutionary Optimisation Crossover Mutation Fittest Evolution ROBUST AND EFFICIENT OPTIMISATION TOOLS Large Multimodal Non-Convex Many Local Optimum Discontinuous
Good for all of the above Easy to paralellise Robust towards noise Explore larger search spaces Good for multi-objective problems
EVOLUTION ALGORITHMS What are EAs. There are many evolutionary methods and algorithms.
The complex task of MDO requires . Crossover Mutation Fittest Evolution Based on the Darwinian theory of evolution populations of individuals evolve and reproduce by means of mutation and crossover operators and compete in a set environment for survival of the fittest.
A Robust and efficient evolutionary optimisation method. DRAWBACK OF EVOLUTIONARY ALGORITHMS A typical MDO problem relies on CFD and FEA for aerodynamic and structural analysis. CFD/FEA Computation are time consuming Our research addresses these issue in some detail Evolution process is time consuming/ high number of function evaluations are required. Hierarchical Asynchronous Parallel Evolutionary Algorithms (HAPEA) ROBUST OPTIMISATION METHODS Multi-objective Parallel Evolutionary Algorithm
Hierarchical Topology
Asynchronous Approach
Features of the Method: Our Contribution.. MULTI-OBJ ECTIVE OPTIMISATION (1) Aeronautical design problems normally require a simultaneous optimisation of conflicting objectives and associated number of constraints. They occur when two or more objectives that cannot be combined rationally. For example:
Drag at two different values of lift. Drag and thickness. Pitching moment and maximum lift. Best to let the designer choose after the optimisation phase. Maximise/ Minimise Subjected to constraints Objective functions, output (e.g. cruise efficiency).
x: vector of design variables, inputs (e.g. aircraft/wing geometry)
g(x) equality constraints and h(x) inequality constraints: (e.g. element von Mises stresses); in general these are nonlinear functions of the design variables.
( ) N i x f i ... 1 = ( ) ( ) K k x h N j x g k i ... 1 0 ... 1 0 = s = = ( ) x f i MULTI-OBJ ECTIVE OPTIMISATION (2) F 2 F 1 Pareto Optimal Front Non-Dominated Dominated Feasible region Infeasible region A set of solutions that are non-dominated w.r.t all others points in the search space, or that they dominate every other solution in the search space except fellow members of the Pareto optimal set. PARETO OPTIMAL SET
EAs work on population based solutions can find a optimal Pareto set in a single run
HIERARCHICAL TOPOLOGY- MULTIPLE MODELS Model 1 precise model Model 2 intermediate model Model 3 approximate model Exploration Exploitation
We use a technique that finds optimum solutions by using many different models, that greatly accelerates the optimisation process.
Interactions of the layers: solutions go up and down the layers.
Time-consuming solvers only for the most promising solutions.
Asynchronous Parallel Computing
Hierarchical Topology ASYNCHRONOUS EVALUATION Suspend the idea of generation Solution can be generated in and out of order Processors Can be of different speeds Added at random Any number of them possible Methods of solutions to MO and MDO -> variable time to complete. Time to solve non-linear PDE - > Depends upon geometry
Why asynchronous?? How: Evolution Algorithm Evaluator PROBLEM FORMULATION AND EXECUTION
The Method is applicable to integrated or distributed MDO analysis
Single or multi-objective problems can be analysed
EAs require no derivatives of the objective function
The coupling of the algorithm with different analysis codes is by simple function calls and input and output data files.
Different programming languages C, C++, Fortran 90, and Fortran 77. and CFD and FEA software: FLO22 FLOPS, ADA, XFOIL, MSES, CalculiX ARCHITECTURAL DESIGN AND INFORMATION ACCESS Design Modules
Design of Experiments
Post-processing
Parallel Computing
Optimisation Tools
DESIGN AND OPTIMISATION MODULES Wing Design Aircraft Design RESULTS SO FAR Evaluations CPU Time Traditional 2311 224 152m 20m New Technique 504 490 (-78%) 48m 24m (-68%) The new technique is approximately three times faster than other similar EA methods.
We have successfully coupled the optimisation code to different compressible and incompressible CFD codes and also to some aircraft design codes CFD Aircraft Design HDASS MSES XFOIL Flight Optimisation Software (FLOPS) FLO22 Nsc2ke ADS (In house) A testbench for single and multi-objective problems has been developed and tested Aircraft Conceptual Design and Multidisciplinary Optimisation 2D Nozzle Inverse Optimisation Transonic Wing Design UAV Aerofoil Design Shock Control Bump Optimisation CURRENT AND ONGOING OPTIMISED INDUSTRIAL APLICATIONS Propeller Design High Lift Aircraft System Transonic aerofoil optimisation using Grid-free solvers AF/A-18 Flutter Model Validation F3 Rear Wing Aerodynamics M 1.11389 1.06934 1.02478 0.980227 0.935671 0.891115 0.84656 0.802004 0.757448 0.712892 0.668337 0.623781 0.579225 0.534669 0.490113 0.445558 0.401002 0.356446 0.31189 0.267335 0.222779 0.178223 0.133667 0.0891115 0.0445558 CURRENT AND ONGOING OPTIMISED INDUSTRIAL APLICATIONS MULTIDISCIPLINARY AND MULTI-OBJECTIVE WING DESIGN OPTIMISATION Mach Number 0.69 Cruising Altitude 10000 ft C l 0.19 Wing Area 2.94 m 2
MOO OF TRANSONIC WING DESIGN FOR AN UNMANNED AERIAL VEHICLE (UAV) Objective: Minimisation of wave drag and wing weight
( ) ( )
min min 2 1 weight w sparcap d total f c f = = DESIGN VARIABLES
16 Design variables on three span wise aerofoils 9 Design variables on three span wise aerofoil section 57 design variables , , , , , , , rb bt l rb bt r b t ARw b A A I I I + Description Lower Bound Upper Bound Wing Aspect Ratio [AR] 3.50 15.00 Break to root Taper [br] 0.65 0.80 Break to tip Taper [bt] 0.20 0.45 Wing 1/4 Chord inboard Sweep, deg [i] 10.00 25.00 Break Location, [bl] 0.30 0.45 DESIGN VARIABLES Minimum thickness Position of Maximum thickness Fitness functions CONSTRAINTS & OBJECTIVE FUNCTIONS
( ) / 14% ,12% int ,11% tip t c root ermediate > ( ) / 20% 55% t c x s s ( ) 1 2 min( ) min w weight f Cd f totalsparcap = = Approach one : Traditional EA with single population model Computational Grid 96 x 12 x 16 Approach two : HAPEA Exploitation Population size = 30 Exploration Population size = 30 Intermediate Population size = 30 Grid size 96 x 12 x 16 Grid size 72 x 9 x 12 Grid size 48 x 6 x 8 Six machines were used in all calculations IMPLEMENTATION The algorithm was run five times for 2000 function evaluations and took about six hours to compute PARETO FRONTS AFTER 2000 FUNCTION EVALUATIONS MULTIDISCIPLINARY WING DESIGN
Best for Objective One Best for Objective Two Pareto Solutions RESULTS Aerofoil Geometries at 0, 20 and 100% semispan UAV DESIGN AND OPTIMISATION Minimise two objectives:
Operational Fuel Weight min(OFW) Endurance min (1/E)
Subject to: Takeoff length < 1000 ft Alt Cruise > 40000 ft Endurance > 24 hrs
With respect to: External geometry of the aircraft Mach = 0.3 Endurance > 24 hrs Cruise Altitude: 40000 ft DESIGN VARIABLES In total we have 29 design variables Design Variable Lower Bound Upper Bound Wing Area (sq ft) 280 330
Aspect Ratio 18 25.2 Wing Sweep (deg) 0.0 8.0 Wing Taper Ratio 0.28 0.8 13 Configuration Design variables Aerofoil-Wing Geometry W i n g
16 Design variables for the aerofoil + DESIGN VARIABLES
Twist Horizontal Tail Area (sq ft) 65.0 85.0 HT Aspect Ratio 3.0 15.0 HT Taper Ratio 0.2 0.55 HT Sweep (deg) 12.0 15.0 Vertical Tail Area (sq ft) 11.0 29.0 VT Aspect Ratio 1.0 3.2 VT Taper Ratio 0.28 0.62 VT Sweep (deg) 12.0 34.0 Fuselage Diameter 2.6 5.0 Tail Fuselage MISSION PROFILE
Structural & weight analysis A compromise on fidelity models Vortex induced drag: VLMpc Viscous drag: friction.f Aerofoil Design Xfoil Evolutionary Algorithms (HAPEA) Optimisation Aircraft design and analysis Aerodynamic Analysis Analytically by FLOPS Flight Optimsation System (FLOPS) NASA CODE DESIGN TOOLS
IMPLEMENTATION Population size: 20
Population size: 20
Population size: 20 Grid 141 x 74 x 36 on aerofoil, 20 x 6 on Vortex model Grid 109 x 57 x 27 on aerofoil, 17 x 6 on Vortex model Grid 99 x 52 x 25 on aerofoil, 15 x 6 on Vortex model Aircraft Design and Optimisation Module
The new technique facilitates the process of conceptual and preliminary MDO studies
The new technique with multiple models: Lower the computational expense dilemma in an engineering environment (three times faster)
Direct and inverse design optimisation problems have been solved for one or many objectives.
Some Multidisciplinary Design Optimisation (MDO) problems have been solved.
OUTCOMES (2) The algorithms find traditional classical results for standard problems, as well as interesting compromise solutions.
In doing all this work, no special hardware has been required Desktop PCs networked together have been up to the task.
No problem specific knowledge is required The method appears to be broadly applicable to different analysis codes.
Work to be done on approximate techniques and use of higher fidelity models.
Acknowledgements Mourad Sefrioui, Dassault Aviation for fruitful discussions on Hierarchical EAs and his contribution to the optimization procedure.
Steve Armfield and Patrick Morgan at the University of Sydney for providing the cluster of computing facilities.
We would like to thank Arnie McCullers at NASA LaRC who kindly provided the FLOPS software.
Questions Thank you for your attention Additional Slides Acknowledgements Multidisciplinary design problems involve search space that are multi-modal, non-convex or discontinuous.
Traditional methods use deterministic approach and rely heavily on the use of iterative trade-off studies between conflicting requirements.
Problems in MDO (1) Traditional optimisation methods will fail to find the real answer in most real engineering applications, (Noise, complex functions).
The internal workings of validated in-house/ commercial solvers are essentially inaccessible from a modification point of view (they are black-boxes). Problems in MDO The process of MDO is complex and involves several considerations as robust optimisation tools, problem formulations, parallel computing visualization tools. A software system or framework is desired Parallelization Module
Classification of our Model: The algorithm can be classified as a hierarchical Hybrid pMOEA model [CantuPaz] uses a Master slave PMOEA but incorporate the concept of isolation and migration trough hierarchical topology binary tree structure where each level executes different MOEAs/parameters (heterogeneous) The distribution of objective function evaluations over the salve processors is where each slake performs k objective function evaluations. Parallel Processing system characteristics: We use a Cluster of maximum 18 PCs with Heterogeneous CPUs, RAMs , caches, memory access times , storage capabilities and communication attributes. Inter-processor communication: Using the Parallel Virtual Machine (PVM) EAs The selection operator is a novel approach to determine whether an individual x is to be accepted into the main population Create a tournament Q Where B is the selection buffer. Population Tournament Q Asynchronous Buffer Evaluate x If x not dominated x Pareto Tournament Selection | | B n B B q q q Q n 2 1 6 1 ; .... , 2 1 s s c = Evolutionary Algorithms Explore large search spaces. Robust towards noise and local minima Easy to parallelise Map multiple populations of points, allowing solution diversity. A number of multi-objective solutions in a Pareto set or performing a robust Nash game.
UAV design
Pareto Optimal configurations The Challenge The use of higher fidelity models is still prohibitive, research on surrogate modeling/approximation techniques is required.
MDO is a challenging topic, the last few year have seen several approaches for Design and optimization using Evolutionary techniques but research indicate that it is problem dependent and it is still an open problem.
Access to Dell Linux Cluster is limited for benchmarking purposes. Use of higher fidelity models is still prohibitive. Work in Progress Master of Engineering
Rotor Blade design and Optimisation using evolutionary Techniques Adaptive Transonic Wing/Aerofoil Design and MDO using Evolutionary Techniques Grid-less Algorithms for Design and optimisation in Aeronautics
Undergraduate Projects
Transonic wing design using DACE (Design of Experiments-approximation Theories) An empirical study on DSMC for within evolutionary Optimisation