Optimization of Blowing and Suction Control on NACA 0012 Airfoil Using Genetic Algorithm

1
Optimization of Blowing and Suction Control on
NACA 0012 Airfoil Using Genetic Algorithm

• L.Huang, R.P. LeBeau, P.G. Huang
• University of Kentucky
• Th. Hauser
• Utah State University

Supported by Kentucky Science and Engineering
Foundation Kentucky NASA EPSCoR
University of Kentucky, 1/5/2004
2
Road Map

• Background
• Genetic Algorithm
• Case Setup
• Two-Jet Control System Optimization

University of Kentucky, 1/5/2004
3
Background

• Flow Control Design Process

Design Quality? Time Limit?
Trial Error
Knowledge
Jet Control
Solution CFD Genetic Algorithm
University of Kentucky, 1/5/2004
4
Background
Solution CFD Genetic Algorithm

• Encode flow control jet parameters (amplitude,
  angle, location) into genetic algorithm
  individuals
• Evaluate each individual (set of jet parameters)
  in CFD simulation to determine lift and drag
• Use genetic algorithm to guide evolution to

• configuration of jets that yield maximum lift,
  minimum drag.
• Test case NACA0012 airfoil with two flow-control
  jets

University of Kentucky, 1/5/2004
5
Background

• Challenges in Genetic Algorithm

• Traditional binary-coded GA limitation
• Convergence speed
• Preliminary convergence to local optima

• Real-Coded EARND Genetic Algorithm

Explicit Adaptive Range Normal Distribution
University of Kentucky, 1/5/2004
6
Genetic Algorithm

• Optimization process

Genetic Algorithm
Crossover
Mutation
University of Kentucky, 1/5/2004
7
Genetic Algorithm

• General Process and Major Improvement

NND
EARND
Real-Coded
University of Kentucky, 1/5/2004
8
Genetic Algorithm

• Binary-Coded Real-Coded (Erased)

Example Encode a five-variable solution in to a
binary string
for
Binary String Chromosome


University of Kentucky, 1/5/2004
9
Genetic Algorithm

• Binary-Coded Real-Coded (eraseContinue)

C Notation
struct MyChromosome double x5 //
five-variable double fitRaw // raw fitness
double fitScale // scale fitness MyChromosome
X X.x00.714285 X.x10.571429 X.x20.42
8571 X.x30.285714 X.x40.142857
Real-Coded Using double
University of Kentucky, 1/5/2004
10
Genetic Algorithm

• Roulette Wheel Selection Operator

4 individuals per generation
Example Maximum function Optimization
Space
Basic
and if
Space
Improve
University of Kentucky, 1/5/2004
11
Genetic Algorithm

• Crossover Operator

Binary-Coded One Cut Point Crossover

Real-Coded Crossover
University of Kentucky, 1/5/2004
12
Genetic Algorithm

• Mutation Operator

Binary-Coded One Point Mutation

Improved Real-Coded Mutation
is scale up factor
if
is even generation
or
if
is odd generation
University of Kentucky, 1/5/2004
13
Genetic Algorithm

• Normal Distribution Major Improvement 1

University of Kentucky, 1/5/2004
14
Genetic Algorithm

• Explicit Adaptive Range Major Improvement 2

During the last 50 evolution
University of Kentucky, 1/5/2004
15
Genetic Algorithm

• Test Function 1 Ackleys Function

Genetic Coefficients

University of Kentucky, 1/5/2004
16
Genetic Algorithm

• Test Function 1 Ackleys Function (Cont.)

NND-Basic Genetic Algorithm
EARND-Improved Genetic Algorithm
University of Kentucky, 1/5/2004
17
Genetic Algorithm

• Test Function 2 Rastringins function

Genetic Coefficients
University of Kentucky, 1/5/2004
18
Genetic Algorithm

• Test Function 2 Rastringins function (Cont.)

NND-Basic Genetic Algorithm
EARND-Improved Genetic Algorithm
University of Kentucky, 1/5/2004
19
Case Setup

• Grid Numerical Scheme

210,000 grid points
Re500,000
Incompressible CFD Code Ghost Finite Volume
Structure 2-D Convective Terms Quick Diffusive
Terms Central Difference
University of Kentucky, 1/5/2004
20
Case Setup

• Computation and Experiment Comparison

University of Kentucky, 1/5/2004
21
Case Setup

• Control Parameter Selection

Jet Width 0.025c
University of Kentucky, 1/5/2004
22
Case Setup

• Control Parameter Selection (Cont.)

Two-Jet Control System Suction Jet Blowing Jet
5 Control Variable
fixed Suction Amplitude 0.03
University of Kentucky, 1/5/2004
23
Two-Jet Control System Optimization

• Objective

• Understand the GA Optimization Process
• 2. Understand the Flow Control Physics

University of Kentucky, 1/5/2004
24
Two-Jet Control System Optimization

• Understand the GA Optimization Process

Genetic Coefficient
Aggregate Fitness Function


Suction Location Suction Angle Blowing
Location Blowing Angle Blowing Amplitude
Computation Time
Computation Time/Case (Individual) 1.5 Hour on
16 Intel Xeon 2.66G Processor
Total Time (100 Generation) 45 days on 64 Intel
Xeon 2.66G Processor
around 69,000 hours (Wall Time)
University of Kentucky, 1/5/2004
25
Two-Jet Control System Optimization

• Understand the GA Optimization Process

• Convergence History

University of Kentucky, 1/5/2004
26
Two-Jet Control System Optimization

• Understand the GA Optimization Process

• Convergence History Movie (1)

University of Kentucky, 1/5/2004
27
Two-Jet Control System Optimization

• Understand the GA Optimization Process

• Convergence History Movie (2)

University of Kentucky, 1/5/2004
28
Two-Jet Control System Optimization

• Understand the GA Optimization Process

• Statistics History

University of Kentucky, 1/5/2004
29
Two-Jet Control System Optimization

• Understand the GA Optimization Process

• Top 100 fit Individuals

Suction Location
Suction Angle
Blowing Amplitude
Blowing Angle
Blowing Locations
University of Kentucky, 1/5/2004
30
Two-Jet Control System Optimization

• Understand the Flow Control Physics

• Most fit Individual and variations

Split Single Suction Jet Single Blowing Jet
University of Kentucky, 1/5/2004
31
Two-Jet Control System Optimization

• Understand the Flow Control Physics

• Single Suction Control

University of Kentucky, 1/5/2004
32
Two-Jet Control System Optimization

• Understand the Flow Control Physics

• Single Suction Location

Suction Angle -90
Suction Amplitude 0.173
Suction Location 0.1, 0.333, 0.567
University of Kentucky, 1/5/2004
33
Two-Jet Control System Optimization

• Understand the Flow Control Physics

• Two-Jet Control

University of Kentucky, 1/5/2004
34
Two-Jet Control System Optimization

• Understand the Flow Control Physics

• Two-Jet Control

Split Single Suction Jet Single Blowing Jet
University of Kentucky, 1/5/2004
35
Two-Jet Control System Optimization

• Conclusion

• Algorithm
• EARND is an efficient algorithm, which is proved
  to have the ability to
• identify and optimize the control factors in
  sequence.
• Suction/Blowing Jet Control
• Suction jet is dominant, the blowing jet is
  secondary to the overall
• fitness improvement.
• Control Parameters
• The most important and fastest converging
  parameters are the
• suction location and angle.
• The blowing location is of secondary importance,
  while the blowing angle
• and blowing amplitude are the parameters least
  well-constrained and
• least critical to the overall performance of the
  two-jet system.