Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
3 August 2000 SPIE 4095-04 1
Design of a Gradient-index Beam Shaping System Via a Genetic Algorithm Optimization Method
Neal C. Evans and David L. Shealy University of Alabama at Birmingham
Department of Physics, 1530 3rd Avenue South, CH310Birmingham, AL 35294-1170 USA
3 August 2000 SPIE 4095-04 2
Concepts and Context – Laser Beam Shaping
3 August 2000 SPIE 4095-04 3
Global Optimization
• General Concepts• Fitness Landscape
– Discrete and continuous variables
3 August 2000 SPIE 4095-04 4
Merit Function Topography• Merit Function
– Optical Design Parameters– Contributions from Individual Elements
ΡN
µ
M
M=(Irradiance Profile calculations on Output Surface) * (Radius of Output Surface) * (Wavefront Checks)
3 August 2000 SPIE 4095-04 5
Overview of Optimization Algorithms in Optics and Optical Design1
•Damped least-squares
•Expert Systems –ZEMAX
–SYNOPSYS
–Optics Toolbox
–CODE V
•Simulated Annealing
•Tabu Search
1. See D. Shafer, “Global Optimization in Optical Design,” Computers in Physics 8, 188-195 (1994).
•Proprietary Merit Functions
•Difficult to Scale
3 August 2000 SPIE 4095-04 6
Genetic Algorithms (GAs)• Useful Tool in Geometrical Optics?• Advantages
– Ease of implementation and broad applicability– Relative immunity to local extrema– Easily made parallel
• Drawbacks– Random starting designs– “This planet does not lack for life forms, but
there is a paucity of intelligent ones.2”2. Shafer,192.
3 August 2000 SPIE 4095-04 7
GA Concepts
• Biological Paradigm and Nomenclature• Expressing solutions in terms of binary
strings: genetic code
3 August 2000 SPIE 4095-04 8
GA Concepts
• Generating solutions• Populations and
individuals• Parents and children:
producing new populations
Init ialize GA byrandomly pickingnew individuals
Evaluate MeritFunct ion for eachindividual ingenerat ion
P erform genet ic operat ions(reproduct ion, mutat ions,cross-overs); produce newgenerat ion
Is the populat ionstagnant? (Micro-GAcheck)
Keep best individual andreplace the remainder withrandomly-selected individuals
Terminat ioncrit erion reached?
End
Y
N
Y
N
Step eligible for parallelizat ion
3 August 2000 SPIE 4095-04 9
Irradiance Profile Calculations
• A fast, efficient method is needed to calculate the irradiance profile on a specified Output Surface
• The Geometrical Optics form of the Energy Conservation Law is harnessed for these calculations:
( )2 2 20 ( ) 0n k e∇ + =r
0 0( ) ( ) exp ( )e e ik S = r r r
( )2 2S n∇ = 2 20 0 02 0e S e e S∇ ⋅∇ + ∇ =
( )f f f∇⋅ = ∇⋅ + ⋅∇v v v( ) 0I∇⋅ =a
1 1 2 2I dA I dA=
Source
1 1I dA
2 2I dA
3 August 2000 SPIE 4095-04 10
( ) ( )ˆ ˆ( ) ( ) ( ) ( ) ( ) ( ) ,in in out out
I O
E da u dAσ ρ ρ ρ= ⋅ = Ρ Ρ ⋅ Ρ∫ ∫n v n v
( ) ( )11
cos( ) cos( )( ) .
cos( ) ( )
in outi i i ii
i i outi i i i
iu
i
ρ ρ ρ χσ ρ −
−
− Ρ = Ρ Ρ −Ρ 1 1 2 2I dA I dA=
3 August 2000 SPIE 4095-04 11
Beam Profile on Input Plane
( )2( ) expi iσ ρ αρ= −
( )σ ρ
,mmρ
3 August 2000 SPIE 4095-04 12
Merit Function Topography for Beam Shaper/Projector
ΡN
µ
M
uN
u ii
N
==∑1
1
( ).Ρ
2
1
1 { ( ) }N
ii
u uN
µ=
= Ρ −∑
21 exp 0.01{50 } ,NMµ
= − − Ρ
3 August 2000 SPIE 4095-04 13
Results for GA-optimized Beam Shaper/Projector System
• CODE V used to evaluate merit function• Simultaneous execution on 4 Sun Sparcs• Manually Tweaking the Merit Function
Lanscape• Total search time of 7 hours• “Best” solution found
3 August 2000 SPIE 4095-04 14
Zoomed Area (above)
Thin Lens Element Shaping
Element
Output Surface Input P lane
GA Solution for Beam Shaper/Projector System
3 August 2000 SPIE 4095-04 15
GA-determinedsurface
z
z h c hk c h
A hjj
j
( )( )
,=+ − +
+=∑
2
2 2 22
2
5
1 1 1
Surface Parameter Constraintc −10 20tok −10 0. toA4 − × ×−10 10 10 103 3. .toA6 − × ×−10 10 10 105 5. .toA8 − × ×−10 10 10 106 6. .toA10 − × ×−10 10 10 107 7. .to
Beam-shaping element
3 August 2000 SPIE 4095-04 16
Beam Profile on Output Surface
( )u Ρ
,mmΡ
( ) ( )11
cos( ) cos( )( ) .
cos( ) ( )
in outi i i ii
i i outi i i i
iu
i
ρ ρ ρ χσ ρ −
−
− Ρ = Ρ Ρ −Ρ
3 August 2000 SPIE 4095-04 17
Two-lens Beam Shaper Problem
W. Jiang, Shealy, Martin (1993)
Differential Equation Design Method: P. Rhodes and Shealy (1980)
3 August 2000 SPIE 4095-04 18
Lens 1
Lens 2
GA Solution to the Two-lens Shaper Problem
3 August 2000 SPIE 4095-04 19
Gradient-Index (GRIN) Shaper Problem*
21 1 1 1n a b z c z= + + 2
2 2 2 2n a b z c z= + +
*C. Wang and Shealy, “Design of gradient-index lens system for laser beam shaping,” Applied Optics 32, 4763-4769, 1993.
3 August 2000 SPIE 4095-04 20
GA Solution to GRIN:Part 1
Lens 2: LightPath G1SF, negat ive gradient
Lens 1: LightPath G1SF, posit ive gradient
Connector
3 August 2000 SPIE 4095-04 21
Free Form GRIN Laser Beam Shaping System
3 August 2000 SPIE 4095-04 22
Parameters Optimized for Free-form GA-designed GRIN ProblemParameter Description Type Limits[i]
1 Number of Elements Discrete 1-4 (integer)2 Radius of curvature of left surface of Element 1 Continuous -100 to 100 3 Radius of curvature of right surface of Element 1 Continuous -100 to 1004 Thickness of Element 1 Continuous 1 to 10 5 Distance between Element 1 and Element 2 Continuous 1 to 10 6 GRIN glass type for Element 1 Discrete 1-6 (integer)7 Positive or Negative GRIN for Element 1 Discrete 0 or 18 Radius of curvature of left surface of Element 2 Continuous -100 to 100 9 Radius of curvature of right surface of Element 2 Continuous -100 to 100 10 Thickness of Element 2 Continuous 1 to 10 11 Distance between Element 2 and Element 3 Continuous 1 to 10 12 GRIN glass type for Element 2 Discrete 1-6 (integer)13 Positive or Negative GRIN for Element 2 Discrete 0 or 114 Radius of curvature of left surface of Element 3 Continuous -100 to 100 15 Radius of curvature of right surface of Element 3 Continuous -100 to 100 16 Thickness of Element 3 Continuous 1 to 10 17 Distance between Element 3 and Element 4 Continuous 1 to 10 18 GRIN glass type for Element 3 Discrete 1-6 (integer)19 Positive or Negative GRIN for Element 3 Discrete 0 or 120 Radius of curvature of left surface of Element 4 Continuous -100 to 100 21 Radius of curvature of right surface of Element 4 Continuous -100 to 100 22 Thickness of Element 4 Continuous 1 to 10 23 Distance between Element 4 and Surface 10 (a dummy surface) Continuous 1 to 10 24 GRIN glass type for Element 4 Discrete 1-6 (integer)25 Positive or Negative GRIN for Element 4 Discrete 0 or 126 Distance from Surface 10 (a dummy surface) to the Output Plane Continuous 1 to 100
3 August 2000 SPIE 4095-04 23
Determining when a Solution is Found
3 August 2000 SPIE 4095-04 24
Free-form GA-designed GRIN Beam Shaper•Average evaluation time per generation: 7.80s. Total execution time: 26.8 hrs
•Integrating Output Profile over Output Surface yields 21.9 units; integrating Input Profile over Input Surface yields 21.7 units
3 August 2000 SPIE 4095-04 25
Free-form GA-designed GRIN Solution
Element 1
Element 2 Element 3
3 August 2000 SPIE 4095-04 26
Summary and Conclusions• Assumption: there are some problems in Geometrical
Optics that are intractable for Analytical or Conventional Optimization Techniques
• Proof-of-principle of GA optimization stage: the Beam Shaper/Projector and the Two-Lens Shaper
• Creative stage of GA optimization: the GRIN problems• Machine Learning Codes, Genetics Algorithms in
particular, are well-suited for these problems where discrete parameters are involved, or where the form of the solution is not known a priori
3 August 2000 SPIE 4095-04 27
Free Form 3-element GRIN