-
1 © 2014 ANSYS, Inc. November 4, 2015
Einsatz von optiSLang in der Elektrotechnik
Gerd Prillwitz
Senior Application Eng.
ANSYS Inc.
-
2 © 2014 ANSYS, Inc. November 4, 2015
Outline
Challenges in Virtual Prototyping
Methods in optiSLang
Example Application
-
3 © 2014 ANSYS, Inc. November 4, 2015
Outline
Challenges in Virtual Prototyping
Methods in optiSLang
Example Application
-
4 © 2014 ANSYS, Inc. November 4, 2015
In reality tolerances, uncertainties andvariability exist caused by environmental influences, load variation or human inaccuracy.
• These kind of unpredictable influences cause uncertainties in the desired product functionalities
• Goal: Understand the these scattering parameters guarantee the product functionalities of an optimized design in spite of these uncertainties
Why do we need Robust Design Optimization?
Uncertainties
-
6 © 2014 ANSYS, Inc. November 4, 2015
Design Exploration
• Simulation shows result for certain design point
• What happens if parameter are changed?
• Response Surface shows result for parameter space
Single Point What If?Response Surface
?
??
-
7 © 2014 ANSYS, Inc. November 4, 2015
Outline
Challenges in Virtual Prototyping
Methods in optiSLang
Example Application
-
8 © 2014 ANSYS, Inc. November 4, 2015
Start
Methodology Definition
-
9 © 2014 ANSYS, Inc. November 4, 2015
Start
Methodology Definition
-
10 © 2014 ANSYS, Inc. November 4, 2015
Optimization Strategy
• Design Optimization
– Gradient Based
– Generic
– Evolutionary
– …
• Design of Experiments
– Data Sampling
– Detecting Correlations
– Detecting Important Parameters
– Parameter Space Reduction
– Response Surface
• Design Optimization
– …
General Procedure:
-
13 © 2014 ANSYS, Inc. November 4, 2015
Design of Experiments, Sampling
Systematic Sampling
Monte Carlo
SamplingLatin
Hypercube
Sampling
“Random
Correlation”
Low effort for
high number
of parameter
-
16 © 2014 ANSYS, Inc. November 4, 2015
Linear,quadratic
RegressionMoving Least Square
Basic PointsTest Points
Meta-Model of Optimal Prognosis, MoP
2N
k k2
ˆy y
k 1
ˆ ˆY YY Y
ˆy yˆY YCoP
N 1
-
18 © 2014 ANSYS, Inc. November 4, 2015
Design Optimization
Gradient-Based
Algorithms
Evolutionary
Algorithm
Pareto Optimization
Adaptive
Response Surface
Generic
Algorithm
Optimization
Algorithms:
Which one is the best?
-
19 © 2014 ANSYS, Inc. November 4, 2015
Adaptive Response Surface Method
• Start Point
• Initial Sample
– Approximated Response Surface
– Best Point
– New Sample with smaller Range
– …
Input 1
Input 1
Output
Inp
ut
2
-
20 © 2014 ANSYS, Inc. November 4, 2015
Evolutionary Algorithms
-
21 © 2014 ANSYS, Inc. November 4, 2015
Pareto Optimization
• Initial Generation
– Select best
• Second Generation
– Select best
• Third Generation
– Select best
• Fourth Generation
– Select best
• …
Ob
jecti
ve A
Objective B
-
22 © 2014 ANSYS, Inc. November 4, 2015
Start
Methodology Definition
-
23 © 2014 ANSYS, Inc. November 4, 2015
X1X5X4
Robustness Evaluation
User Interaction
Input Parameter Variation
Statistical LHS-
Sampling
Output Parameter Variation
Check CoP/MoP
-
24 © 2014 ANSYS, Inc. November 4, 2015
Outline
Challenges in Virtual Prototyping
Methods in optiSLang
Example Application
-
25 © 2014 ANSYS, Inc. November 4, 2015
Maxwell 2D Setup
We need a valid setup including Electrical Machine Toolkit UDO‘s
-
26 © 2014 ANSYS, Inc. November 4, 2015
Setup in OptiSlang
Use graphical description of dataflow in OptiSlang
Design of Experiments
Build mathematical model
Optimisation
-
27 © 2014 ANSYS, Inc. November 4, 2015
Use script to dump out all parameters
-
28 © 2014 ANSYS, Inc. November 4, 2015
Define min and max values for parameters
-
29 © 2014 ANSYS, Inc. November 4, 2015
Define Outputs
-
30 © 2014 ANSYS, Inc. November 4, 2015
Edit script to call Maxwell
-
31 © 2014 ANSYS, Inc. November 4, 2015
-
32 © 2014 ANSYS, Inc. November 4, 2015
-
33 © 2014 ANSYS, Inc. November 4, 2015
-
34 © 2014 ANSYS, Inc. November 4, 2015
-
35 © 2014 ANSYS, Inc. November 4, 2015
-
36 © 2014 ANSYS, Inc. November 4, 2015
-
37 © 2014 ANSYS, Inc. November 4, 2015
Run Sensitivity
-
38 © 2014 ANSYS, Inc. November 4, 2015
Build a Metamodel
-
39 © 2014 ANSYS, Inc. November 4, 2015
Metamodel Data
-
40 © 2014 ANSYS, Inc. November 4, 2015
Run Optimizer
-
41 © 2014 ANSYS, Inc. November 4, 2015
Time Domain VerificationCritical Net electronics
-
42 © 2014 ANSYS, Inc. November 4, 2015
PCB Transmission Channels
-
43 © 2014 ANSYS, Inc. November 4, 2015
PCB Structures
-
44 © 2014 ANSYS, Inc. November 4, 2015
Parametric results
-
45 © 2014 ANSYS, Inc. November 4, 2015
DSO runs multiple designs in parallel
-
46 © 2014 ANSYS, Inc. November 4, 2015
Minimization of the Resonance Frequency
• Patch divided into "bits"
• Bits turned on and off
“1001010010101”
-
47 © 2014 ANSYS, Inc. November 4, 2015
Minimization of the Resonance Frequency
• Resonance Definition: Peak of the Re[Zin]
• Optimization does not include the location of the feed point
• Cases investigated:– Symmetry NOT imposed– Symmetry imposed in the y-axis
– Two different level of discretization: • 21 by 21
• 41 by 41
-
48 © 2014 ANSYS, Inc. November 4, 2015
Symmetry
Asymmetric Patch Symmetric Patch
-
49 © 2014 ANSYS, Inc. November 4, 2015
Animation of GeometricSolution Convergence
• 21 X 21 Symmetric Patch
-
50 © 2014 ANSYS, Inc. November 4, 2015
850 MHz1900 MHz2400 MHz
.85 GHz 1.85 GHz
Current Distribution on Antenna Elements
Watch Performance on Human Hand
Courtesy Mike Schaaf, Synapse
Successful Smart Watch Design Using ANSYS