Robust Design - Dynardo GmbH
Robust Design - Dynardo GmbH Robust Design - Dynardo GmbH
DYNARDO • © Dynardo GmbH 2012 Robustness & Reliability Analysis 1
- Page 2 and 3: DYNARDO • © Dynardo GmbH 2012 Ro
- Page 4 and 5: DYNARDO • © Dynardo GmbH 2012 Ro
- Page 6 and 7: DYNARDO • © Dynardo GmbH 2012 Un
- Page 8 and 9: DYNARDO • © Dynardo GmbH 2012 De
- Page 10 and 11: DYNARDO • RDO forming simulation
- Page 12 and 13: DYNARDO • © Dynardo GmbH 2012 Im
- Page 14 and 15: DYNARDO • © Dynardo GmbH 2012 Wh
- Page 16 and 17: DYNARDO • © Dynardo GmbH 2012 Ad
- Page 18 and 19: DYNARDO • © Dynardo GmbH 2012 Ro
- Page 20 and 21: DYNARDO • ECCOMAS 2012 congress
- Page 22 and 23: DYNARDO • © Dynardo GmbH 2012 Wh
- Page 24 and 25: DYNARDO • © Dynardo GmbH 2012 St
- Page 26 and 27: DYNARDO • © Dynardo GmbH 2012 So
- Page 28 and 29: DYNARDO • © Dynardo GmbH 2012 So
- Page 30 and 31: DYNARDO • © Dynardo GmbH 2012 St
- Page 32 and 33: DYNARDO • © Dynardo GmbH 2012 Ro
- Page 34 and 35: DYNARDO • © Dynardo GmbH 2012 Su
- Page 36 and 37: DYNARDO • © Dynardo GmbH 2012 Ro
- Page 38 and 39: DYNARDO • © Dynardo GmbH 2012 Ro
- Page 40 and 41: DYNARDO • © Dynardo GmbH 2012 Re
- Page 42 and 43: DYNARDO • © Dynardo GmbH 2012 De
- Page 44 and 45: DYNARDO • © Dynardo GmbH 2012 Ho
- Page 46 and 47: DYNARDO • © Dynardo GmbH 2012 Pr
- Page 48 and 49: DYNARDO • © Dynardo GmbH 2012 De
- Page 50 and 51: DYNARDO • © Dynardo GmbH 2012 Ro
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness & Reliability Analysis<br />
1
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong> <strong>Design</strong> Optimization<br />
<strong>Robust</strong> <strong>Design</strong><br />
Variance based<br />
<strong>Robust</strong>ness Evaluation<br />
Probability based<br />
<strong>Robust</strong>ness Evaluation,<br />
(Reliability analysis)<br />
Optimization<br />
Sensitivity Study<br />
CAE process (FEM, CFD, MBD, Excel, Matlab, etc.)<br />
<strong>Robust</strong> <strong>Design</strong> Optimization Methodology<br />
Start<br />
Single & Multi objective<br />
(Pareto) optimization<br />
2
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
How to define <strong>Robust</strong>ness of a <strong>Design</strong><br />
• Intuitively: The performance of a robust design is largely<br />
unaffected by random perturbations<br />
• Variance indicator: The coefficient of variation (CV)<br />
of the objective function and/or constraint values is smaller than<br />
the CV of the input variables<br />
• Sigma level: The interval mean+/- sigma level does not reach an<br />
undesired performance<br />
(e.g. design for six-sigma)<br />
• Probability indicator: The probability of reaching undesired<br />
performance is smaller than an acceptable value<br />
3
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation / Reliability Analysis<br />
• Variance-based robustness evaluation<br />
measure product serviceability<br />
– Safety and reliability for 1 & 2 (3)<br />
Sigma levels and identifies the most<br />
sensitive stochastic variables<br />
– Possible with high number of<br />
stochastic variables<br />
• Probability-based robustness evaluation<br />
(reliability analysis) measure product<br />
serviceability<br />
– Safety and reliability for high<br />
reliability levels (3,4,5,6-Sigma) with<br />
small number of variables<br />
– Possible with a limited number of<br />
stochastic variables<br />
4
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Definition of Uncertainties<br />
5
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Uncertainties and Tolerances<br />
• <strong>Design</strong> variables<br />
• Material, geometry, loads,<br />
constrains,…<br />
• Manufacturing<br />
• Operating processes (misuse)<br />
• Resulting from Deterioration<br />
• …<br />
Property SD/Mean<br />
%<br />
Metallic materiales, yield 15<br />
Carbon fiber rupture 17<br />
Metallic shells, buckling<br />
strength<br />
6<br />
14<br />
Bond insert, axial load 12<br />
Honeycomb, tension 16<br />
Honeycomb, shear, compression 10<br />
Honeycomb, face wrinkling 8<br />
Launch vehicle , thrust 5<br />
Transient loads 50<br />
Thermal loads 7.5<br />
Deployment shock 10<br />
Acoustic loads 40<br />
Vibration loads 20<br />
Klein, Schueller et.al. Probabilistic Approach to Structural Factors of Safety in Aerospace.<br />
Proc. CNES Spacecraft Structures and Mechanical Testing Conf., Paris 1994
DYNARDO • ECCOMAS 2012 congress• © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Definition of Uncertainties<br />
Translate know how about uncertainties into proper scatter definition<br />
Distribution functions define<br />
variable scatter<br />
Correlation is an important<br />
characteristic of stochastic<br />
variables.<br />
Yield stress<br />
Tensile strength<br />
Correlation of single uncertain values<br />
spatial correlated<br />
thickness distribution<br />
7
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Design</strong> “single” scatter shapes<br />
• Use of CAD-parameter or mesh morphing functions to<br />
design “single” scatter shapes<br />
Measured ed imperf imperfection modelled imperfection<br />
Imperfection of cylindricity of truck wheel component<br />
by courtesy of<br />
Suchanek, J.; Will, J.: Stochastik analysis as a method to evaluate the robustness of light<br />
truck wheel pack; Proceedings WOSD 6.0, 2009, Weimar, Germany, www.dynardo.de]<br />
8
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Random Field Parametric<br />
• Introduction of scatter of spatially correlated scatters need parametric of<br />
scatter shapes using random field theory.<br />
The correlation function represents the<br />
measure of “waviness” of random fields.<br />
The infinite correlation length reduced the<br />
random field to a simple random variable.<br />
Usually, there exist multiple scatter shapes<br />
representing different scatter sources.<br />
9
DYNARDO • RDO forming simulation • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Use forming simulation to generate scatter fields<br />
4. Mapping to crash mesh<br />
3. Generation of n-imperfect formed<br />
parts using scatter field parametric<br />
2. Identification of most important<br />
scatter fields for scattering forming<br />
result values (thickness,<br />
hardening)<br />
5. Run <strong>Robust</strong>ness Evaluation of<br />
assembly including scatter from forming<br />
1. Running <strong>Robust</strong>ness<br />
evaluation of forming simulation<br />
10
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Use measurements to generate Random Fields<br />
3. Identification of most<br />
important random fields for<br />
measured quantity (geometry)<br />
4. Generate imperfect geometries using<br />
Random Field parametric<br />
2. Trimming of initial FE-mesh regarding<br />
measurement or realization<br />
5. Run <strong>Robust</strong>ness Evaluation to identify<br />
the most sensitive random fields<br />
•<br />
•<br />
• •<br />
•<br />
•<br />
•<br />
••<br />
•<br />
•<br />
•<br />
•<br />
•<br />
1. Multiple Measurements<br />
from Manufacturing<br />
11
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Implementation of Random Field Parametric<br />
Introduction of spatial correlated scatter to CAE-Parameter (geometry,<br />
thickness, plastic astic values)<br />
3. Generation of multiple imperfect<br />
structures using Random Field parametric<br />
2. Generation of scatter shapes using<br />
Random field parametric, quantify scatter<br />
shape importance<br />
4. Running <strong>Robust</strong>ness Evaluation<br />
including Random Field effects<br />
1. Input: process simulation or<br />
measurements<br />
12
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Variance-based <strong>Robust</strong>ness Analysis<br />
13
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Which <strong>Robust</strong>ness do You Mean?<br />
<strong>Robust</strong>ness evaluation due to naturally given scatter<br />
Goal: measurement of variation and correlation<br />
Methodology: variance-based robustness evaluation<br />
Positive side effect of robustness evaluation: The measurement of<br />
prognosis quality of the response variation answer the question - Does<br />
numerical scatter significantly influence the results?<br />
14
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Statistic Measurements<br />
• Evaluation of robustness with statistical<br />
measurements<br />
– Variation analysis (histogram,<br />
coefficient of variation, standard<br />
variation, distribution fit,<br />
probabilities)<br />
– Correlation analysis (linear,<br />
quadratic, nonlinear) including<br />
principal component analysis<br />
– Forecast quality: Evaluation of<br />
Coefficient of Prognosis (CoP)<br />
including coefficients of<br />
determination (CoD)and coefficient<br />
of importance (CoI)<br />
15
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Advanced Statistic Measurements<br />
• Advanced histogram options<br />
– PDF fit (automatic/manual)<br />
– Number of histogram classes<br />
– PDF values ready for optiSLang<br />
input<br />
– Limits with probabilities<br />
– Probabilities with limit<br />
16
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Standardized and Automated Post Processing<br />
Example how the post<br />
processing is<br />
automated for passive<br />
safety at BMW<br />
The maximum from<br />
the time signal was<br />
taken.<br />
17
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation using optiSLang<br />
1) Define the robustness space using<br />
scatter range, distribution and<br />
correlation<br />
5) Identify the most<br />
important scattering<br />
variables<br />
4) Check the CoI/CoP CoI<br />
2) Scan the robustness space by<br />
producing and evaluating n<br />
(100) <strong>Design</strong>s<br />
3) 3) Check the variation interval<br />
18
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation of NVH Performance<br />
Start in 2002, since 2003 used for Production Level<br />
How does body and suspension system scatter influence the NVH<br />
performance?<br />
• Consideration of scatter of body in<br />
white, suspension system<br />
• Prognosis of response value scatter<br />
• Identify correlations due to the input<br />
scatter<br />
• CAE-Solver: NASTRAN<br />
• Up-to-date robustness evaluation of<br />
body in white have 300 .. 600<br />
scattering variables<br />
• Using filter technology to optimize the<br />
number of samples<br />
by courtesy of<br />
Will, J.; Möller, J-St.; Bauer, E.: <strong>Robust</strong>ness evaluations of the NVH comfort using full vehicle models<br />
by means of stochastic analysis, VDI-Berichte Nr.1846, 2004, S.505-527, www.dynardo.de<br />
19
DYNARDO • ECCOMAS 2012 congress• © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation of break systems<br />
Start in 2007, since 2008 used for Production Level<br />
How does material and geometric scatter influence the break noise<br />
performance ?<br />
• Consideration of material<br />
and geometry scatter<br />
• Break noise is a instability<br />
problem<br />
• Identify correlations due to<br />
the input scatter<br />
• CAE-Solver: NASTRAN,<br />
ABAQUS, ANSYS<br />
• Up-to-date robustness<br />
evaluation have 20 ..30<br />
scattering variables<br />
• Integration of geometric<br />
scatter via random fields<br />
1. Define the<br />
Uncertainties<br />
4. Post<br />
processing<br />
2. Simulate random m parameters<br />
p<br />
3. Generate set of random<br />
specimen, replace in FE assembly<br />
to compute<br />
by courtesy of<br />
Will, J.; Nunes, R.: <strong>Robust</strong>ness evaluations of the break system concerning squeal noise<br />
problem, Weimarer Optimierungs- und Stochastiktage 2009, www.dynardo.de<br />
X2<br />
20<br />
X1
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation of Forming Simulations<br />
• Consideration of process<br />
and material scatter<br />
• Determination of process<br />
robustness based on 3-<br />
Sigma-values of quality<br />
criteria<br />
• Projection and<br />
determination of<br />
statistical values on<br />
FE-structure<br />
necessary<br />
CAE-Solver: LS-DYNA,<br />
AUTOFORM and others<br />
Start in 2004 - since 2006 used for production level<br />
1. Variation der Eingangsstreuungen<br />
mittels geeigneter<br />
Samplingverfahren<br />
3. statistische Auswertung<br />
und <strong>Robust</strong>heitsbewertung<br />
Will, J.; Bucher, C.; Ganser, M.; Grossenbacher, K.: Computation and visualization of<br />
statistical measures on Finite Element structures for forming simulations; Proceedings<br />
Weimarer Optimierung- und Stochastiktage 2.0, 2005, Weimar, Germany<br />
2. Simulation inkl.<br />
Mapping<br />
auf einheitliches<br />
Netz<br />
by courtesy of<br />
21
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Why SoS and What is SoS ?<br />
Why: Engineers need to evaluate statistical Das Bild kann zurzeit nicht angezeigt werden. data on the structure to locate<br />
„hot spots“ of variation as well as investigate correlations<br />
What: A post processor for Statistics on finite element Structures<br />
• Visualization of descriptive statistics on the structure<br />
• Visualization of correlations between random input<br />
and structural results<br />
• Visualization of quality performance (QCS)<br />
• Identification of scatter shapes using Random Fields<br />
Key features:<br />
• locale „hot spots“ of variation<br />
• Data reduction and smoothing by mesh coarsening<br />
• Identification of relevant scatter shapes<br />
• Vizualisation statistics of eroded (failed) elements<br />
22
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Statistical post-processing on FE-meshes<br />
Motivation: - Visualization of statistical measurements<br />
- Identification of hot spots,<br />
- Identification of correlation structure (random fields)<br />
1. Import data<br />
(multiple simulation on runs<br />
or measurements) ments)<br />
3. Export local statistics<br />
and imperfection<br />
amplitudes to<br />
optiSLang for further ther<br />
post-processing<br />
2. Visualize variation, v correlation,<br />
identify ify random fi field shapes<br />
23
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Statistical post-processing on FE-meshes<br />
SoS is the tool to answer the questions:<br />
Where? Locate hot spots of highest variation and/or extreme values, which<br />
may cause lack of performance or quality.<br />
Why? Find the input parameters which cause scatter of the results, by<br />
analysing correlation/CoD between random inputs and spatially distributed<br />
results, export to optiSLang p g for MoP/CoP / analysis. y<br />
24
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
SoS application crash simulation<br />
• Stringer in a car body subject to crash simulation<br />
• 55 random inputs: sheet thickness and material parameters (also of<br />
other parts), load parameters (velocity, barrier angle etc.)<br />
• Observed result: remaining effective plastic strain<br />
[Bayer, V.; Will, J.: Random Fields in <strong>Robust</strong>ness and Reliability Assessment of Structural Parts.<br />
SIMVEC, Verein Deutscher Ingenieure VDI, Baden-Baden 2010.]<br />
25
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
SoS application crash simulation<br />
• Analysis of a car structure subject to<br />
random crash load case<br />
• Decomposition of plastic strain<br />
random field<br />
• Analysis of largest influence by<br />
MoP/CoP in optiSLang<br />
[Bayer, V.; Will, J.: Random Fields in <strong>Robust</strong>ness and Reliability Assessment of Structural Parts.<br />
SIMVEC, Verein Deutscher Ingenieure VDI, Baden-Baden 2010.]<br />
26
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
SoS - Eroding elements<br />
General eroding information.<br />
Green area: no elements eroded.<br />
Red area: there were eroded elements.<br />
Eroding percent plot shows the<br />
percentage of eroded elements.<br />
27
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
SoS – Validated Global and Local Statistics<br />
Variation, mean value, correlation, Histogram, linear & quadratic<br />
QCS using SoS on FE-structure correlation, CoD, CoI, CoP and<br />
Anthill plots using optiSLang at local<br />
nodal or element values<br />
Selection of nodal/element results in<br />
SoS and export to optiSLang *.bin<br />
CoD linear base<br />
Correlation<br />
R-value<br />
Cotrrelation<br />
Y-tolerance<br />
28
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
SoS - Applications<br />
- use SoS for robustness evaluation of forming processes<br />
- use SoS for visualization and hot spot investigation at robustness<br />
evaluations in crashworthiness or drop test applications<br />
- use SoS for the identification of random fields from simulation or<br />
measurements<br />
by courtesy of<br />
29
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Standardized and Automated Post Processing<br />
Productive Level needs standardized and automated post processing!<br />
1. Check variation of<br />
plasticity, failure,<br />
intrusions.<br />
3. Summarize variation<br />
and correlation<br />
2. Identify the beginning of the<br />
phenomena in time and use SoS to<br />
identify the source of variation<br />
30
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Summary SoS – Statistic on Structure<br />
The post processor for Statistics on finite element Structures<br />
- Statistic Measurements<br />
- Single <strong>Design</strong>s<br />
- Differences between <strong>Design</strong>s<br />
- Variation interval<br />
- Minimum/Maximum<br />
- Mean Value<br />
- Standard deviation<br />
- Coefficient of variation<br />
- Quantile (± ����<br />
- Correlation & CoD<br />
- Linear correlation & CoD<br />
- At nodal/element level<br />
- Process quality criteria Cp, Cpk<br />
process indices<br />
- Random field generation<br />
- Scatter shape extraction and<br />
visualisation<br />
31
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation of Passive Safety<br />
• Consideration of scatter of material<br />
and load parameters as well as test<br />
conditions<br />
• Prognosis of response value<br />
variation = is the design robust!<br />
• Identify correlations due to the input<br />
scatter<br />
• Quantify the amount of numerical<br />
noise<br />
• CAE-Solver: MADYMO, ABAQUS<br />
Start in 2004, since 2005 used<br />
for productive level<br />
Goal: Ensuring Consumer<br />
Ratings and Regulations &<br />
Improving the <strong>Robust</strong>ness of a<br />
System<br />
Will, J.; Baldauf, H.: Integration of Computational <strong>Robust</strong>ness Evaluations in Virtual<br />
Dimensioning of Passive Passenger Safety at the BMW AG , VDI-Berichte Nr. 1976, 2006,<br />
Seite 851-873, www.dynardo.de<br />
by courtesy of<br />
32
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Evaluation Crashworthiness<br />
Start in 2004 – since 2007 use for Production Level<br />
• Consideration of scatter of thickness,<br />
strength, geometry, friction and test<br />
condition<br />
• CAE-Solver: LS-DYNA, Abaqus<br />
• Prognosis of intrusions, failure and<br />
plastic behavior<br />
• Identify Coefficient of Prognosis and<br />
nonlinear correlations<br />
• Check model robustness<br />
• 100 .. 200 scattering variables<br />
• Introduction of forming scatter via<br />
Random Fields<br />
by courtesy of the Daimler AG<br />
In comparison to robustness evaluations for NVH, forming or passive<br />
safety, crashworthiness has very high demands on methodology and<br />
software!<br />
by courtesy of<br />
Will, J.; Frank, T.: <strong>Robust</strong>ness Evaluation of crashworthiness load cases at Daimler AG;<br />
Proceedings Weimarer Optimierung- und Stochastiktage 5.0, 2008, Weimar, Germany,<br />
www.dynardo.de<br />
33
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Summary <strong>Robust</strong>ness Evaluation<br />
• optiSLang + SoS have completed the necessary methodology to run<br />
robustness evaluation for NVH, passive safety, forming simulation or<br />
crashworthiness<br />
Success Key:<br />
• Necessary distribution types and correlation definitions<br />
available<br />
• Optimized LHS sampling<br />
• Reliable measurements of variation, correlation and determination<br />
including filter technology<br />
• Projection of statistic onto the FE-structure<br />
Main customer benefit:<br />
• Identification of problems early in the virtual prototyping stage<br />
• Measure, verify and finally significantly improve the modeling quality<br />
(reduce numerical scatter and modeling errors)<br />
34
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
What is necessary for successful implementation?<br />
1. Introduction of realistic<br />
scatter definitions<br />
� Distribution function<br />
� Correlations<br />
� Random fields<br />
2. Using of reliable stochastic methodology<br />
methodology<br />
� Variance-based robustness<br />
evaluation using optimized LHS<br />
3. Development of reliable robustness<br />
measurements<br />
� Standardized post processing<br />
� Significance filter<br />
� Reliable variation and correlation<br />
measurements<br />
Acceptance of method/result documentation/communication!<br />
35
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness and Stability of the Model<br />
Which quantity of „numerical noise“ is acceptable?<br />
� Quantification via coefficients of prognosis (CoP)<br />
� Estimation of numerical noise: 100% - CoP<br />
Experience in passive safety, CFD or<br />
crashworthiness tells that result values<br />
with lower CoP than 80% show:<br />
- High amount of numerical noise<br />
resulting from numerical approximation<br />
method (meshing, material, contact,..)<br />
- Problems of result extractions<br />
- Physically instable behavior<br />
36
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Insurance Crash Loading<br />
Response CoV<br />
CoD<br />
lin[%]<br />
CoD lin<br />
adj [%]<br />
CoD<br />
quad [%]<br />
CoD<br />
quad adj<br />
[%]<br />
max_int_energy 0.036 99.0 99.0 100.0 99.0<br />
max_hourglass_energy 0.066 91.0 87.0 98.0 94.0<br />
x_1129403_1129404 0.217 54.0 31.0 93.0 79.0<br />
x_4000558_1129403 1.383 86.0 79.0 93.0 80.0<br />
x_4139132 0.123 89.0 83.0 97.0 91.0<br />
x_4144932 0.131 88.0 82.0 97.0 90.0<br />
y_4139132 0.224 92.0 89.0 98.0 93.0<br />
y_4144932 0.252 95.0 93.0 98.0 95.0<br />
z_4139132 0.348 33.0 0.015 90.0 71.0<br />
z_4144932 1.035 49.0 24.0 91.0 74.0<br />
max_sec29_rforce 0.032 87.0 80.0 96.0 88.0<br />
max_sec29_xforce 0.032 86.0 78.0 96.0 87.0<br />
max_sec33_rforce 0.034 89.0 84.0 97.0 90.0<br />
max_sec33_xforce 0.034 90.0 84.0 97.0 90.0<br />
max_plast 0.211 88.0 82.0 95.0 86.0<br />
summ_plast 0.26 91.0 87.0 97.0 92.0<br />
In qualified FE-models numerical scatter is not<br />
dominating important response values!<br />
Model guidelines, model verification and robustness<br />
evaluation ensure a reasonable model quality.<br />
LS-DYNA low speed structural<br />
crash case<br />
<strong>Robust</strong>ness evaluation against<br />
structure and loading scatter<br />
show coefficients between 71<br />
and 99 %<br />
37
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness check of optimized designs<br />
• With the availability of parametric modeling environments like<br />
ANSYS workbench an robustness check becomes very easy!<br />
• Menck see hammer for oil and gas exploration (up to 400m deep)<br />
• <strong>Robust</strong>ness evaluation against tolerances, material scatter and working<br />
and environmental conditions<br />
• 60 scattering parameter<br />
<strong>Design</strong> Evaluations: 100<br />
Process chain: ProE-ANSYS workbench- optiSLang<br />
38
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness evaluation as early as possible<br />
Goal: Tolerance check before any<br />
hardware exist!<br />
Classical tolerance analysis tend to be very<br />
conservative<br />
<strong>Robust</strong>ness evaluation against production<br />
tolerances and material scatter (43<br />
scattering parameter) shows:<br />
- Press fit scatter is o.k.<br />
- only single tolerances are important (high<br />
cost saving potentials)<br />
Production shows good agreement!<br />
<strong>Design</strong> Evaluations: 150<br />
solver:<br />
ANSYS/optiSLang<br />
by courtesy of<br />
Suchanek, J.; Will, J.: Stochastik analysis as a method to evaluate the robustness of light<br />
truck wheel pack; Proceedings WOSD 6.0, 2009, Weimar, Germany, www.dynardo.de<br />
39
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Reliability Analysis<br />
40
© <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Reliability analysis<br />
• Limit state function g(x) divides random variable space X<br />
in safe domain g(x)>0 and failure domain g(x) ��<br />
• Multiple failure criteria (limit state functions) are possible<br />
• Failure probability is the probability that at least one failure criteria is<br />
violated (at least one limit state function is negative)<br />
• Integration of joint probability density function over failure domain<br />
41
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Definition of limit state functions<br />
<strong>Dynardo</strong> • optiSlang Seminar<br />
Example damped oscillator<br />
42
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Reliability Analysis Algorithms<br />
Gradient-based<br />
algorithms = First<br />
Order Reliability<br />
algorithm (FORM)<br />
ISPUD Importance<br />
Sampling using <strong>Design</strong><br />
Point<br />
Adaptive Response<br />
Surface Method<br />
Monte Carlo Sampling Latin Hypercube Sampling Directional Sampling<br />
X2<br />
X1<br />
43
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
How choosing the right algorithm?<br />
<strong>Robust</strong>ness Analysis provide the<br />
knowledge to choose the<br />
appropriate algorithm<br />
<strong>Robust</strong>ness & Reliability Algorithms<br />
44
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Application Example ARSM for Reliability<br />
• Fatigue life analysis of Pinion shaft<br />
• Random variables<br />
• Surface roughness<br />
• Boundary residual stress<br />
• Prestress of the shaft nut<br />
• Target: calculate the probability of<br />
failure<br />
• Probability of Failure:<br />
• Prestress I: P(f)=2.3 10 -4 (230<br />
ppm)<br />
• Prestress II: P(f)=1.3 10 -7 (0.13<br />
ppm)<br />
by courtesy of<br />
Solver: Permas<br />
Method: ARSM<br />
75 Solver evaluations<br />
sigma = +/-5kN<br />
sigma = +/-10kN<br />
45
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Probability Assessment of Railways<br />
• Goal: investigate the crosswind stability of Railway<br />
systems<br />
• Sensitivity analysis against stochastic loading<br />
parameters followed by probability analysis<br />
• Random variables<br />
• crosswind amplitude, duration<br />
• aerodynamic coefficients<br />
Solver: ADAMS + Matlab<br />
Methods: LHS, FORM, ISPUD<br />
Proppe, C.Wetzel, C.; Probabilitic Assessment of the Crossswind Stability of Railway<br />
Systems; Proceedings Weimarer Optimierung- und Stochastiktage 4.0, 2007, Weimar,<br />
Germany, www.dynardo.de<br />
46
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong> <strong>Design</strong> Optimization<br />
47
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Design</strong> for Six Sigma<br />
• Six Sigma is a concept to optimize the manufacturing processes such<br />
that automatically parts conforming to six sigma quality are<br />
produced<br />
• <strong>Design</strong> for Six Sigma is a concept to optimize the design such that<br />
the parts conform to six sigma quality, i.e. quality and reliability are<br />
explicit optimization goals<br />
• Because not only 6 Sigma values have to be used as measurement<br />
for a robust design, we use the more general classification <strong>Robust</strong><br />
<strong>Design</strong> Optimization<br />
48
© <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
© <strong>Dynardo</strong> <strong>GmbH</strong><br />
Sigma level vs. failure probability<br />
• The sigma level can be used to calculate the probability of exceeding<br />
a certain response value<br />
• Since the distribution type is often unknown, this estimate may be<br />
very inaccurate for small probabilities<br />
• The sigma level deals with single limit values, whereas the failure<br />
probability quantifies the event, that any of several limits is exceeded<br />
� Reliability analysis should be applied to proof the required safety level<br />
Distribution Required sigma level (CV=20%)<br />
p F = 10 -2 p F = 10 -3 p F = 10 -6<br />
Normal 2.32 3.09 4.75<br />
Log-normal 2.77 4.04 7.57<br />
Rayleigh 2.72 3.76 6.11<br />
Weibull 2.03 2.54 3.49<br />
49
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong> <strong>Design</strong> Optimization<br />
<strong>Robust</strong> <strong>Design</strong><br />
Variance based<br />
<strong>Robust</strong>ness Evaluation<br />
Probability based<br />
<strong>Robust</strong>ness Evaluation,<br />
(Reliability analysis)<br />
Optimization<br />
Sensitivity Study<br />
CAE process (FEM, CFD, MBD, Excel, Matlab, etc.)<br />
<strong>Robust</strong> <strong>Design</strong> Optimization Methodology<br />
Start<br />
Single & Multi objective<br />
(Pareto) optimization<br />
50
<strong>Robust</strong> <strong>Design</strong><br />
<strong>Robust</strong> <strong>Design</strong> Optimization (RDO) optimize the design performance<br />
with consideration of scatter of design (optimization) variables as<br />
well as other tolerances or uncertainties.<br />
As a consequence of uncertainties the location of the optima as well as<br />
the contour lines of constraints scatters.<br />
To proof <strong>Robust</strong> <strong>Design</strong>s safety distances are quantified with variance<br />
or probability measurements using stochastic analysis.<br />
51
Methods for <strong>Robust</strong> <strong>Design</strong> Optimization<br />
Variance-based RDO<br />
• Safety margins of all critical responses<br />
are larger than a specified sigma level<br />
(e.g. <strong>Design</strong> for Six Sigma)<br />
Reliability-based RDO<br />
• Failure probability with respect to given<br />
limit states is smaller as required value<br />
Taguchi-based RDO<br />
• Taguchi loss functions<br />
• Modified objective function<br />
52
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong> <strong>Design</strong> Optimization - RDO<br />
<strong>Robust</strong> <strong>Design</strong> Optimization combines optimization and <strong>Robust</strong>ness<br />
Evaluation. From our experience it is often necessary to investigate both<br />
domains separately to be able to formulate a RDO problem. optiSLang<br />
offers you either iterative or automatic RDO flows.<br />
<strong>Robust</strong>ness <strong>Robust</strong>ne evaluation<br />
SSensitivity it analysis<br />
Define safety factors<br />
<strong>Robust</strong> design optimization<br />
<strong>Robust</strong>ness proof!<br />
53
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Iterative RDO Application Connector<br />
2) The DX Six Sigma design was checked in<br />
the space of 36 scattering variables using<br />
optiSLang <strong>Robust</strong>ness evaluation. Some<br />
Criteria show high failure probabilities!<br />
1) From the 31 optimization<br />
parameter the most effective<br />
one are selected with<br />
optiSLang Sensitivity<br />
analysis.<br />
3) From optiSLang <strong>Robust</strong>ness<br />
Evaluation safety margins are<br />
derived.<br />
4) Three steps of optimization<br />
using optiSLang ARSM and EA<br />
optimizer optimizer improve the design to<br />
an optiSLang Six sigma design.<br />
5) Reliability proof using ARSM<br />
to account the failure probability<br />
did proof six sigma quality.<br />
Start: Optimization using 5 Parameter using DX Six Sigma, then customer<br />
asked: How save is the design?<br />
by courtesy of<br />
54
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
Using <strong>Design</strong>Modeler parametric<br />
was rebuild with 11 Parameter.<br />
Using optiSLang sensitivity<br />
analysis 7 effective parameter<br />
are identified. Using ARSM<br />
design was optimized to have<br />
minimal volume and sufficient<br />
pressure at all contact points.<br />
Initial design<br />
Including Uncertainty of<br />
loading and material 6<br />
Sigma quality was violated.<br />
Therefore iterative RDO<br />
increasing safety margins<br />
was performed to reach 3<br />
Sigma design level.<br />
<strong>Robust</strong> design<br />
Using optiSLang Reliability<br />
Analysis 3 sigma quality was<br />
proven.<br />
Customer started optimization with two parameter and found minimal improvement<br />
potential.<br />
55
DYNARDO • NAFEMS conference 2012 • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Summary<br />
• Stochastic analysis needs a balance between input definitions,<br />
stochastic analysis method and post processing<br />
• Use specific/concrete robustness/reliability measurements<br />
• RDO using one global response surface approximation for<br />
optimization and robustness space usually underrepresents the<br />
influence of scattering variables<br />
• Because all RDO strategies will try to minimize solver runs for<br />
robustness measures, a final proof of robustness/reliability is<br />
mandatory<br />
• Carefully translation and introduction of material scatter is crucial<br />
• Start with robustness evaluation of start/optimal candidate design<br />
• Continue with iterative RDO approach using safety distances<br />
• If safety distances in optimization space vary significant automatic<br />
RDO is necessary<br />
56
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Part 7<br />
Applications<br />
57
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Application Crashworthiness<br />
AZT Insurance Crash Load Case<br />
• Scatter definition (40..60 scattering<br />
parameter)<br />
– Velocity, barrier angle and position<br />
– Friction (Road to Car, Car to<br />
Barrier)<br />
– Yield strength<br />
– Spatially correlated sheet metal<br />
thickness<br />
• Main result: Prognosis of plastic<br />
behavior<br />
• CAE-Solver: LS-DYNA<br />
Deterministic analysis show no problems with an AZT load case. Tests<br />
frequently show plastic phenomena which Daimler would like to minimize.<br />
Motivation for the robustness evaluation was to find the test phenomena<br />
in the scatter bands of robustness evaluations, to understand the sources<br />
and to improve the robustness of the design.<br />
58
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Did You Include All Important Scatter?<br />
Scatter of uniform sheet<br />
thickness thickness (cov=0.05),<br />
yield strength, friction, test est<br />
conditions<br />
We could not find<br />
or explain the test<br />
results!<br />
Introduction of sheet metal<br />
thickness scatter per part<br />
- 100 LS-DYNA simulation<br />
- Extraction via LS-PREPOST<br />
SoS - post processing<br />
Statistics_on_Structure<br />
59
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Definition of Scatter is the Essential Input!<br />
Which degree of forming scatter discretization is becomes necessary?<br />
Level 1 - No distribution information: - increase uniform coil thickness<br />
scatter cov=0.02 to cov=0.03..0.05<br />
Level 2 - Use deterministic distribution information: - use thickness<br />
reduction shape from deterministic forming simulation and superpose coil<br />
(cov=0.02) and forming process scatter (cov=0.01..0.03)<br />
60
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Did You Include All Important Scatter?<br />
Scatter of sheet<br />
thickness, forming<br />
process scatter<br />
covmax=0.05<br />
yield strength, friction,<br />
n,<br />
test conditions<br />
We could find and<br />
explain the test results!<br />
+<br />
Introduction of spatial correlated<br />
forming process scatter<br />
- 100 LS-DYNA simulation<br />
- Extraction via LS-PREPOST<br />
SoS<br />
- post prozessing<br />
Statistics_on_Structure<br />
61
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Iterative <strong>Robust</strong> <strong>Design</strong> Optimization<br />
• <strong>Design</strong> optimization task<br />
• 4 continuous, 2 binary design<br />
variables<br />
• Single objective formulations<br />
(maximal removal power)<br />
• 2 optimization restrictions<br />
• <strong>Robust</strong>ness of performance should<br />
be preserved<br />
by courtesy of<br />
Könning, M; Plotzitza, A.: Optimierung und <strong>Robust</strong>heitsbewertung mit optiSLang bei der<br />
Robert Bosch <strong>GmbH</strong>, Proceeding Weimarer Optimierung- und Stochastiktage 1.0, 2004,<br />
www.dynardo.de<br />
62
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Iterative <strong>Robust</strong> <strong>Design</strong> Optimization<br />
• With help of the sensitivity analysis,<br />
the optimization task was<br />
formulated.<br />
• With help of genetic algorithms, the<br />
optimization was performed.<br />
• With robustness evaluation,<br />
robustness and reliability were<br />
checked.<br />
• The performance was increased by<br />
25 %!<br />
• The results of the digital product<br />
development were proven by<br />
testing!<br />
63
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
RDO Centrifugal Compressor<br />
Parameterization<br />
Parametric geometry definition using ANSYS BladeModeler<br />
(17 geometric parameter)<br />
Model completion and meshing using ANSYS Workbench<br />
by courtesy of<br />
64
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
RDO Centrifugal Compressor<br />
Fluid Structure Interaction (FSI) coupling<br />
Parametric fluid simulation setup using ANSYS CFX<br />
Parametric mechanical setup using ANSYS Workbench<br />
by courtesy of<br />
65
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
RDO Centrifugal Compressor<br />
Optimization goal: increase efficiency<br />
Constraints: 2 pressure ratio’s, 66 frequency constraints, <strong>Robust</strong>ness<br />
Initial SA ARSM I EA I ARSM II ARSM III<br />
Total Pressure Ratio 1.3456 1.3497 1.3479 1.3485 1.356 1.351<br />
Efficiency [%] 86.72 89.15 90.62 90.67 90.76 90.73<br />
#<strong>Design</strong>s - 100 105 84 62 40<br />
by courtesy of<br />
Input Parameter 21<br />
Output Parameter 43<br />
Constraints 68<br />
Tolerance<br />
limit<br />
1.34
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
RDO Centrifugal Compressor<br />
<strong>Robust</strong> <strong>Design</strong> Optimization with respect to 21 design parameters and 20<br />
random geometry parameters, including manufacturing tolerances. <strong>Robust</strong><br />
<strong>Design</strong> was reached after 400+250=650 design evaluations consuming.<br />
<strong>Robust</strong>ness evaluation<br />
Sensi + first optimization step<br />
by courtesy of<br />
RDO optimization<br />
<strong>Robust</strong>ness proof using Reliability<br />
Analysis<br />
67
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
Parametric model and Sensitivity Study<br />
• Customer investigated Seal optimization using only 2 parameter<br />
• Using <strong>Design</strong>Modeler parametric was rebuild with 11 parameter<br />
Initial design<br />
Sensitivity study<br />
• Using optiSLang sensitivity approach 7 effective optimization<br />
parameter are identified<br />
68
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
Deterministic optimization<br />
Using optiSLang ARSM algorithm<br />
design was optimized to have minimal<br />
volume and sufficient pressure at all<br />
contact points.<br />
Initial design<br />
Including Uncertainty of loading and<br />
material 6 Sigma quality was violated.<br />
Therefore iterative RDO increasing safety<br />
margins of contact forces was performed<br />
to reach 3 Sigma design level.<br />
Optimal design<br />
69
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
Iterative <strong>Robust</strong> <strong>Design</strong> Optimization<br />
• Required sigma level = 3 sigma<br />
• 4 iterations of sequential optimization and robustness analysis<br />
• Deterministic optimum (140 design evaluations): 14% failure<br />
• <strong>Robust</strong> optimum (540 design evaluations): 0.3% failure<br />
• Young’s modulus is most important scattering variable<br />
70
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
Iterative RDO – Reliability proof<br />
• 7 geometry + 2 material =9 important parameters detected<br />
from robustness analysis<br />
Directional Sampling on ARSM<br />
• FORM failed<br />
• ARSM: 116 design evaluations<br />
• Failure probability<br />
= 0.07%, 3.2 sigma<br />
71
RDO on global response surface<br />
• Approximation of model responses in<br />
mixed optimization/stochastic space<br />
• Simultaneous RDO is performed on<br />
a global response surface<br />
• Applicable to variance-, reliabilityand<br />
Taguchi-based RDO<br />
• Approximation quality significantly<br />
influences RDO results<br />
� Final robustness/reliability proof<br />
is required<br />
• Pure stochastic variables have small<br />
influence compared to design variables<br />
� Important local effects in the stochastic<br />
space may be not represented<br />
72
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
RDO on global response surface<br />
• Optimization variables are varied in optimization space<br />
• Pure stochastic variables are varied +/- 4 sigma<br />
• 200 LHS samples generated with uniform distribution<br />
• Approximation quality between 93% and 97%<br />
• Young’s modulus is in RDO space<br />
minor important but in robustness<br />
space most important<br />
� Pure stochastic variables<br />
are unimportant<br />
� Poor local approximation quality<br />
for robustness analysis<br />
73
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
RDO on global response surface<br />
• Evolutionary algorithm (444 designs) + 50 <strong>Robust</strong>ness samples for<br />
each optimization design<br />
• Mean values and standard deviation of contact pressures as<br />
robustness measures<br />
• RDO-Optimization with constraints for all contacts: 3.5 sigma<br />
74
<strong>Robust</strong> <strong>Design</strong> Optimization of a rubber seal<br />
RDO on global response surface – <strong>Robust</strong>ness proof<br />
Small approximation error Large approximation error<br />
Failure probability:<br />
• Upper left contact pressure 2.2% (2.0 sigma)<br />
• Lower left contact pressure 1.4% (2.2 Sigma)<br />
� Optimal design is not robust!<br />
75
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Iterative RDO Application Connector<br />
Goal: high safety level of connector<br />
by courtesy of<br />
10 contact forces have to be checked,<br />
failure may happen if N< 1 N<br />
The failure probability of single contact should<br />
be lower than 10%<br />
The failure probability of 5 contacts should be<br />
less than 1 out 4.300.000. (6 Sigma<br />
<strong>Design</strong>)<br />
� Status quo: pre optimized design<br />
using ANSYS DX and 5 optimization<br />
parameter<br />
Question:<br />
how optimal and robust is the design<br />
Solver: ANSYS Workbench<br />
76
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Step 1 - Sensitivity analysis<br />
The Sensitivity Analysis id done in the design<br />
space of 31 potential CAD (ProE) design<br />
parameters<br />
by courtesy of<br />
Identification of n=15<br />
most important<br />
design parameters<br />
77
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Step 2 - <strong>Robust</strong>ness analysis<br />
The failure probability of<br />
single contact failure<br />
is checked with<br />
<strong>Robust</strong>ness<br />
evaluation.<br />
The <strong>Robust</strong>ness Analysis<br />
is done in the<br />
<strong>Robust</strong>ness space of<br />
36 CAD tolerances<br />
• Global variancebased<br />
robustness<br />
analysis using<br />
Advanced Latin<br />
hypercube sampling<br />
with 90 design<br />
evaluations<br />
by courtesy of<br />
Performance<br />
critical contact<br />
force F3o_v<br />
With failure<br />
probability of<br />
89 %!<br />
78
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Step 3 - Optimization<br />
• n=15 most important CAD design<br />
parameters<br />
• Objective: minimal failure distance<br />
of every contact<br />
by courtesy of<br />
Step 1: ARSM<br />
- Stagnation after 5 Iterations (126<br />
<strong>Design</strong> evaluations), contact force<br />
F3o_v still violating criteria<br />
Step 2: introducing of additions constraints<br />
Restart of Evolutionary Optimization using<br />
best <strong>Design</strong>_ARSM, stop after 390<br />
<strong>Design</strong> evaluations with feasible<br />
design, but now other contact forces<br />
become critical<br />
Step 3: modifying the design space range<br />
and introducing additional constraints<br />
Restart using ARSM with best <strong>Design</strong>_EA,<br />
stop after 172 <strong>Design</strong> evaluations<br />
79
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Step 4 <strong>Robust</strong>ness Analysis • 36 scattering CAD tolerances<br />
by courtesy of<br />
• Global variance-based robustness<br />
analysis using Advanced Latin<br />
hypercube sampling with N=50 design<br />
Failure probabilities of two forces higher<br />
than 1%<br />
Performance critical contact force F3o_v<br />
with failure probability 9 %!<br />
Contact force F2o_h with failure<br />
probability 1 %!<br />
80
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Step 5 - Reliability analysis<br />
Identification of n=12 most important random parameters using coefficients of<br />
importance<br />
Defining the limit state condition for violation more than 50% of the contact<br />
forces are lesser than 1N<br />
by courtesy of<br />
Reliability analysis using ARSM with<br />
N=137 D-optimal design of<br />
experiment<br />
Adaptive sampling on the MLS<br />
surrogate model without samples<br />
in the unsafe domain<br />
Probability of failure is near zero,<br />
assumption: normal distributed<br />
random parameters<br />
Optimized design is an Six Sigma<br />
<strong>Design</strong><br />
81
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Results<br />
� Variance and probability-based robust design optimization with 31<br />
optimization parameters and 36 scattering parameters<br />
� Increasing the performance critical contact force F3o_v according<br />
failure probability 89% -> 9%<br />
� Failure probabilities of the other contact forces lesser than 1%<br />
� System failure probability (more than 50% of the contact forces are<br />
lesser than 1) is near zero! (Six Sigma <strong>Design</strong>)<br />
� N=950 parallel finite element calculations<br />
� Total calculation time 1 week<br />
by courtesy of<br />
82
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
<strong>Robust</strong>ness Analysis drop test of mobile phone<br />
A robustness evaluation is a so called global variance-based stochastic<br />
analysis which addresses the question how much the response values<br />
scatter under the consideration of input parameter scatter.<br />
• 209 scattering variables (drop position, material (E,<br />
ro, yield), geometry (thickness, radius), environment<br />
(friction)<br />
• 150 Latin Hypercube Samples<br />
Statistical Post Processing include:<br />
• The variation of the responses<br />
• Coefficient of Determination/Importance, checking<br />
linear and non-linear correlation<br />
• Coefficient of Prognosis<br />
• Statistics on Structure for critical part<br />
by courtesy of<br />
83
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Process automation – drop test<br />
Run the robustness<br />
analysis in optiSLang.<br />
209 scattering inputs<br />
by courtesy of<br />
Run ABAQUS which creates *.csv<br />
files containing the results.<br />
150 designs<br />
Run the robusteval script<br />
after the robustness analysis<br />
is finished. All the designs<br />
will be processed and *.rpt<br />
files are created.<br />
Post processing in<br />
optiSLang.<br />
- 36 responses<br />
84
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Post Processing <strong>Robust</strong>ness - drop test<br />
CoD lin<br />
adj<br />
CoD quad<br />
adj<br />
CoD lin<br />
Adj<br />
Spearman<br />
• The Histogram shows a variation between 45.35<br />
and 121.1<br />
• Probability of violating the limit is 9.33 %<br />
(1-0.9067=0.0933)<br />
CoP<br />
48 48 46 66<br />
• Maximum Cop is 66 %, Inputs ANGLE_Y and<br />
ANLGE_X have the most significant influence<br />
by courtesy of<br />
• Reference<br />
• Maximum<br />
85
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Statistics on Structure (SoS)<br />
• Based on <strong>Robust</strong>ness analysis results (optiSLang)<br />
• Visualization of statistics (Maxima, Minima, Mean value, Coefficient of<br />
variation, …) to locate „hot spots“ of variation at the FE Structure<br />
• Benefits:<br />
• Visualization of Min/Max and scatter range on FE Structure<br />
• Visualization of spatial Distribution of Variation and Correlation on FE-<br />
Structure<br />
• Selecting and visualizing important regions and their correlations<br />
We have to select one time<br />
frame for every design<br />
by courtesy of<br />
86
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Combining SoS and optiSLang<br />
• The picture above shows the maximum of S11 (positive - tension)<br />
• In SoS it is possible to select elements at hot spots and generate optiSLang<br />
*.bin file (see black circles)<br />
• Use the result_monotoring in optiSLang to identify local variation and<br />
correlation, The CoP plots below show that ANGLE_X and ANGLE_Y have<br />
strongest strong influence nfluence on S11 2to:Histogram for the to selected check elements<br />
variation<br />
Structure in<br />
ABAQUS viewer<br />
Element El sets – all glass element sets<br />
by courtesy of<br />
87
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Part 8<br />
Costumers<br />
88
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
Our Costumers agree<br />
89
DYNARDO • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
What‘s the Difference to Others?<br />
Methodology<br />
• Sensitivity analysis and optimization for large (number of variables)<br />
non-linear problems<br />
• COP/MOP to minimize the number of solver calls<br />
• Optimization with robust defaults (ARSM, EA,GA,PARETO)<br />
• Complete methodology suite to run robustness evaluation, reliability<br />
analysis and robust design optimization<br />
Key applications<br />
• Optimization with large number of parameter (>10) having non<br />
linear effects, noise, design failure<br />
• Model update and parameter identification using sensitivity study<br />
and optimization<br />
• oS (+SOS) have completed the functionality for robustness<br />
evaluation and reliability analysis and robust design optimization to<br />
be used in production<br />
We do not just offer a tool, we deliver a process.<br />
We are the ones who implemented robustness evaluation at<br />
different industries.<br />
90
WOSD 2012 • © <strong>Dynardo</strong> <strong>GmbH</strong> 2012<br />
optiSLang goes International<br />
DYNARDO, CADFEM (Germany, Austria, Swiss)<br />
SVSFEM (Czech Republic, Slovakia, Hungary)<br />
Pera Global (China, Shanghai)<br />
CADFEM India (India, Hyderabad)<br />
CAE Technology Inc. (Korea, Seoul)<br />
TaeSung S&E Inc. (Korea, Seoul)<br />
CADFEM CIS (Russia, Moscow)<br />
CADFEM US (USA)<br />
Tecosim Japan (Japan, Saitama)<br />
Introduction Weimar Optimization and Stochastic days 2012<br />
91