Design Exploration and Robust Design using ANSYS DesignXplorer

Design Exploration and Robust Design using ANSYS DesignXplorer Design Exploration and Robust Design using ANSYS DesignXplorer

<strong>Design</strong> <strong>Exploration</strong> <strong>and</strong> <strong>Robust</strong><br />

<strong>Design</strong> <strong>using</strong> <strong>ANSYS</strong> <strong>Design</strong>Xplorer


Doesn’t account for<br />

physics interactions<br />

Single Simulation<br />

• Single Solve<br />

– Solves a single simulation with<br />

only a single physics<br />

– Engineers are interested in<br />

solution robustness, speed,<br />

accuracy, ease of use <strong>and</strong><br />

engineering results<br />

“Virtual Prototyping” typically requires looking<br />

at multiple physics, This single physics<br />

approach is often not enough.


Single Multiphysics Solve<br />

– Solves a single simulation<br />

involving multiple physics<br />

– Engineers are still interested in<br />

solution robustness, speed,<br />

accuracy, ease of use <strong>and</strong><br />

engineering results<br />

Doesn’t provide<br />

direction for design<br />

engineering results<br />

improvement<br />

– And the ease <strong>and</strong> power of the<br />

physics coupling<br />

How can I improve performance?<br />

Can I reduce weight or cost?<br />

What is limiting performance?<br />

Is this a robust design?


?<br />

?<br />

?<br />

Solutions compared,<br />

but design is not well<br />

understood <strong>and</strong> no<br />

optimum is found<br />

“What If” Study<br />

• User adjusts inputs <strong>and</strong> investigates<br />

results<br />

• Builds on previous expectations,<br />

adds requirement of easy <strong>and</strong> robust<br />

parametric updates <strong>and</strong> comparative<br />

reports<br />

Need a more scientific <strong>and</strong> automated way to<br />

decide which points to solve<br />

Need a way to interpolate between these points


Response<br />

Surface<br />

Provides design<br />

underst<strong>and</strong>ing, But ….<br />

<strong>Design</strong> <strong>Exploration</strong><br />

• Scientific methods to explore the<br />

design space fully<br />

• Allows user for: sensitivity <strong>and</strong><br />

correlation, DOE <strong>and</strong> response surface<br />

technology, mesh morphing, charting<br />

<strong>and</strong> reporting<br />

• For complex solutions, user can take<br />

advantage of robust efficient &<br />

affordable <strong>ANSYS</strong> distributed solver<br />

Technologies.<br />

Optimize a design with selected inputs <strong>and</strong> goals


Solutions may be too<br />

sensitive to input<br />

variability<br />

Arriving at a set of Optimal<br />

design solutions<br />

• Optimization<br />

– Searches the design space for<br />

optimal c<strong>and</strong>idates, given userdefined<br />

goals <strong>and</strong> priorities<br />

– Adds requirements for: advanced<br />

optimization algorithms to<br />

efficiently search for c<strong>and</strong>idates,<br />

comparative reporting<br />

Real-world inputs typically have some variation<br />

<strong>and</strong> may require a more “robust design” goal


<strong>Robust</strong> <strong>Design</strong><br />

• Taking the variation of inputs into account, <strong>and</strong><br />

seeking a design with a probabilistic goal<br />

Input<br />

distribution<br />

Output<br />

distribution<br />

• RDO => Min st<strong>and</strong>ard deviation of the results<br />

• Six Sigma => Optimal design within a safe domain<br />

• In order to arrive at <strong>Robust</strong> design, User needs to<br />

- Specify probabilistic parameters<br />

- Use probabilistic optimization algorithms<br />

<strong>Robust</strong> <strong>Design</strong> optimizes design variables to achieve a<br />

particular probabilistic level such as Six Sigma, which<br />

translates into 3.4 failures in one million parts.


The Path to <strong>Robust</strong> <strong>Design</strong><br />

Single Physics<br />

Solution<br />

•Accuracy,<br />

robustness, speed…<br />

Multiphysics<br />

Solution<br />

•Integration<br />

Platform<br />

“What if”<br />

Study<br />

•Parametric<br />

Platform<br />

<strong>Design</strong> <strong>Exploration</strong><br />

•DOE, Response<br />

Surfaces, Correlation,<br />

Sensitivity, etc.<br />

Optimization<br />

•Algorithms<br />

<strong>Robust</strong> <strong>Design</strong><br />

•Probabilistic<br />

Algorithms<br />

•Adjoint solver<br />

methods


<strong>Design</strong> <strong>Exploration</strong> <strong>and</strong><br />

<strong>Robust</strong> <strong>Design</strong> at <strong>ANSYS</strong>


Optimization tools at <strong>ANSYS</strong><br />

– <strong>ANSYS</strong> <strong>Design</strong>Xplorer<br />

• Unified Workbench<br />

solution<br />

– <strong>ANSYS</strong> Fluent<br />

• Has built-in morphing <strong>and</strong><br />

optimization tools<br />

• Has an adjoint solver<br />

– <strong>ANSYS</strong> MAPDL<br />

• DX VT<br />

– And more<br />

Baseline <strong>Design</strong><br />

Optimized <strong>Design</strong>


<strong>ANSYS</strong> <strong>Design</strong>Xplorer<br />

Integral with Workbench<br />

• Parametric multiphysics<br />

modeling with automated<br />

updates<br />

• Bi-directional CAD, RSM,<br />

scripting, reporting <strong>and</strong><br />

more...


<strong>Design</strong>Xplorer in Workbench


<strong>ANSYS</strong> <strong>Design</strong>Xplorer<br />

<strong>Design</strong>Xplorer is everything<br />

under this Parameter bar…<br />

• Low cost & easy to use!<br />

• It drives Workbench<br />

• Improves the ROI!<br />

DX<br />

<strong>ANSYS</strong><br />

Workbench<br />

Solvers


<strong>ANSYS</strong> <strong>Design</strong>Xplorer <strong>Design</strong> of<br />

Experiments<br />

With little more effort than for a single<br />

run, you can use <strong>Design</strong>Xplorer to<br />

create a DOE <strong>and</strong> run many variations.


Correlation Matrix<br />

Underst<strong>and</strong> how your parameters are<br />

correlated/influenced by other parameters!


Sensitivity<br />

Underst<strong>and</strong> which<br />

parameters your<br />

design is most<br />

sensitive to!


Response Surface<br />

Underst<strong>and</strong> the<br />

sensitivities of the output<br />

parameters (results) wrt<br />

the input parameters.<br />

3D Response<br />

2D Slices Response


Goal-Driven Optimization<br />

Use an optimization algorithm<br />

or screening to underst<strong>and</strong><br />

tradeoffs or discover optimal<br />

design c<strong>and</strong>idates!


Output<br />

parameters<br />

vary also!<br />

<strong>Robust</strong>ness Evaluation<br />

Input parameters have variation!<br />

Underst<strong>and</strong> how your<br />

performance will vary<br />

with your design<br />

tolerances?<br />

Predict how<br />

many parts will<br />

likely fail?<br />

Make sure your<br />

design is robust!<br />

Six Sigma, TQM<br />

Underst<strong>and</strong> which<br />

inputs require the<br />

greatest control?


Challenges<br />

• Engineers face numerous obstacles that prevent them from <strong>using</strong><br />

optimization fully<br />

• Our plans are largely based on removing those obstacles so our<br />

customers can get more out of simulation<br />

41.9%<br />

21.8%<br />

Do you use optimization<br />

software?<br />

18.6%<br />

Matlab<br />

16%


The Path to <strong>Robust</strong> <strong>Design</strong><br />

• Build on the solid foundation of Workbench as<br />

a platform for parametric simulation<br />

• Develop DX as an optimization platform<br />

• Build up to probabilistic optimization<br />

Single Physics<br />

Multiphysics<br />

“What if” Study<br />

Optimization<br />

<strong>Design</strong> <strong>Exploration</strong><br />

<strong>Robust</strong> <strong>Design</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!