Why Simcenter HEEDS ?

New Paradigm for Design Exploration

HEEDS Optimization

Benefits of Using HEEDS Optimization

1. Innovate & Extract Higher Value From Simulation
2. Redefine Simulation Strategy & Focus on Driving Performance
3. Discover Better Designs, Faster

Here, we would like to share with you ways to help your engineers use HEEDS modeling & simulation to discover better designs faster…or in other words how to help you drive product innovation.
 

Featured Design Exploration Case Study Using HEEDS Optimization:

Underhood pipe flow design

Objective

Design an underhood pipe to Maximize uniformity of outlet velocity
Design variables (7): Cross-section center, radii, angle

HEEDS Results:

Increased flow uniformity by 30%
Decreased swirl by 50%

Let’s look at how the SHERPA framework in HEEDS allows us to approach problems in a whole new manner. This is a relatively simple underhood pipe design problem that we’ll show in a product demo later.
The objective is to design a pipe that has a uniform outlet velocity at the left end.

With HEEDS, we don’t need to simplify the problem.  We just parametrically define the pipe shape with seven design variables, connect the CAD model to a CFD model, and allow HEEDS to automatically search for good designs as shown here.
In this case, we easily find a design that has a very uniform outlet flow, but we notice that it also has significant swirl. We take a look at the streamlines to understand why the swirl occurs. So, we simply add a constraint to the problem definition, re-run the design study, and find a design that has lower swirl.

Now, it becomes clear that once we added a swirl constraint, we got lower uniformity, so there must be a trade-off between swirl and uniformity. So, we can just tell HEEDS to perform a multi-objective study to examine the trade-off between swirl and uniformity. 

Now, we can find a design with both high uniformity and relatively low swirl.  These three design studies were done in less than a day, very simply. They’re illustrative of the way users typically work with HEEDS.

Now, think about how we would have considered this problem with the traditional approach.  We normally would have screened the variables down to a smaller number by performing a sensitivity analysis as shown here. In this case, we probably would have kept x1, x2, and r2 and eliminated angle, y1, r1, and y2.  Now, in this case, just for illustration, we also took a look at the sensitivity of the best design to the same variables.

What we see is that without keeping the variable “angle,” we could never have actually arrived at the best design as we did easily with HEEDS.  Not only did HEEDS easily identify a much better design that would traditionally have been found, but it also identified the sensitivity of the design to all variables – all without forcing the engineer to have any special expertise in design search or statistics.

Innovation Through Design Exploration

With STAR-CCM+ You Are Able to Automate The Design Exploration and Optimization Process

Design Search Strategy – SHERPA Search Framework

Hybrid:

1. Blend of search strategies used simultaneously
2. Global and local search performed together
3. Leverages the best of fall methods

Adaptive:

1. Tunes itself to the design space
2. Efficient with simple and complex spaces
3. Very  cost effective for complex problems

• The basic premise is that traditional approach is untenable
• We don’t know the terrain (design space response surface)
• We can’t know the best approach apriori
• And picking one exclusive approach is nonsensical when the terrain varies
• Describe the basic idea of SHERPA to:
• Combine different strategies “on the fly” as needed
• Tailor and tune the approach to the problem at hand

 
To help convey the difference between HEEDS and the traditional design exploration, let’s consider an analogy.
Suppose you’re tasked with climbing to a peak of an unknown mountain range.  That’s essentially what you face when you have a simulation model with an unknown design space or response surface that changes every time you make a small change to the objectives, constraints, design variables, or variable ranges.  Even though you don’t know much about what terrain to expect, suppose that you have to choose one set of climbing gear and one climbing strategy.  That’s the traditional approach to design exploration.  Even if software solutions offer hybrid strategies, they still contain pre-defined algorithms, instead of adjusting to the problem at hand.
What you need is a SHERPA who knows all of the climbing strategies, carries all of your equipment, and can tailor your climbing approach to the terrain encountered.  That’s what HEEDS do.
 
HEEDS contains the SHERPA framework that combines and weights multiple search strategies, changes strategies as needed, and tunes the strategy to each problem. The SHERPA framework is both hybrid and adaptive and it eliminates all of the previously described issues with the traditional approach.
If a new search strategy is developed in the research world, we simply add it to the SHERPA framework and allow it to lend its advantages to any problem in combination with other strategies.

SHERPA Search Strategy – Boeing Benchmark Results

Challenge:

Find minimal function value in the least number of evaluations

Results:

In 2000 evaluations, SHERPA performed >10% better than any other algorithm and >30% better than the nearest hybrid algorithm
 

• SHERPA approach combines the best of all search strategies
• The results Show here are from Boeing benchmark
• Each different search strategy has benefits
• SHERPA combines all of these benefits to solve any problem better than anyone algorithm can do, in general
• Reinforce that these results are typical of all benchmark problems

 
To fully appreciate the advantage that SHERPA affords to HEEDS users, let’s look at one of the standard benchmark problems used to assess HEEDS performance. This was a problem provided by Boeing during an evaluation of HEEDS.
 
We’re considering a sample mathematical function shown here that is very multi-modal, meaning it has lots of peaks and valleys.  In fact, the graph of this function for just two variables is shown in the upper right.  But, we’re considering a case with 20 variables which is much more difficult to visualize.  The challenge for a search strategy is to find the minimum value of this function among all the valleys.  Here we’ve compared the performance of all of the standard techniques in competitive software with the performance of the SHERPA strategy in HEEDS.  You can see how SHERPA combines the strengths of all of these strategies to deliver the best performance by far.
 
Upon clicking, we see an image showing how SHERPA searches the 2-variable problem.  This makes it clear how efficiently SHERPA can search the design space and perform more of its evaluations in areas of high interest.
 
This result is typical of the performance of HEEDS on any problem.  In each case, different single algorithms perform better or worse depending on the shape of the response surface.  But, in each case, the SHERPA framework in HEEDS is able to combine the strengths of these different algorithms in a unique way to find the most efficient path through the design space.