modeFRONTIER is a CAE integration platform used for multiple objective and multi-disciplinary optimizations and Design of Experiments (DoE). For example, if you are simulating with ANSYS or CDAdapco’s STAR-CMM+, you can incorporate simulation data into an optimization process using modeFRONTIER. The way it controls 3rd party software is through nodes added to the modeFRONTIER workflow to define a design space, or to optimize a design using various algorithms.

modeFRONTIER inputs values into the CAE tool and then collects output based on the simulation. modeFRONTIER can chain a series of CAE tools together in an iterative loop, working through each until all the parameters of the DoE are met or an optimization is completed. modeFRONTIER can also save time and computational resources using Response Surface Models (RSM) to limit the calls to various third party programs.

“modeFRONTIER offers a wide range of DoE and optimization algorithms. The user can choose an algorithm up front from a list. For instance, when setting up a problem the user can pick purely “random” and the program will ask how many random points are needed. Then modeFRONTIER would generate the DoE points.” said Alexander Duggan, Sr. Application Engineer at ESTECO.

Nodes Connect 3^rd Party CAE Solutions

“Direct integration nodes are built by our development team. These nodes cover many of the standard CAE tools in the industry. The hundreds of other tools that users have to interface with, from third party tools to in-house applications, are accessed using indirect integration. This connection is made by the user. It can link to scripting languages, text files, DOS, Shell, Java, and more. We also have a feature, called EasyDriver, where user can work with in-house code and create a custom integration to be very similar to a direct integration,” explained Duggan.

Direct nodes and EasyDriver options can cover all the applications of the target tool. Nodes are continuously maintained according to third party software updates. Custom-made integrations may need some re-scripting with each upgrade. Users have also been known to re-script these custom integrations when having to access a new application of the third party tool.

New in modeFRONTIER: DOE Designer Node

The September 2014 release outlines various improvements such as new DOE and optimization tools.

Danilo Di Stefano, Product Manager at ESTECO said, “these are the targets we are pursuing on parallel tracks: a simplified optimization experience, modularity and flexibility of the workflow and new features that open the platform to web-based collaboration.”  

The first highlight of the new release is the DoE designer node. The DoE Designer Node helps designers create specific workflows and reusable modules. As it runs, different DoE configurations are created. Users can then expand or focus on regions of the design space.

Duggan explained, “You can run the first DoE and then input a new design space based on the initial DoE. This can be run on the fly.” This method is best used in the exploration of problematic assessments.

New in modeFRONTIER: pilOPT

Perhaps the most ambitious announcement of the new release is pilOPT. Duggan explained, “pilOPT is very easy to use. You don’t have to configure any parameter or create a DoE. This is all done by the algorithm. It can then adjust how many designs are run through the software vs. how many RSM generated designs to run. The idea is to find an improvement as quickly as possible. If something runs quickly then pilOPT will rely less on the RSM approximation and more on third party tools.”

Di Stefano added, “The introduction of pilOPT offers a real one-click optimization experience.  By specifying just a single parameter the algorithm is able to quickly achieve a set of good and robust solutions. This way, you do not need to be an optimization expert to benefit from the most advanced optimization technology.”

There is another added benefit to such an optimization algorithm, especially when facing very complex problems. By limiting the inputs users make, it limits the dangers of ‘garbage in, garbage out’ assessments. “It can be dangerous to not understand what is going on in the background. That is why for many years optimization has been done by specialists … With pilOPT the DoE and optimization strategy is chosen for the user,” said Duggan.

He was however, quick to add that the new algorithm shouldn’t be used to replace experts or other algorithms that may apply better to the situation at hand: “You typically need an optimization expert or someone with knowledge on how to use an algorithm. Or they can also ask us if they are unsure”.

New in modeFRONTIER: Powell, Stepwise Regression

Though the Powell method has existed for a while, it has recently joined the list of optimization algorithms for modeFRONTIER. “It works by generating a certain number of DoE points. Based off of these points, the algorithm will find a search direction for improvement,” explained Duggan.

The Powell method can be used as an alternative to local search as it doesn’t use derivatives within the algorithm. It is typically used to find local minimums in complex continuous functions which are not well defined mathematically. Additionally as the algorithm doesn’t use gradients, it can save computational time.

The Stepwise Regression optimization is located in modeFRONTIER’s RSM algorithm list. It approximates RSMs using polynomials. The algorithm will automatically, and iteratively select an optimized polynomial regression model to minimize error.

Duggan said, “We have implemented the Stepwise Regression to simplify the complexity of higher order polynomial functions which can have thousands of terms in the equation. We simplify this equation by eliminating the less important terms. The algorithm maintains the error between the RSM and the actual points. We don’t want to include terms that cause overfitting, however, as this will introduce error.”

As for the future of modeFRONTIER, Di Stefano related that they are, “Moving forward to widen the integration options with CFD technology, starting with the new interface with EXA Powerflow in modeFRONTIER 2014 and we are planning to further expand our library of direct nodes with other CFD tools, including established software (for example SC/Tetra from Cradle Software) and newer products providing innovative approaches.”

Who Should Use modeFRONTIER

“If you have a hundred variables, or even 10 variables, with a large number of constraints and objectives you cannot solve them and find the best solution without a tool such as modeFRONTIER optimization,” said Dr Tayeb Zeguer, CAE Manager at Jaguar Land Rover. “It allows you to change hundreds of variables and receive output which contains many attributes. It would be impossible to do this manually. Multi-disciplinary optimization is a design by objective approach that is faster than the traditional CAE uses trial and error analysis … This will be a norm CAE process in the future and will help to achieve a CAE driven (design) process.”

This multi-objective, multidisciplinary nature of modeFRONTIER makes it a useful tool across many industries. Dr. Zeguer noted that across industries business priorities are changing. The need to get products out to market faster is ever growing. However, users want more robust products. This requires producers to create products through iteration. He adds, "Simulation is a very important enabler to help the delivery of greater product faster. Business drivers require ever shorter development times. This is why we use modeFRONTIER. It integrates and automates different design activities from concept, design, to production.”

For example, Dr. Zeguer remarks on their latest Land Rover Product. The goal was to make the product more sustainable through reduced CO₂ production, fuel consumption and weight. The team verified the reduced carbon footprint by lifecycle assessment tools. Jaguar moved to lightweight materials like plastics and aluminum but they needed to ensure the change did not affect the safety of their customers.

“By connecting tools through modeFRONTIER for simulation and optimization of the structure, we sped up the process. Without it we would look at one optimization at a time. With modeFRONTIER you can optimize many attributes at one time. With it, we were able to help drop the weight of the product from 498 kg (1098 lbs) to 288 kg (658 lbs),” exclaimed Dr. Zeguer.

“The Curse of Dimensionality”

Dr. Zeuger and Duggan both outlined that all optimization software is limited by ‘the curse of dimensionality.’ Essentially, the more variables and objectives in the optimization the slower it runs.

To combat this limitation, Duggan suggested that, “as you get into higher dimensions, about ten or more, things become slower. For a user the key thing is to find the most important inputs and limit the design space to those inputs. The number of objectives is also a limitation: identifying them and formulating the objective functions correctly are key to a successful optimization.”

He added that “modeFRONTIER does offer a number of tools to help identify the important inputs and reduce the number of dimensions.” These tools can go a long way to speed up those more complicated optimizations within modeFRONTIER.

To learn more about the new release of modeFRONTIER watch this webinar.

ESTECO has sponsored promotion of modeFRONTIER on ENGINEERING.com. They have no editorial input to this post - all opinions are mine.  Shawn Wasserman