FE simulation correlation: a dedicated platform

As an engineer, correlation of finite element simulation often requires lots of data management within Excel or with in-house Python scripts.

As the European leader in space launchers, Ariane Group participates with EikoSim in multiple Research and Development projects, including a RAPID R&D project (“MUTATION”) funded by the Direction Générale de l’Armement.

This project aims at developing an industrial platform for test-simulation correlation to meet the challenges of faster and safer development of industrial applications.

finite Element Analysis Software and correclation metrics to meet challenges of industrial applications

The Galileo dispenser project

Within this R&D project, one of the key use cases was the qualification test of the Galileo Dispenser in the Ariane 6 version.

A dispenser is a system placed under the launcher fairing, which is designed to release one or several satellites during the launcher mission and to put them into orbit.

The test is performed on a flight model, meaning only qualification load cases are applied to the structure without ever reaching failure.

The objective was, therefore, twofold:

  • Validating the behavior of the structure for these load cases.
  • Showing the adequacy of the FE simulation model on the latter use case allows us to have a satisfactory correlation to extrapolate the forecasts on more complex and hardened loads.
simulation correlation metrics using finite element method or finite element analysis for fluid structure interaction

The finite element model was composed of a Shell model completed with 3D sub-models for critical areas.

In practice, the post-test analysis had to demonstrate the model’s ability to predict the overall shape and behavior material properties of finite elements of the structure as well as an acceptable difference between it and the test results through the various instrumentations put in place, notably concerning the linearity of the overall behavior.

The MUTATION project was organized to allow direct testing of the implementation that was made within the platform.

With this point of view, this test responded to an opportunity in three steps: assessing a new instrumentation process, embedding a wide range of instrumentation and correlation metrics, and allowing a post-test analysis with an optimized test-simulation correlation through a “digital twin for solid mechanics.”

How Ariane Group seeks to save up to 30% of analysis time with EikoSim

By engaging in this R&D project, with the support of Direction Générale de l’Armement, ArianeGroup was seeking to make the validation and analysis process more fluid and efficient in order to allow growing confidence in simulation models and remove unnecessary physical tests.

One of the key elements identified by the project leaders was that the current processes still involved a lot of do-it-yourself activities, such as the very common post-processing of sensors in Excel.

This is nearly an industry standard at this point, which means a lot of room for improvement, and especially for structural tests that involve a large number of sensors,” explains Florent Mathieu.

Indeed, for this project, simulation engineers spent hundreds of hours just for data management within Excel.

Some sensor management was already available in EikoTwin DIC but not usable for non-DIC users.

The work with ArianeGroup helped identify forgotten pain points and define the use scenarios that will yield the most long-term value for our partner,” states Pierre Baudoin, Research Engineer and Project Leader at EikoSim. 

“Post-processing of a large number of sensors clearly was one of these situations, and that’s how EikoTwin Lite was born.”

simulation correlation, correlation coefficient and correlation metrics via finite element analysis of material behavior

On paper, the software has a pretty simple value proposition: aggregating all sensors around the FE model and providing a global comparison.

In practice, a lot of operations are needed to actually analyze all the available data, including the sensor calculation itself, but also aggregation features or results in visualization tools.

With Excel, these operations necessitate a lot of specific in-house development for each case.      

Due to the large scale of the component and the fact that it was imperative to verify that it was not damaged during testing, over 200 strain gauges were disposed of in the dispenser.

The correlation of this large amount of experimental data to simulation presents its own set of challenges. First, the necessity to provide simulation previsions across this set of sensors, and to update them quickly when the model is modified.

Second, dedicated tools were required to automate the import of experimental data across this large number of strain gauges and display the test data in-simulation comparisons in efficient ways.

The Galileo dispenser qualification test was thus an opportunity to validate different strain evaluation strategies for strain gauges. Strain predictions on the most sensitive areas were derived from higher-order 3D elements.

To estimate strain values at the strain gauge location, a series of local coordinate systems were defined so that their first axis matched the axial direction of the specified gauges gauge.

Then, the 3D strain tensor was expressed in this local coordinate system for each strain gauge.

Finally, the linear first component of this tensor was expressed at the relevant simulation nodes, and the strain gauges previsions were derived for each load case.

This method was tailored for this particular test, and from experience on the previous qualification campaign on the Galileo 5 dispenser.

In parallel, an alternative method was also evaluated for strain gauge previsions, based on surface displacement fields obtained from the initial simulation.

In this method, strain computation is performed automatically in the plane of the surface element.

The local coordinate system is defined according to the Abaqus convention: axis 1 is obtained by projecting the global X-axis onto the surface of the element, or by projecting the global Z-axis if the global X-axis is normal to the element’s surface.

Thus, only one angle has to be specified to define the gauge orientation.

Finally, strain gauge distribution predictions are determined by computing a least square best fit of a bi-parabolic displacement field across the gauge area. The local strain field is obtained directly from the polynomial coefficients.

Note that this second method is more generic than the finite element method and can be extended to a number of cases independently of the finite element application.

The results of the two methods are presented in Figure 7. All results have been normalised. Although their form and implementation significantly differ, the two methods above yield close results for the studied set of strain gauges.

simulation correlation of finite elements looking at residual stresses, eikosim modeling methods
ImFigure 8: Normalised strain estimates for a series of strain gauges on the first load case.age

 “With this software, we can already expect a time saving of about 40% for data management alone, which represents hundreds of hours,” says Jérémy Pradelli, CAE engineer at ArianeGroup.

The next developments could increase that saving to more than 60% of data management time.

visualization of test data from finite element method using correlation metrics to save time in physical system

How Ariane Group improves the correlation of simulation with EikoTwin

The H3d (Altair Hyperworks) format has been chosen to provide a pathway between Hyperworks, the pre-post solution chosen by Ariane Group, and EikoTwin.

The connection was developed in a matter of months and now allows our engineers to import and export data to the EikoTwin platform without any data conversion,” says Nicolas Swiergiel, Photomechanics Expert at Ariane Group. “To keep the platform robust, it is essential to have a streamlined process for data flow between our tools of choice”.

We can also expect more robustness and far fewer human errors” continues Jérémy Pradelli.

The problem with Excel for CAE engineers is that it requires them to build everything from scratch for the first step of each new project.

This can create user errors after hundreds of hours spent copying/pasting data, despite the engineers’ best efforts.

Using a more integrated solution also ensures everyone uses the same post-processing algorithm for a given sensor, and that this algorithm has been vetted by experts.

As a result, engineers will spend less time worrying about data management and cut the chase to the correct parameters for their simulations.

The natural continuation of this work is to allow full integration with the latest Verification, Validation, and Uncertainty Quantification (VVUQ) techniques.

The work has already been started by EikoSim, especially by integrating the measurement uncertainties into the analysis process, which will be paramount for CAE engineers to justify modelling choices.

Finally, the platform is modular by nature, as other measurement techniques are already being integrated into physical systems, such as marker tracking or optical fibers measurement of linear strains.

Posted in

Leave a Comment