Session: Poster Presentation

Paper Number: 155556

155556 - Validation Hierarchy and Building a Credibility Case Using Validation Suites 

Abstract: 

The ASME standard VVUQ 10 Verification, Validation, and Uncertainty Quantification in Computational Solid Mechanics shows an organizing principle of a validation hierarchy: single physics models feed into more complex models and ideally a validation assessment is made as part of a blind prediction for an integrated experiment.

As an organizational principle, the validation hierarchy works well.  As a practical matter for engineering problems, the details quickly become overwhelming for more than a trivial case.  One method to make details more manageable is to think about the big question: What types of evidence would be useful for assessing the credibility of the simulations for this context of use and what tools are needed to obtain that evidence?

Identifying the typical types of evidence may then lead to a toolbox that could be reused for different contexts of use without starting from scratch for every new question.  We are calling sets of reusable computational tools “validation suites”.  An example of use may be flyer plate experiments for different materials as the single physics that explores some of the thermodynamic and mechanical regimes, a Richtmyer-Meshkov instability experiment for more integrated physics, and then a third level like multilayered ejecta experiments.

By having multiple suites that explore different aspects of the physics, one could then do some material model evaluations on different types of experiments to see patterns and accumulate credibility evidence.  For example, one could imagine having flyer plates with copper, aluminum, and tantalum individually; RMI experiments with singletons and pairs of the materials to explore new physics as well as combinations; and then multilayered ejecta experiments that combine all materials along with new physics.

Nothing prevents doing each of these simulations individually every time they would be useful.  However, having the machinery to rerun everything by simply doing a swap of copper model A for copper model B with one line with no further human intervention needed is very handy.

This technique lessens the need for the “big green validated for everything” check mark and the related attitude of “we did one of those V&V exercises three code versions ago; those results are probably still good” while preserving the intent of “why can’t VVUQ be easier for things we do all the time?”

This work will present the progress made for such validation suites in the Common Modeling Framework at Los Alamos National Laboratory along with some lessons learned because nothing is easy.

Presenting Author: Joanne Budzien Los Alamos National Laboratory

Presenting Author Biography: Joanne Budzien has a PhD in materials engineering and has worked in various aspects of materials science and engineering in academia and national laboratories. Dr. Budzien currently works at Los Alamos National Laboratory in the Verification and Analysis group in the Computational Physics division.

Authors: