Predictive maturity of computer models using functional and multivariate output

Sez Atamturktur, François Hemez, Cetin Unal, Brian Williams

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Scopus citations

Abstract

Computer simulations are valued in science and engineering because they enable us to gain knowledge about phenomena that would otherwise be difficult to understand. Dependency on simulations primarily stems from our inability to conduct a sufficient number of experiments within the desired settings or with sufficient detail. However, if one were able to conduct a large-enough number of experiments, it is reasonable to envision that a simulation model could be calibrated to the point that its predictive uncertainty is reduced down to uncontrolled, natural variability. We inductively conclude that, as new experimental information is used for calibration, the calibrated parameters should stabilize, and thus, the disagreement between simulations and experiments should be reduced down to "true" bias. We propose to use the stabilization of the incremental improvement to assess the predictive maturity of a model. Accordingly, we develop a Prediction Convergence Index (PCI) that approximates the convergence of predictions to their "true" or stabilized values or, conversely, can be used to estimate the number of experimental tests that would be required to reach stabilization of predictions. The application of the PCI is illustrated using a Preston-Tonks-Wallace material model for Tantalum and six experimental datasets in the form of strain and stress curves. Once the predictive maturity of a model has been assessed, we argue that it is acceptable to extrapolate its predictions away from settings or regimes where validation tests have been conducted as long as the physics involved and modeled by the code remains unchanged. For the given model, the extent to which extrapolation and interpolation are acceptable is investigated. The results agree with our hypothesis and suggest that the approach proposed can prove useful for claiming completion of the calibration phase and providing insight into the predictive maturity of numerical models.

Original languageEnglish (US)
Title of host publicationIMAC-XXVII
Subtitle of host publicationConference and Exposition on Structural Dynamics - Model Verification and Validation
Publication statusPublished - Dec 1 2009
Event27th Conference and Exposition on Structural Dynamics 2009, IMAC XXVII - Orlando, FL, United States
Duration: Feb 9 2009Feb 12 2009

Publication series

NameConference Proceedings of the Society for Experimental Mechanics Series
ISSN (Print)2191-5644
ISSN (Electronic)2191-5652

Other

Other27th Conference and Exposition on Structural Dynamics 2009, IMAC XXVII
CountryUnited States
CityOrlando, FL
Period2/9/092/12/09

    Fingerprint

All Science Journal Classification (ASJC) codes

  • Engineering(all)
  • Computational Mechanics
  • Mechanical Engineering

Cite this

Atamturktur, S., Hemez, F., Unal, C., & Williams, B. (2009). Predictive maturity of computer models using functional and multivariate output. In IMAC-XXVII: Conference and Exposition on Structural Dynamics - Model Verification and Validation (Conference Proceedings of the Society for Experimental Mechanics Series).