AN UPDATED TOOLBOX FOR VALIDATION AND UNCERTAINTY QUANTIFICATION OF NONLINEAR FINITE ELEMENT MODELS

It is becoming commonplace to use numerical simulations supported by limited experimentation for the characterization of physical phenomena. This trend, with its perceived potential for reducing costs, is the basis for the simulation-based procurement initiatives currently gaining momentum within the government and industry. Insuring the quantitative viability of a simulation-based procurement still requires some experimental data upon which the assessment of simulation accuracy can be based. In addition, it requires minimizing the differences between corresponding analytical and experimental results in physically meaningful ways, and characterizing the ability of the models to predict future events. The purpose of model validation and uncertainty quantification is to confirm the correctness and credibility of numerical simulations, so that the underlying models may be used with greater confidence to extrapolate limited test experience to a range of practical applications. In this paper, an advanced principal components-based computational procedure is demonstrated by validating the DYNA models used to achieve HFPB numerical simulations of physical processes important to assessing weapon- target interaction. Bayesian statistical parameter estimation is used to estimate material parameters that cannot be measured directly, such as strain rate enhancement and shear dilatency in reinforced concrete structures. This demonstration is performed using an updated MATLAB® Nonlinear Model Validation and Verification Toolbox. The work reported in this paper has resulted in improvements to the original Toolbox. A multi-level parameter estimation procedure is implemented to sequentially accumulate information from prior estimates in a Fisher information matrix for use in subsequent parameter estimates. The use of a generic uncertainty model in estimating the predictive accuracy of future DYNA simulations is enhanced through the use of a reduced set of principal component metrics and a basis augmentation technique.