John Emery, Sandia National Laboratories
Coleman Alleman, Sandia National Laboratories
Thomas Seidl, Sandia National Laboratories
Calibration of material constitutive models typically requires solving an inverse problem to determine the best set of model parameters to match some observed/measured response. Often, this is accomplished with constrained optimization algorithms. This poses two significant challenges. First, it is difficult in general to define objective functions that are smooth and convex with a unique global minimum, so local and gradient-based optimization techniques can be inadequate. Second, evaluation of the objective function and its derivatives typically require finite element calculations for each iteration. These may be quite expensive for calibration of plasticity and damage models, where structural effects are not entirely separable from the material response. In addition, analysis of uncertainty typically requires distributions for each calibrated parameter, leading to a number of difficulties associated with high-dimensionality, non-uniqueness, etc.
For this minisymposium, we are soliciting contributions that address one of these challenges, or challenges with model calibration more generally. We are particularly interested in research that addresses one or more of the following 1) constrained optimization in the context of model calibration; 2) machine learning and associated techniques to provide surrogate or reduced-order models for increased efficiency in the fitting process; 3) methods that provide quantification of uncertainty in fit parameter values; 4) approaches to address multiphysics and multi-fidelity aspects of model calibration.