Presented at X-4 Seminar, May 12, 2003, Los Alamos National Laboratory

Inference of material-model parameters from experimental data

Kenneth M. Hanson
Los Alamos National Laboratory

Abstract

I will describe an analysis of quasi-static and Hopkinson bar data for a high-strength steel, HSLA 100. The purpose of the analysis is to determine the parameters for a Zerilli-Armstrong model that are suitable for a room-temperature, high strain-rate application, namely a Taylor impact test. It is required to be able to predict the uncertainties in simulations of the application scenario. This example may appear to be just a straightforward data-fitting problem, albeit a nonlinear one. On the other hand, because of disagreement between the data and the Z-A model, a number of decisions need to be made. The Bayesian approach to uncertainty quantification provides the necessary tools to combine the experimental data with expert judgment about the experiments and material models. Part of the solution to obtaining an acceptable result is to introduce systematic uncertainties to account for sample-to-sample variations in behavior. In the end, it becomes apparent that the Z-A model is inadequate to describe the material characterization data, specifically the stress-strain behavior at high temperatures. The measured profile of the deformed Taylor cylinder is used to adjust the Z-A parameters in a way consistent with the analysis of the basic experiments. However, the results only demonstrate further the difficulties with the Z-A model. My purpose in this presentation is to induce an open discussion of the various ways to cope with and estimate the uncertainties in analyses of experimental data.

Keywords: simulation uncertainty, uncertainty quantification, systematic uncertainty, Bayesian analysis, Taylor impact test, material characterization experiments, Zerilli-Armstrong model, HSLA 100 steel

Viewgraphs for this paper (pdf, 733 KB)
Send e-mail to author at kmh@hansonhub.com

Return to Hanson presentations