Parameter covariance and non-uniqueness in material model calibration using the Virtual Fields Method

• Published in: Computational materials science 2018-09, Vol.152 (C), p.268-290
• Main Authors: Jones, E.M.C., Carroll, J.D., Karlson, K.N., Kramer, S.L.B., Lehoucq, R.B., Reu, P.L., Turner, D.Z.
• Format: Article
• Language: English
• Subjects: 304L stainless steel; Bammann-Chiesa-Johnson (BCJ) material model; Digital Image Correlation (DIC); Material identification; MATERIALS SCIENCE; MATHEMATICS AND COMPUTING; Virtual Fields Method (VFM); Viscoplasticity
• ISSN: 0927-0256; 1879-0801
• DOI: 10.1016/j.commatsci.2018.05.037

[Display omitted] Traditionally, material identification is performed using global load and displacement data from simple boundary-value problems such as uni-axial tensile and simple shear tests. More recently, however, inverse techniques such as the Virtual Fields Method (VFM) that capitalize on heterogeneous, full-field deformation data have gained popularity. In this work, we have written a VFM code in a finite-deformation framework for calibration of a viscoplastic (i.e. strain-rate dependent) material model for 304L stainless steel. Using simulated experimental data generated via finite-element analysis (FEA), we verified our VFM code and compared the identified parameters with the reference parameters input into the FEA. The identified material model parameters had surprisingly large error compared to the reference parameters, which was traced to parameter covariance and the existence of many essentially equivalent parameter sets. This parameter non-uniqueness and its implications for FEA predictions is discussed in detail. Finally, we present two strategies to reduce parameter covariance – reduced parametrization of the material model and increased richness of the calibration data – which allow for the recovery of a unique solution.