Confidence intervals of inversely identified material model parameters: A novel two-stage error propagation model based on stereo DIC system uncertainty

Abstract

Digital image correlation (DIC) is a powerful tool for characterising materials and determining material model parameters. To assess the reliability of the full-field measurement-based inverse identification procedures, it is crucial to investigate the impact of the measurement errors on the identified material model parameters. Literature indicates that conventional error propagation models, which rely on Gaussian noise-contaminated data, significantly overestimate the confidence for inversely identified material model parameters, resulting in misleadingly narrow confidence intervals. A more precise assessment of systematic errors originating from the experimental setup leads to an improved prediction of the confidence intervals, but this requires specific information about the DIC equipment, post-processing details, and a skilled experimentalist. In this work, we propose an alternative two-stage error propagation model that yields more realistic confidence interval predictions based solely on a database of past mechanical experiments conducted with the specific stereo DIC system set up in a particular way. We have validated the proposed procedure numerically using the open-hole test and an orthotropic elastic material model. Our predictions reveal a non-Gaussian probability distribution of the inversely identified material model parameters, with confidence intervals considerably wider than those obtained by considering random Gaussian noise. Furthermore, these predictions were experimentally validated in an extensive experimental campaign investigating a pultruded carbon fibre epoxy composite.