Information-Theoretic Approach for Identifiability Assessment of Nonlinear Structural Finite-Element Models

This paper presents an information-theoretic approach for identifiability assessment of model parameters in nonlinear finite-element (FE) model updating problems. Rooted in the Bayesian inference method, the proposed approach uses the Shannon information entropy as a measure of uncertainty in the model parameters. The difference in the entropy of a priori and a posteriori probability distribution functions of model parameters, which is referred to as the entropy gain, is used as a measure of information contained in each measurement channel about the model parameters. The entropy gain approach can be used for selection of estimation parameters, optimal sensor placement, and design of experiment. In this study, an approximate expression for the entropy gain is derived, and a three-step process is suggested for the identifiability assessment. The application of the proposed approach is demonstrated for a nonlinear structural system identification problem. Although the focus of this study is on nonlinear structural FE model identifiability, the provided approach can be used for identifiability assessment of other types of linear/nonlinear dynamic models.