Parameter identifiability through information theory

In this paper, we address the problem of assessing the identifiability of model parameters in a mechanical system, i.e., whether unknown parameters can be estimated given a set of measurements collected through sensor networks. Practical identifiability can arise due to either a lack of sensitivity or a joint effect of the parameters on the measurements. Information theory can be used to detect the sources of non-identifiability, with the purpose of establishing an efficient sensor network design. Mutual Information between the parameter and the measured outputs, and Conditional Mutual Information of each parameter couple, conditioned on the measurements, are considered. Adoption of these indices is overviewed for practically assessing the identifiability of the mechanical properties of a non-linear structural model.