From full-field data to spatial uncertainty

“Lighter is better” is quickly becoming a new mantra in mechanical engineering designs, leading to structures that are becoming increasingly designed up to their performance limits. As a result, uncertainty quantification (UQ) is quickly gaining in importance to assess the reliability of these structures, given typical variability and/or uncertainty an analyst is faced with. However, these reliability estimates are only as accurate as the uncertainty models that feed them, which can be problematic when faced with model quantities that are not directly measurable. In this case, inverse UQ methods have to be applied. In the context of model quantities where the uncertainty is not constant over the model domain and/or tensor valued (e.g., orthotropic material properties), such inverse quantification can be highly challenging, especially when not sufficient measurement data are available to construct a full probabilistic description of this uncertain field. This project therefore proposes to develop a set of new interval techniques to perform such inverse quantification on tensor valued spatially uncertain model properties. Hereto, it is proposed to (1) develop a general methodological framework for the modelling and propagation of tensor valued, nonstationary interval fields, (2) extend a recently introduced framework for inverse interval quantification and (3) collaborate closely with Flemish industry to apply these methods to industrially relevant case studies.