Abstract

Advances in machine learning and sensors have improved how we model and monitor structures, yet challenges remain in extracting meaningful information from high-dimensional, noisy data to create interpretable low-order models. This talk presents data-centric methodologies that combine Gaussian Processes (GPs) as a statistical regression tool with low-order models for probabilistic inference of structural behaviour. These approaches are designed to address problems in structural (aero-)elasticity, including modelling nonlinear aeroelastic forces acting on bluff bodies, inferring the behaviour of elastic beams and plates based on heterogeneous data, and the inverse reconstruction of dynamic loads. Additionally, a novel Bayesian method utilising structured variational inference is introduced for parameter and state estimation of dynamical systems. Potential applications of this method include reverse engineering low-order dynamical systems, such as limit cycle attractors in vortex-induced vibrations. In closing, and somewhat loosely related to the main topics, a probabilistic structural analysis of a reconstruction of the Lefkandi-Toumba Building, an ancient Greek longhouse from the Greek Dark Ages, will be presented.