Bayesian geometric inverse problems

Project description

The aim of this project is to develop computational Bayesian techniques for the solution of geometric inverses problems that arise in a wide range of applications such as subsurface geophysics, manufacturing engineering and the built environment. Examples of geometric inverse problem that will be addressed with the techniques developed in this project are: (i) inference of fracture networks and/or conduits in Karst aquifers during the injection of CO2 for its geologic storage; (ii) detection of defects in reinforced preform during the resin infusion process in the fabrication of composite materials; (iii) inference of internal structures (e.g. cavities) in building structures such as walls with the aim of improving estimates of energy consumption. These problems have an underlying (forward) model described by Partial Differential Equation(s) (PDE) with input parameters associated to some (unknown) physical property of interest; the inverse problem is to infer this property from noisy observations of the solution of the PDE. In the context of problems (i)-(iii), sophisticated geometric parameterizations are often required to enable an accurate and realistic characterization of these properties (e.g. channelized structures in Karst networks). This project will develop computational hierarchical Bayesian methodologies to infer those geometry-constrained properties within an infinite-dimensional Bayesian framework for PDE-constrained inverse problems.