High-dimensional bayesian optimization for finite element model updating and fault detection

Abstract

Within the field of Structural Engineering, Finite Element Analysis is ubiquitous, however discrepancies betweennumerical models and the actual structure in question will always exist to some extent. To remedy the inaccurateparameterization inherent to some of the discrepancies compared with the experimental data, the Finite ElementModel Updating Problem has to be solved. In this thesis the use of Bayesian Optimization as applied to FiniteElement Model Updating problems will be investigated and compared with the state-of-the-art Black BoxOptimizer: Covariance Matrix Adaptation-Evolutionary Strategy. Various acquisition functions will be examinedand a framework to extend Bayesian Optimization to high-dimensional problems will be implemented: TrustRegion Boundary Optimization. Finally Finite Element Model Updating will be used within the context of faultdetection in structures, to predict the location and weight of a point mass placed on the structure in question.