Efficient uncertainty quantification and propagation in performance-based earthquake engineering
No Thumbnail Available
Date
2025-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de los Andes
Abstract
Description
In recent decades, the constant deterioration of existing infrastructure and the increasing exposure to natural hazards driven by geological processes and changing climate conditions have motivated the development of a new philosophical approach to structural engineering, known as performance-based engineering. Its goal is to provide a rigorous, science-based framework through a comprehensive assessment of structural risk, ultimately delivering a decision variable that is useful for practical decision-making. To this end, performance-based engineering establishes a probabilistic framework that aims to address uncertainty regarding (i) the hazards to which the structure is exposed, (ii) the actual behavior of the structure versus that predicted by the engineering model, and (iii) the damage caused when certain intensity levels are exceeded. At each of these stages, properly quantifying uncertainty and subsequently propagating it through the following stages is critical for a successful risk assessment. In this context, methodological progress has been gradual, supported by technological advances that have enabled the implementation of probabilistic methods. However, the cost of adopting a probabilistic framework has been high, especially due to the need for largescale simulation of finite element models, which requires significant computational and time investment. For this reason, developing methods that enable efficient uncertainty quantification and propagation in performance-based engineering remains an open challenge and a key area of current research. This thesis presents two approaches aimed at providing efficient methods for uncertainty quantification and propagation by supporting structural simulations with machine learning surrogate models using Gaussian processes. The first approach focuses on quantifying and decomposing parameter-induced uncertainty in structural responses under specific hazard scenarios. Its goal is to support probabilistic sampling-based analyses, including model calibration and updating, iterative performancebased design, and sensitivity analysis. The second approach focuses on the quantification, propagation, and decomposition of uncertainty in structural vulnerability assessment under a broad range of seismic events. This approach implements and discusses the performance-based engineering framework from a philosophical standpoint, although applied to a real-world case study. Available definitions of damage states in bridge components and the relationships between these and their consequences are discussed. Both approaches are developed in a fully probabilistic setting, including probabilistic seismic hazard analysis, probabilistic structural modeling, and uncertainty decomposition.