A novel framework for probabilistic model updating using load test data and Copula-Based Bayesian Networks

Abstract

Accurate model updating is essential for reliable structural assessment and performance prediction of civil infrastructure. Classical Bayesian model updating approaches, while powerful, are often limited by their reliance on predefined parametric distributions, and by the computational burden of sampling-based inference, which becomes prohibitive as the number of updating parameters increases. These limitations hinder their applicability to complex structures where high-dimensional parameter spaces and uncertain parameter-output relations are prevalent. The restriction on the number of updating parameters often leads to a mismatch between the global-level formulation of the updating problem and the local structural behavior or failure mechanisms of interest. To address these limitations, this paper proposes a novel model updating framework that integrates load test data, Finite Element Modelling (FEM), and Copula-Based Bayesian Networks (CBBNs). The approach uses load test measurements to calibrate FEM parameters, and employs CBBNs to efficiently capture complex dependencies among model variables. This paper presents the framework’s formulation and implementation, along with key verification steps. The flexibility and computational tractability of the CBBN-based framework make it particularly suited for structural engineering applications, contributing to more reliable data-driven asset management.