Abstract

Population-based structural health monitoring (PBSHM) aims to share valuable information within a population, such as normal- and damage-condition data, to improve predictions regarding the members’ health states. Homogeneous populations, comprised of nominally-identical structures, and heterogeneous populations, characterised by greater disparities among the members, both exhibit dynamic variability because of factors such as material properties, geometry, boundary conditions, and environmental effects.

Many SHM strategies rely on monitoring these dynamic properties, so benign variations pose challenges to system implementation and generalisation. Hierarchical (multilevel) Bayesian models with partial pooling have been shown to be effective for homogeneous systems. These models simultaneously consider population-level similarities and individual-level differences, leading to more robust statistical inferences and reduced variance in parameter estimates, particularly when data are sparse. Heterogeneous populations have members that are farther apart in the feature space, and information transfer within these populations is considerably more challenging. However, geometrical transfer approaches (e.g., geodesic flows) are an exciting prospect, as they are naturally equipped to capture the complex curvature of these spaces.

This talk outlines the application of hierarchical Bayesian models to SHM and PBSHM via experimental case studies. Preliminary findings on a transfer method for heterogeneous populations are also introduced.