On Regression in Population based Structural Health Monitoring

Complex engineering infrastructure is commonplace among modern society. Fatigue, erosion, and adverse environmental factors degrade these infrastructure over time, requiring maintenance to restore proper function. Structural Health Monitoring (SHM) provides proactive solutions to ensure structural safety and reliability, SHM systems consist of data acquisition and processing systems to enable the detection of damage in monitored structures. Data-based approaches to SHM have demonstrated considerable promise, but their widespread adoption remains limited by a fundamental challenge: the scarcity of labelled data. Acquiring observations that capture the full spectrum of structural health conditions — from the healthy state through to advanced damage — is costly, disruptive, and often practically unfeasible for in-service assets. Population-based structural health monitoring has emerged as a principled response to this challenge. By treating a collection of structures jointly rather than as isolated assets, PBSHM enables information to be shared across population members — allowing data-rich assets to inform inference for data-scarce ones. While considerable research effort has been directed toward developing population-based technologies for classification tasks, comparatively little attention has been paid to regression, where the quantity of interest is a continuous health variable rather than a discrete damage category. To continue the development of PBSHM systems, regression based technologies are required. Regression further enables damage severity to be quantified, degradation trajectories to be tracked, and predictions to be made about the remaining useful life of a structure. These capabilities are essential for the informed, cost-effective operation and maintenance of structural assets, and their development for population-based settings is the central concern of this thesis.