Detecting and localising bridge stiffness loss using autoregressive models and curvature profiles

Ensuring the safety and serviceability of bridges is vital for the reliability of transportation networks, yet continuous Structural Health Monitoring (SHM) remains challenging due to the high cost and data burden of dense sensor networks. Many vibration-based approaches require extensive instrumentation and centralized processing, limiting their practical scalability. This thesis develops a sparse-sensor SHM framework that integrates unsupervised Autoregressive (AR) models with curvature-based analysis for damage detection, localization, and scour assessment in bridges. The proposed AR method performs sensor-level detection using Mahalanobis distance as a damage-sensitive feature, enabling decentralized decision-making without dependence on a global model. The approach was validated using numerical and experimental bridge data, demonstrating comparable sensitivity and robustness compared with frequency-based techniques. Complementing this, a novel AR-residual approach reconstructs quasistatic curvature profles from vehicle passages, enabling high-resolution damage localization in multi-span structures. Finally, the framework is extended to scour detection around bridge foundations by linking curvature variations to bendingmoment changes, enabling identifcation of both the location and severity of foundation damage. Together, these developments establish a low-cost, data-efcient SHM protocol that transitions from local unsupervised detection to global structural diagnosis. The framework enhances the feasibility of continuous, autonomous bridge monitoring using sparse sensors and ordinary trafc excitation.