Xue, Zechang; (2025) Vibration-based Structural Health Monitoring for Three Towers Cable-stayed Bridges Using Deep-learning. Doctoral thesis (Ph.D), UCL (University College London).

Text Xue_10203537_Thesis.pdf Access restricted to UCL open access staff until 1 February 2026. Download (7MB)

Abstract

This thesis enhances structural health monitoring (SHM) for large-span cable-stayed bridges by integrating Finite Element (FE) technology, SHM sensor data, and advanced deep learning techniques. This research focuses on the Queensferry Crossing, located in East Scotland, which is the world's longest three-tower cable-stayed bridge and was opened in 2017. It develops and validates a simplified yet highly accurate 3D FE model that optimizes data generation for AI-based structural analysis and significantly reduces computational demands while maintaining precision. Leveraging a comprehensive network of multi-sensors, the study utilizes both real-time sensor measurements and simulated data—acceleration signals obtained from FE models simulating damage under white noise excitation—to train and refine a Transfer Learning Convolutional Neural Network (TL-CNN). TL-CNN is an advanced machine learning approach that improves learning efficiency by applying knowledge gained from previous tasks to new but related tasks. This dual-source approach ensures extensive data availability at low cost and enables the TL-CNN to perform precise defect detection and structural integrity assessments under varied environmental conditions, including different wind speeds and temperatures. Over a year-long monitoring period and through extensive numerical simulation testing using FE model-generated datasets, the TL-CNN model demonstrated exceptional accuracy in localizing structural damages, validating its potential for monitoring large infrastructures. These advancements significantly enhance predictive capabilities for bridge defects without disrupting ongoing bridge operations and address critical gaps in data sufficiency and operational feasibility. This thesis confirms the potential of high-fidelity FE models and deep learning algorithms for real-time, precise structural assessments that could significantly extend the service life of cable-stayed bridges and propel further developments in this area of research.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item