Automated and intelligent road monitoring system using embedded fibre bragg grating sensor networks

Abstract

Modern road infrastructure must operate safely and efficiently despite increasing traffic demands, ageing materials, and changing environmental conditions. Ensuring its long-term performance requires reliable information about how pavements respond internally over time. However, conventional inspection and testing methods provide only limited snapshots of structural condition and are often costly, disruptive, and reactive. Embedded sensing technologies, particularly fibre Bragg grating (FBG) sensors, enable continuous in-situ pavement monitoring. Nevertheless, their practical application remains limited by the large volumes of data generated and the difficulty of extracting decision-oriented information from raw measurements. This thesis develops a comprehensive framework for automated road structural health monitoring using embedded FBG sensor networks and artificial intelligence. Adopting a system-level perspective, the framework integrates sensing, data management, and learning-based interpretation to transform raw measurements into compact and meaningful representations of pavement condition. The framework introduces two complementary monitoring levels. At the local level, data-driven anomaly detection identifies abnormal responses through feature extraction and unsupervised pattern recognition. At the global level, deep-learning models extract long-term trends and generate health indicators (HIs) that summarise structural condition over extended monitoring periods. A key focus is the automation and reduction of long-term monitoring data through representation learning. The proposed approach combines signal processing with neural-network architectures, including autoencoders and recurrent models, to compress high-frequency measurements while preserving information related to traffic loading, environmental effects, and structural response. Particular attention is given to distinguishing temperature-driven behaviour from traffic-induced responses and learning compact latent representations that evolve with structural condition. The methodology is validated using field measurements from instrumented road sections, laboratory experiments, and simulated data. This multi-source validation demonstrates that the proposed methods are physically meaningful and applicable to real monitoring conditions. The results show that large volumes of sensing data can be converted into lightweight and interpretable indicators suitable for continuous monitoring and future edge-based implementation. Overall, this dissertation bridges the gap between embedded sensing and artificial intelligence, providing a scalable and automated pathway towards proactive pavement monitoring, improved maintenance planning, reduced lifecycle costs, and more resilient road networks. This thesis addresses this challenge by developing a comprehensive framework for automated road structural health monitoring using embedded FBG sensor networks combined with artificial intelligence methods. Rather than focusing only on sensor installation or isolated analysis techniques, the research adopts a system-level perspective that integrates sensing, data management, and learning-based interpretation to transform raw measurements into compact and meaningful representations of pavement condition. The central premise of the dissertation is that continuous monitoring is only effective when the collected data can be efficiently processed, learned from, and translated into information that supports management and maintenance decisions. To achieve this goal, the thesis introduces a two-level monitoring concept. At the local level, data-driven anomaly detection methods are used to identify abnormal structural responses through feature learning and unsupervised pattern recognition. At the global level, deep learning models are employed to extract long-term trends and generate HIs that summarise structural condition over extended monitoring periods. A key focus of the research is placed on automation and data reduction through representation learning. Long-term monitoring campaigns can generate extremely large datasets, making storage, transmission, and manual interpretation impractical. The proposed framework combines signal processing with neural network architectures such as autoencoders and recurrent models to compress high-frequency sensor data while preserving information related to traffic loading, environmental effects, and structural response. Particular attention is given to separating temperature-driven behaviour from traffic-induced responses and to learning compact latent representations that evolve with structural condition. The methodology is validated using a combination of field measurements from instrumented road sections, laboratory experiments, and simulated data. This multi-source validation ensures that the proposed learning-based methods are both physically meaningful and applicable to real monitoring scenarios. The results demonstrate that large volumes of raw sensing data can be transformed into lightweight and interpretable indicators suitable for long-term monitoring and future in-situ or edge-based implementation. Overall, this dissertation contributes to the development of intelligent and data-driven road infrastructure by bridging the gap between continuous sensing technologies and modern artificial intelligence. The proposed framework provides a scalable and automated pathway toward proactive pavement monitoring, with the potential to improve maintenance planning, reduce lifecycle costs, and enhance the long-term resilience of road networks.