A hybrid machine learning framework for bridge damage detection and localization using vibration signals

The care, durability, and endurance of modern bridge structures are vital aspects as they are complex engineered systems involving advanced materials, design, and embedded technologies. Over time, they are exposed to various types of damage, including loosened connections, cracks, and degradation due to various environmental factors. This can lead to a decline in the integrity and serviceability of the infrastructure. For detecting such damage, visual inspection and manual analysis are often labor-intensive and lack various ability to perform in real-time. To address this gap, this research develops a robust, hybrid machine learning model for precise and accurate damage identification as well as localization in the bridge infrastructure. This paper mentions an investigation on a real life KW51 bridge. This paper integrates IoT-enabled vibration data acquisition, denoising through Autoencoders, and feature engineering, followed by a hybrid ML architecture that combines SVM and GNN, optimized using PSO and DANN. Findings demonstrate that the suggested approach acquired 98% accuracy in damage detection and precisely locates the damage on the structure. To successfully differentiate between the normal and damaged condition of the bridge, a unified DI is used. This research concludes a scalable, real-time method for monitoring bridge health. Enhancing the model’s generalizability, on-site edge computing, and adoption in infrastructure management can be a focus in the future.