Speaker
Description
Bridges are complex structural systems that can undergo significant changes of their structural conditions over time. Due to dynamic loads, they are particularly vulnerable to fatigue, which can lead to losses of financial resources and human lives. Traditionally, rainflow diagrams based on strain time histories have been used to assess fatigue, relying on direct strain measurements. However, this method has several limitations, including the equipment costs and difficulty in accessing critical areas. An AI-based approach offers a promising alternative. This study proposes a hybrid approach that combines low-cost accelerometer data with prior structural knowledge to predict strain signals. By leveraging cutting-edge deep learning algorithms such as Graph Neural Networks (GNNs), we present a framework that learns the complex relationship between acceleration and strain, enhanced by supporting physical information. The creation of virtual sensors allows strain signals to be obtained at critical structural positions, leading to an accurate assessment of the remaining useful life of the structure.
