When flowing water erodes material from bridge foundations – exacerbated by heavy flooding and extreme weather events – that’s scour. Heavier storms and changing climate patterns are adding stress, on infrastructure, scour can threaten the stability of bridge foundations and eventually lead to serious structural problems. Researchers Thiago Moreno Fernandes, Leonardo Minski, Pedro Vinicius Gasparotti de Souza, Diogo Rodrigo Ferreira Ribeiro, Leandro Fleck Fadel Miguel, and Rafael Holdorf Lopez address this persistent threat to bridge safety in their paper “Early Scour Damage Detection Using Drive-By Monitoring Data through Supervised Learning.” To address this risk, the authors investigated whether artificial intelligence and data collected from ordinary vehicles crossing a bridge could be used to identify early signs of damage before they become visible through traditional inspection methods.
Using computer-simulated drive-by monitoring data, the team developed a supervised learning approach designed to distinguish between an undamaged bridge and several levels of scour-related damage. Rather than focusing on costly or difficult-to-install monitoring systems, the research looks at how information gathered from passing vehicles could provide insight into a bridge's condition. The study offers a glimpse into how machine learning may help engineers monitor infrastructure more efficiently and prioritize inspections where they are needed most. Learn more about how emerging bridge monitoring tools can support improved infrastructure resilience in the Journal of Structural Design and Construction Practice at https://ascelibrary.org/doi/10.1061/JSDCCC.SCENG-1785. The abstract is below.
Abstract
Early detection of damage is crucial for maintaining the safety and integrity of infrastructure. Scour at bridge foundations is particularly critical, as it poses a significant risk of collapse, which is heightened by the increasing frequency of extreme weather events such as floods. This paper introduces a supervised learning approach for early scour damage detection using drive-by monitoring data considering a small number of vehicle passages. A convolutional neural network model is employed to classify different levels of scour damage using two methodologies. The first methodology incorporates only vehicle acceleration measurements and the second methodology includes both acceleration and vehicle speed information. To handle data variability and neural network randomness, classification accuracy is evaluated through confusion matrices and boxplots. The model’s performance in classifying scour scenarios is assessed from sensors positioned on the car body and the front bogie of the first vehicle. The results show high accuracy in detecting scour for the sensor positioned in the car body, achieving approximately 100% accuracy in classifying all scenarios when vehicle speed information is included.
Learn more about this novel way to detect bridge scour earlier in the ASCE Library: https://ascelibrary.org/doi/10.1061/JSDCCC.SCENG-1785.
