An Artificial Neural Network Approach to Structural Condition Assessment of a Reinforced Concrete Viaduct

Resumo

Early damage detection is essential for asset managers to implement less invasive, cost-effective maintenance strategies and extend service life. Finite element models enable the simulation of complex scenarios that could be impractical to test on full-scale structures, making them valuable tools for managing these assets. The variability of operational and environmental conditions that influence the structural response can effectively conceal the effects of damage, thereby increasing the uncertainty associated with the damage detection task. Artificial Neural Networks (ANNs) specialize in pattern recognition and can handle large amounts of data. However, a key consideration regarding their use concerns the selection of input variables, as environmental effects can interfere with input data quality. This study presents a methodology based on Artificial Intelligence (AI) methodology for detecting and locating damage in a reinforced concrete viaduct. The case study investigates a viaduct located on the SP-340 highway, an important connection between the cities of Rio Claro and São Carlos, Brazil. Artificial damage scenarios with varying severity levels were created in critical structural elements were simulated to train ANNs. Different levels of white noise were introduced into the mode shapes extracted from numerical models. This step aimed to approximate these data to those captured in field tests and test the algorithm’s robustness to noise. Two ANNs were developed: one for detecting and locating damage, and another for estimating its severity. This AI-driven methodology aims to provide a support tool for structural health monitoring, combining numerical modeling and machine learning to enhance early damage detection.