Drive-by Damage Detection In Railway Bridges Using 1d Convolutional Neural Networks

Resumo

The aging of civil engineering infrastructure emphasizes the importance of structural health monitoring
(SHM) in railway bridges. Data-driven models are a prominent approach in this field. However, network-wide

bridge instrumentation is logistically difficult and expensive, leading to the development of the drive-by or indi-
rect monitoring method. In the drive-by approach, the instrumented vehicle acts as the SHM system’s actuator

and receiver. Environmental and operational conditions can affect structure properties and measured acceleration
signals, making it challenging to infer the real condition of the structure. To address this, we propose a drive-by

damage detection method using a classification model with a Convolutional Neural Network (CNN). CNNs under-
stand connectivity patterns between neurons, inspired by the animal visual cortex. The 1D CNN identifies damage

from raw acceleration signals at the front boogie of a train. Training data generated by a finite element method

considers healthy and damaged bridge conditions. The paper focuses on identifying scour-induced damage, result-
ing from the loss of stiffness in the bridge support due to erosion. Extensive numerical experiments evaluate the

effectiveness and robustness of the 1D CNN approach.