Detectability of Bridge-Structural Damage Based on Fiber-Optic Sensing through Deep-Convolutional Neural Networks

Improving the accuracy and efficiency of damage detection of bridge structures is a major challenge in engineering practice. This paper aims to address this issue by monitoring the continuous bridge deflection based on the fiber optic sensing technology and applying a deep-learning algorithm to perform structural damage detection. With a scaled-down bridge model, three categories of damage scenarios plus an intact state were simulated. A 13-layer supervised learning model based on the deep convolutional neural networks was proposed. After the training process of original continuous deflection under 10-fold cross-validation, the model accuracy can reach 96.9% for damage classification with the performance outperforming that of the other four methods (random forest = 81.6%, support vector machine = 79.9%, k-nearest neighbor = 77.7%, and decision tree = 74.8%). The proposed model also demonstrated its decent abilities in automatically extracting damage features and distinguishing damage from structurally symmetrical locations.