Damage-Detection Approach for Bridges with Multi-Vehicle Loads Using Convolutional Autoencoder

Abstract

Deep learning has been widely employed in recent studies on bridge-damage detection to improve the performance of damage-detection methods. Unsupervised deep learning can be effectively utilized to increase the applicability of damage-detection approaches. Hence, the authors propose a convolutional-autoencoder (CAE)-based damage-detection approach, which is an unsupervised deep-learning network. However, the CAE-based damage-detection approach demonstrates only satisfactory accuracy for prestressed concrete bridges with a single-vehicle load. Therefore, this study was performed to verify whether the CAE-based damage-detection approach can be applied to bridges with multi-vehicle loads, which is a typical scenario. In this study, rigid-frame and reinforced-concrete-slab bridges were modeled and simulated to obtain the behavior data of bridges. A CAE-based damage-detection approach was tested on both bridges. For both bridges, the results demonstrated satisfactory damage-detection accuracy of over 90% and a false-negative rate of less than 1%. These results prove that the CAE-based approach can be successfully applied to various types of bridges with multi-vehicle loads.