ABSTRACT
Vehicles represent the main operational loadings of bridge structures since they are designed and constructed to sustain passing vehicles. Among others, structural responses due to the moving vehicles play a crucial role in bridge health monitoring, condition assessment, maintenance. Traditionally, vehicle-induced dynamic responses can be obtained through a variety of numerical computational models. Depending on the complexity of existing models, they could become either computationally expensive or hard to be implemented in regular engineering practice. Moreover, as physical-based methods, a good number of parameters associated with materials, mechanics, and geometry are required to establish a reliable model reflecting the physical structure behaviour. Recently, the literature has seen an increasing research efforts in dynamic response estimation of civil structures subjected to vehicle, seismic, or wind load using machine learning techniques. This paper presents an investigation of predicting dynamic responses of bridges by neural network under vehicle loadings. Representative analysis model is employed to create training data pairs for learning parameters in the network. The state-of-art deep learning technique, namely, convolutional neural network is adopted to develop the metamodel of the vehicle and bridge system, capable of delivering time history response to be compared with results from numerical experiments. The results show that the predictive model yields high prediction accuracy and is robust to data noise.
