Bridge Displacement Prediction from Dynamic Responses of a Passing Vehicle Using CNN-GRU Networks

Abstract

Dynamic displacement is an important indicator for bridge safety estimation but is difficult to measure due to economic or technological limitations. Dynamic responses of a passing vehicle include the bridge dynamic response information. This study proposes a framework utilizing artificial neural networks to efficiently and accurately predict bridge displacements from the dynamic response of a passing vehicle. The input and the output of the networks are the vehicle acceleration responses and the bridge dynamic displacement responses, respectively. The implemented framework consists of convolutional neural network (CNN) and gated recurrent units (GRU). CNN is adept at feature extraction in the microcosm of short-term time series, revealing intricate nuances. As a complement, GRU plays a crucial role in extracting features of macroscopic long-term time series. The CNN-GRU network can efficiently extract higher-order features contained in the input data. Numerical experiments are conducted using the developed vehicle-bridge interaction (VBI) system model to obtain requisite data for training the deep neural network. The impact of the presence or absence of roadway irregularities and the number of vehicles are discussed, indicating the accuracy of the framework. Furthermore, a laboratory experiment is conducted to further assess the performance of the CNN-GRU network. Results indicate that the CNN-GRU network offers an effective alternative for bridge displacement measurements.