Bridge structural damage identification based on parallel CNN-GRU

Abstract

Structural damage identification has been the focus of engineering fields, while the existing damage identification methods heavily depend on extracted “hand-crafted” features. Recently, due to the powerful feature learning capability of deep learning, it has been widely used in structural damage identification. However, those methods only consider the local dependence or temporal relation of data. Thus, in this paper, a structural damage identification method by combining the convolutional neural network (CNN) and gated recurrent unit (GRU) is proposed. The CNN model is used to extract the local dependence of data, and the GRU model is used to extract the temporal feature of data. These two extracted feature matrices are spliced horizontally to a fused eigenvector. The eigenvector is input to the final softmax classifier layer to identify the structural damage state. Experiments on a scale model of the three-span continuous rigid frame bridge shown that the CNN-GRU model performs significantly better than CNN, LSTM, and GRU models for structural damage identification.

Operational Modal Analysis on Bridges: A Comprehensive Review

“…Moreover, Duan et al [310] applied CNNs for damage identification in hanger cables of a tied-arch bridge. Zou et al [311] proposed a CNN model by incorporating temporal features from the gated recurrent unit (GRU) model, demonstrating significant improvement in structural damage identification compared to other models. Yessoufou et al [312] used a one-class convolutional neural network (OC-CNN) model capable of detecting bridge damage across various vehicle weights and speeds.…”

A New Framework for Isolating Sensor Failures and Structural Damage in Noisy Environments Based on Stacked Gated Recurrent Unit Neural Networks

“…In addition to the input and output layers, the parameters of the stacked GRU layer, FC1 layer, FC2 layer and dropout layers need to be set. We have optimized the parameters using the vertical comparison method [37], i.e., when analyzing the effect of one of the parameters on the prediction results, the rest of the parameters are fixed and the optimal range is derived from literature [21,[38][39][40]. The optimized network parameters were set as follows: q 1 (stacked GRU) = 100, q 2 (FC1) = q 3 (FC2) = 128, and the dropout layers were added to the stacked GRU, FC1 and FC2 layers at a rate of 0.3.…”