Dual-Graph Collaboration: Bidirectional Fusion Graph Convolution Network for Structure Multidefect Positioning and Assessment

Graph convolution network can extract structural multidefect information well, and has been widely concerned in the field of structural damage detection. However, it is difficult to locate and evaluate defects of different sizes only using a single graph structure, and the similarity of nodes is destroyed in the aggregation process. Therefore, this article proposes a dual-graph collaborative bidirectional fusion graph convolution network. First, a dual-graph structure including fixed graph and feature graph is constructed to make up for the lack of information caused by a single structure. Then, the multiscale adaptive fusion of the dual-graph structure is carried out in the horizontal direction, and the node similarity is effectively maintained while using the graph structure. The related information of dual-graph structure and time is deeply mined in the vertical direction, and the multilevel and multitemporal spatio-temporal information is flexibly aggregated. Finally, the accurate location and evaluation of the multiple damage of the structure are realized. Two typical structural multidamage case studies are conducted to verify the performance of the proposed model.