Research on Modeling Strategy of Ancient Stone Arch Bridges Based on Masonry Structure Gap Image Recognition

The conservation of ancient stone arch bridges is severely hindered by drawing deficiency, difficult on-site survey, and structural deterioration. These obstacles make it difficult to acquire the geometric parameters required for refined mechanical models and to reproduce the real damage state of individual blocks. As a result, refined mechanical models can hardly be established effectively. To solve the problem, a finite-element (FE) modeling strategy of ancient stone arch bridges based on masonry structure gap image recognition is proposed. First, a dataset with a large number of labeled contours of blocks in stone arch bridges was constructed, and a trained YOLOv8 convolutional neural network was used for instance segmentation of the block contours on the bridge images. Second, the recognition results were post-processed with the Douglas–Peucker algorithm, and key geometric information of individual blocks was extracted. Finally, a parametric modeling procedure was developed: an ABAQUS parametric modeling script was developed to automatically generate a separate FE model that faithfully replicates the actual masonry structure, with contact interfaces defined between blocks for subsequent FE simulation analysis. The results show that under self-weight and deck loads, the peak principal stress in the arch rib predicted by the separate FE model is about 1.2 times that given by a conventional monolithic FE model, and conspicuous stress concentrations appear at masonry defects. The separate model can more accurately reproduce the block distribution and local defects of the actual bridge. It has significant advantages for revealing the damage mechanism of the masonry structure of ancient bridges, and provides a new perspective and method for the mechanical simulation study of ancient bridge protection.