فهرست نشریات علمی دسترسی آزاد ایران

In recent decades, numerous innovative methods have been developed to improve the quality, speed, cost-effectiveness, and efficiency of structural damage detection. Among these, computer vision has emerged as a  an auspicious approach, particularly due to recent advancements in machine learning. Many successful models have been developed to detect a wide range of structural damages such as cracks, spalling, corrosion, rusting, and bolt loosening. However, shear buckling damage has been almost entirely neglected in the literature. This type of damage commonly occurs in thin steel plates used in structural members such as steel plate shear walls, and its accurate detection and localization are essential for reliable post-event condition assessment, especially following seismic loading. This study investigates the application of computer vision for both detection and instance-level localization of shear buckling damage using an instance segmentation framework. A key challenge in this task is the limited availability of labeled real-world images. To address this issue, an innovative data augmentation strategy is proposed that combines synthetic images generated using finite element analysis (FEA) with visually enhanced synthetic images created using 3D modeling software. These synthetic datasets are then combined with real experimental images to form larger and more diverse training datasets. In total, five datasets were considered, including 208 real images, 343 synthetic images, and 551 combined real–synthetic images. All models were trained using the YOLO11 instance segmentation algorithm. Results demonstrate that a model trained solely on real images achieved strong segmentation performance, with a precision of 0.87 and a recall of 0.81. The best-performing model, trained using a combination of real and visually enhanced synthetic images, achieved a precision of 0.90 and a recall of 0.84, corresponding to an improvement of approximately 3% in mAP50 compared to the real-only model. These findings confirm that high-fidelity synthetic data can effectively mitigate data scarcity and significantly enhance shear buckling detection and localization performance.