AI Data-Centric 3D Machine Vision for Structural Element and Damage Identification

Traditional bridge inspection is a manually performed visual process that is time-consuming, costly, and requires significant support from equipment and resources. Recent advances in artificial intelligence (AI), represented by image-based deep learning (DL) methods, have accelerated the advances toward inspection automation enabled by robotic imaging and machine vision. Many researchers have reported that damage types and their locations can be identified in two-dimensional (2D) images with complex scenes through a DL-based semantic object detection or segmentation process. However, recent efforts in image-based structural damage assessment have not achieved a level of applicability as meaningful as those from traditional bridge inspection. We state that structural elements and damage patterns must be identified in a 3D space to approach the intelligence level of engineer-produced inspection. In this work, we adopt the data-centric principle of AI development by exploiting off-the-shelf AI models and then develop a novel 3D structural identification framework. Combining a 3D geometric learning network and a transformer-based object-detection network, the proposed DL models can output 3D maps of structural elements and structural damage, enabling truly engineering-meaningful damage assessment and mapping. By conducting numerical experiments and validation, it is expected that the proposed methodology can contribute to the vision of realizing autonomous and low-cost structural inspection in the near future.