ARTIFICIAL INTELLIGENCE - BASED METHODOLOGY FOR BRIDGE PATHOLOGY ANALYSIS APPLIED TO DIGITAL TWINS
BIM; digital twin; bridge inspection; IFC-based semantic enrichment
Bridge asset management is crucial for cost-effective maintenance management, although it entails specialized human resources that are costly and comes with limitations, such as subjective interpretation and lack of precision. With technological advancements in deep learning algorithms, computer vision-based BIM software as strategy analytics results, enabling the detection of damages, small cracks, corrosion, or even the prediction of pathological manifestations. This study aims to investigate and develop techniques and approaches based on artificial intelligence to improve and optimize the process of mining data from inspections of bridges to improve design, operation and maintenance stages using BIM methodologies, integrating Digital Twins. This works involved collecting monitoring data from 560,000 inspection images from the Special Works Management System (SGO) of the Brazilian National Department of Transportation Infrastructure (DNIT) for training Convolutional Neural Networks (CNNs) with detecting pathological manifestations. YOLOv5 was trained for object labeling and SAM for parameter segmentation, as well as their respective dependencies. Thus, an IFC-based semantic enrichment process from data collected with BIM and DT aims share information and automate models’ information in real time. A scan-to-BIM modeling of a viaduct located in Brasília - Brazil, is presented as a case study to validate the approach. This methodology should real-time update the model, based on routine bridge inspection. The AI applied on this, must use scan-to-BIM data to recognize the elements, identify pathologies and export 3D features and tabular information for updating an IFC model. Connecting these results in a Common Data Environment (CDE) will allow as-is model visualization shared from IA and its real-time updating.