Structural damage detection using an unmanned aerial vehicle-based 3D model and deep learning on a reinforced concrete arch bridge
Visual inspection is a common method for detecting structural damage, but has limitations in terms of subjectivity, time, and access. This research proposes an innovative approach to identify cracks using a 3D model generated from photographs of an unmanned aerial vehicle (UAV) and the use of a conv...
| Autores: | ; ; ; ; |
|---|---|
| Tipo de documento: | Artigo |
| Data de publicação: | 2025 |
| País: | España |
| Recursos: | Universitat Politècnica de Catalunya (UPC) |
| Repositório: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | Inglês |
| OAI Identifier: | oai:upcommons.upc.edu:2117/424657 |
| Acesso em linha: | https://hdl.handle.net/2117/424657 https://dx.doi.org/10.3390/infrastructures10020033 |
| Nível de acesso: | Acesso aberto |
| Palavra-chave: | Structural damage detection Bridge 3D model Binary segmentation Deep learning Convolutional neural networks Àrees temàtiques de la UPC::Enginyeria civil::Materials i estructures::Tipologies estructurals |
| Resumo: | Visual inspection is a common method for detecting structural damage, but has limitations in terms of subjectivity, time, and access. This research proposes an innovative approach to identify cracks using a 3D model generated from photographs of an unmanned aerial vehicle (UAV) and the use of a convolutional neural network (CNN). These networks are effective in detecting complex patterns, improving the accuracy and efficiency of damage identification based on simple visual inspection. The case study is the old Villena Rey bridge in Lima, Peru. The methodology covers (i) the development of a 3D model of the bridge structure, (ii) the extraction of photographs of the model and its binary segmentation, (iii) the application of deep learning through the training and testing phase of a CNN to achieve crack detection in photographs, and (iv) damage location within the 3D model. An 88.4% accuracy was achieved in crack detection, identifying 18 damage points, of which 3 turned out to be false positives. Additionally, it was determined that the left pillar in the southern area of the bridge presented the highest concentration of damage, which underlines the effectiveness of the method used. |
|---|