Bibliographic citations
Espinola, L., (2024). Detección y medición automática del descascaramiento superficial de estructuras de concreto armado mediante aprendizaje profundo [Tesis, Universidad de Ingeniería y Tecnología]. https://hdl.handle.net/20.500.12815/395
Espinola, L., Detección y medición automática del descascaramiento superficial de estructuras de concreto armado mediante aprendizaje profundo [Tesis]. PE: Universidad de Ingeniería y Tecnología; 2024. https://hdl.handle.net/20.500.12815/395
@misc{renati/900647,
title = "Detección y medición automática del descascaramiento superficial de estructuras de concreto armado mediante aprendizaje profundo",
author = "Espinola Diaz, Luis Alexander",
publisher = "Universidad de Ingeniería y Tecnología",
year = "2024"
}
In Peru, bridges play an essential role in road infrastructure, but often receive little attention. This neglect is reflected in the limited information available on these bridges, as only a quarter of the registered bridges have been inspected. Conventional methods for bridge inspection require specialized personnel and considerable investment, which is unfeasible for the frequency required in the Peruvian context. The objective of this thesis is to propose a system to accelerate the preliminary inspection of bridges, facilitating the collection and identification of concrete spalling. The proposed method uses convolutional neural networks for the semantic segmentation of images of spalling on concrete surfaces. Two architectures, YOLOv8 and Unet, trained on 200 images of civil structures, were evaluated. The results indicate that the YOLOv8 model achieved an F1- score of 0.68, while the Unet model obtained an F1-score of 0.89 in the spalling segmentation. In terms of diameter prediction according to its F1-score, YOLOv8 had 52%, and Unet 61%. It was observed that prediction worsened with increasing diameter and geometric variability of the defect, which is attributed to the imprecise definition of diameter in the inspection guides and the template used. Analysis of the RGB scale images showed that damage to the concrete surface can be identified by differences in colour intensity, although no clear correlation was found between damage depth and intensity values.
This item is licensed under a Creative Commons License