| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 11.07 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A presente dissertação tem como objetivos o desenvolvimento de ferramentas e procedimentos para a deteção de anomalias em estruturas de betão armado de forma semi automatizada recorrendo a técnicas de Deep Learning. Das várias anomalias que podem ser encontradas em estruturas de betão armado, destacam-se neste documento as fissuras e as armaduras expostas, para as quais foi desenvolvido um método alternativo de inspeção, baseado no processamento avançado de imagens, de modo a determinar se seria viável e vantajosa a substituição deste método pelo método de inspeção visual tradicional. As ferramentas desenvolvidas assentam em técnicas de Inteligência Artificial, em particilar as redes neuronais convolucionais (RNC), mais concretamente na utilização do transfer learning recorrendo à RNC Alexnet. Foram conduzidas experiências em diversas construções de modo a aferir a eficiência do método, recorrendo a veículos aéreos não tripulados (VANT), e em condições de acesso condicionado. Os resultados obtidos mostram ser bastante promissores, sendo previsível que do aprimoramento do método, possa resultar na adoção desta tecnologia como auxiliar nas inspeções visuais num futuro próximo.
This dissertation aim is to develop tools and procedures for the detection of anomalies in reinforced concrete structures in a semi-automated way using Deep-Learning techniques. Of the various anomalies that can be found in reinforced concrete structures, this document highlights the exposed rebars and cracks, for which an alternative inspection method was developed, based in advanced image processing, in order to determine whether it would be feasible and advantageous to replace this method with the traditional visual inspection method. The tools developed are based on artificial intelligence techniques, in particular, convolutional neural networks (CNN), more specifically in the use of transfer-learning using the CNN Alexnet. Experiments were conducted in several constructions in order to assess the efficiency of the method, using unmanned aerial vehicles (UAV), under conditions of conditioned access. The results show to be quite promising, and it is foreseeable that its improvement may result in the adoption of this technology for aid in visual inspections in the near future.
This dissertation aim is to develop tools and procedures for the detection of anomalies in reinforced concrete structures in a semi-automated way using Deep-Learning techniques. Of the various anomalies that can be found in reinforced concrete structures, this document highlights the exposed rebars and cracks, for which an alternative inspection method was developed, based in advanced image processing, in order to determine whether it would be feasible and advantageous to replace this method with the traditional visual inspection method. The tools developed are based on artificial intelligence techniques, in particular, convolutional neural networks (CNN), more specifically in the use of transfer-learning using the CNN Alexnet. Experiments were conducted in several constructions in order to assess the efficiency of the method, using unmanned aerial vehicles (UAV), under conditions of conditioned access. The results show to be quite promising, and it is foreseeable that its improvement may result in the adoption of this technology for aid in visual inspections in the near future.
Description
Keywords
Inspeção Anomalias Betão armado Deep-Learning Processamento de imagem Redes Neuronais Convolucionais (RNC) Veículos aéreos não tripulados Inspection Anomalies Reinforced concrete Image processing Convolutional Neural Networks (CNN) Unmaned Aerial Vehicles