| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 15.26 MB | Adobe PDF |
Advisor(s)
Abstract(s)
A monitorização e inspeção de ativos industriais são essenciais para garantir a segurança dos
utilizadores, no entanto caracterizam-se por serem processos morosos, dispendiosos e sujeitos
a erro humano. Com o avanço tecnológico, é possível efetuar a realização de inspeções mais
seguras e eficientes, minimizando o trabalho e o risco para os intervenientes. A aplicação de
algoritmos de IA na segmentação de instâncias em imagens destaca-se na deteção de anomalias
estruturais, como são os casos da Mask R-CNN e do YOLO. Quando combinadas com
plataformas UAV, estas redes tornam-se ferramentas poderosas para inspecionar ativos em
zonas de difícil acesso, assegurando maior precisão. A tecnologia de Ray Casting complementa
este processo, permitindo mapear danos identificados pela inteligência artificial nos modelos
3D concebidos através de tecnologias como o LiDAR ou a Fotogrametria. Neste trabalho foi
desenvolvida uma metodologia com capacidade de mapear corrosão, choques mecânicos e
acumulações de água em nuvens de pontos de ativos industriais com revestimentos em painéis
sandwich, utilizando TLS e UAV. Os modelos treinados através do Detectron2 e do YOLOv8
conseguem segmentar os três tipos de danos com uma precisão de 68,8 % e 72,5 %.
The monitoring and inspection of industrial assets are essential to guarantee the safety of users, but they are time-consuming, costly and subject to human error. With technological advances, it is possible to carry out safer and more efficient inspections, minimizing work and risk for those involved. The application of AI algorithms in the segmentation of instances in images stands out in the detection of structural anomalies, as is the case with Mask R-CNN and YOLO. When combined with UAV platforms, these networks become powerful tools for inspecting assets in hard-to-reach areas, ensuring greater precision. Ray Casting technology complements this process, allowing damage identified by artificial intelligence to be mapped onto 3D models designed using technologies such as LiDAR or photogrammetry. This study developed a methodology capable of mapping corrosion, mechanical shocks and water accumulations in point clouds of industrial assets with sandwich panel cladding, using TLS and UAVs. The models trained using Detectron2 and YOLOv8 are able to segment the three types of damage with an accuracy of 68.8% and 72.5%.
The monitoring and inspection of industrial assets are essential to guarantee the safety of users, but they are time-consuming, costly and subject to human error. With technological advances, it is possible to carry out safer and more efficient inspections, minimizing work and risk for those involved. The application of AI algorithms in the segmentation of instances in images stands out in the detection of structural anomalies, as is the case with Mask R-CNN and YOLO. When combined with UAV platforms, these networks become powerful tools for inspecting assets in hard-to-reach areas, ensuring greater precision. Ray Casting technology complements this process, allowing damage identified by artificial intelligence to be mapped onto 3D models designed using technologies such as LiDAR or photogrammetry. This study developed a methodology capable of mapping corrosion, mechanical shocks and water accumulations in point clouds of industrial assets with sandwich panel cladding, using TLS and UAVs. The models trained using Detectron2 and YOLOv8 are able to segment the three types of damage with an accuracy of 68.8% and 72.5%.
Description
Financiado
Keywords
Monitoring LiDAR Photogrammetry Detectron2 YOLOv8 Ray casting Monitorização Fotogrametria