A carregar...
Publicação
Deteção de Partículas Contaminantes Presentes em Amostras de Óleo Lubrificante
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 32.09 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Os lubrificantes industriais desempenham um papel essencial no bom funcionamento e
longevidade das máquinas. No entanto, a sua eficácia pode ser gravemente comprometida pela contaminação com detritos metálicos resultantes do desgaste de componentes
das máquinas. Os métodos atuais de deteção são demorados e dispendiosos, podendo
atrasar decisões de manutenção e aumentar riscos operacionais.
Esta dissertação de mestrado, desenvolvida no âmbito do projeto MicroLubProbe da
GALP, propõe uma metodologia automatizada para deteção e medição de partículas metálicas contaminantes presentes em imagens de microscopia de óleos e massas lubrificantes, recorrendo a técnicas de processamento digital de imagem. O algoritmo integra
etapas de conversão de espaços de cor, aumento de contraste, redução de ruído e segmentação, implementados em Python com a biblioteca Open Source Computer Vision Library (OpenCV). Foram testadas várias técnicas de filtragem e realce avaliadas com mais
do que uma métrica quantitativa.
O desenvolvimento inicial do algoritmo foi realizado com imagens obtidas em laboratório, uma vez que na fase inicial ainda não estavam disponíveis imagens recolhidas no
terreno. Nessas condições controladas, e em imagens capturadas com foco adequado e
iluminação homogénea, o algoritmo apresentou bom desempenho, identificando e quantificando partículas com precisão e consistência. No entanto, quando aplicado posteriormente às imagens de terreno, surgiram limitações significativas, o desempenho foi afetado por grandes diferenças no tipo de dados e nas condições de aquisição. Perante estes
resultados, foi testada uma nova abordagem mas, dado que o conjunto de imagens disponível para esta fase foi extremamente reduzido (apenas duas imagens), não permitiu
validar o método nem confirmar o seu potencial para integração em sistemas de monitorização em tempo real. São necessários mais dados de campo e ajustes de calibração para
garantir a fiabilidade da solução fora do ambiente laboratorial.
Industrial lubricants play a key role in the proper functioning and longevity of machines. However, their effectiveness can be severely compromised by contamination with metallic debris resulting from the wear of machine components. Current detection methods are time-consuming and expensive, which may delay maintenance decisions and increase operational risks. This Master’s dissertation, developed within the scope of the MicroLubProbe project at GALP, proposes an automated methodology for the detection and measurement of metallic contaminant particles in microscopy images of oils and greases, using digital image processing techniques. The algorithm integrates steps of colour space conversion, contrast enhancement, noise reduction, and segmentation, implemented in Python with the OpenCV library. Several filtering and enhancement techniques were tested and evaluated using more than one quantitative metric. The initial development of the algorithm was carried out using laboratory-acquired images, since at that stage field images were not yet available. Under these controlled conditions, and with images captured with proper focus and homogeneous lighting, the algorithm performed well, accurately and consistently identifying and quantifying particles. However, when later applied to field images, significant limitations emerged, as performance was affected by substantial differences in data type and acquisition conditions. In light of these results, a new approach was tested; however, the dataset available for this phase was extremely limited (only two images), preventing the validation of the method or confirmation of its potential for integration into real-time monitoring systems. Additional field data and calibration adjustments are required to ensure the reliability of the solution outside the laboratory environment.
Industrial lubricants play a key role in the proper functioning and longevity of machines. However, their effectiveness can be severely compromised by contamination with metallic debris resulting from the wear of machine components. Current detection methods are time-consuming and expensive, which may delay maintenance decisions and increase operational risks. This Master’s dissertation, developed within the scope of the MicroLubProbe project at GALP, proposes an automated methodology for the detection and measurement of metallic contaminant particles in microscopy images of oils and greases, using digital image processing techniques. The algorithm integrates steps of colour space conversion, contrast enhancement, noise reduction, and segmentation, implemented in Python with the OpenCV library. Several filtering and enhancement techniques were tested and evaluated using more than one quantitative metric. The initial development of the algorithm was carried out using laboratory-acquired images, since at that stage field images were not yet available. Under these controlled conditions, and with images captured with proper focus and homogeneous lighting, the algorithm performed well, accurately and consistently identifying and quantifying particles. However, when later applied to field images, significant limitations emerged, as performance was affected by substantial differences in data type and acquisition conditions. In light of these results, a new approach was tested; however, the dataset available for this phase was extremely limited (only two images), preventing the validation of the method or confirmation of its potential for integration into real-time monitoring systems. Additional field data and calibration adjustments are required to ensure the reliability of the solution outside the laboratory environment.
Descrição
Palavras-chave
Deteção de Partículas Processamento de Imagens Lubrificantes Suavização Operações Morfológicas Diâmetro Maior