| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 47.59 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A nĆvel industrial, o processo de controlo de qualidade de produtos Ć© realizado com diferentes tĆ©cnicas e especificaƧƵes, adequadas a cada processo. No entanto, por vezes, as tĆ©cnicas utilizadas tĆŖm falhas que se traduzem num prejuĆzo para as empresas. Os avanƧos tecnolĆ³gicos seguem num ritmo acelerado e o mercado tem vindo a absorver as inovaƧƵes nos seus mais diversos setores. Numa procura pela transformaĆ§Ć£o digital, as empresas passaram a investir mais em soluƧƵes que gerem um diferencial competitivo frente Ć concorrĆŖncia. Nesse sentido, o conceito de InteligĆŖncia Artificial (IA) ganhou bastante importĆ¢ncia. Nos Ćŗltimos anos temos assistido a uma adoĆ§Ć£o acelerada do Machine Learning (ML) como parte integrante da IndĆŗstria 4.0, na qual a digitalizaĆ§Ć£o estĆ” refazendo a indĆŗstria. Essa Ćŗltima onda de iniciativas Ć© marcada pela introduĆ§Ć£o de sistemas inteligentes e autĆ³nomos, alimentados por grandes quantidades de dados e por Deep Learning (DL). Uma poderosa geraĆ§Ć£o de IA que promove a inspeĆ§Ć£o de qualidade no chĆ£o de fĆ”brica. Este trabalho visa investigar e implementar tĆ©cnicas supervisionadas de Deep Learning, aliadas Ć visĆ£o computacional, para a implementaĆ§Ć£o de um sistema de classificaĆ§Ć£o automĆ”tico de imperfeiƧƵes de Capply. Para esse efeito, inicialmente, foi realizada uma revisĆ£o bibliogrĆ”fica sobre o estado da arte, passando de seguida Ć implementaĆ§Ć£o e comparaĆ§Ć£o do desempenho de vĆ”rias arquiteturas utilizando as mĆ©tricas adequadas. Para a execuĆ§Ć£o desta tarefa foi necessĆ”rio recolher e fazer um prĆ©-processamento dos dados (imagens de bobines de capply). Foi ainda, desenvolvida uma aplicaĆ§Ć£o Web que permite testar e avaliar os resultados e por Ćŗltimo, foi tambĆ©m desenvolvido e implementado um sistema de classificaĆ§Ć£o em contexto real. Resumidamente, os resultados deste trabalho demonstraram o grande potencial das metodologias de Deep Learning aplicadas ao controlo de qualidade na indĆŗstria.
At an industrial level, the product quality control process is carried out using different techniques and specifications, suitable for each process. However, sometimes the techniques used have flaws that translate into a loss for companies. Technological advances continue at an accelerated pace and the market has been absorbing innovations in its most diverse sectors. In a search for digital transformation, companies began to invest more in solutions that generate a competitive edge against the competition. In this sense, the concept of Artificial Intelligence (AI) has gained a lot of importance. In recent years we have seen an accelerated adoption of Machine Learning (ML) as an integral part of Industry 4.0, in which digitalization is remaking the industry. This latest wave of initiatives is marked by the introduction of intelligent and autonomous systems, powered by large amounts of data and by Deep Learning (DL). A powerful generation of AI that drives quality inspection on the shop floor. This work aims to investigate and implement supervised Machine Learning techniques, combined with computer vision, for the implementation of an automatic classification system for Capply imperfections. For this purpose, initially, a bibliographic review was carried out on the state of the art, followed by the implementation and comparison of the performance of several architectures using the appropriate metrics. To perform this task, it was necessary to collect and preprocess the data (capply reel images). A web application was also developed that allows testing and evaluating the results and finally, a classification system in real context was also developed and implemented. Briefly, the results of this work demonstrated the great potential of Machine Learning methodologies applied to quality control in the industry.
At an industrial level, the product quality control process is carried out using different techniques and specifications, suitable for each process. However, sometimes the techniques used have flaws that translate into a loss for companies. Technological advances continue at an accelerated pace and the market has been absorbing innovations in its most diverse sectors. In a search for digital transformation, companies began to invest more in solutions that generate a competitive edge against the competition. In this sense, the concept of Artificial Intelligence (AI) has gained a lot of importance. In recent years we have seen an accelerated adoption of Machine Learning (ML) as an integral part of Industry 4.0, in which digitalization is remaking the industry. This latest wave of initiatives is marked by the introduction of intelligent and autonomous systems, powered by large amounts of data and by Deep Learning (DL). A powerful generation of AI that drives quality inspection on the shop floor. This work aims to investigate and implement supervised Machine Learning techniques, combined with computer vision, for the implementation of an automatic classification system for Capply imperfections. For this purpose, initially, a bibliographic review was carried out on the state of the art, followed by the implementation and comparison of the performance of several architectures using the appropriate metrics. To perform this task, it was necessary to collect and preprocess the data (capply reel images). A web application was also developed that allows testing and evaluating the results and finally, a classification system in real context was also developed and implemented. Briefly, the results of this work demonstrated the great potential of Machine Learning methodologies applied to quality control in the industry.
Description
Keywords
Controlo de Qualidade InteligĆŖncia Artificial Machine Learning Deep Learning VisĆ£o Computacional IndĆŗstria 4.0 ClassificaĆ§Ć£o CNN Quality Control Artificial Intelligence Computer Vision Industry 4.0 Classification