Loading...
Publication
Validação de sistemas de inspeção automática por visão artificial e implementação de processos de monitorização em linhas de engarrafamento
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.41 MB | Adobe PDF |
Authors
Abstract(s)
A indústria vinícola enfrenta, atualmente, uma crescente competitividade, o que exige a implementação de processos que assegurem a qualidade e segurança alimentar dos produtos e dos processos envolvidos, em simultâneo com um aumento da eficiência produtiva. A presente dissertação teve como principal objetivo a validação de sistemas de inspeção automática por visão artificial instalados em linhas de produção da Symington Family Estates, Vinhos, S.A., como alternativa ao método tradicional de inspeção manual, sujeito a erros humanos. Adicionalmente, desenvolveram-se metodologias de monitorização para avaliar o desempenho dos sistemas de inspeção ao longo do tempo. Foram avaliados quatro sistemas de inspeção automática por visão artificial para deteção de defeitos em garrafas vazias, controlo do nível de enchimento, de verificação do produto final (presença de cápsula e rótulo) e de deteção de progressões capilares de vinho nas rolhas de cortiça após estágio em garrafeira. Com base nos resultados obtidos, foi possível construir a matriz de confusão para cada ensaio, permitindo calcular métricas de desempenho como a exatidão, erro, precisão, sensibilidade, especificidade e taxa de falso alarme. Estas métricas possibilitaram uma análise objetiva e comparativa da performance dos sistemas de inspeção em diferentes condições operacionais. Os resultados demonstraram que o sistema de inspeção de produto final apresentou o melhor desempenho, com exatidão superior a 99% e taxa de falsos negativos nula, evidenciando elevada fiabilidade na deteção de não conformidades. O sistema de nível de enchimento também
revelou elevada eficácia, com exatidão de 94,4%, precisão de 93,3% e sensibilidade de 96,2%. Por outro lado, o sistema de inspeção de garrafas vazias mostrou desempenho variável consoante o modelo de garrafa, sendo o modelo de garrafa A6 o que apresentou resultados significativamente inferiores, com exatidão de 78,3%.
A avaliação das não conformidades permitiu identificar os tipos de defeitos mais difíceis de detetar, como os de pequena dimensão e baixo contraste visual, que contribuíram para o aumento de falsos negativos. Já o sistema de progressões capilares revelou maior variabilidade nos resultados, mesmo após intervenção do fabricante, não atingindo os níveis de desempenho esperados. A análise estatística entre ensaios revelou diferenças significativas em métricas como precisão, sensibilidade e taxa de falso alarme, evidenciando a necessidade de nova intervenção e estudo das características estruturais das garrafas que influenciam o desempenho. A implementação de sistemas de inspeção automática por visão artificial contribui para a fiabilidade do processo, para a conformidade legal e expectativas do consumidor. No entanto, a sua eficácia depende de uma validação cuidadosa e de uma monitorização sistemática, face à complexidade técnica e à variabilidade das condições operacionais. Estes processos são essenciais para garantir que os sistemas mantêm um desempenho consistente ao longo do tempo, mesmo perante alterações nos produtos inspecionados ou nas condições de produção.
The wine industry is currently facing increasing competition, which requires the implementation of processes that ensure the quality and food safety of products and processes involved, while simultaneously increasing production efficiency. The main objective of this dissertation was to validate automatic inspection systems using artificial vision installed on production lines at Symington Family Estates, Vinhos, S.A., as an alternative to the traditional method of manual inspection, which is subject to human error. In addition, monitoring methodologies were developed to evaluate the performance of the inspection systems over time. Four automatic inspection systems using artificial vision were evaluated for detecting defects in empty bottles, controlling the filling level, checking the final product (presence of capsule and label), and detecting capillary progression of wine in cork stoppers after aging in the cellar. Based on the results obtained, it was possible to construct a confusion matrix for each test, allowing the calculation of performance metrics such as accuracy, error, precision, sensitivity, specificity, and false alarm rate. These metrics enabled an objective and comparative analysis of the inspection system’s performance under different operating conditions. The results showed that the final product inspection system performed best, with an accuracy of over 99% and a false negative rate of zero, demonstrating high reliability in detecting nonconformities. The fill level system also proved highly effective, with an accuracy of 94.4%, precision of 93.3%, and sensitivity of 96.2%. On the other hand, the empty bottle inspection system showed variable performance depending on the bottle model, with the A6 bottle model showing significantly lower results, with an accuracy of 78.3%. The evaluation of non-conformities identified the types of defects that were most difficult to detect, such as those that were small and had low visual contrast, which contributed to an increase in false negatives. The capillary progression system showed greater variability in results, even after intervention by the manufacturer, failing to achieve the expected performance levels. Statistical analysis between tests revealed significant differences in metrics such as accuracy, sensitivity, and false alarm rate, highlighting the need for further intervention and study of the structural characteristics of bottles that influence performance. The implementation of automatic inspection systems using artificial vision contributes to the reliability of the process, legal compliance, and consumer expectations. However, their effectiveness depends on careful validation and systematic monitoring, given the technical complexity and variability of operating conditions. These processes are essential to ensure that systems maintain consistent performance over time, even in the face of changes in the products inspected or production conditions.
The wine industry is currently facing increasing competition, which requires the implementation of processes that ensure the quality and food safety of products and processes involved, while simultaneously increasing production efficiency. The main objective of this dissertation was to validate automatic inspection systems using artificial vision installed on production lines at Symington Family Estates, Vinhos, S.A., as an alternative to the traditional method of manual inspection, which is subject to human error. In addition, monitoring methodologies were developed to evaluate the performance of the inspection systems over time. Four automatic inspection systems using artificial vision were evaluated for detecting defects in empty bottles, controlling the filling level, checking the final product (presence of capsule and label), and detecting capillary progression of wine in cork stoppers after aging in the cellar. Based on the results obtained, it was possible to construct a confusion matrix for each test, allowing the calculation of performance metrics such as accuracy, error, precision, sensitivity, specificity, and false alarm rate. These metrics enabled an objective and comparative analysis of the inspection system’s performance under different operating conditions. The results showed that the final product inspection system performed best, with an accuracy of over 99% and a false negative rate of zero, demonstrating high reliability in detecting nonconformities. The fill level system also proved highly effective, with an accuracy of 94.4%, precision of 93.3%, and sensitivity of 96.2%. On the other hand, the empty bottle inspection system showed variable performance depending on the bottle model, with the A6 bottle model showing significantly lower results, with an accuracy of 78.3%. The evaluation of non-conformities identified the types of defects that were most difficult to detect, such as those that were small and had low visual contrast, which contributed to an increase in false negatives. The capillary progression system showed greater variability in results, even after intervention by the manufacturer, failing to achieve the expected performance levels. Statistical analysis between tests revealed significant differences in metrics such as accuracy, sensitivity, and false alarm rate, highlighting the need for further intervention and study of the structural characteristics of bottles that influence performance. The implementation of automatic inspection systems using artificial vision contributes to the reliability of the process, legal compliance, and consumer expectations. However, their effectiveness depends on careful validation and systematic monitoring, given the technical complexity and variability of operating conditions. These processes are essential to ensure that systems maintain consistent performance over time, even in the face of changes in the products inspected or production conditions.
Description
Keywords
Confusion Matrix Non-conformities Performance metrics Capillary Progression Matriz de confusão Não conformidade Métricas de desempenho Progressões capilares