Publicação
Real-time quality control of heat sealed bottles
| Resumo: | The present document describes a system for controlling the quality of heat sealed bottles. The system detects defective seals to identify bottles that can not be sold. A prototype was developed to validate and test the system proposed. In the production line, the bottles are filled with a toxic substance and can only be sold when properly sealed. A leak can be harmful to humans and the environment. Because the seals are not visible from outside the bottle, images from each seal are obtained using a thermal camera. The hot glue used in the sealing process makes the seal visible in the infrared image. The image is cleaned and converted to black and white only keeping the seal in the final image. Black pixels present the value 0 and white pixels present the value 1. Then a signature composed by two arrays containing the sum of the number of white pixels in each column and in each row is calculated. Both arrays present a U shape when the bottle is sealed. The signature is then fed to an artificial neural network which was trained to identify correctly sealed bottles. The classification results are stored in a database. The trained neural net presented an accuracy of 98.7 % and an F1 score of 96.0 % in the testing phase. The results shows the inspection process is effective in identifying defective seals and because it is automated it can be scaled up to large bottle processing plants. All classified images can be seen though a web application where a user has the option of validating the operation and identifying errors which will be individually fitted to improve the machine learning model performance. The system is non invasive, automated, and can be applied to common conveyor belts currently used in industrial plants. It can also be adapted to detect different prob lems in bottles of different shapes. |
|---|---|
| Autores principais: | Cruz, Samuel Silva da |
| Assunto: | Quality-control Machine learning Computer vision Thermal images Artificial neural networks |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Instituto Politécnico de Coimbra |
| Idioma: | inglês |
| Origem: | Instituto Politécnico de Coimbra |
| Resumo: | The present document describes a system for controlling the quality of heat sealed bottles. The system detects defective seals to identify bottles that can not be sold. A prototype was developed to validate and test the system proposed. In the production line, the bottles are filled with a toxic substance and can only be sold when properly sealed. A leak can be harmful to humans and the environment. Because the seals are not visible from outside the bottle, images from each seal are obtained using a thermal camera. The hot glue used in the sealing process makes the seal visible in the infrared image. The image is cleaned and converted to black and white only keeping the seal in the final image. Black pixels present the value 0 and white pixels present the value 1. Then a signature composed by two arrays containing the sum of the number of white pixels in each column and in each row is calculated. Both arrays present a U shape when the bottle is sealed. The signature is then fed to an artificial neural network which was trained to identify correctly sealed bottles. The classification results are stored in a database. The trained neural net presented an accuracy of 98.7 % and an F1 score of 96.0 % in the testing phase. The results shows the inspection process is effective in identifying defective seals and because it is automated it can be scaled up to large bottle processing plants. All classified images can be seen though a web application where a user has the option of validating the operation and identifying errors which will be individually fitted to improve the machine learning model performance. The system is non invasive, automated, and can be applied to common conveyor belts currently used in industrial plants. It can also be adapted to detect different prob lems in bottles of different shapes. |
|---|