A carregar...
Publicação
Automatic detection of defects in ceramic parts
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 6.18 MB | Adobe PDF |
Autores
Resumo(s)
Quality inspection is an important aspect of modern industrial manufacturing. In the context of industrial product surfaces the main objective is to automatically detect defects. Often small defects are difficult to detect without a careful human inspection. Therefore, automatic surface defect detection has had a sharp growth based on computer vision and image processing techniques. This process if accurate, presents several advantages to industrial companies, such as an improvement in the products’ quality due to a more accurate quality control process, as well as production cost savings. There are already several studies and implementations on different types of surfaces, such as metallic, concrete and wood surfaces, however, there are few for other kinds of more specific surfaces like ceramic. This dissertation describes and details a study and a complete solution for defect detection on ceramic pieces in an industrial environment using a computer vision system. The system developed was tested in an important Portuguese porcelain company named Grestel S.A. We structured and specified a complete architecture solution from sensors and hardware to machine learning algorithms.
The implemented system includes an industrial camera and hardware for image acquisition, a developed software platform that allows employees to perform image labeling to be used in the machine learning process, an image pre-processing process and finally a machine learning algorithm based on convolutional neural networks (CNNs). The implemented CNNs perform defect detection and classification in real-time using images acquired in the production line. To do a detailed study, the research process used several additional techniques such as data augmentation, transfer learning and fine-tuning. The collaboration between the Polytechnic of Leiria and GRESTEL S.A. resulted in a system for defect detection in ceramic surfaces, which was implemented and tested in the factory facilities with promising results. Further investigation is being carried out to make improvements and continue the research towards a more accurate and faster solution.
Descrição
Palavras-chave
Defect detection Deep learning: CNN Industrial surface Automatic surface inspection Quality inspection Computer vision