Proposta de automação da identificação de juntas soldadas em linhas de produção de tubos e perfis com visão computacional e Raspberry Pi/OpenCV

Abstract

This work presents a system for inspecting welded joints in industrial production lines of tubes and profiles. The system is based on computer vision techniques and is capable of detecting non-conforming products in real-time. The results showed that out of 300 tests performed, the system had an error rate of 4.67%, indicating that it can identify non-conforming products in most cases. However, there were 14 type-two errors, demonstrating that there is still room for improvement in reducing these errors. Furthermore, the data also showed that most errors occurred in galvanized material, followed by cold and hot-refined material. Despite achieving an accuracy rate of 95.33%, the system still has a vulnerability, mainly in lighter-toned materials. This is due to the higher reflectivity of the material, which can cause difficulties in recognition, leading to a test failure. The system's potential for use in difficult-to-reach environments, which can increase operational staff safety, is also highlighted. Future research can investigate non-dye-based detection methods, which can reduce operating costs, as the dyeing process involves a cost in time and money. Therefore, the presented system is a promising solution for detecting non-conforming products in industrial production lines, and the study opens up several avenues for future research.