In metalworking industries dedicated to the manufacture of parts, a large number of welds are required, yet not all industries employ advanced technologies for detecting welding defects. As a result, quality control is often performed manually by workers, leading to longer processing times and a higher likelihood of misidentification of defects due to human error. This introduces additional costs to the manufacturing process. This article presents the implementation of a welding defect detection system for metalworking parts using Digital Image Processing (DIP) techniques combined with deep learning. The proposed system utilizes Convolutional Neural Networks (CNNs) trained to identify defects in analyzed metal parts, such as porosities, holes, cracks, bubbles, among others. Additionally, the system integrates a user interface designed to display detected defects in real time and alert supervisors, enabling timely decision-making in production. Finally, this research includes a cost-benefit analysis comparing the proposed system to the traditional method, with the aim of facilitating future real-world testing. The results demonstrate that this technology reduces production times and costs in metalworking welding plants.
A Thu, study studied this question.