Classificação de defeitos superficiais em placas de aço laminadas a quente com assistência de uma rede neural

Abstract

This work addresses the use of neural networks for the detection of surface defects in hot-rolled steel plates. Initially, the importance of defect detection in the industry is discussed, highlighting its impacts on product quality and structural safety. Next, the basic concepts of neural networks are presented, including their structure, functioning, and learning from data. The main types of neural networks used in this context are mentioned, with a focus on convolutional neural networks (CNNs), which are suitable for processing image data. The methodology employed in the work is detailed, covering data preprocessing, selection of relevant features, and model training. The use of transfer learning is highlighted, which involves leveraging prior knowledge from neural networks trained on large generic datasets. Previous studies that used neural networks for defect detection in steel plates are presented, demonstrating the effectiveness of this approach. Additionally, the main advancements, challenges, and limitations in this research area are discussed. Finally, the commonly encountered types of surface defects in hot-rolled steel plates, such as crazing, inclusion, patches, pitted, rolled, and scratches, are addressed. Previous research related to these types of defects is also mentioned. Overall, the work provides a comprehensive overview of the detection of surface defects in steel plates using neural networks, discussing everything from basic concepts to practical application and the challenges faced in this field of study.