U-net Architecture-Based Image Detection Model Development for Defect Detection in Hot Rolling Mill

As global steel demand increases after the pandemic, competition in the steel industry has intensified, and it is necessary to secure product competitiveness. This study aims to product quality deviation due to strip waves generated in the hot rolling process of steel plants by applying deep learning-based image detection technology. A high-definition camera was installed within the rolling mill to collect strip wave image data. The collected data were preprocessed using the concatenation technique to partition the high-performance images. This study developed an image detection model using the U-net architecture, a deep learning based Fully Convolutional Network (FCN) algorithm proposed for image segmentation. As a result of testing the detection accuracy, the developed model achieved a 97.91% of detection rate. The deep learning-based image detection model developed in this study can potentially enhance the quality and productivity of steel products.