Detection Methods of Surface Defects of Sintering Material Based on Machine Vision

Wensi Liu,Xiao-Yu Tang,Yi Yang,Liang Zhao,Chunjie Yang

DOI: https://doi.org/10.1109/cfasta57821.2023.10243366

2023-01-01

Abstract:The iron and steel industry is one of the pillar industries in the national economy. Blast furnace (BF) ironmaking is the core step in long-process ironmaking. Before BF production, to improve the ore quality, sintering is adopted to provide highquality man-made ore from raw materials. In the sintering process, the material surface state is the main indicator for monitoring the ignition effect. During the actual production process, due to the influence of the ingredients and the ignition effect, there are defects on the sintering material surface, including yellow mud, mouse holes, and cracks. Detection of defect information on the material surface is conducive to ensuring the stable quality of the sinter and the stable operation of sintering production. This paper designs the corresponding detection methods for the three defects based on machine vision algorithms, 1) Yellow mud detection: separate the material surface area of the sintering trolley by semantic segmentation, and judge whether there is yellow mud according to the HSV histogram correlation combined with the shape judgment condition of yellow mud; 2) Mouse hole detection: use the target detection method of YOLOv5 without Mosaic detects mouse holes and improves the detection ability of small targets; 3) Crack detection: transfer learning and improved DeepLabv3+ network semantic segmentation are used for crack detection to improve the accuracy of crack detection. Finally, all methods are applied to the actual production line, and the detection accuracy and effectiveness of the above methods are verified through image testing and calculation of evaluation indicators, which can meet the needs of on-site use.