热轧带钢表面缺陷自动判定系统及其应用  

Automatic judgment system for surface defects of hot rolled strip and its application

作　　者：付光[1] 焦会立[1] 吴耐[1] 吴强 彭纯琪 吴冰[2] FU Guang;JIAO Huili;WU Nai;WU Qiang;PENG Chunqi;WU Bing(Beijing Shougang Co.,Ltd.,Tangshan 064400,China;Beijing Shougang Automation Information Technology Co.,Ltd.,Beijing 100043,China)

机构地区：[1]北京首钢股份有限公司,河北唐山064404 [2]北京首钢自动化信息技术有限公司,北京100043

出　　处：《轧钢》2023年第3期97-102,共6页Steel Rolling

摘　　要：带钢表面缺陷形成机理复杂、发生频次高,对成品质量的影响大,是最重要、最难控制的质量指标之一。针对当前卷积神经网络模型存在系统消耗大、处理时间长、无有效特征输出,以及热轧带钢表面缺陷数据量庞大、伪缺陷众多,不能及时、准确地判断其表面缺陷的问题,基于卷积神经网络深度学习技术,开发了一种带钢表面缺陷自动判定系统。介绍了该系统中在线采集模块、多通道结合分析模块、典型特征提取模块、缺陷严重性细化分类模块、缺陷自动评审模块的功能,缺陷分类准确率约90%,可以实现热轧带钢表面缺陷的快速、准确分类及自动判定。The formation mechanism of surface defects of strip is complex,its occurrence frequency is high,and its impact on the quality of finished products is great.It is one of the most important and difficult index to control quality indicators.Aiming at the problems of large system consumption,long processing time and no effective feature output in the current convolution neural network model,as well as the large amount of data on the surface defects of hot rolled strip and the large number of false defects,which can not judge the surface defects timely and accurately,an automatic judgment system for the surface defects of strip was developed based on the deep learning technology of convolution neural network.The functions of online acquisition module,multi-channel combined analysis module,typical feature extraction module,defect severity refinement and classification module,and defect automatic evaluation module in the system were introduced.The accuracy rate of defect classification was about 90%,which could realize rapid and accurate classification and automatic determination of surface defects of hot rolled strip.

关 键 词：热轧带钢 表面缺陷 卷积神经网络 深度学习 缺陷分类 自动判定 

分 类 号：TG335.56[金属学及工艺—金属压力加工]