燕山大学学报2026，Vol.50Issue(2)：156-168,188,14.DOI:10.3969/j.issn.1007-791X.2026.02.007
结合区域距离和循环密度聚类的钢板缺陷归集方法
Method for collecting steel plate defects based on improved DBSCAN algorithm and region distance
摘要
Abstract
Defects such as dents and scratches frequently occur on the surface of steel plates produced in steel mills due to factors like process and equipment control precision.Manual grinding is typically required to address such defects in actual production operations.In recent years,automated grinding has emerged as a predominant trend to enhance efficiency.However,defects on steel plate surfaces are often characterized by uneven distribution and varied shapes and sizes.In order to effectively collect defect dense areas and improve subsequent grinding efficiency,a new method called RD(Region Distance)+DB-DBSCAN(Double Density-Based Spatial Clustering of Applications with Noise)is introduced in this paper.RD is combined with the DB-DBSCAN algorithm in this approach to effectively clustering surface defects.RD is introduced to resolve the limitation of traditional clustering methods in handling data points with area attributes.The DB-DBSCAN conducts initial clustering using DBSCAN,partitioning the data into density-differentiated partitions.Subsequently,Eps values for each partition were determined through K-distance graphs,followed by local clustering conducted within these regions.Noise points were merged into relevant clusters during the process.The limitations of the traditional DBSCAN algorithm are addressed by the method,enhancing clustering performance in scenarios with uneven data density.Experiments conducted on typical UCI datasets and synthetic datasets demonstrated that the DB-DBSCAN algorithm achieved superior NMI and ARI scores compared to DBSCAN and its extended clustering algorithms across most datasets.Furthermore,experiments performed on steel mill datasets revealed that the RD+DB-DBSCAN method proposed in this study exhibited exceptional clustering performance for handling complex and highly concentrated defect distributions.关键词
钢板缺陷/聚类算法/区域距离/循环密度聚类Key words
steel plate defects/clustering algorithm/region distance/DB-DBSCAN分类
信息技术与安全科学引用本文复制引用
李晨,杨明永,于露,冯昊吉,刘天歌,何海涛..结合区域距离和循环密度聚类的钢板缺陷归集方法[J].燕山大学学报,2026,50(2):156-168,188,14.基金项目
太原市关键核心技术攻关"揭榜挂帅"项目(2024TYJB0104) （2024TYJB0104）
河北省自然科学基金资助项目(F2023203030) （F2023203030）
河北省教育厅科学研究项目(QN2024010) （QN2024010）
