基于Mask-RCNN的转炉炉口形貌检测  

Morphology detection of converter mouth based on Mask-RCNN

作　　者：戴张杰 黄成永 刘威[1] 夏建超[3] 杨树峰[1] 李京社[1] DAI Zhang-jie;HUANG Cheng-yong;LIU Wei;XIA Jian-chao;YANG Shu-feng;LI Jing-she(School of Metallurgical and Ecological Engineering,University of Science and Technology Beijing,Beijing 100083,China;Equipment Department,Shanghai Meishan Iron and Steel Co.,Ltd.,Nanjing 210039,Jiangsu,China;Steelmaking Plant,Shanghai Meishan Iron and Steel Co.,Ltd.,Nanjing 210039,Jiangsu,China)

机构地区：[1]北京科技大学冶金与生态工程学院,北京100083 [2]上海梅山钢铁股份有限公司设备部,江苏南京210039 [3]上海梅山钢铁股份有限公司炼钢厂,江苏南京210039

出　　处：《钢铁》2023年第3期73-78,共6页Iron and Steel

基　　金：国家自然科学基金资助项目(52104318,52074030);中央高校基本科研业务费资助项目(FRF-IDRY-20-007)。

摘　　要：随着炼钢工序自动化要求的进一步提高,实现过程自动化加料具有重要意义。由于转炉冶炼过程时常伴随溢渣现象,炉口处易形成结瘤,致使内径减小,从而影响下一炉次加料工序进行。目前现场主要通过人工对炉口形貌进行观测以判断是否需要进行修补炉口操作,这种传统人工检测效率低,并且由于炉内光强以及技术人员的主观判断等因素,致使检测结果不稳定。随着人工智能领域中深度学习方法的快速发展及并在各行业中发挥重大作用,提出了一种基于Mask-RCNN的转炉炉口形貌检测方法。该方法基于网络模型的输出结果,通过图像处理对炉口轮廓的周长与面积进行测算,并结合最小二乘圆拟合法(LSCM),通过圆度指标对形貌进行更加合理定量的表征。试验表明,随着冶炼的连续进行,炉口面积与周长都将持续减小,对应圆度误差值持续增加,这说明过程中炉口形貌由于黏渣现象发生而持续改变。自动检测方法能够针对加料或出渣时不同转动倾角的炉口进行实时稳定检测,其识别率高达99%,具有较高的检测性能,相较于人工检测具有稳定、准确、高效的优势,大幅度提升了工艺稳定性。With the further improvement of the automation requirements of the steelmaking process,it is of great significance to realize the automatic feeding of the process.Because the converter smelting is often accompanied by the phenomenon of slag overflow,the furnace mouth is easy to form nodules,resulting in the reduction of the inner diameter,which affects the next furnace feeding process.At present,the site mainly observes the morphology of the furnace mouth manually to determine whether it is necessary to repair the furnace mouth.This traditional manual detection is inefficient,and the detection results are unstable due to factors such as the light intensity in the furnace and the subjective judgment of the technicians.With the rapid development of deep learning methods in the field of artificial intelligence and its important role in various industries,a converter mouth shape detection method based on Mask-RCNN is proposed.Based on the output results of the network model,the method calculates the perimeter and area of the furnace mouth contour through image processing,and combines the least squares circle fitting meth-od(LSCM)to characterize the morphology more reasonably and quantitatively through the roundness index.The experiment shows that with the continuous smelting,the area and perimeter of the furnace mouth will continue to decrease,and the corresponding roundness error value will continue to increase,which indicates that the morphology of the furnace mouth will continue to change due to the phenomenon of slag adhesion.The automatic detection meth-od can perform real-time stable detection on the furnace mouth with different rotation angles when feeding or slag discharge.The recognition rate is as high as 99%,which has high detection performance.Compared with manual detection,it has the advantages of stability,accuracy and efficiency,and greatly improves the process stability.

关 键 词：转炉炉口形貌 Mask-RCNN 深度学习 图像处理 智能炼钢 

分 类 号：TF713[冶金工程—钢铁冶金]