摘要
钢铁微观组织检测分析人工依赖程度大,随着钢铁行业的快速发展,组织分析检测任务量逐年增加,通常完成一项晶粒度评级任务耗时1 h以上。数字智能化赋能材料检测创新成为发展必然趋势,人工智能技术在钢铁材料检测分析领域的应用,为满足材料基因和高通量实验发展需求、推动微观组织分析智能化带来新的机遇。介绍了一种基于深度学习的铁素体晶粒度评级方法。首先对图像进行预处理,利用U-Net网络对组织图像进行晶界提取,采用一种基于Zhang的快速并行细化算法的目标提取方法,获得无毛刺的铁素体晶界骨架图像,通过阈值截点识别方法识别截线与晶界的交点类型,从而确定截点数,自动计算晶粒度级别数。实现晶粒度自动评级一键化,大大提高了晶粒度评级的效率,有效节约了人员成本和时间成本。经对比,该评级方法识别精度完全满足±0.25级的标准要求,为孪晶、混晶智能评级做出了有益实践。
With the rapid development of the steel industry,the number of tissue analysis and detection tasks is increasing year by year,and it usually takes more than 1 hour to complete a grain size grading task.The application of artificial intelligence technology in the field of steel material detection and analysis has brought new opportunities to meet the development needs of material genes and high-throughput experiments and promote the intelligence of microstructure analysis.In this paper,a ferrite grain size rating method based on deep learning is introduced.Firstly,the image was preprocessed,the U-Net network was used to extract the grain boundaries of the tissue image,and a target extraction method based on Zhang's fast parallel refinement algorithm was used to obtain the burr-free ferrite grain boundary skeleton image,and the intersection type of the intercept line and the grain boundary was identified by the threshold intercept recognition method,so as to determine the number of intercept points and automatically calculate the number of grain size levels.The one-click automatic grain size rating is realized,which greatly improves the efficiency of grain size rating and effectively saves personnel costs and time costs.After comparison,the recognition accuracy of the rating method fully meets the standard requirements of±0.25,which makes a useful practice for the intelligent rating of twin and mixed crystals.
作者
宋月
安治国
白丽娟
谷秀锐
严文谨
刘丽君
Song Yue;An Zhiguo;Bai Lijuan;Gu Xiurui;Yan Wenjin;Liu Lijun(HBIS Materials Technology Research Institute,Shijiazhuang 050023,Hebei)
出处
《河北冶金》
2023年第11期73-77,共5页
Hebei Metallurgy
