基于AI图像分析的铸造原砂粒度粒形检测法
AI-Based Image Analysis for Particle Size and Shape Detection in Casting Sand
李贝贝 1郭树人 1曹华堂 1刘鑫旺 1董选普 1张宇辉 2钟飞升 2吴执成2
作者信息
- 1. 华中科技大学材料科学与工程学院,材料成形与模具技术全国重点实验室,湖北武汉 430074
- 2. 辽宁联信铸砂投资集团有限公司,辽宁沈阳 110000
- 折叠
摘要
铸造原砂的粒形和粒度分布是原砂生产和使用过程中的重要衡量指标.针对筛分法存在的测量效率和测量误差问题,本文提出一种基于人工智能(Artificial Intelligence:AI)图像分析的铸造原砂粒度粒形测试方法,即采用工业相机采集原砂图像,通过AI图像处理技术对图像进行实例分割,然后对图像进行特征提取,统计出原砂的粒度分布和粒形分布.研究结果表明:基于AI技术的BlendMask实例分割模型能有效地分离粘连砂粒;采用圆形度、形状因子和方形度三个特征参数对原砂粒形进行K-means聚类,能够准确分析砂粒的粒形特征;采用等效椭圆法、面积占比等效质量占比法,可准确测得烘焙砂、烘干砂和宝珠砂三种铸造原砂的粒度分布,满足行业精度要求.
Abstract
The shape and particle size distribution of casting sand are important factorsin the production and utilization of the casting sand.To address the issues of measurement efficiency and measurement errors associated with sieving methods,this study proposed a casting sand particle size and shape testing method based on artificial intelligence(AI)image analysis,which involves capturing images of the casting sand using an industrial camera and employing AI image processing techniques for instance segmentation.Subsequently,the images were subjected to feature extraction to statistically determine the particle size and shape distribution of the casting sand.The research findings indicated that the AI-based BlendMask instance segmentation model,could effectively separate agglomerated sand particles.By utilizing three feature parameters-circularity,shape factor,and rectangularity-for K-means clustering of casting sand particles,the method analyzed the particle shape characteristics accurately.Furthermore,employing the equivalent ellipse method and the area proportion equivalent mass ratio method enabled precise measurement of the particle size distribution for three types of the casting sand:baked sand,dried sand,and zircon sand,respectively,which meets the industry accuracy requirements.
关键词
铸造原砂/粒度分布/粒形/图像法/实例分割/深度学习Key words
casting sand/particle size distribution/particle shape/image-based method/instance segmentation/deep learning引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据