基于MCMC算法的粗集料粒径分布推断研究  

Coarse Aggregate Particle Size Distribution of Inference Based on MCMC Algorithm Research

作　　者：童欣 陆艺[1] 李静伟 范伟军[1] TONG Xin;LU Yi;LI Jingwei;FAN Weijun(College of Metrology Measurement and Instrument,China Jiliang University,Hangzhou,Zhejiang 310018,China;Hangzhou Wolei Intelligent Technology Co.Ltd,Hangzhou,Zhejiang 310018,China)

机构地区：[1]中国计量大学计量测试与仪器学院,浙江杭州310018 [2]杭州沃镭智能科技股份有限公司,浙江杭州310018

出　　处：《计量学报》2024年第9期1273-1280,共8页Acta Metrologica Sinica

基　　金：国家自然科学基金(52075511)。

摘　　要：运用机器视觉对振动盘掉落的粗集料进行数据采集,得到了动态集料的‘伪三维’图像。针对图像信息不能准确表达集料真实三维情况,引入贝叶斯统计推断的思想对集料进行粒径分布的推断,选取等效Feret短径作为图像特征,由于集料在大粒径情况下其Feret短径与实际粒径误差较大,故增加了等效椭圆短径作为第二特征。为获得准确的后验分布和高效的工程计算能力,采用了MCMC算法,解决了传统贝叶斯统计推断在高维计算不足的问题,从而得到了集料粒径分布的后验分布。实验结果表明:该方法的针对合格集料的粒径分布测量误差保持在±2.5%以内,不合格集料误差保持在±3.5%以内。Machine vision is used to collect data on the coarse aggregate dropped from the vibrating plate,and a‘pseudo-three-dimensional’image of the dynamic aggregate is obtained.Since the image information cannot accurately express the aggregate.Therefore,the idea of Bayesian statistical inference is introduced to infer the particle size distribution of aggregates.The equivalent Feret short diameter is selected as the image feature,but the error between the Feret short diameter and the actual particle size of the aggregate is large when the particle size is large,so the equivalent elliptical short diameter is added as the second feature.In order to obtain accurate posterior distribution and efficient engineering computing capabilities,the Markov-Monte Carlo(MCMC)algorithm is used,thus breaking through the problem of insufficient high-dimensional calculations of traditional Bayesian statistical inference,and thus obtaining the aggregate Posterior distribution of particle size distribution.Experimental results show that the particle size distribution measurement error of this method for qualified aggregates is maintained within±2.5%,and the error for unqualified aggregates is maintained within±3.5%.

关 键 词：几何量计量 粗集料 粒径分布 视觉跟踪 贝叶斯推断 MCMC算法 级配检测 

分 类 号：TB92[一般工业技术—计量学]