基于砂铸生产大数据的铸件缺陷质量分析  

Quality Analysis of Data-driven Defects Based on Sand Casting Production Data

作　　者：官邦 汪东红 疏达 计效园[2] 周建新[2] 朱守琴 GUAN Bang;WANG Donghong;SHU Da;JI Xiaoyuan;ZHOU Jianxin;ZHU Shouqin(Shanghai Key Laboratory of Advanced High-Temperature Materials and Precision Forming,School of Materials Science and Engineering,Shanghai Jiaotong University,Shanghai 200240;State Key Laboratory of Materials Processing and Die&Mould Technology,Huazhong University of Science and Technology,Wuhan 430074;Hefei Casting and Forging Factory of Anhui Heli Co.,Ltd.,Hefei 230061)

机构地区：[1]上海交通大学材料科学与工程学院,上海市先进高温材料及其精密成形重点实验室,上海200240 [2]华中科技大学材料成形与模具技术全国重点实验室,武汉430074 [3]安徽合力股份有限公司合肥铸锻厂,合肥230061

出　　处：《特种铸造及有色合金》2024年第11期1471-1478,共8页Special Casting & Nonferrous Alloys

基　　金：国家重点研发计划资助项目(2020YFB1710100,2022YFB3706800);国家科技重大专项资助项目(J2019-Ⅵ-0004-0117);国家自然科学基金资助项目(52090042);长寿命高温材料国家重点实验室开放基金资助项目(DECSKL202109)。

摘　　要：针对砂型铸造工艺复杂,各工艺参数相互影响,铸件质量难以控制,废品率高的问题,提出基于大数据模型减少铸件缺陷和提高生产效率的分析方法。采集工厂的数据包括4种缺陷类型及其相应的工艺参数,进行数据预处理,将高维数据降至三维后,通过三维散点图可视化直观地展示数据的分布、聚类、趋势等信息,建立随机森林(RF)分类模型。结果表明,所有缺陷类别的召回率都在90%以上,通过模型的基尼不纯度分析得到不同工艺参数对缺陷的影响程度。最后结合蒙特卡罗(MC)仿真法进行试验模拟,预测出较优的工艺参数分布。Aiming at complex sand-casting process combined with the interactions between process parameters and difficulty in controling casting quality as well as high scrap rate,a data-driven model was proposed to reduce casting defects and improve production efficiency.The collected data were utilized for data preprocessing,including four types of defects and corresponding process parameters.The high-dimensional data was reduced to 3D,and the distribution,clustering,trend and other information were visually displayed through 3D scatter diagram.The random forest(RF)classification model was established,and the results indicate that the recall rates of all categories reach above 90%.The influence degree of different process parameters on defects was obtained by Gini impurity analysis of the model.Finally,Monte Carlo(MC)simulation method was employed to predict the optimal distribution of process parameters.

关 键 词：砂型铸造 数据驱动 分类模型 质量预测 特征重要性 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TG249[自动化与计算机技术—计算机科学与技术]