SMOTE数据预处理算法在砂型铸造复杂铸件缺陷预测中的应用  

Application of SMOTE Data Preprocessing Algorithm in Defect Prediction of Complex Castings in Sand Casting

作　　者：潘徐政 刘迎辉 李文[1] 计效园[1] 殷亚军[1] 吴来发 解明国[2] 周建新[1] PAN Xu-zheng;LIU Ying-hui;LI Wen;JI Xiao-yuan;YIN Ya-jun;WU Lai-fa;XIE Ming-guo;ZHOU Jian-xin(State Key Laboratory of Material Forming and Die Technology,Huazhong University of Science and Technology,Wuhan 430074,Hubei China;Anhui Heli Co.,Ltd.,Hefei 230601,Anhui,China)

机构地区：[1]华中科技大学材料成形与模具技术国家重点实验室,湖北武汉430074 [2]安徽合力股份有限公司,安徽合肥230601

出　　处：《铸造》2024年第10期1473-1479,共7页Foundry

基　　金：国家重点研发计划项目(2020YFB1710100);国家自然科学基金(52275337、52090042、51905188)。

摘　　要：针对实际生产过程采集的复杂转向桥铸件工艺数据中冷隔、气孔、砂眼、缩孔等缺陷类别的数据量严重不平衡、复杂铸件缺陷预测模型准确率不高的问题,结合砂型铸造实际工况,引入了SMOTE(Synthetic Minority Oversampling Technique)数据预处理算法,探究其在砂型铸造复杂铸件缺陷预测中的应用。根据采集到的复杂铸件不平衡数据集的特点,基于SMOTE数据预处理算法,科学扩充了不平衡数据集,创建了可用于训练复杂铸件缺陷预测模型的平衡数据集,数据预处理前后的模型预测准确率从86.50%提高至97.91%。In view of the problems that the data amount of defects such as cold shut,blowhole,sand inclusion and shrinkage was seriously imbalance,and the accuracy of defect prediction model of complex casting was not high in the technological data of complex steering bridge casting collected during actual production processes,combined with the actual operating conditions of sand casting,the SMOTE(Synthetic Minority Oversampling Technique)data preprocessing algorithm was introduced to investigate its application in the defect predictions of complex castings manufactured by using of sand casting.According to the characteristics of the collected imbalance data set of complex castings,SMOTE data preprocessing algorithm was used to expand the imbalance data set scientifically and create a balance data set that can be used to train the defect prediction models of complex castings.The accuracies of the model predictions before and after data preprocessing increased from 86.50%to 97.91%.

关 键 词：转向桥铸件 砂型铸造 不平衡数据集 数据预处理 SMOTE算法 缺陷预测 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]