基于数据驱动的镁合金压铸件质量智能预测  

Intelligent Quality Prediction of Magnesium Alloy Die-Casting Parts Based on Data-Driven Method

作　　者：汪星辰 王鑫 付彭怀[1] 童胜坤 陈滨 彭立明[1] Wang Xingchen;Wang Xin;Fu Penghuai;Tong Shengkun;Chen Bin;Peng Liming(School of Materials Science and Engineering,Shanghai Jiao Tong University,Shanghai 200240;Meridian Lightweight Technologies,Shaoxing 312500)

机构地区：[1]上海交通大学材料科学与工程学院,上海200240 [2]万丰镁瑞丁新材料科技有限公司,绍兴312500

出　　处：《汽车工艺与材料》2025年第4期40-45,共6页Automobile Technology & Material

基　　金：国家重点研发计划项目(2021YFB3701000);宁波科技项目(20241ZDYF020400);自然科学基金项目(U21A2048、51821001);广东省重点领域研发计划(2020B010186001);上海市科委重点项目(21DZ1208200);广东省基础与应用基础研究基金(2022B1515120046)。

摘　　要：为实现镁合金压铸件质量的智能预测,降低人工下线检测成本,提升镁合金压铸产业智能化水平,通过收集镁合金大型薄壁压铸件“工艺参数-质量参数”大数据,采用随机森林模型建立工艺参数与铸件产生的缺陷种类间的关系,分析了工业数据中的标签长尾分布现象对机器学习模型预测性能的影响,通过“随机降采样+SMOTE过采样”算法对数据集分布进行均衡化,最终获得了准确率为89.54%、受试者工作特征曲线(ROC)下面积为0.9838、平均真正率为87.65%的准确预测模型,实现了极少数含缺陷样本的精准检出,并获得了镁合金压铸关键工艺参数重要性排序。In order to achieve intelligent predication of magnesium alloy die-casting parts,reduce offline labor inspection cost,and improve intelligent level of magnesium alloy die-casting industry,this paper collects big data on“process parameters-quality parameters”of large thin-walled magnesium alloy castings,and uses random forest model to establish the relationship between process parameters and the types of defects in castings,and analyzes the effect of long-tailed distribution of labels in the industrial data on the predictive performance of machine learning models.Then the“Random Downsampling+SMOTE Over-sampling”algorithm is emptoyed to balance the distribution of the data set.Finally,an accurate prediction model with an accuracy of 89.54%,an area under ROC curve of 0.9838,and an average true rate of 87.65%are obtained,which achieves a precise detection of a small number of defective samples,and obtains the ranking of the importance of key process parameters for magnesium alloy casting.

关 键 词：高压铸造 镁合金 机器学习 质量智能预测 

分 类 号：TG292[金属学及工艺—铸造] TG249.2