铸造工艺数据驱动的工程机械铸件缺陷预测OACSTPCD
Casting Process Data-Driven Defect Prediction for Construction Machinery Castings
针对砂型铸造过程缺陷难以寻因、工艺数据类别不平衡问题,提出了一种基于特征重分布与代价敏感学习的卷积神经网络缺陷预测方法.首先,根据样本特征相关性,对特征向量排列顺序进行了优化;其次,基于不平衡工艺数据样本设计了代价敏感正则项,对模型损失函数进行了修正;最后,构建了缺陷预测模型(FR-CS-CNN).测试结果表明,本研究构建的FR-CS-CNN在总体预测精度上达到了93.67%,相比卷积神经网络提升了2.96%.
Directing at the problems of difficult to find defect cause,and category imbalance of technical data during sand casting process,a convolutional neural network defect prediction method based on feature redistribution and cost sensitive learning was proposed to solve the defects in sand casting process.Firstly,according to the feature correlation of samples,the sequence of feature vectors is optimized.Secondly,the cost sensitive regular term is designed based on the unbalanced process data sample,and the model loss function is modified.Finally,a defect prediction model(FR-CS-CNN)is constructed.The test results show that the overall prediction accuracy of FR-CS-CNN constructed in this study reaches 93.67%,which is 2.96%higher than that of convolutional neural network.
刘迎辉;周建新;余朋;潘徐政;朱守琴;计效园;吴来发;殷亚军;沈旭;解明国
华中科技大学材料成形与模具技术全国重点实验室,湖北武汉 430074安徽合力股份有限公司合肥铸锻厂,安徽合肥 230022
金属材料
砂型铸造缺陷预测代价敏感学习卷积神经网络
sand castingdefect predictioncost-sensitive learningconvolutional neural networks
《铸造》 2024 (009)
1329-1335 / 7
国家重点研发计划项目(2020YFB1710100);国家自然科学基金(52275337、52090042、51905188).
评论