RBF与SVM在注塑件翘曲预测中的对比分析  

Comparison and Analysis of Prediction in Warpage of Injection Molding Parts by RBF and SVM

作　　者：党玉春[1] 刘鸿滨[2] 翟秀云[1] 

机构地区：[1]攀枝花学院机电工程学院,四川攀枝花617000 [2]昆明理工大学机电工程学院,云南昆明650000

出　　处：《塑料》2013年第3期79-82,共4页Plastics

基　　金：攀枝花市科技项目计划(2011CY-G-5)

摘　　要：为了快速且较准确地预测注塑件的翘曲值,以某电脑显示器的外壳为例,分别利用RBF神经网络和SVM模型建立了显示器的翘曲值预测模型,并通过测试样本验证了2种模型的预测精度。结果表明:RBF网络模型和SVM模型的绝对百分比误差都在2%以内,说明二者都具有较好的预测性能;但从最大绝对百分比误差和最小绝对百分比误差分析得出,SVM模型比RBF模型更稳定,且预测精度更高,表明支持向量机的预测模型更适合处理此类问题。In order to predict the warpage of injection molded parts quickly and accurately,a computer monitor was taken as an example. Prediction models of warpage were set up based on the model of RBF neural network and the model of SVM respectively, and the accuracy of these two models were verified by test data. It was concluded that both the neural network model and the SVM model got the absolute percentage error within 2 % , which showed they both had a good prediction performance. The model of SVM had better stability and higher accuracy than the model of RBF neural network ,in terms of the value of maximum absolute percentage error and minimum absolute percentage error. It was shown that the prediction model of SVM was more suitable to deal with such problems.

关 键 词：注塑成型 翘曲 工艺参数 RBF SVM 

分 类 号：TQ320.662[化学工程—合成树脂塑料工业]