欢迎访问过程工程学报, 今天是

›› 2008, Vol. 8 ›› Issue (5): 957-961.

• 系统与集成 • 上一篇    下一篇

基于神经网络规则抽取的带钢热镀锌质量监控模型

王建国 阳建宏 张文兴 徐金梧   

  1. 北京科技大学、内蒙古科技大学 北京科技大学机械工程学院 内蒙古科技大学机械工程学院 北京科技大学机械工程学院
  • 收稿日期:2008-06-05 修回日期:2008-07-14 出版日期:2008-10-20 发布日期:2008-10-20
  • 通讯作者: 王建国

Strip Hot-dip Galvanizing Quality Monitoring Model Based on Neural Network Rule Extraction

WANG Jian-guo YANG Jian-hong ZHANG Wen-xing XU Jin-wu   

  1. Mechanical Engineering School, University of Science and Technology Beijing Mechanical Engineering School, University of Science and Technology Inner Mongolia Mechanical Engineering School, University of Science and Technology Beijing
  • Received:2008-06-05 Revised:2008-07-14 Online:2008-10-20 Published:2008-10-20
  • Contact: WANG Jian-guo

PDF

被引次数

9

摘要： 为了克服传统神经网络产品质量监控模型中解释性差的困难，提出了基于神经网络规则抽取的带钢热镀锌质量监控模型. 以带钢热镀锌生产中锌层重量监控为研究对象，利用神经网络规则抽取方法对样本数据进行学习，以知识规则的形式给出模型中输入(原料参数及生产控制参数)与输出(产品质量)间的定量关系，用于对生产控制参数的设定与更新. 选取756个训练样本和376个测试样本分别对网络进行了训练和检验，结果表明，新模型中的知识规则覆盖率达到94.8%，并可根据输出变量的目标区间快速地设定各输入变量的范围，为产品质量的自动控制提供了有效的方法.

关键词: 神经网络, 规则抽取, 带钢热镀锌, 质量监控

Abstract: To overcome the difficulty of production quality monitoring model based on traditional neural network which is usually used poorly, a strip hot-dip galvanizing quality monitoring model based on neural network rule extraction is proposed. Taking the quality monitoring of zinc coating weight in strip hot-dip galvanizing as the investigated subject, the sample datasets are trained by neural network rule extraction method to obtain the quantitative relationships in the form of knowledge rules among input variables (such as the parameters of raw materials and control parameters of production) and output ones (the quality parameters), with which the production control parameters can be set and updated easily. 756 training and 376 testing examples are chosen as variables of the network. The results show that the new model has a rule-overcast-ration of 94.8% and has provided an effective tool for auto-control of product quality, because in the new model the range of each input variable can be readily set up based on the target range of the output variables.

Key words: neural network, rule extraction, strip hot-dip galvanizing, quality monitoring

中图分类号: