工业工程

工业工程 ›› 2012, Vol. 15 ›› Issue (4): 12-16.

• 专题论述 • 上一篇    下一篇

基于神经网络的SPC/EPC整合过程监测方法研究

  

  1. （郑州航空工业管理学院 管理科学与工程学院,河南 郑州  450015）
  • 出版日期:2012-08-31 发布日期:2012-09-19
  • 作者简介:王秀红（1974-）,女，河北省人，副教授，主要研究方向为质量工程和技术创新.
  • 基金资助:

    国家自然科学基金资助项目（70771102）；航空科学基金资助项目（2010ZG55025）；河南省科技攻关资助项目（122102210512）

Method of Monitoring Abnormality under Integrated SPC/EPC Based on Neural Network Techniques

  1. （Zhengzhou Management Institute of Aviation Industry ,Zhenzhou 450015，China）
  • Online:2012-08-31 Published:2012-09-19

摘要： 为解决统计过程控制（SPC）/工程过程调整（EPC）整合引起的传统SPC控制图监测异常扰动效率低的问题，提出了采用神经网络技术监测SPC/EPC整合过程的策略，并对神经网络模型结构和参数设置进行分析，构建过程输入、过程输出及两者的协方差为输入参数，异常扰动发生与否为输出参数的3层神经网络模型。为验证该方法的性能，进行了大量的比较实验：即对相同的样本，分别采用Shewhar图、CUSUM图和上述神经网络模型进行监测。实验结果表明：神经网络模型能准确监测幅度大于2的阶跃扰动和大于2的过程漂移，平均运行步长（ARL）为1；传统SPC监测技术只能较准确地（监测率大于90%）监测幅度大于5的阶跃扰动和大于2的过程漂移，ARL大于2。与传统监测方法相比，该方法能快速有效地监测异常扰动的发生。

关键词:  , 统计过程控制, 工程过程控制, 神经网络

Abstract: It is known that, under the integrated scheme of  statistical process control（SPC） and engineering process control（EPC）, the SPCs capability of monitoring the feedback-controlled process is low.To resolve this problem, neural network techniques are introduced into the integrated SPC/EPC method. Based on structural analysis and parameter setting, a three-layer neural network model is presented. For model training, the input data include process inputs, process outputs, and their covariance, and the output dada are whether an abnormality occurs. A number of tests are done to compare with Shewhart chart and CUSUM chart methods. Results show that the proposed model outperforms traditional SPC methods. It can accurately monitor a process for step disturbance with change over 2 and process drift with range over 2, and average run length(ARL) value equal to 1. While the traditional SPC methods can correctly monitor a process (monitoring rate > 90%) only for step disturbance with change over 5 and process drift with range over 2, and ARL value greater than 2.

Key words: statistical process control, engineering process control, neural networks