工业工程

工业工程

• 实践与应用 • 上一篇    下一篇

基于数据驱动仿真的两设备系统元模型

  

  1. 上海大学 机电工程与自动化学院，上海 200072
  • 出版日期:2016-10-31 发布日期:2017-02-21
  • 作者简介:夏蓓鑫（1984-），男，浙江省人，讲师，博士，主要研究方向为制造系统建模与仿真.
  • 基金资助:

    国家自然科学基金资助项目（71401098）；上海市科学技术委员会科研计划资助项目（14511108303）；上海市高校青年教师培养资助计划资助项目（ZZSD15047）

A Study of Metamodeling of Twomachine Systems Based on Datadriven Simulation Technique

  1. School of Mechatronics Engineering and Automation, Shanghai University, Shanghai 200072, China
  • Online:2016-10-31 Published:2017-02-21

摘要：

半导体封装测试系统等复杂制造系统的性能分析是项非常困难的任务。利用仿真模型构建两设备系统元模型，并以元模型为基石构建面向大规模复杂系统的近似解析方法是分析复杂制造系统的有效手段。为了快速准确地构建两设备系统元模型，提出了一种基于数据驱动仿真技术及人工神经网络的元模型构建方法。该方法以考虑缓存输送时间的两设备制造系统为研究对象，采用AREAN的二次开发技术实现仿真模型的自动配置、运行、统计，以生成人工神经网络所需案例，并通过比较分析BP、RBF和Chebyshev这3类典型的函数逼近神经网络确定最优的人工神经网络模型。实验结果表明径向基函数密度为120的RBF神经网络模型表现最优，其结果误差最小，能够成为大规模复杂制造系统近似解析方法的基石。

关键词: 数据驱动仿真, 元模型, 人工神经网络

Abstract:

It is a difficult task to analyze the performance of complex manufacturing systems like semiconductor assembly and testing systems. To fulfill the task, an efficient approach to develop a twomachine metalmodel based on simulation model as well as an approximate analytical method for large systems based on the developed metalmodel is proposed. A study of metamodeling of twomachine systems is carried out by using datadriven simulation technique in order to find out a fast and accurate method to build the metalmodel. Twomachine systems taking into account transfer delays in buffers are taken as a research object. To obtain cases for artificial neural network, secondary development based on ARENA is made to automatically configure and run simulation models and gather statistics. Three typical artificial neural networks for function approximation (BP, RBF and Chebyshev) are compared and optimized. The experiment results show that RBF model with 120 spread is the best. The low rate of error of that model indicates that it is accurate enough to be the building block of approximate analytical methods for the analysis of large systems.

Key words: data driven simulation, metamodel, artificial neural network