中国塑料 ›› 2010, Vol. 24 ›› Issue (08): 64-71 .DOI: 10.19491/j.issn.1001-9278.2010.08.017

• 加工与应用 • 上一篇    下一篇

基于神经网络集的注射成型工艺参数多目标优化

胡泽豪 卫炜 刘娟 刘琨   

  1. 中南林业科技大学
  • 收稿日期:2010-04-12 修回日期:2010-06-07 出版日期:2010-08-26 发布日期:2010-08-26

Multi-objective Optimization of Injection Molding Processing Parameters Based on Neural Networks Ensemble

  

  • Received:2010-04-12 Revised:2010-06-07 Online:2010-08-26 Published:2010-08-26

摘要: 以计算机辅助工程(CAE)数值仿真正交试验所得工艺参数与质量指标的数据作为训练样本,对经过优化的BP神经网络进行训练,得到工艺参数与制品质量指标之间的神经网络集近似计算代理模型,该模型快速准确,有明确的数学公式,可以利用遗传算法进行全局寻优,得到使多个质量指标综合最优的工艺参数组合。通过对比验证,这种多目标优化方法可以在正交试验结果数据较少的情况下较大程度地提高制品的多个质量指标。

关键词: 多目标优化, 注塑成型, 工艺参数, 神经网络集, 代理模型, 遗传算法

Abstract: The data from CAE simulation orthogonal test was used as training samples to establish the neural networks ensemble approximate calculation agent model. With clear mathematical formula, the model may carry out global optimization by genetic algorithm quickly and accurately, and the optimal process parameters were obtained. By comparing and verification, this multi-objective optimization method can improve multiple quality indicators in the case of lacking sufficient orthogonal data.

Key words: multi-objective optimization, injection molding, processing parameter, neural networks ensemble, agent model, genetic algorithm