基于神经网络和遗传算法的冲压成形多目标优化技术  

Study on multi-objects optimization design techniques of stamping based on neural network and genetic algorithm

作　　者：彭必友[1] 胡腾[1] 肖兵[1] 殷国富[2] 

机构地区：[1]西华大学材料科学与工程学院,成都610039 [2]四川大学制造科学与工程学院,成都610065

出　　处：《塑性工程学报》2010年第1期57-61,共5页Journal of Plasticity Engineering

基　　金：国家自然科学基金资助项目(50275100);校人才引进基金资助项目(R0820103)

摘　　要：在冲压成形质量控制中,目标质量间常常是相互冲突的(如破裂和起皱)。传统求解多目标优化问题取决于设计人员对优化模型的理解程度、实践经验等,求解的结果在工程中并非为最合理。文章提出一种集数字化分析技术、神经网络和遗传算法于一体的冲压成形多目标优化设计技术。其以数字化分析的大量结果作为神经网络的学习样本,遗传算法所需的目标函数值由神经网络模型预测,该技术实现了多目标优化过程中遗传算法个体适应度值的动态求解,从而解决了数字化分析计算量大的缺陷。实例验证了该优化技术的有效性,为冲压成形优化设计提供了一种新的方法。There is often conflict between the quality goals in stamping forming quality control, such as crack and wrinkle. The solving of traditional multi-objective optimization problem is dependent on the understanding on the optimization model and the practical experience of the designer, therefore the result is not often so reasonable. The multi-objects optimization techniques for stamping process was put forward in this paper, which integrates digitized analysis, neural network and genetic algorithms. The artificial neural network was trained by learning samples coming from digital analysis. The fitness values are obtained on the basis of a multilayer BP neural network. This technique realized the dynamic solving for the individual fitness value of genetic algorithm in the multi-objects optimization process and solved the problem of too much calculation in digital analysis. Examples demonstrate the effectiveness of the optimization techniques, which provide a new method for stamping forming design.

关 键 词：冲压 多目标优化 神经网络 遗传算法 

分 类 号：TG386[金属学及工艺—金属压力加工]