神经网络与遗传算法在拉延筋参数反求中的应用

APPLICATION OF NEURAL NETWORK AND GENETIC ALGORITHM TO INVERSE SOLUTION OF PARAMETERS OF DRAWBEAD

以某车型前地板角支撑板的拉延工序为例,讨论BP神经网络技术与遗传算法在拉延筋几何参数反求中的综合应用问题,建立能描述反映成形效果的三个参数与半圆形拉延筋几何参数之间非线性映射关系的神经网络模型,并运用遗传算法对神经网络结构进行了优化。提出逐次局部密化样本点的样本点设计方法。该方法有助于加快神经网络的设计进程,提高神经网络的模拟精度。当训练样本数据可通过有限元法自动获得时,使用该方案则更为便利。

The application of neural network and genetic algorithm to inverse solution of parameters of drawbead used in drawing procedure of corner backstop of front floor of a truck is discussed. A neural network model, the structure of which is optimized in genetic algorithm, is designated to simulate the nonlinear mapping relation between parameters used to reflect metal forming effect and geometric parameters of semi-circular drawbead. A sample design method with gradual local den-sifing samples is developed. The method is helpful to shorten design period of neural network and to improve simulating degree of neural network. It is especially suitable to the case of sample datum being gotten by finite analysis automatically.