基于鱼群RBF神经网络和改进蚁群算法的拉深成形工艺参数优化

Process parameters optimization of deep drawing based on fish RBF neural network and improved ant colony algorithm

针对基本蚁群算法只能应用于离散问题的缺陷,借鉴鱼群算法中人工鱼移动机制,提出了一种连续蚁群算法。为了减少板料成形工艺参数优化时间,建立工艺参数与成形目标之间的RBF神经网络近似模型。以板料成形中分块压边力为设计变量,以板料成形后增厚和减薄为成形目标,通过拉丁超立方抽样,运用Dynaform软件进行仿真以获得训练样本,基于人工鱼群算法建立分块压边力与成形质量间RBF神经网络近似模型,最后采用连续蚁群算法对该模型进行优化,获得最优分块压边力。并以油底壳零件为例进行工艺参数优化,通过与整体压边力对比,证明该方法可以有效提高成形件质量,并为快速计算最优分块压边力提供了依据。

For the shortcoming of the basic ant colony algorithm which can only apply to discrete problems and by referring to artificial fish moving mechanism of fish swarm algorithm, a new ant colony algorithm for continuous problems was put forward. To i＇educe the time of process parameters optimization of sheet forming, the approximate model of RBF neural network was established between the process pa- rameters and the forming target. Several segmental blank holder forces were set as the design variables in sheet forming, and thickening and thinning were regarded as forming target after sheet forming. Latin hypereube was used to sample, and the software of Dynaform was simula- ted to get the training samples. RBF neural network approximate mode between several segmental blank holder forces and forming quality was established based on the artificial fish swarm algorithm. The continuous ant colony algorithm was applied to optimize the model, and the opti- mal forces of several segmental blank holders were obtained. Taking oil pan for example, by contrast with the whole blank holder force, the method effectively improves the forming quality, and provides the basis for calculating the optimal blank holder force quickly.