基于神经网络无铆钉自冲铆接头力学性能预测.pdfVIP

基于神经网络无铆钉自冲铆接头力学性能预测 龙江启，兰凤崇，陈吉清 （华南理工大学 机械与汽车工程学院，广东 广州 510640） 摘 要：为了研究钢铝一体化结构车身无铆钉自冲铆接接头力学性能，引入反向传播（BP ）神经网 络模型来描述板材厚度、板材硬度及成形接头底部直径等工艺参数与接头剪切力及剥离力强度等力 学性能的映射关系。由于标准BP 网络存在训练精度低、收敛速度慢及泛化能力差等缺陷，采用归一 化法与Levenberg-Marquardt (LM)算法结合的算法来优化神经网络预测模型连接权值，从而提高了神 经网络模型的预测精度和泛化能力。由BTM®Tog-L-Loc系统采用不同连接工艺参数组合生成训练样 本与测试样本，把训练样本的试验数据提供给神经网络学习，然后用测试样本试验数据对神经网络 的预测结果进行检验。结果表明，训练后的神经网络模型能够准确有效地预测无铆钉自冲铆接接头 力学性能，证实了神经元网络应用于无铆钉自冲铆接接头力学性能预测的可行性与可靠性。预测结 果可为设计出优质的钢与铝无铆钉自冲铆接接头提供依据。 关键词：无铆钉自冲铆；接头；反向传播神经网络；预测 TH164 文献标识号: A 中图分类号： Neural network-based mechanical property prediction in the mechanical clinching joints LONG Jiang-qi, LAN Feng-chong, CHEN Ji-qing (School of Mechanical Automotive Engineering, South China University of Technology, Guangzhou 510640, China) Abstract: In order to investigate the property of mechanical clinching joints in the steel-aluminum hybrid structure car body, back-propagation (BP) neural networks are used to describe the mapping relationship between joining technique parameters such as sheet thickness, sheet hardness, joint bottom diameter etc., and joints mechanical properties of shearing and peeling. To overcome the existing disadvantages of the standard propagation algorithm in lower training precision, slow convergence process and bad generalization capability, in this study the algorithms of normalization and Levenberg-Marquardt are combined to optimize the standard BP neural network connection weights and improve the model prediction precision and generalization capability. The training and validating samples are made by the ® BTM Tog-L-Loc system with different combination of technique parameters. Then the parameters of tra