基于BP神经网络的激光弯曲成形工艺参数优化  

Process parameters optimization of laser bending based on BP neural network

作　　者：张雅晶 董文彬 杨拓宇 陈丰 ZHANG Ya-jing;DONG Wen-bin;YANG Tuo-yu;CHEN Feng(College of Mechanical Engineering,Anhui Science and Technology University,Bengbu 233000,China)

机构地区：[1]安徽科技学院机械工程学院,安徽蚌埠233000

出　　处：《塑性工程学报》2020年第8期66-71,共6页Journal of Plasticity Engineering

基　　金：安徽省教育厅重点项目(KJ2020A0071,KJ2020A0072);安徽省留学人员创新项目(2018-830246)。

摘　　要：利用MATLAB软件构建了用于预测激光弯曲成形过程中激光扫描速度和激光功率的3层BP神经网络,其中,输入层、隐含层和输出层节点数分别为2、12和1。将实验获得的15组样本数据进行归一化处理后对所构建的BP神经网络进行训练和验证,并将扫描速度的预测误差控制在8%以下,激光功率的预测误差控制在6%以下。利用训练后的BP神经网络对获得0.1°~0.5°板材弯曲角所需的激光扫描速度和激光功率进行预测,得到30组工况下的预测数据。计算每种工况下的线能量大小并进行比较,以最小线能量原则筛选出获得每种板材弯曲角所需的最佳工艺参数组合。MATLAB software was employed to establish a three-layer BP neural network for predicting the laser scanning speed and laser power during laser bending process. The number of neurons in input layer,hidden layer and output layer was 2,12 and 1,respectively.Fifteen groups of experimental data were normalized to train and verify the established BP neural network. The prediction error of laser scanning speed was controlled below 8% and the prediction error of laser power was controlled below 6%. The trained BP neural network was used to predict the laser scanning speed and laser power needed to obtain the sheet bending angle of 0. 1°-0. 5°,and thirty groups of prediction data were obtained. The line energy under each working condition was calculated and compared,and the optimal combination of process parameters was selected on the basis of the minimum line energy principle.

关 键 词：激光弯曲 MATLAB BP神经网络 预测误差 工艺参数 

分 类 号：TG665[金属学及工艺—金属切削加工及机床]