基于RBF神经网络的飞机翼肋橡皮囊成形回弹预测  

Springback prediction of rubber bladder forming for aerofoil rib based on RBF neural network

作　　者：张凌云 石绍秋 王晓玉 徐淑赢 范作鹏[3] ZHANG Ling-yun;SHI Shao-qiu;WANG Xiao-yu;XU Shu-ying;FAN Zuo-peng(Key Laboratory of Fundamental Science for National Defense of Aeronautical Digital Manufacturing Process,Shenyang Aerospace University,Shenyang 110136,China;Air China Limited,Beijing 101318,China;A VIC Shenyang Aircraft Industry(Group)Co.,Ltd.,Shenyang 110034,China)

机构地区：[1]沈阳航空航天大学航空制造工艺数字化国防重点学科实验室,辽宁沈阳110136 [2]中国国际航空股份有限公司,北京101318 [3]航空工业沈阳飞机工业(集团)有限公司,辽宁沈阳110034

出　　处：《塑性工程学报》2021年第5期218-225,共8页Journal of Plasticity Engineering

摘　　要：为了对回弹量进行精准预测和控制,以飞机翼肋件为研究对象,结合有限元模拟和RBF神经网络对橡皮囊成形的飞机翼肋凸翻边回弹情况进行预测。首先利用PAM-STAMP 2G软件进行翼肋的橡皮囊成形模拟,将模拟得到的翻边回弹值与同参数下试验件的实际回弹测量值对比,验证了模拟试验代替实际成形的可行性。然后借助正交试验获得不同参数组合下的进行模拟试验的影响因素数组,并通过有限元模拟成形获得测量点的回弹值作为神经网络的训练样本,建立RBF神经网络并对其训练测试得到最优的网络模型。最后采用与神经网络输入相同的工艺参数进行翼肋的橡皮囊成形试验,将成形试验测得的回弹值与神经网络预测的回弹值进行对比,两种成形压力下的最大相对误差分别为3.54%和3.78%,远小于工业要求误差,验证了RBF神经网络对钣金件回弹预测的可靠性。To accurately predict and control the springback amount,aerofoil rib was taken as the research object,combined with finite element simulation and RBF neural network,the springback of aerofoil rib shrink flanging was predicted. Firstly,the simulation of rubber bladder forming of aerofoil rib was carried out by using PAM-STAMP 2 G software,and the simulated springback value was compared with the actual springback measurement value of the test part under the same parameters,which verifies the feasibility of replacing the actual forming by simulation test. Then,with the help of orthogonal test,the array of influencing factors for simulation test under different parameter combinations was obtained,and the springback value of measured points was obtained by finite element simulation forming as the training sample of neural network,and RBF neural network was established and trained to obtain the optimal network model. Finally,using the same process parameters as neural network input,the rubber bladder forming test of aerofoil rib was carried out. The springback value measured by forming test was compared with that predicted by neural network,the maximum relative errors under the two forming pressures are 3. 54% and 3. 78% respectively,which are much less than the industrial requirement error. The reliability of RBF neural network for predicting the springback of sheet metal parts is verified.

关 键 词：橡皮囊成形 回弹 有限元模拟 正交试验 RBF神经网络 

分 类 号：TG386[金属学及工艺—金属压力加工]