基于DPSO-BP神经网络的V形自由折弯成形角度和回弹预测  

Prediction on V-shaped free bending angle and springback based on DPSO-BP neural network

作　　者：占少伟 龚俊杰[1] 韦源源[1] 王金荣[2] 陈扬东 Zhan Shaowei;Gong Junjie;Wei Yuanyuan;Wang Jinrong;Chen Yangdong(College of Mechanical Engineering,Yangzhou University,Yangzhou 225000,China;Jiangsu Yawei Machine Tool Co.,Ltd.,Yangzhou 225200,China)

机构地区：[1]扬州大学机械工程学院,江苏扬州225000 [2]江苏亚威机床股份有限公司,江苏扬州225200

出　　处：《锻压技术》2023年第8期151-157,共7页Forging & Stamping Technology

基　　金：江苏省自然科学基金青年基金项目(BK20190869)。

摘　　要：采用基于DPSO算法优化BP神经网络(DPSO-BP)的机器学习算法建模,提出一种考虑材料参数和几何参数的V形自由折弯成形角度和回弹的预测方法。该方法主要引入非线性惯性权重改进粒子群(PSO)算法,进一步优化神经网络的初始权值和阈值,构建神经网络预测模型。以不同批号的SUS304不锈钢板料为研究对象,通过设计正交试验得到45个训练样本数据,验证所建立的预测模型的准确性。结果表明:采用DPSO-BP神经网络模型预测的成形角和回弹角的平均误差分别为0.150°和0.120°,与未优化的PSO-BP神经网络模型相比,预测的成形角和回弹角的平均误差明显减小,且计算耗时由14.0 min大幅缩短至0.8 min,同时实现了高预测精度和高计算效率。A prediction method about V-shaped free bending angle and springback considering material parameters and geometric parameters was proposed,according to the machine learning algorithm modelling based on the BP neural network optimized by the DPSO algorithm(DPSO-BP).The method mainly introduced the nonlinear inertia weight to improve particle swarm(PSO)algorithm,further optimized the initial weight and threshold of the neural network,and constructed the neural network prediction model.Then,the different batches of SUS304 stainless steel sheets was taken as the research object,45 training sample data were obtained by designing orthogonal experiment,and the accuracy of the constructed prediction model was verified.The results show that the average errors of the forming angle and the springback angle predicted by the DPSO-BP neural network model are 0.150°and 0.120°,respectively.Compared with the PSO-BP neural network model before optimization,the average errors of the forming angle and the springback angle are significantly reduced,and the calculation time is greatly shortened from 14.0 min to 0.8 min,achieving high prediction accuracy and high calculation efficiency at the same time.

关 键 词：回弹 V形自由折弯 BP神经网络 PSO算法 回弹角 

分 类 号：TG386[金属学及工艺—金属压力加工]