基于深度学习神经网络的铝合金型材回弹预测  

Springback prediction of aluminum alloy profile based on deep learning neural network

作　　者：王鹏鹏 程子詹 凌强 刘宇[1] 王春举 吴子彬 长海博文 Wang Pengpeng;Cheng Zizhan;Ling Qiang;Liu Yu;Wang Chunju;Wu Zibin;Nagaumi Hiromi(School of Mechanical and Electrical Engineering,Soochow University,Suzhou 215131,China;Performance Metal Structural Materials Research Institute,Soochow University,Suzhou 215131,China;Weiqiao Lightweight Research Center at Soochow,Suzhou 215131,China)

机构地区：[1]苏州大学机电工程学院,江苏苏州215131 [2]苏州大学高性能金属结构材料研究院,江苏苏州215131 [3]魏桥轻量化(苏州)科技有限公司,江苏苏州215131

出　　处：《锻压技术》2024年第7期105-111,共7页Forging & Stamping Technology

基　　金：高精度型材成型加工工艺技术(P114401222)。

摘　　要：铝合金型材在各个领域均具有广泛的应用,为解决了6061铝合金方管在生产与加工过程中出现的回弹问题,构建了以Python语言作为开发环境、Keras作为深度学习框架的ANN(Artificial Neural Network)算法,使用包含两个隐藏层的4层全连接神经网络模型进行数据训练。算法后端的数据库内容通过弯曲回弹试验获得,采用结构化的MySQL关系型数据库系统存取和管理试验所得的198条弯曲回弹数据记录。最后,通过足量的模型训练与实际预测可得,该算法的角度回弹预测均方误差MSE的平均值为0.044、曲率回弹预测平均绝对百分比误差MAPE的平均值为4.255。算法训练和比较验证的结果表明,该回弹预测系统具有满足误差要求的预测精度,其预测结果可为铝合金型材的弯曲回弹与补偿提供有效参考。Aluminum alloy profiles are widely used in various fields,in order to solve the springback problem in the production and processing of 6061 aluminum alloy square tubes,an artificial neural network(ANN)algorithm was established by using Python as the development environment and Keras as the deep learning framework,and the data training was conducted by using a four-layer fully connected neural network model with two hidden layers.Then,the backend database content for the algorithm was derived by bending springback tests,and the 198 bending springback data records obtained from these tests were stored and managed by using a structured MySQL relational database system.Finally,through sufficient model training and actual prediction,the average value of a mean squared error(MSE)of angle springback prediction for this algorithm was 0.044,and the average of a mean absolute percentage error(MAPE)of curvature springback prediction for this algorithm was 4.255.The results of algorithm training and comparative validation show that this springback prediction system achieves the requisite accuracy of error requirement,which provides an effective reference for the bending springback and compensation of aluminum alloy profiles.

关 键 词：铝合金型材 回弹预测 深度学习 人工神经网络 Keras MYSQL 

分 类 号：TG386[金属学及工艺—金属压力加工]