• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2022, Vol. 58 ›› Issue (10): 78-86.doi: 10.3901/JME.2022.10.078

• 材料科学与工程 • 上一篇    下一篇

扫码分享

基于改进遗传算法优化人工神经网络的304不锈钢流变应力预测准确性研究

丁军1, 古愉川1, 黄霞1, 宋鹍1, 路世青1, 王路生2   

  1. 1. 重庆理工大学机械工程学院 重庆 400054;
    2. 合肥工业大学材料科学与工程学院 合肥 230009
  • 收稿日期:2021-06-28 修回日期:2021-12-26 出版日期:2022-05-20 发布日期:2022-07-07
  • 通讯作者: 丁军(通信作者),男,1978年出生,博士,教授,硕士研究生导师。主要研究方向为基于机器学习的材料力学性能预测、材料失效破坏的多尺度模拟方法等。E-mail:dingjun@cqut.edu.cn
  • 作者简介:古愉川,男,1996年出生,硕士研究生。主要研究方向为机械设计及理论。E-mail:guyuchuan@2019.cqut.edu.cn
  • 基金资助:
    国家自然科学基金委员会与中国工程物理研究院联合(U1530140)、重庆市自然科学基金面上(cstc2020jcyj-msxmX0286)和重庆市教育委员会科学技术研究(KJQN202001126)资助项目。

Research on Prediction Accuracy of Flow Stress of 304 Stainless Steel Based on Artificial Neural Network Optimized by Improved Genetic Algorithm

DING Jun1, GU Yuchuan1, HUANG Xia1, SONG Kun1, LU Shiqing1, WANG Lusheng2   

  1. 1. College of Mechanical Engineering, Chongqing University of Technology, Chongqing 400054;
    2. School of Materials Science and Engineering, Hefei University of Technology, Hefei 230009
  • Received:2021-06-28 Revised:2021-12-26 Online:2022-05-20 Published:2022-07-07

摘要: 随着航空航天和武器装备等国防军工领域对金属材料高温、高应变率使役条件下的力学性能提出了更高要求。以304不锈钢为例,提出一种基于改进遗传算法选择算子的优化人工神经网络预测金属在复杂使役环境下流变应力的新模型。以应变范围为0.1~0.5、温度变化范围为20~600℃、应变率区间为0.001~100 s-1下的304不锈钢流变应力试验数据为基础,构建了304不锈钢流变应力预测模型,并将预测结果与决策树、线性回归和未改进遗传神经网络模型进行对比,以平均绝对误差MAE和决定系数R2为检验参数来评价所建立模型的准确性。结果显示,改进遗传神经网络模型在测试集数据上的MAE和R2最佳,表明该模型能够很好预测304不锈钢流变应力。

关键词: 304不锈钢, 遗传算法, 人工神经网络, 流变应力

Abstract: The development in the defense and military industry such as the aerospace and weaponry require the higher mechanical performance for the metallic materials under the elevated temperature coupled with the high strain rate conditions. In the light of the values for 304 stainless steel from the experimental measurements, a new artificial neural network(ANN) model is proposed, which is optimized by modifying the selection operator in the genetic algorithm to predict the values for the flow stress of metallic materials in the complicated service condition. The improved model of the flow stress prediction is established on the basis of new ANN method from the experiments for the strain range of 0.1-0.5, temperature range of 20-600 ℃ and strain rate range of 0.001-100 s?1. Taken the mean absolute error(MAE) and the determination coefficient(R2) as the criterions, the results calculated from the improved model are compared with those from the regression tree model(RR), linear regression model(LR) and the unimproved genetic neural network model(GNN). The MAE and R2 for the ANN optimized by improved genetic Algorithm model shows the minimum value of 21.91 and the maximum of 0.97, respectively, in comparison with RR, LR and GNN model, which indicates that it can accurately predict the flow stress of 304 stainless steel.

Key words: 304 stainless steel, genetic algorithm, artificial neural network, flow stress

中图分类号: