复合材料科学与工程 ›› 2024, Vol. 0 ›› Issue (12): 119-125.DOI: 10.19936/j.cnki.2096-8000.20241228.017

• 应用研究 • 上一篇    下一篇

基于机器学习的L形复合材料长桁固化变形预测

孙晓辉1, 吕毅1,2*, 王建军2, 张先芝3, 谢佳庆1   

  1. 1.西安工业大学 机电工程学院,西安 710021;
    2.西安航空学院 民航学院,西安 710077;
    3.哈德斯菲尔德大学 计算机与工程学院,英国HDI 3DH
  • 收稿日期:2023-09-12 出版日期:2024-12-28 发布日期:2025-01-14
  • 通讯作者: 吕毅(1981—),男,博士,教授,硕士生导师,研究方向为复合材料结构强度分析与验证,复合材料结构制造工艺,lyuyi@xaau.edu.cn。
  • 作者简介:孙晓辉(1998—),男,硕士,研究方向为复合材料结构固化变形。

Curing deformation prediction of L-shaped composite stringer based on machine learning

SUN Xiaohui1, LÜ Yi1,2*, WANG Jianjun2, ZHANG Xianzhi3, XIE Jiaqing1   

  1. 1. School of Mechatronic Engineering, Xi’an Technological University, Xi’an 710021, China;
    2. School of Civil Aviation, Xi’an Aeronautical University, Xi’an 710077, China;
    3. Department of Computing and Engineering, University of Huddersfield, Huddersfield HDI 3DH, UK
  • Received:2023-09-12 Online:2024-12-28 Published:2025-01-14

摘要: 针对复合材料结构在制造成型过程中固化变形机理十分复杂,涉及的影响参数较多且在固化过程中不断变化的情况,提出一种基于机器学习的方法,预测L形复合材料长桁在成型过程中的固化变形。使用ABAQUS有限元软件模拟L形复合材料长桁固化成型过程,建立以标准固化工艺温度曲线中的两段升温速率、两段保温温度、两段保温时间六个参数为特征的L形复合材料长桁固化回弹角数据集,通过构建径向基函数(Radical Basis Function,RBF)神经网络进行固化变形预测。结果表明,该方法拥有较高的预测精度和效率,预测误差在3%以内且模型耗时仅需1.25 s。

关键词: 复合材料, 固化变形, 径向基函数, 机器学习, L形长桁

Abstract: Aiming at the problem that the curing deformation mechanism of composite structure in the process of manufacturing is very complicated, and many parameters are involved and constantly change during the curing process, a method based on machine learning was proposed to predict the curing deformation of L-shaped composite stringer in the molding process. ABAQUS finite element software was used to simulate the curing molding process of L-shaped composite stringer in autoclave, and a data set of curing spring-in angle of L-shaped composite stringer was established, which was characterized by six parameters of the curing process temperature curve: two-stage heating rate, two-stage holding temperature and two-stage holding time. Then RBF neural network was constructed and curing deformation prediction was carried out. The results show that this method has high prediction accuracy and efficiency, the prediction error is less than 3%, and the model time is only 1.25 s.

Key words: composite, curing deformation, radical basis function, machine learning, L-shaped stringer

中图分类号: