航空学报 > 2025, Vol. 46 Issue (19): 531574-531574   doi: 10.7527/S1000-6893.2025.31574

基于数据融合-知识迁移方法的高温数字图像相关试验应力表征

郭程洁1,2, 徐典1,2, 李进宝1,2, 程超宇1,2, 郭硕昌1,2, 李锐1,2()   

  1. 1.大连理工大学 力学与航空航天学院,大连 116024
    2.大连理工大学 工业装备结构分析优化与CAE软件全国重点实验室,大连 116024
  • 收稿日期:2024-11-26 修回日期:2024-12-24 接受日期:2025-01-24 出版日期:2025-07-01 发布日期:2025-02-10
  • 通讯作者: 李锐 E-mail:ruili@dlut.edu.cn
  • 基金资助:
    国家自然科学基金(12372067);国家自然科学基金(12022209);国防基础科研计划(JCKY2021205B003);辽宁省自然科学基金杰出青年基金(2025JH6/101100005);大连市杰出青年科技人才项目(2024RJ005)

Stress characterization of high-temperature digital image correlation experiments based on a data fusion-knowledge transfer method

Chengjie GUO1,2, Dian XU1,2, Jinbao LI1,2, Chaoyu CHENG1,2, Shuochang GUO1,2, Rui LI1,2()   

  1. 1.School of Mechanics and Aerospace Engineering,Dalian University of Technology,Dalian 116024,China
    2.State Key Laboratory of Structural Analysis,Optimization and CAE Software for Industrial Equipment,Dalian University of Technology,Dalian 116024,China
  • Received:2024-11-26 Revised:2024-12-24 Accepted:2025-01-24 Online:2025-07-01 Published:2025-02-10
  • Contact: Rui LI E-mail:ruili@dlut.edu.cn
  • Supported by:
    National Natural Science Foundation of China(12372067);National Defense Basic Scientific Research Program of China(JCKY2021205B003);Liaoning Science Fund for Distinguished Young Scholars(2025JH6/101100005);Dalian Science Fund for Distinguished Young Scholars(2024RJ005)

摘要:

数字图像相关(DIC)试验系统因其能够实现对全场应变的非接触式精确测量,被广泛应用于高温环境下结构的力学性能评估。然而,基于DIC测量技术的试验方法难以直接表征结构应力场,且高温环境下材料参数与温度密切相关,有限元法高度依赖材料参数而导致应力结果精度不足。因此,如何结合高温DIC试验与有限元法实现结构全场应力的精确表征,对理解、测评结构的力学性能具有重要意义。提出了一种高温DIC试验应力表征的数据融合-知识迁移方法,针对航空结构中典型高温钛合金搭建了高温DIC试验系统、高精度有限元模型、迁移学习框架,实现了基于小样本试验数据的高精度应力场表征。首先,基于多层感知机神经网络对大样本仿真数据进行预训练,捕捉复杂高温环境下结构坐标与应力间的映射关系;其次,基于小样本高温DIC试验数据对预训练模型进行微调,表征了高温环境下的结构应力场,将有限元模型中的知识有效迁移至试验数据;最后,利用均方误差、决定系数等指标,验证了所提方法用于结构应力场预测的准确性、适用性,为表征高温环境下的试验应力场提供了一种新思路。

关键词: 高温数字图像相关, 数据融合, 迁移学习, 应力表征, 多层感知机神经网络

Abstract:

The Digital Image Correlation (DIC) experimental system is widely used for mechanical performance evaluation of structures in high-temperature environments due to its ability to achieve non-contact precise measurements of full-field strain. However, experimental methods based on DIC measurement technology face challenges in directly characterizing the stress field of structures, particularly in high-temperature environments where material parameters are closely related to temperature, and the heavy reliance of the Finite Element Method (FEM) on these parameters leads to insufficient accuracy of stress outcomes. Therefore, it is of great significance to understand and evaluate the mechanical properties of structures by combining high-temperature DIC experiments and FEM to accurately characterize the full-field stress of structures. This paper proposes a data fusion-knowledge transfer method for stress characterization in high-temperature DIC experiments. A high-temperature DIC experimental system, a high-precision FEM model, and a transfer learning framework are developed for typical high-temperature titanium alloys in aerospace structures, achieving accurate characterization of stress fields based on small-sample experimental data. Firstly, a multi-layer perceptron neural network is employed to pre-train large-sample simulation data, capturing the mapping relationship between structural coordinates and stress in complex high-temperature environments. Secondly, based on small-sample high-temperature DIC test data, the pre-trained model is fine-tuned to characterize a structural stress field in high-temperature environments, and effectively transfer knowledge from the finite element model to the experimental data. Finally, the accuracy and applicability of the proposed method for predicting structural stress fields are verified using indicators such as the mean square error and the coefficient of determination, which provides a new approach for characterizing experimental stress fields in high-temperature environments.

Key words: high-temperature digital image correlation, data fusion, transfer learning, stress characterization, multilayer perception neural network

中图分类号: