摘要
材料的多轴疲劳寿命预测研究是保证部件结构完整性的关键要素之一。近年来机器学习尤其是神经网络在疲劳寿命预测领域得到了广泛应用。然而,疲劳数据的不足阻碍了神经网络在疲劳预测中的进一步应用。为了解决这一问题,考虑疲劳先验物理知识的物理信息神经网络逐渐受到关注。首先,概述了机器学习算法的分类及神经网络模型在多轴疲劳寿命预测中的应用。随后,重点对基于物理信息神经网络的材料疲劳寿命预测研究进行了深入探讨。最后,从基于物理信息的输入特征、基于物理信息的损失函数构建和基于物理信息的网络框架开发等3个方面对物理信息神经网络模型的发展进行介绍。相关研究表明,在材料多轴疲劳寿命预测过程中,物理信息神经网络可以表现出更好的物理一致性和预测性能。
The research on multiaxial fatigue life prediction of materials is one of the critical elements in ensuring the structural integrity of components.In recent years,machine learning,especially neural networks,has been widely applied in fatigue life prediction.However,the scarcity of fatigue data has limited the further application of neural networks in fatigue prediction.To address this issue,physics-informed neural networks that consider prior physical knowledge of fatigue have gradually gained attention.Firstly,provided an overview of the classification of machine learning algorithms and the application of neural-network models in multiaxial fatigue life prediction.Then,it focused on a deep exploration of the research on material fatigue life prediction based on physics-informed neural networks.Finally,the development of physics-informed neural networks was introduced from three aspects:physics-informed input features,the construction of physics-informed loss functions,and physics-informed network frameworks.Relevant studies show that physics-informed neural networks can exhibit better physical consistency and prediction performance in the process of multiaxial fatigue life prediction of materials.
作者
张颛利
孙兴悦
陈旭
ZHANG Zhuanli;SUN Xingyue;CHEN Xu(Industrial Protection Engineering Center,CNOOC Energy Development Equipment Technology Co.,Ltd.,Tianjin 300457,China;School of Chemical Engineering,Tianjin University,Tianjin 300350,China)
出处
《机械强度》
北大核心
2025年第2期44-52,共9页
Journal of Mechanical Strength
基金
国家自然科学基金项目(12302098)
国家资助博士后研究人员资助计划项目(GZB20230508)。
关键词
物理信息神经网络
多轴疲劳
寿命预测
机器学习
Physics-informed neural network
Multiaxial fatigue
Life prediction
Machine learning
