|
| 基于物理信息神经网络的大跨度桥梁主梁疲劳寿命预测 |
| Fatigue life prediction of Orthotropic Steel Bridge Deck for long-span bridges based on physics-informed neural networks |
| 投稿时间：2026-04-07 修订日期：2026-05-18 |
| DOI： |
| 中文关键词: 结构健康监测 疲劳寿命预测 物理信息神经网络 正交异性钢桥面板 智能预警 |
| 英文关键词: Structural health monitoring Fatigue life prediction Physics-informed neural network Orthotropic steel bridge deck Intelligent early warning |
| 基金项目: |
|
| 摘要点击次数: 154 |
| 全文下载次数: 0 |
| 中文摘要: |
| 大跨径桥梁钢梁面板在运营期间的疲劳寿命预测至关重要,从健康监测的有限数据中准确获取主梁参数和响应是评估其性能及结构安全的关键。传统基于物理的方法需要理想环境条件,而纯数据驱动方法在泛化能力和可解释性方面存在局限。为解决这些问题,本文提出一种物理信息神经网络(PINN),将物理原理与数据驱动技术相结合用于主梁的疲劳寿命预测。与传统神经网络不同,PINN通过在损失函数中引入物理约束,选取Miner线性累积损伤准则与Paris 裂纹扩展公式作为核心物理约束,结合结构力学平衡方程,构建多维度的物理约束体系,使得网络在学习过程中不仅拟合观测数据,同时遵循本身的物理规律,从而提高预测的可解释性和物理一致性,实现对结构动态响应中疲劳寿命的高效预测。本文以安庆长江公路大桥主梁面板监测数据为案例,使用PINN物理信息神经网络进行深度学习训练,结果表明该模型在预测精度与鲁棒性方面均优于传统预测方法,为桥梁健康监测系统的智能预警与运维决策提供了有力的技术支持。 |
| 英文摘要: |
| Predicting the fatigue life of steel bridge decks in long-span bridges during operation is crucial, and accurately extracting key girder parameters and responses from limited health monitoring data is essential for assessing structural performance and safety. Traditional physics-based methods require ideal environmental conditions, while purely data-driven approaches suffer from limitations in generalization ability and interpretability. To address these issues, this paper proposes a Physics-Informed Neural Network (PINN) that integrates physical principles with data-driven techniques for fatigue life prediction of main girders. Unlike conventional neural networks, PINN introduces physical constraints into the loss function, adopting Miner’s linear cumulative damage rule and Paris’ crack growth law as core physical constraints, combined with structural mechanical equilibrium equations to establish a multi-dimensional physical constraint system. This enables the network to not only fit observed data but also adhere to underlying physical laws during training, thereby enhancing prediction interpretability and physical consistency, and enabling efficient fatigue life prediction under dynamic structural responses. Using monitoring data from the main girder deck of the Anqing Yangtze River Highway Bridge as a case study, the PINN model is trained via deep learning. Results demonstrate that the proposed model outperforms traditional forecasting methods in both prediction accuracy and robustness, providing strong technical support for intelligent early warning and maintenance decision-making within bridge health monitoring systems. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
