泄漏(经济)
人工神经网络
参数统计
反问题
润滑
计算流体力学
趋同(经济学)
外推法
反向
模拟
计算机科学
收敛问题
一致性(知识库)
收敛速度
机械
雷诺数
机械工程
一般化
MATLAB语言
物理
雷诺方程
物理系统
戒指(化学)
天然气
边值问题
计算模型
算法
数学优化
人工智能
作者
Jie Xu,Baoyue Dai,Jiaxin Si,Xuexing Ding,Chaojun Deng
摘要
Predicting leakage rates in floating ring gas film seals has long been recognized as one of the most challenging problems in sealing technology. Although traditional numerical methods perform excellently in solving the Reynolds equation for individual operating conditions, they still face challenges such as high repetitive computational costs and low data utilization efficiency in scenarios requiring extensive parameterization or inverse problem solving. To this end, this paper proposes a leakage prediction model for floating ring gas film seals based on physics-informed neural networks (PINNs), which achieves a natural fusion of parametric modeling and data-driven physical constraints. A dynamic loss-weighting strategy is adopted, initially prioritizing data fitting to accelerate convergence and subsequently strengthening physical constraints to improve generalization. After training on leakage-rate sample data, the model achieves high-precision leakage prediction while ensuring consistency with lubrication physics. Comparative studies with purely physics-driven and purely data-driven models show that the PINN approach delivers markedly higher prediction accuracy (R2 > 0.90), improved physical interpretability, improved computational efficiency, and superior generalization across operating conditions. This method integrates physical rigor with intelligent learning, offering a practical foundation for the intelligent development of floating ring gas film sealing technology.
科研通智能强力驱动
Strongly Powered by AbleSci AI