航空发动机

赵普1 ，毛宇凡 1 ，李嘉2 ，李金花 1 ，王晓放 1 ，刘海涛 1.有限数据下基于物理信息神经网络的平面叶栅流场重构预测方法[J].航空发动机,2025,51(2):84-90

有限数据下基于物理信息神经网络的平面叶栅流场重构预测方法

Plane Cascade Flow Field Reconstruction and Prediction via IPINN Under Limited Data

DOI：

英文关键词:ntegrated Physics-Informed Neural Network limited data flow field prediction plane cascade

基金项目:国家自然科学基金面上项目（52375231）、辽宁省科技联合计划项目（2024011954-JH4/4800、2023JH2/ 101800028）、国防科工局稳定支持科研项目、中央高校基本科研业务费（DUT24BK034）资助

作者	单位
赵普1 ，毛宇凡 1 ，李嘉2 ，李金花 1 ，王晓放 1 ，刘海涛 1	1.大连理工大学能源与动力学院，辽宁大连 116024；2.中国航发四川燃气涡轮研究院，成都 621000

摘要点击次数: 476

全文下载次数: 372

中文摘要:

为了快速准确地获取平面叶栅流场分布，提升现代压气机敏捷数字化设计效率，建立了基于有限数据的2维平面叶栅流场快速重构预测方法。该方法采用集成的物理信息神经网络（IPINN）架构，通过将叶栅流场的Navier-Stokes方程以损失函数的形式融入模型训练，嵌入物理先验知识，实现了有限数据下的高精度流场预测。结果表明：与纯粹数据驱动的模型相比，物理信息神经网络（PINN）模型融入物理信息后，对叶栅流场中速度轴向分量、速度垂直分量、压力3个物理场预测误差分别减小了 13.0%、 25.5% 和 76.3%；在此基础上，IPINN-cascade 模型进一步优化，针对这3个物理场的预测值误差相较于 PINN 模型的又分别减小了 14.8%、19.8% 和 17.5%。所提出的方法在有限数据条件下能够有效捕捉叶栅流场的主要特征，为压气机设计提供了新的技术手段。

英文摘要:

In order to quickly and accurately obtain the flow field distribution of a plane cascade and enhance the efficiency of modern compressor agile digital design, a rapid reconstruction and prediction method for the two-dimensional plane cascade flow field based on limited data was established. The method employs an Integrated Physics-Informed Neural Network (IPINN) architecture and achieves high-precision flow field prediction under limited data by incorporating the Navier-Stokes equations of the cascade flow field as a loss function into the model training process and embedding physical prior knowledge. The results show that, compared to purely data- driven models, the Physics-Informed Neural Network (PINN) model reduces the prediction errors of the axial velocity, vertical velocity, and pressure fields in the cascade flow by 13.0%, 25.5%, and 76.3%, respectively. Furthermore, the optimized IPINN-cascade model has the prediction errors for these three physical fields decreased by 14.8%, 19.8%, and 17.5% in comparison to the PINN model. The method proposed can effectively capture the main characteristics of the cascade flow field under limited data conditions, providing a new technological approach for compressor design.

查看全文查看/发表评论下载PDF阅读器