摘要
涡轮冷却技术是涡轮设计的关键技术之一。涡轮冷却结构的表面温度和压力均为复杂的三维分布,但目前的设计体系仍基于低保真度的一维管网,深度学习可能是提升涡轮冷却设计保真度的一种有效手段。本文总结了几种纯数据的神经网络模型和几种与物理规律结合的神经网络模型在涡轮冷却问题中的应用。通过巧妙地将涡轮冷却问题看待为图像、序列和参数的建模,神经网络模型可以实现不同保真度的拟合。尽管目前深度学习方法在涡轮冷却领域已取得一些进展,但本领域过小的数据量与神经网络方法的严重冗余特性阻碍了此类方法的发展。结合数据、物理规律、计算流体力学、分布式测量技术和偏微分方程理论的综合数学模型可能是涡轮冷却设计工具的未来。
Turbine cooling technology is one of the key technologies in turbine design. The surface temperature and pressure of the turbine cooling structure are complex three-dimensional distributions, but the current design system is still based on a low-fidelity one-dimensional flow network.Deep learning may be an effective means to improve the fidelity of the turbine cooling design. The present study summarizes the applications of several pure data-based neural networks and physical informed neural networks in turbine cooling problems. By cleverly treating the turbine cooling problem as the modeling of images, sequences and parameters, the neural network can achieve different fidelity fittings. Although deep learning methods have made some progress in the field of turbine cooling, the small-scale of data in this field and the severe redundancy characteristics of neural networks hinder the development of such methods. A comprehensive mathematical model that combines data, physical laws, computational fluid dynamics, distributed measurement technology, and partial differential equation theory may be the future of turbine cooling design tools.
作者
汪奇
杨力
饶宇
WANG Qi;YANG Li;RAO Yu(Institute of Turbomachinery,School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《工程热物理学报》
EI
CAS
CSCD
北大核心
2022年第3期656-662,共7页
Journal of Engineering Thermophysics
基金
国家自然科学基金资助项目(No.51906139)
上海市青年科技英才扬帆计划项目(No.19YF1423200)。
关键词
涡轮冷却
物理规律
神经网络
深度学习
turbine cooling
physical laws
neural networks
deep learning
作者简介
汪奇(1995-),男,博士研究生,主要从事航空发动机、燃气轮机传热与流动的深度学习研究。;通信作者:杨力,副教授,liy59@sjtu.edu.cn。
