融合相似性原理的涡轮叶型流场预测方法研究  

STUDY ON FLOW FIELD PREDICTION OF TURBINE BLADES BY COUPLING SIMILARITY PRINCIPLE

作　　者：郭振东[1] 成辉 陈云 蒋首民 宋立明[1] 李军[1] 丰镇平[1] Guo Zhendong;Cheng Hui;Chen Yun;Jiang Shoumin;Song Liming;Li Jun;Feng Zhenping(School of Energy and Power,Xi’an Jiaotong University,Xi’an 710049,China;AECC Shenyang Engine Research Institute,Shenyang 110015,China)

机构地区：[1]西安交通大学能源与动力工程学院,西安710049 [2]中国航发沈阳发动机研究所,沈阳110015

出　　处：《力学学报》2023年第11期2647-2660,共14页Chinese Journal of Theoretical and Applied Mechanics

基　　金：国家科技重大专项(2019-II-0008-0028);国家自然科学基金(51936008,52306048)资助项目。

摘　　要：计算流体力学(CFD)方法是涡轮叶片等设计阶段性能评估的重要手段.然而,基于CFD的数值仿真方法通常比较耗时,难以满足涡轮叶型设计阶段快速迭代的需求.为实现快速性能评估并克服纯数据驱动预测模型泛化能力不足的问题,受到物理增强的机器学习思路的启发,将相似性原理与深度学习模型相结合,提出了一种泛化能力强的涡轮叶型流场预测新方法.以涡轮叶片表面等熵马赫数分布预测为例,提出采用相似性原理对叶型几何变量和气动参数进行归一化,进而在归一化参数空间构建训练样本集与深度学习预测模型,由此建立统一的流场预测模型,对几何尺寸、边界条件差异较大的叶型气动性能进行评估.在完成模型训练后,对归一化条件下不同工况/不同形状叶型的流场、真实环境下不同工况/不同尺寸叶型的流场以及GE-E3低压涡轮不同截面叶型的流场进行预测,结果表明预测结果的分布曲线与CFD评估结果吻合良好,平均相对误差在1.0%左右,由此验证了所提出的融合相似性原理的流场预测模型的精度与泛化能力.Computational fluid dynamics(CFD)is an important tool to evaluate the performance of turbine blades and etc.in the design stage.However,the numerical simulation of turbine blades that based on CFD method can be very time-consuming,which makes it rather difficult to meet the need of rapid iteration in the design process of turbine blades.In order to evaluate the performance of turbine blades rapidly and overcome the problem of insufficient generalization ability of pure data-driven prediction models as well,inspired by the concept of physics augmented machine learning,a novel method for turbine blade flow field prediction with strong generalization ability is proposed,by combining the similarity principle with deep learning model.Taking the prediction of the isentropic Mach number distribution at the surface of turbine blades as an example,we propose to make use of the similarity principle to normalize the geometric variables and aerodynamic parameters of turbine blades,and then prepare the training sample set and train the deep learning-based prediction model in the normalized parameter space.And accordingly,a unified prediction model based deep learning can be obtained,which can quickly predict the aerodynamic performance of turbine blades that in very different geometric size and have different boundary condition values.After finishing the model training,the trained prediction model is used to predict the flow fields of the turbine blades that works under different operation condition and of different shape in normalized design space,the flow fields of real-world blades of different size/different working conditions,and the flow fields of different section profiles of GE-E3 low-pressure turbines.The results showed that the predicted results were in good agreement with the CFD evaluation results,and the averaged relative error was less than 1.0%,which verify the accuracy and generalization ability of the proposed flow field prediction model coupling the similarity principle.

关 键 词：深度学习流场预测 物理增强的机器学习 相似性原理 数据驱动模型 气动分析 

分 类 号：TK14[动力工程及工程热物理—热能工程]