高超声速乘波前体-进气道三维复杂流场的POD-BPNN快速预测方法  

POD-BPNN prediction on the three-dimensional complex flow field of hypersonic waverider forebody/inlet

作　　者：关开港 苏纬仪[1] 崔晟 张文强 马航宇 安航 GUAN Kaigang;SU Weiyi;CUI Sheng;ZHANG Wenqiang;MA Hangyu;AN Hang(College of Energy and Power Engineering Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学能源与动力学院,南京210016

出　　处：《航空动力学报》2025年第2期99-115,共17页Journal of Aerospace Power

基　　金：先进航空动力工作站项目“可调进气道/预冷器多尺度流动换热机理及匹配设计方法研究(HKCX2020-02-021)”;中国航发四川燃气涡轮研究院外委课题“飞-发一体化试验智能设计研究”。

摘　　要：针对乘波前体-进气道三维构型,构建了基于本征正交分解(proper orthogonal decomposition,POD)和反向传播神经网络(back propagation neural network,BPNN)的流场快速预测模型,并对不同马赫数、攻角和反压下高超声速三维流场结构进行了快速预测研究。研究发现,快速预测模型能够准确预测样本空间内非样本工况的流场信息,对称面马赫数分布的预测误差小于0.5%、压力分布的预测误差小于3.8%、温度分布的预测误差小于1.1%。乘波体周围三维流场、壁面压力分布及主要性能参数的预测结果与CFD(computational fluid dynamics)计算结果高度一致。预测模型对样本空间外工况具备一定预测能力。流场数值模拟采用定常计算,而隔离段内激波串结构受到分离涡影响,具备非定常特性,预测模型在隔离段激波串区域存在较大的预测误差。A fast prediction model for the flow field was constructed based on proper orthogonal decomposition(POD)and back propagation neural network(BPNN)for a three-dimensional integrated waverider forebody-inlet.Moreover,the hypersonic three-dimensional flow fields were predicted under different Mach number,angle of attack and back pressure.The research showed that the prediction model can accurately predict the flow field of non-sample conditions in the sampling space.The errors of fast prediction model for the Mach number,pressure,and temperature were less than 0.5%,3.8%and 1.1%,respectively.The predictive results of the flow field for the waverider forebody,pressure distribution and the main performance parameters were highly consistent with the CFD(computational fluid dynamics)results.The prediction model had certain prediction ability for the external working conditions of the sampling space.However,the flow field numerical simulation adopted steady calculation,while the shock string structure in the isolator was affected by the separated vortex and had unsteady characteristics,indicating a large prediction error in the shock string area of the isolator.

关 键 词：乘波布局 本征正交分解 反向传播神经网络 多参数 快速预测 

分 类 号：V221.3[航空宇航科学与技术—飞行器设计]