考虑燃烧室出口温度分布的航空发动机部件级模型

Aero-engine component level model considering combustion chamber outlet temperature distribution

燃烧室出口温度分布不均匀会使涡轮叶片受到不均匀的热载荷,严重影响涡轮叶片的工作寿命。本文提出一种具有燃烧室出口温度分布预测功能的部件级模型建模方法,为燃烧室出口温度分布控制研究提供了仿真平台。以变循环发动机为研究对象,根据其设计点参数设计燃烧室三维模型,通过CFD数值模拟的方法,计算得到该燃烧室三维模型在地面不同工作状态下的燃烧室出口温度分布场,组成温度分布场训练数据集。提出基于Inception-反卷积网络的燃烧室出口温度分布场重建方法,基于该方法构建了燃烧室出口温度分布场预测模型。建立了适用于全包线、全状态,可以预测燃烧室出口温度分布场的部件级模型,与传统的部件级模型相比,该模型能够预测发动机在不同工作状态、不同包线点下的燃烧室出口温度分布场。结果表明:Inception-反卷积网络在训练集和测试集上的均方误差比常规反卷积降低11.83%和5.6%,比WGAN-GP降低87%和90%;部件级模型预测温度分布场和CFD仿真温度分布场的温度分布趋势基本一致;所提出的Inception-反卷积网络预测精度高于常规反卷积网络和WGAN-GP网络预测精度,在热斑处温度点误差更小,在亚声速巡航点(H=8 km,Ma=0.7),(H=8 km,Ma=0.9)和超声速巡航点(H=10 km,Ma=1.4),(H=10 km,Ma=1.6)时,Inception-反卷积网络预测温度分布场的平均温度误差分别为0.05 K,-1.38 K,-1.54 K和4.44 K,均方误差分别为6.7×10^(-4),1.9×10^(-4),3.0×10^(-4)和1.4×10-3,热斑温度误差分别为-3.91 K,-3.67 K,-5.34 K和0.85 K。

The uneven distribution of temperature at the outlet of the combustion chamber will cause the turbine blades to suffer from uneven thermal load and seriously affect the working life of the turbine blades.A component-level modeling method with predictive function of temperature distribution at combustor outlet is proposed,which provides a simulation platform for the research of temperature distribution control at combustor outlet.Firstly,the variable cycle engine was taken as the research object,and the combustion chamber 3D model was designed according to its design point parameters.Through CFD numerical simulation,the temperature distribution field of the combustion chamber outlet of the 3D model under different working conditions on the ground was calculated,and the training data set of temperature distribution field was composed.Then,a reconstruction method of the outlet temperature distribution field of the combustion chamber based on the Inception-deconvolution network is proposed,and a prediction model of the outlet temperature distribution field of the combustion chamber is constructed based on this method.Finally,a component-level model is established that is applicable to the entire envelope and state and can predict the temperature distribution field at the outlet of the combustion chamber.Compared with the traditional component level model,the model can predict the temperature distribution field at the outlet of the combustion chamber under different working conditions and different enveloping line points.The results show that the mean square error of the initiative-deconvolution network on training set and test set is 11.83%and 5.6%lower than that of conventional deconvolution,and 87%and 90%lower than that of WGANGP.The trend of temperature distribution field predicted by component level model is basically the same as that simulated by CFD.The prediction accuracy of the proposed Inception deconvolution network is higher than that of conventional deconvolution network and WGAN-GP network,and the error of temperature points at hot spots is smaller.At subsonic cruise(H=8 km,Ma=0.7),(H=8 km,Ma=0.9)and supersonic cruise point(H=10 km,Ma=1.4),(H=10 km,Ma=1.6),the mean temperature errors of the predicted temperature distribution field by the Inception-deconvolution network are 0.05 K,-1.38 K,-1.54 K and 4.44 K,and mean square errors are 6.7×10^(-4),1.9×10^(-4),3.0×10^(-4) and 1.4×10-3,respectively.The temperature errors of hot spots were-3.91 K,-3.67 K,-5.34 K and 0.85 K,respectively.