涡轮叶片厚壁带肋通道流动与传热性能的预测和优化

Prediction and Optimization on Flow and Heat Transfer Performance of Ribbed Thick-Wall Channel in Turbine Blade

为提高燃气轮机涡轮叶片中弦区带肋通道的传热效果,基于已有实验数据开展了雷诺数为10000~60000、宽高比为0.25~4.00和肋角度为30°~90°的涡轮叶片厚壁带肋通道流动与传热性能的优化研究。利用Sobol方法分析了厚壁带肋通道流动及传热性能对雷诺数、宽高比、肋角度的全局敏感性;采用BP神经网络对60组实验数据进行了多输入、多输出的非线性拟合,得到了可以准确预测厚壁带肋通道平均努塞尔数、摩擦系数和综合热力系数的神经网络模型,并结合遗传算法对厚壁带肋通道在研究范围内的不同工况下进行结构优化。结果表明:厚壁带肋通道的传热性能对进口雷诺数变化的敏感性很高,而对通道宽高比和肋角度变化的敏感性较低;流动性能对通道宽高比变化的敏感性非常高,对肋角度变化的敏感性较低,而对进口雷诺数变化的不敏感。训练得到的BP神经网络模型对平均努塞尔数、摩擦系数和综合热力系数的最大预测偏差分别为3.9%、2.8%和4.6%。最优宽高比和最优肋角度都随着雷诺数的增大而增大。不同雷诺数时,基于平均努塞尔数优化得到的最优宽高比和肋角度分别在2.23~2.75和41.12°~55.16°;基于综合热力性能优化得到的最优宽高比和肋角度分别在1.75~2.11和56.68°~65.2°。该研究可为未来重型燃机叶片冷却结构设计提供参考。

To improve heat transfer effect of internal cooling channel in the mid-chord region of gas turbine blade,the optimization of flow and heat transfer performance of ribbed thick-wall channel in a turbine blade with Reynolds number ranging from 10000 to 60000,aspect ratio ranging from 0.25 to 4.00 and rib angle ranging from 30°to 90°is investigated based on the existing experimental data.Sobol method is used to analyze the global sensitivity of flow and heat transfer performance of the ribbed thick-wall channel to Reynolds number,aspect ratio and rib angle.A BP(back propagation)neural network is used to perform multi-input and multi-output nonlinear fitting for 60 groups of experimental data,and a multi-input and multi-output neural network model is obtained which can accurately predict average Nusselt number,friction factor and comprehensive thermal coefficient of the ribbed thick-wall channel.Then structural parameters of the ribbed thick-wall channel are optimized by using a genetic algorithm under different working conditions.Results show that the heat transfer performance of the ribbed thick-wall channel is highly sensitive to the change of inlet Reynolds number,but is less sensitive to the change of channel aspect ratio and rib angle.The flow performance is highly sensitive to the change of channel aspect ratio,but is less sensitive to the change of rib angle,and insensitive to the change of inlet Reynolds number.The maximum prediction deviations of the trained BP neural network model for average Nusselt number,friction factor and comprehensive thermal coefficient are 3.9%,2.8%and 4.6%respectively.The optimal values of aspect ratio and rib angle increase with the increase of Reynolds number.At different Reynolds numbers,the optimal aspect ratio and rib angle based on the optimization of average Nusselt number are from 2.23 to 2.75 and 41.12°to 55.16°,respectively,while the optimal aspect ratio and rib angle based on the optimization of comprehensive thermal performance are from 1.75 to 2.11 and 56.68°to 65.2°,respectively.This research may provide reference for the design of blade cooling structure of heavy duty gas turbine in future.