E3高压涡轮三维流场快速预测方法研究

Rapid Prediction Method of Three Dimensional Flow Field for E3 High Pressure Turbine

关键词:  高压涡轮  本征正交分解  代理模型  神经网络  快速预测  相关分析

Key words:  High pressure turbine  Proper orthogonal decomposition  Surrogate model  Neural network  Rapid prediction  Correlation analysis

摘要:

快速获得温度场和压力场载荷环境是航空发动机涡轮寿命预测的关键。在本征正交分解的基础上，分别采用响应面法、径向基函数、Kriging方法和BP神经网络，构建了E³涡轮三维流场拓扑结构的多种快速预测方法，为载荷环境实时预测提供了途径。结果表明，本征正交分解能成功地实现E³涡轮三维旋转流场的降阶，基于响应面法、径向基函数、Kriging方法和BP神经网络能精确预测涡轮流场预测结构，但在预测精度、速度等方面存在差异。在样本空间范围内点预测上，10阶模型下四种方法预测出的压力场和温度场误差均小于1%，流量、效率预测误差低于0.4%；在样本空间范围以外点的预测上，径向基函数和Kriging方法的表现不稳定。涡轮壁面流场相关性分析表明，压力场预测精度与转速、进口压力高度相关；温度场与转速、进口压力的相关性弱于压力场。

Abstract:

Quickly obtaining the temperature and pressure load environment is the key for the life prediction of aeroengine. Based on proper orthogonal decomposition， this paper utilized the response surface method， radial basis function， Kriging method and BP neural network to construct for rapid prediction methods for the flow field of E³ turbine， which provide real-time prediction for the load environment. The results show that proper orthogonal decomposition has successfully achieved the order reduction for the 3D rotating flow field of E³ turbine. Response surface method， radial basis function， Kriging method and BP neural network have different prediction accuracy and speed for turbine flow field. For the cases under the 10 modes in the sampling range， the predicted errors of pressure and temperature by the four prediction methods are all less than 1%，and the prediction errors of mass flow and efficiency are less than 0.4%. For the cases out of the sampling range， the performance of radial basis function and Kriging method is unstable. Correlation analysis for turbine wall flow field shows that the prediction accuracy of pressure is highly correlated with the rotational speed and incoming pressure. The correlation of temperature with rotational speed and incoming pressure is weaker than that of pressure.