神经算子增强的双级低压涡轮子午面流场全景式预测模型

打开文本图片集

关键词：双级低压涡轮；子午面流场；全景式预测；神经算子网络中图分类号：TK14文献标志码：ADOI：10.7652/xjtuxb202506011 文章编号： 0253-987X（2025）06-0103-09

A Neural Operator Enhanced Panoramic Prediction Model for Meridional Flow Field of Two-Stage Low Pressure Turbine

JIANG Shoumin1'²，CHENG Hui1 ， SONG Limingl， ZHU Ruoyu²，GUO Zhendongl （1. School of Energy and Power Engineering，Xi'an Jiaotong University，Xi'an 71o049，China; 2.AECC Shenyang Engine Research Institute，Shenyang 11ool5，China）

Abstract：To address the ineficiency and poor engineering applicability of traditional flow field prediction models due to separate construction required for each performance parameter of the turbine，an efficient panoramic prediction framework is proposed to predict accurately any key performance parameter of the turbine stage meridional plane. With a two-stage low-pressure turbine taken as an example，a high-precision panoramic prediction model for the turbine meridional plane was established. Specifically，the proposed prediction framework first predicted the six basic physical parameters of the turbine stage meridian plane，such as temperature， pressure，density and velocity，and then predicted the overall performance parameters of the turbine stage and the distribution of key cross-section performance parameters along the blade height as wellas the meridian plane contour of the key parameters. Meanwhile，in order to improve the prediction accuracy of the performance parameters of meridional plane， Transformer was integrated into a Neural Operator network under the framework of panoramic prediction，and a Transformer enhanced Neural Operator （TNO） prediction model was established. The test results of the two-stage meridian prediction model based on TNO showed that， the TNO prediction model could accurately predict the overallperformance parameters such as mass flow rate/turbine power/expansion ratio of the turbine stage，the distribution of outlet flow angle/stage reaction along the span，and the meridian contour of entropy，etc. The relative prediction error was less than 1% ， and the prediction accuracy of TNO was significantly higher than that of the panoramic prediction model based on the classical UNet network. Thus，the effectiveness of the proposed panoramic prediction framework and model has been well demonstrated.

Keywords： two-stage low-pressure turbine; flow field of meridian plane; panoramic prediction; neural operator network

由于涡轮性能评估所依赖的纳维-斯托克斯（N-S）方程的通用解析解仍然未知，通常采用计算流体动力学（CFD）及试验等方法开展涡轮性能分析与设计。（剩余12971字）