基于深度神经网络的压气机叶型性能预测

杜周; 徐全勇; 马玉林; 蒋耀东

doi:10.13224/j.cnki.jasp.20240123

基于深度神经网络的压气机叶型性能预测

doi: 10.13224/j.cnki.jasp.20240123

杜周^{1, 2,},
徐全勇^2,*, ,,
马玉林¹,
蒋耀东^{1, 2}

1.
辽宁工程技术大学力学与工程学院，辽宁阜新 123032
2.
清华大学航空发动机研究院，北京 100084

基金项目: 国家科技重大专项（J2019-Ⅴ-0001-0092，J2019-Ⅴ-0013-0108）；清华大学自主科研计划（20234616001）

详细信息

作者简介:
杜周：杜周（1996-），男，博士生，主要从事压气机设计与人工智能结合相关方面的研究。E-mail：321115728@qq.com

通讯作者:
徐全勇（1980-），男，副研究员，博士，主要从事航空发动机气动热力学方面的研究。E-mail：xuquanyong@tsinghua.edu.cn

中图分类号: V231.3
计量
- 文章访问数: 682
- HTML浏览量: 686
- PDF量: 129
- 被引次数: 0
出版历程
- 收稿日期: 2024-03-03
- 网络出版日期: 2024-09-02

Performance prediction of compressor blade profile based on deep neural network

DU Zhou^{1, 2
,},
XU Quanyong^{2,*
, ,},
MA Yulin¹,
JIANG Yaodong^{1, 2}

1.
School of Mechanics and Engineering，Liaoning Technical University，Fuxin Liaoning 123032，China
2.
Institute of Aircraft Engine，Tsinghua University，Beijing 100084，China

摘要

摘要:
基于深度学习的方法建立了一种压气机叶型在非设计工况下的总压损失系数和落后角预测流程。以NACA65系列叶型为例，通过拉丁超立方采样建立一种综合考虑叶型设计变量和来流工况的定制叶型生成方法，共计15500组工况，将对应来流工况下的叶栅算例进行了二维仿真计算，将计算得到的总压损失系数和落后角数据进行数据分析和数据处理后，将处理过的数据与相对应的设计变量映射提供给神经网络进行训练，得到叶栅的总压损失系数和落后角模型，总压损失系数和落后角平均绝对误差分别为0.00129和0.18673°，并与传统的非设计损失和落后角经验模型进行了对比验证。结果显示：基于深度学习的可以准确预测在非设计工况下的损失与落后角，对比经验模型预测结果具有更高的准确度，该代理模型可用于压气机气动设计。
- 轴流压气机 /
- 叶型性能 /
- 总压损失系数 /
- 落后角 /
- 深度学习
Abstract:
A process for predicting the total pressure loss coefficient and deviation angle of axial compressor blades under off-design conditions was established using deep learning methods. Taking the NACA65 series blades as an example, a customized blade generation method considering blade design variables and incoming flow conditions was developed using Latin hypercube sampling, under a total of 15500 operating conditions. Two-dimensional simulation calculations were performed for the corresponding blade cases under these flow conditions. The calculated total pressure loss coefficient and deviation angle data were analyzed and processed, and the processed data, along with the corresponding design variables, were provided to a neural network for training. The final model for blade total pressure loss coefficient and deviation angle was obtained, with a total pressure loss coefficient and deviation angle mean absolute error of 0.00129 and 0.18673°, respectively. A comparison with traditional empirical models for off-design total pressure loss coefficient and deviation angle was conducted for validation. The results showed that the deep learning-based approach can accurately predict loss and deviation angle under off-design conditions with higher accuracy compared with empirical models. This surrogate model can be applied to aerodynamic design of axial compressors.
- axial compressor /
- blade profile performance /
- total pressure loss coefficient /
- deviation angle /
- deep learning

HTML

图 1 NACA65系列叶型中线分布和厚度分布

Figure 1. Camber and thickness distribution of NACA65-series blades

下载: 全尺寸图片幻灯片

图 2 不同升力系数和最大相对厚度的叶型几何

Figure 2. Blade geometry with different lift coefficients and maximum relative thickness

下载: 全尺寸图片幻灯片

图 3 典型的NACA65叶栅几何

Figure 3. Distribution of centerline and thickness of NACA65 series leaf profiles

下载: 全尺寸图片幻灯片

图 4 叶型设计空间

Figure 4. Blade design space

下载: 全尺寸图片幻灯片

图 5 叶栅典型算例网格

Figure 5. Typical grid for a cascade example

下载: 全尺寸图片幻灯片

图 6 NACA65-（12）10叶型压力系数分布结果对比

Figure 6. Comparison of pressure coefficient distribution results for NACA65-（12）10 blade

下载: 全尺寸图片幻灯片

图 7 落后角和总压损失系数分布

Figure 7. Distribution of deviation angle and total pressure loss coefficient

下载: 全尺寸图片幻灯片

图 8 数据相关性分析

Figure 8. Data dependence analysis

下载: 全尺寸图片幻灯片

图 9 神经网络结构模型

Figure 9. Neural network architecture model

下载: 全尺寸图片幻灯片

图 10 损失函数收敛曲线图

Figure 10. Loss function convergence curve

下载: 全尺寸图片幻灯片

图 11 损失函数收敛曲线图

Figure 11. Loss function convergence curve

下载: 全尺寸图片幻灯片

图 12 总压损失函数预测结果

Figure 12. Predicted results of total pressure loss function

下载: 全尺寸图片幻灯片

图 13 落后角预测结果

Figure 13. Predicted results of deviation angle

下载: 全尺寸图片幻灯片

图 14 S₁和S_t敏感度

Figure 14. Sensitivity of S₁ and S_t

下载: 全尺寸图片幻灯片

图 15 不同马赫数下非设计落后角

Figure 15. Non-design deviation angle at different Mach numbers

下载: 全尺寸图片幻灯片

图 16 总压损失系数预测结果

Figure 16. Predicted results of total pressure loss function

下载: 全尺寸图片幻灯片

表 1 NACA65系列叶型设计空间

Table 1. NACA65 series blade design space

设计变量	变量下界	变量上界
来流马赫数Ma	0.3	0.7
气流入口角$ {\beta _1} $/（°）	30	70
攻角$ i $/（°）	−2	8
最大相对厚度$ {t_{\text{b}}} $/%	4	14
升力系数$ {C_{{\text{lo}}}} $	0	30
稠度$\sigma $	0.5	2

下载: 导出CSV

表 2 性能预测网络参数设计

Table 2. Performance prediction network parameter design

层类型	输入形状	输出形状	参数数量
全连接层	（32, 6）	（32, 32）	（6×32）+32=224
全连接层	（32, 32）	（32, 64）	（32×64）+64=2112
全连接层	（32, 64）	（32, 128）	（64×128）+128=8320
全连接层	（32, 128）	（32, 256）	（128×256）+256=33024
全连接层	（32, 256）	（32, 512）	（256×512）+512=131584
一维卷积层	（32, 4, 128）	（32, 8, 64）	（4×8×3）+8=104
一维卷积层	（32, 8, 64）	（32, 16, 32）	（8×16×3）+16=400
一维卷积层	（32, 16, 32）	（32, 32, 16）	（16×32×3）+32=1552
一维卷积层	（32, 32, 16）	（32, 64, 8）	（32×64×3）+64=6208
一维卷积层	（32, 64, 8）	（32, 128, 4）	（64×128×3）+128=24704
平均一维池化	（32, 128, 4）	（32, 128, 1）	0
展平		（32, 128）	0
全连接层	（32, 128）	（32, 1）	（128×1）+1=129

下载: 导出CSV

参考文献(21)

[1]	程荣辉,桂幸民,等. 航空发动机风扇压气机设计[M]. 北京: 科学出版社,2022. CHENG Ronghui,GUI Xingmin,et al. Design of aero-engine fan compressor[M]. Beijing: Science Press,2022. (in Chinese CHENG Ronghui, GUI Xingmin, et al. Design of aero-engine fan compressor[M]. Beijing: Science Press, 2022. (in Chinese)
[2]	SCHNOES M,NICKE E. A database of optimal airfoils for axial compressor throughflow design[J]. Journal of Turbomachinery,2017,139(5): 051008. doi: 10.1115/1.4035075
[3]	杜周,马玉林,徐全勇,等. 基于ResNet数据驱动的压气机叶型优化[J]. 航空动力学报,2023,38(7): 1592-1603. DU Zhou,MA Yulin,XU Quanyong,et al. Optimization of compressor blade profile based on ResNet data drive[J]. Journal of Aerospace Power,2023,38(7): 1592-1603. (in Chinese DU Zhou, MA Yulin, XU Quanyong, et al. Optimization of compressor blade profile based on ResNet data drive[J]. Journal of Aerospace Power, 2023, 38(7): 1592-1603. (in Chinese)
[4]	JOHNSEN I A,BULLOCK R O. Aerodynamic design of axial-flow compressors[M]. Washington: National Aeronautics and Space Administration,1965.
[5]	王何建,刘波,张博涛. 基于一维模型的跨声速轴流压气机特性预测研究[J]. 推进技术,2021,42(6): 1256-1264. WANG Hejian,LIU Bo,ZHANG Botao. Characteristics prediction of transonic axial compressor based on one-dimensional model[J]. Journal of Propulsion Technology,2021,42(6): 1256-1264. (in Chinese WANG Hejian, LIU Bo, ZHANG Botao. Characteristics prediction of transonic axial compressor based on one-dimensional model[J]. Journal of Propulsion Technology, 2021, 42(6): 1256-1264. (in Chinese)
[6]	费腾,季路成,周玲. 神经网络模型在压气机通流特性分析中的应用[J]. 航空动力学报,2022,37(6): 1260-1272. FEI Teng,JI Lucheng,ZHOU Ling. Application of neural network model in compressor through-flow analysis[J]. Journal of Aerospace Power,2022,37(6): 1260-1272. (in Chinese FEI Teng, JI Lucheng, ZHOU Ling. Application of neural network model in compressor through-flow analysis[J]. Journal of Aerospace Power, 2022, 37(6): 1260-1272. (in Chinese)
[7]	马博文,巫骁雄,于洋. 基于机器学习方法的压气机落后角与总压损失预测代理模型[J]. 航空动力学报,2023,38(7): 1675-1690. MA Bowen,WU Xiaoxiong,YU Yang. Surrogate model for deviation angle and total pressure loss prediction of compressor based on machine learning methods[J]. Journal of Aerospace Power,2023,38(7): 1675-1690. (in Chinese MA Bowen, WU Xiaoxiong, YU Yang. Surrogate model for deviation angle and total pressure loss prediction of compressor based on machine learning methods[J]. Journal of Aerospace Power, 2023, 38(7): 1675-1690. (in Chinese)
[8]	唐天全. 压气机二维性能预测模型的研究[D]. 西安: 西北工业大学,2018. TANG Tianquan. Study on models of two dimensional performance prediction for compressor[D]. Xi’an: Northwestern Polytechnical University,2018 (in Chinese). TANG Tianquan. Study on models of two dimensional performance prediction for compressor[D]. Xi’an: Northwestern Polytechnical University, 2018 (in Chinese).
[9]	杜周,徐全勇,宋振寿,等. 基于深度学习的压气机叶型气动特性预测[J]. 航空动力学报,2023,38(9): 2251-2260. DU Zhou,XU Quanyong,SONG Zhenshou,et al. Prediction of aerodynamic characteristics of compressor blade profile based on deep learning[J]. Journal of Aerospace Power,2023,38(9): 2251-2260. (in Chinese DU Zhou, XU Quanyong, SONG Zhenshou, et al. Prediction of aerodynamic characteristics of compressor blade profile based on deep learning[J]. Journal of Aerospace Power, 2023, 38(9): 2251-2260. (in Chinese)
[10]	RAISSI M,PERDIKARIS P,KARNIADAKIS G E. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J]. Journal of Computational Physics,2019,378: 686-707. doi: 10.1016/j.jcp.2018.10.045
[11]	冯云阳,宋西镇,袁巍,等. 基于物理嵌入神经网络的叶栅损失模型[J]. 航空动力学报,2023,38(7): 1615-1625. FENG Yunyang,SONG Xizhen,YUAN Wei,et al. Physics-informed neural networks based cascade loss model[J]. Journal of Aerospace Power,2023,38(7): 1615-1625. (in Chinese FENG Yunyang, SONG Xizhen, YUAN Wei, et al. Physics-informed neural networks based cascade loss model[J]. Journal of Aerospace Power, 2023, 38(7): 1615-1625. (in Chinese)
[12]	AUNGIER R H. Axial-flow compressors: a strategy for aerodynamic design and analysis[M]. New York: American Society of Mechanical Engineers Press,2003.
[13]	HERRIG L,EMERY J,ERWIN J. Systematic two-dimensional cascade tests of NACA 65-series compressor blades at low speeds: 93R21643 [R]. Washington: National Aeronautics and Space Administration,1957.
[14]	GLOROT X,BORDES A,BENGIO Y. Deep sparse rectifier neural networks[J]. Journal of Machine Learning Research,2011,15: 315-323.
[15]	KINGMA D P,BA J. Adam: a method for stochastic optimization[EB/OL]. (2017-01-30) [2024-01-03]. http://arxiv.org/abs/1412.6980
[16]	HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al. Delving deep into rectifiers: surpassing human-level performance on ImageNet classification[C]//Proceedings of IEEE International Conference on Computer Vision. Piscataway,US: Institute of Electrical and Electronics Engineers,2015: 1026-1034.
[17]	MASTERS D,LUSCHI C. Revisiting small batch training for deep neural networks[EB/OL]. (2018-4-20) [2024-1-3]. https://arxiv.org/abs/1804.07612
[18]	LIEBLEIN S. Incidence and deviation-angle correlations for compressor cascades[J]. Journal of Basic Engineering,1960,82(3): 575-584. doi: 10.1115/1.3662666
[19]	陈鹏. 船用燃机低压压气机气动布局方法研究[D]. 哈尔滨: 哈尔滨工程大学,2021. CHEN Peng. Research on aerodynamic configuration method of low pressure compressor of marine gas turbine[D]. Harbin: Harbin Engineering University,2021. (in Chinese CHEN Peng. Research on aerodynamic configuration method of low pressure compressor of marine gas turbine[D]. Harbin: Harbin Engineering University, 2021. (in Chinese)
[20]	CARMODY R,CREVELING H F. Axial flow compressor computer program for calculating off design performance/Program 4/: 69N10873 [R]. Indianapolis,US: National Aeronautics and Space Administration,1968.
[21]	LIEBLEIN S. Loss and stall analysis of compressor cascades[J]. Journal of Basic Engineering,1959,81(3): 387-397. doi: 10.1115/1.4008481