基于XGBoost的叶型表面转捩位置预测新方法

李昌林; 童歆; 虞培祥; 欧阳华

doi:10.13224/j.cnki.jasp.20220210

基于XGBoost的叶型表面转捩位置预测新方法

doi: 10.13224/j.cnki.jasp.20220210

李昌林^1,,
童歆¹,
虞培祥^{1, 2,*, ,},
欧阳华^{1, 2}

1.
上海交通大学机械与动力工程学院，上海 200240
2.
上海交通大学燃气轮机与民用航空发动机教育部工程研究中心，上海 200240

基金项目: 国家科技重大专项（2017-Ⅱ-0007-0021，J2019-Ⅱ-0005-0025）；中国联合重燃专项（19UGTC037）

详细信息

作者简介:
李昌林（1997-），男，硕士，主要从事压气机可控扩散叶型流场研究

通讯作者:
虞培祥（1986-），男，助理研究员，博士，主要从事计算流体力学及计算气动声学研究。 E-mail：pxyu@sjtu.edu.cn

中图分类号: V232.4
计量
- 文章访问数: 141
- HTML浏览量: 46
- PDF量: 27
- 被引次数: 0
出版历程
- 收稿日期: 2022-04-14
- 网络出版日期: 2024-06-29

New method for predicting the transition position of airfoil surface based on XGBoost model

LI Changlin^1
,,
TONG Xin¹,
YU Peixiang^{1, 2,*
, ,},
OUYANG Hua^{1, 2}

1.
School of Mechanical Engineering，Shanghai Jiao Tong University，Shanghai 200240，China
2.
Engineering Research Center of Gas Turbine and Civil Aero Engine，Ministry of Education，Shanghai Jiao Tong University，Shanghai 200240，China

摘要

摘要:
针对叶型表面边界层转捩问题，基于机器学习XGBoost模型的层流/湍流界面识别方法，建立了一种不依赖于指定阈值的预测转捩位置的新方法。在本方法中，根据大涡模拟计算得到的可控扩散叶型绕流的高精度流场，考虑到流动的间歇性，利用机器学习方法统计出不同时刻下边界层中不同位置处层流状态的比例，并依据其在叶型弦长方向上的变化率，得出转捩区域位置。通过对不同影响参数的考察，检验该方法的有效性，从而验证了其具有较好的通用性。与传统判据相比，本方法预测转捩位置准确，且在结果研判上不依赖于主观判断。此外，利用当前方法，发现对于可控扩散叶型，其边界层转捩除了受湍动能影响较大之外，也取决于涡量的大小及其空间分布。
- 转捩位置预测 /
- XGBoost模型 /
- 分离诱导转捩 /
- 可控扩散叶型 /
- 边界层
Abstract:
For identification of the transition position on the blade surface, a turbulence/non-turbulence interface identification method based on XGBoost model without specified thresholds was introduced. According to this method, the high-precision flow field around the controlled diffusion airfoil was solved by the large eddy simulation method. Considering the intermittent flow, the proportions of laminar flow state at different positions in the boundary layer at different times were calculated by the machine learning method, and the transition position was obtained according to the change rate in the chord length direction of the airfoil. The method was verified by investigating different influencing parameters. Compared with traditional criteria, this method could accurately predict transition positions without subjective judgment. In addition, using the present method, it was found that for a controlled diffusion airfoil, the boundary layer transition depended not only on the turbulent energy, but also on the size of vortices and the space distribution feature.
- prediction of transition position /
- XGBoost model /
- separation-induced transition /
- controlled diffusion airfoil /
- boundary layer

HTML

图 1 $ {Re} = 5 \times 1{0^5} $，$ \alpha = {0{\text{°}}} $层流/湍流识别结果

Figure 1. $ {Re} = 5 \times 1{0^5} $，$\alpha = {0{\text{°}}}$ laminar/turbulence recognition results

下载: 全尺寸图片幻灯片

图 2 不同弦长位置处的层流节点占比

Figure 2. Laminar flow nodes rates on different chord length position

下载: 全尺寸图片幻灯片

图 3 不同采样长度转捩位置识别结果

Figure 3. Transition position identification results with different sampling time spans

下载: 全尺寸图片幻灯片

图 4 不同采样长度25%弦长层流节点占比识别结果

Figure 4. Transition position identification results on 25% chord length with different time length

下载: 全尺寸图片幻灯片

图 5 不同采样分辨率转捩位置识别结果

Figure 5. Transition position identification results with different time resolution

下载: 全尺寸图片幻灯片

图 6 不同法向空间尺度转捩位置识别结果

Figure 6. Transition position identification results with different normal spatial scale

下载: 全尺寸图片幻灯片

图 7 不同流向空间尺度转捩位置识别结果

Figure 7. Transition position identification results with different streamwise spatial scale

下载: 全尺寸图片幻灯片

图 8 不同雷诺数工况转捩位置识别结果

Figure 8. Transition position identification results with different Reynolds number condition

下载: 全尺寸图片幻灯片

图 9 不同攻角工况转捩位置识别结果

Figure 9. Transition position identification results with different angle of attack condition

下载: 全尺寸图片幻灯片

图 10 判据累计增益

Figure 10. Total_Gains of different criterions

下载: 全尺寸图片幻灯片

图 11 不同雷诺数下判据权重

Figure 11. Criterion weight under different Reynolds numbers

下载: 全尺寸图片幻灯片

表 1 叶栅和叶型参数

Table 1. Cascade and airfoil parameters

变量	数值
安装角/（°）	41.95
几何进口角/（°）	59.61
弦长/mm	60
稠度	1.105
叶片最大无量纲厚度	0.079
叶型进气角/（°）	59.61
叶型折转角/（°）	26.84
叶型前缘曲率	4.0
叶型椭圆轴比	1.925

下载: 导出CSV

表 2 判据表达式

Table 2. Criterion expression

变量	表达式
$ X_{0} / （\mathrm{kg}\cdot {\mathrm{m}}^{2} / {\mathrm{s}}^{2}） $	$ k $
$ X_{1} / {\mathrm{s}}^{-2} $	$ I_{2} （S'） $
$ X_{2} / {\mathrm{s}}^{-2} $	$ I_{3} （S'） $
$ X_{3} / {\mathrm{s}}^{-2} $	$ I_{2} （\varOmega'） $
$ X_{4} /{\mathrm{ s}}^{-4} $	$ I_{2} （S' \cdot S'） $
$ X_{5} / {\mathrm{s}}^{-4} $	$ I_{3} （S'\cdot S'） $
$ X_{6} / {\mathrm{s}}^{-4} $	$ I_{2} （\varOmega'\cdot \varOmega'） $
$ X_{7} / {\mathrm{s}}^{-4} $	$ I_{2}\left(S^{\prime}\cdot \varOmega^{\prime}+\varOmega^{\prime}\cdot S^{\prime}\right) $

下载: 导出CSV

表 3 训练集流向空间尺度规划

Table 3. Streamwise training set spatial scale planning

组合	${L_{\mathrm{l}}} $/%	${L_{\mathrm{t}}}$/%	层流/湍流数据量比值/%
1	0～5	90～100	0.501
2	0～10	80～100	0.636
3	0～15	70～100	0.818
4	0～20	60～100	0.930
5	0～10	90～100	1.358
6	0～5	60～100	0.135

下载: 导出CSV

表 4 涡量判据在不同阈值下的识别结果

Table 4. Identification results of vorticity criterion under different thresholds %

识别结果	阈值
100	1
98	2
90	3
37	5
21	10
10	20
8	99

下载: 导出CSV

表 5 不同判据识别结果对比

Table 5. Comparison of identification results of different criteria

判据	识别结果/%	阈值/%
$ K $	33	5
涡量	37	5
$ \lambda_{2} $	35	5
$ Q $	40	98
$ \varPhi $	32	1
$ c_{p} $	33
XGBoost	30～35

下载: 导出CSV

表 6 对不同雷诺数的不同判据识别结果对比

Table 6. Comparison of identification results of different criteria for different Reynolds numbers

判据	识别结果/%			阈值/%
判据	Re=1×10⁵	Re=5×10⁵	Re=8×10⁵	阈值/%
$ K $	39	33	27	5
涡量	38	37	28	5
$ \lambda_{2} $	42	35	22	5
$ Q $	41	40	31	98
$ \varPhi $	32	32	26	1
$ c_{p} $	38	33	27
XGBoost	35～40	30～35	25～30

下载: 导出CSV

表 7 对不同攻角的不同判据识别结果对比

Table 7. Comparison of identification results of different criteria for different angles of attack

判据	识别结果/%				阈值/%
判据	α=−3°	α=0°	α=3°	α=6°	阈值/%
$ K $	37	33	28	27	5
涡量	42	37	21	16	5
$ \lambda_{2} $	37	35	24	21	5
$ Q $	40	40	30	22	98
$ \varPhi $	36	32	24	21	1
$ c_{p} $	37	33	25	识别失败
XGBoost	35～40	30～35	25～30	20～25

下载: 导出CSV

参考文献(23)

[1]	BISSET D K,HUNT J C R,ROGERS M M. The turbulent/non-turbulent interface bounding a far wake[J]. Journal of Fluid Mechanics,2002,451: 383-410. doi: 10.1017/S0022112001006759
[2]	WESTERWEEL J,FUKUSHIMA C,PEDERSEN J M,et al. Mechanics of the turbulent-nonturbulent interface of a jet[J]. Physical Review Letters,2005,95(17): 174501.1-174501.4.
[3]	JOCKSCH A,KLEISER L. Growth of turbulent spots in high-speed boundary layers on a flat plate[J]. International Journal of Heat and Fluid Flow,2008,29(6): 1543-1557. doi: 10.1016/j.ijheatfluidflow.2008.08.008
[4]	DA SILVA C B,PEREIRA J C F. Invariants of the velocity-gradient,rate-of-strain,and rate-of-rotation tensors across the turbulent/nonturbulent interface in jets[J]. Physics of Fluids,2008,20(5): 55101.1-55101.18.
[5]	BORRELL G,JIMÉNEZ J. Properties of the turbulent/non-turbulent interface in boundary layers[J]. Journal of Fluid Mechanics,2016,801: 554-596. doi: 10.1017/jfm.2016.430
[6]	LEE Jin,ZAKI T A. Detection algorithm for turbulent interfaces and large-scale structures in intermittent flows[J]. Computers & Fluids,2018,175: 142-158.
[7]	DE SILVA C M,PHILIP J,CHAUHAN K,et al. Multiscale geometry and scaling of the turbulent-nonturbulent interface in high Reynolds number boundary layers[J]. Physical Review Letters,2013,111(4): 044501.1-044501.5.
[8]	MENTER F R,LANGTRY R B,LIKKI S R,et al. A correlation-based transition model using local variables: Part Ⅰ model formulation[J]. Journal of Turbomachinery,2006,128(3): 413. doi: 10.1115/1.2184352
[9]	ANAND R K,BOERSMA B J,AGRAWAL A. Detection of turbulent/non-turbulent interface for an axisymmetric turbulent jet: evaluation of known criteria and proposal of a new criterion[J]. Experiments in Fluids,2009,47(6): 995-1007. doi: 10.1007/s00348-009-0695-5
[10]	MCMULLAN W A,PAGE G J. Large eddy simulation of a controlled-diffusion cascade blade at varying flow inlet angles[C]//Proceedings of the ASME Turbo Expo 2009: Power for Land,Sea,and Air. Orlando,Florida,US: ASME,2009: 453-466.
[11]	MCMULLAN A,PAGE G. Loss coefficient estimation in a controlled diffusion cascade using large eddy simulation[C]//Proceedings of the 48th AIAA Aerospace Sciences Meeting Including the New Horizons Forum and Aerospace Exposition. Reston,Virigina: AIAA,2010: 315.1-315.17.
[12]	MCMULLAN W A,PAGE G J. Large eddy simulation of a controlled diffusion compressor cascade[J]. Flow,Turbulence and Combustion,2011,86(2): 207-230. doi: 10.1007/s10494-010-9314-z
[13]	宋寅,奉凡,顾春伟. 间断Galerkin方法计算叶型转捩流动[J]. 航空动力学报,2013,28(5): 1057-1065. SONG Yin,FENG Fan,GU Chunwei. Transitional flow on turbomachine blade using discontinuous Galerkin method[J]. Journal of Aerospace Power,2013,28(5): 1057-1065. (in Chinese SONG Yin, FENG Fan, GU Chunwei. Transitional flow on turbomachine blade using discontinuous Galerkin method[J]. Journal of Aerospace Power, 2013, 28(5): 1057-1065. (in Chinese)
[14]	曾瑞慧,高丽敏,杨冠华,等. 层流叶片在压气机中的应用研究[J]. 工程热物理学报,2017,38(11): 2348-2356. ZENG Ruihui,GAO Limin,YANG Guanhua,et al. Application research of laminar blade in compressor[J]. Journal of Engineering Thermophysics,2017,38(11): 2348-2356. (in Chinese ZENG Ruihui, GAO Limin, YANG Guanhua, et al. Application research of laminar blade in compressor[J]. Journal of Engineering Thermophysics, 2017, 38(11): 2348-2356. (in Chinese)
[15]	施恒涛,刘宝杰,于贤君. 基于多项式的曲率连续前缘造型方法及应用[J]. 航空动力学报,2020,35(2): 397-409. SHI Hengtao,LIU Baojie,YU Xianjun. Polynomial-based continuous-curvature leading edge design method and its application[J]. Journal of Aerospace Power,2020,35(2): 397-409. (in Chinese SHI Hengtao, LIU Baojie, YU Xianjun. Polynomial-based continuous-curvature leading edge design method and its application[J]. Journal of Aerospace Power, 2020, 35(2): 397-409. (in Chinese)
[16]	张子良,张明明. 仿生减阻翼型的气动性能[J]. 航空动力学报,2021,36(8): 1740-1748. ZHANG Ziliang,ZHANG Mingming. Aerodynamic performance for bionic drag-reducing airfoil[J]. Journal of Aerospace Power,2021,36(8): 1740-1748. (in Chinese ZHANG Ziliang, ZHANG Mingming. Aerodynamic performance for bionic drag-reducing airfoil[J]. Journal of Aerospace Power, 2021, 36(8): 1740-1748. (in Chinese)
[17]	郭志远. 基于大涡模拟的可控扩散叶型转捩问题研究[D]. 上海: 上海交通大学,2020. GUO Zhiyuan. The study of transition on controlled diffution airfoil using large simulation[D]. Shanghai: Shanghai Jiao Tong University,2020. (in Chinese GUO Zhiyuan. The study of transition on controlled diffution airfoil using large simulation[D]. Shanghai: Shanghai Jiao Tong University, 2020. (in Chinese)
[18]	CHEN Tianqi,GUESTRIN C. XGBoost: a scalable tree boosting system[EB/OL]. (2016-08-13)[2022-04-01]. https://doi.org/10.1145/2939672.2939785
[19]	LI Binglin,YANG Zixuan,ZHANG Xing,et al. Using machine learning to detect the turbulent region in flow past a circular cylinder[J]. Journal of Fluid Mechanics,2020,905: A10.1-A10.26.
[20]	王会社,钟兢军,王仲奇. 多级压气机中可控扩散叶型研究的进展与展望第二部分可控扩散叶型的实验与数值模拟[J]. 航空动力学报,2002,17(1): 16-22. WANG Huishe,ZHONG Jingjun,WANG Zhongqi. Development of controlled diffusion airfoils for multistage compressor application part 2 test and numerical simulation of controlled diffusion airfoils[J]. Journal of Aerospace Power,2002,17(1): 16-22. (in Chinese doi: 10.3969/j.issn.1000-8055.2002.01.002 WANG Huishe, ZHONG Jingjun, WANG Zhongqi. Development of controlled diffusion airfoils for multistage compressor application part 2 test and numerical simulation of controlled diffusion airfoils[J]. Journal of Aerospace Power, 2002, 17(1): 16-22. (in Chinese) doi: 10.3969/j.issn.1000-8055.2002.01.002
[21]	童歆,羌晓青,虞培祥,等. 基于曲率分布控制的叶型前缘设计方法[J]. 航空学报,2021,42(7): 124712. TONG Xin,QIANG Xiaoqing,YU Peixiang,et al. Leading edge design method based on curvature distribution control[J]. Acta Aeronautica et Astronautica Sinica,2021,42(7): 124712. (in Chinese TONG Xin, QIANG Xiaoqing, YU Peixiang, et al. Leading edge design method based on curvature distribution control[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(7): 124712. (in Chinese)
[22]	REHILL B,WALSH E J,BRANDT L,et al. Identifying turbulent spots in transitional boundary layers[J]. Journal of Turbomachinery,2013,135(1): 011019. doi: 10.1115/1.4006395
[23]	SCHLICHTING H,GERSTEN K. Boundary-layer theory[M]. Karlsruhe: Springer Science & Business Media,2003.