文章快速检索 高级检索

1. 中国民航大学 航空工程学院, 天津 300300;
2. 中国民航大学 工程训练中心, 天津 300300

Aeroengine fault diagnosis based on multi-classification AdaBoost
CAO Huiling1, GAO Sheng1, XUE Peng2
1. College of Aeronautical Engineering, Civil Aviation University of China, Tianjin 300300, China;
2. Engineering Training Center, Civil Aviation University of China, Tianjin 300300, China
Received: 2017-12-13; Accepted: 2018-03-09; Published online: 2018-03-27 13:11
Foundation item: the Fundamental Research Funds for the Central Universities (3122014D010)
Corresponding author. CAO Huiling, E-mail:hlcao@cauc.edu.cn
Abstract: The data mining of aeroengine operational data is an important research for engine fault diagnosis. Due to the limitations of various algorithms, the accuracy of fault classification is difficult to be greatly enhanced with a single algorithm. Using a combination of classifications and diagnosis of multiple classification models, AdaBoost algorithm is a good method to improve the fault recognition accuracy. This paper combined the AdaBoost algorithm and its improved algorithm, and established a multi-classification AdaBoost algorithm. Support vector machine (SVM) was taken as the basic classifier, and a comprehensive diagnostic model was established. Fault identification data in statistics of fingerprint maps were processed with unit vector, ratio coefficient and correlation coefficient, and the training data for fault diagnosis with few effects of fault degrees were obtained. Then the model was constructed. The experimental results illustrate that the AdaBoost based combination algorithm can significantly improve the performance of classifier. With the actual fault cases, it is verified that the established diagnostic model can be well applied to engine fault diagnosis.
Keywords: AdaBoost     support vector machine (SVM)     unit vector     ratio coefficient     correlation coefficient     fault diagnosis

 (1)
 (2)
 (3)
 (4)

SVM基于统计学习理论，采用结构风险最小化，以提高学习机器泛化能力。在小样本、非线性、高维模式的情况下，能获得良好的统计规律，适用于航空发动机故障样本数目少的情况。在模式识别的问题中，SVM的基本思想是：寻找最大分类间隔的最优分类面。

 (5)
 (6)

 (7)
 (8)

 (9)

 (10)

1.3 发动机多分类故障诊断的流程设计

 图 2 发动机故障诊断的多分类AdaBoost算法流程图 Fig. 2 Flowchart of multi-classification AdaBoost algorithm for engine fault diagnosis

2 发动机故障诊断模型 2.1 数据预处理

2.1.1 数据来源

 图 3 发动机故障诊断指印图 Fig. 3 Fingerprint map for engine fault diagnosis

 故障序号 故障类别 ΔEGT/℃ ΔFF/% ΔN2/% ΔN1/% 1 +5℃ TAT -17.0 -1.4 -1.0 -1.0 2 -5℃ TAT 17.0 1.4 1.0 1.0 3 +0.02MACH 2.0 -2.2 -0.1 -0.1 4 -0.02MACH -2.0 2.2 0.1 0.1 5 +500 ALT 0 2.4 0 0 6 -500 ALT 0 -2.4 0 0 7 -2% HPC 12.0 1.6 0 0  24 -2% LPT -2.0 -2.1 0.7 -1.7

 故障序号 ΔEGT/℃ ΔFF/% ΔN2/% ΔN1/% ΔEGT/ΔFF 1 -0.993 -0.082 -0.058 -0.058 12 2 0.993 0.082 0.058 0.058 12 3 0.672 -0.739 -0.034 -0.034 -1 4 -0.672 0.739 0.034 0.034 -1 5 0 1 0 0 0 6 0 -1 0 0 0 7 0.991 0.132 0 0 8  24 -0.583 -0.612 0.204 -0.495 1

2.1.2 数据噪声的添加方法

1) 在表 1所示原始偏差数据加入一定程度的随机噪声，然后用比值系数法等方法处理后所得数据，作为训练和测试样本。诊断时，需要将实际参数的偏差数据进行比值方法等处理得到类似表 2中的转化数据，再进行诊断。

2) 在表 2的数据中，直接根据已经转化后的数据进行噪声添加。此时如果直接引入同一程度的随机误差，显然对各标识数据影响程度不同。因此应添加自身数值一定程度(比例)的偏差，来保证噪声数据一定程度也呈故障的线性比例。

 图 4 不同训练数据下单个SVM模型的正确率 Fig. 4 Accuracy of single SVM under different training data

 % 故障诊断模型 弱分类器最高正确率 应用AdaBoost算法后正确率 比值系数法 77.00 97.3 相关系数法 70.10 87.50 单位向量法 64.38 86.52 单位向量法(加入ΔEGT/ΔFF) 83.04 87.45

 图 5 M=50时各弱分类器的训练误差 Fig. 5 Training errors of each weak classifier when M=50

 图 6 不同模型诊断错误率随弱分类器个数的变化 Fig. 6 Variation of diagnosis error rate of different models with number of weak classifier

 故障序号 1 2 3 4 5 6 7 8 9 … 24 1 1 -1 -0.803 0.803 0.311 -0.311 -0.994 -0.995 -0.993 … 0.377  7 -0.994 0.994 0.734 -0.734 -0.208 0.208 1 0.999 0.999 … -0.435 8 -0.995 0.995 0.740 -0.740 -0.217 0.217 1 1 1 … -0.422 9 -0.994 0.994 0.734 -0.734 -0.208 0.208 0.999 1 1 … -0.405  24 0.377 -0.377 0.031 -0.031 -0.414 0.414 -0.435 -0.422 -0.405 … 1

2.3 案例诊断

 故障诊断模型 故障序号 案例1 案例2 案例3 比值系数法(1) 7 7 20 比值系数法(2) 7 7 1 相关系数法(2) 8 7 1 单位向量法(1) 7 7 1 单位向量法(2) 7 12 1 单位向量法(加入ΔEGT/ΔFF)(2) 7 7 2

 图 7 案例1的实际排故检测结果 Fig. 7 Actual detection and troubleshooting results of Instance 1

3 结论

 [1] WU X, KUMAR V. The top ten algorithms in data mining[M]. Boca Raton: CRC Press, 2009: 127-149. [2] 曹莹, 苗启广, 刘家辰, 等. AdaBoost算法研究进展与展望[J]. 自动化学报, 2013, 39 (6): 745–758. CAO Y, MIAO Q G, LIU J C, et al. Advance and prospects of AdaBoost algorithm[J]. Acta Automatica Sinica, 2013, 39 (6): 745–758. (in Chinese) [3] 徐启华, 杨瑞. 基于AdaBoost算法的故障诊断仿真研究[J]. 计算机工程与设计, 2005, 26 (12): 3210–3212. XU Q H, YANG R. Simulation research on fault diagnosis using AdaBoost algorithm[J]. Computer Engineering and Design, 2005, 26 (12): 3210–3212. DOI:10.3969/j.issn.1000-7024.2005.12.016 (in Chinese) [4] 夏利民, 戴汝为. 基于Boosting模糊分类的滚动轴承故障诊断[J]. 模式识别与人工智能, 2003, 16 (3): 323–327. XIA L M, DAI R W. Fault testing on rolling bearing based on Boosting fuzzy classification[J]. Pattern Recognition and Artificial Intelligence, 2003, 16 (3): 323–327. DOI:10.3969/j.issn.1003-6059.2003.03.011 (in Chinese) [5] 孙超英, 刘鲁, 刘传武. 基于Boosting-SVM算法的航空发动机故障诊断[J]. 航空动力学报, 2010, 25 (11): 2584–2588. SUN C Y, LIU L, LIU C W. Aero-engine fault diagnosis based on Boosting-SVM[J]. Journal of Aerospace Power, 2010, 25 (11): 2584–2588. (in Chinese) [6] 胡金海, 谢寿生, 蔡开龙, 等. Diverse AdaBoost-SVM分类方法及其在航空发动机故障诊断中的应用[J]. 航空学报, 2007, 28 (5): 1085–1090. HU J H, XIE S S, CAI K L, et al. Classification method of diverse AdaBoost-SVM and its application to fault diagnosis of aero-engine[J]. Acta Aeronautica et Astronautica Sinica, 2007, 28 (5): 1085–1090. DOI:10.3321/j.issn:1000-6893.2007.05.010 (in Chinese) [7] 胡金海, 骆广琦, 李应红, 等. 一种基于指数损失函数的多类分类AdaBoost算法及其应用[J]. 航空学报, 2008, 29 (4): 811–816. HU J H, LUO G Q, LI Y H, et al. An AdaBoost algorithm for multi-class classification based on exponential loss function and its application[J]. Acta Aeronautica et Astronautica Sinica, 2008, 29 (4): 811–816. DOI:10.3321/j.issn:1000-6893.2008.04.007 (in Chinese) [8] KEARNS M, VALIANT L. Cryptographic limitations on learning Boolean formulae and finite automata[J]. Journal of the ACM, 1994, 41 (1): 67–95. DOI:10.1145/174644.174647 [9] FREUND Y, SCHAPIRE R E. A decision-theoretic generalization of on-line learning and an application to Boosting[J]. Journal of Computer and System Sciences, 1997, 55 (1): 119–139. DOI:10.1006/jcss.1997.1504 [10] FREUND Y, SCHAPIRE R E.Experiments with a new Boosting algorithm[C]//Proceedings of the 13th Conference on Machine Learning, 1996: 148-156. [11] 李斌, 王紫石, 汪卫, 等. AdaBoost算法的一种改进方法[J]. 小型微型计算机系统, 2004, 25 (5): 869–871. LI B, WANG Z S, WANG W, et al. Enhancing method for AdaBoost[J]. Mini-Micro Systems, 2004, 25 (5): 869–871. DOI:10.3969/j.issn.1000-1220.2004.05.020 (in Chinese) [12] 廖红文, 周德龙. AdaBoost及其改进算法综述[J]. 计算机系统应用, 2012, 21 (5): 240–244. LIAO H W, ZHOU D L. Review of AdaBoost and its improvement[J]. Computer Systems & Applications, 2012, 21 (5): 240–244. (in Chinese) [13] ZHU J, ZOU H, ROSSET S, et al. Multi-class AdaBoost[J]. Statistics and Its Interface, 2009, 2 (3): 349–360. DOI:10.4310/SII.2009.v2.n3.a8 [14] 李应红, 尉询楷. 航空发动机的智能诊断、建模与预测方法[M]. 北京: 科学出版社, 2013: 82-85. LI Y H, WEI X K. The methods of aero-engine's modeling, intelligent fault diagnosis and prognosis[M]. Beijing: Science Press, 2013: 82-85. (in Chinese) [15] FRIEDMAN J, HASTIE T, TIBSHIRANI R. Additive logistic regression:A statistical view of boosting[J]. Annals of Statistics, 2000, 28 (2): 337–407. [16] 张卓.基于SVM多分类的PW4000故障诊断研究[D].天津: 中国民航大学, 2015: 42-49. ZHANG Z.Research on fault diagnosis of PW4000 based on SVM multi-classification[D].Tianjin: Civil Aviation University of China, 2015: 42-49(in Chinese). [17] 杨新武, 马壮, 袁顺. 基于弱分类器调整的多分类AdaBoost算法[J]. 电子与信息学报, 2016, 38 (2): 373–380. YANG X W, MA Z, YUAN S. Multi-class AdaBoost algorithm based on the adjusted weak classifier[J]. Journal of Electronics & Information Technology, 2016, 38 (2): 373–380. (in Chinese) [18] 曹惠玲, 庞思凯, 薛鹏, 等. 基于双发差异的航空发动机故障诊断方法研究[J]. 中国民航大学学报, 2014, 32 (3): 41–44. CAO H L, PANG S K, XUE P, et al. Research of aero-engine fault diagnosis method based on monitoring of twin differences[J]. Journal of Civil Aviation University of China, 2014, 32 (3): 41–44. DOI:10.3969/j.issn.1674-5590.2014.03.010 (in Chinese)

#### 文章信息

CAO Huiling, GAO Sheng, XUE Peng

Aeroengine fault diagnosis based on multi-classification AdaBoost

Journal of Beijing University of Aeronautics and Astronsutics, 2018, 44(9): 1818-1825
http://dx.doi.org/10.13700/j.bh.1001-5965.2017.0774