-
摘要:
航空发动机在使用寿命周期内会不断磨损最终出现故障，通过对发动机油液监测铁谱分析数据的挖掘可实现磨损故障的诊断。本文研究免疫算法优化的支持向量机（SVM）在航空发动机磨损故障诊断中的运用。首先，总结了支持向量机和免疫算法的运行流程和关键算法。然后，用改进的免疫算法优化支持向量机惩罚因子、松弛变量及核函数参数。某型航空发动机的油液铁谱分析数据和加入噪声数据验证结果表明，该方法可有效实现航空发动机磨损故障诊断且具有较好的鲁棒性。最后，研究了核函数、多分类决策方法、初始种群大小、亲和力计算公式、支持向量机优化方法和归一化方法对磨损故障诊断准确率的影响，得到了最佳诊断方法。
-
关键词:
- 航空发动机 /
- 磨损故障诊断 /
- 铁谱分析 /
- 免疫算法 /
- 支持向量机（SVM）
Abstract:Aircraft engine is wearing during its service life and will finally break down. The wear fault can be diagnosed by analyzing the ferrography data of oil monitoring. The use of immune algorithm optimized support vector machine (SVM) in aircraft engine wear fault diagnosis was researched in this paper. First, the process and algorithm of SVM and immune algorithm were summarized. Then, the optimization of SVM's penalty factor, slack variable and kernel function parameters by immune algorithm was researched. The verification results of an engine's oil ferrography analysis data and adding noise data show that the method can effectively diagnose the aircraft engine wear fault and has good robustness. Finally, the impact of kernel function, multi-classification decision method, initial population size, affinity calculation formula, optimization algorithm and normalization method on diagnosis accuracy was analyzed, and the best algorithm was achieved.
-
图 6 初始种群个数对检测器成熟时间和诊断准确率的影响
Figure 6. Impact of initial population number on detector maturation time and accuracy
表 1 样本数据
Table 1. Sample data
状态 编号 层状磨粒 疲劳剥块 严重滑动磨粒 正常(Ⅰ) 1 0.0298 0.0975 0.0741 2 0.0021 0.1429 0.0941 ⋮ ⋮ ⋮ ⋮ 49 0.0025 0.0539 0.0157 50 0.0311 0.1375 0.1223 轴承疲劳磨损(Ⅱ) 1 0.5015 0.4725 0.0495 2 0.4627 0.5026 0.0581 ⋮ ⋮ ⋮ ⋮ 49 0.6005 0.5780 0.1854 50 0.4219 0.4275 0.1721 齿轮过载疲劳(Ⅲ) 1 0.0067 0.7018 0.1182 2 0.0132 0.6125 0.0094 ⋮ ⋮ ⋮ ⋮ 49 0.0147 0.6251 0.2142 50 0.0024 0.5276 0.0241 齿轮胶合或擦伤(Ⅳ) 1 0.0072 0.0089 0.8053 2 0.0261 0.1055 0.7218 ⋮ ⋮ ⋮ ⋮ 49 0.0179 0.1403 0.6732 50 0.0188 0.0852 0.7829 
下载: 导出CSV
表 2 支持向量机最优参数向量、迭代次数及准确率
Table 2. Optimal parameter vector, iteration times and accuracy of SVM
分类 最优参数向量v=(C, ξ, σ) 迭代次数 准确率/% Ⅰ/Ⅱ (3.58, 0.41, 1.88) 30 98.3 Ⅰ/Ⅲ (14.63, 5.01, 9.25) 27 99.5 Ⅰ/Ⅳ (7.47, 11.12, 1.84) 35 97.9 Ⅱ/Ⅲ (35.48, 18.12, 4.96) 68 94.4 Ⅱ/Ⅳ (42.03, 7.15, 2.48) 54 93.9 Ⅲ/Ⅳ (27.50, 15.42, 4.83) 73 95.2
下载: 导出CSV
表 3 测试样本量及准确率
Table 3. Number and diagnostic accuracy of testing sample
样本量 100 200 300 400 500 600 准确率/% 98 98.5 98.3 98.25 97.8 98.1
下载: 导出CSV
表 4 不同核函数准确率
Table 4. Accuracy of different kernel functions
核函数 公式 准确率/% 线性核函数 K(x, xi)=xTxi 94.8 多项式核函数 K(x, xi)=(x·xi+1)d 95.5 感知器核函数 K(x, xi)=tanh(βxi+b) 95.2 高斯径向基核函数 K(x, xi)=exp(-||x-xi2/2σ2) 98.3 注：d、β和b代表系数和常数。
下载: 导出CSV
表 5 不同多分类决策方法准确率
Table 5. Accuracy of different multi-classification decision methods
多分类决策方法 向量机个数 运算时间/s 准确率/% 一对多 4 11.07 97.6 一对一 6 12.45 98.3 DAGSSVM 3 10.36 93.5
下载: 导出CSV
表 6 亲和力计算方法对诊断准确率的影响
Table 6. Impact of affinity calculation method on accuracy
计算方法 计算公式 准确率/% 欧氏距离 D1=[(x1-x2)2+(y1-y2)2+ (z1-z2)2)1/2 95.7 曼哈顿距离 D2=|x1-x2|+|y1-y2|+|z1-z2| 90.4 切比雪夫距离 D3=max(|x1-x2|, |y1-y2|, |z1-z2|) 80.2 本文方法 D4=1/(1+D1) 98.3
下载: 导出CSV
表 7 不同算法优化性能
Table 7. Optimization performance of different algorithms
优化算法 优化时间/s 准确率/% 遗传算法 17.98 98.2 粒子群算法 11.58 96.4 交叉验证 9.83 92.9 免疫算法 12.45 98.3
下载: 导出CSV
表 8 不同归一化方法对诊断准确率的影响
Table 8. Impact of different normalization methods on accuracy
归一化区间 归一化方式 准确率/% (-∞，+∞) 不归一化 80.9 [-1, 1] (x-xmean)/(xmax-xmin) 97.4 (x-xmean)/xvar 94.1 [0, 1] x/xmax 82.2 (x-xmin)/(xmax-xmin) 98.3
下载: 导出CSV
-
[1] 李艳军, 左洪福, 吴振锋.基于磨粒分析方法的发动机磨损故障智能诊断技术[J].南京航空航天大学学报, 2001, 33(3):221-226. http://www.cnki.com.cn/Article/CJFDTOTAL-NJHK200103003.htmLI Y J, ZUO H F, WU Z F.Intelligent diagnostics for engine wear failure based on debris analysis[J]. Journal of Nanjing University of Aeronautics and Astronautics, 2001, 33(3):221-226(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-NJHK200103003.htm [2] 王汉功, 陈桂明.铁谱图像分析理论与技术[M].北京:科学出版社, 2005:245.WANG H G, CHEN G M.Ferro graphic image analysis theory and technology[M]. Beijing:Science Press, 2005:245(in Chinese). [3] 李民赞.光谱分析技术及其应用[M].北京:科学出版社, 2006:1.LI M Z.Spectral analysis technique and application[M]. Beijing:Science Press, 2006:1(in Chinese). [4] 吴振锋, 左洪福, 孙有朝.航空发动机磨损故障的常用监控手段及其对比[J].航空工程与维修, 2000, 60(5):25-26. http://www.cnki.com.cn/Article/CJFDTOTAL-KONG200005010.htmWU Z F, ZUO H F, SUN Y C.The common monitoring techniques of aero engine and their comparison[J]. Aviation Maintenance & Engineering, 2000, 60(5):25-26(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-KONG200005010.htm [5] 邓明, 金业壮.航空发动机故障诊断[M].北京:北京航空航天大学出版社, 2012:187-191.DENG M, JIN Y Z.The aircraft engine fault diagnosis[M]. Beijing:Beihang University Press, 2012:187-191(in Chinese). [6] 杨云, 朱家元, 张恒喜.基于新型机器学习的电子装备系统智能故障诊断研究[J].计算机工程与应用, 2003, 39(22):210-213. doi: 10.3321/j.issn:1002-8331.2003.22.068YANG Y, ZHU J Y, ZHANG H X.Electronic equipment systems intelligent fault diagnosis based on new machine learning approach[J]. Computer Engineering and Applications, 2003, 39(22):210-213(in Chinese). doi: 10.3321/j.issn:1002-8331.2003.22.068 [7] 孙铁轩, 邵春福, 计寻, 等.基于ARIMA与信息粒化SVR组合模型的交通事故时序预测[J].清华大学学报(自然科学版), 2014, 54(3):348-354. http://www.cnki.com.cn/Article/CJFDTOTAL-QHXB201403011.htmSUN T X, SHAO C F, JI X, et al.Urban traffic accident time series prediction model based on combination of ARIMA and information granulation SVR[J]. Journal of Tsinghua University(Science and Technology), 2014, 54(3):348-354(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-QHXB201403011.htm [8] 王旭辉, 黄圣国, 曹力, 等.基于LS-SVM的航空发动机气路参数趋势在线预测[J].吉利大学学报(工学版), 2008, 38(1):239-244. http://www.cnki.com.cn/Article/CJFDTOTAL-JLGY200801048.htmWANG X H, HUANG S G, CAO L, et al.LS-SVM based online trend prediction of gas path parameters of aero engine[J]. Journal of Jilin University (Engineering and Technology Edition), 2008, 38(1):239-244(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-JLGY200801048.htm [9] 陈立波, 宋兰琪, 宋科, 等.基于支持向量机的航空发动机磨损趋势预测[J].润滑与密封, 2008, 33(5):84-87. http://www.cnki.com.cn/Article/CJFDTOTAL-RHMF200805025.htmCHEN L B, SONG L Q, SONG K, et al.Wear trend forecast of aviation engine based on support vector machine model[J]. Lubrication Engineering, 2008, 33(5):84-87(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-RHMF200805025.htm [10] 张周锁, 李凌均, 何正嘉.基于支持向量机的机械故障诊断方法研究[J].西安交通大学学报, 2002, 36(2):1303-1306. http://cdmd.cnki.com.cn/Article/CDMD-10407-2010020205.htmZHANG Z S, LI L J, HE Z J.Research on diagnosis method of machinery fault based on support vector machine[J]. Journal of Xi'an Jiaotong University, 2002, 36(2):1303-1306(in Chinese). http://cdmd.cnki.com.cn/Article/CDMD-10407-2010020205.htm [11] SHIN H J, EOM D H, KIM S S.One-class support vector machines an application in machine fault detection and classification[J]. Computers & Industrial Engineering, 2005, 48(2):395-408. [12] 徐启华, 师军.基于支持向量机的航空发动机故障诊断[J].航空动力学报, 2005, 20(2):298-302. http://cdmd.cnki.com.cn/Article/CDMD-10143-1015534019.htmXU Q H, SHI J.Aero-engine fault diagnosis based on support vector machine[J]. Journal of Aerospace Power, 2005, 20(2):298-302(in Chinese). http://cdmd.cnki.com.cn/Article/CDMD-10143-1015534019.htm [13] 孙超英, 刘鲁, 刘传武, 等.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. http://www.cnki.com.cn/Article/CJFDTOTAL-HKDI201011026.htmSUN C Y, LIU L, LIU C W, et al.Aero-engine fault diagnosis based on Boosting-SVM[J]. Journal of Aerospace Power, 2010, 25(11):2584-2588(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-HKDI201011026.htm [14] 莫宏伟.人工免疫系统原理与应用[M].哈尔滨:哈尔滨工业大学出版社, 2003:48-57.MO H W.Principle and application of artificial immune system[M]. Harbin:Harbin Institute of Technology Press, 2003:48-57(in Chinese). [15] ENDOH S, TOMA N, YAMADA K.Immune algorithm for n-TSP[C]//IEEE International Conference on Systems, Man and Cybernetics.Piscataway, NJ:IEEE Press, 1998, 4:3844-3849. [16] SASAKI M, KAWAFUKU M, TAKAHASHI K.An immune feedback mechanism based adaptive learning of neural network controller[C]//ICONIP'99, 6th International Conference on Neural Information Processing.Piscataway, NJ:IEEE Press, 1999, 2:502-507. [17] 蒋加伏, 陈荣元, 唐贤瑛, 等.基于免疫-蚂蚁算法的多约束QoS路由选择[J].通信学报, 2004, 25(8):89-95. http://www.cnki.com.cn/Article/CJFDTOTAL-TXXB200408011.htmJIANG J F, CHEN R Y, TANG X Y, et al.A multiple constrained QoS routing based on immune-ant algorithm[J]. Journal of Communications, 2004, 25(8):89-95(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-TXXB200408011.htm [18] 颜瑞, 张祥.基于改进免疫算法优化支持向量机的钢材消费预测[J].工业工程, 2013, 16(5):90-95. http://www.cnki.com.cn/Article/CJFDTOTAL-GDJX201305017.htmYAN R, ZHANG X.Forecasting of steel demands by using support vector machine and immune algorithm[J]. Industrial Engineering, 2013, 16(5):90-95(in Chinese). http://www.cnki.com.cn/Article/CJFDTOTAL-GDJX201305017.htm [19] 高文军. 基于人工免疫算法优化支持向量机的电力变压器故障诊断研究[D]. 太原: 太原理工大学, 2012.GAO W J.Study on fault diagnosis foe power transformer based on support vector machine of artificial immune algorithm[D]. Taiyuan:Taiyuan University of Technology, 2012(in Chinese). -
下载:
点击查看大图
计量
- 文章访问数: 889
- HTML全文浏览量: 203
- PDF下载量: 482
- 被引次数: 0
