基于LMKL和OC-ELM的航空电子部件故障检测方法

doi:10.3969/j.issn.1001-506X.2020.06.29

摘要/Abstract

摘要：

针对航空电子部件故障样本获取困难以及检测准确率不高的问题,提出基于局部多核学习(localized multiple kernel learning, LMKL)和一类超限学习机(one-class extreme learning machine, OC-ELM)的故障检测方法。仅运用正常状态的小样本数据,给出了LMK-OC-ELM的数学表达形式,并在不同的门模型下推导了LMK-OC-ELM中局部核权重的优化方法;在获取局部核权重的基础上,定义了离线故障检测所需的统计检验量与阈值,以便工程实现。将所提方法应用于某型接收机,结果表明，在训练时间可控的前提下,与4种常见的一类分类(one-class classification, OCC)算法相比,所提方法可均衡地提高召回率、查准率和特异度,以LMK-OC-ELM-sig为代表,其在F1、曲线下方面积(area under curve, AUC)、G-mean和准确率4个指标上,比最近提出的局部多核异常检测(localized multiple kernel anomaly detection, LMKAD)方法分别提高了1.60%、1.57%、1.53%和2.23%。

关键词: 超限学习机, 局部多核学习, 一类分类, 故障检测

Abstract:

In consideration of the difficulty in acquiring real fault samples and the low detection accuracy of the avionics, a fault detection method based on localized multiple kernel learning (LMKL) and one-class extreme learning machine (OC-ELM) is proposed. The mathematical expression of LMK-OC-ELM is given onlyby using small sample data in normal state, and the localized kernel weights in LMK-OC-ELM is deduced under different gating models. On the basis of obtaining the localized kernel weights, the test statistic and threshold required for offline fault detection are defined to facilitate the engineering implementation. The proposed method is applied to the receiver. On the premise of controllable training time, the proposed method can improve recall, precision and specificity equitably compared with the other four common one-class classification algorithms. Taking LMK-OC-ELM-sig as the representative, the indicators of F1, area under curve (AUC), G-mean and accuracy are increased by 1.60%, 1.57%, 1.53% and 2.23% respectively compared with the localized multiple kernel anomaly detection (LMKAD) which is recently proposed.

Key words: extreme learning machine (ELM), localized multiple kernel learning (LMKL), one-class classification (OCC), fault detection

中图分类号:

V243

朱敏, 刘奇, 刘星, 许晴. 基于LMKL和OC-ELM的航空电子部件故障检测方法[J]. 系统工程与电子技术, 2020, 42(6): 1424-1432.

Min ZHU, Qi LIU, Xing LIU, Qing XU. Fault detection method for avionics based on LMKL and OC-ELM[J]. Systems Engineering and Electronics, 2020, 42(6): 1424-1432.

图/表 10

表1

表2

各OCC在UCI数据集上的实验结果"

数据集	算法	F1	AUC	G-mean	准确率	参数取值
Ecoli	SVDD	0.448 3	0.807 9	0.601 8	0.920 2	σ=3.439 5
	PCA	0.342 5	0.774 1	0.421 3	0.736 5	n=3
	OC-SVM	0.459 8	0.816 2	0.520 6	0.797 7	γ=0.047 5
	OC-KELM	0.778 4	0.903 2	0.780 3	0.960 5	C=1, σ=3.223 5
	LMKAD	0.594 8	0.857 8	0.633 4	0.865 8	-
	LMK-OC-ELM-so	0.782 4	0.896 3	0.783 4	0.962 1	C=0.1
	LMK-OC-ELM-r	0.564 3	0.829 4	0.605 1	0.817 1	C=0.001
	LMK-OC-ELM-sig	0.793 1	0.894 2	0.793 4	0.964 7	C=0.1
Sonar	SVDD	0.584 2	0.623 6	0.616 0	0.559 5	σ=17.438 5
	PCA	0.532 2	0.551 4	0.569 2	0.476 0	n=3
	OC-SVM	0.570 3	0.617 9	0.593 9	0.570 4	γ=0.002 5
	OC-KELM	0.513 3	0.483 2	0.575 2	0.352 0	C=0.01, σ=8.439 1
	LMKAD	0.573 1	0.621 3	0.594 3	0.583 9	-
	LMK-OC-ELM-so	0.598 0	0.679 1	0.601 4	0.691 4	C=10
	LMK-OC-ELM-r	0.542 0	0.582 3	0.569 9	0.560 9	C=1
	LMK-OC-ELM-sig	0.605 9	0.648 9	0.638 6	0.528 0	C=1
Diabetes	SVDD	0.686 0	0.601 7	0.710 4	0.590 3	σ=3.104 0
	PCA	0.667 5	0.572 6	0.693 2	0.560 7	n=6
	OC-SVM	0.662 6	0.575 8	0.684 7	0.565 1	γ=0.007
	OC-KELM	0.678 7	0.608 5	0.696 8	0.599 1	C=0.1, σ=2.249 2
	LMKAD	0.672 1	0.614 7	0.684 3	0.610 0	-
	LMK-OC-ELM-so	0.690 8	0.607 4	0.715 6	0.596 0	C=0.1
	LMK-OC-ELM-r	0.692 0	0.607 1	0.717 7	0.595 5	C=0.1
	LMK-OC-ELM-sig	0.702 3	0.627 6	0.722 8	0.623 6	C=0.01
Liver	SVDD	0.414 1	0.527 0	0.474 5	0.394 3	σ=2.817 9
	PCA	0.422 1	0.531 7	0.490 1	0.377 9	n=5
	OC-SVM	0.411 7	0.508 2	0.487 6	0.330 0	γ=0.012 6
	OC-KELM	0.421 0	0.522 6	0.497 6	0.344 3	C=0.01, σ=1.878 6
	LMKAD	0.419 9	0.530 1	0.486 4	0.380 5	-
	LMK-OC-ELM-so	0.422 2	0.524 2	0.498 9	0.345 8	C=0.001
	LMK-OC-ELM-r	0.421 5	0.527 4	0.493 2	0.362 9	C=0.01
	LMK-OC-ELM-sig	0.429 0	0.540 0	0.500 4	0.378 1	C=0.1
Spectf	SVDD	0.372 4	0.739 3	0.441 7	0.688 3	σ=9.644 2
	PCA	0.386 0	0.721 5	0.428 7	0.750 9	n=3
	OC-SVM	0.375 4	0.727 7	0.427 7	0.724 5	γ=0.013 1
	OC-KELM	0.370 5	0.752 9	0.453 6	0.658 1	C=0.01, σ=5.526 6
	LMKAD	0.414 4	0.779 3	0.461 0	0.723 4	-
	LMK-OC-ELM-so	0.413 2	0.747 0	0.457 1	0.768 9	C=1
	LMK-OC-ELM-r	0.404 8	0.741 8	0.450 7	0.756 8	C=0.1
	LMK-OC-ELM-sig	0.426 9	0.752 5	0.467 3	0.778 7	C=0.1
Abalone	SVDD	0.449 8	0.659 3	0.453 6	0.744 1	σ=3.672 1
	PCA	0.478 9	0.715 5	0.544 6	0.597 1	n=7
	OC-SVM	0.465 1	0.700 0	0.531 3	0.580 7	γ=0.010 8
	OC-KELM	0.452 7	0.686 0	0.520 3	0.563 1	C=1, σ=1.419 9
	LMKAD	0.500 8	0.736 4	0.561 0	0.632 0	-
	LMK-OC-ELM-so	0.525 6	0.732 2	0.548 4	0.730 9	C=10^-5
	LMK-OC-ELM-r	0.462 9	0.697 0	0.528 0	0.581 1	C=0.000 1
	LMK-OC-ELM-sig	0.506 4	0.737 1	0.562 2	0.643 3	C=10^-5
Breast	SVDD	0.882 2	0.894 7	0.888 4	0.898 6	σ=0.941 8
	PCA	0.873 4	0.876 6	0.873 5	0.876 2	n=7
	OC-SVM	0.924 4	0.929 0	0.925 6	0.930 7	γ=0.031 6
	OC-KELM	0.934 9	0.939 0	0.936 9	0.941 2	C=0.1, σ=0.638 9
	LMKAD	0.937 8	0.941 4	0.939 6	0.943 6	-
	LMK-OC-ELM-so	0.945 1	0.948 2	0.946 6	0.950 0	C=0.000 1
	LMK-OC-ELM-r	0.925 0	0.930 3	0.927 6	0.932 9	C=0.000 1
	LMK-OC-ELM-sig	0.945 8	0.948 8	0.947 2	0.950 7	C=10^-6

表2

图1

图2

图3

图4

图5

表3

图6

图7

参考文献 23

1	孙伟超, 李文海, 李文峰. 融合粗糙集与D-S证据理论的航空装备故障诊断[J]. 北京航空航天大学学报, 2015, 41 (10): 1902- 1909.
	SUN W C , LI W H , LI W F . Avionic devices fault diagnosis based on fusion method of rough set and D-S theory[J]. Journal of Beijing University of Aeronautics and Astronautics, 2015, 41 (10): 1902- 1909.
2	JAMIL T , MOHAMMED I . Simulation of VICTOR algorithm for fault diagnosis of digital circuits[J]. International Journal of Computer Theory and Engineering, 2015, 7 (2): 103- 107. doi: 10.7763/IJCTE.2015.V7.939
3	SMART E, BROWN D, AXEL-BERG L. Comparing one and two class classification methods for multiple fault detection on an induction motor[C]//Proc.of the IEEE Symposium on Industrial Electronics & Applications, 2013: 132-137.
4	IQBAL R , MANIAK T , DOCTOR F , et al. Fault detection and isolation in industrial processes using deep learning approaches[J]. IEEE Trans.on Industrial Informatics, 2019, 15 (5): 3077- 3084. doi: 10.1109/TII.2019.2902274
5	SCHOLKOPF B , PLATT J C , SHAWE-TAYLOR J , et al. Estimating the support of a high-dimensional distribution[J]. Neural Computation, 2001, 13 (7): 1443- 1471. doi: 10.1162/089976601750264965
6	戴海发, 卞鸿巍, 马恒, 等. 基于一类SVM的综合导航系统信息故障检测方法[J]. 中国惯性技术学报, 2017, 25 (4): 555- 560.
	DAI H F , BIAN H W , MA H , et al. Information fault detection for integrated navigation systems using one-class support vector machine[J]. Journal of Chinese Inertial Technology, 2017, 25 (4): 555- 560.
7	TAX D M J , DUIN R P W . Support vector data description[J]. Machine Learning, 2004, 54 (1): 45- 66. doi: 10.1023/B:MACH.0000008084.60811.49
8	NAVARRO-ACOSTA J A , RESéNDIZ-FLORES E O . Gompertz binary particle swarm optimization and support vector data description system for fault detection and feature selection applied in automotive pedals components[J]. International Journal of Advanced Manufacturing Technology, 2017, 92, 2311- 2314. doi: 10.1007/s00170-017-0333-y
9	MANSOURI M , NOUNOU M , NOUNOU H , et al. Kernel PCA-based GLRT for nonlinear fault detection of chemical processes[J]. Journal of Loss Prevention in the Process Industries, 2016, 40, 334- 347. doi: 10.1016/j.jlp.2016.01.011
10	HUANG G B , ZHOU H , DING X , et al. Extreme learning machine for regression and multiclass classification[J]. IEEE Trans.on Systems, Man, and Cybernetics, Part B: Cybernetics, 2011, 42 (2): 513- 529. doi: 10.1109/tsmcb.2011.2168604
11	LENG Q, QI H, MIAO J, et al. One-class classification with extreme learning machine[EB/OL].[2019-11-04]. https://www.hindawi.com/journals/mpe/2015/412957.
12	IOSIFIDIS A , MYGDALIS V , TEFAS A , et al. One-class classification based on extreme learning and geometric class information[J]. Neural Processing Letters, 2017, 45 (2): 577- 592. doi: 10.1007/s11063-016-9541-y
13	DAI H , CAO J , WANG T , et al. Multilayer one-class extreme learning machine[J]. Neural Networks: the Official Journal of the International Neural Network Society, 2019, 115, 11- 22. doi: 10.1016/j.neunet.2019.03.004
14	GAUTAM C , TIWARI A , LENG Q . On the construction of extreme learning machine for online and offline one-class classification-an expanded toolbox[J]. Neurocomputing, 2017, 261, 126- 143. doi: 10.1016/j.neucom.2016.04.070
15	邴其春, 龚勃文, 杨兆升, 等. 一种组合核相关向量机的短时交通流局域预测方法[J]. 哈尔滨工业大学学报, 2017, 49 (3): 150- 155.
	BING Q C , GONG B W , YANG Z S , et al. A short-term traffic flow local prediction method of combined kernel function relevance vector machines[J]. Journal of Harbin Institute of Technology, 2017, 49 (3): 150- 155.
16	李军, 李大超. 基于优化核极限学习机的风电功率时间序列预测[J]. 物理学报, 2016, 65 (13): 39- 48.
	LI J , LI D C . Wind power time series prediction using optimized kernel extreme learning machine method[J]. Acta Physica Sinica, 2016, 65 (13): 39- 48.
17	FENG J , JIAO L C , SUN T , et al. Multiple kernel learning based on discriminative kernel clustering for hyperspectral band selection[J]. IEEE Trans.on Geosciences and Remote Sensing, 2016, 54 (11): 6516- 6530. doi: 10.1109/TGRS.2016.2585961
18	GONEN M , ALPAYDIN E . Multiple kernel learning algorithms[J]. Journal of Machine Learning Research, 2011, 12, 2211- 2268.
19	GONEN M , ALPAYDIN E . Localized algorithms for multiple kernel learning[J]. Pattern Recognition, 2013, 46 (3): 795- 807. doi: 10.1016/j.patcog.2012.09.002
20	TAX D M J, MULLER K R. A consistency-based model selection for one-class classification[C]//Proc.of the 17th International Conference on Pattern Recognition, 2004: 363-366.
21	TAX D M J. DDtools, the data description toolbox for Matlab[EB/OL].[2013-07-24]. http://prlab.tudelft.nl/david-tax/dd_tools.html.
22	BISHOP C M . Neural networks for pattern recognition[M]. Oxford: Oxford University Press, 1995.
23	GAUTAM C , BALAJI R , SUDHARSAN K , et al. Localized multiple kernel learning for anomaly detection: one-class classification[J]. Knowledge-Based Systems, 2019, 165, 241- 252. doi: 10.1016/j.knosys.2018.11.030

算法	F1	AUC	G-mean	准确率	平均训练时间/s	平均测试时间/s
SVDD	0.922 2±0.030 7	0.885 4±0.028 8	0.923 8±0.028 8	0.886 5±0.039 8	0.031 4	0.006 9
PCA	0.918 5±0.020 9	0.912 6±0.018 2	0.921 0±0.019 7	0.883 5±0.028 0	0.060 7	0.004 4
OC-SVM	0.922 8±0.028 6	0.898 1±0.020 5	0.924 4±0.026 8	0.888 1±0.037 3	-	-
OC-KELM	0.925 0±0.019 2	0.930 5±0.014 9	0.927 8±0.018 7	0.893 1±0.029 9	0.001 7	0.001 5
LMKAD	0.930 1±0.023 7	0.933 8±0.021 4	0.932 5±0.022 3	0.900 0±0.031 5	-	-
LMK-OC-ELM-so	0.934 8±0.020 8	0.935 5±0.019 3	0.936 8±0.019 4	0.906 2±0.030 2	0.142 0	0.006 3
LMK-OC-ELM-sig	0.946 1±0.024 4	0.949 5±0.021 3	0.947 8±0.022 8	0.922 3±0.035 4	0.181 0	0.006 9

[1]	刘星, 王文双, 赵建印, 朱敏. 自适应在线增量ELM的故障诊断模型研究[J]. 系统工程与电子技术, 2021, 43(9): 2678-2687.
[2]	戴金玲, 许爱强. 基于动态软聚类的航空电子部件LMKELM诊断模型[J]. 系统工程与电子技术, 2021, 43(3): 637-646.
[3]	刘星, 熊厚情, 赵建印, 朱敏. 基于改进稀疏KELM的在线非平稳动态系统状态预测方法[J]. 系统工程与电子技术, 2020, 42(9): 2022-2032.
[4]	李晨, 杨俊安, 刘辉. 基于信息熵和GA-ELM的调制识别算法[J]. 系统工程与电子技术, 2020, 42(1): 223-229.
[5]	谢皓宇, 邱静, 杨鹏. 考虑单元互测的测试性指标分配方法[J]. 系统工程与电子技术, 2019, 41(12): 2899-2904.
[6]	彭雅奇, 许承东, 牛飞, 李臻, 范国超. 基于鲁棒扩展卡尔曼粒子滤波的RAIM算法[J]. 系统工程与电子技术, 2018, 40(12): 2790-2796.
[7]	马龙, 代超璠, 裴昕, 黄超, 张鸿燕, 苏志刚. MEMS陀螺仪冗余配置故障检测方法[J]. 系统工程与电子技术, 2017, 39(10): 2298-2304.
[8]	张伟, 许爱强, 平殿发. 基于稀疏化核方法的非线性动态系统在线辨识[J]. 系统工程与电子技术, 2017, 39(1): 223-230.
[9]	赵玉新，常帅，万程程. 考虑故障检测效用与修正效用的软件可靠性增长模型[J]. 系统工程与电子技术, 2015, 37(4): 969-973.
[10]	冯丽媛, 姚绪梁, 曹然, 邹艾利. 基于分布式系统可修复控制方法可靠性和安全性分析[J]. 系统工程与电子技术, 2015, 37(11): 2663-2668.
[11]	王佳伟, 沈毅, 王振华. 基于未知输入观测器方法的非线性切换系统故障检测[J]. 系统工程与电子技术, 2015, 37(10): 2322-2328.
[12]	陈芙蓉，尹燕燕，刘飞. 部分转移概率未知的Markov跳变系统鲁棒故障检测[J]. 系统工程与电子技术, 2014, 36(9): 1819-1825.
[13]	刘聪，李颖晖. 基于高增益滑模观测器的非线性系统执行器鲁棒故障检测[J]. 系统工程与电子技术, 2014, 36(8): 1626-1631.
[14]	钱华明，彭宇，崔玫，富振铎. 基于LMI的不确定系统鲁棒故障检测观测器设计[J]. 系统工程与电子技术, 2014, 36(6): 1162-1168.
[15]	马奥家，董朝阳，王青，牛尔卓. 基于异步切换的飞行器网络控制系统故障检测[J]. 系统工程与电子技术, 2014, 36(5): 940-946.