基于支持向量机的飞行器多余物信号识别

孟偲; 李阳刚; 张国强; 赵长兴

doi:10.13700/j.bh.1001-5965.2019.0266

基于支持向量机的飞行器多余物信号识别

doi: 10.13700/j.bh.1001-5965.2019.0266

孟偲^{1, 2, ,},
李阳刚¹,
张国强³,
赵长兴³

1.
北京航空航天大学宇航学院, 北京 100083
2.
北京航空航天大学生物医学工程高精尖创新中心, 北京 100083
3.
航天科工防御技术研究试验中心, 北京 100854

详细信息

作者简介:
孟偲, 男, 博士, 副教授, 博士生导师。主要研究方向:机器视觉、机器人智能系统; E-mail:Tsai@buaa.edu.cn

李阳刚, 男, 硕士研究生。主要研究方向:图像处理、机器学习

张国强, 男, 硕士研究生。主要研究方向:无损检测

赵长兴, 男, 高级工程师。主要研究方向:无损检测

通讯作者:
孟偲, E-mail:Tsai@buaa.edu.cn

中图分类号: TP274⁺.3;TP274⁺.5;V19
计量
- 文章访问数: 577
- HTML全文浏览量: 93
- PDF下载量: 285
- 被引次数: 0
出版历程
- 收稿日期: 2019-05-28
- 录用日期: 2019-08-30
- 网络出版日期: 2020-03-20

Signal recognition of loose particles inside aerobat based on support vector machine

1.
School of Astronautics, Beihang University, Beijing 100083, China
2.
Beijing Advanced Innovation Center for Biomedical Engineering, Beihang University, Beijing 100083, China
3.
Defense Technology Research and Test Center, China Aerospace Science and Industry Corporation, Beijing 100854, China

More Information

Corresponding author: MENG Cai, E-mail:Tsai@buaa.edu.cn

摘要

摘要:
针对飞行器控制电路在生产制造过程中可能引入金属线头等微小多余物，从而留下短路等安全隐患的问题，提出了一种基于微粒碰撞噪声检测（PIND）的飞行器多余物材质识别方法。首先，利用短时自相关函数提取PIND信号的脉冲部分；然后，提取多种时频域统计特征，并与梅尔频率倒谱系数（MFCC）特征结合起来；最后，训练多分类支持向量机模型实现材质分类。为验证所提方法的有效性，采集了3种不同材质多余物的PIND信号进行模型训练及测试，实验结果表明，所提方法材质识别准确率达98%，优于同类方法的相关结果。
- 多余物检测 /
- 微粒碰撞噪声检测(PIND) /
- 机器学习 /
- 信号识别 /
- 支持向量机
Abstract:
Loose particles such as metal fragments and wires may be left in the control circuit of the aerobat during the process of manufacture, which will cause potential danger like short circuits. To solve this problem, a method of identifying material of loose particles in the aerobat based on particle impact noise detection (PIND) is proposed. This method firstly uses short-time autocorrelation function to obtain the pulse part of PIND signal, and then extracts various statistic features in time domain and frequency domain, which is combined with Mel frequency cepstral coefficient (MFCC) feature, and finally trains support vector machine model for material classification. In order to verify the effectiveness of the proposed method, loose particles' PIND signals with three different types of material are acquired and used for model training and tests. Test results show that the accuracy of identification can reach 98% which is better than related papers' results, verifying the effectiveness of the proposed method.
- loose particle detection /
- particle impact noise detection (PIND) /
- machine learning /
- signal recognition /
- support vector machine

HTML全文

图 1 多余物检测系统

Figure 1. Loose particle detection system

下载: 全尺寸图片幻灯片

图 2 不同材质的多余物信号及对应频谱

Figure 2. Loose particle signals of different material and corresponding spectra

下载: 全尺寸图片幻灯片

图 3 一段PIND信号样例及主次峰比

Figure 3. An example of PIND signal and main-secondary peak ratio

下载: 全尺寸图片幻灯片

图 4 基于SVM的材质识别方法流程

Figure 4. Process of method for material identification based on SVM

下载: 全尺寸图片幻灯片

图 5 MFCC提取过程

Figure 5. Process of MFCC extraction

下载: 全尺寸图片幻灯片

图 6 Mel滤波器组

Figure 6. Mel filter set

下载: 全尺寸图片幻灯片

图 7 SVM多分类策略

Figure 7. SVM multi-classification strategy

下载: 全尺寸图片幻灯片

图 8 搜索网格

Figure 8. Grid for search

下载: 全尺寸图片幻灯片

图 9 网格搜索过程

Figure 9. Process of grid search

下载: 全尺寸图片幻灯片

图 10 实验设备

Figure 10. Experiment equipment

下载: 全尺寸图片幻灯片

表 1 仅使用时频域特征的识别结果

Table 1. Identification results by using features in time and frequency domain

材质	样本数	正确识别数	错误识别数	识别准确率/%
锡渣	181	174	7	96.13
环氧	169	150	19	88.76
细丝	150	143	7	95.33

下载: 导出CSV

表 2 时频域特征与MFCC组合的识别结果

Table 2. Identification results by using features in time and frequency domain combined with MFCC

材质	样本数	正确识别数	错误识别数	识别准确率/%
锡渣	181	176	5	97.24
环氧	169	167	2	98.82
细丝	150	148	2	98.67

下载: 导出CSV

表 3 与同类方法的结果对比

Table 3. Result comparison with other methods

方法	材质类别	颗粒质量/mg	信号长度/s	识别准确率/%
文献[5]	金属与非金属	1.8~4.4	10	85
文献[6]	铝、锡、导线、芯片	1~4	5	91
本文	锡渣、环氧、细丝	0.1~4	0.128	98

下载: 导出CSV

参考文献(15)

[1]	王国涛, 王强, 韩笑, 等.密封电子元器件多余物检测技术综述[J].机电元件, 2017, 37(1):55-63. doi: 10.3969/j.issn.1000-6133.2017.01.014 WANG G T, WANG Q, HAN X, et al.Summarize of loose particles detection method for sealed electronic components[J].Electromechanical Components, 2017, 37(1):55-63(in Chinese). doi: 10.3969/j.issn.1000-6133.2017.01.014
[2]	西印.多余物导致发射失败质子号发动机厂受到批评[J].导弹与航天运载技术, 2000(2):63. http://www.cnki.com.cn/Article/CJFDTotal-DDYH200002017.htm XI Y.Loose particles cause launch failure and proton engine factory is criticized[J].Missle and Space Vehicles, 2000(2):63(in Chinese). http://www.cnki.com.cn/Article/CJFDTotal-DDYH200002017.htm
[3]	United States Department of Defense.Test method standard electronic and electrical component parts: MIL-STD-202G[S]. Washington, D.C.: United States Department of Defense, 2002: 4-7.
[4]	张龙.星载电源多余物检测系统的研究[D].哈尔滨: 哈尔滨工业大学, 2010. ZHANG L.Research on particles detection and identification system for aerospace power[D].Harbin: Harbin Institute of Technology, 2010(in Chinese).
[5]	陈蕊.航天电子装置多余物检测与材质识别方法研究[D].哈尔滨: 哈尔滨工业大学, 2014. CHEN R.Research on detection and classification method of remainders in aerospace electronic equipments[D].Harbin: Harbin Institute of Technology, 2014(in Chinese).
[6]	陈金豹.密封电子设备多余物检测的信号特征识别与置信度评价[D].哈尔滨: 哈尔滨工业大学, 2015. CHEN J B.Signal pattern recognition and confidence evaluation for loose particle detection of sealed electronic devices[D].Harbin: Harbin Institute of Technology, 2015(in Chinese).
[7]	HSUNG T C, LUN D P K, SIU W C.Denoising by singularity detection[J].IEEE Transactions on Signal Processing, 1999, 47(11):3139-3144. doi: 10.1109/78.796450
[8]	WU Y G, WANG S C, WANG S J.Research on wavelet threshold de-noising method of remainder detection for stand-alone electronic equipments in satellite[C]//International Conference on Pervasive Computing Signal Processing & Applications.Piscataway, NJ: IEEE Press, 2010: 1013-1017.
[9]	王强.密封电子元器件多余物检测信号调理优化与组件信号识别[D].哈尔滨: 哈尔滨工业大学, 2018. WANG Q.Signal conditioning optimization and component signal recognition for redundant particles detection of sealing electronic components[D].Harbin: Harbin Institute of Technology, 2018(in Chinese).
[10]	DAVIS S B, MERMELSTEIN P.Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences[J].IEEE Transactions on Acoustics Speech and Signal Processing, 1980, 28(4):357-366. doi: 10.1109/TASSP.1980.1163420
[11]	CORTES C, VAPNIK V.Support-vector networks[J]. Machine Learning, 1995, 20(3):273-297. http://d.old.wanfangdata.com.cn/Periodical/hwyhmb200803006
[12]	周志华.机器学习[M].北京:清华大学出版社, 2016:121-146. ZHOU Z H.Machine learning[M].Beijing:Tsinghua University Press, 2016:121-146(in Chinese).
[13]	KNERR S, PERSONNAZ L, DREYFUS G.Single-layer learning revisited:A stepwise procedure for building and training a neural network[M].Berlin:Springer, 1990:41-50.
[14]	SUN L, BAO J, CHEN Y Y, et al.Research on parameter selection method for support vector machines[J].Applied Intelligence, 2018, 48(2):331-342. doi: 10.1007/s10489-017-0975-3
[15]	STAELIN C.Parameter selection for support vector machines: HPL-2002-354R1[R].Palo Alto: Hewlett-Packard Company, 2003.