Study on deep learning-based anomaly detection method for wind tunnel balance force data
-
摘要: 风洞天平测力试验数据异常检测有助于分析试验异常原因、排查设备故障、改进试验方案，为解决目前人工检测时间成本高、效率低的问题，提出一种基于深度学习的异常检测框架。首先针对异常零样本问题，对风洞试验常见的异常类型及规律进行总结；然后为解决不同车次数据维度不同的问题，提出基于统计特征的标准化特征表示方案；最后利用神经网络学习异常特征，完成异常检测。试验结果表明：基于深度学习的异常检测方法对风洞异常数据检测的准确率和检出率分别达到了81.7%和72.6%，能够较好地识别孤立跳点异常和多点异常。Abstract: Anomaly detection for wind tunnel balance force data is beneficial to analyze anomaly reasons, improve test schema and troubleshoot equipment problems. To solve the high time cost and low-efficiency problems of the manual detection method, a deep learning-based anomaly detection method is proposed. To solve the problem of no abnormal data, we summarize the most common abnormal types in the wind tunnel test. For the problem that the dimensions of data in different experiments are different, a standardization scheme based on statistical characteristics is proposed. Finally, a deep learning model is utilized to learn abnormal features and detect abnormal data. Experimental results show that our deep learning-based anomaly detection method can achieve 81.7% accuracy and 72.6% recall, and has a good detection performance for isolated jump points and multipoint anomalies.
-
Key words:
- wind tunnel test /
- balance force /
- anomaly detection /
- anomaly rule /
- deep learning
-
图 1 基于深度学习的异常检测算法框架图
Figure 1. The framework of our deep learning-based anomaly detection method
图 3 异常检测方法准确率–召回率曲线图
Figure 3. The precision-recall curve of anomaly detection methods
表 1 风洞天平测力试验数据异常信息
Table 1. The information of wind tunnel balance force anomaly data
异常类型 序号 异常名称 异常规律 孤立跳点异常 1 A孤立跳点 $\left| { {A_{{\rm{abnormal}},{\alpha _i} } } - {A_{{\rm{normal}},{\alpha _i} } } } \right| \geqslant \left| { {A_{{\rm{normal}},{\alpha _i} } } } \right| \times a$ 2 N孤立跳点 $\left| { {N_{{\rm{abnormal}},{\alpha _i} } } - {N_{{\rm{normal}},{\alpha _i} } } } \right| \geqslant \left| { {N_{{\rm{normal}},{\alpha _i} } } } \right| \times a$ 多点异常 3 N模型碰支杆 $\left| { {N_{ {\rm{abnormal} },\alpha \geqslant { {20}^{^\circ} } } } } \right| = c$， $ c $为一常数 4 N多个跳点 $\left| { {N_{{\rm{abnormal}},{\alpha _i} } } - {N_{{\rm{normal}},{\alpha _i} } } } \right| \geqslant \left| { {N_{{\rm{normal}},{\alpha _i} } } } \right| \times a$，$ {\alpha _i} $值有多个 整组试验异常 5 N斜率异常 $\left| { {k_{N,{\rm{abnormal}}} } - {k_{N,{\rm{normal}}} } } \right| \geqslant \left| { {k_{N,{\rm{normal}}} } } \right| \times a$ 6 A整体偏大 ${A_{{\rm{abnormal}},{\alpha _i} } } > {A_{{\rm{normal}},{\alpha _i} } }$ 
下载: 导出CSV
表 3 风洞天平测力试验数据异常检测结果
Table 3. The performance of anomaly detection methods for wind tunnel test data
方法 准确率/% 召回率/% F1分数/% OC–SVM 51.1 ± 1.9 68.5 ± 3.3 58.5 ± 2.4 IF 42.8 ± 1.1 52.7 ± 2.2 47.2 ± 1.0 Deep SVDD 29.5 ± 3.8 35.5 ± 9.5 30.4 ± 4.5 Current work 81.7 ± 1.7 72.6 ± 2.4 76.9 ± 1.4
下载: 导出CSV
表 4 各种类异常识别率
Table 4. The detection results of all anomaly types
异常类型 召回率/% 1 77.6 2 89.9 3 89.7 4 95.0 5 38.0 6 63.3
下载: 导出CSV
-
[1] 谢艳,李平,蒋鸿. 大数据分析方法在风洞试验中的应用[J]. 空气动力学学报,2019,37(6):1004-1009. doi: 10.7638/kqdlxxb-2018.0105XIE Y,LI P,JIANG H. Application of big data analytics approach in wind tunnel test[J]. Acta Aerodynamica Sinica,2019,37(6):1004-1009. doi: 10.7638/kqdlxxb-2018.0105 [2] BELMAHDI M,ZEGADI R,BOUHARATI S,et al. Modeling of air flow in wind tunnel using artificial intelligence techniques[J]. Wulfenia,2013,20(6):94-101. [3] 竹朝霞,惠增宏,金承信. 虚拟仪器技术在风洞测控智能化中的应用[J]. 实验技术与管理,2006,23(9):76-79. doi: 10.3969/j.issn.1002-4956.2006.09.026ZHU Z X,HUI Z H,JIN C X. Virtual instrument application for intelligent system in wind tunnel test[J]. Experimental Technology and Management,2006,23(9):76-79. doi: 10.3969/j.issn.1002-4956.2006.09.026 [4] VLACHAS P R,BYEON W,WAN Z Y,et al. Data-driven forecasting of high-dimensional chaotic systems with long short-term memory networks[J]. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences,2018,474(2213):20170844. doi: 10.1098/rspa.2017.0844 [5] MIFSUD M,VENDL A,HANSEN L U,et al. Fusing wind-tunnel measurements and CFD data using constrained gappy proper orthogonal decomposition[J]. Aerospace Science and Technology,2019,86:312-326. doi: 10.1016/j.ast.2018.12.036 [6] YU W F, WANG N. Research on credit card fraud detection model based on distance sum[C]//Proc of the 2009 International Joint Conference on Artificial Intelligence, 2009 : 353-356. doi: 10.1109/JCAI.2009.146 [7] PURARJOMANDLANGRUDI A,GHAPANCHI A H,ESMALIFALAK M. A data mining approach for fault diagnosis: an application of anomaly detection algorithm[J]. Measurement,2014,55:343-352. doi: 10.1016/j.measurement.2014.05.029 [8] BAUDER R A,KHOSHGOFTAAR T M. Multivariate outlier detection in medicare claims payments applying probabilistic programming methods[J]. Health Services and Outcomes Research Methodology,2017,17(3):256-289. doi: 10.1007/s10742-017-0172-1 [9] BARNETT V, LEWIS T. Outliers in statistical data[M]. 3rd ed. Chichester: Wiley & Sons, 1994. [10] SCHÖLKOPF 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 [11] LIU F T, TING K M, ZHOU Z H. Isolation forest[C]//Proc of the 2008 Eighth IEEE International Conference on Data Mining. 2008: 413-422. doi: 10.1109/ICDM.2008.17 [12] AKCAY S, ATAPOUR-ABARGHOUEI A, BRECKON T P. GANomaly: semi-supervised anomaly detection via adversarial training[M]//Computer Vision - ACCV 2018. Cham: Springer International Publishing, 2019: 622-637. doi: 10.1007/978-3-030-20893-6_39 [13] RUFF L, VANDERMEULEN R, GÖERNITZ N, et al. Deep one-class classification[C]//International Conference on Machine Learning, 2018: 4393-4402. [14] ROBBINS H,MONRO S. A stochastic approximation method[J]. The Annals of Mathematical Statistics,1951,22(3):400-407. doi: 10.1214/aoms/1177729586 [15] NAIR V, HINTON G E. Rectified linear units improve restricted boltzmann machines[C]//Proc of the Proceedings of the 27th International Conference on International Conference on Machine Learning. 2010: 807-814. -
点击查看大图
图(4) / 表(4)
计量
- 文章访问数: 347
- HTML全文浏览量: 158
- PDF下载量: 56
- 被引次数: 0
