文章快速检索 高级检索

1. 北京航空航天大学自动化科学与电气工程学院, 北京 100083;
2. 先进航空发动机协同创新中心, 北京 100083;
3. 解放军 95809部队93分队, 沧州 061736

Performance evaluation of fault diagnosis system based on Bayesian network
YU Jinsong1,2 , SHEN Lin1, TANG Diyin1, LIU Hao1,3
1. School of Automation and Electrical Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100083, China;
2. Collaborative Innovation Center of Advanced Aero-Engine, Beijing 100083, China;
3. Unit 93, Army 95809 of PLA, Cangzhou 061736, China
Received: 2015-01-31; Accepted: 2015-05-08; Published online: 2015-06-04
Abstract:Assessing whether a newly developed fault diagnosis system is effective is an important issue to ensure diagnosis system performance.Due to the requirement of evaluating the performance of the fault diagnosis system based on Bayesian network (BN), an evaluation method using a modified binomial distribution was developed, considering the real distribution of diagnosis results. The parameters of the modified binomial distribution were estimated using training data during the training process of fault diagnosis system, and both diagnosis accuracy and confidence interval of a diagnostic system could be calculated simultaneously by this evaluation method. The quantitive evaluation indices provided by the proposed evaluation method greatly contributed to the evaluation of acceptability and reliability of a Bayesian network-based diagnosis system, and were of great significance in supporting diagnosis system training. In conclusion, the effectiveness of the proposed evaluation method was validated by an example concerning a fault diagnosis system for the aircraft fuel system.
Key words: Bayesian network (BN)     diagnosis     performance     accuracy     confidence interval

3.2 改进的置信区间计算方法

1) 在相同置信水平下改进方法所求得的置信区间相较于传统方法更狭窄。

2) 尽管在h=100时传统方法所得边界处置信区间要窄于改进方法,但传统方法的置信上下限超出了概率值的有效范围[0,1],不符合置信区间含义。

3) 随着实验样本的增加,置信区间会趋于狭窄,并且2种方法的区间范围也会更加接近。

 图 1 2种方法求得的置信区间对比Fig. 1 Comparison of confidence intervals calculated from two methods

 h 方法 准确度 0.01 0.10 0.30 0.50 0.70 0.90 0.99 100 传统方法 0.0390 0.1176 0.1796 0.1960 0.1796 0.1176 0.0390 改进方法 0.0426 0.0999 0.1491 0.1622 0.1491 0.0999 0.0426 1000 传统方法 0.0123 0.0372 0.0568 0.0620 0.0568 0.0372 0.0123 改进方法 0.0107 0.0312 0.0476 0.0519 0.0476 0.0312 0.0107 10000 传统方法 0.0039 0.0118 0.0180 0.0196 0.0180 0.0118 0.0039 改进方法 0.0033 0.0099 0.0151 0.0164 0.0151 0.0099 0.0033

 图 2 燃油系统故障诊断模型Fig. 2 Fault diagnostic model for a fuel system

 图 3 缺失训练数据对诊断准确度的影响Fig. 3 Influence of missing training data on diagnostic accuracy

 图 4 置信区间验证Fig. 4 Verification of confidence interval
5 结 论

1) 采用准确度与置信区间相结合的指标计算方法可以较为客观和全面地衡量贝叶斯网络的故障诊断性能,运用该指标能够观察到训练数据规模以及数据缺失程度对于诊断系统性能的影响。

2) 对已有置信区间的计算方法进行改进,放宽了置信区间对测试数据规模的限制条件,减小准确度指标因受测试数据制约而造成的误差,计算所得指标可靠性更高,并且更加符合工程实际应用。

3) 通过燃油系统故障诊断实例对诊断准确度和置信区间组成的综合评价指标体系进行验证评估,证明该指标能够实现实际故障诊断系统的评价及改进优化。该方法可以进一步推广到其他故障诊断模型,成为故障诊断技术性能分析的评价指标。

 [1] HECKERMAN D.Learning in graphical models[M].Berlin:Springer Netherlands,1998:301-354. [2] LUXHØJ J T,COIT D W.Modeling low probability/high consequence events:An aviation safety risk model[C]//Proceedings of the Reliability & Maintainability Symposium (RAMS).Washington,D.C.:IEEE Computer Society,2006:215-220. Click to display the text [3] 徐璡,许朝霞,许文杰,等.基于贝叶斯网络原理的835例冠心病病例中医证候分类研究[J].上海中医药杂志,2014,48(1):10-13. XU J,XU Z X,XU W J,et al.Classification of TCM syndromes in 835 cases of coronary heart disease:On the basis of Bayesian networks principle[J].Shanghai Journal of Traditional Chinese Medicine,2014,48(1):10-13(in Chinese). Cited By in Cnki (3) [4] DALLA V L,GIUDICI P.A Bayesian approach to estimate the marginal loss distributions in operational risk management[J].Computational Statistics,2008,52(6):3107-3127. Click to display the text [5] BASIR O,YUAN X H.Engine fault diagnosis based on multisensor information fusion using Dempster-Shafer evidence theory[J].Information Fusion,2007,8(4):379-386. Click to display the text [6] 李业波,李秋红,黄向华,等.航空发动机气路部件故障融合诊断方法研究[J].航空学报,2014,35(6):1612-1622. LI Y B,LI Q H,HUANG X H,et al.Research on gas fault fusion diagnosis of aero-engine component[J].Acta Aeronautica et Astronautica Sinica,2014,35(6):1612-1622(in Chinese). Cited By in Cnki (1) [7] DAS S,HARRIS M.Estimating accuracy and confidence interval of an intelligent diagnostic reasoner system[C]//Proceedings of the 2009 IEEE Autotestcon.Piscataway,NJ:IEEE Press,2009:288-291. Click to display the text [8] WINTERBOTTOM A.The interval estimation of system reliability from component test data[J].Operations Research,1984,32(3):628-640. Click to display the text [9] CHOI A,DARWICHE A,ZHENG L,et al.Machine learning and knowledge discovery for engineering systems health management[M].Boca Raton,FL:Chapman and Hall/CRC Press,2011:39-66. [10] 段荣行,董德存,赵时旻.采用动态故障树分析诊断系统故障的信息融合法[J].同济大学学报(自然科学版),2011,39(11):1699-1704. DUAN R X,DONG D C,ZHAO S M.Information fusion method for system fault[J].Journal of Tongji University(Natural Science),2011,39(11):1699-1704(in Chinese). Cited By in Cnki (3) [11] ZHANG J J.Empirical likelihood ratio confidence interval for positively associated series[J].Acta Mathematicae Applicatae Sinica,English Series,2007,23(2):245-254. Click to display the text [12] 蒋灵,何小荣.BP神经网络的置信度分析[J].计算机与应用化学,1999,16(3):55-60. JIANG L,HE X R.Confidence bounds prediction for backpropagation neural network[J].Computers and Applied Chemsitry,1999,16(3):55-60(in Chinese). Cited By in Cnki (17) [13] KOHAVI R.A study of cross-validation and bootstrap for accuracy estimation and model selection[C]//Proceedings of the 14th International Joint Conference in Artificial Intelligence,Volof the International Joint Conference in Artificial Intelligence.San Francisco,CA:Morgan Kaufmann Publishers Inc.,1995,2:1137-1143. Click to display the text [14] 李俭川,胡茑庆,秦国军,等.贝叶斯网络理论及其在设备故障诊断中的应用[J].中国机械工程,2003,14(10):896-900. LI J C,HU N Q,QIN G J,et al.Bayesian network theory and its application in equipment fault diagnosis[J].China Mechanical Engineering,2003,14(10):896-900(in Chinese). Cited By in Cnki (122) [15] 李鸿.二项分布的参数估计问题研究[J].应用数学学报,2010,33(3):385-394. LI H.The research about binomial distribution parameter estimation problem[J].Acta Mathematicae Applicatae Sinica,2010,33(3):385-394(in Chinese). Cited By in Cnki (5) [16] 孟昭为.二项分布参数的置信区间[J].工科数学,1995,11(4):169-171. MENG Z W.The confidence interval of the binomial distribution parameters[J].Journal of Mathematics for Technology,1995,11(4):169-171(in Chinese). Cited By in Cnki (5) [17] 王正武,任喜风,张瑞平.置信区间分析法在BP神经网络中的应用研究[J].数理统计与管理,2006,25(2):156-160. WANG Z W,RENG X F,ZHANG R P.The research about the application of the confidence interval analysis in BP neural network[J].Application of Statistics and Management,2006,25(2):156-160(in Chinese). Cited By in Cnki (2) [18] MADSEN A L,LANG M,KJARULFF U B,et al.The Hugin Tool for learning Bayesian networks[M].Symbolic and Quantitative Approaches to Reasoning with Uncertainty.Berlin:Springer,2003:594-605.

#### 文章信息

YU Jinsong, SHEN Lin, TANG Diyin, LIU Hao

Performance evaluation of fault diagnosis system based on Bayesian network

Journal of Beijing University of Aeronautics and Astronsutics, 2016, 42(1): 35-40.
http://dx.doi.org/10.13700/j.bh.1001-5965.2015.0070