RAT非线性递归特征提取及故障检测中的应用  

Nonlinear RAT Feature Extraction and Application in Fault Detection

作　　者：赵竞雄[1] 王晓菊[1] 

机构地区：[1]华北科技学院计算机学院,北京东燕郊101601

出　　处：《科技通报》2014年第4期44-46,49,共4页Bulletin of Science and Technology

摘　　要：提出使用平均互信息算法和虚假最近邻点算法提取非线性时间序列相空间重构的最优化重构参数。在研究递归图算法的基础上,提出使用递归图中的递归率与确定性的比值RAT作为一种新的非线性递归特征量,对其算法进行描述。对涡轮发动机涉及到气缸压缩、供油系统和燃烧室等涡轮机子系统3类典型故障进行了故障诊断实验。仿真实验结果表明,使用RAT特征能有效实现3类故障下的发动机故障的聚类和诊断,故障诊断准确率为95.7%,具有绝对优越的诊断性能,具有较强的工程实践意义。The average mutual information algorithm and false nearest neighbors algorithm were presented for calculatingthe optimal parameters of phase space reconstruction. On the basis of recurrence plot, the ratio of recurrence rate to the determinism which called RAT in this paper was proposed as a new nonlinear recurrence feature. And the algorithm of RATwas researched in detail. Three typical types of fault such as gas compressor fault, fuel apply fault and burning fault. Simulation result shows that the RAT feature can realize the fault diagnosis of engine effectively, and the fault diagnosis precision can reach to 95.7%. According to the research result, it shows predominant performance and good engineering value inapplication.

关 键 词：非线性时间序列 递归分析 故障检测 特征提取 

分 类 号：TH165.3[机械工程—机械制造及自动化]