非高斯噪声下基于Unscented粒子滤波器的非线性系统故障诊断方法  

Unscented Particle Filter-based Fault Diagnosis of Non-linear System with Non-Gaussian Noises

作　　者：葛哲学[1] 杨拥民[1] 胡政[1] 陈仲生[1] 

机构地区：[1]国防科技大学机电工程研究所,湖南长沙410073

出　　处：《兵工学报》2007年第3期332-335,共4页Acta Armamentarii

基　　金：国家自然科学基金资助项目(50375153);部委预先研究项目

摘　　要：非高斯噪声下非线性系统的故障诊断中,一般是基于粒子滤波器的方法,但普通粒子滤波器通常会发生“退化”现象,严重影响故障的检测和诊断品质。本文通过引入Unscented粒子滤波器方法,利用Unscented变换对随机分布的非线性概率传递能力来产生建议分布,能明显地改善普通粒子滤波器的性能;然后,提出了基于该滤波器的序贯式故障诊断策略,采用负对数似然比方法监控系统的运行状态,故障发后利用状态联合估计器进行故障隔离。计算实例表明,该新方法能实时检测诊断出非线性系统的故障,同时能抑制非高斯噪声的影响。The traditional method of fault diagnosis of nonlinear system with non-Gaussian noises is based on particle filter. However, ordinary particle filter has the problem of degeneracy and therefore deteriorates the fault diagnosis performance. Based on generic particle filter, a new Unscented particle filter method was brought forward to estimate the system true state. The estimation performance of the new method was markedly improved by generated importance proposal distribution. A sequential strategy of fault diagnosis was presented and negative log likelihood ratio was used to detect the fault. When a fault occurred, a new joint estimation method was used to isolate the fault. Computational results demonstrate that the proposed method can detect and diagnose faults Of a nonlinear system, and suppress non-Gaussian noises.

关 键 词：机械制造自动化 故障诊断 粒子滤波 非高斯噪声 Unscented变换 联合估计 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]