非线性机械故障诊断的核分布粒子滤波方法

针对非线性、非高斯环境下的机械系统故障诊断问题,提出了一种基于粒子滤波的故障诊断方法.通过粒子滤波得到状态估计值的全概率分布信息可用于故障检测.运用核密度估计方法得到传感器输出信号的概率密度函数估计,计算输出信号与标准信号间Bhatta-charyya(BH)系数,并提出一种基于BH系数值的机械故障判别准则.在各观测值独立同分布的条件下,提出了故障诊断算法.该算法较常规粒子滤波对设备运行状态的跟踪过程BH系数提高30%～50%.

Mechanical Fault Diagnosis of a Non-linear System Based on Kernel Distribution Particle Filter

For the fault diagnosis in nonlinear and non-Gaussian mechanical systems,a particle filtering-based approach is developed.One of its advantages is that the complete probability distribution information of state estimation is well described by particles.First,the estimation of probability density of the sensors＇ output sighals is obtained through the kernel density esfimations,and then the BH（Bhattacharyya） coefficient between the two output signals and the standard signal is calculated.A discriminating approach of mechanical faults is proposed based on the BH coefficient.Under the assumption of independent and identically distributed observed values,a fault detection algorithm is proposed.Compared with conventional partical filters,BH coefficient of the proposed algorithm can be improved by 30%～50% during the tracking state of the mechanical system.