基于K-L散度的机械或传感器故障判别方法

DISCRIMINANT APPROACH TO THE MACHINERY OR SENSOR FAULT BASED ON THE K-L DIVERGENCE

机械故障诊断系统中,对同一监测部位通常采用双传感器配置(如水平和垂直方位)。文中首先运用核密度估计方法得到两传感器输出信号的概率密度函数估计,然后计算两输出信号间K-L(Kullback-Leiber)散度,并提出一种基于K-L散度值的机械或传感器故障判别准则。通过对一个齿轮减速箱实测振动信号和模拟的传感器故障信号的计算,可以发现,与无故障状态时K-L散度相比,监测部位出现机械故障时两传感器输出信号间K-L散度显著减小;而两传感器之一出现故障时其K-L散度显著增大。因此,两信号间K-L散度的变化可用于区别机械和传感器故障。

It is usually the case for a mechanical fault diagnosis system to configure two sensors in the same monitoring position. The two sensors＇ output signals are used to estimate their respective probability density using kemel density estimation method, and then the K-L（Kullback-Leiber） divergence between two output signals is calculated. A discriminating approach of machinery or sensor faults is proposed based on the K-L divergence. Through evaluating real vibration signals measured on a gearbox and some simulated sensor fault signals, it is shown that contrasted to the K-L divergence between two sensors＇ output signals with fault-free condition, the K-L divergence decreases significantly when a progressed pitting in gears occurs; while it increases significantly when one of two sensors shows some faults. Therefore, the K-L divergence could be an index to distinguish between machinery and sensor faults.