改进粒子滤波在汽轮机故障诊断中的应用夏飞,郝硕涛

针对汽轮机的振动信号容易受到较为复杂的随机噪声污染，提出了一种改进粒子滤波的振动信号降噪方法。首先建立采集振动信号的数学模型，将其作为粒子滤波的状态方程；然后利用小波分析提取采集振动信号的背景噪声，将其和状态信号一起作为观测信号，得到观测方程，把降噪问题转化成在状态空间模型下的滤波问题。由于采用序贯重要性采样的粒子滤波存在着样本退化问题，在重采样阶段采用了一种权值排序、优胜劣汰的重采样算法，就是对各粒子的归一化权值从小到大的排列顺序，并根据权值方差大小淘汰粒子，从而得到了改进的粒子滤波算法，在一定程度上解决了标准粒子滤波的退化问题。进而运用改进粒子滤波算法对振动信号进行降噪处理，降噪前信号和降噪后信号分别通过小波包分解系数求取频带能量，根据各个频带能量的变化提取故障特征向量浓缩了汽轮机振动故障的全部信息，对提取的故障特征向量应用诊断识别算法进行故障模式识别。通过对比降噪前信号和降噪后信号的故障诊断识别率，证明了改进粒子滤波在汽轮机故障诊断中的应用效果更佳。

Improved particle filter applied in fault diagnosis of steam turbineXia Fei, Hao Shuotao

In view of the steam turbine vibration signal being vulnerable to more complex random noise pollution, it puts forward an improved particle filter method of vibration signal de-noising. First, it establishes the mathematical model of the vibration signal acquisition as the state equation in the particle filter. Then it uses wavelet analysis to extract the background noise of the signal acquisition, signal of the state together as the observed signal obtained observation equation. It converts into the problem under state space model. Because the particle filter using sequential importance sampling exists the problem of sample degradation. I n the resampling stage, it proposes a resampling algorithm of weight sorting and survival of the fittest. It normalizes weight of each particle from small to large. It eliminates the large variance of the particles and keeps the small variance of the particles. It gets an improved particle filter algorithm. To a certain extent, it solves the degradation problem of the particle filter. It uses the improved particle filter algorithm to the vibration fault signal. The signal and de-noised signal are decomposed by the wavelet packet to obtain the frequency band energy. According to the change of each frequency band energy extracts fault tag. The fault symptoms condense the whole information of turbine sets vibration faults. For these symptoms, applying diagnosis and identification algorithms can diagnose fault pattern. By comparing the fault diagnosis and recognition rate of the signal and de-noised signal, the improved particle filter is proved to be better in the fault diagnosis of steam turbine.