用KL信息数在频域识别机器状态

机器振动信号包含有关机器状态的大量信息,可用来识别机器状态。W.Gersch1]和屈梁生2]用Kallback-Leibler信息数处理振动信号以识别机器状态,取得了较好效果。但是,对一些实际情况,尤其是对于恒转速机器,Gersch方法未能充分利用已知信息。本文用KL数在频域识别恒转速机器状态。由例行检测和试验,可以得到机器各状态的样本功率谱,从其中任取一个作为基准谱。在识别过程中,求出状态样本和待识别样本功率谱与基准谱的差,称为“差谱”。差谱中只包含各状态中互不相同的成分,提高了分析信号的信噪比。用KL数度量各状态差谱间的距离,并用聚类分析方法识别状态,提高了识别精度。经初步实验验证,效果良好。

APPLICATION OF KULLBACK-LEIBLER INFORMATION NUMBER TO RECOGNIZING MACHINERY OPERATING CONDITIONS IN FREQUENCY DOMAIN

Mechanical vibration signals contain large amount of information about the operating conditions of a machine and can often be used to monitor the con- ditions. Gersch 1] and Qu 2] used Kullback-Leibler information number to recognize conditions in time domain and some good results have been achieved. But Gersch's method is not very effective for some practical usage, espacially for constant-revolution machinery. Instead, authors have used Kullback-Leibler information number in the frequency domain. Through routine monitoring and experiments, the power spectrum sample of each condition has been built and stored in memory. In recognition procedure, power spectrum of present ma- chine condition is inputted. Take any one of the oondition samples as a base spectrum by which all sample spectra are substracted, thus so called "diffe- rence spectra" are obtained. These difference spectra contain only the dif- ference information among each samples, so that the signal-noise ratio is increased. Using K-L number to measure the divergence of each pair of sam- ples, and using cluster analysis to recognize conditions can increase the pre- cision of recognition.