基于时频图像的不变矩神经网络故障诊断方法

A Neural network fault diagnosis method based on moment invariants of time-frequency image

在设备发生故障时,其故障信号往往表现为时变特性。传统的基于FFT变换的分析方法对这些时变信号往往会得出错误的结论。在此,应用一种新的时频分析方法,局域波法,对故障振动信号进行描述。由于局域波时频谱可以表示为灰度图像。因此,利用图像信息的不变矩进行故障特征提取。并且以径向基函数神经网络(RBFNN)作为故障分类器,对实际故障信号进行测试,实验证明了该方法的可行性。

The fault signals always showed the characteristic of non-stationary. The analysis methods based on FFT can not get correct resuits for this signals. Here, we adopt a new time-frequency methods-Local wave method, to analysis the fault vibration signals. Because the Local wave T-F spectrum can be showed in the gray image, so we use the moment invariants of time-frequency image as the fault features, And we use the RBFNN as the fauh classification tool to test the real fauh signals. The experiments show that this method is feasible.