张量模型区分度函数在轴承故障诊断中的应用

Application of the Tensor Model and Discrimination Function in Bearing Fault Diagnosis System

目前对故障信号的刻画方法多以向量模型为主,信号的表现形式单一,存在着小样本和维数灾难等问题。另外,多数特征提取算法在处理新样本时,通常需要同时利用新、旧样本构建新数据空间,并进行重新计算,算法处理新样本的效率较低。为此,提出了一种新的基于张量模型的故障诊断方法,采用小波变换和相空间重构建立轴承故障信号的张量模型,通过高阶奇异值分解获得轴承振动信号的初始特征。在此基础上提出了张量模型初始特征最优分类点的区分度函数,实现了轴承故障的快速诊断。分别利用轴承试验平台和凯斯西储大学轴承数据集进行实验,实验结果证明了所提算法能够提取显著的特征,并具有诊断速度快和识别精度高等优点,适合于实际工程应用。

Currently,vector model is mostly used to describe fault signals,which exists the problems of monotonous form,small samples and curse of dimensionality and so on.Furthermore,when dealing with newly coming samples,new and old samples should be used to construct a new data space that will be recomputed.Hence,a novel feature extraction algorithm based on tensor model is proposed.This algorithm first employs wavelet transformation and phase space reconstruction to construct the tensor model of bearing signal.Then,the HOSVD method is introduced to extract the rough features,by which a discrimination function is given.Finally,the type of bearing fault is quickly recognized.We perform a series of experiments on two real bearing fault data sets,and the results show that the proposed method can extract significant feature.Moreover,it can also efficiently deal with newly coming samples and has high recognization accuracy.Hence,the proposed method is suitable in practical engineering application.