基于WMRMR的滚动轴承混合域特征选择方法

Fault diagnosis of rolling bearings in mixed domain based on WMRMR

为充分利用时域、频域以及时频域中的有效特征,提高滚动轴承故障诊断准确率,提出一种混合域特征集构建方法,利用原始信号分别生成时域和频域特征集,通过经验模式分解提取固有模态函数的排列熵和Hilbert谱的奇异值作为时频域特征集,使得混合域特征集比单域特征更能全面准确反映轴承运行状态。针对混合域特征集存在维数过高、特征之间冗余性严重的问题,采用加权最大相关最小冗余的特征选择方法,以支持向量机分类正确率为依据,选取7个有效特征向量。实验结果表明:基于WMRMR的混合域特征选择方法的分类准确率可达98%,能够有效的识别轴承故障信息。

In order to improve the accuracy of rolling bearings fault diagnosis by making full use of effective features in time domain,frequency domain and time-frequency domain, a mixed domain feature construction approach was proposed. With it,time domain and frequency domain features were generated using the original signals,permutation entropies of intrinsic mode functions obtained with EMD and singular values of Hilbert spectrum were extracted as timefrequency domain feature sets,and mixed domain feature sets were made to more fully and accurately reflect bearing running states than the single domain features do. Aiming at mixed domain feature sets having shortcomings of too high dimensions and serious redundancy,a feature selection method based on weighted minimal redundancy maximal relevance（ WMRMR） was proposed,it could select seven major feature vectors based on the classification accuracy of support vector machine. The test results showed that the classification accuracy of mixed domain feature selection can reach 98%based on WMRMR,and it can effectively identify the bearing fault information.