基于Stockwell变换的滚动轴承故障诊断方法

Fault Diagnosis Method for Rolling Bearing Based on Stockwell Transform

针对滚动轴承产生磨损时振动信号表现出非平稳的特点,采用继承短时傅里叶变换和小波变换优良性质的Stockwell变换特征提取方法。为解决Stockwell变换后得到的二维矩阵阶数过高问题,采用奇异值分解方法对结果进行降维,并进一步矢量化,然后构成特征矩阵。将特征矩阵分别输出至多分类支持向量机、神经网络和近邻算法模型进行训练。对比测试结果,表明多分类支持向量机在准确率和识别速度上均有优势,从而证明基于Stockwell变换的滚动轴承故障诊断方法的有效性。

Aiming at the non-stationary vibration signal generated by the rolling bearing, the Stockwell transform feature extraction method inheriting STFT and wavelet transform with excellent properties was adopted. In order to solve the problem that the two-dimensional matrix order was too high after the Stockwell transform, the SVD method was used for dimensionality reduction and for vectorization in further step, and then the feature matrix was formed. The feature matrices are respectively output to a multi-classification SVM, neural network and neighbor algorithm model for training. By comparing the test results, it shows that the multi-classification SVM has advantages in accuracy and identification speed, which proves the effectiveness of the fault diagnosis method for rolling bearings based on Stockwell transform.