旋转机械振动信号频域随机压缩与故障诊断

Frequency Domain Random Compression of Vibration Signal and Fault Detection for Rotation Machinery

提出一种旋转机械故障诊断方法,该方法由频域随机压缩和稀疏表示分类两部分组成。频域随机压缩实现了故障特征的提取,首先通过傅里叶变换得到振动信号的幅值序列,然后构造随机测量矩阵对幅值序列进行压缩测量,压缩测量值作为故障特征向量。在稀疏表示分类中,以有故障标签的特征向量构成故障特征库,将待测特征向量的分类问题转化为稀疏优化问题,应用正交匹配追踪求得待测特征在故障特征库上的表示系数,然后利用表示系数求出待测特征的类重构偏差,根据类重构偏差可以得到诊断结果。齿轮和轴承故障诊断实验证实了本文所提方法的有效性。

The fault detection method based on random compression of frequency domain and sparse representation classification for rotation machinery is proposed. The random compression of frequency domain is a way to achieve feature extraction. Fourier Transform converts vibration signal to get the amplitude sequence. And then, the compressive measurement of amplitude sequence is implemented with random matrix as fault feature vector. In sparse representation classification, the fault feature library is composed of fault feature vectors of which fault pattern is known. The classification of test feature vector is converted to a sparse optimization problem. The sparse representation coefficient of test feature vector under fault feature library is obtained using Orthogonal Matching Pursuit. With the sparse representation coefficient, the reconstruction residual of test feature vector under each fault pattern is obtained and the fault detection is done. The effectiveness of the proposed fault detection method is verified through the experiment of gear and bearing vibration.