基于流形子带特征映射的转子复合故障特征提取方法

发布时间：2018-06-04 03:12

本文选题：转子系统 + 流形子带　；参考：《振动与冲击》2017年16期

【摘要】：针对复合故障特征易被噪声信号淹没,传统时频分析和流形学习方法不能完整有效的挖掘故障潜在信息和进一步实现故障特征提取。在流形学习的基础上提出了一种流形子带思想并将其应用到转子复合故障特征提取研究中,进而得出了一种基于流形子带特征映射的转子复合故障特征提取方法。对故障原始信号序列进行相空间重构,结合小波包对噪声的强烈抑制性和对信号分辨率高的特点,将重构信号分解成不同频带即子带。将同故障多种工况下的同一频带融合成频带矩阵并估计其本征维数,并通过拉普拉斯特征映射算法以本征维数为依据将子带降维获取低维特征向量并提取信息熵,进一步实现故障特征提取。实验表明,相对于经典的局部线性嵌入和拉普拉斯特征映射等算法,流形子带特征映射算法不仅对单故障而且对复合故障特征进行了更完整有效的挖掘和提取。
[Abstract]:Because complex fault features are easily submerged by noise signals, traditional time-frequency analysis and manifold learning methods can not fully and effectively mine fault potential information and further implement fault feature extraction. On the basis of manifold learning, an idea of manifold subband is proposed and applied to the research of rotor complex fault feature extraction, and a rotor composite fault feature extraction method based on manifold subband feature mapping is proposed. Based on the phase space reconstruction of the original fault signal sequence, the reconstructed signal is decomposed into sub-bands in different frequency bands, which combines the strong suppression of wavelet packet to noise and the high resolution of signal. The same frequency band under the same fault condition is fused into a frequency band matrix and its intrinsic dimension is estimated. Based on the intrinsic dimension, the subband reduction dimension is obtained by the Laplace feature mapping algorithm, and the information entropy is extracted. Furthermore, fault feature extraction is realized. Experiments show that compared with the classical local linear embedding and Laplace feature mapping algorithms, the Manifold subband feature mapping algorithm not only can mine and extract the single fault but also the composite fault feature more completely and effectively.
【作者单位】：湖南科技大学机械设备健康维护湖南省重点实验室;中交第二航务工程局有限公司深圳分公司;中芯国际集成电路制造(深圳)有限公司;
【基金】：国家自然科学基金项目(51575178;11572125)
【分类号】：TH17;TP181

【相似文献】