基于PCA-FastICA的故障信号分离识别方法

PCA-FastICA based fault signal separation and identification method

针对燃气轮机转子系统故障信号成分分类复杂、预处理困难的问题,提出了基于改进PCA-FastICA的故障信号识别方法。首先用PCA算法对信号进行零均值处理,然后根据协方差矩阵的特征值与特征向量和奇异值分解原理的奇异值(向量)之间的计量关系,获取降序排列的特征值及对应的特征向量,简化PCA计算流程,将优化的低维特征矩阵用于FastICA进行分类训练。实验结果表明:降维后的数据能保留原始信号大部分有用信息,通过FastICA能有效分离识别故障信号,对进一步分析转子故障信号提供一种有效的途径。

In view of the complex classification of the fault signal components of the gas turbine rotor system and the difficulty in preprocessing,an improved PCA-based Fast ICA fault signal identification method was proposed.The signal was processed with zero by PCA algorithm,and according to the measurement relationship between the eigenvalue and the eigenvector and the singular value(vector)of the singular value decomposition principle,the eigenvalues and the corresponding eigenvectors in descending order was obtained,the PCA calculation process was simplified,and the optimized low-dimensional eigenmatrix was used for Fast ICA classification training.The experimental results show that the data after reducing dimensionality can retain most of the useful information of the original signal,and fault signal can be effectively separated and identified with Fast ICA,which provides an effective way for the further analysis of the rotor vibration fault signal.