基于时频图像特征约简的柴油机故障特征提取新方法

A Novel Fault Feature Extraction Scheme for Diesel Engine Valves Based on Feature Reduction of Time-frequency Images

针对柴油机振动响应信号时频表示特征提取困难的问题,提出了一种直接对柴油机振动谱图像进行特征约简的时频特征提取新方法。通过现有的时频分析方法生成柴油机振动谱图像,采用三次卷积插值法对图像矩阵的维度进行压缩,并将矩阵分块计算的方法与稀疏非负矩阵分解算法(sparse non-negative matrix factorization,SNMF)相结合,用来直接对压缩后的振动谱图像进行特征约简以获取蕴含在其内的低维特征。将提出的方法应用于4种不同状态的柴油机气门故障特征提取试验中,结果证明该方法可准确、快速提取柴油机气门故障特征。

Aiming at the difficulty in the feature extraction of time-frequency representation of diesel engine vibration response signals,a novel time-frequency feature extraction scheme with feature reduction directly from the vibration spectra images of diesel engine is put forward. The vibration spectra images of diesel engine are generated by existing time-frequency analysis methods,the dimension of image matrix is reduced by using cubic convolution interpolation,and by combining matrix block calculation method with parse non-negative matrix factorization algorithm,feature reduction is conducted directly on the vibration spectra images compressed to get the low-dimension features implied in them. The results of test applying the scheme proposed to fault feature extraction of diesel engine valves at four different states demonstrate that the scheme can accurately and rapidly extract the fault features of diesel engine valves.