柴油机时频图像双向二维特征编码与故障识别

Two-Directional Two-Dimensional Feature Encoding Recognition Method for Time-Frequency Images of Diesel Engine

针对柴油机故障特征提取困难的问题,提出一种基于时频图像双向二维特征编码识别的柴油机智能故障诊断方法.将内燃机故障诊断问题转化为故障信号时频图像的识别问题,分别利用短时傅里叶变换、小波包、魏格纳分布(WVD)、伪魏格纳分布(PWVD)与平滑伪魏格纳分布(SPWVD)生成柴油机振动时频图像,提出了自适应匹配追踪(AMP)算法与魏格纳相结合的AMP-WVD时频表征方法;为进一步获取包含于柴油机振动时频图像内部的低维特征参量,在二维非负矩阵分解的基础上提出了双向二维非负矩阵分解(TD2DNMF)算法,将数据矩阵行、列维信息融合到一个判别分析框架中,将不同类别的数据信息并行运算,对柴油机时频图像样本进行特征编码,并将支持向量机作为分类器,实现了时频图像的自动分类识别.在6135G型柴油机上模拟了8种不同气门状态,利用时频图像双向二维特征编码与故障识别方法进行柴油机运行状态判别,结果表明:AMP-WVD时频图像可描述柴油机运行状态信息,各时频分量的物理意义更加明确;TD2DNMF方法有较好的特征提取能力,可提取柴油机故障信息.

In order to deal with the fault feature extraction problems for diesel engine, an intelligent fault diagnosis method based on two-directional two-dimensional feature encoding recognition of time-frequency images was pro- posed. It converts the fault diagnosis issue into the recognition of time-frequency images. The time-frequency images are generated by short time Fourier transform （STFT）, wavelet packet, Wigner-Ville distribution （WVD）, pseudo WVD （PWVD） and smooth pseudo WVD （SPWVD） respectively. Based on adaptive matching pursuit （AMP） algo- rithm and Wigner-Ville, an AMP-WVD time-frequency characterization method was proposed. In order to obtain the low-dimensional characteristic parameters embedded in the time-frequency images, the TD2DNMF algorithm was proposed. The row and column information of the data matrix were fused into a discriminant analysis framework, and different types of data information were run in parallel to characterize the diesel time-frequency image samples, and SVM was used as a classifier to realize the automatic classification and recognition of time-frequency im- ages. Experiments were carried out on the 6135G diesel engine with eight different valve status simulated, the pro- posed method was used to judge the diesel engine operating status. The results show that the AMP-WVD time- frequency images can well describe the diesel engine operating status information. The physical meaning of each time- frequency component in AMP-WVD images is much clearer. TD2DNMF method has a good feature extraction capa- bility and can be used to effectively extract the diesel engine fault information.