基于 EMD-WVD 与 LNMF 的内燃机故障诊断

摘要

内燃机的振动信号是复杂非平稳信号，准确提取内燃机振动信号中的特征信息进行模式识别，是对振动信号进行故障诊断的关键。基于经验模态分解的维格纳时频分析方法，不但保留了维格纳分布的所有优良特，而且还避免了交叉项的干扰，能够有效地提取内燃机振动信号的特征信息；在此基础之上，针对传统非负矩阵分解非正交的基矩阵导致数据冗余性较大、影响后续故障分类准确率提高的问题，提出采用局部非负矩阵分解的方法，直接对 EMD-WVD 时频图像的矩阵进行分解，计算用于内燃机故障诊断的特征参数，并利用特征参数进行故障分类。对内燃机4种不同工况的振动信号进行实验，证明基于 EMD-WVD 与局部非负矩阵分解的方法对内燃机气门间隙的故障诊断的有效性。%IC engine vibration signals are complex and non-stationary,therefore,how to accurately extract the feature information of IC engine vibration signals for pattern recognition is very important for fault diagnosis of an IC engine.Here,Wigner-Ville time-frequency distribution analysis method based on empirical mode decomposition (EMD-WVD)was used for IC engine vibration fault diagnosis.The method of EMD-WVD could not only avoid interferences of cross-terms,but also retain all the excellent characteristics of Wigner-Ville distribution to effectively extract fault features of IC vibration signals.A new method of locals non-negative matrix factorization (LNMF)was directly used for feature extraction due to the traditional NMF non-orthogonal basis matrix with larger data redundancy,and being not conducive to subsequently improving the correct rate of faut classification.The application of LNMF in practical IC engine fault diagnosis showed that LNMF algorithm's fault classification accuracy is better than that of the traditional NMF algorithm. The test results for IC engine vibration signals under 4 different operational conditions verified the effectiveness of EMD-WVD and LNMF methods in IC engine fault diagnosis.