改进时频分析和特征融合在内燃机故障诊断中的应用

针对基于内燃机振动信号的故障识别诊断问题,首先提出一种基于阈值筛选的变分模态分解(VMD)、玛基诺-希尔时频分布(MHD)的时频分析方法,该方法针对Cohen类时频分布存在的交叉干扰项问题,通过阈值筛选法确定VMD算法的分解层数,从而将内燃机振动信号分解成一系列单分量模态信号,然后对单分量信号进行MHD时频表征及线性叠加得到时频聚集性优良、物理意义明确的振动信号时频谱图。再通过局部非负矩阵分解(LNMF)对时频图像特征进行提取,将提取的特征与振动信号时域参数进行特征融合,得到融合特征向量。对支持向量机(SVM)采用改进粒子群优化算法进行参数优选,然后对特征向量进行训练和测试,实现了内燃机的故障识别诊断。将该方法应用于内燃机气门间隙故障8种工况下缸盖振动信号的识别诊断试验,结果表明,该方法能够对不同工况振动信号进行有效识别分类。通过参数优选,最高识别率达到了99.17%,同时对比传统的最近邻分类器的分类结果,证明了该方法的优越性。

Applications of Improved Time-frequency Analysis and Feature Fusion in Fault Diagnosis of IC Engines

Aiming at the problems of fault identification and diagnosis based on the IC engine vibration signals, a time-frequency analysis method was proposed based on variational mode decomposition(VMD) threshold filtering，Margenau-Hill time-frequency distribution(MHD). The method aimed at the problems of cross-interference terms of the Cohen time-frequency distribution, determining decomposition level of the VMD algorithm by the threshold filtering method, the IC engine vibration signals were decomposed into a series of single-component modal signals; then the MHD time-frequency characterization and linear superposition were applied to the single-component signals to get the time-frequency images of the vibration signals with good time-frequency clustering and clear physical meaning.Then，the local non-negative matrix factorization(LNMF) was used to extract features of time-frequency images，and the extracted features were combined with the time domain parameters of the vibration signals to obtain the fused feature vectors. The improved particle swarm optimization algorithm was used to optimize the parameters of the support vector machine(SVM)，and then the feature vectors were trained and tested by SVM to realize the fault identification and diagnosis of the IC engines. The method was applied to the identification and diagnosis tests of the cylinder cover vibration signals under 8 working conditions of valve clearances. The results show that the method may effectively identify and classify the vibration signals of different working conditions. Through parameter optimization, the highest recognition rate reaches 99.17%，and compared the classification results of the traditional nearest neighbor classifier，the superiority of the method is proved.