基于KVMD-PWVD与LNMF的内燃机振动谱图像识别诊断方法

发布时间：2018-09-17 07:36

【摘要】：为了直接对内燃机振动谱图像进行诊断识别,提出一种基于改进变分模态分解(VMD)、伪魏格纳时频分析(PWVD)与局部非负矩阵分解(LNMF)的内燃机振动谱图像识别诊断方法。该方法首先针对VMD分解过程中的层数选取问题,提出了一种中心频率筛选的VMD分解层数改进方法(KVMD),然后将内燃机振动信号利用KVMD分解成一组单分量模态信号,并对生成的各个单分量信号进行伪魏格纳分析处理后表征成振动谱图像;在此基础上,对生成的内燃机KVMD-PWVD振动谱图像分别采用非负矩阵分解(NMF)和LNMF形成编码矩阵,并采用最近邻分类器、朴素贝叶斯分类器和支持向量机对上述编码矩阵直接进行模式识别,以实现内燃机振动谱图像的自动诊断。最后,将该方法应用在内燃机故障诊断实例中,结果表明:该方法改进了传统图像模式识别中的特征参数方法,能有效诊断出内燃机气门间隙故障,三种分类器识别精度均大于93%,其中支持向量机的分类精度最高,达到99.8%,且采用LNMF形成的编码矩阵识别精度整体高于NMF,为内燃机振动诊断探索了一条新途径。
[Abstract]:In order to diagnose and identify the internal combustion engine vibration spectrum image directly, a new method based on modified variational mode decomposition (VMD),) pseudo-Wigner time-frequency analysis (PWVD) and local nonnegative matrix decomposition (LNMF) is proposed. To solve the problem of selecting the number of layers in the process of VMD decomposition, an improved method of VMD decomposition layer number based on center frequency selection is proposed in this method, (KVMD),. Then, the vibration signal of internal combustion engine is decomposed into a group of single-component modal signals by KVMD. The generated single component signals are characterized by pseudo-Wigner analysis, and then the generated KVMD-PWVD vibration spectrum images are decomposed by non-negative matrix factorization (NMF) and LNMF respectively to form coding matrix. The nearest neighbor classifier, naive Bayesian classifier and support vector machine are used to recognize the coding matrix directly to realize the automatic diagnosis of internal combustion engine vibration spectrum image. Finally, the method is applied to the fault diagnosis of internal combustion engine. The results show that the method improves the characteristic parameter method of traditional image pattern recognition, and can effectively diagnose the valve clearance fault of internal combustion engine. The recognition accuracy of the three classifiers is greater than 93s, and the classification accuracy of SVM is the highest, reaching 99.8. The recognition accuracy of coding matrix formed by LNMF is higher than that of NMF, and a new way for internal combustion engine vibration diagnosis is explored.
【作者单位】：第二炮兵工程大学五系;
【基金】：国家自然科学基金青年基金项目(51405498) 陕西自然科学基金项目(2013JQ8023) 中国博士后基金资助项目(2015M582642)
【分类号】：TP391.41;TK407

【参考文献】