基于EEMD排列组合熵的SVM转子振动故障诊断研究

SVM Rotor Vibration Fault Diagnosis Based on EEMD Permutation Entropy

对汽轮机转子故障状态进行准确判别一直是工程领域研究的重点。在使用支持向量机作为模式识别方法进行故障诊断的过程中,提取能明显区别不同故障的信号特征参数,构建高质量的样本可以较大提高支持向量机(support vector machine,SVM)模型的分类正确率。针对此问题,提出一种总体平均经验模态分解(ensemble empirical mode decomposition,EEM D)、排列组合熵和SVM相结合的汽轮机转子振动多故障诊断方法。方法首先引入有向无环图建立了多故障诊断模型,利用EEMD将振动信号分解成单一无混叠的内禀模态函数(intrinsic mode function,IMF)分量,然后计算对振动信号变化非常敏感的IMF排列组合熵作为特征向量,并应用到有向无环图SVM进行多故障状态识别。实验结果表明,该方法实现了汽轮机转子的振动多故障诊断,同时与基于EEMD能量法提取的特征向量进行对比,通过实验证明,该方法具有更加准确的识别率。

The accurate identification of the fault conditions of steam turbine rotor has been the research focus in the field of engineering. In the process of fault diagnosis by using support vector machine（ SVM）,extracting the signal characteristic parameters,which can clearly distinguish different fault signals to construct high-quality samples,plays a significant role in improving the classification accuracy of SVM model. To solve these problems,we propose a multiple fault diagnosis method for steam turbine rotor based on ensemble empirical mode decomposition（ EEM D）,permutation entropy and SVM. Firstly,this method applies directed acyclic graph to establish multiple faults diagnosis model,and uses EEM D to decompose the vibration signals into single and unmixed IM F components. Then,the permutation entropy of IM F component,which is very sensitive to the changes in vibration signal,is calculated as eigenvectors,and applied in directed acyclic graph SVM for multiple fault state recognition. The experimental results showthat this method can realize the multiple faults diagnosis of turbine rotor vibration. M eanwhile, compared with the extracted eigenvectors based on EEM D energy method, the experiment proves that this method has more accurate recognition rate.