基于EMD与JADE的设备状态特征提取方法  

Equipment Status Feature Extraction Method Based on EMD and JADE

作　　者：陈凤林[1] 刘永斌[1,2] 方健[1] 许强[1] 

机构地区：[1]安徽大学电气工程与自动化学院,合肥230039 [2]中国科学技术大学精密机械与精密仪器系,合肥230027

出　　处：《计算机工程》2015年第7期305-309,共5页Computer Engineering

基　　金：国家自然科学基金资助项目(11274300);安徽省教育厅基金资助重点项目(KJ2013A010);安徽省自然科学基金资助项目(1408085ME81)

摘　　要：针对机械设备在不同状态下振动信号频率特性的差异,基于经验模式分解(EMD)与特征矩阵联合近似对角化的方法提取设备状态特征参数。采用EMD将信号分解为不同频率成分,计算不同频段信号的频域相关系数,构造信号谱相关特征矩阵,运用联合相似对角化方法对特征矩阵降维,提取设备状态特征参数,研究机械设备故障诊断方法。使用实验实测信号进行验证,并基于支持向量机方法对滚动轴承4种状态特征进行识别,结果表明,该方法提取的特征参数分类正确率达到95%以上,可以有效表征设备状态。According to the different frequency characteristics of vibration signals in different conditions, a feature extraction method for machinery fault diagnosis is proposed based on Empirical Mode Decomposition （EMD） and Joint Approximate Diagonalization of Eigen-matrices （JADE）. Vibration signals ae decomposed into different frequency components which are called stationary Intrinsic Mode Functions（IMFs） using EMD. The correlation coefficients of the IMFs and the original spectrum are calculated to construct a feature matrix of spectrum correlation. Then, the dimension of feature matrix is reduced using JADE. Simulation experimental signals are used to verify the effectiveness of the proposed method. The extracted features using this method are applied to machinery fault diagnosis. The features extracted from bearing signals on four conditions are classified by Support Vector Machine（SVM） ,and the correct rate of classification is more than 95%. The results show that the features extracted by the presented method can effectively characterize machine conditions.

关 键 词：特征提取 故障诊断 经验模态分解 特征矩阵联合相似对角化 谱相关 支持向量机 

分 类 号：TH133.3[机械工程—机械制造及自动化]