基于优化集合EMD的滚动轴承故障位置及性能退化程度诊断方法

Diagnosis method of fault location and performance degradation degree of rolling bearing based on optimal ensemble EMD

为了更有效地同时诊断出滚动轴承故障位置及不同性能退化程度,提出了对滚动轴承不同状态振动信号进行特征提取和智能分类的故障诊断方法。该方法对各状态振动信号进行集合经验模态分解,但其效果依赖于总体平均次数和加入噪声的大小这2个重要参数,因此,提出集合经验模态分解中加入白噪声的准则。将分解后的一系列固有模态函数结合奇异值分解获取各状态的奇异值,并组成特征向量矩阵。将其输入到改进的超球结构多类支持向量机进行分类,从而实现滚动轴承正常、不同故障位置及性能退化程度的多状态同时智能诊断。实验结果表明,提出的集合经验模态分解方法中加入白噪声准则,可避免人为确定分解参数,提高其分解效率。基于优化参数的集合经验模态分解结合奇异值分解的智能诊断方法比已有的基于经验模态分解结合自回归模型的诊断方法识别率高。

In order to more effectively diagnose the rolling bearing fault position and different performance degradation degrees simultaneously, a fault diagnosis method is introduced to achieve feature extraction and intelligent classi- fication for the vibration signals of rolling bearing under different conditions. In this method, the vibration signal in each condition is decomposed using ensemble empirical mode decomposition （EEMD） , however, the result depends on two important parameters, i. e. the number of ensemble trials and the amplitude of the added white noise. So, a rule of adding white noise in EEMD is presented. Using the series of intrinsic mode functions （IMF） obtained with EMD method and combined with singular value decomposition （SVD）, the singular values for different conditions are ob- tained, which form the feature vector matrix. The obtained feature vector matrix is then used as the input of the im- proved hyper-sphere multi-class support vector machine for classification. Thereby, the multi-status intelligent diagno- sis of normal rolling bearings and the faulty rolling bearings at different fault locations and with different performance degradation degrees ean be achieved simultaneously. Experiment results show that the presented rule of adding white noise in EEMD method can avoid artificially determining decomposition parameters and improve the decomposition ef- ficiency. Compared with the diagnosis method based on EMD combined with autoregressive （AR） model, the intelligent diagnosis method based on optimal parameter EEMD combined with SVD has higher recognition rate.