多重分形近似熵与减法FCM聚类的研究及应用

Application of multifractal approximate entropy and subtractive FCM clustering in gearbox fault diagnosis

提出了一种基于多重分形与近似熵相结合的信号特征量提取方法,应用于齿轮箱的故障信号诊断中。针对齿轮箱的故障信号的复杂性,先用减法聚类对提取到的信号特征量进行处理,得到初始聚类中心,然后再用模糊C均值聚类(FCM)作进一步处理,实现齿轮箱故障的自动诊断和识别。多重分形谱提取的特征量如谱宽,可以表示信号的波动程度,而近似熵可以表示信号的复杂程度。两者结合可以得到更加准确的齿轮箱故障信号模式。减法聚类可以有效解决FCM容易陷入局部最优的问题,还可以提高收敛速度。提取的特征参数作为聚类分析的数据,通过计算数据点与聚类中心的隶属度判定所属类型,实现齿轮箱故障类型聚类以及模式识别。通过风力发电机齿轮箱故障诊断实验,证明该方法的可行性和有效性。为齿轮箱故障诊断提供了一种新的有效途径。

A feature extraction method based on muhifractal approximate entropy was presented and used in gearbox fault diagnosis. Considering the complexity of gearbox fault data, the subtractive clustering was used to obtain an initial cluster center of characteristics. Then the fuzzy C-means clustering （ FCM ） was used for further processing to achieve automatic gearbox fault diagnosis and identification. The volatility of a signal was expressed by feature values extracted by muhifractal spectrum, such as spectral width, and the complexity of the signal was represented by approximate entropy （ApEn）. The combination of the two representations can make the patterns of gearbox faults more accurate. The problem of easily falling into local optimum during the FCM clustering process was effectively solved by applying subtractive fuzzy clustering, which also improves the convergence rate. The characteristic parameters extracted were taken as the data used in clustering analysis. In order to achieve gearbox fault clustering and recognition, the membership grades of data points and cluster center were calculated. To prove the feasibility and effectiveness of the method proposed, fault diagnosis experiments on a wind turbine gearbox were implemented. The study provides a new effective way for gearbox fault diagnosis.