摘要
齿轮箱故障信号通常是具有多标度行为的非平稳信号,去趋势波动分析(Detrended Fluctuation Analysis,DFA)不能准确揭示隐藏在这类信号中的动力学行为。多重分形去趋势波动分析(Multifractal Detrended Fluctuation Analy-sis,MF-DFA)是DFA方法的拓展,能够有效地揭示隐藏在多标度非平稳信号中的动力学行为。利用MF-DFA计算齿轮箱故障信号的多重分形奇异谱,而多重分形奇异谱的宽度、最大奇异指数、最小奇异指数和极值点对应的奇异指数都具有明确的物理意义,能够表征齿轮箱故障信号的内在动力学机制,适合作为齿轮箱振动信号的故障特征。提出一种基于MF-DFA的齿轮箱故障特征提取方法,将该方法用于包含正常、轻度磨损、中度磨损和断齿故障齿轮箱的故障诊断,并与DFA方法的结果进行了对比。结果表明,提出的方法对齿轮箱故障状态的变化非常敏感,能够完全分离相近的故障模式,有效地克服了传统DFA方法存在的缺陷,为齿轮箱的故障特征提取提供了一种新方法。
Gearbox fauh data are usually characterized by nonstationarity and multiple scaling behaviors, a detrended fluctuation analysis （DFA） often fails to uncover their underlying dynamical mechanism. Multifraetal DFA （MF-DFA） is an extension of DFA and able to effectively reveal their underlying dynamical mechanism hidden in nonstationary data with multiple scaling behaviors. To start with, MF-DFA was used to compute the multifractal singularity spectrum of gearbox fault data. Next, four characteristic parameters including muhifractal spectrum width, maximum singularity exponent, minimum singularity exponent and singularity exponent corresponding to extremum of muhifractal spectrum had clear physical meaning, they could express underlying dynamical mechanism of gearbox fault data and could be employed as fault features of gearbox fault data. Consequently, a novel method for feature extraction of gearbox fault data was proposed based on MF-DFA. Besides, the proposed method together with DFA was utilized to separate the normal, the slight-worn, the medium-worn and the broken-tooth vibration data from a four-speed motorcycle gearbox. The results showed that the proposed method overcomes the deficiencies of DFA, it is sensitive to small changes of gearbox fault conditions, it can totally separate the fault patterns close to each other and is a feasible method for feature extraction of gearbox fault data.
出处
《振动与冲击》
EI
CSCD
北大核心
2013年第2期97-101,共5页
Journal of Vibration and Shock
基金
山东省自然科学基金资助项目(ZR2012EEL07)
关键词
多重分形
去趋势波动分析
齿轮箱
特征提取
multifractal
detrended fluctuation analysis
gearbox
feature extraction
