基于改进MF-DFA和随机森林的液压阀故障诊断

Fault Diagnosis of Hydraulic Valve Based on Improved MF-DFA and Random Forest

多重分形去趋势波动分析(MF-DFA)能够有效地揭示隐藏在非线性和非平稳振动信号中的多重分形特征,而液压阀磨损产生的泄露故障信号往往具有非线性、非平稳,且不同严重程度故障信号特征难以辨识,MF-DFA扩展了液压阀的特征提取及故障诊断方法。然而MFDFA去趋势多项式阶数选取的不恰当往往会出现欠拟合或过拟合现象从而产生新的波动误差。为此,提出了一种改进MF-DFA方法实现故障特征提取。通过建立低阶多项式信号轮廓去趋势拟合曲线和不同时间尺度固有模态函数(IMF)之间的相关性,选取最优的IMF模态分量的累计和将其作为信号轮廓的趋势项,进而提取分型谱参数特征。最后,通过随机森林分类器进行故障模式识别。实验结果证实了所提出的方法在电液换向阀内泄漏故障诊断中的有效性。

Multifractal detrended fluctuation analysis(MF-DFA)can effectively uncover multifractality buried in nonlinear and nonstationary vibration signals,while the leakage fault signals caused by hydraulic valve wear are often nonlinear and non-stationary.MF-DFA makes it possible for feature extraction and fault diagnosis of hydraulic valves because the characteristics of fault signals of different severity are very close to those of difficult to identify.However,the improper selection of detrending polynomial orders in MF-DFA may lead to under-fitting or over-fitting,which may result in new fluctuation error.Therefore,an improved MF-DFA method is proposed to realize fault feature extraction.By establishing the correlation between the detrended fitting curve of the low order polynomial signal contour and the intrinsic modal function(IMF)of different time scales,the optimal IMF modal components are accumulated and used as the trend term of the signal contour,then,the characteristic of spectral parameters was extracted.Finally,the fault pattern recognition is carried out by random forest classifier and the experimental results verify the effectiveness of the proposed method in the electro-hydraulic directional valve internal leakage fault.