基于VMD和MF-DFA的往复压缩机气阀故障特征提取方法

A Feature Extraction Method for Fault Diagnosis of Reciprocating Compressor Valve Based on VMD and MF-DFA

针对往复压缩机气阀故障振动信号在进行多重分形分析时易受时间序列非平稳趋势影响,无法准确揭示其多重分形特征的难题,提出了基于变分模态分解(Variational Mode Decomposition,VMD)和多重分形去趋势分解(Multifractal Detrended Fluctuation Analysis,MF-DFA)的往复压缩机气阀故障特征提取方法。首先,利用VMD方法对往复压缩机气阀信号进行分解,根据互相关系数法选取模态分量进行信号重构,可有效消除噪声干扰;然后采取MF-DFA方法对重构后信号进行分析,以反映结构特征和局部振动信号尺度行为的特征向量参数Δα、α(fmax)、fmax、Δf和B为模式识别向量,极限学习机(Extreme Learning Machine,ELM)为故障分类器对往复压缩机气阀的4种状态进行分类识别。研究结果表明:该方法能够揭示往复压缩机气阀振动信号的多重分形特性,具有较强的辨识能力。

Considering the fact that multifractal analysis of the reciprocating compressor valve＇ s fault vibration signals is easily affected by the non-stationary trend of time series and it fails to accurately reveal the signal＇ s muhifractal characteristics, an integrated teature extraction method based on variational mode decomposition （VMD） and muhifractal detrended fluctuation analysis（MF-DFA） was proposed tor the valve fault diagnosis of reciprocating compressor. Firstly, having the VMD method used to decompose valve signals and having the cross-correlation number method based to select mode components and reconstruct the signal so as to eliminate the noise interterence effectively;and secondly having the MF-DFA method adopted to analyze the signals reconstructed and having the parameters △α, α （fmax） ,fmax, Af and B which reflecting the structure characteristics and local scale behavior of the vibration signals employed as the pattern recognition vectors;finally having the extreme learning machine（ELM）taken as the fault classifier to classify and identitfy four types of fault teatures of the reciprocating compressor valve. The results show that, this method can reveal the muhifractal characteristics of the vibration signals of reciprocating compressor valve and has strong identification ability.