基于LMD多尺度熵和LSSVM的往复压缩机故障诊断方法研究

Research on the Fault Diagnosis Method for Reciprocating Compressor Based on LMD,MSE and LSSVM

针对往复压缩机振动信号的非线性和非平稳多源冲击性,提出一种基于局部均值分解(LMD)、多尺度熵(MSE)和最小二乘支持向量机的诊断方法。首先,利用LMD将不同状态振动信号分解为一系列乘积函数(PF)分量,然后根据各PF分量与原信号的互信息值,选择相关性较大且包含故障状态主要信息分量,计算其相应的多尺度熵值,并构造能够定量描述往复压缩机状态的特征向量,最后利用LSSVM作为模式分类器,对上述不同状态下的特征向量样本进行训练和识别,诊断得出往复压缩机气阀故障类型。进一步与小波多尺度熵、EMD多尺度熵方法所提取特征向量识别结果进行对比,结果表明:该方法具有更高的识别率,为往复压缩机故障诊断提供了一种新途径。

According to the multiple source impact characteristics of nonlinearity and non-stationarity for the vibration signal of recip-rocating compressors, a fault diagnosis approach for reciprocating compressor based on local mean decomposition （LMD）, multiscaleentropy （MSE） and least square support vector machines （LSSVM） is proposed. Firstly, The LMD method is used for obtaining a se-ries of product functions （PFs） which are decomposed from the vibration signals in different state. Then the main PFs that containslarge correlation and useful fault information are selected by comparing the mutual information between each PF component and theoriginal signal. Next, the multiscale entropy of each main PF is calculated and the feature vectors which can quantitavely describethe state of reciprocating compressor are established. Finally, the LSSVM is taken as the pattern classifier, some feature vectors sam-ples are used as training set and the others used as test set for diagnosing the fault type of reciprocating compressor gas valve. Com-pared with wavelet＋MSE and EMD＋MSE method, the fault feature extracted by the proposed method can obtain a higher recognitionrate. Therefore, it supplya new way for fault diagnosis of reciprocating compressor.