基于多尺度排列熵和线性局部切空间排列的机械故障诊断特征提取

Feature extraction of mechanical fault diagnosis based on MPE-LLTSA

机械设备监控系统收集的大量信号通常是包含多种自然振荡模式的非线性信号,这意味着单尺度特征提取无法表征这些非线性信号。而对于高维特征矩阵,也需要进一步提取主要的低维特征。针对这两个问题,提出了一种结合多尺度排列熵和线性局部切线空间排列(MPE-LLTSA)的非线性特征提取方法。首先通过MPE计算信号以获得具有高维度的多尺度特征。然后利用LLTSA挖掘嵌入的内在结构,实现低维特征提取。最后引入最小二乘支持向量机(LSSVM)来训练和识别低维特征。试验结果表明了该方法在机械模式分类和故障识别领域的应用潜力。

A large number of signals collected by mechanical equipment monitoring system are usually nonlinear signals with multiple natural oscillation modes, so the single-scale feature extraction can’t characterize these nonlinear signals. For high dimensional feature matrix, its main lower dimensional features need to be further extracted. Here, to solve these two problems, a nonlinear feature extraction method combining multiscale permutation entropy and linear local tangent space alignment(MPE-LLTSA) was proposed. Firstly, signals were calculated using MPE to obtain multi-scale features with high dimensions. Then, LLTSA was used to excavate the embedded intrinsic structure, and realize low dimensional feature extraction. Finally, least squares support vector machine(LSSVM) was introduced to train and recognize low dimensional features. The test results showed that the proposed method has application potential in fields of mechanical pattern classification and fault recognition.