基于流形学习的转子系统运行状态识别理论研究

Research of Running State Recognition Theory for Rotor System Based on Manifold Learning

流形学习是一种机器学习算法,它能够提取隐藏在数据中的内在几何结构与规律特性,因此在图像、信号等处理方面得到广泛应用。运用主成分分析法与拉普拉斯特征映射算法对转子系统正常、碰磨、偏心三种运行状态的振动数据构造的高维特征空间进行降维处理,提取其中的低维流形。结果表明拉普拉斯特征映射算法能有效区分转子正常、碰磨、偏心三种运行状态。

Manifold learning is a kind of machine learning algorithms. The intrinsic geometric structure and regular features which hidden in the data can be extracted by manifold learning so it has been widely used in the aspect of image and signal processing. PCA and Laplacian Eigenmap are applied to distinguish the three types of normal,misalignment,rotor rub-impact for the test-bed. The result shows that Laplacian Eigenmap algorithm can distinguish the three kinds of running state effectively.