一种基于非线性流形学习的故障特征提取模型

A fault feature extraction model based on nonlinear manifold learning

流形学习作为一种挖掘高维非线性数据内在几何分布特征的有效方法,可用于故障信号的特征提取。针对机械故障诊断中的非线性、故障征兆复杂的诊断问题,提出一种基于非线性流形学习的故障特征提取模型。该模型针对采集样本的不同处理情形,分别运用Laplacian Eigenmaps算法及其增量、监督算法,进行故障样本的特征提取与分类,由于采用非线性的维数约简方式,极大地保留了故障信号中的整体几何结构信息,增强了故障模式识别的分类性能。最后通过工程实例应用,表明了所提特征提取模型的可行性和有效性。

Manifold learning is one of effective methods to obtain geometric distribution feature in high-dimensional nonlinear data, it can be used for fault signal feature extraction and diagnosis. Aiming at the diagnostic problems with nonlinearity and complex failure symptoms in mechanical fault diagnoses, a feature extraction model based on manifold learning method was proposed. With the model, aiming at different processing cases of the collected sample, Laplacian Eigen-maps algorithm and its incremental, and a supervision algorithm were applied to implement feature extraction and classification of fault samples. As a result of the non-linear dimension reduction method, the model greatly retained the overall geometry information of fault signals, it significantly enhanced the classification performance of fault pattern recognition. The test results of fault diagnosis of an air compressor demonstrated the feasibility and effectiveness of the proposed model.