【作者】 蒋全胜； 贾民平； 胡建中； 许飞云；

【Author】 JIANG Quan-sheng1,2 JIA Min-ping1 HU Jian-zhong1 XU Fei-yun1 (1. School of Mechanical Engineering, Southeast University, Nanjing 211189, China; 2. Department of Physics, Chaohu College, Chaohu 238000, China)

【机构】 东南大学机械工程学院； 安徽巢湖学院物理系；

【摘要】 利用流形学习方法能有效提取高维非线性数据中嵌入的低维流形特征。将其引入到设备故障诊断领域,应用于故障模式识别问题,提出了一种基于流形学习的故障模式识别新方法。运用基于拉普拉斯特征映射的非线性降维算法直接对原始故障信号进行学习,提取数据内在的流形特征,极大地保留了信号中内含的整体几何结构信息,有效克服了常规模式识别方法仅能获得局部线性结构的不足,明显改善了故障模式识别的分类性能。仿真和工程实例结果表明了所提方法的可行性和有效性。更多还原

【Abstract】 Making use of manifold learning method can effectively extract the low dimension feature inbuilt from high dimension nonlinear data. A new method of fault pattern recognition based on manifold learning (ML-FPR) is proposed, while manifold learning is introduced into equipment fault diagnosis fields and applied to fault pattern recognition problem. A nonlinear dimensionality reduction algorithms based on laplacian eigenmap is used to learn original fault signal directly and extract intrinsic manifold feature in data set. The method can hugely hold the whole geometry structure information embeded into the signal, and availably overcome the flaw of general pattern recognition methods which it only obtain data’s local linear structure, obviously improve classify performance of fault pattern recognition. The simulation and instance results demonstrate the feasibility and effectiveness of the method.更多还原