基于流形学习的涡轮泵海量数据异常识别算法

为了获取海量试车数据中的信息以分析涡轮泵的健康状态,提出一种基于流形学习的海量数据异常识别算法.该算法将反映涡轮泵状态的振动数据重构到高维空间中,利用扩散映射方法直接对其进行学习,提取出数据内在的低维流形特征,以可视化的方式直观地识别出涡轮泵数据中的异常状态.仿真与试车数据验证结果表明了所提算法的可行性和有效性.该算法克服了传统方法解决非线性问题不足的缺点,为试车后涡轮泵的健康分析提供了一条新的途径.

Abnormal recognition algorithm based on manifold learning for turbopump mass data

To extract the information from turbopump mass data and analyze its health condition,the paper presented a mass data abnormal recognition algorithm based on manifold learning.The algorithm reconstructed turbopump vibration data in high-dimensional space.Then,the diffusion map method was used to directly learn the high-dimensional data and extract intrinsic low-dimensional manifold feature in data set.The abnormal condition in turbopump mass data was discerned visually.The validation results of simulation and test data demonstrate the feasibility and effectiveness of the presented algorithm.The algorithm conquers the shortcoming of the traditional methods that are insufficient for nonlinear problems,and gives a new solution of turbopump post-test health analysis.