基于局部线性嵌入和云神经网络的转子故障诊断方法

A method for rotor fault diagnosis based on local linear embedding and cloud neural network

提出一种基于流形学习的汽轮机转子故障诊断方法。利用振动信号构造一个能够表示该信号的矩阵作为流形学习的输入数据,使用局部线性嵌入算法对矩阵进行维数约简,实现了高维数据向低维空间的嵌入,从而有效提取了故障特征。使用云神经网络分类器测试LLE算法输出维数大于3时的故障诊断率,并分析了各个参数对诊断率的影响。该方法克服了在样本较少的情况下故障诊断的困难,能在有限的故障数据中发掘故障特征并进行故障诊断。

A fault diagnosis method based on local linear embedding（LLE） was proposed here to deal with turbine rotor fault, a feature matrix was built with original vibration signals as the input data of manifold learning, and the higher dimensional data were embedded into the lowerdimensional space, the dimensionality reduction was performed with LLE algorithm. The rate of fault diagnosis was calculated with cloud neural network when the output dimensionality of LLE algorithm was greater than 3, and the effects of the parameters on the fault diagnosis rate were analyzed here. It was shown that the proposed method can overcome difficulties to find the fault features within the finite fault data.