基于分维LLE和Fisher判别的故障诊断方法

Fault diagnosis approach based on fractal dimension LLE and Fisher discriminant

针对非线性系统故障诊断难以解决的问题,通过改进的局部线性嵌入映射算法解决了非线性数据的特征映射问题。首先,通过线性拟合改进了基于分形维估计的内在维数的估计。然后,将故障状态与空间分布结合起来,通过确定数据点在空间超球内的分布完成故障的检测,在这个过程中将超球的确定与LLE算法中基于核函数的样本外数据扩展结合起来,大大减少了计算量,提高了算法的实时性。然后,利用Fisher判别分析进行故障匹配,通过计算最优的投影向量与历史故障数据投影向量的相似度的计算,完成故障识别,从而为复杂非线性系统故障诊断提供了一种新的有效的方法。

Aiming at the difficult problem of nonlinear system fault diagnosis,an improved method based on LLE is used to solve the problem of character projection of nonlinear data.Firstly,the method of intrinsic dimension estimation based on fractal dimension is improved by linear fitting.And then sensor fault state is combined with spatial distribution to complete the fault detection.Out of sample extension is also considered and combined with fault detection algorithm,which can reduce computation task obviously and improve real time capability of the algorithm.Then,Fisher discriminant is used for fault type matching by computing the similarity between the optimal projection vectors and historical projection vectors,and fault diagnosis is completed.The proposed algorithm provides an effective method for nonlinear sensor fault diagnosis.