Abstract:

In order to improve the classification performance of fault pattern recognition,this paper proposes a local orthogonal locality preserving projection(OLPP) algorithm based on orthogonal manifold feature extraction method. For high dimensional nonlinear data, it can effectively extract the low dimensional manifold feature vector, and will not change the intrinsic attributes of data. Construct a high-dimensional symptom matrix with the vibration signal of the rotor, the orthogonal locality preserving projection was used to reduce the dimension of this high dimension matrix, extract the low dimensional feature vector matrix, mapping in the visual space, this can achieve the result of fault classification effectively, enhance the accuracy in  fault  diagnose. Finally, through experiment and data dimension reduction simulation process shows that the orthogonal locality preserving projection is effective and feasible.

Key words: pattern recognition, orthogonal locality preserving projection, feature extraction, fault diagnosis

摘要：

为了改善故障模式识别的分类性能,提出了一种基于正交局部保持映射算法的多流形特征提取方法。对于高维的非线性数据可以有效地提取低维流形特征向量，并且不会改变数据的内在属性。利用转子的振动信号构造一个高维多征兆矩阵，然后在应用正交局部保持映射将这个高维矩阵进行降维，提取低维特征向量矩阵，映射在可视空间里，从而可以有效地达到故障分类的效果，提高故障诊断的准确率。最后通过实验和数据降维仿真证明了正交局部保持映射算法的有效性和可行性。

关键词: 模式识别, 正交局部保持映射, 特征提取, 故障诊断