增量LTSA算法在转子故障数据集降维中的应用

摘要

The traditional learning algorithm does not have incremental learning ability, so it is unlikely to deal with ad-ditional new data and large data sets. In this paper, an incremental local tangent space alignment (LTSA) algorithm for me-chanical rotor fault diagnosis was put forward. In this method, the LTSA algorithm was used for dimension reduction of the original training samples, and the corresponding low-dimension configuration was obtained. Then, using the incremental learning algorithm, the additional new samples were processed, and the embedded low-dimensional coordinates of the data were obtained. Finally, the rotor fault datasets verified the feasibility of the method, and a good classification effect was ob-tained.%针对传统流形学习算法不具有增量学习能力；故难以处理新增数据与大规模海量数据集的问题，由此，提出一种用于机械转子故障数据集降维的增量局部切空间的排列算法(ILTSA)。该算法首先采用局部切空间排列算法对原始训练样本进行降维处理，获得其低维流形结构，然后通过增量学习算法对新增样本进行处理。得到所有数据的低维嵌入坐标，最后通过转子故障数据集验证了该方法的有效性，取得了良好的分类效果，有利于实时动态故障监测与诊断。