基于核函数的SOM及在齿轮故障聚类识别中的应用

结合统计学习理论（SLT）和支持向量机（SVM）的原理，在传统的采用欧氏距离作为竞争评价函数的自组织映射（SOM）基础上，提出了一种基于核函数的自组织映射（KSOM）方法，并把它应用于齿轮故障的聚类识别。该方法先把特征空间映射到高维的像空间，使得特征数据在像空间的映像能够较好地分开，然后在特征空间利用核函数来构造像空间中自组织竞争的评价函数以及权值的调整方法，最终在特征空间实现高维像空间中数据的聚类。试验结果表明，应用KSOM方法来聚类识别齿轮的故障，比传统SOM取得的效果更好、更稳定。

Kernel Self-Organizing Maps and Its Application in Gear Failures Clustering and Recognition

Based on statistical learning theory and support vector machine, a new technique, Kernel Self organizing Maps (KSOM), is presented in this paper. Unlike conventional SOM, KSOM firstly maps the feature space to a high dimensional one to enable the mapped features separable, and then by the use of kernel function it constructs the competitive discrimination criterion and tunes the weights in the feature space. As a result, it can cluster the mapped features in the feature space. KSOM is also applied to clustering and recognition industrial gearbox vibration signals of different operating conditions, and experiment results indicate that the proposed method is superior to conventional SOM.