基于SOM特征映射图上引力场的故障模式识别  

Fault Pattern Recognition Based on Gravitation Field of Self-organizing Feature Map

作　　者：陈文宇[1] 刘井波[1,2] 程小鸥[1] 孙世新[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都610054 [2]重庆三峡学院数学与计算机科学学院,重庆404000

出　　处：《计算机科学》2009年第4期273-276,共4页Computer Science

基　　金：国家自然科学基金(60471055);电子科技大学青年基金(JX7019)资助

摘　　要：对聚类方法SOM特征映射图使用引力场进行区域划分,实现远程故障识别。首先由SOM得到输入模式的具有不同响应特性的聚类区域;对SOM特征图进行邻域相关性分析并使用阀值划分种子区域;以种子区域为引力源在SOM特征图中构造引力场,根据特征图中点沿在引力场所受力的方向运动而收殓到的种子区域进行区域划分,并以此进行模式识别。该方法不用考虑区域边界上点的分类问题且很容易扩展到多维空间。对12种典型飞机起落架故障进行远程检测仿真,取得了较满意的效果,较大地提高了正确识别率。This paper proposed the Remote Fault Diagnosis based on Gravitation Field of Self-Organizing Feature Map （SOM）. In order to find the region of different response characteristic, the input vector can be processed by SOM. This Gravitation Field method only uses the position characteristics of the regions of SOM to find their central seeds, and then construct gravitation field by employing the central seeds as gravitation origin. The convergent region of dot which moves along the direction of gravitation field in SOM is looked as region segmentation. Furthermore, the method needn＇t consider the classification of boundary point and easy to be extended to n-dimensional situation. The proposed methods had been successfully evaluated using twelve different datasets, and had greatly improved the rate of correct classification.

关 键 词：SOM 引力场 区域分割 远程故障诊断 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]