Visual analysis of SOM network in fault diagnosis

Abstract

SOM network (self-organizing feature map neural network) learning with no instructors which has self-adaptive, self-learning features. The advantage is to maintain the topology of original data. It is in extensive application in the field of the data classification, knowledge acquisition, process monitoring fault identification and so on. SOM network is used for rotor fault diagnosis. The U matrix map and D matrix is used as visualization tools to simulate and analyses the classification results, and it is com-pared with the general SOM network clustering results. The conclusion is that the SOM network visualization method is simple and easy to understand, and has high rate in fault discrimination. © 2011 Published by Elsevier B.V.