Diagnosis of Electric Parts by Using Competitive Learning with Confidential Voting

Abstract

This paper describes a new approach for competitive learning network and its application to fault diagnosis of electric parts using image analysis. Conventional competitive learning such as RCE network proposed by Cooper et. al. has a shortcoming with its decision making process. Thus, we introduce the confidential voting process in order to avoid the ambiguous decision making in the network. Generally speaking, reliability of electric product is evaluated with some average properties such as MTTF. However, if we can find an efficient method for fault diagnosis of individual electric product, the more efficient maintenance and quality control can be performed. The basic concept of this study is that the performance degradation of carbon film resistor can be estimated from their images. We first discuss some essential issues to be considered in reliability problems. Experimental results by using the proposed competitive network with confidential voting are also described. It is shown that the new competitive learning approach give a good performance.