ニューラルネットワークによる空気式制御弁異常診断 : (第2報)

概要

論文の詳細を見る
Our objective is to diagnose the failure of actuators of air-operated control valves that are frequently used in process-control plants. In our first report, a prototype failure diagnosis system using neural network was developed and its recognition capability was discussed. Some failures were difficult to diagnose because their characteristics were very similar to each other or to the normal state. This paper deals with the fuzzy network (the "teacher" data is determined by fuzzy clustering)used to improve recognition capability and the selection of effective sensors by means of the multivariate analysis method. The amount of square errors of the output from the fuzzy network can be considerably reduced for the failures difficult to diagnose with the network with the binary digit "teacher" data. Three effective sensors were selected out of more than 10 sensors, and a new network system utilizing these three sensor signals was developed. The recognition Capability of this network is compared with that of the network utilizing full sensor signals. Its recognition capability is good, even though the number of connections is reduced by approximately 60%.
社団法人日本機械学会の論文
1993-12-25