ガスセンサとニューラルネットワークモデルによるディーゼルエンジンの故障予知 (<小特集>ファジィ/ニューラルネットのダイナミックスと制御への応用)

概要

論文の詳細を見る
Sensing the density of the exhaust gas elements influenced by faults for ten minutes of engine start-up, the abnormalities of a diesel engine are predicted by a feature-sampling and discriminating processors, which are described by the neural network models. Three pieces of time response density data are transformed into feature data by the feature-sampling processor, which has learned the normal response data of the three gas sensors. At a result, we obtain a pattern which is composed of ten sampling data values at intervals of one minute. Using the pattern, the discriminating processors, which have learned the respective normal or abnormal patterns, calculate certain factors and discriminate and predict the abnormality of the engine combustion by consensus based on certain factors. We expect that this discrimination method will have robustness against the change of gas sensors as days go by, as well as against the start-up conditions of the engine.
一般社団法人日本機械学会の論文
1992-06-25