多点分析技術による配管サポート異常の検出
Multivariate analysis method to detect abnormality of pipe supports
著者:
角皆 学 Manabu TSUNOKAI 萱田 良 Ryo KAYATA 高瀬 健太郎 Kentaro TAKASE
発刊日:
公開日:
キーワードタグ:
Anomaly DetectionCondition MonitoringMultivariatePipe SupportSimilarity Based Modeling
概要
In order to investigate effective methods to detect the abnormality of pipe supports, vibration data of piping was acquired under various operating pressure and conditions of supports. Single feature value such as RMS of vibration acceleration, vibration velocity, vibration displacement, was not always able to detect the abnormality. To make the feature value more informative, the feature vector was defined as the combination of 3 aforementioned feature value. Then the difference of feature vectors comparing to the normal state was evaluated with Similarity Based Modeling. Through this procedure, all abnormality was detected regardless of mounted positions of sensors. This method was also able to detect all abnormality even when the normal model includes data of different operating pressure. Moreover, by making models from abnormal data, specification of abnormal condition was successfully achieved.