車両機器振動監視のための機械学習を用いた異常検知手法における大規模データ学習法

カテゴリ: 論文誌(論文単位)

グループ名: 【D】産業応用部門

発行日: 2020/06/01

タイトル(英語): Large-Scale Data Learning Method for Anomaly Detection using Machine Learning for Monitoring Vibration in Vehicle Equipment

著者名: 近藤　稔（鉄道総合技術研究所）

著者名(英語): Minoru Kondo (Railway Technical Research Institute)

キーワード: 状態監視，1クラスサポートベクターマシン，代表データ選択　　condition monitoring，one class support vector machine，prototype selection

要約(英語): Vibration monitoring is effective for the early detection of equipment failure, and a vibration monitoring system for vehicle equipment has been developed from the viewpoint of enhancing the reliability and safety of railways. In the proposed system, abnormality detection is performed by applying the One Class Support Vector Machine (OCSVM) to the octave band analysis results of vibration. However, it is difficult to train OCSVM and optimize its hyperparameters for large-scale datasets due to limited computer resources. Therefore, we propose to combine prototype selection (PS) and OCSVM. In this paper, OCSVM is applied to actual vibration data, and the abnormality detection results and calculation time are compared with those of the conventional method. As a result, it was verified that abnormality detection equivalent to that of the conventional method can be achieved using OCSVM with PS.

本誌: 電気学会論文誌D（産業応用部門誌） Vol.140 No.6 （2020）

本誌掲載ページ: 480-487 p

原稿種別: 論文／日本語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejias/140/6/140_480/_article/-char/ja/