2021 Fiscal Year Final Research Report
Development of practical anomaly detection based on robust sparse modeling
| Project/Area Number |
18K13953
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 25010:Social systems engineering-related
|
|---|
| Research Institution | Toyo University (2019-2021)
Waseda University (2018) |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2018-04-01 – 2022-03-31
|
| Keywords | 異常検出 / ロバスト統計 / スパース・モデリング / タグチメソッド / MTシステム |
|---|
| Outline of Final Research Achievements |
We consider the problems accompanied with the anomaly detection for sensor data. Since sensor data is automatically acquired and accumulated in real time, there is a possibility that a large amount of anomaly data is mixed in the learning data. Such a learning data can bring significant reduction of the performance of anomaly detection, even if the conventional statistical modeling method is applied. Therefore, we conduct theoretical research so that we apply robust and sparse modeling that can estimate statistical models without being affected by mixed anomaly data to our statistical anomaly detection procedures.
|
| Free Research Field |
複合領域
|
| Academic Significance and Societal Importance of the Research Achievements |
統計的異常検出法は製造業における設備機器の状態監視保全の中核をなす技術であるだけでなく，様々な製品・サービスに応用され，安心・安全な社会システムの構築に重要な役割を担っている．本研究の成果である統計的異常検出法は，特にセンサーから自動で取得・蓄積されたデータを対象とした場合の異常検出性能を飛躍的に向上させるとともに，その原因の特定に有益な情報を同時に提供するものである．この研究成果により，統計的異常検出法の応用可能性が広がり，設備機器の故障予測や重篤な事故の未然防止等の様々な社会問題の解決につながることが期待される．
|
