2022 Fiscal Year Final Research Report
Study on transfer learning and explainability of data-driven health monitoring
Project/Area Number |
19K12094
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 61030:Intelligent informatics-related
|
Research Institution | The University of Tokyo |
Principal Investigator |
Yairi Takehisa 東京大学, 先端科学技術研究センター, 教授 (90313189)
|
Project Period (FY) |
2019-04-01 – 2023-03-31
|
Keywords | データ駆動型健全性管理 / 機械学習 / 動的システム学習 / 異常検知 |
Outline of Final Research Achievements |
This study tackled two issues associated with data-driven health monitoring methods for artificial systems. The first issue is the difficulty or high cost of obtaining a sufficient amount and quality of training data in real-world artificial systems. The second issue is the lack of explanatory power in data-driven models obtained through machine learning, as they diverge from the domain knowledge of the target artificial system, making them less practical. In this study, we addressed these two problems by integrating interpretable latent variable-state space models, which are highly interpretable for engineers and experts, with state-of-the-art machine learning techniques. This integration enabled us to reduce the required amount of training data through the utilization of domain knowledge and improve the explanatory power of data-driven health monitoring.
|
Free Research Field |
健全性管理
|
Academic Significance and Societal Importance of the Research Achievements |
人工知能・機械学習はインフラや生産システムなどが安全かつ効率的に運用されているかどうかを監視する目的においても大いに期待されているが、学習に膨大な訓練データが必要であること、および、モデルがブラックボックスになり説明性に欠けることが大きな懸念事項である。本研究は、最新の機械学習と伝統的な状態空間・潜在空間モデルを統合して動的なシステムのモデルを学習する方法を開発することによって、これらの問題の解決に貢献した。
|