Investigation of anomaly detection method for remote monitoring of highway bridges

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K15072
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 22020:Structure engineering and earthquake engineering-related
• Research Institution: Kyoto University
• Principal Investigator: Goi Yoshinao 京都大学, 工学研究科, 助教 (30831359)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: 構造物ヘルスモニタリング / 損傷検知 / ベイズ統計

Outline of Final Research Achievements

This study proposed machine learning to detect anomalies in long-term measurements, focusing on feature extraction using posterior probabilities. For temperature change, we attempted to evaluate the uncertainty of fluctuation by the posterior probability.
Knowledge of modal property fluctuation caused by damage was accumulated through field experiments on steel truss bridges, steel girder bridges, signposts, steel box girders, and so on. A method was also proposed to quantify the uncertainty in the vibration characteristics based on the measured data.
The data collection method and validation proceeded with the cloud server. In addition, a flow diagram was created to provide a basis for third parties to apply the proposed method.

Free Research Field

構造工学

Academic Significance and Societal Importance of the Research Achievements

振動特性に基づく異常検知の試みは既往の研究において数多くなされてきたが，特性を推定しそれらを比較する過程で技術者の経験が求められ，その手順は十分に一般化されてこなかった．このため，異常検知の可否は構造同定や統計の手法に依存する結果となり，これまで実務での振動モニタリングの利用は限定的であった．
本研究の成果により異常検知の手順を一般化することで，主観的な判断に基づく誤検出および見落としを避けやすくなると期待される．また，自動化により多数の橋梁についてスクリーニングを実施することが可能となる．以上より異常検知技術の社会実装において意義のある研究成果が得られたと考えられる．