Track Evaluation from Train Responses by Data Assimilation and Machine Learning

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K04570
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 22020:Structure engineering and earthquake engineering-related
• Research Institution: The University of Tokyo
• Principal Investigator: Su Di 東京大学, 大学院工学系研究科(工学部), 特任准教授 (40535796)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 軌道評価 / データ同化 / 機械学習 / 携帯情報端末

Outline of Final Research Achievements

Early detection of rail abnormalities is important to prevent accidents caused by track irregularity. However, track inspections, usually conducted once a year, are infrequent and do not adequately monitor the progress of track deterioration. This study aims to construct a simple and high-frequency monitoring system by estimating track conditions based on vibration responses of car bodies of commercial vehicles observed by inexpensive mobile devices, which have rapidly become popular in recent years. The research method is to classify track irregularity into long- and short-wavelength components, combine data assimilation and machine learning methods, verify the results using numerical analysis models, and apply the results to actual trains. The evaluation method of track conditions is constructed by numerical analysis and actual vehicle measurement, and the practicality and accuracy of the method are clarified.

Free Research Field

構造工学

Academic Significance and Societal Importance of the Research Achievements

軌道の変状特性によって応答への影響が異なる．本研究は対象とする軌道状態を表現するために十分に詳細な，かつ，逆解析に耐える車両モデルを提案する．さらに，長波長変状成分をデータ同化から直接逆推定を適用するともに，推定困難の短波長成分は機械学習より特徴検出を試みた．
本システムでは，営業車両と携帯情報端末を利用するため，特殊な車両と計測機器を必要としないことから，実装は容易である．また，通常営業中に頻繁な計測，データ収集が可能であるため，本研究は地方中小鉄道事業者に低廉かつ簡便なモニタリング手法を提供し，資産価値の維持と向上に資する重要な基礎技術と位置付けられる．