－ －5Published PaperCondition Monitoring of Railway Tracks from Car-Body Vibration Using a Machine Learning TechniqueHitoshi TSUNASHIMAWith the recent development of sensors and information technology (IT), conditions in railway facilities can be monitored continually by using sensors installed in the rolling stock and in areas adjacent to the track. This, in turn, has spurred the interest in using this type of monitoring to create maintenance plans or schedule condition-based maintenance when the track conditions indicate deterioration. The condition of railway tracks is an important factor in ensuring the safe operation of trains. To ensure that railway travel is safe and comfortable, it is necessary to maintain and manage tracks properly; this includes preventive maintenance. Further, it is desirable to monitor tracks frequently. The deformations that occur in tracks, referred to as track irregularities, are closely connected to the riding quality and safety of railway vehicles. On regional railway lines, many operators are unable to implement adequate inspections owing to problems such as cost and personnel constraints. Hence, low-cost monitoring systems have been developed based on on-board sensing devices and global navigation satellite system (GNSS) data. In this system, the sensing device is installed in an in-service vehicle to measure car-body vibrations. The diagnosis is performed by rating the track conditions based on the measurement data. In this study, we analysed the effects of railway track irregularities on car-body vibrations through railway vehicle travel simulations conducted by using the SIMPACK software package. We subsequently developed an algorithm based on machine learning to diagnose track conditions by extracting features to detect track irregularities. We used the developed algorithm to detect track faults from the car-body vibrations generated by the simulation. Additionally, we examined the possibility of diagnosing track conditions by using real-world data collected by measurements on an actual regional railway line. The results of this study indicated that the developed algorithm could automatically detect faults associated with longitudinal level, alignment, and cross level irregularities by using measured car-body vibrations. In future work, we plan to create higher-precision training data to improve detection performance. We will also attempt to increase the effectiveness of the current approach by increasing the number of training data clusters and developing an algorithm that can detect the magnitude of track irregularities.Keywords: railway, condition monitoring, fault detection, preventive maintenance, machine learning機械学習を用いた車体振動加速度からの軌道状態診断 鉄道を安全かつ快適に走行させるためには，軌道状態を監視，保守することが重要である．現在，軌道の監視には保線係員の巡回や軌道検測車が用いられているが，このような軌道検査には多くの費用，人員や時間を要するため地方鉄道などでは十分な検査を行えない事業者も多い．そのため，先行研究では車両に設置したセンサ類，GPSなどで構成された車体動揺計測装置を用いることで，安価かつ常時軌道状態の診断ができる軌道状態診断システムが開発されている．車体動揺と軌道の歪みである軌道不整には相関性があるため，このシステムでは車体の上下加速度，左右加速度，ロール角速度から，軌道状態に関する評価値であるRMS値を算出することで，軌道不整の診断を行っている．しかし，営業列車から取得できる大量のデータを効率よく計算処理するためは，軌道状態の診断・予測の自動化が必要となる． 本論文では，車体の上下加速度，左右加速度，ロール角速度のRMS値から構成される特徴空間から，機械学習の手法であるサポートベクターマシンを用いて軌道の状態を判別するアルゴリズムを構築し，軌道の上下方向の歪みである高低不整，左右方向の歪みである通り不整，左右レールの高低差である水準不整を自動化に診断する手法を提案し，その有効性を，走行シミュレーションおよび現車実験により検証した．キーワード：鉄道，状態監視，故障検出，予防保全，機械学習Journal（掲載誌）Appl. Sci. 2019, 9, 2734.
元のページ ../index.html#7