Zobrazit v EDS

Using machine learning for long-term track bed behaviour analysis and maintenance scheduling optimisation

Uloženo v:
Podrobná bibliografie
Název: Using machine learning for long-term track bed behaviour analysis and maintenance scheduling optimisation
Autoři: Popov, Konstantin, Pankaj
Přispěvatelé: Forde, Mike, Chai, Hwa Kian, Pankaj, Pankaj, Loram UK Ltd., Scottish Research Partnership in engineering
Informace o vydavateli: The University of Edinburgh, 2024.
Rok vydání: 2024
Témata: Rail traffic levels, KMeans clustering algorithm, machine learning, long-term track bed behaviour analysis, evaluating railway track quality, machine learning algorithms, high-speed line in the UK, single-layer Artificial Neural Network, high-speed line, Autoencoder, long-term track bed behaviour, track geometry data, maintenance scheduling optimisation
Popis: The purpose of this study is to present a novel approach for evaluating railway track quality using machine learning algorithms. The work will focus mainly on track geometry data from a high-speed line in the UK, as well as two conventional-speed lines which have been used as case studies. Rail traffic levels have been increasing steadily in the UK over the years and this is expected to continue, as the country looks to shift to more sustainable transportation. Greater traffic loads will lead to quicker declining of track components and overall, more deteriorated state of the infrastructure. In addition, changing climate conditions will further add to this because of more frequent occurrence of events such as track buckling due to high temperatures and embankment instability due to increasing levels of precipitation. As a result, more frequent maintenance interventions will be required, as well as a more detailed approach of track quality assessment. Maintenance in the form of tamping is used to re-position the track at a level, which reduces train vibrations and produces a smooth journey. Remedial works are often scheduled using either corrective or preventive maintenance techniques, which do not always provide the highest efficiency. These can result in over-tamping of the track, which can degrade ballast particles more quickly and prematurely result in highly expensive renewal works. On the other hand, proactive maintenance identifies poorly performing segments and investigates the reason behind it, so an intervention can be targeted to remove that reason and improve quality. This ensures greater track stability moving forward and therefore, fewer interventions in the future. Less tamping will preserve the quality of the ballast for a longer period of time and extend the lifespan of the track, thus reducing annual costs. In an attempt to achieve proactive maintenance, track inspections have become increasingly digital, which has resulted in the accumulation of large data sets. Such information helps to investigate track response under various maintenance activities and other factors, such as climate conditions. For the purpose, a single-layer Artificial Neural Network was applied in this study. Based on available data from the high-speed track, it was found that tamping works covering long sections of track (>100-200m) were less efficient at correcting geometry than works focusing on shorter spans (
Druh dokumentu: Doctoral thesis
Popis souboru: application/pdf
Jazyk: English
DOI: 10.7488/era/4352
Přístupová URL adresa: https://hdl.handle.net/1842/41621
Přístupové číslo: edsair.dedup.wf.002..165dfaaf6c69e73f19147b79c7f5223a
Databáze: OpenAIRE
Popis
Abstrakt:The purpose of this study is to present a novel approach for evaluating railway track quality using machine learning algorithms. The work will focus mainly on track geometry data from a high-speed line in the UK, as well as two conventional-speed lines which have been used as case studies. Rail traffic levels have been increasing steadily in the UK over the years and this is expected to continue, as the country looks to shift to more sustainable transportation. Greater traffic loads will lead to quicker declining of track components and overall, more deteriorated state of the infrastructure. In addition, changing climate conditions will further add to this because of more frequent occurrence of events such as track buckling due to high temperatures and embankment instability due to increasing levels of precipitation. As a result, more frequent maintenance interventions will be required, as well as a more detailed approach of track quality assessment. Maintenance in the form of tamping is used to re-position the track at a level, which reduces train vibrations and produces a smooth journey. Remedial works are often scheduled using either corrective or preventive maintenance techniques, which do not always provide the highest efficiency. These can result in over-tamping of the track, which can degrade ballast particles more quickly and prematurely result in highly expensive renewal works. On the other hand, proactive maintenance identifies poorly performing segments and investigates the reason behind it, so an intervention can be targeted to remove that reason and improve quality. This ensures greater track stability moving forward and therefore, fewer interventions in the future. Less tamping will preserve the quality of the ballast for a longer period of time and extend the lifespan of the track, thus reducing annual costs. In an attempt to achieve proactive maintenance, track inspections have become increasingly digital, which has resulted in the accumulation of large data sets. Such information helps to investigate track response under various maintenance activities and other factors, such as climate conditions. For the purpose, a single-layer Artificial Neural Network was applied in this study. Based on available data from the high-speed track, it was found that tamping works covering long sections of track (>100-200m) were less efficient at correcting geometry than works focusing on shorter spans (
DOI:10.7488/era/4352