Benchmarking on railway safety performance using Bayesian inference, decision tree and petri-net techniques based on long-term accidental data sets

Research output: Contribution to journalArticlepeer-review

Authors

Abstract

Not only has the railway accidental prevention been a prime focus, but it has also become a key challenge for the industry in recent years. For many decades, rail authorities have attempted to significantly improve rail safety, whilst facing various passengers’ risks and uncertainties. The overarching goal of this study is to develop a new posterior probability model to quantify uncertainties for benchmarking. This is the world’s first to establish new insights from the benchmarking of risk and safety across different rail networks. The insights will point out the advantages and practicability of launching safety policies and reducing railway accidents for other rail networks. The new model has been developed using unparalleled long-term accidental data sets, including ‘a trailer an accident’ and ‘causes of the accident’. The investigation adopts a Bayesian approach (via Python) to codify the novel model. The new findings lead to the better understanding into the uncertainty of railway accidents. Five notable rail networks have been selected as case studies. This study has also compared the effectiveness of the decision tree and Petri-net models using the posterior probability and number of injuries and fatalities. Based on the benchmarking outcomes, Chinese and Japanese railway systems denote the lowest risk over other networks, followed by Spanish, French and South Korean rail networks. The study also demonstrates that the novel benchmarking criteria can effectively measure and compare any rail networks’ risk and uncertainties. Its adoption will lead to performance improvement in terms of safety, reliability and maintenance policies of railway networks globally.

Bibliographic note

Funding Information: The first author gratefully acknowledges the Royal Thai Government for the PhD scholarship at the University Of Birmingham (UOB), United Kingdom and the RISEN funding for one year at University of California, Berkeley. The first author also thanks the second and third authors for giving recommendation during studying PhD at UOB. The third author acknowledges the Australian Academy of Science (AAS) and the Japan Society for the Promotion of Sciences (JSPS), for the JSPS Invitation Fellowship for Research (Long-term), Grant No. JSPS-L15701, at the Railway Technical Research Institute (RTRI) and the University of Tokyo, Japan. The authors are sincerely grateful to the European Commission for the financial sponsorship of the H2020-RISEN Project No. 691135 ‘RISEN: Rail Infrastructure Systems Engineering Network’, which enables a global research network that tackles the grand challenge of railway infrastructure resilience and advanced sensing in extreme environments (www.risen2rail.eu). Publisher Copyright: © 2021 Elsevier Ltd

Details

Original languageEnglish
Article number107684
Number of pages13
JournalReliability Engineering and System Safety
Volume213
Early online date16 Apr 2021
Publication statusE-pub ahead of print - 16 Apr 2021

Keywords

  • Bayesian inference, Decision tree, Petri-nets, Railway accident, Risk and uncertainty