Dynamic Hybrid Model for Comprehensive Risk Assessment: A Case Study of Train Derailment Due to Coupler Failure

Abstract

Comprehensive risk assessment plays a significant role in railway rolling stock safety planning to prevent accidents, including rail derailment and collision. Several methods of evaluating individual sources of railway system risk, ranging from human factors to inherent system failure and environmental hazards, exist in the literature. However, the lack of a hybrid technique to integrate these multiple sources of risk holistically, including their interdependent effects, as a single framework for robust, accurate, and comprehensive risk assessment can limit risk perception and risk mitigation actions. This report proposes a dynamic hybrid model (DHM) that incorporates the Bayesian convolutional factorization and elimination method as a compound aggregation of frequency and severity distributions. The DHM validates predicted risk using Bayesian expectation–maximization machine learning with evidenced-based propagation from expert knowledge and learned data. It also incorporates sensitivity analysis to improve the predicted risk further by prioritizing the hazards with the maximum impact on the estimated risk due to organization resource constraints. A railway case study in the UK revealed that risk prediction using the DHM provided a holistic view of the risk. The results showed that the quantitative risk prediction using the DHM was significantly more robust, accurate, and holistic than that of the conventional risk-assessment method based on the inherent failure rate. This research will facilitate the comprehensive development of risk-mitigation strategies, such as improvements in staff training and wiring insulation, to decrease the likelihood of train derailment caused by semi-permanent coupler failure.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 19 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.