Abstract
{ "background": "The reliability of railway maintenance depot systems is critical for operational continuity and safety in developing transport networks. Current reliability assessments often lack the flexibility to incorporate multi-level operational data and quantify epistemic uncertainties inherent in such complex infrastructure.", "purpose and objectives": "This article presents a novel Bayesian hierarchical modelling framework designed to evaluate the reliability of railway maintenance depot systems. The objective is to provide a robust methodology for integrating heterogeneous data sources and quantifying uncertainty in reliability parameters for maintenance decision support.", "methodology": "A three-level hierarchical model is developed, where observed failure times at the depot level are modelled using a Weibull distribution, whose shape and scale parameters are themselves modelled by higher-level distributions capturing variability across different subsystems and environmental conditions. The core reliability model for depot $i$ is $Ti \\sim \\text{Weibull}(\\beta, \\etai)$, with $\\log(\\etai) = \\alpha + u{\\text{subsystem}[i]} + v_{\\text{location}[i]}$. Parameters are estimated using Hamiltonian Monte Carlo sampling, with inference providing posterior distributions for all reliability metrics.", "findings": "The application of the framework to a case study demonstrates its utility in quantifying uncertainty, yielding a 95% credible interval for system availability between 0.87 and 0.92. A key finding is the dominant influence of power supply subsystem failures, which were associated with a posterior probability of 0.78 of being the primary reliability bottleneck.", "conclusion": "The proposed Bayesian hierarchical model offers a statistically rigorous and adaptable methodology for reliability analysis of maintenance depots, effectively integrating sparse and multi-level data while fully characterising parameter uncertainty.", "recommendations": "Adoption of this modelling framework is recommended for infrastructure asset managers seeking to move beyond point estimates in reliability assessments. Future work should focus on integrating real-time sensor data into the model's observational layer.", "key words": "Bayesian inference, hierarchical modelling, system reliability, maintenance engineering, railway infrastructure, uncertainty