{ "background": "The reliability assessment of industrial machinery fleets is critical for operational efficiency and cost management in heavy industries. Traditional reliability models often fail to account for heterogeneity across different machine types and operational sites, leading to imprecise maintenance strategies.", "purpose and objectives": "This study conducts a comparative methodological evaluation of approaches for modelling fleet reliability. Its primary objective is to demonstrate the superiority of a Bayesian hierarchical framework over conventional pooled and separate models in handling multi-level data from heterogeneous fleets.", "methodology": "We developed a Bayesian hierarchical Weibull model, $T{ij} \\sim \\text{Weibull}(\\alphaj, \\lambda{ij})$, $\\log(\\lambda{ij}) = \\beta0 + \\beta1 x{ij} + uj$, where $T{ij}$ is time-to-failure for machine $i$ in fleet $j$, $\\alphaj$ is the fleet-specific shape parameter, and $uj \\sim N(0, \\sigma^2u)$ are random effects. This was applied to a novel dataset of over 15,000 failure and maintenance events from mining and construction fleets. Model performance was compared using the Watanabe-Akaike Information Criterion (WAIC).", "findings": "The hierarchical model provided a substantially better fit (WAIC difference >120) than conventional alternatives, effectively pooling information across fleets while accounting for their inherent differences. A key concrete result is that it reduced uncertainty in mean-time-failures (MTBF) estimates by approximately 40% compared to analysing each fleet separately, with the 95% credible intervals for fleet-specific parameters being notably narrower.", "conclusion": "The Bayesian hierarchical approach offers a rigorous methodological framework for fleet reliability analysis, yielding more precise and robust inferences than standard methods by formally modelling multi-level data structures.", "recommendations": "Practitioners should adopt hierarchical modelling for heterogeneous fleet data to improve maintenance planning. Future research should integrate covariate information at