Abstract
{ "background": "Manufacturing systems in developing economies face complex, interdependent risks that challenge conventional reliability engineering methods. Current diagnostic frameworks often lack the probabilistic rigour to quantify uncertainty and integrate multi-level operational data, hindering effective risk reduction strategies.", "purpose and objectives": "This article presents a novel Bayesian hierarchical modelling framework designed to diagnose and measure risk reduction in complex industrial systems. The objective is to provide a robust, evidence-based methodology for quantifying improvements in system safety and reliability.", "methodology": "A four-level hierarchical model is developed, integrating plant-wide, subsystem, component, and observational data. The core structure is defined by $y{ij} \\sim \\text{Bernoulli}(\\theta{ij}), \\; \\text{logit}(\\theta{ij}) = \\alpha + \\betaj + \\gammak + \\epsilon{ij}$, where $\\betaj \\sim N(0, \\sigma^2\\beta)$ represents random effects for subsystems. Inference uses Hamiltonian Monte Carlo sampling, with model performance validated via posterior predictive checks and leave-one-out cross-validation on simulated and field data.", "findings": "The model successfully decomposes total system risk into constituent sources, identifying subsystem-level interventions as the most influential for overall risk reduction. Application to diagnostic data from three plants showed a median reduction in predicted failure probability of 34% (95% credible interval: 28% to 41%) following targeted maintenance protocols. The hierarchical structure effectively captured heterogeneity across different production lines.", "conclusion": "The proposed Bayesian hierarchical model provides a statistically coherent and operationally actionable framework for diagnosing risk in manufacturing systems. It formally accounts for uncertainty and data structure, offering a superior alternative to deterministic or single-level analyses.", "recommendations": "Practitioners should adopt this modelling approach for systematic risk diagnostics and prioritising mitigation investments. Further research should integrate time-series data to model risk evolution and expand the framework to incorporate economic cost parameters.", "key words": "Bayesian inference, hierarchical modelling, risk diagnostics