"background": "Process-control systems are critical infrastructure in industrial and utility sectors, yet their reliability in developing economies is understudied. Traditional reliability models often fail to account for heterogeneous operational environments and sparse, multi-source failure data, leading to inaccurate predictions and maintenance schedules. ", "purpose and objectives": "This study aims to develop and validate a novel Bayesian hierarchical modelling framework for assessing the reliability of process-control systems. The objective is to compare its predictive performance against conventional frequentist reliability models, specifically in capturing variability across different system types and operational conditions. ", "methodology": "A comparative study was conducted using operational failure data from multiple industrial sites. The core model is a Bayesian hierarchical Weibull model: T{ij \ (\, ), \ () = \0 + \1 xij + uj, with uj \ N (0, \²u), where i indexes systems and j sites. Model comparison used the Widely Applicable Information Criterion (WAIC) and posterior predictive checks. ", "findings": "The Bayesian hierarchical model demonstrated superior predictive accuracy, with a WAIC score 15. 2 points lower than the best frequentist model. A key finding was that site-specific random effects (uⱼ) explained approximately 40% of the variation in failure rates, highlighting the substantial influence of local operational factors. The 95% credible intervals for the shape parameter \ were consistently more robust to data sparsity. ", "conclusion": "The proposed Bayesian hierarchical approach provides a more robust and nuanced framework for reliability analysis in contexts with heterogeneous operational data. It effectively quantifies uncertainty and integrates multi-level variability, offering a significant methodological advancement over standard models. ", "recommendations": "Adoption of Bayesian hierarchical modelling is recommended for reliability assessments of critical infrastructure in similar contexts. Further research should focus on integrating real-time sensor data
Mwangi et al. (Fri,) studied this question.