Interpreting maintenance decisions in age replacement models with LIME

Proper maintenance of non-repairable systems is essential for ensuring uptime, minimising cost inefficiencies, and meeting performance targets. This study presents a hybridised framework that combines analytical age-replacement models, Monte Carlo simulation, and explainable machine learning (XAI) for improved maintenance decision-making. Empirical inter-failure-time data from a 20-kilowatt radio transmitter system were analysed using EasyFit, which identified the Birnbaum-Saunders distribution as the best-fit for modelling the system’s failure rate. The optimal preventive replacement threshold was computed at τ = 113 hours, with a minimum expected maintenance cost of 122.93 Naira, outperforming baseline models. To overcome the limitations of discrete analytical models, we simulated 10,000 maintenance events using Monte Carlo simulation, enabling the modelling of continuous operational metrics such as uptime, downtime, repair probability (RP) and maintainability index (MI). Classification targets, including Reliability, Availability, and Maintainability, were interpreted using the Local Interpretable Model-Agnostic Explanations (LIME) technique. LIME analysis revealed that inter-failure time (t) and reliability (R(t)) ≤ 0.75 were the strongest negative contributors to low-reliability classification, while RP > 0.69 and MI > 0.69 had the highest positive contributions in maintainability predictions. The findings demonstrate that integrating analytical modelling with XAI offers both theoretical consistency and operational transparency, bridging the gap between optimisation and explainability. This integration enables data-driven, interpretable, and auditable maintenance recommendations suitable for high-reliability systems.