Development of a Novel Theoretical Basis for Context-Aware Predictive Maintenance Decision Support in Smart Manufacturing Systems Under Nigerian Industrial Conditions

Smart manufacturing systems incorporating cyber-physical production system architectures generate rich multi-source condition monitoring data streams that have the potential to enable highly accurate predictive maintenance, yet the theoretical basis for translating heterogeneous industrial IoT data into optimal maintenance decisions that account for the full operational and business context of Nigerian manufacturing facilities, including spare parts supply chain uncertainty, technical workforce availability constraints, and episodic power supply interruptions, has not been established. This dissertation develops a novel theoretical basis for context-aware predictive maintenance decision support grounded in the integration of deep learning-based remaining useful life estimation, partially observable Markov decision process-based maintenance policy optimization, and supply chain uncertainty propagation theory. A key original contribution is a context integration layer that represents time-varying operational context variables, including spare part lead time distributions, technician availability schedules, and grid power reliability forecasts, as exogenous state variables in the POMDP formulation, enabling maintenance decisions that are optimal with respect to both equipment health state uncertainty and operational context uncertainty jointly. The theory is developed and validated using four years of condition monitoring data, maintenance records, and operational context logs from a plastics injection moulding facility in Lagos State. Against standard rule-based and non-context-aware predictive maintenance baselines, the proposed framework reduces total maintenance cost by 28.6 percent and unplanned downtime by 41.3 percent over the validation period. Formal proofs of the POMDP policy convergence properties under the proposed context integration are provided. Keywords: predictive maintenance, smart manufacturing, POMDP, remaining useful life, industrial IoT Nigeria