A reliability-driven digital twin framework for maintenance scheduling: from asset criticality to hazard-based decision optimisation

Digital Twin research in maintenance has matured around fault detection and diagnostics, yet the pathway from DT-derived risk estimates to defensible maintenance schedules remains underdeveloped. This paper proposes a four-stage, reliability-driven DT framework that specifies a particular method at each stage, rather than presenting the framework only as a concept. Stage 1 ranks assets through a Delphi-Analytic Hierarchy Process procedure that produces a Risk Priority Index integrating expert priority vectors with a percentile-based severity-occurrence rating. Stage 2 characterises system structure with a BowTie model that hybridises Fault Tree Analysis and Reliability Block Diagrams, yielding minimal cut sets, barrier effectiveness, and an explicit single-point-of-failure rule. Stage 3 combines a Cox proportional hazards model, with condition-monitoring covariates, and a cross-fitted residual-learning correction, yielding a hybrid linear predictor and a calibrated baseline cumulative hazard. Stage 4 converts the hazard outputs into decision quantities and optimises the preventive-maintenance interval and risk threshold via parametric simulation subject to an availability constraint. The framework is implemented as an offline, advisory Digital Twin with unidirectional data flow, calibrated and validated on synthetic and historical data using the C-index, Brier score, and expected calibration error. The contribution is a method-locked, decision-oriented architecture that links criticality, reliability structure, hazard modelling, and scheduling optimisation in a single, auditable workflow. The framework is conceptual at this stage; full empirical validation is the next step in the research programme.