A Self-Adaptive Digital Twin Architecture for Dynamic Resource Management

Abstract

Digital Twins (DTs) are increasingly used to manage systems under fluctuating demand, yet many remain static and cannot adjust their internal models or control policies as the targeted real system changes. We present a self-adaptive DT architecture that supports runtime reconfiguration of resources. The architecture integrates semantic reflection to keep the DT’s runtime model aligned with the structural representation of the real system, lifecycle-based state management to trigger reconfiguration at appropriate times, and penalty-guided optimisation for decision-making under constraints, balancing resilience and operational cost when capacity must be reorganised. We realise the approach in DynResDT, a resilient hospital ward prototype for bed bay allocation of patients. Through simulation with realistic patient-arrival patterns and stress peaks, the DT maintains model consistency, opens and closes overflow capacity when needed, and allocates patients while minimising costly room usage and unnecessary moves. Our results show a practical trade-off between correctness and responsiveness: timely and principled adaptions offset potential overhead through reflection and lifecycle logic. The architectural pattern is applicable beyond healthcare to other dynamic resource-management domains.