Digital twins (DTs) enable real-time modeling of physical entities but face challenges in multi-domain applications due to distributed, heterogeneous, and privacy-sensitive data. This paper proposes the Privacy-Preserving Dynamic Ontology Evolution (PPDOE) framework, a novel federated DT approach integrating dynamic ontology updates, federated learning (FL), reinforcement learning (RL), and privacy-preserving techniques. PPDOE constructs a semantic graph database using OWL and RDF standards, mapping multi-domain data (e.g., healthcare vitals, aircraft maintenance logs) into a unified, evolving ontology. It leverages FL for secure model aggregation, RL for graph topology optimization, and privacy mechanisms like differential privacy (ε = 0.1) and secure multi-party computation (SMPC) to comply with GDPR and HIPAA5,6. PPDOE enables adaptive, scalable, and privacy-aware collaboration across domains (e.g., health, aviation, manufacturing), transforming DTs into dynamic knowledge systems. It addresses limitations of static, single-domain twins and supports cross-domain reasoning in resource-constrained settings like Small Island Developing States (SIDS).
Paul et al. (Wed,) studied this question.