This paper introduces a conceptual framework to advance Self-Generating Digital Twins (SGDTs) by integrating Hierarchical Digital Twin (HDT) architectures, Physics- Informed Machine Learning (PIML), Generative Artificial Intelligence (Gen-AI), and dynamic diffusion models. Current SGDT implementations face significant challenges in real-time Structural Health Monitoring (SHM) due to the complexity of processing vast, multi-dimensional data streams and the inherent difficulties in achieving accurate predictive insights across large-scale systems. The reliance on a singular SGDT framework often results in inefficiencies when addressing multi-level structural complexities and managing uncertainties, highlighting the necessity for a scalable and layered approach. The proposed framework leverages HDTs to decompose global and local structural behaviours into interconnected layers, simplifying computational demands while enhancing interpretability. By embedding fundamental physical laws and constraints into the learning process, PIML improves the efficiency and accuracy of Gen-AI models. Gen-AI autonomously refines the digital twin models, synthesising actionable insights from real-time data streams, while dynamic diffusion models facilitate precise damage evolution predictions under varying operational and environmental conditions. This proposal addresses critical gaps in real-time SHM by overcoming the limitations of standalone SGDT systems and introducing a methodology that is both scalable and adaptable. It offers a novel pathway for optimised resource allocation, reduced downtime, and improved sustainability in infrastructure management. Although validation through real-world application is a future objective, the framework provides a robust foundation for advancing SHM technologies to meet the growing demands of mobility, autonomy, and sustainability.
Sasan Darius Adib (Tue,) studied this question.