Digital twins for energy wells through multiscale physics-data fusion: Review, challenges, and a conceptual framework

Energy wells underpin critical technologies in the global energy transition, including geothermal energy extraction, carbon capture and storage, and underground hydrogen storage. In recent years, digital twin (DT) concepts have been increasingly explored to support well design, monitoring, and integrity management. However, most existing DT implementations for energy wells remain limited in scope and maturity, with incomplete integration of multiscale physics, evolving field data, and adaptive operational decision support. These limitations arise from heavy reliance on high-fidelity physics-based simulations with prohibitive computational costs, or on purely data-driven models that exhibit limited generalizability under evolving geological and operational conditions. To clarify the current state of the field, this paper first reviews recent DT developments for energy wells across oil and gas, geothermal, and subsurface energy storage applications. Building on this, we further synthesize advances in molecular dynamics, pore-scale simulation, continuum-scale thermo–hydro–mechanical–chemical modeling, and physics-informed machine learning, highlighting how multiscale physical processes governing clay–cement–fluid interactions, deformation, damage, and transport are currently treated in isolation. Motivated by these gaps, we propose a conceptual multiscale physics–data fusion DT framework for energy wells. The framework outlines an integrated architecture in which molecular- and mesoscale descriptors are systematically upscaled into continuum constitutive relationships, while top-down geological and operational constraints are incorporated to ensure cross-scale physical consistency. A central artificial intelligence layer is introduced to conceptually combine physics-informed surrogate models with ensemble-based data assimilation of heterogeneous monitoring data (e.g., pressure, temperature, deformation, and acoustic responses), enabling real-time updating primarily at the continuum-scale THMC and surrogate-model layers, while molecular- and pore-scale modules provide offline mechanistic priors and event-triggered feedback within a hierarchical multiscale DT architecture. An illustrative lifecycle workflow based on a geological CO 2 storage well is presented to demonstrate how the proposed framework could, in principle, support design optimization, operational anomaly diagnosis, and long-term integrity assessment. This study provides a unifying methodological foundation for the future development of predictive and data-adaptive DTs for subsurface energy systems.