ScholarWorks@Y-Scholar Hub

초록

The integration of Computational Fluid Dynamics (CFD) and Artificial Intelligence (AI) have offered a trans-formative solution to CFD's prohibitive computational cost in built environment application. While AI surrogates successfully accelerate simulations, a systematic review of 229 studies revealed a fundamental disconnect: studies are overwhelmingly distributed across a single spatial scale, such as a room, a building, or a city. True multi-scale integration, essential for capturing the interconnected physics of the built environment, remains critically under-explored. This core challenge is compounded by persistent hurdles in model generalization, physical consistency, and research reproducibility. This review synthesizes these limitations to propose a comprehensive framework that pivots from isolated models toward a unified, physically consistent approach. We advocate for developing hierarchical, physics-informed learning strategies built upon sophisticated benchmark datasets to ensure reliable and scalable information transfer across room-, building-, and city-scales. This work provides the necessary roadmap to bridge this critical gap, enabling robust, multi-scale simulation for sustainable design and operation.