CEE PhD Defense: Juliet Nwagwu Ume-Ezeoke, "Integrating Physics-Based Reduced Order Modeling into Early Stage Design"

Abstract:

The long-run thermal and energy performance of buildings and outdoor spaces is influenced by local wind velocity fields. These flow fields govern whether occupants can rely on natural ventilation, and whether pedestrians experience comfortable streets. Yet fluid-dynamical information typically enters the architectural design process at later stages, only once geometry has been cemented. Flow fields thus serve as a check of design decisions, rather than a driver. Bringing velocity data into early-stage design could enable long-term thermal performance to materially shape design intent. A challenge of incorporating velocity data is that early-stage design requires iteration over flexible, under-resolved geometries at a pace incompatible with high-fidelity simulation.

This thesis proposes two physics-aware workflows for early-stage design, each illustrating a different approach to reduced-order modeling. The first, AURA, enables designers to convert sketches into building energy models with low effort, exposing the physics-based Airflow Network for pressure-driven analysis of natural ventilation. In a case study of three apartment floor plans, switching layouts shifted the plan-averaged temperature by 0.32 °C, an effect comparable to or larger than that of changing window sizes and constructions, highlighting the value of resolving geometrical layouts from sketches. AURA is then applied to a dataset of 500+ floor plans to derive generalizable rules to inform natural-ventilation plan design. The second approach enables representations of urban geometry and geometric interventions as 2D polygons in a CFD domain. A data-driven surrogate model can then quickly be learned from snapshots of the flow, with sufficient expressiveness to find new optimal designs not in the training data.

Together, these approaches demonstrate the viability of using reduced-order modeling, whether physics-based or data-driven, in early-stage design to drive decision-making. This public defense will be of particular interest to building scientists, architects, and urban designers interested in advancing sustainable built environments; however, all are welcome.