Machine Learning-Based Intelligent CFD Surrogates for Interactive Design Exploration of Built Environments

This study focuses on the application of reduced order models (ROMs) and various machine learning (ML) techniques to create surrogate computational fluid dynamics (CFD) models for the built environment. By using functional principal component analysis (FPCA), these surrogates can reconstruct flow fields for different geometries significantly faster than traditional CFD methods, providing insights into factors such as thermal comfort and ventilation. The goal is to develop a model that can predict the effects of different geometries or indoor space configurations and produce accurate fluid/air flow results without the need for retraining, enabling rapid, interactive design exploration and optimization.