Hybrid framework for rapid evaluation of wind environment around buildings through parametric design, CFD simulation, image processing and machine learning

Abstract

High-efficient evaluations of building performance are often required for comparisons of different design alternatives in architectural sustainable design processes. General Computational Fluid Dynamics (CFD) simulations are usually complicated and time-consuming for wind environment investigation and evaluation. A hybrid framework for rapid evaluation of pedestrian-level wind environment will be proposed in the present work. This framework will then be formulated by integrating parametric design, CFD simulation, image processing, and machine learning, and it could immediately predict the Low-Velocity Areas (LVAs) around rectangular-form buildings. A large amount of data of 300 building cases generated by parametric design, CFD simulation, and image processing to train a Machine Learning Model (MLM) could be applied for the prediction of LVAs. In the case investigations, MLM was tested in the prediction of the other new 24 building cases with random geometric parameters. The comparison of MLM and CFD results showed that their solutions were close to each other. Efficiency and accuracy of the hybrid framework were further demonstrated through quantitative analysis of statistical discrepancies of MLM and CFD results. Hybrid framework was an original attempt to integrate multiple emerging computational tools, and it could provide high-efficient quantitative analysis of wind environment and give practical design optimization information in the early stage.