CFD–Machine Learning Optimization of High-Rise Airflow and Pollution

Field

Architecture, Urban Physics and Environmental Engineering

Study type

Computational design optimization study

Challenge

The study needed to connect building geometry, severe winter air-pollution conditions, CFD performance metrics, explainable machine learning, and multi-objective optimization without inventing simulation results.

Approach

A parameterized design-of-experiments workflow linked geometry variables to CFD, ventilation/pollution metrics, machine-learning surrogates, SHAP interpretation, NSGA-II optimization, and CFD verification of selected solutions.

Methods

Design of experiments · Computational Fluid Dynamics · Machine learning surrogate modelling · SHAP · NSGA-II · Sensitivity analysis

Software / workflow

ANSYS Fluent · Python

Project outcome

A reproducible end-to-end research framework and figure architecture were established, with explicit separation between measured/climate inputs, simulated outputs, ML interpretation, optimization, and final CFD verification.

This case study is intentionally anonymized. Client names, institutions, exact titles, locations, and confidential source materials are not published.