Fire Performance Prediction of Naturally Ventilated Double-Skin Façades Using CFD and Machine Learning

Double-skin façade (DSF) systems are important for energy efficiency because they effectively utilize natural ventilation and daylight. However, the uninterrupted vertical gaps in these systems may pose safety risks in the event of a fire by causing the rapid spread of smoke and hot gases. This study presents a hybrid approach that combines computational fluid dynamics (CFD)-based simulations and machine learning (ML) techniques to predict heat flow and fire-room control-volume heat release rate (FR-HRR). Within the scope of the study, 400 different scenarios were modeled with different combinations of basic natural ventilation design parameters consisting of gap width, gap height, window opening area, and air inlet and outlet area. The data obtained were evaluated with different ML models, including Fine Tree, Bagged Tree, Support Vector Machine, and Artificial Neural Network models; in particular, the Fine Tree model gave the most successful results with high accuracy rates (R2 = 0.99 for FR-HRR; R2 = 0.91 for heat flow). The analysis showed that DSF gap width provided a dominant model-based contribution within the investigated CFD-generated dataset. This approach provides a preliminary CFD-informed ML framework for the rapid comparative assessment of fire-related responses in open-boundary naturally ventilated DSF configurations during the early design stage.