Air quality in educational environments significantly impacts students’ cognitive performance and well-being. Hence, developing accurate and swift predictive models is essential for mitigating indoor air quality risks during the early-stage design of ventilation system duct layouts. This study investigates the potential of Conditional Generative Adversarial Networks (CGANs) as surrogate models for predicting airflow and temperature distribution, offering a significantly faster alternative to conventional computational fluid dynamics (CFD) simulations. A validated CFD model of a reference classroom serves as the baseline for generating datasets by varying air inlet locations. The pix2pix architecture is trained on paired image datasets, with model performance evaluated using the Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR). The results demonstrate prediction accuracies of up to 0.89 SSIM and 27.7 PSNR for temperature distribution, and 0.91 SSIM and 25.5 PSNR for airflow velocity. Notably, the trained models synthesize flow field images in under one second, compared to the two-hour runtime of conventional steady-state, non-isothermal RANS CFD simulations on standard hardware. This significant reduction in computation time highlights the potential of CGAN-based models as valuable decision-support tools, facilitating the exploration of more ventilation system designs in the early design stages and promoting healthier, energy-efficient indoor environments.
Moradi et al. (2026) studied this question.