AI surrogate model for room ventilation trained on CFD
Simulation661 MB
Input case list for CFD runs, ML inputs and outputs, a trained neural network and an inference script for predicting room ventilation conditions, including humidity. Useful as a ventilation surrogate-modelling example.
CC BY 4.0Free to use and share, including commercially, as long as you credit the creators.
Size
661 MB
How to cite
Grupo Fluidos Multifásicos (GFM), Universitat Jaume I (UJI). (2025). AI-Driven Surrogate Model for Room Ventilation dataset [Data set]. Zenodo. https://doi.org/10.5281/zenodo.15025264
Check the repository for the citation its authors prefer.
Processed URANS outputs (regional temperature rise, passive tracer, mean-field sections), surrogate models, reference geometry, input dictionaries and analysis code comparing fan allocations at equal power in a normalised industrial hall. Useful for industrial ventilation design and surrogate modelling.
Air velocities measured on an HTM 03-01 grid at 1 m and 2 m below the canopy (three repeats), processed means and uncertainties, and the CAD geometry. Useful for validating CFD of ultra-clean ventilation airflow and particle entrainment.
Field CO2 and PM2.5 measurements in a 71 m3 occupied office, with PINN training/analysis code and weights used to derive air change rate, penetration factor and field CADR. Useful for ventilation-rate inference and indoor particle modelling.
Environment, training and evaluation code, 776 trained RL policies and PINN CO2/PM2.5 surrogate weights for schedule-aware control of windows, doors and an air purifier. Useful for ML-based indoor ventilation control studies.
Data files for several air purifier positions in two elevator cabin models, giving mean age of air on a plane at 1.6 m and infiltration results. Useful for ventilation effectiveness studies and for validating indoor airflow models in small enclosures.
From Zenodo, added 2026-11-15.
Listed there as “AI-Driven Surrogate Model for Room Ventilation dataset”. Tags proposed by AI and not yet checked by hand. Spotted a mistake? Tell the portal contacts.