This project developed a practical, AI-enabled framework to quantify and reduce the risk of airborne infection in offices using simulations and data-driven design.

The team created a new probability-of-infection metric by combining high-fidelity computational fluid dynamics (CFD) with models for exhaled CO₂ and the “age of air”. The age of air measures how effectively fresh air reaches occupants.

The methodology combines particle emissions, exposure time and clinical viral-load data to produce spatial and temporal risk maps that evolve over time.

The study then trained machine-learning models to predict infection risk using two easily measured inputs: indoor CO₂ concentration and the supplied ventilation rate.

In a validated office case, the analysis identified ventilation effectiveness, rather than simply the volume of outside air supplied, as the key factor. Recirculation zones can create hot spots of stale air and trapped aerosols.

Contour plots on page 10 show how quanta, or infectious particles, accumulate in areas where airflow stalls. These areas are not always identified by CO₂ measurements alone.

Volume-averaged curves show that, under the tested layouts, infection probability can rise quickly despite nominal ventilation requirements being met. This highlights the importance of diffuser placement and effective airflow distribution.

For the data-driven element, an optimised random forest model predicted the CFD-derived infection risk with a high level of accuracy. It achieved a coefficient of determination of approximately 0.99 and a low root mean square error.

This supports a closed-loop system in which smart building technology can estimate infection risk using CO₂ sensors and adjust ventilation in real time, without running computationally expensive simulations.

The approach also supports AI-enabled simulation and data-driven design workflows. Surrogate models and sensor data can help teams rapidly assess diffuser layouts, occupancy policies and ventilation control settings.

Beyond offices, the framework could contribute to digital twins for schools, hospitals and public venues. This could accelerate the development of safer, energy-efficient ventilation systems while reducing modelling requirements.