Fast prediction of indoor airflow distribution inspired by synthetic image generation artificial intelligence

Prediction of indoor airflow distribution often relies on high-fidelity,computationally intensive computational fluid dynamics(CFD)simulations.Artificial intelligence(Al)models trained by CFD data can be used for fast and accurate prediction of indoor airflow,but current methods have limitations,such as only predicting limited outputs rather than the entire flow field.Furthermore,conventional Al models are not always designed to predict different outputs based on a continuous input range,and instead make predictions for one or a few discrete inputs.This work addresses these gaps using a conditional generative adversarial network(CGAN)model approach,which is inspired by current state-of-the-art Al for synthetic image generation.We create a new Boundary Condition CGAN(BC-CGAN)model by extending the original CGAN model to generate 2D airflow distribution images based on a continuous input parameter,such as a boundary condition.Additionally,we design a novel feature-driven algorithm to strategically generate training data,with the goal of minimizing the amount of computationally expensive data while ensuring training quality of the Al model.The BC-CGAN model is evaluated for two benchmark airflow cases:an isothermal lid-driven cavity flow and a non-isothermal mixed convection flow with a heated box.We also investigate the performance of the BC-CGAN models when training is stopped based on different levels of validation error criteria.The results show that the trained BC-CGAN model can predict the 2D distribution of velocity and temperature with less than 5%relative error and up to about 75,ooo times faster when compared to reference CFD simulations.The proposed feature-driven algorithm shows potential for reducing the amount of data and epochs required to train the Al models while maintaining prediction accuracy,particularly when the flow changes non-linearlywith respectto an input.

Quantifying spatiotemporal variability in occupant exposure to an indoor airborne contaminant with an uncertain source location

Well-mixed zone models are often employed to compute indoor air quality and occupant exposures.While effective,a potential downside to assuming instantaneous,perfect mixing is underpredicting exposures to high intermittent concentrations within a room.When such cases are of concern,more spatially resolved models,like computational-fluid dynamics methods,are used for some or all of the zones.But,these models have higher computational costs and require more input information.A preferred compromise would be to continue with a multi-zone modeling approach for all rooms,but with a better assessment of the spatial variability within a room.To do so,we present a quantitative method for estimating a room’s spatiotemporal variability,based on influential room parameters.Our proposed method disaggregates variability into the variability in a room’s average concentration,and the spatial variability within the room relative to that average.This enables a detailed assessment of how variability in particular room parameters impacts the uncertain occupant exposures.To demonstrate the utility of this method,we simulate contaminant dispersion for a variety of possible source locations.We compute breathing-zone exposure during the releasing(source is active)and decaying(source is removed)periods.Using CFD methods,we found after a 30 minutes release the average standard deviation in the spatial distribution of exposure was approximately 28%of the source average exposure,whereas variability in the different average exposures was lower,only 10%of the total average.We also find that although uncertainty in the source location leads to variability in the average magnitude of transient exposure,it does not have a particularly large influence on the spatial distribution during the decaying period,or on the average contaminant removal rate.By systematically characterizing a room’s average concentration,its variability,and the spatial variability within the room important insights can be gained as to how much uncertainty is introduced into occupant exposure predictions by assuming a uniform in-room contaminant concentration.We discuss how the results of these characterizations can improve our understanding of the uncertainty in occupant exposures relative to well-mixed models.