Recent SBS Lab graduate Cary Faulkner’s journal article, titled “fast prediction of indoor airflow distribution inspired by synthetic image generation artificial intelligence,” was published in the journal Building Simulation. This article builds off Cary’s previous research in which he developed a conditional generative adversarial network (CGAN) model to decrease the amount of time needed to predict indoor airflow distribution, which can be prohibitively slow with computational fluid dynamics (CFD) modeling. This CGAN model uses AI image generation technology to predict airflow distribution much faster than physics-intensive CFD models. The research in this article modifies the original CGAN model by allowing a continuous input variable (such as a boundary condition) instead of a single parameter. Additionally, the new model, called BC-CGAN (boundary condition CGAN), was tested with a novel feature-driven algorithm that reduces the amount of computationally-intensive, potentially redundant data, allowing the model to be trained more quickly without sacrificing the quality of the training set. As a result, the BC-CGAN model predicted 2D air velocity and temperature distribution with less than 5% error and up to 75,000 times faster than reference CFD simulations.
Congrats to Cary on publishing this impressive research! The full paper can be found here.
