Generative Adversarial Network for Optimizing Air Flow and Particulate Dissipation in Buildings

Due to an increased focus on public health, building optimization has become more common. One example is air quality: a building can be designed for optimum air flow to more effectively dissipate unhealthy air. This is beneficial in protecting the health of occupants in the event of fires, epidemics, and chemical release. Typically, designers must use Computational Fluid Dynamics (CFD) to solve turbulent flow with advection-diffusion. However, CFD relies on slow numerical methods and multiple iterations via parametric modeling, making it extremely slow. Instead, our research has discovered a faster way using Generative Adversarial Networks (GANs). GANs are magnitudes faster than CFD because they represent the solution operator to the entire system, instead of iterating through every step. As a proof-of-concept, we developed a GAN that can accept any arbitrary building design and solve for the subsequent smoke dissipation nearly $\mathbf{5 0, 0 0 0}$ times faster than CFD simulation.