Physics Informed Neural Networks (PINNs) for rapid contamination dispersion predictions

This paper investigates the application of Physics-Informed Neural Networks (PINN) in predicting contamination dispersion, contrasting it with traditional Computational Fluid Dynamics (CFD). Notably, PINN demonstrates advantages when boundary conditions are unknown or frequently changing, a scenario often encountered in emergency response missions. By integrating appropriately positioned sensors, PINN can rapidly estimate dispersion maps. The study utilizes 2D case to analyze the strengths and limitations of sensor-augmented PINN versus traditional grid-based CFD. The findings reveal that for simple 2D flows, the convergence time of PINN is comparable to that of CFD. However, PINN's independence from critical boundary conditions such as the location of the source release and wind conditions renders it particularly attractive for specific applications.