A physics-informed data-driven model applied for gas dispersion

Abstract

Gas dispersion calculations are essential for numerous applications. While the gas flow behaviour can be theoretically described by the Navier–Stokes equations, obtaining numerical solutions poses significant computational challenges, due to the demanding computational time involved. In this study, we tackle these challenges by leveraging the power of physics-informed neural networks (PINNs). PINNs integrate the underlying physics of the problem directly into the architecture of the neural network. By incorporating the Navier–Stokes equations within the framework of neural networks, our approach accounts for the fundamental physics governing gas dispersion. We use an in-house Computational Fluid Dynamics (CFD) code and commercial software to generate the required dataset. Our results demonstrate that the model is robust and capable of providing rapid solutions to gas dispersion problems. This efficiency is particularly noteworthy when compared to the considerable computational time required for traditional CFD calculations. Therefore, our approach offers a promising alternative for efficient and accurate gas dispersion simulations in process safety applications.