evansoong517/pinns

Solving diffusion equation using physics informed neural network.

PINNs

Application of neural networks to improve the modelling of cleaning processes

Solving diffusion equations for the cleaning processes in the food industry using numerical methods is a recursive task and demanding in terms of both simulation duration and computer processing power. Current studies have shown that neural networks can solve differential equations, thus bringing rise to the physics informed neural network (PINN) framework. This ability is exploited in a novel data-driven approach to the diffusion equations presented in this paper. The PINN is applied in solving the diffusion equations given its only initial and boundary conditions. This research project aims to apply PINN to solve the diffusion equations to significantly decrease the time to the solution while preserving the accuracy of traditional numerical methods and then deploy it in a C++ environment. PINN for both one and two dimensions have been proposed, and their corresponding numerical solutions and PINN prediction results have been analyzed. PINN is suggested to be more promising than the numerical method for solving PDEs. The same was validated in the current research project while solving diffusion equations, which was true. It is assumed that integrating PINN and C++ can transform the physics simulation area by allowing real-time physics simulations in a cost and time-effective manner.