A Convolutional Neural Network Based Approach for Computational Fluid Dynamics

Abstract

Computational fluid dynamics (CFD) is the visualisation of how a fluid moves and interacts with things as it passes by using applied mathematics, physics, and computational software. The project is designed to simulate fluid flow of a particle based on provided boundary conditions using High Performance Computing (HPC), with two-dimensional picture files as input to the software and fluid flow of a particle generated based on these image data. The Naiver Stokes Equation and the Lattice Boltzmann Equation are used to create these fluid flow particles.The governing equations based on the conservation law of fluid physical characteristics lead the primary structure of thermofluids investigations. Fluid flow is created depending on the item using the three governing equations from the conservation laws of physics. CFD simulation, on the other hand, which is a Iterative process is frequently computationally costly, memory-intensive, and time-consuming. A model based on convolutional neural networks, is proposed for predicting non-uniform flow in 2D to over come these disadvantages. The proposed approach thus aims to aid the behaviour of fluid particles on a certain system and to assist in the development of the system based on the fluid particles that travel through it. At the early stages of design, this technique can give quick feedback for real-time design revisions. In comparison to previous approximation methods in the aerodynamics domain, CNNs provide for efficient velocity field estimate and took less time then the previous approximation method. As CFD based CNN model is more effective to 2D design(2D aeroplane dataset) as it is in research stage lot more experiments have to be made for 3D designs. Designers and engineers may also use the CFD based CNN model directly in their 2D design space exploration.