Skip to content

Predicting the pressure and velocity fields of a fluid using the RANS approach with Physics-Informed Neural Networks, specifically Physics Residual Adaptive Networks (PirateNET).

Notifications You must be signed in to change notification settings

Gokay1904/PINN-TurbulentFlow-PirateNET

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Estimation of Turbulent Flow Using Physics-Informed Neural Networks (PINNs)

Overview

This repository contains code and datasets for estimating turbulent flow using Physics-Informed Neural Networks (PINNs). The project focuses on solving the Navier-Stokes equations for fluid dynamics and modeling velocity fields around periodic hills using PirateNET, a specialized deep learning framework. The study aims to demonstrate the integration of physical laws into neural networks for accurate turbulence modeling.

Purpose

  • Solve the Navier-Stokes equations for fluid flow simulations.
  • Use PINNs to model complex phenomena in fluid dynamics.
  • Compare the performance of PirateNET against traditional methods like Multi-Layer Perceptrons (MLPs).
  • Validate model predictions using key metrics such as L2 loss and L2 accuracy scores.

Key Methods

  • Physics-Informed Neural Networks (PINNs): Integrating physical principles (Navier-Stokes) directly into the training process.
  • PirateNET: Residual adaptive network used for fluid dynamics modeling.
  • Optimization:
    • Adam Optimizer: Used for training the model.
    • L-BFGS Optimizer: Employed for fine-tuning.
  • Accuracy Evaluation:
    • L2 Loss: A common metric to measure model accuracy.
    • L2 Accuracy Scores: Used to assess model's ability to predict velocity fields.

Data

  • The dataset includes simulated fluid dynamics data based on turbulence models: k-ε, k-ω, k-ε-ϕt-f, k-ω SST.
  • Focus on the periodic hills configuration for turbulence modeling.

Applications

  • Computational Fluid Dynamics (CFD): Solving complex flow simulations.
  • Physics-Informed Deep Learning: Integrating physical laws into AI models.
  • Turbulence Modeling: Predicting and analyzing turbulent flows with deep learning.
  • Data Efficiency: Reducing the need for large datasets by incorporating physics.

Keywords:

Physics-Informed Neural Networks (PINNs), PirateNET, Navier-Stokes Equations, Fluid Dynamics, Turbulence Modeling, Velocity Fields, Deep Learning, Artificial Intelligence (AI), Optimization, Adam Optimizer, L-BFGS Optimizer, Residual Networks, L2 Loss, L2 Accuracy Scores, Turbulence Models, k-ε, k-ω SST, Computational Fluid Dynamics (CFD), Data Efficiency, Machine Learning, Physics-Informed Deep Learning, Simulation Data, Turbulent Flow Prediction.

About

Predicting the pressure and velocity fields of a fluid using the RANS approach with Physics-Informed Neural Networks, specifically Physics Residual Adaptive Networks (PirateNET).

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published