Using Physics-Informed Deep Learning for Transport in Porous Media

This is success story of modeling the flow and transport in porous media using NVIDIA PhysicsNeMo.

NVIDIA logo
NVIDIA
•
Cedric G. Fraces
•7 min read•intermediate•
--
•View Original
Deep LearningDockerKerasPythonTensorFlow

Overview

The article discusses the use of NVIDIA PhysicsNeMo, a simulation toolkit that integrates AI and physics to model flow and transport in porous media. It highlights a case study by Cedric Frances, a PhD student, who explores mesh-free reservoir simulations using physics-informed neural networks (PINNs) to improve efficiency and accuracy in various industrial applications.

What You'll Learn

1

How to utilize NVIDIA PhysicsNeMo for simulating flow in porous media

2

Why physics-informed neural networks (PINNs) are beneficial for reservoir simulations

3

When to apply mesh-free simulation techniques in engineering projects

Prerequisites & Requirements

  • Understanding of physics-informed neural networks
  • Familiarity with Python and deep learning frameworks like TensorFlow and Keras(optional)

Key Questions Answered

What is NVIDIA PhysicsNeMo and how does it assist in simulations?
NVIDIA PhysicsNeMo is a simulation toolkit that combines AI and physics to streamline the simulation process in various scientific and engineering domains. It addresses challenges like high computational times and complex setup efforts, allowing researchers to focus more on the physics of their problems.
How did Cedric Frances use PhysicsNeMo in his research?
Cedric Frances utilized PhysicsNeMo to implement fully connected networks with tanh activation functions for modeling fluid flow in porous media. This framework enabled him to efficiently test various methods and achieve quicker results compared to previous implementations.
What challenges did researchers face when using PINNs for porous media simulations?
Researchers previously struggled with using PINNs to solve hyperbolic problems with nonconvex flux terms, leading to unsuccessful attempts. Cedric's work highlights how PhysicsNeMo helped overcome these limitations by providing a more robust framework.
What are the applications of the transport problem in porous media?
The transport problem in porous media has applications in various fields, including hydrocarbon production, carbon dioxide storage, and waste management. Cedric's research focuses on improving predictions related to these industrial energy projects.

Technologies & Tools

Some links below are affiliate links. We may earn a commission if you make a purchase.

Simulation Toolkit
Nvidia Physicsnemo
Used for modeling flow and transport in porous media
Programming Language
Python
Used to develop the reservoir simulator
Deep Learning Framework
Tensorflow
Utilized for implementing PINNs
Deep Learning Framework
Keras
Also used for implementing PINNs

Key Actionable Insights

1
Leverage NVIDIA PhysicsNeMo to streamline your simulation processes, reducing setup time and complexity.
Using a unified framework like PhysicsNeMo allows engineers to focus on the physics of their problems rather than the technicalities of implementation, which can lead to significant time savings.
2
Explore the capabilities of physics-informed neural networks (PINNs) for solving complex transport problems in porous media.
PINNs can provide more accurate predictions in scenarios where traditional methods struggle, making them a valuable tool for researchers and engineers in the field.

Common Pitfalls

1
Many researchers face difficulties in implementing various neural network architectures due to high overhead in environment setup and configuration.
This can lead to inefficiencies and wasted time. Using a well-maintained framework like PhysicsNeMo can help mitigate these issues.

Related Concepts

Physics-informed Neural Networks (pinns)
Mesh-free Simulation Techniques
Flow And Transport In Porous Media