Goals of the Website

Goals

  • Introduce Surrogates: Give an introduction to surrogates of physics-based models; initially, we will focus on steady-state systems.
  • Explore Surrogate Architectures: Create surrogates for various configurations, scales, and situations.
  • Provide Compelling Use Cases: Use the created surrogate model to solve exciting engineering challenges.
  • Collaborate: Provide a space for collaboration and interaction.

Why use MATLAB?

Engineering professionals widely use MATLAB in industry, and personal licenses are available for hobbyists. Simulink also provides the flexibility to create different systems for a variety of domains. Furthermore, its Deep Learning Toolbox provides all the necessary functionality for creating surrogate models. However, the methods and ideas discussed are generally applicable.

What are Surrogate Models?

Neural Network Surrogate Models

Physics-based surrogate models are computational models that approximate the behavior of complex physical systems using simplified mathematical equations derived from the underlying physics. These models capture the essential dynamics and interactions within a system while reducing computational complexity compared to full-scale simulations or solvers. In the context of neural networks, these models leverage the expressive capacity of neural networks to approximate the behavior of complex physical systems while ensuring adherence to fundamental physical laws. Physics-based neural network surrogate models offer enhanced accuracy and generalization capabilities by integrating domain knowledge into the neural network architecture or training process. Physics-based neural network surrogate models enable efficient representation and prediction of system dynamics, supporting tasks like optimization, uncertainty quantification, and control across various engineering, physics, and materials science domains.

Benefits of Surrogate Models

The articles on this website showcase the following benefits:
  1. Getting Answers Quicker: We show surrogates to be much quicker than their physics-based counterparts (200x-50,000x)
  2. Expanded Workflows: Surrogates improve or enable vital engineering workflows such as operations optimization and parameter estimation
  3. Portability: The surrogate models have a small footprint, can be executed in diverse environments, and only require basic mathematical operations to solve