W24-07 Automated Model Discovery
Instructor(s): Ellen Kuhl, Stanford University; Skyler St. Pierre, Sanford University; Mathias Peirlinck, TU Delft, Kevin Linka, RWTH Aachen
Description:
In this course, you will learn how to discover models from data using machine learning enriched by the fundamental laws of physics. For more than 100 years, chemical, physical, and material scientists have proposed competing models to best characterize the behavior of soft matter systems in response to mechanical loading. Now, computer science offers a universal solution: neural networks. This course introduces a special family of neural networks, constitutive neural networks, that inherently satisfy common kinematic, thermodynamic, and physical constraints and, at the same time, constrain the design space of admissible functions to robustly discover interpretable models with physically meaningful parameters, even in the presence of sparse data. You will receive a library of discovery notebooks to implement and train your own constitutive neural networks and analyze and interpret classical benchmark data of man-made materials like rubber and living materials like skin, muscle, arteries, and the human brain.
You are welcome to bring your own data!
You will observe that your constitutive neural network autonomously selects the best material model, parameters, and experiment to characterize your material.
You will learn
- to implement and train your own constitutive neural networks
- to discover the best models, parameters, and experiments from data
- to translate model discovery into finite element analysis
- to analyze and interpret classical benchmark data of soft matter systems
