Lehrinhalte
The lecture first covers the fundamentals of nonlinear material model derivation. The main focus is on viscoelastic and plastic behavior. Next, the basics of machine learning algorithms will be covered with a special focus on artificial neural networks. The student at this point will learn how to implement simple material routines and set up a simple feed-forward neural network. We plan to use open-source packages in Python for this regard which will be taught to the students through the exercise session. Finally, we discuss how different classes of ANN can be utilized as a new method for predicting the nonlinear stress-strain response of various materials in different disciplines. Having the data from other simulations or experimental measurements, one learns how to properly apply ANN for a consistent and physical data-driven predictions.
I. Material Mechanics (MM)
Motivation and basics of material mechanics and multi-scale analysis
Derivation of a visco-elastic constitutive formulation
Derivation of a advance plasticity constitutive model
II. Machine learning (ML)
Principle of ML and data-driven approaches
Review on different types of neural networks (FFNN, RNN, CNN, etc.)
Learning how to setup an artificial neural network (architecture, loss function, minimization methods and etc.)
III. Application of Physics guided ML in MM
Idea behind physics-informed / physics guided neural networks (PINN)
Application of PINN in consistent prediction of complex material stress-strain response
Literatur
1. S. Kollmannsberger, D. D'Angella, M. Jokeit and Leon Herrmann, Deep Learning in Computational Mechanics. Springer, 2021.
Voraussetzungen
There is no specific requirement or background needed for this course.
Offizielle Kursbeschreibung
The student will learn how to derive consistent material models to predict the nonlinear stress-strain response of various types of materials at the microscale. Next, the concept behind machine learning should become clear. Furthermore, they learn how to work and when to use various types of artificial neural networks. Finally, the students will learn how to apply their understanding from the last two points to utilize new computational technologies such as ML in the prediction of complex material responses.
Online-Angebote
moodle
The lecture first covers the fundamentals of nonlinear material model derivation. The main focus is on viscoelastic and plastic behavior. Next, the basics of machine learning algorithms will be covered with a special focus on artificial neural networks. The student at this point will learn how to implement simple material routines and set up a simple feed-forward neural network. We plan to use open-source packages in Python for this regard which will be taught to the students through the exercise session. Finally, we discuss how different classes of ANN can be utilized as a new method for predicting the nonlinear stress-strain response of various materials in different disciplines. Having the data from other simulations or experimental measurements, one learns how to properly apply ANN for a consistent and physical data-driven predictions.
I. Material Mechanics (MM)
Motivation and basics of material mechanics and multi-scale analysis
Derivation of a visco-elastic constitutive formulation
Derivation of a advance plasticity constitutive model
II. Machine learning (ML)
Principle of ML and data-driven approaches
Review on different types of neural networks (FFNN, RNN, CNN, etc.)
Learning how to setup an artificial neural network (architecture, loss function, minimization methods and etc.)
III. Application of Physics guided ML in MM
Idea behind physics-informed / physics guided neural networks (PINN)
Application of PINN in consistent prediction of complex material stress-strain response
Literatur
1. S. Kollmannsberger, D. D'Angella, M. Jokeit and Leon Herrmann, Deep Learning in Computational Mechanics. Springer, 2021.
Voraussetzungen
There is no specific requirement or background needed for this course.
Offizielle Kursbeschreibung
The student will learn how to derive consistent material models to predict the nonlinear stress-strain response of various types of materials at the microscale. Next, the concept behind machine learning should become clear. Furthermore, they learn how to work and when to use various types of artificial neural networks. Finally, the students will learn how to apply their understanding from the last two points to utilize new computational technologies such as ML in the prediction of complex material responses.
Online-Angebote
moodle
- Lehrende: Anonym
- Lehrende: Baixiang Xu
Semester: ST 2022