Contents
- Physics-aware machine learning (ML) combines classical, physics-based modeling approaches with ML methods to improve the generalization capabilities, interpretability, robustness, reliability and efficiency of ML methods in engineering applications
- Introduction to ML methods and their essential theoretical properties, including in particular artificial neural networks (approximation capabilities, training, gradients, etc.)
- Foundations of physics-based modeling and simulation using differential equations and suitable temporal and spatial discretization methods (time integration and finite elements)
- Physics-based and data-driven model order reduction and surrogate modeling (e.g. modal analysis, orthogonal decompositions, kriging, kernel methods, etc.)
- Mathematical knowledge representations of conservation equations & quantities, symmetries, invariances, etc. for physics-aware ML
- Construction principles for informing or augmenting ML methods through appropriate design of training data, hypotheses for input and output variables of ML models, ML model architectures, or learning or training algorithms
- Methods include e.g. Sobolev training, convex & monotonic NNs, physics informed NNs (PINNs), Langrangian NNs, neural operators, stochastic NNs, recurrent NNs, convolutional NNs, graph NNs, autoencoders, generative NNs, Gaussian processes & kernel methods, etc.
- Applications and examples for solid mechanics, structural dynamics, material modeling, dynamic systems, multiscale and multiphysics problems, (additive) manufacturing processes, digital twins, etc.
Summer term 2024
The lecture is offered for the first time in summer term 2024.
Details
Usability of this module
- Master in Mechanical Engineering (Electives Area II)
- Master in Aerospace Engineering
- Master in Computational Engineering
- Master in Mechanics
- Master in Mechatronics