Surrogate models for uncertainty quantification in engineering sciences
OPEN ACCESS
Loading...
Author / Producer
Date
2024-11-12
Publication Type
Other Conference Item
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
Computational models play a crucial role in modern engineering and applied sciences by enabling the prediction and optimization of natural and man-made systems. These models, such as high-fidelity finite element simulations, help engineers evaluate performance virtually, reducing the need for costly physical testing. However, the complexity and computational expense of running these simulations, especially when accounting for uncertainties in their governing parameters, environmental conditions, and operational variables, present significant challenges. Traditional methods like Monte Carlo simulations, while accurate, are prohibitively time-consuming, requiring a vast number of simulations for reliable results.
To address these challenges, surrogate modelling has emerged as a powerful alternative for uncertainty quantification (UQ). Surrogate models construct an efficient approximation of a simulator's output by utilizing a limited number of runs, strategically chosen through the design of experiments, and using advanced learning algorithms. This allows for significant reductions of the computational costs while preserving accuracy. In this presentation, we will introduce the fundamental principles of surrogate models and explore their ties to machine learning techniques. Focus will be placed on polynomial chaos expansions, including their sparse version designed to handle high-dimensional problems. We will also discuss recent developments in extending these methods to dynamical systems (e.g. wind turbine simulators) and present practical applications to global sensitivity analysis.
Permanent link
Publication status
published
External links
Editor
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich, Chair of Risk, Safety and Uncertainty Quantification
Event
Dutch Computational Science Day (DUCOMS 2024)
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Organisational unit
03962 - Sudret, Bruno / Sudret, Bruno