August 29th – September 1st, 2022
Organizers:
- Silesian University of Technology (SUT), Gliwice, PL
Priority Research Area #1 Computational Oncology and Personalized Medicine - Norwegian University of Science and Technology (NTNU), Trondheim, NO
under auspices of the Committee of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences
Participation in the school is free of charge. The school is addressed to PhD students and young researchers interested in vascular flow simulations. The working language will be English. Basic knowledge of MatLAB, Python and ANSYS is required.
The number of participants is limited and cannot exceed 25 in the theoretical part (lectures) and 15 individuals in the hands-on training. Priority will be given to PhD students of SUT and NTNU. The organizers reserve the right to select participants based on the application form submitted.
Registration form (deadline July 20th, 2022):
https://forms.office.com/r/4hUtstDkhN
Contacts:
Prof. Ziemowit Ostrowski, Summer School Coordinator (ziemowit.ostrowski@polsl.pl)
Prof. Ryszard Białecki, Project Coordinator & PI (ryszard.bialecki@polsl.pl)
Summer School Program
Inverse Methods
Instructors: Prof. Ryszard Białecki, Prof. Ziemowit Ostrowski
SUT, Gliwice, PL
2 h lectures + 4 h assignment to simulate examples
Software: MatLAB, Python
Direct boundary value problems have a unique solution, provided that a set of data is known, known as the unambiguity conditions. This set encompasses the geometry of the domain, the boundary and initial conditions, material properties, and internal sources that act in the domain. The solution is a physical field, such as velocity, pressure, temperature, etc. If any element of the unambiguity data is unknown, the problem does not have a unique solution. To produce a unique solution, additional data are included in the process. These data come from the measurements of the fields that would be the solution to the associated direct problems. Computed tomography, image deblurring, and determination of unknown tissue properties belong to the class of inverse problems.
Topics:
- Definition and classification of inverse problems
- Examples of applications
- Ill posendess and regularization
- Methods of solution
- Solution of simple numerical examples.
Sensitivity Analysis and uncertainty quantification via the DAKOTA Code
Instructor: Prof. Adam Klimanek
SUT, Gliwice, PL
2 h lectures + 4 h assignment to simulate examples
Software: Dakota, Python, Matlab
Numerical simulations of physical processes yield the fields with a certain accuracy. The model parameters, as well as boundary and initial conditions, are uncertain and contribute to the uncertainty of the model predictions. The goal of sensitivity analysis is to identify the extent to which simulation results depend on model input and to determine the most important input data. The goal of uncertainty quantification is to reveal the variability and statistics of model predictions, which allow assessing the probability of typical and extreme outcomes. Using the methods, the question of how reliable the obtained results are can be addressed, which is a crucial aspect of engineering. Dakota, a Sandia Lab freeware, offers a flexible interface between the user’s simulation code and iterative methods for uncertainty quantification, design optimization, parameter estimation, design of experiments, and sensitivity analysis.
Topics:
- Sensitivity Analysis
- Uncertainty Quantification
- Solution methods
- Applications
- Solution of simple numerical examples
Machine learning
Instructors: Prof. Joanna Polańska, Dr Michał Marczyk, Aleksandra Suwalska
SUT, Gliwice, PL
2 h lectures + 4 h assignment to simulate examples
Software: Python
The classic approach to building a mathematical model requires detailed knowledge of the processes involved. Often, however, we do not have such knowledge, or it is limited, and the analyzed phenomena are complex and/or highly nonlinear. Then, machine learning techniques, especially those using artificial neural networks (ANN), can be helpful. Regression ANNs predict an output variable as a function of the inputs. The input (explanatory) variables can be categorical or numerical, but we require a response variable to be numeric. We will focus on the regression analysis using ANN in the TensorFlow environment. The hands-on example will support the theoretical knowledge presented during the lecture.
Topics:
- Statistical versus machine learning approach
- Structure of artificial neural networks
- Model training and optimization
- Performance evaluation scenarios
Fluid-structure interaction
Instructors: Prof. Joris Degroote, Prof. Patrick Segers
Ghent University, Ghent, BE
2 h lectures + 4 h assignment to simulate examples
Software: Ansys Workbench
The interaction between a fluid flow and a moving or deforming structure is called fluid-structure interaction (FSI). In the arterial system, there is FSI between the blood flow and the arterial walls. This can be simulated in a partitioned way, by coupling a computational fluid dynamics (CFD) solver with a computational structural mechanics (CSM) solver. However, robust coupling techniques have to be applied to stabilize the coupling iterations between the solvers. Furthermore, an accurate simulation requires insight in the physics of the problem, with attention for the boundary conditions, the geometry from medical imaging, the meshing and the tissue properties.
Topics:
- Definition and classification of fluid-structure interaction problems
- Techniques for a deforming fluid domain
- Added mass effect
- Coupling techniques for partitioned simulation
- Wave propagation in elastic arteries
- Patient-specific simulations based on medical imaging
Download:
Summer School Flyer and Programme is available for download here (https://enthral.pl/wp-content/uploads/2022/06/ENTHRAL_SummerSchool_2022_flyer_v1_09062022_h1511-1.pdf)