13/12/2023 Shervin Bagheri, https://www.kth.se/profile/sherwinb

Title: Boundary conditions between flows and complex surfaces

Abstract:

The no-slip condition is evidently the appropriate macroscopic boundary condition for flows over simple surfaces. However, realistic surfaces have texture, chemical contrasts, fouling, pores, and compliance, for which we mostly lack effective boundary conditions. This limits our ability to efficiently model flows over complex surfaces and for describing transport processes between domains that separated by an interface. In this talk, we describe measures and boundary conditions for increasingly complex fluid-surface systems, including inhomogeneous (e.g. change in chemistry), textured and porous interfaces. We discuss the regime of validity and high-order extensions of the boundary conditions for modelling Stokes, laminar and turbulent flows. Finally, we present applications, where different forms of the macroscopic boundary conditions enable us to; i) discover new physics, such as how small surface contrasts modify the lubrication forces on particles; ii) accurately model problems with engineering relevance, such as transport processes between free flows and porous materials, and finally; iii) devise semi-empirically models for turbulent flows, such as rough-wall turbulence.

6/12/2023 Mark Carpenter, https://scholar.google.com/citations?user=PPRgFsIAAAAJ&hl=en

Title: Intrastep, Stage-Value Predictors for Diagonally-Implicit Runge--Kutta Methods

Abstract:

To better identify the necessary attributes of good stage-value predictors (SVPs), numerous SVPs are designed for an existing: ESDIRK4(3)7L[2]SA (Kennedy2019) and a new: ESDIRK4(3)8L[2]SA (eight stages) scheme. Both are stiffly-accurate, stage-order two, explicit, singly-diagonally implicit Runge--Kutta (ESDIRK) schemes. Tradeoffs are studied in the parameter spaces enforcing the constraints on accuracy, linear stability, nonlinear stability and coefficient size to determine which objectives correlate with effective predictors. The SVPs are tested in challenging external aerodynamics simulations ($10^7$ DoFs) using the compressible Navier-Stokes equations (CNSE). An entropy stable spectral collocation formulation is used for discretizing the spatial terms in the equations. Simulations are performed at a wide variety of temporal error tolerances.

At lax temporal error tolerances, the most efficient SVPs are those designed with second-order accuracy and the stability properties: A-stability, L-stability, rather than high accuracy constraints. Simulations requiring tight error tolerances are better suited for SVPs designed using high accuracy constraints. Designing SVPs with enhanced stability properties is tedious but worthwhile. Simulation times are reduced with optimal SVPs by as much as 100% on some stages, with combined stepwise improvements of between 50--100% for both methods. A comparative study is performed with the two aforementioned methods as well as four other ESDIRKs. The newly designed ESDIRK4(3)8L[2]SA with $\gamma \approx 1/10$, proves to be the most efficient of the six tested ESDIRK schemes simulating the CNSE.

29/11/2023 Mattias Liefvendahl, see bio below

Title: Computational Challenges in Naval Hydrodynamics

Abstract:

Marine applications and naval hydrodynamics require the solution of a range of challenging fluid dynamics problems, including turbulent very high-Re-number flows, two-phase flow (cavitation, bubbly flows and waves) and maneuvering simulations, possibly including rudder and propeller motion. The presentation will give an overview of a number of difficult modelling problems and outline the current simulation state-of-the art. An application-oriented perspective is taken to motivate the relevance and indicate the parameter ranges of interest for the different mathematical models.

Short speaker bio:
- Ph.D. in Numerical Analysis, Royal Institute of Technology, Stockholm, Sweden, 2001.
- Post-doc., EU Marie Curie Industry Host Fellowship, Mecalog SARL/Ecole centrale, Paris, 2001-2003.
- Research Director Hydrodynamics, Swedish Defence Research Agency (FOI), 2003-2021.
- Lead Researcher Naval Research, RISE Research Institutes of Sweden AB, 2021-present.
- Adjunct professor in scientific computing, Uppsala University, 2013-2018.

Research interests: Theoretical and computational hydrodynamics, with application to both design and operational aspects of marine platforms.

22/11/2023 Ida-Maria Sintorn, https://www.katalog.uu.se/empinfo/?id=N0-606

Title: Two attempts at bridging the gap between proof of concepts and real-world deployment of image based deep learning in microscopy

Abstract:

AI and deep learning offer the great possibility to learn relevant information from examples. It has the drawback that a priori expertise and information not represented in the training examples will not be incorporated in the model. It is also difficult to decipher and validate what information a model actually uses to reach a decision. These drawbacks practically limit AI deployment in many real-world microscopy applications. In reality one often face a scenario with limited amount of training images, lack of/unreliable ground truth, non-representative or too narrow training sets, reliability of results, and lack of explainability for trustworthiness.

I will present two tools addressing some of these shortcomings. 1) Microscopy based analyses often result in large image sets with rich but sparse information content. Rare or unexpected effects often require explorative analysis to investigate how common they are and where/when they appear. I will present a similarity search framework- SimSearch, for quick and easy user-guided training of a deep neural network. 2) Commonly used AI networks are very self-confident in their predictions, also when the evidence for a certain decision is dubious. This results in silent failing i.e., misclassifications and unreliability due to too narrow training sets, as well as missing things in the sample not directly asked for. I will describe a simple approach for out of distribution (OOD) detection that does not require retraining of the network.

15/11/2023 Stefan Widgren, https://www.sva.se/en/what-we-do/research-at-sva/researchers-at-sva/researchers/stefan-widgren

Title: Data analysis in the African swine fever outbreak

Abstract:

On September 6, Sweden's first case of African swine fever (ASF) in wild boar was confirmed. ASF is a viral disease of pigs and wild boar that causes high mortality in affected animals and there is no vaccine or treatment. During the outbreak, we collect data for epidemiological analysis and research. For example, locations of carcasses found, and manual search efforts. We would also like to assess if drones equipped with thermal sensors combined with computerised image analysis could provide an efficient, reliable, and practical method to detect living wild boar and identify carcasses. This data will be used to inform transmission models developed for ASF in the future. However, in this talk I will focus on challenges in data collection and setting up a transmission model in the acute phase of an outbreak.

8/11/2023 Steffi Burchardt https://steffiburchardt.com and Christoph Hieronymus https://www.katalog.uu.se/profile/?id=N5-1495

Title: From fieldwork to numerical modelling – how to deal with the uncertainties of geological data

Abstract:

The complex systems that comprise planet Earth are the ultimate examples of how physical laws play out in nature. However, due to the complexity and scales of the processes involved, as well as the time scales on which they unfold, Earth science research by definition struggles to quantify all parameters involved. In this talk, we give some examples from the field of volcanology, based on our long-term collaboration. Our research focusses on the physics of how magma is transported and stored inside volcanoes. We will give examples of how we conduct empirical research on extinct and eroded volcanoes to identify the physical laws behind the natural phenomena and how we identify input parameters for numerical models. The models – usually finite element models – aim to simulate physical and chemical processes that cannot be directly observed in nature, for example because they occur inside magma chambers or over millions of years. Besides examples, we will discuss challenges in dealing with the large uncertainties that comes from dealing with geological input parameters.

18/10/2023 Orcun Göksel, http://goksel.org/

Title: Numerical problems and deep learning in the modeling and simulation of imaging and biomechanics

Abstract:

In this talk, I will give highlights from our research on different numerical problems. I'll start with the modeling and simulation of biomechanics for deformable soft tissues. I will demonstrate finite-element based real-time, interactive simulators for medical training as well as the problem of generating discretized 3D anatomical models for such simulations. I will present a reinforcement learning solution for the dynamic control of complex musculoskeletal biomechanical models.

While biomechanical simulations solve forward problems, biomechanical tissue characterization such as for clinical diagnosis often involves inverse problems, aiming to estimate intrinsic tissue parameters given observations typically from medical imaging such as ultrasound. In this context and for the general solution of inverse problems, I will present variational neural networks for loop unrolling iterative optimization techniques. These can then utilize known forward models and lead to non-black-box solutions that are explainable and generalizable.

4/10/2023 Anders Bergman, https://www.katalog.uu.se/empinfo/?id=N1-1159

Title: Numerical modelling of magnetic phenomena at the atomic scale

Abstract:

Magnetic phenomena exist over a vast range of scales, ranging from the geomagnetic reversal that stretches over millennia to sub-picosecond electron dynamics. In this seminar we will introduce how dynamical processes of magnetic materials can be simulated on the atomic scale. We will further discuss key ingredients for these simulations including how to efficiently and accurately solve the governing Landau-Lifshitz equation and how to realistically model stochastic temperature fluctuations. To showcase the complex phenomena that can be described with these methods, we will also present recent results on novel spin-glass systems.

27/9/2023 Josefin Ahlkrona, https://www.su.se/profiles/joah8451-1.417812

Title: Numerical methods for ice sheet models

Abstract:

Better computer models of ice sheets in contact with the warming ocean are needed in order to reduce the uncertainty in estimates of future sea level rise. We focus on finite element modelling of ice and its interaction with the adjacent ocean, using mathematical models of high accuracy. Such accurate models are needed to describe the feedbacks between ice the warming ocean, but are too expensive to use for large-scale simulations of Greenland and Antarctica over the relevant time spans of hundreds to thousands of years. This talk gives an overview of our work on improving the efficiency of models. We present work on stable time stepping of the moving ice/atmosphere interface, coupling with the ocean, and preconditioning of the arising (non)-linear systems.