Uncertainty Quantification techniques are by now mature enough to address realistic, large scale problems of significant relevance. In this minisymposium, we focus in particular on complex fluid dynamics problems for engineering and environmental applications, that are fields in which computational science has traditionally played a major role. Several different kinds of UQ analyses naturally arise in these fields: forward UQ, optimization under uncertainty, inverse problem and data assimilation (e.g. for real-time control). In these scenarios, non-standard randomness might occur, and the complexity of the governing PDE equations further introduces significant and fascinating theoretical and computational challenges. In particular, polynomial-based UQ methods like Polynomial Chaos or Sparse grids collocation might not work well, in which case one has to resort to sampling methods, control variates, and more recently, machine learning techniques. Ad-hoc algorithms for high-performance computing are also relevant in this framework and welcome in this minisymposium.
08:30
Uncertainty quantification in numerical modeling of volcanic ash dispersal in the atmosphere
Mattia de' Michieli Vitturi | Istituto Nazionale di Geofisica e Vulcanologia (INGV) | Italy
Show details
Authors:
Mattia de' Michieli Vitturi | Istituto Nazionale di Geofisica e Vulcanologia (INGV) | Italy
Federica Pardini | Istituto Nazionale di Geofisica e Vulcanologia (INGV) | Italy
Samantha Louise Engwell | British Geological Survey | United Kingdom
Sara Barsotti | Vedurstofa Íslands, Icelandic Met Office | Iceland
Augusto Neri | Istituto Nazionale di Geofisica e Vulcanologia (INGV) | Italy
The behavior of ash plumes associated with volcanic eruptions is complex and strongly dependent on uncertain eruptive source parameters. It is also well known that the atmospheric environment interacts with volcanic plumes produced by explosive eruptions in a number of ways. Assessing the way in which these elements influence plume height and subsequent ash dispersal patterns is of fundamental importance for volcanic hazard mitigation. Indeed, through numerical models of ash plumes, Volcanic Ash Advisory Centres produce ash concentration charts and determine hazardous area for flight safety. Here, we present the results of uncertainty quantification and sensitivity
analysis, aimed at identifying and quantifying the source parameters exerting a major role on ash concentration and size distribution in the atmosphere. The analysis is applied to the combined model obtained coupling a column model describing the rise of a volcanic gas-particles mixture with a dispersal model simulating the transport and dispersion of particles into the atmosphere. The study is focused on the effects that uncertain eruptive source parameters have on ash dispersion in the atmosphere and deposition on the ground. The uncertainty quantification analysis allows to
construct statistical particle size distributions, both in the air and on the ground, at various distances from the volcanic vent, and also to quantify the uncertainty in the satellite observations of ash concentration in the atmosphere.
09:00
- NEW - Adaptive Multi-fidelity Surrogates for Uncertainty Quantification of Noisy CFD Data
Riccardo Pellegrini | Istituto di Ingegneria del Mare (CNR-INM) | Italy
Show details
Authors:
Riccardo Pellegrini | Istituto di Ingegneria del Mare (CNR-INM) | Italy
Andrea Serani | Istituto di Ingegneria del Mare (CNR-INM) | Italy
Simone Ficini | Istituto di Ingegneria del Mare (CNR-INM) | Italy
Jeroen Wackers | Ecole Centrale de Nantes (LHEEA Lab) | France
Matteo Diez | Istituto di Ingegneria del Mare (CNR-INM) | Italy
The assessment of the performance of aerial, ground or marine vehicles, requires high-fidelity computational fluid-dynamics (CFD) solvers, especially for innovative designs and extreme operational conditions. These solvers are often computationally expensive, making the design/operational-space exploration a challenge. Metamodeling can be used to reduce the computational cost by reducing the number of simulations required for the exploration. Multi-fidelity metamodels (MFM) can further reduce the computational cost by using mainly low-fidelity evaluations, while maintaining a high prediction accuracy using few high-fidelity evaluations. Finally, adaptive sampling methods increase the efficiency of the MFM training by adding samples only where it is most useful. Unfortunately, numerical noise is often present in CFD outputs that negatively affects the adaptive sampling process, deteriorating the MFM quality.
Four approaches for MFM, namely Stochastic Radial-Basis Functions (MF-SRBF) and three Gaussian Process (MF-GP) formulations, with training sets affected by noise are investigated for forward Uncertainty Quantification (UQ) problems. MFM approaches have auto-tuning capabilities to address the noise in the training sets. MF-SRBF use additive correction of a low-fidelity metamodel with least-square approximation, with leave one out cross validation and k-means clusterin for auto-tuning. MF-GPs use additive correction, additive and multiplicative, and training sets fusion.
09:30
On Numerical and Statistical Uncertainties in Scale-Resolving Simulations of Wall Turbulence
Saleh Rezaeiravesh | Royal Institute of Technology (KTH) | Sweden
Show details
Authors:
Saleh Rezaeiravesh | Royal Institute of Technology (KTH) | Sweden
Ricardo Vinuesa | Royal Institute of Technology (KTH), Swedish e-Science Research Centre | Sweden
Philipp Schlatter | Royal Institute of Technology (KTH), Swedish e-Science Research Centre | Sweden
The present study focuses on investigating how the accuracy and certainty of the quantities of interest (QoIs) of canonical wall-bounded turbulent flows are sensitive to various numerical parameters and time averaging. To perform scale-resolving simulations, we use Nek5000, an open-source high-order spectral-element code. The numerical parameters are taken to be the grid spacing in different directions (element size and number of spectral points per element), as well as the filtering parameters. For uncertainty propagation, different techniques including polynomial chaos expansion (PCE) and Gaussian Process (GP) regression (Kriging) are employed. As a complement to these uncertainty quantification (UQ) forward problems, global sensitivity analyses (with Sobol indices) are performed. To estimate uncertainties due to finite time-averaging, the use of classical, batch-based and autoregressive models is discussed. The variation in accuracy of the QoIs as a result of changing the numerical parameters is demonstrated and discussed. Motivated by this, a set of best-practice guidelines are derived to obtain simulations with minimum deviation from reference Direct Numerical Simulation. Moreover, the possibility of using Bayesian optimization techniques to find optimal combination of numerical parameters for obtaining QoIs with a given target accuracy is studied. Finally, comparisons of the certainty and accuracy aspects between Nek5000 and a low-order code (OpenFOAM) are presented.
10:00
Impact of uncertainties in inlet conditions in numerical simulations of the blood flow in the thoracic aorta
Maria Vittoria Salvetti | University of Pisa | Italy
Show details
Authors:
Maria Vittoria Salvetti | University of Pisa | Italy
Alessandro Mariotti | University of Pisa | Italy
Maria Nicole Antonuccio | BioCardioLab - Heart Hospital, Fondazione Toscana G. Monasterio | Italy
Simona Celi | BioCardioLab - Heart Hospital, Fondazione Toscana G. Monasterio | Italy
Computational Fluid Dynamics (CFD) is a useful complementary tool to in-vivo measurements to investigate hemodynamics on a patient-specific basis. Indeed, variables difficult to be obtained from measurements, as e.g. wall shear stresses, which play an important role in many cardiovascular diseases, can be quantified through CFD. The accuracy of CFD predictions however depends on modeling assumptions and computational set-up. For the simulation of blood flow in the thoracic aorta, a critical issue is given by the inflow boundary conditions, which are often imposed in terms of flow rate. The time behavior of the inlet flow rate can indeed significantly vary among different patients. A first stochastic analysis of the impact of uncertainties in the stroke volume and in the period of the cardiac cycle is presented. The PDFs of these quantities are evaluated fitting the clinical data of 23 patients. A continuous response surface of the output quantities of interest in the parameter space is recovered from a few deterministic simulations through the generalized Polynomial Chaos. A similar investigation is then carried out for the spatial distribution of the inlet velocity. From the analysis of in-vivo data, a parametric truncated-conical velocity distribution is considered. The uncertain parameters are the ratio of the base surfaces (equal to one for plug flow) and the position of the center of the top base, which is related to the eccentricity of the velocity distribution.