-
Notifications
You must be signed in to change notification settings - Fork 93
User day #14 2021
Expect most presentations to be spoken in french, but at least the slides will be in english:
Presentations will be made available at https://github.com/openturns/presentation#openturns-presentations
This year the user day is split into 2 online sessions:
-
session 1: Friday 11 june 9h30-12h30 CEST
-
session 2: Monday 21 june 9h30-12h30 CEST
Program session 1:
- 9h30-10h30: Sensitivity analysis base on HSIC dependence measures, A. Marrel (CEA)
- 10h30-11h: Learning wall models for turbulent flow computations, X. Garnaud, R. Lebrun (Airbus)
- 11H-11H30: otbenchmark: an open source Python package for benchmarking and validating uncertainty quantification algorithms, E. Fekhari (EDF)
- 11h30-12h: Probabilistic models for penstock integrity assessment, Ph. Bryla (EDF), A. Dumas (PhiMECA), E. Ardillon (EDF), A. Dutfoy (EDF)
Program session 2:
- 9h30-10h: New OpenTURNS features, K. Delamotte (IMACS), J. Schueller (PhiMECA)
- 10h-10h30: Surrogate model of water temperature, M. Baudin, Fabien Souillé (EDF)
- 10h30-11h: Adaptive kriging based system reliability analysis applied to space variant problems, Ch. Amrane (Univ. Clermont Auvergne)
- 11h-11h30: Sensitivity analysis and uncertainty in CFD simulations of multiphase flow, C. Henry, A. Dupré, M. Bossy. Université Côte d'Azur, INRIA, CaliSto laboratory, Sophia-Antipolis, France (INRIA)
- 11h30-12h: Interfacing a Modelica modeling tool and Persalys with OpenTURNS for the analysis of a solar collector, C-E. Gerrer (Phimeca)
- 12h-12h30: Decoding Neural Signatures in an Emergency Driving Situation using OpenTURNS, C. Obando (Airbus)
Sensitivity analysis base on HSIC dependence measures, A. Marrel (CEA) In the framework of propagation of uncertainties in numerical simulation, global sensitivity analysis aims at studying the impact of the input uncertainties on the output of the model. For this, dependence measures based on reproducing kernel Hilbert spaces (namely Hilbert-Schmidt Independence Criterion denoted HSIC), are very efficient statistical tools. HSIC can be used via sensitivity indices for ranking purpose or via independence tests for screening and ranking purposes. This presentation will introduce the theoretical and statistical concepts around HSIC. Their practical use and interest in industrial applications will be described and illustrated. Some recent advanced R&D work will also be presented, to further extend their use and increase their robustness. Remaining challenges will be highlighted. Finally, a progress report on the implementation of HSIC in OT platform will be proposed.
Learning wall models for turbulent flow computations, X. Garnaud,R. Lebrun (Airbus) Turbulence models used in CFD simulation require meshes that are fine enough to capture the viscous sublayers at the walls, with typical thickness of the order to 1 to 10 microns. This imposes significant requirements in terms of mesh size and maximum allowable (pseudo) time step, and complicates the mesh generation process. Wall models may be used to represent the flow in the near wall region and increase the cell sizes at the wall by a factor 10 or more. Such models are typically based on experimental measurements for flat-plate boundary layers, and do not provide the required level of accuracy for aerodynamic applications. We will present metamodels built from wall-resolved simulations to improve the accuracy of wall-modeled RANS computations for industrial configurations.
otbenchmark: an open source Python package for benchmarking and validating uncertainty quantification algorithms, E. Fekhari (EDF) This presentation introduces a new Python package, called “otbenchmark”, offering tools to evaluate the performance of a large panel of uncertainty quantification algorithms. It provides benchmark classes containing problems with their reference values. Two categories of benchmark classes are currently available: reliability estimation problems (i.e.,estimating failure probabilities) and sensitivity analysis problems (i.e.,estimating sensitivity indices such as the Sobol’ indices). This package can either be used for validating a new algorithm or automatically comparing various algorithms on a set of problems. Additionally, the package provides several convergence and accuracy metrics to compare the performance of each algorithm. To face high-dimensional problems, otbenchmark offers graphical tools to draw multi-dimensional events, functions and distributions based on cross-cuts visualizations. Finally, to ensure otbenchmark’s accuracy, a test-driven software development method has been adopted (using, among others, Git for collaborative development, unit tests and continuous integration). Ultimately, otbenchmark is an industrial platform gathering problems with reference values of their solutions and various tools to achieve a robust comparison of uncertainty management algorithms.
Probabilistic models for penstock integrity assessment, Ph. Bryla (EDF), A. Dumas (PhiMECA), E. Ardillon (EDF), A. Dutfoy (EDF) A probabilistic model for penstock integrity assessment has been implemented in the Persalys tool. This model takes into account uncertainties and scatter of input data concerning material characteristics and pipe wall thinning due to corrosion. Two different failure criteria have been implemented : the 1st one is a plastic collapse criterion, whereas the 2nd criterion also considers ruptures due to the presence of manufacturing planar flaws. Both criteria allow to calculate upper bounds of the annual failure probability of a penstock. For both plastic collapse and fracture mechanics, large calculation grids have been applied for covering most penstock diagnoses configurations. For plastic collapse, these calculations have shown that the evaluation of the Margin Factor with Q2,5% quantiles generally leads to a very conservative assessments. The study also allowed to estimate the increase of failure probability resulting from the presence of non-detected surface manufacturing flaws. These latter calculations led to measure the influence of two major factors on the penstock reliability: the relief of residual stresses in welds and of the detectability performance of NDT. In these configurations with residual flaws, the upper bound of annual failure probability remains lower than the allowable threshold in presence of welding manufacturing flaws under conditions that residual stress were sufficiently relieved and that efficient NDT were applied after manufacturing. However, the evaluation of conditional failure probabilities have shown that a successful hydrostatic pressure test can lead to a drastic reduction of the annual failure probability since the average cumulated corrosion thinning remains moderate.
Surrogate model of water temperature, M. Baudin, Fabien Souillé (EDF) The goal of this study is to assess the sensitivity of the Blayais nuclear power plant to cold temperatures of water in the Gironde estuary. It is possible to consider the system of partial differential equations associated with the model and, combined with initial and boundary conditions, predict the temperature depending on the time. The CPU cost of such an approach is, however, very high. In this talk, we present a method based on measured data which is used by a statistical model to predict the water temperature time series. The method is based on a surrogate model which parameters are estimated from the data. This surrogate model predicts the future water temperature time series depending on the past. The surrogate model uses a Karhunen-Loève decomposition to reduce the dimension of the input and output time series and uses a polynomial chaos in order to predict the coefficients in the reduced space.
Adaptive kriging based system reliability analysis applied to space variant problems, Ch. Amrane (Univ. Clermont Auvergne) Reliability analysis aims at propagating input uncertainties within a system in order to evaluate its failure probability. Such uncertainties may result from the inhomogeneity of its material properties, dimensions, boundary conditions, etc. The input parameters are in this case random and spatially inhomogeneous. In practice, the failure location is supposed to be unique in space and selected a priori, sometimes based on one particular realization of the involved random variables. If the determination of this unique location is carried out on an aberrant or unlikely realization of the random variables, the failure probability estimation may be totally erroneous. In fact, the uncertain spatial variability may generate multiple failure locations that are often neglected. So as to accurately estimate the failure probability, all the potential critical locations should be considered. Moreover, the spatial correlation between the different locations implies that the failure at any location can lead to the failure of the whole system. This behavior is similar to that of series systems. Therefore, system reliability methods seem to be appropriate for conducting reliability analysis of problems subject to spatial variability. Among the existing methods, the system reliability methods based on surrogate models are distinguished because of their effectiveness in the context of expensive performance functions. Hence, the Active learning and Kriging-based SYStem reliability (AK-SYS) framework (Fauriat and Gayton 2014) is here extended to such problems. The proposed approach named AK-SYSs combines the enrichment process of the AK-SYS method with an active search strategy for failure loca- tions. From a practical point of view, the method requires the calibration of Kriging surrogate models (one per one location) which is carried out with the OpenTURNS toolbox (Baudin, et al. 2015). Furthermore, the application of the method on a finite elements problem having dif- ferent configurations of the uncertain spatial variability requires a coupling of a finite elements software with python which is also presented in the proposed talk.
Sensitivity analysis and uncertainty in CFD simulations of multiphase flow, C. Henry (INRIA) We present recent developments that have been made within the framework of the VIMMP EU project (Virtual Materials Market Place). Our objective is to set up a methodology to analyse the sensitivityand then quantify uncertainty in numerical simulations of multiphase flows to a number of input variables. In this talk, we start by introducing the workflow used for the multiphase flow simulations. The physical case studied here consists in the point-source dispersion of particles by a turbulent pipe flow. Numerical simulations are performed by coupling a CFD simulation of the turbulent pipe flow (using standard turbulence models) to a particle-tracking simulation (using a stochastic Lagrangian model). The simulations are performed in Code_Saturne CFD [1]. The workflow is launched using tools from the Salome Platform, which allows to handle the coupling of the fluid phase simulation and the particle-phase simulation. The results obtained are then analysed using existing tools within OpenTurns [2]. For that purpose, a dataset is obtained by running the workflow with a range of input variables (e.g. the fluid velocity, number of particles injected, size of particles) and accounting for the intrinsic stochasticity of each run. Then, we use sensitivity analysis techniques (especially Sobol sensitivity index [3] through meta-model) to identify the key parameters affecting the observed results. By considering several observables, we also highlight the need to define a clear observable before running such analysis.
Interfacing a Modelica modeling tool and Persalys with OpenTURNS for the analysis of a solar collector, C-E. Gerrer (Phimeca) Numerical experiments make it possible to better understand the functioning of a physical model, as well as to reduce its computing time. We show the interfacing between a Modelica modeling tool and Persalys. The Modelica model of a solar collector, exported in FMU format, is analyzed and metamodelized in Persalys. The created metamodel is then exported as a Modelica (or FMU) model in order to be re-used in the Modelica modeling tool.
Decoding Neural Signatures in an Emergency Driving Situation using OpenTURNS, C. Obando (Airbus) The increasing level of automation brings new challenges in the understanding of human machine interaction that are pivotal to aircraft pilotage. The use of wearable devices to measure brain activity are a tool to quantify human factors in this context. However, neurophysiological data are often noisy; moreover underlying neural processes are hard to disentangle. In this talk, we present methods used to decode pilot’s electroencephalogram (EEG) signatures exhibited during an emergency situation in a virtual racing vehicle. One one hand, spatiotemporal models for EEG signatures corresponding to the different pilot’s reaction to an emergency. On the other hand, a neural network which detects an emergency situation from the neurophysiological data faster than from other sensors, such as muscle or motor reactions.