User day #15 2022

This year the user day is Friday 10 june 9h30-17h00 CEST at EDF Lab Saclay

The slides will be in english and we will speak english if necessary.

Here is the TEAMS link if you are unable to travel: https://teams.microsoft.com/l/meetup-join/19%3ameeting_NGVkZjU5OTQtYWM3Yy00YzlkLTllOGUtZTYzYzJlZmY2ZWI3%40thread.v2/0?context=%7b%22Tid%22%3a%22e242425b-70fc-44dc-9ddf-c21e304e6c80%22%2c%22Oid%22%3a%225efb0673-cf7d-413d-8080-288612be4a9a%22%7d

Here is the Discourse announce: https://openturns.discourse.group/t/journee-utilisateurs-openturns/201

Program:

8h30-9h30: Breakfast

• 9h30-10h: Introduction, L. Cambier (Scientific Director ONERA)
• 10h-11h:
  • New features in OpenTURNS, J. Schueller (PhiMECA), R. Lebrun (Airbus)
  • New features in Persalys, M. Baudin (EDF)
• 11h-12h: Point process-based approaches for robust reliability analysis of systems modeled by expensive simulators, G. Perrin (G. Eiffel University)

12h-13h30: Lunch

• 13h30 - 15h45: Studies OpenTURNS inside (1/2)
  • Surrogate models and sensitivity analysis for simulation of electron guns of high-frequency & high-power amplification devices, F. Molenda (Thales)
  • Copula-based iterative approximate Bayesian calibration for urban building energy modeling and impact of time resolution, X. Faure and O. Pasichnyi (KTH Institute of Techology Department of Sustainable Development Environmental Science and Engineering), R. Lebrun (Airbus)
  • Sensitivity studies OD-1D/3D of thermo aeraulic models of industrial premises, assistance in the identification of influential phenomena and preparation of experimental protocols for the Zephyr test platform, P. Borel (EDF)
  • Correction of surface reflectance retrieval in the tree shadow by machine learning, S. Ollivier, S. Lefebvre (ONERA)
  • Variance based sensitivity analysis for functional inputs - Application to pilots' actions, C. Obando (Airbus), J. Schueller (PhiMECA)

15h45 - 16h15: Gourmet break

• 16h15 - 17h: Studies OpenTURNS inside (2/2)
  • 1D site effects including spatial variability of soil properties, V. Alves-Fernandes (EDF), J. Berger (EDF)
  • The plugin otagrum: learning non-parametric Copula Bayesian Networks, M. Lasserre

Presentations are available at https://github.com/openturns/presentation#openturns-presentations

Abstracts:

Point process-based approaches for robust reliability analysis of systems modeled by expensive simulators, G. Perrin (Université G. Eiffel)

This presentation will focus on the guarantee by simulation of the correct functioning of complex systems using expensive simulators. These guarantees are most often based on the fact of being able to ensure that the probability of occurrence of undesired events is lower than a risk that is considered acceptable. With this objective, this work will first propose a method to bound this probability with a specified confidence, while requiring only a reduced number of calls to the simulator. This method is based on the substitution of the costly code by a surrogate model, the use of order statistics, and the exploitation of a dedicated Poisson process. A sequential strategy will then be presented, allowing to refine this bound by adding, in an optimized way, new learning points for the surrogate model. Indicators will finally be introduced to evaluate the robustness of the proposed estimates to small changes of the input distributions.

Surrogate models and sensitivity analysis for simulation of electron guns of high-frequency & high-power amplification devices, F. Molenda (Thales)

Thales AVS'MIS Business Line (Microwaves and Imaging Subsystems) designs and builds electron amplification devices (for high power and high frequency purposes). Commonly called hyper frequency amplifier tubes. Examples are TWT (Travelling Waves Tubes) and klystrons. In these devices an electron beam interacts with a wave and their energy is transferred to wave so this one is amplified. These devices are formed by 3 main subparts : 1) the electron gun (which produces the electron beam) 2) the interaction line (in which input wave and electron gun interact and wave exits amplified) 3) the collector (which is a garbage for electrons). Within Thales AVS-MIS we have developed our own FEM softwares in order to simulate involved physics by these phenomena (for stationary cases). In current study we focus our talk on electron gun simulations. From cathode's surface electrons are emitted and propagated through the gun composed by a wehnelt, one or two anodes and along the tube line. This physics is managed by a coupled system : Poisson equation for electric field (locally disturbed by electron beam charges) and Vlasov equations for electron macro-particles motion (under Lorentz force). Simulations are generally always done in best or nominal configurations. Newly we are wondering how simulation results are sensitive to manufacturing tolerances and bias (assembly defaults), which are the most and least influential inputs parameters (geometry characteristics, but also electric or magnetic characteristics). Because during manufacturing sometimes lot of sensitivity is observed (but not really measured). We are wondering also if such tools (as OpenTurns, Persalys and so on) could be integrated in our design simulations line. For this purpose we used OpenTurns and Persalys to generate designs of experiments, build surrogate models (kriging and polynomial chaos) and compute Sobol global sensitivity indices. These were applied for a specific electron gun simulations. Some our results will be displayed.

Copula-based iterative approximate Bayesian calibration for urban building energy modeling and impact of time resolution, X. Faure and O. Pasichnyi (KTH Institute of Techology Department of Sustainable Development Environmental Science and Engineering), R. Lebrun (Airbus)

Urban Building Energy modelling (UBEM) has emerged the last decade as an important tool to change the pace of energy transition in the building sector. To face one of its major usages, forecasting energy savings from potential energy conservation measures at an urban scale, several issues are still to address. Two of the most important are 1) the calibration of the model have to deal with a large amount of unknown parameters and a large set of buildings (being either building per building or through the archetype paradigm); 2) the scarcity of measured data at high time resolution question the ability of models being calibrated on yearly values to forecast energy savings. These two issues are address in this study and illustrated with a use case of 35 buildings in a district in Stockholm, Sweden. A new iterative Approximate Bayesian Calibration (ABC) is proposed using a copula-based sampling process a teach iteration for issue 1). The new calibration process is applied with three different time resolutions to address the influence of data resolution on the forecasted energy of a classical energy conservation measure for issue 2). Over the 35 buildings, the new method�s efficiency is demonstrated. It enables to populate very quickly a final joint distribution for the nine unknown parameters considered in each building. The margins can be strongly influenced by the time resolution, but the forecasted energy is identical for the three time resolution based cases. A noticeable difference is still shown when the ECM concerns a formerly unknown or known parameter.

Sensitivity studies OD-1D/3D of thermo aeraulic models of industrial premises, assistance in the identification of influential phenomena and preparation of experimental protocols for the Zephyr test platform, P. Borel (EDF R&D Prisme)

The availability of control equipment / electrical cabinets in industrial premises depends, among other things, on the good evacuation of the heat generated by these components. This requires temperature conditions that are low enough to allow the proper functioning of the equipment. The present study aims to characterize, using 0D-1D (Modelica TAeZoSysPro library) and 3D (Code Saturne) numerical approaches, the physical phenomena influencing the distribution of the temperature field in a typical industrial room, the control room of the Zephyr test platform. This characterization is carried out by the evaluation and analysis of experimental designs of identified parameters.

Correction of surface reflectance retrieval in the tree shadow by machine learning, S. Ollivier, S. Lefebvre (ONERA)

In urban environment, many remote sensing applications such as classification of impermeable surfaces and vegetation and their physical characterization require very good accuracy in surface reflectance retrieval, both for illuminated and shadowed areas. 3D atmospheric correction models can handle opaque shadows but fail for semi-transparent shadows such as tree ones, because they do not take into account the contribution of the transmitted irradiance through the tree crown and the structural traits of the tree. From simulations performed with a 3D radiative transfer code, DART, on different tree structural parameters and environmental conditions, we suggest to build a Gaussian process metamodel to predict a correction factor that accounts for both spectral and spatial variability of the surface reflectance. For this purpose, we have performed a sensitivity analysis to identify the most significant input variables.

Variance-based sensitivity analysis for functional inputs - application to pilots' actions, C. Obando (Airbus), J.Schueller (Phimeca)

Pilots' behavior and performance are considered key elements in the design and operation of cockpits. Aircraft sensors are a rich and readily available source of data that can be used to characterize pilots' actions. These multivariate temporal signals are highly dimensional and often it is difficult to disentangle each input parameter's effect on the actions. In this talk, we present a methodology that exploits aircraft multivariate temporal signals to detect anomalies in pilots' actions and incorporates a variance-based sensitivity analysis to quantify the contribution of each functional input.

The plugin otagrum: learning non-parametric Copula Bayesian Networks, M. Lasserre

In the context of learning Bayesian networks with continuous data, the solution that is most often used is to learn a discrete model from the discretized data. As such, the obtained model does not allow to sample new continuous values to be able, for example, to make approximate inferences via Monte-Carlo Markov Chains. To do so, continuous parametric models can be used but at the cost of the model expressivity. On the other side, continuous non-parametric models are difficult to learn for high-dimensional problems and can lead to computationally expensive and time-consuming calculations. Copula Bayesian Networks (CBNs) leverage both Bayesian networks (BNs) and copula theory to compactly represent continuous distributions as a set of local low dimensional copula functions, allowing to use non-parametric models such as the empirical Bernstein Copula (EBC). After a short introduction to copula theory and CBNs, we will present the OTaGrUM plugin that allows to learn CBN using the libraries OpenTURNS and aGrUM.