Scientific Computing Seminar

Date and Place: Thursdays in Room 32-349. For detailed dates see below!

Content

In the Scientific Computing Seminar we host talks of guests and members of the SciComp team as well as students of mathematics, computer science and engineering. Everbody interested in the topics is welcome.

List of Talks

  • Tue
    03
    Nov
    2015

    10:30SC Seminar Room 32-349

    Prof. Andrea Walther, Institut für Mathematik, Universität Paderborn

    Title:
    On the optimization of piecewise smooth functions

    Abstract:

    This talk presents the abs-normal form for piecewise linear functions, which can be used for the optimization of piecewise linear function. Such an optimization serves as inner loop for the solution of general piecewise smooth optimization problems using successive piecewise linearization (SPL). For the convex as well as the non-convex case the latest results with respect to the convergence analysis of this new SPL approach will be presented. Furthermore, first numerical results will be shown.

  • Thu
    05
    Nov
    2015

    11:30SC Seminar Room 32-349

    Simon Gottschalk, Fraunhofer ITWM

    Title:
    The One-shot method for optimization problems with underlying PDEs

    Abstract:

    Optimization problems with underlying PDEs appear in disciplines like geophysics, medical imaging, atmospheric science and aerodynamic shape design. Conventional approaches -known from the nonlinear optimization theory- are in general extremly costly. This talk introduces the One-shot method in order to solve such optimization problems with less expense. The main idea of this approach is presented in the case of a steady PDE as a constraint. It is explained how the Karush-Kuhn-Tucker conditions can be used to find an iterative method which is able to solve this problem. In the second part optimization problems with underlying unsteady PDEs are considered. It is shown how one can transform this problem into a new form such that the One-shot method is applicable.

  • Thu
    19
    Nov
    2015

    11:30SC Seminar Room 32-349

    Marc Schwalbach, von Karman Institute (VKI)

    Title:
    Cold-to-hot transformation for adjoint optimization of turbomachinery components

    Abstract:

    Due to the centrifugal forces experienced by turbomachinery components in operation, structural deformations occur such that the running “hot” structure differs from the resting “cold” structure. Typically, an optimization is performed in the “hot” state and the resulting CAD model has to be deformed to the “cold” state as a post-processing step for manufacturing. In this project, a cold-to-hot transformation is to be included in the optimization loop to compute the sensitivities of the objective function with respect to the “cold” CAD parameters directly using adjoints. This talk will give a summary of the overarching project, which involves the CAD-based adjoint optimization of turbomachinery components under mechanical constraints, and discuss the cold-to-hot transformation.

  • Thu
    26
    Nov
    2015

    11:30SC Seminar Room 32-349

    Junis Abdel Hay, SciComp

    Title:
    Modelling Real Acoustic Absorbers for Computational Aeroacoustics

    Abstract:

    Acoustic liners are widely used as acoustic absorbers in the inlet and bypass duct of aero-engines. Commonly used acoustic liners are Helmholtz resonator arrays and consist of a rectangular or honeycomb cell structure which is covered by a perforated facesheet. As far as aero-engines are concerned drainage slots are cut into the resonator walls. Therefore, these resonators are exchanging fluid and are considered to behave non-locally reacting. The behaviour of locally reacting liners can be described by the acoustic impedance. Currently, there is no method to directly measure the impedance under flow conditions but it can be calculated indirectly by an impedance eduction. Therefore, a liner model is used to include lined surfaces in Computational AeroAcoustic (CAA) simulations. The model’s parameters must be optimized to reproduce the measured liner characteristics. The optimized parameters can be used to predict the liner behaviour for complex geometries and with flow. For some novel liner concepts or non-locally reacting liners, the behaviour is too complex to be accurately reproduced by models based on simple physical assumptions. This issue is investigated based on impedance eductions for locally and non-locally reacting liners using different liner models and optimization methods. The effects of the non-locally reacting behavior will be discussed and the applicability of these different methods is reviewed. An analytical model for two coupled non-locally reacting Helmholtz resonators is presented and the structure of the model parameters will be discussed. The effects generated by the propagation of sound through the cavities are identified. The model is validated based on CAA simulations of this configuration with normal sound incidence. Furthermore, the results are compared to the experimental data and the model is extended empirically.

  • Mon
    07
    Dec
    2015

    11:30SC Seminar Room 32-349

    Dr. Torsten Bosse, Argonne National Laboratory

    Title:
    An asynchronous Oneshot method with load-balancing

    Abstract:

    We propose a ‘blurred’ one-shot method for the minimization of design optimization problems using a parallel and asynchronous evaluation of the three basic tasks: the state-, the adjoint- and the control-update. In particular, for each task it is allowed to always use and/or override the latest information of another task, i.e., rather than waiting until the fixed-point iteration provides a new state update it is assumed that parts of the corresponding adjoint iteration already use the latest information from the simulation code. Naturally, this cross-communication between the three tasks will lead to inconsistencies and any mathematical convergence theory for such an approach is far from being obvious.

    Nevertheless, one can expect convergence of the method in certain cases. The key for the success of such a method relies on an optimal distribution of the different tasks for a given amount of available resources on a igh performance cluster. This assignment problem yields a possibility to influence the contraction rates of the primal and dual updates as well as the change in the control variables. Also, it can be be shown that the blurred Oneshot algorithm is a generalization of the previously presented Jacobi- and (Multistep-) Seidel-Oneshot method, which can be recovered by a suitable allocation of resources. The blurred method can be applied on (discretized) optimal control problems, which might also include unsteady PDEs.

  • Thu
    10
    Dec
    2015

    11:30SC Seminar Room 32-349

    Prof. Domenico Quagliarella, Italian Aerospace Research Center (CIRA)

    Title:
    Risk measures and optimization under uncertainty

    Abstract:

    Many industrial optimization processes must take account of the stochastic nature of the system and processes to be designed or re-designed and have to consider the random variability of some of the parameters that describe them. Thus it is necessary to characterize the system that is being studied from various points of view related to the treatment of uncertainty. This talk is related to the use of various risk measures in the context of robust and reliability based optimization. We start from the definition of risk measure and its formal setting and then we show how different risk functiontional definitions can lead to different approaches to the problem of optimization under uncertainty. In particular, the application of value-at-risk (VaR) and conditional value-at-risk (CVaR), also called quantiles and superquantiles, is here illustrated. These risk measures originated in the area of financial engineering, but they are very well and naturally suited to reliability-based design optimization problems and they represent a possible alternative to more traditional robust design approaches. We will then discuss the implementation of an efficient risk-measure based optimization algorithm based on the introduction of the Weighted Empirical Cumulative Distribution Function (WECDF) and on the use of methods for changing the probability measure. Finally we will discuss the problems related to the error in the estimation of the risk function and we will illustrate the “bootstrap” computational statistics technique to get an estimate of the standard error on VaR and CVaR. Finally, we will report some simple application examples of this approach to robust and reliability based optimization.

  • Thu
    17
    Dec
    2015

    11:30SC Seminar Room 32-349

    Thomas Dick, SciComp

    Title:

    Development of a Discrete Adjoint method for unsteady PDE optimization

    Abstract:

    In the design and optimization of aerospace systems, there are many problems which are unsteady by nature. For example active flow control, aeroacoustic or turbo machinery. For these problems the assumption, that the underlying physical problem is steady, is not valid. In this talk we will show how one can adapt an adjoint framework for unsteady problems. Therefore we will generalize the first discretize then optimize approach for this case. This will allow us to construct a Lagrangian and to derive the KKT condition, which will ultimately give us a discrete adjoint equation.

  • Thu
    28
    Jan
    2016

    11:30SC Seminar Room 32-349

    Prof. Claus-Dieter Munz, IAG, Uni Stuttgart

    Title:
    High Order Discontinuous Galerkin Schemes for Industrial Flow Problems

    Abstract:

    High-order methods are attractive for simulations of multiscale problems. Due to their low dispersion and dissipation errors, they minimize the number of points, required to resolve a wavelength or another structure with given accuracy. In this talk a survey is given about the construction of the class of spectral element discontinuous Galerkin schemes. Several building aspects in this construction are considered in more detail including shock capturing and de-aliasing. Their high computational efficiency, especially on high performance computers is motivated and shown. As applications the fluid flow around a rear mirror of a car is shown and the noise generation by an acoustic feedback identified. Other large scale simulations include the direct numerical simulation of supersonic boundary layers and an example of plasma flow.

  • Mon
    08
    Feb
    2016

    15:00SC Seminar Room 32-349

    Dr. Roland Ewert, DLR Institut für Aerodynamik und Strömungstechnik

    Title:
    Strömungsgeräuschsimulation mittels stochastischer Quellen

    Abstract:

    Der Vortrag präsentiert hybride aeroakustische Simulationsmethoden und ihre Anwendung auf Umströmungsgeräuschprobleme der Luftfahrt. In einem ersten Schritt wird das turbulente Umströmungsproblem mit Reynolds gemittelten Navier-Stokes (RANS) Gleichungen simuliert. Die resultierende Schallgenerierung und Schallpropagation wird in einem zweiten Schritt mit Hilfe von Störungsgleichungen und Methoden der Numerischen Aeroakustik (CAA) gelöst. Die Simulationen verwenden instationäre Geräuschquellen, die stochastisch auf Basis der RANS Turbulenzstatistik realisiert werden. Durch die hohe Simulationseffizienz im Vergleich zu skalenauflösenden Verfahren ist die Analyse einer größeren Anzahl von Konfigurationen möglich. Die Methode bietet damit eine geeignete Grundlage für einen optimierten aeroakustischen Entwurf.

    Zur effizienten stochastischen Generierung synthetischer Turbulenz für Schallquellen wird die Fast Random Particle-Mesh (FRPM) Methode eingesetzt. Ziel der stochastischen Rekonstruktion ist eine möglichst genaue lokale Reproduktion der verschiedenen statistischen Zielgrößen aus der RANS. Der derzeitige Status der Rekonstruktionsqualität der teilweise gegenseitig abhängigen Zielgrößen wird präsentiert und zukünftige Verbesserungsmöglichkeiten diskutiert.

    Verschiedene aeroakustische Propagationsverfahren werden diskutiert. Lineare Störungsgleichungen (APE) können die Refraktion des Schallfeldes im nichtuniformen Strömungsfeld beschreiben. Eine Diskretisierung mit Finite-Differenzen (DRP Verfahren) auf strukturierten Multiblock Gittern wird einer Diskretisierung mittels der Diskontinuierlichen Galerkin (DG) Methode auf unstrukturierten Gittern gegenüber gestellt. Die Vorhersagequalität wird für generischen Hinterkantenlärm, die Schallabstrahlung von Hochauftriebskonfigurationen und Strahllärm für lärmoptimierte Düsenrandmodifikationen demonstriert.

    Des Weiteren wird als Propagationsverfahren ein schnelles Multipole Boundary Element Verfahren für die konvektive Wellengleichung auf der Basis von unstrukturierten Oberflächennetzen vorgestellt, dass die Simulation größerer umströmter Komponenten (z.B. Halbflügel eines Verkehrsflugzeuges) bei vereinfachter Propagationsphysik ermöglicht. Abschließend werden Ergebnisse für skalenauflösende Simulationen mit Navier-Stokes Gleichungen in Störungsform (NLDE) gezeigt, wobei eine stochastische Quelle als ein aktives Backscattering-Modell eingesetzt wird.