Scientific Computing Seminar

Date and Place: Thursdays and hybrid (live in 32-349/online via Zoom). 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. Everybody interested in the topics is welcome.

List of Talks

  • Thu
    18
    Apr
    2019

    11:30SC Seminar Room 32-349

    Patrick Mischke, TU Kaiserslautern

    Title:
    Implementation of a POD Based Surrogate Model for Aerodynamic Shape Optimization

    Abstract:

    Computational Fluid Dynamics (CFD) simulations are often used with aerodynamic shape optimization in mind. There exist several approaches to perform optimization, for example employing one shot methods or using adjoint computations by utilizing algorithmic differentiation techniques while running the flow solver. However, it may sometimes be preferred to inspect the flow fields across the whole design space, for instance if those strategies seem to operate too local for the problem on hand. This issue can be targeted by the use of surrogate models. A surrogate model predicts the flow field for a given design input in a computationally cheaper way, but also with less accuracy than the full fluid dynamics simulation. The surrogate model uses a set of training data to find reasonable model parameters for future design inputs. The goal of my bachelor thesis, that I present in this talk, was the implementation of an surrogate model for CFD using Proper Orthogonal Decomposition (POD) and Kriging regression. The challenges encountered are discussed, and the NACA 0012 airfoil at transonic flow conditions is presented as test case. The characteristic shock front and the high number of design parameters describing the airfoil geometry of its flow field lead to a rather expensive training process for the surrogate model. However, the discussed approach may still be valuable for other geometries or as foundation for surrogate models using other techniques.

  • Thu
    02
    May
    2019

    11:30SC Seminar Room 32-349

    Dr.-Ing. Bernhard Eisfeld, DLR Braunschweig, Institut für Aerodynamik und Strömungstechnik, C²A²S²E Center for Computer Applications in AeroSpace Science and Engineering

    Title:
    Reynolds-Stress Modelling – Concepts, Advances and Challenges

    Abstract:

    The Reynolds-Averaged Navier-Stokes (RANS) equations are still the backbone of numerical flow simulations in industrial applications. Hence, a turbulence model is required for closure, which decides about the accuracy of the predictions.
    Many models are based on the assumption of a flow dependent eddy viscosity added to the molecular viscosity of the fluid. While agreeing with the observation of enhanced momentum transfer due to turbulent fluctuations, this is a significant simplification of the physics of turbulent flow, limiting the predictive accuracy in complex flow situations.
    Improvement is expected by Reynolds-stress modelling based on the transport equation for the individual components of the Reynolds-stress tensor and for an additional length-scale providing variable. In this case, the modelling is restricted to the different terms of the Reynolds-stress transport equation and the length-scale equation that is usually taken over from corresponding eddy-viscosity models and considered the weakest link of the approach.
    The presentation will introduce the Reynolds-stress transport equation, explain the physical significance of its terms and outline the corresponding modelling approaches.
    Recent advances have been achieved by developing a length-scale correction. The underlying idea will be presented and its improvement on the prediction of separated flows will be demonstrated.
    Turbulence modelling is challenged by the variety of flow phenomena that need to be treated. This will be underlined by a theoretical analysis of self-similar free-shear flows, predicting a layer of constant Reynolds-stress anisotropy. Experimental data confirm its existence, revealing differences in the turbulence structure between different flows. Hence, a self-adaptive modelling strategy is required, applying tailored models to automatically identified regions of the flow field. An example will be given, demonstrating the potential of such tailored modelling.

  • Thu
    16
    May
    2019

    11:30SC Seminar Room 32-349

    Matthias Freimuth, TU Kaiserslautern/MTU Aero Engines

    Title:
    The multiphysics coupling tool preCICE in the context of adjoint-based aeroelastic designs

    Abstract:

    In the emerging field of coupled numerical simulations including two or more physical fields the capabilities of the multiphysics coupling tool preCICE are discussed with respect to aeroelastic design in this talk. With an adapter for the structural solver within the multiphysics solver SU2 a new link between SU2 and preCICE is presented. The development process and the key ingredients for multiphysics coupling are explained and the result is shown with a simple testcase. A smaller aspect also addressed in this talk is the discrete adjoint method for the gradient computation to efficiently optimize in a fluid structure interaction framework. The talk is based on the content of my masters thesis at MTU Aero Engines in Munich and will be given in english.

  • Thu
    06
    Jun
    2019

    14:30SC Seminar Room 32-349

    Prof. Andrea Walther, Institut für Mathematik, Universität Paderborn

    Title:
    Minimization by Successive Abs-Linearization: Recent Developments

    Abstract:

    For finite dimensional problems that are unconstrained and piecewise smooth the optimization based on successive abs-linearisation is well analysed yielding for example linear or even quadratic convergence under reasonable assumptions on the function to be optimised. In this talk we discuss the extension of this approach to the more general class of nonsmooth but still Lipschitz continuous functions covering also the Euclidean norm. For this purpose, we introduce the so-called clipped root linearisation and present first numerical results.
    Furthermore, we sketch the extansion of this approach to the infinite dimensional setting.

  • Fri
    14
    Jun
    2019

    11:30SC Seminar Room 32-349

    Manfred Schneider, retired senior confirmed advisor Flight-Physics, Airbus Defense & Space Deutschland GmbH

    Title:
    Noise Simulation at FTEG high-lift airfoil using hybrid RANS/LES Model

    Abstract:

    This study focuses on the development, validation and application of the interdisciplinary computational fluid dynamics/computational aeroacoustics (CFD/CAA) method with the name Flight-Physics Simulator AEOLus (FPS-AEOLus). FPS-AEOLus is based on enhanced conservative, anisotropic, hybrid Reynolds-averaged Navier-Stokes/ Large-Eddy Simulation (RANS/LES) techniques to solve an aerodynamic flow field by applying the unsteady, compressible, hyperbolic Navier–Stokes equations of second order. The two-layer SSG/LRR-ω differential Reynolds stress turbulence model presented, combining the Launder-Reece-Rodi (LRR) model near walls with the Speziale-Sarkar-Gatski (SSG) model further apart by applying Menter’s blending function F1. Herein, Menter’s baseline ω-equation is exploited for supplying the length scale. Another emphasis is put on the anisotropic description of dissipation at close distance to the solid wall or in wake area for describing the friction-induced surface-roughness behaviour in viscous fluid physics and swirling wake effects. For that purpose, the SSG/LRR-ω seven-equations Reynolds stress turbulence model with anisotropic extension was realized, therefor the theory is described in general. Beyond that, a modified delayed detached-eddy simulation (MDDES) and a scale adaptive simulation (SAS) correction to capture the stochastic character of a large-eddy-type unsteady flow with massive flow separations in the broad band is implemented. To demonstrate the time-dependent noise propagation having wave interference a linearized Euler equation (LEE) model using a combined Momentum- and Lamb-vector source have been applied into the CFD/CAA – method.

    The DLR 15 wing, a High-Lift device in landing configuration having a deployed slat and landing flap is studied experimentally and numerically. The first part of the application deals with the steady flow investigation; however, the same turbulence model is used for the unsteady flow case without the enclosed time derivatives. The second part concentrates on unsteady modelling for the Navier–Stokes and Linearized Euler field. With this new combined CFD/CAA – method, steady and unsteady numerical studies for the high-lift wing configuration for discovering the aerodynamic and –acoustic propagation effects are shown, discussed and when experimental data were available validated. The High-Lift wing has a constant sweep angle of Λ=30° to investigate possible cross-flow; to realize this, periodic boundary conditions were set in spanwise direction.

  • Thu
    04
    Jul
    2019

    11:30SC Seminar Room 32-349

    Prof. Xun Huan, Mechanical Engineering, University of Michigan

    Title:
    Optimal Experimental Design and Bayesian Neural Networks for Physics-Based Models

    Abstract:

    Models and data are two critical components of scientific research: models provide predictions of what data we might observe, and data in turn help refine and advance our models. In this talk, we focus on two important interactions between models and data for complex physical systems: (1) optimal experimental design (OED) for finding the most useful data, and (2) Bayesian neural network (BNN) data-driven models for accelerating expensive predictions with uncertainty quantification.

    First, we present the OED framework that systematically quantifies and maximizes the value of experiments. Indeed, some experiments produce more useful data than others, and well-chosen experiments can provide substantial resource savings. We describe a general mathematical framework that accommodates nonlinear and computationally intensive (e.g., ODE- and PDE-based) models. The formalism employs Bayesian statistics and an information-theoretic objective, and we develop tractable numerical methods with demonstrations on designing combustion kinetic experiments and sensor placement for contaminant source inversion.

    Next, we introduce BNNs as data-driven surrogate models capable of rapid predictions while quantifying uncertainty, which are highly useful for real-time decision-making scenarios. We focus on their use for in-flight detection of rotorcraft blade icing using acoustic signals. With a database of computational fluid dynamic and computational aeroacoustic simulations produced from the open-source SU2 software, a BNN is constructed to directly map the acoustic signals to aerodynamic performance metrics, thus bypassing the expensive inverse problems. The prediction uncertainty is quantified by treating BNN weight parameters as random variables, thus revealing a distribution of predictions instead of a single-value output. This uncertainty distribution reflects the quality and confidence of the BNN prediction, information that is critical for pilot decision-making under potentially dangerous icing flight conditions.

  • Tue
    09
    Jul
    2019

    11:00SC Seminar Room 32-349

    Steffen Schotthöfer, TU Kaiserslautern

    Title:
    Sensitivity analysis in the presence of limit cycle oscillations – Regularizing methods

    Abstract:

    Many unsteady problems equilibrate to periodic behavior. For these problems the sensitivity of periodic outputs to system parameters are often desired and must be estimated from a finite time span. Sensitivities computed in the time domain over a finite time span can take excessive time to converge or fail to do so. In this presentation two approaches will be discussed to overcome these difficulties.
    First, the so called windowing approach uses weighting functions to improve convergence behavior. We will consider long-time and short-time windowing as two aspects of this approach. The idea of a long time window is to average over a big, non-integer number of periods of the weighted output, to archive convergence. On the other hand, the idea of short time windowing is to average over a small, integer multiple of a period. Convergence is archived by refining the period approximation.
    Second, an analytic approach is discussed. Here we set up an additional boundary value problem to exactly compute the influence of parameter changes on amplitude, period and relative phase.
    We will discuss the efficiency of both approaches and their usability in SU2 and aerodynamic applications.

  • Thu
    25
    Jul
    2019

    11:30SC Seminar Room 32-349

    Marc Schwalbach, von Karman Institute for Fluid Dynamics/TU Kaiserslautern

    Title:
    CAD-Based Adjoint Multidisciplinary Optimization Framework for Turbomachinery Design

    Abstract:

    Most adjoint-based optimization frameworks for turbomachinery only consider aerodynamic performance and constraints, leading to designs that need to pass through revisions by structural requirements. Only in recent years, adjoint optimization frameworks have been extended to include structural constraints.

    In this work, a CAD-based parametrization is used for defining the shape freedom, from which the fluid and solid grids are generated. The aerodynamic efficiency is computed using a Reynolds-Averaged Navier-Stokes solver based on the finite volume method. The maximum von Mises stress and eigenfrequencies are computed using a linear stress and vibration solver based on the finite element method. The CFD, stress, and vibration solvers each have adjoint capabilities, allowing an efficient evaluation of the gradients at a cost independent of the size of the design space. The fluid and structural domains are coupled with the CAD kernel to form a CAD-based adjoint multidisciplinary optimization framework (MDO).

    In this presentation, the CAD-based adjoint MDO framework is introduced. In particular, the differentiation of the FEM solver, using the AD tool CoDiPack, is discussed. This includes the differentiation of both the linear stress solver for the maximum von Mises stress gradients, which results in one additional linear system solve, and the vibration solver for eigenvalue gradients, which results in one additional outer product per eigenvalue. Performance results for gradient calculations are presented, as well as aerodynamic shape optimizations of a radial turbine with both maximum von Mises stress and vibrational resonance-reducing constraints.

  • Fri
    02
    Aug
    2019

    11:00SC Seminar Room 32-349

    Alessandro Gastaldi, Airbus Defence and Space

    Title:
    High-fidelity airframe shape and sizing optimisation for maximum aircraft performance

    Abstract:

    The objective of the aircraft development process is to formulate a design which provides the highest possible performance at specific operating conditions, while satisfying many, often competing requirements emerging from the various engineering disciplines involved. Multidisciplinary Design Optimization (MDO) can be used to systematically identify and efficiently resolve complex trade-offs at every stage of the aircraft design process. In this context, the performance of an aircraft is measured in terms of e.g. range or endurance, for a given mission or at specific flight regimes. To guarantee the feasibility of each solution, it is essential to define a complete criteria model including the design-driving constraints emerging from the various engineering disciplines involved. Allowing simultaneous variation of the internal structural layout and the external aerodynamic shape enables improved solutions which satisfy such structural, aerodynamic and aero-elastic criteria, beyond what is achievable by a sequential approach. Naturally, the analysis models employed in the optimisation must capture the phenomena and interactions which determine these quantities of interest. The ultimate objective of the present work is to combine the above elements in a robust and flexible software framework for aircraft performance optimisation through simultaneous shape and sizing optimisation, using fully coupled high-fidelity aerodynamic and structural solvers. This task includes the investigation of best practices for the integration of such a demanding multidisciplinary analysis in an optimisation process, in order to ensure manageable computational cost and robust, automated model evaluations.