FlowHead

The EC-funded research project FlowHead, http://flowhead.sems.qmul.ac.uk, has developed adjoint-based optimisation methods for shape and topology optimisation of fluid flow with application to the automotive industry. The focus in FlowHead was on:

Invited speakers are:

Prof. R.P. Dwight (TU Delft)
"Discrete Adjoint Methods and Applications: Beyond Optimization"

The implementation of a robust discrete adjoint is fraught with issues. We examine some of the more intractable problems and how they may be mitigated or circumvented. Given the high effort required to implement and maintain an adjoint capability, it's desirable to get the most out of it. Applications other than optimization will be discussed, in particular error estimation, uncertainty quantification, surrogate modelling, and data assimilation/model updating.

Prof. K. Giannakoglou (NTUA Athens):
"Development and application of continous adjoint methods."

Prof. R. Loehner (George Mason Univ., Washington):
"Adjoint design optimisation"

Adjoint-based design optimization has seen major developments over the last two decades. Due to its very low computational cost, it has become the method of choice for optimization problems that are amenable to gradient-based techniques. And experience in engineering, where controlability and smooth, predictable performance are key requirements, has shown that adjoint-based techniques can be used advantageously for many more problem classes than initially thought.

After a brief introduction, the talk will focus on adjoint- based techniques for transient problems. As the adjoint proceeds backwards in time (from the end to the start of the flow simulation), the main difficulty encountered for large-scale applications is the massive amount of data storage required. Techniques such as checkpointing (i.e. trading CPU for storage) and data compression can aleviate this core difficulty to a large extent.

Several cases from design for blast-structure protection will be used to exemplify the techniques developed.

Prof. U. Naumann (RWTH Aachen):
"Toward First- and Higher-Order Discrete Adjoint Flow Solvers"

Traditionally, Algorithmic Differentiation (AD) has been regarded as a method for the computationof derivatives FOR use within numerical algorithms, such as nonlinear solvers and optimizers. More recently, the need for AD OF numerical algorithms has become more apparent. We have been receiving requests for first- and higher-order tangent-linear and adjoint versions of numerical libraries (e.g. the NAG library) as well as of flow solvers such as OpenFOAM. In this talk we discuss first results of black-, grey-, and white-box interpretations of AD principles when applied to numerical algorithms.

Dr. A. Naumovich (DLR)
"Algebraic multigrid for solution of adjoint equations in compressible aerodynamics"

Linear systems arising from linearization of second-order accurate discretizations of compressible Euler and Navier-Stokes equations are often very stiff and their effcient solution is a challenging task.
    Based on the experience made with the linear counterpart of the DLR TAU code, available solution approaches, namely geometric multigrid and ILU-based approaches combined with Krylov methods, have certain disadvantages. Although the geometric multigrid approach is memory-effcient, it often demonstrates an unsatisfactory convergence behavior, especially for turbulent test cases. In the same time, the ILU-based approaches demon- strate rather good convergence rates as soon as a high enough ILU decom- position level is employed. However, they tend to become very memory intensive and thus are not always affordable in practice.
    In this work, we offer an alternative solution approach based on algebraic multigrid (AMG). Unlike the geometric multigrid, AMG does not explicitely rely on the geometry of the grid. It is therefore a very attractive approach for problems dealing with unstructured grids. Originally developed for scalar elliptic problems, AMG needs to be extended and its components must be properly chosen and adjusted for the approach to be applicable to advection- dominated systems of PDEs.In collaboration with Fraunhofer SCAI, using their AMG software (the SAMG library), we develop a solution approach for the adjoint equations.
    We demonstrate application of the suggested solution technique to a range of problems from subsonic Euler equations to transonic Reynolds-averaged Navier–Stokes equations.

Call for Papers

Oral and poster presentations are invited on the topics above. Please submit online a one-page abstract in PDF format by January 9, 2012 if you would like to present a paper or a poster. Please direct your queries to the FlowHead project coordinator Dr. J.-D. Mueller, j.mueller@qmul.ac.uk.

In addition to the topics covered by the project as listed above, submissions are welcome on topics such as:

Organising Committee

The coordinator of the FlowHead project is:

Key Dates

Venue

The conference will be held on the Faculty Club, Lichtenbergstrasse 2a [ directions ], on the Garching campus of TU Munich. Garching is conveniently reached from MUC airport and connected to the Munich U-Bahn/S-Bahn train system, http://www.mvv-muenchen.de

Costs

Cost of the conference is €300 to cover lunches, coffees and a conference dinner on Wed evening at Weihenstephan Brewery including bus transfer.

All participants must pay the conference fee via bank transfer before the conference the account of the Tech. Univ. Munich as shown below. Your payment must be received by TUM's bank before the conference.

You must make sure to indicate in the comment box the cost centre number for this event and the name of the participant:

Accomodation

Block reservations for a limited number of rooms have been taken out at the hotels listed below. The hotels are within easy reach of the U6 Subway stop 'Garching', the TUM Garching campus is one subway stop away at 'Garching-Forschungszentrum'. The reservations expire 28/02/12.

Conference on Industrial Design Optimisation for Fluid Flow, Munich, 28-29 March 2012

Overview