fluid-structure coupling for aerodynamic analysis and design: a

American Institute of Aeronautics and Astronautics

1

Fluid-Structure Coupling for Aerodynamic Analysis and Design – A DLR Perspective

R. Heinrich1 and N. Kroll.2 German Aerospace Center (DLR), Institute of Aerodynamics and Flow Technology

Lilienthalplatz 7, 38108 Braunschweig, Germany

J. Neumann3 German Aerospace Center (DLR), Institute of Aeroelasticity

Bunsenstraße 10, 37073 Göttingen, Germany

and

B. Nagel4 German Aerospace Center (DLR), Air Transportation System and Technology Assessment

Eißendorfer Straße 38, 21073 Hamburg, Germany

Accurate prediction of the aerodynamic performance of an aircraft requires coupled fluid/structure simulations. The aerodynamic loads acting on the aircraft result in deformations of its flexible structure, which in turn causes changes in the fluid flow and thus the loads. Static aeroelastic effects are also present in high Reynolds number wind-tunnel testing. For proper comparison of numerical and experimental data the aeroelastic deformations of the wind tunnel model have to be taken into account or modeled within a CFD simulation. At DLR a fully automatic process chain for both steady and unsteady aeroelastic applications has recently been developed, allowing the coupling between the in-house CFD codes TAU and FLOWer and the commercial FEM codes ANSYS and NASTARAN. The CFD codes have been extended to include functions for aeroelastic simulations, including an interpolation module for the transfer of the aerodynamic loads into structural nodes between the CFD and CSM meshes. Emphasis is put on robust volume mesh deformation translating the structural displacement into the CFD mesh since this is a critical issue in the overall process Steady and unsteady applications are presented to demonstrate the feasibility of the approach for industrial applications, including complex high-lift configurations, a rocket nozzle and maneuvering aircraft. First results for design and optimization based on coupled fluid/structure simulations are also presented. Finally, the paper discusses future developments necessary to further enhance the simulation capabilities for multidisciplinary simulation and optimization.

Nomenclature AMIF = Aerodynamic Mesh Interface Format APDL = ANSYS Parametric Design

Language CFD = Computational Fluid dynamics CSM = Computational Structure Mechanics DES = Detached Eddy Simulation DNW = German-Dutch Wind Tunnel

PSP = Pressure Sensitive Paint RANS = Reynolds Averaged Navier-Stokes

Equations S_BOT = Sizing Robot TE = Trailing Edge WRBN = Wing Root Bending Moment α = Angle of attack

1 Senior Research Scientist, Center for Computer Applications in Aerospace Science and Engineering (C2A2S2E ) 2 Head of Center for Computer Applications in Aerospace Science and Engineering (C2A2S2E ) 3 Research Scientist, Branch Aeroelastic Simulations 4 Head of Branch Aircraft Design

46th AIAA Aerospace Sciences Meeting and Exhibit7 - 10 January 2008, Reno, Nevada

AIAA 2008-561

Copyright © 2008 by Ralf Heinrich. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.


2

DPWS = Drag prediction workshop ETW = European Transonic Wind Tunnel FEM = Finite Element LE = Leading Edge NWB = Low Speed Wind Tunnel LU-SGS = Lower-Upper Symmetric

Gauss-Seidel MDO = Multidisciplinary Optimization PARA_MAM = Parametric, Simple and Fast Mesh

Based Aircraft Modelling Tool

CL = Lift coefficient CD = Drag coefficient g = gravity L/D = Lift to drag ration M = Mach number

rrΔ = Displacement vector Re = Reynolds number wi = Weight of node with index i

I. Introduction The aerodynamic performance of large transport aircraft operating at transonic speeds is highly dependent on

the deformation of their wings under aerodynamic loads. Hence accurate performance predictions require fluid-structure coupled numerical simulations to determine the aerodynamics of the configuration in aeroelastic equilibrium. The influence of aeroelastic effects on the aerodynamics is demonstrated for a generic transport aircraft configuration in Figure 1. The numerical analysis of aircraft performance at cruise conditions is usually based on the flight-shape geometry. In Figure 1 this shape, known as the 1g shape, is represented by the grey shaded geometry. In addition, for the structural design of the wing and for aircraft certification, the root bending moment of the wing due to a 2.5g load is of great interest. Loads of this magnitude can occur during a pull-up maneuver, for example. Usually, for simplicity, the CFD mesh generated for the flight shape is used to compute the flow under a 2.5g load without taking into account the additional deformations of the wing. In the right part of Figure 1 the surface pressure coefficient distributions is plotted for a cut at a spanwise location of 90% of the wing halfspan. The red curve represents the result of the computation using the flight shape (1g) whereas the green curve shows results obtained by a fluid-structure coupled computation. Although the same lift coefficient corresponding to a 2.5g load is targeted in both cases, a large discrepancy in the resulting pressure distribution is observed. The load distribution over the aeroelastically deformed shape, shown in shades of green in the left picture in Figure 1, comes along with a reduced root bending moment compared to the uncoupled computation: in the outboard sections the load is significantly reduced, whereas the load is increased in the inboard area, in order to achieve the same target lift.

Figure 1. Comparison of cp-distributions at 90% half-span for 2.5g load with and without fluid/structure coupling (CL=0.78, M=0.78, Re=25x106, height=22500ft.

Clearly, more accurate prediction of the wing bending moment based on fluid-structure coupled computations can help to reduce the structural weight and, as a consequence, the fuel consumption of an aircraft. Aeroelastic effects can also play a significant role in wind-tunnel testing under high Reynolds number conditions, as experienced within the European project HIRETT1. Due to the high dynamic pressure in the wind tunnel the deformations can reach a magnitude2 which cannot be neglected in the numerical simulation in the framework of code validation.

The improvement of maneuverability and agility is a substantial requirement for modern fighter aircraft. Most of today’s and probably tomorrow’s fighter aircraft will be delta-wing configurations. The flow field over such configurations is dominated by multiple vortices. For a maneuvering aircraft the time lag between the dynamic


3

vortex position and state and the static vortex characeristics at the same instantaneous on-flow conditions can lead to significant phase shifts in the loads distribution. Reliable results for the analysis of the aircraft characteristics can only be achieved by a combined non-linear integration of the unsteady aerodynamics, the flight mechanics and the elastic deformation of the aircraft structure.

Consequently, at DLR, major efforts have been devoted to couple the in-house flow solvers FLOWer and TAU with structural analysis codes. The activities include the development of efficient and robust grid deformation tools, accurate interpolation tools for transferring data between the CFD grid and the FEM grid as well as the implementation of suitable interfaces between the flow solvers and the structural analysis software. Concerning the structural mechanics, both high-fidelity models (ANSYS, NASTRAN) and simplified models (beam model) are considered. To enable realistic large scale applications, it is important that the process chain is able to run automatically and on heterogeneous hardware including parallel batch computers. The coupling procedure was designed for the simulation of steady and unsteady fluid/structure interactions. For the prediction of maneuvering aircraft the numerical environment was extended to the in-house flight mechanics tool SIMULA, which provides basic functionalities for flight mechanics simulations and flight control. Furthermore, an effort was made towards the integration of the fluid/structure coupling into a multidisciplinary optimization process.

This paper describes the fluid/structure coupled process chain implemented by DLR and its application to steady and unsteady problems. The different components of the simulation environment are discussed with special emphasis on the interpolation module between CFD and CSM and the volume mesh deformation being one of the most critical issues in the overall procedure. In terms of applications focus is put on demonstrating that fluid/structure interactions can have a significant impact on the aerodynamic analysis and design of transport and combat aircraft. The simulation of the aeroelastic behavior of an aircraft is not subject of this paper.

II. The Process Chain and its Components In Figure 2 a flow chart shows the sequence of steps followed during a coupled computation to achieve the

aeroelastic equilibrium. Figure 3 shows a corresponding convergence history of the density residual and the lift of the CFD code for a generic wing-body configuration in the transonic regime. Usually in a first step a CFD computation is performed using the undeformed geometry. Output of the CFD computation is the aerodynamic load on the coupling surface (pressure and friction coefficient or force distribution). The aerodynamic loads have to be mapped from the nodes of the CFD surface to the nodes of the CSM surface. Therefore the CFD and the CSM code have to export the coupling surfaces.

For the data transfer the interpolation module is required. It imports the coupling surfaces, the data to be interpolated and performs the mapping from CFD nodes to CSM nodes. Depending on the quantity to be interpolated different interpolation techniques can be selected. After mapping the aerodynamic loads from the CFD to the CSM side the structural analysis code is started. Output of the structural analysis code is the deformation of each structural node of the CSM coupling surface. These have to be interpolated to the nodes of the CFD coupling surface. Due to the change of the CFD surface the volume mesh has to be deformed as well. Input for the volume mesh deformation is the undeformed CFD mesh and the deflections of the surface nodes. Now the second loop of the coupling procedure can be conducted starting again with a CFD computation. The new (deformed) volume mesh and the last CFD solution are used as input for the second CFD computation. In the convergence history (Figure 3) we see that the density residual is increased and the lift coefficient changes its value. After a number of iterations a converged state is reached again. The aerodynamic loads are transferred again to the CSM surface using the interpolation module. Then a CSM solution is computed and so on. The process is repeated until the equilibrium state is reached or a user specified number of iterations have been performed. To determine if the equilibrium state is reached the solutions of the actual and the previous coupling cycle are compared. If the change of aerodynamic coefficients or maximum deflection is smaller than a prescribed value the coupled process is stopped and the aeroelastic equilibrium is assumed to be reached.

For unsteady fluid/structure interactions different time coupling schemes with or without sub-iterations are realized. The baseline scheme is a “Conventional Serial Staggered” (CSS)3 algorithm modified with a predictor-corrector scheme in each time step for the transformed structural forces. For the time integration of the CSM equations explicit or implicit Newmark algorithms are used. More details are discussed in4,5. The design of the coupling procedure also allows an easy plug-in of a third discipline, the flight mechanics. The user can decide, if the resulting set of coupled equations is solved with or without sub-iterations. Current investigations focus on determining which coupling scheme is most beneficial, if all three disciplines (CFD, CSM and flight mechanics) have to be taken into account.


4

Figure 2. Flowchart of a steady coupled computation (aeroelasitc equilibrium).

Figure 3. Convergence history of a coupled fluid/structure computation for the generic DLR F11 wing-body configuration under cruise conditions.

A. Flow Solvers As flow solvers the block-structured FLOWer-Code6,7 and the hybrid (unstructured) TAU-Code8,9,10 are

available. Both codes are developed by the DLR Institute of Aerodynamics and Flow Technology and they are well established tools for aerodynamic applications in DLR, aerospace industry and universities11,12,13. The compressible, three-dimensional, time-accurate Reynolds-Averaged Navier-Stokes equations for rigid bodies in arbitrary motion are solved. For spatial approximation a finite-volume method with second order upwind or central discretization with scalar or matrix artificial dissipation is used. In FLOWer cell centered and cell vertex formulations are provided, whereas TAU uses a vertex centered dual mesh formulation. The discrete equations are integrated explicitly by multistage Runge-Kutta schemes, using local time stepping and multigrid acceleration. In FLOWer the explicit scheme is used in combination with implicit residual smoothing, whereas in TAU the implicit LU-SGS scheme is additionally available. For time accurate computations the implicit dual time stepping method is employed. Preconditioning is used for low speed flow simulations. Various turbulence models are available, ranging from eddy viscosity to full differential Reynolds stress models including options for DES (Detached Eddy Simulation). The Chimera technique enhances the flexibility of FLOWer and TAU with respect to complex geometries or independently moving bodies. For the simulation of aeroelastic phenomena both codes have been extended to allow geometry and mesh deformation. A key feature of TAU is the grid adaptation capability for hybrid meshes based on local grid refinement and wall-normal mesh movement in semi-structured near-wall layers, allowing efficient resolution of detailed flow features. A discrete adjoint solver was developed within TAU enabling efficient gradient-based shape optimization and goal-oriented mesh adaptation. The TAU-Code is not a single code but is composed of a number of modules and libraries which can be used within a Python scripting framework which allows for inter-module communication without file-I/O, i.e, using common memory allocation. Both codes, FLOWer and TAU, have been efficiently parallelized and ported to a variety of platforms.

For the communication with the interpolation module between CFD and CSM, an appropriate interface in the flow solver is needed to transfer the relevant data. To specify the coupling surface the xyz-coordinates of the surface points and a connectivity list are required. For each point the xyz-coordinates are associated with a node index. These indices are used in the connectivity list to specify the surface elements of the coupling surface. Triangles as


5

well as quadrilaterals can be used as element type. Each surface element is associated with a boundary marker. The boundary marker can be used to group a selection of surface elements to components like e.g. wing, slat or flap. This is very useful for setting up a coupled computation (see subsection C). For each surface node variables can be ex- or imported. Export values for the CFD code are currently pressure and friction coefficients in xyz direction or forces. Import values are the deflections in xyz direction. In the current implementation the transfer of data is realized with file IO. As data format the AMIF (aerodynamic mesh interface format) specification of MSC13 can be used. However, it should be emphasized that a change of the data format or a switch to direct communication with the interpolation module can be done with limited effort.

B. Structural Analysis Software On the structure side the well known commercial FEM codes NASTRAN13 and ANSYS15 are used. No source

code is available, so interfaces cannot be directly plugged into ANSYS and NASTRAN. However, within ANSYS a powerful scripting language called APDL can be used. In our case we make use of two ANSYS scripts. One script is used to export the coupling surface in the same format as the CFD code. This is done only once at the beginning of a coupled computation. A second script is used to import the structural loads (which have been interpolated from the CFD nodes to the CSM nodes), to perform a FEM analysis and to export the resulting deflections of the CSM nodes in AMIF format. Besides the commercial FEM codes the beam generator of the Technical University of Aachen is used16. This model is used to automatically create reduced structural models by representing the wing box as a multi-axial Timoshenko beam. The equivalent Timoshenko beam is characterized by the cross-sectional coordinates of the centers of mass, shear and bending.

For unsteady applications usually an in-house discrete approach is preferred, which simplifies the coupling process significantly. Within this approach reduced system matrices MAA (mass matrix) and KAA (stiffness matrix) from a NASTRAN or ANSYS eigenvalue solution are used. To get rid of severe stability restrictions for time integration a Newmark algorithm is used. This allows to synchronize the time stepping schemes on CFD and CSD side. Loose as well as close coupling schemes can be constructed easily. The software package for this approach is written in the MATLAB language5. The tool can also be used independently from the MATLAB environment, if it has been compiled with the appropriate MATLAB compiler in the C or C++ language.

C. Interpolation Module The task of the interpolation module is the mapping of aerodynamic loads from the CFD to the CSM coupling

surface and the mapping of deflections from the CSM coupling surface back to the CFD coupling surface. Depending on the application and quantity to be interpolated, different interpolation techniques can be selected. Main input of the interpolation module is the CFD and CSM coupling surfaces, the quantities to be interpolated or calculated, the selection of the specific interpolation technique and the specification of the interpolation direction (source and destination coupling surface).

In many cases different coordinate systems and scalings are used on the CFD and CSM side. Therefore the user can specify (per input parameters) a number of transformations for each coupling surface resulting in a common coordinate system. Translation, rotation (around user specified axis and angle) and scaling can be selected as transformation type.

Figure 4 shows the outer wing region of the CFD and the CSM coupling surfaces of a generic wing. On both sides a similar mesh resolution is used. The shape of both meshes matches well and a linear interpolation of the pressure or pressure coefficient is sufficient for the mapping of the aerodynamic loads. Based on the interpolated pressure corresponding forces can be calculated for each surface node. In Figure 5 the pressure distribution on the CFD side is compared to the interpolated values on the CSM side. A good agreement of input and interpolated values is achieved. It should be mentioned that the linear interpolation of e.g. pressure will not result in a conservative interpolation with respect to forces, moments and work. It should be noted that this is of secondary interest for steady applications. However, as stated below for unsteady fluid/structure interactions conservative interpolation is of great importance.


6

Figure 4. CFD and CSM coupling surface for generic LANN wing.

Figure 5. Comparison of calculated pressure distribution on CFD side and linear interpolated pressure distribution on CSM side (generic LANN wing, M=0.82, α=0.6°, tip region, inviscid flow).

In cases in which the geometries or meshes on CFD and CSM side do not match or only scattered data are available on the CSM coupling surface, a linear interpolation of pressure or other quantities is not possible. In that case the user can select a nearest neighbour search for the forces, see Figure 6. For a given point “i” on the CFD side the nearest neighbour on the CSM side is searched. If a connectivity on the CSM side is given (like in the figure), the nearest point with the same normal orientation is selected. If the nearest point “j” is found, Fi,CFD is added to the value of Fj,CSM (more than one aerodynamic force can be mapped to the same CSM node). Additionally a moment Fi,CFD x Δrij is mapped to the node j, because the force is moved. This ensures a conservative interpolation scheme with respect to the force and moment balance on the CFD and CSM side.

Figure 6. Mapping of a force from CFD side to nearest neighbor on CSM side. Figure 7 shows the influence of the interpolation scheme for the previous example. On the left we see the CFD surface mesh and a slice through the surface mesh at 90% half wing span. In black we see the resulting profile and pressure coefficient distribution without coupling (rigid surface). Three coupled computations have been made, all until the equilibrium state is reached. In red we see the result using linear interpolation of the pressure coefficient, in green (dashed) the results achieved with the conservative force interpolation described before and in blue (dash dot) if a nearest neighbour search is used for the pressure coefficient. No large differences between the coupled computations can be observed. Also global lift coefficients in equilibrium agree very well.


7

Figure 7. Comparison of surface pressure distribution close to the wing tip in aeroelastic equilibrium obtained with different interpolation techniques (generic LANN wing, M=0.82, α=0.6°, inviscid flow).

Figure 8. Computed and interpolated deflections normal to the wing plane (generic LANN wing, lower and upper surface).

Nearest neighbour search is not an appropriate technique for the interpolation of surface deformations. A linear interpolation of the deflections can also be problematic if the resolution of the coupling surfaces is very different or if no connectivity is given on the CSM side. As is known from literature17,18,19,20 interpolation schemes based on radial basis functions are appropriate for this task, even if only scattered data are available. This approach is also adopted here. These techniques are very well suited for smooth functions, and usually the deformation of aerodynamic components are smooth. Figure 8 shows for the previous example the deformations in y direction (normal to the wing plane). In red lines denote deformations computed by ANSYS, the blue dashed lines indicate the interpolated values. Both results agree very well.

Figure 9. Computed (symbols) and interpolated deflections (red line) normal to wing plane for a high-lift application (DLR F11 wing/slat/flap starting configuration). However, problems can arise if different components of an aircraft have to be taken into account, like for example a high-lift aircraft configuration. Figure 9 shows deflections of the DLR F11 wing/slat/flap configuration. In the upper half of the right figure lines of constant deflection in z direction are indicated by green (computed deflections) and red (interpolated deflections) lines. In the lower half of the figure deflections in wall normal direction are shown in the span wise section η=71 % (half wing span). The green symbols correspond to the


8

deformations computed with the structural analysis tool. The different behaviour of slat, wing and flap is evident. For the same x coordinate the deformation of the slat and flap is slightly higher. In other words: The gaps between the wing and the two high-lift devices have changed. This cannot be handled properly by a single interpolation function, as becomes clear from the interpolated displacement (red line, left). The quality of the interpolation can be significantly improved by using local interpolation functions, which are calculated for each component. In this case the agreement of calculated and interpolated displacement is excellent as shown in the right part of Figure 9.

For the prediction of unsteady fluid/structure interactions the underlying spatial coupling scheme needs special care to ensure conservation with respect to forces, moments and work performed on both the aerodynamic and structure dynamic side. An approach based on the scattered data interpolation mentioned above is used that ensures an equivalence of the potential and kinetic energy of the structure mechanical model and the work performed by the forces on the aerodynamic surface. Several scattered data interpolation methods with and without compact support radius are implemented as described for example in18.

D. Mesh Deformation If the CFD surface mesh is deformed the volume mesh has to be adapted as well. For TAU an algebraic method

has been developed21 in order to avoid time consuming iterations for solving equations based e.g. on linear elasticity or spring analogy. The displacements which are the input for the deformation tool and the rotation of surface points are transported into the interior of the grid by an advancing front technique. Depending on the ratio between the local point displacement and the cell size the displacement is reduced by some fraction in each step of the front. This procedure ends when no more grid points are moved during a sweep. In parallel computations, due to displacements coming from neighbouring domains the sweeps are continued until the grid does not change any more. Since one sweep requires negligible costs only, this is not a significant drawback of the parallel mode where usually an order of 10 to 20 sweeps is needed.

This algebraic method is robust enough for small and sometimes also for medium displacements and can handle e.g. wing tip deflections of one or several chord lengths. It has been observed that the limit, i.e. the deflection which makes the first cell collapse can be extended considerably when accepting the collapse of a few more cells only. Thus, in a second stage of the deformation an algorithm is started which repairs collapsed cells. This increases the robustness considerably and allows for large grid deformations. To do so, each region in the grid containing collapsed cells is marked such that it is bounded by valid cells only. With the shape of this boundary in the deformed and the undeformed grid a transformation can be computed applying radial basis functions like the classical volume spline, which allows rebuilding the collapsed cells as images of the original ones. As long as these regions remain small the additional computational costs for the local volume spline is low.

As an example that this robust approach allows going beyond realistic deformations Figure 10 shows the maximum possible wing tip deflections for a viscous hybrid grid. The grid is composed of 2.5*106 points. CPU time requirements on one single Opteron CPU is less than 2 minutes for small deflections (of about a chord length) and less than 10 minutes for the maximum wing tip deflection.

Maximum wing tip deflections in

parallel (left) and sequential mode (right)

Figure 10. Maximum wing tip deflections and details of the deformed viscous hybrid grid for the DLR F6 configuration (available from DPWII).

Tip detail


9

Recently, an alternative mesh deformation tool based on radial basis function was developed which allows deformation of both block-structured and hybrid unstructured meshes. The basic idea is to apply the interpolation functions calculated for the surface mesh deformation (section C) to the nodes of the volume mesh as well. Additionally, the resulting deflections can be superimposed with a blending function based on the wall distance in order to achieve zero deflections for a specified distance from the wall, for example at the farfield. This idea can also be utilized if more than one interpolation function is used like in the high-lift application shown before where a separate interpolation function was created for each aircraft component. A global interpolation function is created by weighting each component function based on the wall distances. Therefore wall distances are calculated for each relevant component, as sketched in Figure 11.

For a configuration decomposed in n components the deformation rrΔ of a mesh node ),,( zyx is calculated as

( ) ( )∑=

Δ=Δn

iii zyxrwzyxr

1,,,, rr

, where ),,( zyxrir

Δ

denotes the deformation based on the individual interpolation function i, and wi indicates the corresponding weight. The sum of all weights equals one.

Figure 11. Wall distance distribution for each component of 2D high-lift configuration.

Figure 12. Comparison of original and deformed mesh for 2D high-lift configuration.

The weights are calculated as

∑=

⎟⎟⎠

⎞⎜⎜⎝

⎛

⎟⎟⎠

⎞⎜⎜⎝

⎛

=n

j

p

j

p

ii

zyxwalldist

zyxwalldistw

1 ),,(1

),,(1

where walldistj(x,y,z) denotes the wall distance of the mesh node (x,y,z) relative to aircraft component j, p is a user specified exponent. As an example Figure 12 shows the movement of a flap for a 2D high-lift configuration. For the slat and wing component zero deflection is prescribed, whereas for the flap a translation downward and in flow direction is prescribed. The top half shows the original undeformed mesh. The bottom half demonstrates the result of the deformation using the global interpolation function based on the blended component interpolation functions. For accurate coupled fluid/structure computations it is important to conserve the mesh quality while deforming the mesh. This property was checked for a 3-element-airfoil and an isolated swept wing. For both configurations reference computations have been performed. For the airfoil case a hybrid mesh was used and the computation was carried out for M=0.2, Re=4x106, and α = 21.4°. The computation for the wing was performed on a structured mesh at M=0.82, Re=7.17x106 and α = 0.6°.


10

Y

X

Z

slice at 82.5%halfspan

deformed mesh,5° rotated surface

rigid meshsurface

X

cp

0.45 0.5 0.55 0.6

-0.5

0

0.5 rigid meshdeformed mesh

slice at 82.5%halfspan

Figure 13. Comparison of surface pressure distributions of reference configuration and equivalent configuration (3-element airfoil, M=0.22, Re=4x106).

Figure 14. Comparison of surface pressure distributions of reference configuration and equivalent configuration (LANN wing, M=0.82, Re=7.2x106).

In order to test the mesh deformation procedure, surface deflections are computed resulting in surface meshes rotated 5° rigidly. Based on the computed deflections the volume meshes are deformed using the radial basis function approach. Now computations are made for an angle of attack which is reduced by 5°. Both computations should result in equivalent solutions. If the mesh quality is conserved during the deformation process, the computed surface pressure distributions should be the same on the deformed meshes. Figure 13 shows results for the 3-element airfoil. In the upper left corner of the figure both meshes are plotted. In red the rigid starting mesh and in green the deformed mesh is shown. In the upper right part the deformed mesh is rotated -5° by rigid deformation, so that the original surface and the surface of the deformed mesh are the same. In the region close to the surface the meshes almost match. In the lower part of Figure 13 the pressure coefficient distributions are compared. The agreement is very good demonstrating equivalent quality of the original and deformed mesh. A similar behavior is found for the wing application, see Figure 14. In the left part the surface mesh and the symmetry plane of the original mesh are shown. The deformed surface is displayed in grey color. At 82.5% half-span of the wing the surface pressure distributions are compared for the rigid and deformed mesh. Again a very good agreement is achieved. It should be noted that the robustness of the deformation tool based on radial basis function was demonstrated for many applications. It was used for all high-lift applications shown in this paper.

E. Process Chain

Several simulation environments for coupled fluid/structure computations have been realized in different application areas. The most frequently used process chain is controlled by a shell script. All components of the process chain can run locally on the computer where the script is launched. However, usually only a part of the components runs locally. Especially for applications using ANSYS or NASTRAN the CSM code runs on a separate computer (question of license model used). Therefore all input files that are needed are copied automatically to a temporary directory on the remote host (using secure copy). The ANSYS/NASTRAN job is then started via secure shell. After successful computation the file containing the resulting deformations is copied back as input for the interpolation module. For large scale applications the CFD codes run on batch computers in parallel. Depending on the target host, a job script is created automatically. For each supported host a batch-script template exists. In general only a few lines of the batch-script have to be adjusted, like e.g. the number of processors, a path for executables and so on. For each target host an additional procedure for adjusting the script is included in delivery of the coupling environment.. All input files needed for FLOWer / TAU are transferred to a temporary directory of the remote host via “secure copy”, and the resulting surface solution is transferred back to the local (controlling) host. During execution of the CFD run on the batch computer the controlling shell script on the local host is “sent to sleep”. After finishing the CFD computation control is given back to the local shell script.


11

For each coupling cycle the user has the possibility to adjust parameters of the CFD computation. For example the angle of attack can be increased after maybe 10 coupling cycles to perform a coupled computation for the next point of a polar.

A key ingredient for large-scale applications is the ability to complete a coupled computation without the need for user interaction, on a heterogenous network where the CFD simulation runs on batch computers in parallel.

III. Steady Fluid/Structure Interactions Several examples are discussed within this section indicating the range of applications and the level of

complexity which can be handled with the coupled fluid/structure process chain described above.

A. Rectangular swept wing As a first verification test case for the TAU-ANSYS process chain the rectangular swept reference wing of a

German aeroelastic research program conducted at the Technical University of Aachen22 has been selected. For this configuration the DLR FLOWer-Code coupled with a Timoshenko beam model was successfully validated by the Mechanics Department of the Technical University of Aachen during the last years, see for example23,2 The wing, shown in Figure 15 has a span of 1.5m and a sweep angle of 45°. Based on the characteristics of the beam model provided by the Aachen University, a FEM model was created which can be used by the structural analysis solver ANSYS. Coupled TAU-ANSYS computations were performed and the results were crosschecked with validated results of Ballmann et al.2,22. In both computations the same block-structured CFD mesh was used. In the lower half of the figure the undeformed wing and the wing in static equilibrium is shown for a Mach number of 0.22, a Reynolds number of 1.7x106 and an angle of attack of 4.48°. Slices at 90% half span are shown on the top half. The blue lines show the reference results, whereas the results of TAU coupled with ANSYS are indicated with red symbols. The equilibrium shape as well as the pressure coefficient distribution is in excellent agreement with the reference solution.

Figure 15. Comparison of results obtained with TAU-ANSYS and FLOWer coupled with a beam model

B. Wing-body configuration The DLR-F6 model, representing a generic transport aircraft configuration featuring a wing-fuselage and a wing-

fuselage-pylon-nacelle configuration, has been selected as one of the test cases of the AIAA CFD Drag Prediction Workshop (DPW) Series24. The major prediction challenges associated with the F6 configuration are the small areas of flow separations at the wing-fuselage junction, on the upper wing close to the trailing edge, and on the wing lower side at the wing-pylon-nacelle junction. Because the sequence of DPWs showed increasing levels of prediction accuracy, the 3rd workshop (June 2006) clearly indicated the need for highly accurate experimental data


12

of the F6 configuration as these local separations turned out to be the major source of numerical uncertainty. It was decided to test the F6 configuration in the NASA Langley’s National Transonic Facility (NTF).

As the aerodynamic forces associated with the required flow conditions exceed the model’s design load limits various model modifications were implemented and comprehensive fluid-structure coupled analyses were performed for the desired flow conditions in order to accurately determine aerodynamic loads and meet NTF safety requirements25. The CFD geometry used for the F6 flow simulations is based on a half-model of the wing and fuselage. An unstructured hybrid CFD grid with 2.46 million grid points, Figure 16, was used. The structural model consists of the wing and balance block only. The fuselage is not considered as it is not relevant for the analysis of wing loads. The FE discretization consists of 160.000 nodes (Figure 17). Coupling is established on the wings upper and lower surface with approximately 35.000 surface grid points on CFD side and 4.800 surface nodes on CSM side.

Figure 16. CFD grid of the DLR F6 wing-body configuration, source25.

Figure 17. FE model of the DLR F6 wing-body configuration, source25.

The three-dimensional state of stress in a body due to external forces is commonly assessed by scalar formulations which allow to determine the overall stress magnitude and to predict failure. For ductile materials under tri-axial loading the von Mises equivalent stress represents a uni-axial stress equivalent to the body’s actual three-dimensional stress state. In order to prevent failure in the form of plastic deformation von Mises equivalent stress must not exceed the material’s allowable yield stress. In the subsequent structural loads analysis von Mises equivalent stress is determined over the entire model and areas where the material’s allowable yield stress is exceeded are identified.

Figure 18. Von Mises Stress Distribution around Pylon Notch, Comparison of uncoupled (left) and coupled (right) Simulations (Re=5.0×106, CL=0.6, Ta=322K), source25.

In Figure 18 von Mises stress distribution is plotted around the engine pylon notch located on the lower wing surface. Aerodynamic loads for this analysis were taken from an uncoupled CFD simulation on the undeformed geometry (left) and from a coupled analysis (right). Color scaling is matched to the material’s certified yield strength of Rp0.2 = 420N/mm2. The highest stress levels were found immediately around the pylon notch. As the coupled simulation comprises the loads redistribution caused by the wing’s bending and twist deformation a significant stress


13

reduction is observed in both the pylon notch area and surrounding wing surface. However, some plastic deformation still exists indicating that stress peak levels have not changed much. Therefore, the influence of ambient temperature, lift coefficient and Reynolds number reduction on the structural loads was investigated. Based on the fluid-structure analysis two test cases were defined to determine the maximum allowable flow conditions and to set the wind tunnel system test limits for the NTF tests.

C. High-lift configuration under Low and High Reynolds Number Wind Tunnel Conditions Within the European project EUROLIFT I experiments26 were made for high lift applications under high Reynolds number conditions in the ETW wind tunnel in Cologne. During the experiments visible deformations of the wing were noticed. The follow-on project EUROLIFT II27 addressed the question whether the deformations under high loads have a large or only a small, negligible influence on the aerodynamic behavior. In the following, first results achieved with the TAU-ANSYS process chain are shown. Up to now the DLR-F11 configuration in start and landing configuration was investigated. For a coupled computation several difficulties had to be solved, which are partly caused by the high complexity of the geometry and the resulting computational meshes. Some of the aspects have already been addressed in previous chapters. Figure 22 shows some details of the complexity of the landing configuration including fuselage, belly fairing, wing, slat, flap, flap-track fairings, pylon, engine, nacelle strake, slat tracks and pressure tube bundles. The hybrid mesh contains in total 14 million nodes. The starting configuration is slightly simpler (fuselage, belly fairing, wing, slat, flap, flap-track fairings). The mesh contains 7 million nodes.

Structural FEM models were created by the company Leichtwerk28 with ANSYS. Figure 19 shows the surface mesh of the model created for the landing configuration. Mainly quadratic volume elements were used (SOLID95) for meshing. Only for the tracks another element type was selected (SHELL91). Both models have about 200000 degrees of freedom.

For validation and calibration of the FEM models designed a static deformation test was performed29. The slat and flaps were deflected according to the landing settings. The lower side of the wing was equipped with 45 markers suited for deformation detection via a stereo camera system. The wing was loaded with different weights imposed at one single point close to the wing tip (see Figure 20 left). The loads range from 3.5 kg up to 83.5 kg. The right picture of Figure 20 shows a comparison of the measured deflections normal to the wing plane for all 45 markers and the deflections computed with ANSYS for two different loadings (33.5 and 83.5 kg). The agreement of measured and computed data is good.

Figure 20. Left: position of the markers and the point where the force is applied within the static deformation test; right: Measured and computed deflections for the marker positions 1-45 for a load of 83.5 kg.

Figure 19. Surface mesh of the ANSYS FEM model of the DLR F11 landing configuration.


14

The table in Figure 21 gives an overview of the computations done so far. To check the influence of the deformations, reference computations without coupling were performed first. They were used as input for the coupled computations.

DLR F11 high-lift configuration

Angle of attack

No coupling: α=7°,

11.9°, 18.5°, 20°

Starting configuration

M=0.176, R =15x106,

p∞=2.95bar With coupling: α=7°,

11.9°, 18.5°

No coupling: α=16.5° Landing configuration

M=0.203, Re=15x106,

p∞=2.59bar With coupling: α=16.5°

Figure 21. Overview of computations with and without coupling for DLR F11 starting and landing configuration.

Figure 22. Details of 3D DLR-F11 landing configuration including fuselage, belly fairing, wing, slat, flap, flap-track fairings, pylon, engine, nacelle strake, slat tracks and pressure tube bundles.

Figure 23 shows the undeformed shape of the wing close to the tip region in comparison with the shape in aeroelastic equilibrium for three angles of attack (starting configuration). The maximum deflection in z direction ranges from about 22 mm for α=7° up to 32 mm for α=18.5. Putting these values into relation to the wing half span (1400 mm), we end up with maximum deflection of 1.6% for α=7° and 2.3% for α=18.5°. For the landing configuration, see Figure 24 (α=16.5°), 34.6 mm (2.5%) was computed as the maximum deflection.

Figure 23. Comparison of undeformed shape and shape in aeroelastic equilibrium, DLR F11 starting configuration.

Figure 24. Comparison of undeformed shape and shape in aeroelastic equilibrium, DLR F11 landing configuration.

Figure 25 shows isolines of the deformations normal to the wing plane. For both the starting and landing configuration a small gap change for the slat and the flap is visible. Although the maximum deformation of the wing


15

is small even close to the wing tip, the change of the pressure coefficient distribution is visible, see Figure 26 for starting and Figure 27 for landing configuration.

Figure 25. Left: computed deformations for starting configurations for the three angles of attack; right: computed deformations for landing configuration for α=170.

Figure 26. Comparison of surface distributions for flexible (blue) and rigid (red) configuration for DLR F11 starting configuration close to the tip.

Figure 27. Comparison of surface distributions for flexible (blue) and rigid (red) configuration for DLR F11 landing configuration close to the tip.

In Figure 28 the lift over α curve is plotted with and without coupling in comparison to experimental data available from EUROLIFT I. As could be expected, with coupling the lift is decreased slightly, because of the reduction of the wing’s twist caused by bending. All in all it can be summarized that aeroelastic effects are visible for current high lift applications, but they seem to play only a secondary role. Within the EUROLIFT II project experiments were made for 70% higher Reynolds-number. Here a larger impact of aeroelastic effects is expected. Corresponding computations are underway.


16

Within the DLR internal project FORMEX measurements of the deformations of the EUROLIFT landing configuration were made during the wind tunnel tests in the DNW-NWB in Braunschweig. Compared to the EUROLIFT ETW tests the same Mach number (M=0.2) was used but the Reynolds number (Re=1.57x106) was one order of magnitude lower. The static pressure is 1bar compared to 2.59 bar in ETW. Although the static pressure is relatively small, visible deformations up to 15 mm were found. The left picture of Figure 29 shows the comparison of measured and computed deflections normal to the wing plane at the local quarter point for an angle of attack of 11°. The corresponding twist distributions are found in the right hand part of Figure 29. The agreement of the bending as well as the twist is reasonably good. For the comparison of the bending distributions the numerical results have been corrected, because there is a constant shift between numerical and experimental data due to the mounting of the wind tunnel model.

Figure 29. Comparison of measured and computed bending at C/4 (left) and twist (right) for α=110 for DLR F11 landing configuration.

D. Winglet design Increasing fuel prices and environmental issues are key drivers for airliners and the aerospace industry to

consider measures which reduce the fuel consumption and pollution of their aircraft. For existing civil aircraft the use of a retrofit winglet constitutes one of the measures most often used to improve fuel consumption. The design of a retrofit winglet continues to be a good example of a challenging multidisciplinary problem. Under consideration of structural constraints, an appropriate design strategy combined with accurate CFD prediction tools are required to provide the best aerodynamic benefit for installing a winglet in an existing aircraft.

At DLR retrofit winglets were designed for a generic twin engine transonic transport aircraft30,31. The planform and aerodynamic load of the winglets were obtained by optimizing L/D for the 1g cruise condition (M=0.78, h=36000 ft, 0.5<CL<0.55) using a lifting line method in combination with the RANS solver TAU. Airfoil design was used to reduce wave drag. Typical constraints of a retrofit design were imposed. These are a maximum wing root bending moment increase and a maximum wing span increase based on the reference wing featuring a Küchemann-tip. Furthermore, the wing was allowed to be modified only outboard of a specified station. To have a realistic assessment of the designed winglets, coupled fluid-structure simulation were performed. Using the procedure described above the performance and structural loading (fatigue) at cruise as well as ultimate structure

Figure 28. Lift versus angle of attack for DLR F11 starting configuration.


17

loadings (i.e. gusts, pull out of a dive) for deformed shapes were computed. The used structure models are finite element models (FEM) created for the reference configuration with the different wing tip devices using the DLR PARA_MAM-Code32. A short description of the tool is given in chapter V. The whole primary structure was modelled using eight node shell elements whereby the stringers are considered implicitly as stiffness equivalent layer of the wing’s skin. The elastic properties of average aircraft aluminium were implemented. Coupled fluid-structure RANS solutions were obtained for the cruise condition and the n=2.5g load case. The n=2.5g load case (M=0.754, h=22500ft, CL=0.749) simulates a pull out of a dive, therefore for the structural analysis aerodynamic, weight and inertial (centrifugal) forces were considered.

Three winglet tip device configurations were analysed with the coupled TAU-ANSYS approach (see Figure 30). The devices 1 and 2 are two of the promising designed retrofit winglets whereas device 3 is a classical winglet of large size. Figure 31 shows the front view of the deformed shapes with wing tip device 1 for the 1g cruise condition and for the n=2.5g load case computed with and without considering the inertial forces.

Figure 30. Wing tip devices used for the fluid-structure analysis, source30.

Figure 31. Deformed shapes for the wing tip device 1, source30.

The influence of deformation is shown in Figure 32, which includes a comparison of pressure distributions and skin friction lines for the wing tip device 1 case, obtained either using the deformed shape or the 1g rigid shape of the reference configuration. The impact of deformation is very large for the comparison at the same incidence, the results showing a different streamline pattern. The comparison at the same lift (here the n=2.5g lift value) does not show such large differences, however the results indicate a transfer of load from outboard sections to inboard sections. This results in a weaker shock at the outboard sections.

Figure 32. Skin friction lines and pressure distribution contours for upper wing with the wing tip device 1 geometry: RANS solutions for deformed and rigid shapes for M=0.754, h=22.500 ft, source30.

Figure 33 shows a comparison of pressure contours for deformed and rigid shapes for the wing tip device 3. For the n=2.5g load case the rigid shape shows a shock along the complete winglet span, whereas in the deformed shape this shock is reduced to a small region located at the junction between winglet and wing.

1g shape def shape

same lift: CL=0.749 at n=2.5g same incidence: α=4.38o

1g shape def shape


18

Figure 33. Skin friction lines and pressure distribution contours for wing tip device 3: RANS solutions for rigid (left) and deformed (right) shapes for cruise and n=2.5g load case condition, source30.

The relative decrease in drag as function of the relative increase in wing root bending moment (WRBM) is shown in Figure 34. Results are given for cruise and for the n=2.5g load case and are related to the deformed basis configuration. The deformed shapes lead to a decrease in WRBM and an increase in drag in comparison to the rigid shapes. From the three studied devices, the second one shows the largest aerodynamic benefit with a WRBM increase close to the 4% WRBM structural reserve. In addition, at similar structural load the aerodynamic optimum of this device leads to a larger reduction of aerodynamic drag. The wing tip device configurations with the largest WRBM for the rigid shape show the largest decrease in WBRM for the deformed shapes. Especially for the n=2.5g load case the results for the deformed shape lead to large differences in comparison to the rigid shape results, with a reduction of WRBM up to 5.55%. The work showed that under consideration of structural constraints, an appropriate design strategy combined with accurate CFD prediction based on deformed shapes are required to provide the best aerodynamic benefit when installing a winglet into an existing aircraft.

E. Multidisciplinary Optimization

Based on the experience gained by using the high-fidelity fluid-structure coupled analysis for a variety of applications, the coupled process has been recently included within an optimization process, not only for detailed wing design but also for the pre-design of supersonic business jet.

Detailed Wing Optimizations In the frame of the German cooperative CFD effort MEGADESIGN33,34 detailed shape optimization of a transport aircraft wing including static deformation was carried out35. In a first approach, the aeroelastic effects are simulated by coupling TAU-Code and a beam stick model developed at the Technical University of Aachen16.

Δ Cwrbm/ Cwrbm_REF

ΔC

D/C

D_R

EF

[%]

0 0.02 0.04 0.06 0.08

rigid

deformedrigid

cruise

REF= deformed Basic atcorresponding flight condition

deformed

n=2.5g

Device 3

Device 1Device 2rig. filled

def. non filled

2%

Figure 34. Drag reduction as function of wing root bending moment increase for rigid and deformed shapes, source30.

deformed shape rigid shape

n=2.5g cruise

cruise

n=2.5g


19

The process chain used is depicted in Figure 35. The parametrization of the wing is based on a parametric CATIA V5 model. The design parameters selected for the study are thickness, camber and twist in four predefined wing sections, while the wing planform is held fixed. The resulting 12 shape parameters are accessible by the optimizer via the CATIA-Design Table. Additionally, two structure design parameters control the relative thickness change of the wing front and rear spars in combination with the upper and lower metal sheet thicknesses of the wing box (see Figure 36). These structure parameters determine the stiffness and the weight of the wing. In order to account for any change of the wing box geometry, an equivalent beam stick model is automatically generated16. From the CFD side, the variation of the geometry is propagated to the mesh using a deformation procedure. Iterative coupling between aerodynamic forces and weight forces (weight of wing box, fuel, engines, payload, and engine thrust forces are taken into account) and equivalent beam stick bending and twisting is done until a steady state solution is obtained. In general, 12 coupling iterations are carried out using the Volume-Spline technique described above to deform the CFD mesh according to the resulting equivalent beam stick deformation. On top of this chain, a gradient free Downhill Simplex method is coupled for driving the design process. This optimiser is very robust and performs well in case objective function evaluations are subject to random noise.

Wing Box

Front Spar Rear Spar

Wing Box

Front Spar Rear Spar

Figure 36. Wing box structure parameters. Figure 37. Objective function reduction (total drag force) during optimization.

The objective is to minimize the thrust required in horizontal cruise flight that is equivalent to the product of total aircraft weight and drag over lift ratio. In Figure 37 the reduction of the total drag force (objective function) is shown. Figure 38 displays the decrease of total aircraft mass due to wing box mass optimization while at the same time the aerodynamic performance L/D improved (see Figure 39).

Optimiser

CFD +Mesh Deformation.

CSMBeam Stick Model

Loads

Aerodynamic CoefficientsLift, Drag, Moments, etc.

Max Deformation / Stresses(Or max. Stress / Section)

Beam StickGenerator

New Beam Model

New Geometry

Mesh Generation or Deformation

Iterative Calculation of Aero-elastic Steady State Solution

StructureWeights

Deformation

Optimiser

CFD +Mesh Deformation.

CSMBeam Stick Model

Loads

Aerodynamic CoefficientsLift, Drag, Moments, etc.

Max Deformation / Stresses(Or max. Stress / Section)

Beam StickGenerator

New Beam Model

New Geometry

Mesh Generation or Deformation

Iterative Calculation of Aero-elastic Steady State Solution

StructureWeights

Deformation

Figure 35. MDO process chain.


20

Figure 38. Aircraft mass reduction during optimization.

Figure 39. Aerodynamic improvement (lift over drag ratio) during optimization.

In the previous exercise, while the number of design parameters is relatively limited - in particular for the structural design - the design process still requires more than 350 optimization steps to converge. In order to reduce computational costs for larger number of design variables a gradient-based optimization strategy based on efficient evaluation of the sensitivities is preferred. For this purpose, a coupled adjoint aero-structure formulation was developed. Following the approach of Alonso et al36,37 the continuous adjoint formulation implemented in the flow solver FLOWer was coupled with the structure solver MSC NASTRAN38. Exploiting the symmetry of the structural stiffness matrix, only the right-hand side needs to be modified for the adjoint formulation of the structural equations. A first application of the coupled adjoint aero-structure approach was the shape optimization of a transport aircraft wing in inviscid flow taking into account the sensitivities of the static deformation. The Breguet range function was optimized with constraints on lift and angle of attack using 240 design parameters. Details are given in 38. Currently the approach is extended to viscous flows using the discrete adjoint formulation of the hybrid RANS solver TAU39,40.

Pre-design of Supersonic aircraft configuration The society in the “after Concorde” age is facing constant requests for increased mobility but at the same time

bears fundamental debates whether a high-speed transport successor could be operated in an environmentally acceptable way and whether it could be economic. These environmental constraints and the need to obtain a superior economic business case ask for a lightweight and relatively small configuration for the envisaged transport task. Because of these challenges, a new design will need to be highly optimized regarding all main disciplines involved. Thanks to the robustness and efficiency of CFD and CSM codes, it is now possible to include them within a pre-designed environment and to capture more accurately aerodynamics, structure and aeroelasticity effects.

The DLR has accumulated significant experience within various projects41,42 and has in hand a multidisciplinary analysis suite dedicated for supersonic aircraft applications. This procedure, originally developed by NLR43,44 and adapted to the DLR tools, is relying on high-level fidelity for aerodynamics and structure (in supersonic cruise) and on statistic based tools to cover other disciplines (flight mechanic and propulsion) and remaining parts of the mission. This mixed level fidelity multidisciplinary suite is then coupled to an outer optimizer that drives the complete design process according to a given figure of merit. Here only the fluid/structure analysis module used at DLR is described. Technical details of the complete analysis tools as well as their implementation can be found in references43,44.

The fluid/structure analysis module accounts for the aeroelastic wing deformation (bending and torsion) and for sizing the thicknesses of the wing primary structural elements under a given representative load case. The driving scenario retained is a +2.5g pull-up maneuver at maximum take-off weight, dive Mach number / dive altitude. The aircraft loading is configured such that the wing structure experiences maximum bending moments, i.e. maximum payload and maximum fuel in fuselage/inner-wing tanks.

The fluid/structure analysis module consists of the flow solver FLOWer and the finite-element code NASTRAN13 for the structural analysis. A reasonable level of convergence is obtained after five coupling steps. The aerodynamic loads generation is based on an Euler flow solution for the prescribed maneuver lift coefficient. To reduce computing time, this may be done at a relative coarse grid level, as accurate resolution of flow details is not


21

of direct interest. Non-structural mass items (i.e. landing gear, engines, LE-/TE-movables, servo systems, etc.) and fuel masses are identified and connected as discrete mass items to the nearest structural grid points.

During the aeroelastic wing deformation, all the wing structural elements - covers, spars, ribs and stringers - grouped into design areas serve as input for the structural sizing. The objective of this sizing is to obtain the minimal structural weight at a maximum allowable von Mises stress levels and the resulting element thicknesses have to be greater than 2 mm. This inner optimization is performed using NASTRAN’s native gradient-based optimizer and for a total number of 200 design areas, about 5 to 15 iteration loops are necessary to converge. An example of the internal structure elements is given in Figure 40.

The multidisciplinary analysis suite is coupled with an outer optimizer for planform optimization. The resulting multidisciplinary optimization process, sketched in Figure 41, has been successfully applied for the design of small-scale supersonic cruise configurations in the frame of the EU project HISAC42. Here the capability is applied to the range maximization of a low-boom concept based aircraft cruising at M=1.8 (see Figure 40). This aircraft features two engines mounted aside on top of the rear part of the fuselage and is characterised by the significant dihedral of its double delta wing. The basic set of design variables selected for the outer optimizer allows changing the planform of the wing, as presented on the top of Figure 40. However, the fuselage is tailored so tight around the wing root, that varying this part is literally impossible. Thus the wing body intersection is kept unchanged. Additionally, the thickness and the twist at the crank and wing tip are variables. In total, the outer optimizer, a gradient free optimization strategy, manages 11 design variables.

Figure 40. Design variables of a generic wing geometry (top) and internal structural elements (spars and ribs in red), source42.

Figure 41. Optimization chain with emphasis on fluid/structuring coupling

The history of this optimization is shown in Figure 42. Here, 300 iterations are performed which implies the same number of calls for the coupled fluid/structure analysis module. The mission range increases by 5.8% compared to the baseline configuration.

The character of the wing deformation of the datum wing and optimized wing is shown in Figure 43. To give a better illustration of these deflections, a view from behind and above the configuration is chosen. Furthermore the jig-shapes (grey, note the difference in planform) are also shown. Due to the selected 3D-view in this figure, it is reminded that the baseline wing features significant dihedral (Figure 43). In order to quantify the influence of the aeroelastic coupling, the wing group weights during the first (i.e. no coupling), the 4th and the last step of the aeroelastic loop (W1, W4 & W5) are plotted at each step of the outer optimizer, see Figure 42. By taking into account the aeroelastic effects, the low-level structural optimizer is able to reduce the wing weight of about 600 kg more at each top-level optimization step. This may suggest that taking into account aeroelastic effects allows the optimizers to find a better aircraft. However, the computed maximal wing tip deflection increases recognizable from 2 m (datum) to 2.42 m and may suggest that the elastic wing may suffer from flutter. As no license is available within the implemented MDO process to provide a NASTRAN based flutter check,


22

the obvious way around is to impose a maximal deflection constraint for the structural wing optimization. This forces the NASTRAN optimizer to produce a stiffer wing being less susceptible to flutter.

Figure 42. Mission range, wing weights and L/D development, optimization considering aeroelastic deflections during maneuver, source42.

Figure 43. Wing bending under 2.5g maneuver load; left: datum concept; right: optimized concept, source42.

A third optimization has thus been performed with an additional constraint on the maximum wing tip deviation. After the optimization, it turns out that the range increase is lower than for the other two optimizations and is the lowest value obtained (2.9% increase). The aerodynamic improvement is even lower, being only 2.6%. Constraining the aero-elastic deflections to the initial (datum wing) value, results in an optimized wing coming close to the datum planform. In conclusion, the impact of the aero-elasticity effects has been checked for the design of a supersonic small-scale a/c and plays an important role to perform a realistic design.

IV. Unsteady Fluid/Structure Interactions In this chapter some examples for predicting unsteady fluid/structure interactions are highlighted.

A. Rocket nozzle Concerning the requirement of future rocket technologies providing cost-efficient access to orbit, deeper insight into the unsteady phenomena during the start phase of modern launchers is essential. Especially unsteady interactions and resonances of flow separation inside the nozzle structure will play an important role for the design of future main stage propulsion systems. The so called buffeting coupling is one of the most challenging issues during ascent. At DLR a coupled fluid/structure simulation of the Ariane 5 after body with a realistic Vulcain-2 nozzle at transonic conditions were carried out using the hybrid flow solver TAU and the structural analysis code ANSYS45. The complete launcher was computed. In the nozzle region the computational grid was designed to allow detached eddy simulations (DES). Outside this area a coarser grid resolution sufficient for quasi-steady turbulence modeling was chosen, in order to make unsteady coupled simulation affordable. The grid shown in Figure 44 consists of about 6 million nodes and contains the whole launcher with lots of details. The helium tank aside of the central nozzle is modified to provide better grid resolution in the boundary layer by simplifying geometrical constraints.


23

Figure 44. Symmetry plane of the Ariane-5 grid and cut outs of the nozzle section, source45.

The structural behavior of the Vulcain-2 nozzle was predicted with the commercial software ANSYS. A simplified model of the nozzle was used containing tube wall, turbine exhaust gases and outskirt (see Figure 45). The elements used in ANSYS are shell elements for the structure and outskirt stiffeners and beam elements for the tube wall stiffeners. The complete model had 8640 nodes and 9064 elements. The mesh exchanged with the flow solver contained only the shell elements defining the geometry. The stiffeners were modeled in the structural part only. For the simulation wind tunnel conditions of M=0.8 and Re=25x106 related to the launcher length are chosen. All computations were carried out under the assumption of perfect gas, also inside the nozzle. At the nozzle inflow planes the total pressure of 11.7MPa at a total temperature of 400K was specified. As a first step towards a coupled transient fluid/structure simulation a DES was started without the consideration of the structural behavior of the nozzle. The computation ran with a physical time step of 4ms for about 7s. Pressure distribution and instantaneous stream traces of the DES taken from a snapshot after 7s are shown in Figure 46 for three different views of the launcher. The complex flow structure in the nozzle area is clearly visible. The difference of the pressure distribution on the central nozzle for the 900- and 2700-view is significant and obviously influenced by the position of the tubing connection box which is the only non symmetric feature of the configuration in the corresponding xy-plane. The difference in the position of the pressure maximum at the lower part of the Vulcain-2 nozzle for 900 and 2700 is a direct result of the displacement of the wake vortex separating at the edge of the central stage.

Figure 45. ANSYS structural grid of Vulcain-2 nozzle and simplified refined structural grid for coupled fluid-structure simulation, source45.

Tube wall

Turbine Exhaust Gases

Outskirt


24

Figure 46. Snapshot of the ARIANE-5 Cp distribution and instantaneous stream traces, three different views, source45.

Another important feature of the investigated configuration is the asymmetry of the launcher generated by the helium-shell which is obviously of significant size. The numerical results show strong differences between the unsteady flow of the left and right side of the cut-plane between boosters and central nozzle (Figure 46). While in the part without sphere the unsteady flow from the gap between booster and central stage can directly hit the nozzle where a much more refined vortex system develops, this mechanism is blocked by the shell on the opposite side. DES predictions were compared with experimental data in former investigations with a Volvo-S6 nozzle46. Figure 47 shows comparisons for the averaged pressure coefficient distribution and the distribution of standard derivation of the pressure coefficient along circumferential rings around the external shape of the nozzle. The numerical results for the averaged pressure coefficient are in good agreement with the measurements although the numerical predictions are slightly lower. The standard deviation of the pressure is predicted quite well. All maxima and minima are located at the correct positions. The computation for the unsteady fluid/structure analysis of the Vulcain-2 nozzle was performed over several periods of the structural oscillations using a strong coupling strategy. The simulation ran for about 180ms with a physical time step of 2ms. Each coupling step required about 3 hours on 128 Blue-Gene processors. The shapes and the surface deformations taken over a period of 6ms are shown in Figure 48, seen from side and bottom. As visible mainly an ovalization mode of the Vulcain-2 nozzle is superimposed by other modes, resulting in a rotation of the minima and maxima of the deformation. The position of the minimum in the nozzle radius is marked in the lower figures by arrows. The counterclockwise rotation of this position is obvious.

Figure 47. Averaged pressure distribution and standard deviation along circum-ferential rings around the Volvo nozzle, source46.


25

Figure 48. Deformation of Vulcain-2 nozzle at four different times, exaggerated view by factor of 30, source45.

In Figure 49 the displacement components of a single point perpendicular to the flow direction are presented. The deformation of the Vulcain-2 nozzle with an outlet diameter of 2m is in the order of 3mm in amplitude. The resonance frequency of the structure is obviously in the range of 30Hz.

B. Unsteady aircraft maneuver Within the DLR internal project SikMa (Simulation of Complex Maneuvers”) a multidisciplinary simulation environment was developed to predict the unsteady aerodynamic behavior of a free flying elastic combat aircraft47. The numerical tool combines time-accurate aerodynamic, aeroelastic and flight mechanics calculations. The target application is a maneuvering aircraft with its movable control devices.

For the numerical simulation of the flight mechanics, the simulation environment SIMULA developed at the DLR Institute of Flight Systems is currently used48. SIMULA provides the three basic functionalities necessary for flight simulation and flight control purposes: trimming, i.e. the determination of the initial state and control values, linearization and stability analysis, and simulation, i.e. the numerical integration of the equations of motion. Single and multi-body flight mechanics models, ranging from 1 to 6 degrees of freedom, are made available to the simulation by SIMULA. The amount of data that is exchanged between SIMULA and TAU is of a scale that can be easily communicated directly through a TCP/IP socket connection.

Computations were carried out for X-31 military configurations. Figure 50 illustrates the complexity of the vortex flow topology over the clean wing and fuselage at an angle of attack α= 180. A qualitative comparison between measured and computed surface pressure distributions over the X-31 wing including control devices at α= 160 and Re=2.07x106 is shown in Figure 51. The RANS computations were performed with the DLR TAU-Code using the Spalart-Allmaras turbulence model with vortex correction. The comparison indicates that the main footprints of the vortices as well as their location are predicted quite well. Due to the gaps between the leading edge control devices three distinct vortices can be identified. The first vortex comes from the inner wing, the second and

Figure 49. Displacement of a single point on the Vulcain-2 nozzle over simulation time, source45.


26

third from the inner and outer leading edge flap, respectively. The suction-strength of the vortex at the inner leading edge flap is predicted to be stronger and the outer vortex starting at the kink of the wing is predicted to be weaker than in the experiment.

Figure 50. 3D flow field over the X-31 configuration (clean wing) at 18º angle-of-attack, source47.

Figure 51. Left: steady PSP measurement of the pressure distribution over the X-31 wing; right: Steady TAU RANS calculation of the pressure distribution over the X-31 wing, source47.

In order to demonstrate the implemented multidisciplinary simulation capability, first computations for a free-to-roll maneuver of the elastic X-31 configuration around the longitudinal axis were performed. A generic structural model containing the full aircraft wind tunnel configuration with fuselage, wings, and control surfaces was created. The finite element model has 273 nodes with 819 discrete translational degrees of freedom and is depicted in Figure 52. Euler computations were performed on a mesh with 19.4 million tetrahedrons, 3.44 million nodes and 261,792 nodes on the surface. To simulate the free-to-roll maneuver around the longitudinal axis a simplified one degree of freedom flight mechanic model was used. This means that the acceleration of the model depends only on the aerodynamic rolling moment and the moment of inertia around the x-axis of the finite element model.

The simulated maneuver starts from a roll angle of 45° and a pitch angle of 15°. A converged, steady-state solution of the aerodynamic forces and moments was obtained for the undeformed mesh. During the maneuver unsteady non-symmetric loads on the aircraft lead to non-symmetric deformation of the structure. The elastic model is depicted in Figure 52. For a FE model with generic stiffness the deflected finite element model and the corresponding transformed aerodynamic forces are shown.

Figure 52. Structural deformation of the FE-Model and corresponding forces on the aircraft. One time step within an unsteady maneuver simulation, source47.

Figure 53. Corresponding rigid and elastic position of the X-31 configuration during a free-to-roll maneuver around the longitudinal axis, source47.


27

Figure 53 compares the rolling angle for the rigid and elastic model during a real time of 0.5s. It can be seen that for the simulation with the elastic model the rolling angle approaches zero somewhat faster. Due to the one degree of freedom flight mechanic model the acceleration of the rolling angle is only dependent on the rolling moment. The deflection of the elastic model due to the aerodynamic forces induces a higher angle of attack and leads to a higher rolling moment and a higher rolling frequency. This in turn leads to a higher aerodynamic damping of the system, which has the effect of bringing the system to a trimmed position within a shorter period of time than is observed with the rigid model. The comparison of the corresponding pressure distributions for the rigid and the elastic CFD model is shown in Figure 54. The pressure distributions differ between the models. Since the loads on the configuration are asymmetric during the maneuver, the displacements due to the transformed structural loads on the left and right hand side differ from each other. In Figure 55 the difference of the aerodynamic surface between the rigid and the full elastic model during the maneuver simulation is shown. The maximum displacement at the wing tip is approximately 3.8cm for a half span of the wind tunnel model of 50cm. Having demonstrated the functionality of the developed multidisciplinary environment future simulations will concentrate on RANS calculations taking into account the deformation of the control device.

Figure 54. Pressure distribution for the rigid and the elastic CFD-model of the X-31 configuration. Rolling angle at approximately Φ=-20° (at t=0.2s in Fig. 48), source47.

Figure 55. Deformation of elastic CFD-Surface compared to rigid CFD-Surface during the free-to-roll maneuver, source47.

V. Outlook

A. Parametric structure model For future multidisciplinary optimization an automatic and parametric generation of finite element models in a

detailed shell representation with realistic weight optimization of the structure is required. Beam models currently often used have been proved to represent the global structural stiffness in aeroelastic calculations appropriately resulting in realistic global deflections. Besides the wingbox geometry the skin thickness distribution is required for beam models which results from structural sizing. Beam idealizations base on kinematical assumptions on the correlation between global deformations and in-plane strain of the surface. Local stress is derived from local strain. Hence, the primary data for sizing is not calculated explicitly but is only derived from predicted kinematics. Especially if composite material with its anisotropic strength is of concern a more detailed structural analysis is mandatory. Finite shell elements permit a more physical calculation of in-plane strain distributions with a computational effort which is suitable also for early design phases. The challenge with shell models is the necessity of explicit geometric modeling, which can be done automatically e.g. using the DLR model generator PARA_MAM (Parametric, Simple and Fast Mesh Based Aircraft Modelling Tool) 32.

PARA_MAM is a compact and comprehensive MATLAB code which generates input decks for FEM pre-processors such as the ANSYS prep7 including all commands for geometric modeling and meshing. Since the program directly calculates the coordinates of all geometric keypoints, which are constituted by the rib-spar intersections, the required pre-processor operations consist of elementary commands only. This avoids slow and error prone CAD operations and in particular Boolean operations. Furthermore, interfaces to alternate pre-processors

Blue – rigid Red – elastic


28

can be realized easily. The geometric input for the FEM preprocessor is created in such a way, that the ribs and spars (see Figure 56 right) fit exactly into the aerodynamic shape. The input needed by PARA_MAM is the aerodynamic surface of the wing. Therefore either an unstructured CFD surface mesh like in Figure 56 left or a structured CFD surface mesh can be used. The definition of the inner structure is fully parametric and covers the spectrum from simple pre-design modeling, which can be realized with low effort, up to medium complex models for the representation of realistic rib and spar arrangements. Stringers can either be modeled as ‘smeared’ stiffness of the multi-layered skin or explicitly as shell elements. PARA_MAM was used for example in the frame of the design of transport aircraft wing-tip device for aeroelastic analysis (see section III-D).

Figure 56. Unstructured CFD surface mesh and resulting inner structure created by PARA_MAM For structural sizing the code S_BOT (Sizing Robot) was developed at DLR. S_BOT is not a capsulated programme but a modular and open framework for automated FEM analysis and sizing realised as set of APDL macros (ANSYS Parametric Design Language). As input, a FEM model and sets of external loads has to be provided which remain constant during the sizing loop. The process takes benefits from PARA_MAM generated models but does not rely on them. One common type of external loads is the aerodynamic pressure distribution in the same format as used within the presented aeroelastic coupling chain.

B. High-fidelity coupled aero-structural design optimization The major challenge of aero-structural optimization is to optimize the structure for the relevant load cases, followed by a static aeroelastic equilibrium computation at cruise conditions to correctly predict the aerodynamic performance in an acceptable turn-around time. For instance, the sizing of the structure has to be performed not only under a +2.5g pull-up maneuver, as presented in section II-E, but also for other critical load cases, like gusts, fatigue, flutter, touch down. Similarly, accurate aerodynamic performance predictions at cruise require the coupling of a high-fidelity viscous flow solver, a structural analysis code and methods for trimming the aircraft. All these coupled computations imply large turn around times and innovative strategies have to be developed to cut down computational costs while keeping accuracy at an acceptable level. Thus DLR will improve the capabilities of its high-fidelity tools for detailed aero-structural design within the project MDOrmec (MDO of a rear-mounted engine configuration), which is set-up in

Figure 57. Aero-structural design based on high-fidelity methods.


29

close cooperation with ONERA. The primary purpose of this project is to establish an accurate fluid/structure optimization chain for the flexible aircraft with dedicated key technologies. These technologies cover CAD parametrization, efficient mesh generation processes, effective coupling strategies, MDO architectures and optimization techniques. In a second step, the trade-off between accuracy and efficiency of the MDO process will be investigated in detail by testing different fidelity levels of flow solvers (Euler, Euler+boundary layer, Navier-Stokes) and structural analysis methods (FEM, beam, simple models).

C. Process chain The strategies for the process chains described in section II-E have their specific advantages but there are also

common disadvantages. The process chains rely on the combination of a relatively heterogeneous set of tools, which can complicate the porting of the process chain to other platforms. The substitution of components and the support of end-users may also become problematic. For example, the flight mechanics code SIMULA as well as the structured CFD code FLOWer are written in FORTRAN 77, whereas TAU is based on the programming language C. The discrete CSM code has been developed in MATLAB. To make the code more platform independent the MATLAB code can be exported to C or C++, but even so, a platform specific library must be available for the target platform.

Currently, a harmonization of the process chains is under way. A very flexible Python interface is already available for TAU. The interface enables the direct control of the physical time loop of TAU. Data required by external modules, for example global forces and moments for flight mechanics, can be obtained directly through the interface, and in the same way data like the speed and rotation of the aircraft can be sent to TAU through the interface. A coupling manager, which is also a Python script, is used to control the process chain. This coupling manager can either be internal or external to the Python script which controls the TAU simulation loop.

Relatively simple tools like the flight mechanics and the trim modules can be coded directly in Python. The same holds for the discrete structure solver, which is currently coded in MATLAB. The consequent use of Python as the coding language for simple tools and as the interface for complex tools like TAU makes the test and evaluation of multidisciplinary coupling strategies much easier and more efficient. The portability of the code to other platforms is also simplified. The components of the process chain which are directly coded in Python will become part of the TAU suite. This will definitely make the maintenance of the code and the support of end users much easier.

VI. Conclusion The paper presents recent DLR activities to improve the prediction quality of aerodynamic aircraft performance by taking into account aeroelastic effects. A flexible process chain for steady and unsteady coupled fluid/structure simulations was set up, allowing the coupling between the in-house CFD codes TAU and FLOWer and the commercial FEM codes ANSYS and NASTARAN. Different interpolation techniques for transferring aerodynamic loads and surface deformation between CFD and CSM meshes were investigated. A robust volume mesh deformation tool was developed which can be applied for either block-structured or hybrid, unstructured meshes. The fully automatic process chain can run on a single workstation as well as on a heterogeneous network including multiprocessor batch computers for the time consuming CFD computation. The coupled fluid/structure simulation capability was successfully applied to a variety of steady and unsteady problems. First results for aerodynamic design and optimization based on coupled fluid/structure simulations are presented. A fluid mechanics solver was included in the numerical environment in order to allow the simulation of maneuvering aircraft. Future developments necessary to enhance the simulation capability for multidisciplinary simulation and optimization are identified.

Acknowledgments The authors would like to thank their colleagues of the DLR Institute of Aerodynamics and Flow Technologies

for providing results for this overview paper, in particular J. Brezillon, G. Einarsson, Th. Gerhold, U. Herrmann, J. Himisch, S. Keye, H. Lüdeke, S. Melber-Wilkending, A. Schütte, Th. Streit. Also thanks to H. Barnewitz from Airbus Germany for the results of the multidisciplinary wing design.

References 1 Rolston, S., Elsholz, E.: “Initial Achievements of the European High Reynolds Number Aerodynamic Research Project

‘HiReTT’”, AIAA paper 2002-424, 2002.


30

2 Braun, C., Boucke, A., Ballmann, J.: “Numerical Study of the Influence of Dynamic Pressure and Deflected Ailerons on the Deformation of a High Speed Wing Model”, The 7th Results and Review Workshop of the HLRS at the High Performance Computing Center Stuttgart, Germany, 2004.

3 Farhat, C., Lesoinne, M.: “Higher-Order Staggered and Subiteration Free Algorithms for Coupled Dynamic Aeroelasticitiy Problems, AIAA 98-0516, 1998.

4 Neumann, J., Arnold, J., Krüger, W. R.: “Numerische Simulation komplexer aeroelastischer Phänomene und Vergleich mit Windkanalexperimenten”, Proceedings of DGLR Annual Conference, Braunschweig, Germany, 2006.

5 Arnold, J., Einarsson, G., Schütte, A.: “Multibody Simulation of an Aeroelastic Delta Wing in Roll Manoeuvres”, Proc ICAS 2006, Hamburg, Germany; 2006.

6 Raddatz, J., Fassbender, J.K.: “Block Structured Navier-Stokes Solver FLOWer, In: Kroll, N., Fassbender, J. (Eds) MEGAFLOW – Numerical Flow Simulation Tool for Transport Aircraft Design”, Notes on Multidisciplinary Design, Vol. 89, Springer, 2005.

7 Kroll, N., Radepiel, R., Rossow, C.C.: “Accurate and Efficient Flow Solvers for 3D Applications on Structured Meshes”, AGARD R-807, 4.1-4.59, 1995.

8 Galle, M.: “Ein Verfahren zur numerischen Simulation kompressibler, reibungsbehafteter Strömungen auf hybriden Netzen”, DLR-FB 99-04, 1999.

9 Gerhold, Th.: “Overview of the Hybrid RANS TAU-Code”, In: Kroll, N., Fassbender, J. (Eds) MEGAFLOW – Numerical Flow Simulation Tool for Transport Aircraft Design, Notes on Multidisciplinary Design, Vol. 89, Springer, 2005.

10 Schwamborn, D., Gerhold, T., Heinrich, R.: “The DLR TAU-Code: Recent Applications in Research and Industry”. In proceedings of “European Conference on Computational Fluid Dynamics“ ECCOMAS CDF 2006, Delft The Netherland, 2006.

11 Kroll, N.; Fassbender, J. K. [Hrsg.]: “MEGAFLOW - Numerical Flow Simulation for Aircraft Design”, Notes on Numerical Fluid Mechanics and Multidisciplinary Design (NNFM), 89, Springer, Closing Presentation DLR Project MEGAFLOW, Braunschweig, Germany, ISBN 3-540-24383-6, 2002.

12 Kroll, N., Rossow, C.C., Schwamborn, D., Becker, K., Heller, G.: “MEGAFLOW - A Numerical Flow Simulation Tool For Transport Aircraft Design”, ICAS Congress 2002, Toronto, Canada, ICAS-Proceedings, pp 1.105.1-1.105.20, 2002.

13 Kroll, N., Rossow, C.C., Schwamborn, D.: “The MEGAFLOW-Project - Numerical Flow Simulation for Aircraft”. In A. DiBucchianico, R.M.M. Mattheij, M.A. Peletier (Edts.), Progress in Industrial Mathematics at ECMI 2004, Springer, New York, S. 3 – 33, 2005.

14 http://www.mscsoftware.com 15 http://www.ansys.com 16 Reimer, L., Braun, C., Ballmann, J.: “Analysis of the static and dynamic aero-structural response of an elastic swept wing

model by direct aeroelastic simulation”, Annual meeting of the International Council of the Aeronautical Sciences (ICAS), Hamburg, Germany, 2006.

17 Beckert, A.:”Coupling fluid (CFD) and structural (FE) models using finite Interpolation elements”, Aerospace, Science and Technology (AST), No.5082, 1/2000, 2000.

18 Beckert, A., Wendland, H.: “Multivariate interpolation for fluid-structure-interaction problems using radial basis functions”, Aerospace, Science and Technology 5 (AST), pp 125-134, 2001.

19 Hounjet, M.H.L. Meijer, J. J.: “Evaluation of Elastomechanical and Aerodynamic Data Transfer Methods for Non-Planar Configurations in Computational Aeroelastic Analysis”, ICAS-publication, pp. 10.1-10.25, 1994.

20 Wendland, H.: “Scattered Data Approximation”, Cambridge Monographs on Applied and Computational Mathamathics, Cambridge University Press, Cambridge, UK, 2005.

21 Gerhold, T.: “Efficient Algorithms for Mesh, Deformation”, Proceedings of the ODAS Symposium Toulouse, 2006. 22 SFB 401, “Strömungsbeeinflussung und Strömungs-Struktur-Wechselwirkung an Tragflügeln”, Arbeits und

Ergebnisbericht, 2002. 23 Britten, G., Braun. C., Hesse, M., Ballmann, J.: “Computational Aeroelasticity with Reduced Structural models”, Notes on

Numerical Fluid Mechanics and Multidisciplinary Design, Volume 84: Flow Modulation and Fluid-Structure Interaction at Airplane Wings, (Ed. Ballmann, J.), pp. 275-299, 2003.

24 http://aaac.larc.nasa.gov/tsab/cfdlarc/aiaa-dpw/ 25 Keye, S, Sitzmann, M., Brodersen, O., Heinrich, R.: “Fluid-Structure Coupled Loads Analysis of DLR’s F6 Wing-Body

Configuration”, AIAA-2008-0675, 2008. 26 Rudnik, R: “Towards CFD Validation for 3D High Lift Flows – EUROLIFT”, In: Book of Abstracts - ECCOMAS CFD

2001, S. 70 -, ECCOMAS Computational Fluid Dynamics Conference 2001, Swansea, Great Britain, ISBN 0905091116, 2001. 27 Rudnik, R., v. Geyr, H.: “The European High-Lift Program EUROLIFT II”, Aeronautics Days 2006, Vienna, 19.-21. June

2006. 28 http://www.leichtwerk.de 29 Puffert-Meißner “Static Deformation Test”, EUROLIFT II report, 1995. 30 Streit, Th., Himisch, J., Heinrich, R., Nagel, B.; Horstmann, K.H., Liersch, C.: “ Design of Retrofit Winglet for a Transport

Aircraft with Assessment of Cruise and Ultimate Structural Loads”, STAB Conference, Darmstadt 2006. 31 Himisch, J., Horstmann, K. H., Streit, T., Nagel. B.: “Design of a retrofit winglet for a transport aircraft”, ODAS

symposium Toulouse, 2006. 32 Nagel, B.; Rose, M.; Monner, H.P.; Heinrich, R.: “An Alternative Procedure for FE-Wing Modelling”, DGLR-2006-048,

DGLR Annual Conference, Braunschweig, Germany, 2006.


31

33 Kroll, N., Gauger, N., Brezillon, J., Becker, K., Schulz, V.: “Ongoing Activities in Shape Optimization Within The German Project MEGADESIGN”, in: CD of ECCOMAS 2004, Jyväskylä (FI), July 24-28, 2004.

34 Kroll, N.; Becker, K.; Rieger, H. Tiele, F.: Ongoing Activities in Shape Optimization with the German Project MEGADESIGN, Annual meeting of the International Council of the Aeronautical Sciences, ICAS-2006-3.11.1, Hamburg, Germany, 2006.

35 Barnewitz, H., “Flexible Wing Optimization Based on Shapes and Structures”, Final review meeting of the German MEGADESIGN project, Braunschweig, Germany, May 2007.

36 Reuther, J. J., Alonso, J.J, Martin, J. R. R. A., Smith, S. C.: “A Coupled Aero-Structural Optimization Method For Complete Aircraft Configurations”, AIAA Paper 99-0187, 1999.

37 Martins, J. R. R. A., Alonso, J. J.: “Aero-structural wing design optimization using high fidelity sensitivity analysis”, CEAS conference on multidisciplinary analysis and optimization, Cologne, Germany June 2001.

38 Fazzolari, A., Gauger, N., Brezillon, J., “ Efficient aerodynamic shape optimization in MDO context”, Journal of Computational and Applied Mathematics, Vol. 203, pp. 548-560, 2007.

39 Dwight, R.P., Brezillon, J. “Effect of Various Approximations of the Discrete Adjoint on Gradient-Based Optimization”, AIAA 2006-690, 44th AIAA Aerospace Sciences Meeting and Exhibit, Reno, Nevada, 2006.

40 Dwight, R.P., Brezillon, J.; Vollmer, D.: Efficient Algorithms for the Adjoint Compressible Navier-Stokes Equations with Applications, ODAS Conference 2006, Toulouse France, 2006.

41 Herrmann, U., Frhr. von Geyr, H., Werner-Westphal, Ch.: Mixed Fidelity Multi Discipline Optimization of a Supersonic Transport Aircraft, AIAA 2005-534, 43rd AIAA Aerospace Sciences Meeting and Exhibit, Reno, Nevada, Jan. 2005.

42 Herrmann, U., Laban, M., “Multi-Disciplinary Analysis and Optimisation Applied to Supersonic Aircraft; Part 2: Application & Results”, AIAA 2007-1857, 3rd AIAA Multidisciplinary Design Optimization Specialist Conference, Honolulu, Hawaii, April 2007.

43 Laban, M., “Multi-Disciplinary Analysis and Optimization of Supersonic Transport Aircraft Wing Planforms”, AIAA 2004-4542, 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Albany, New York, 2004.

44 Laban, M., Herrmann, U., “Multi-Disciplinary Analysis and Optimisation Applied to Supersonic Aircraft; Part 1: Analysis Tools”, AIAA 2007-1857, 3rd AIAA Multidisciplinary Design Optimization Specialist Conference, Honolulu, Hawaii, April 2007.

45 Lüdeke, H., Calvo, J. B., Filimon, A.: “Multidisciplinary Simulation of the Ariane 5 Nozzle Section”, CEAS Congress, Berlin, Germany, 2007.

46 Lüdeke, H., Calvo, J. B., Filimon, A.: “Fluid Structure Interaction at the Ariane-5 Nozzle Section by Advanced Turbulence Models”, Proceedings of the European Conference on Computational Fluid Dynamics, ECCOMAS, Egmand aan zee, The Netherlands, 2006.

47 Schütte, A., Einarsson, G., Raichle, A.¸Schöning, B., Orlt, M., Neumann, J., Arnold, J., Mönnich, W.¸ Forkert, T., “Numerical Simulation of Maneuvering Aircraft by Aerodynamics, Flight Mechanics and Structural Mechanics Coupling”, AIAA 2007-1070, 2007.

48 Mönnich, W.; Buchholz, J. J.: SIMULA - Ein Programmpaket für die Simulation dynamischer Systeme - Dokumentation und Benutzeranleitung - Version 2 DFVLR Institutsbericht IB 111-91/28, 1991.

fluid-structure coupling for aerodynamic analysis and design: a

Documents