MEGADESIGN and MegaOpt - German Initiatives for Aerodynamic Simulation and Optimization in Aircraft Design

We present an overview of the mesh adaptation facility of the DLR TAU code as well as details of improvements made to it in the recent MEGADESIGN project, in particular focusing on advances made in the core of the adaptation module (for example parallel (de-)refinement and other efficiency improvements) as well as a relatively new type of adaptation indicator based on the adjoint solution (among other things) for a goal-oriented mesh adaption.

These improvements to the existing algorithms already available in the TAU code allow us to produce improved computational meshes in a more distributed manner, which provide more accurate predictions for selected functionals of the flow solution such as drag or lift – in fact, for any for which the necessary adjoint solution can be computed.

Conventionally, two types of wall boundary condition are available for the solution of turbulence transport equations in CFD. These exhibit very different requirements on the wall normal distance of the first grid point and any violation of these requirements leads to a drastic degeneration in the solution quality. This places a very high level of importance on the design of the numerical grid, and contributes to the excessive human resources typically spent on this task. Furthermore, these criteria depend strongly on the local flow field quantities, which means that prior knowledge of the solution is required for correct grid design. In practice this often means that an iterative grid design process is required, causing further grid generation expenditure.

A number of possible techniques to accelerate common CFD processes based on the DLR TAU code [1] is addressed, the outcome is briefly discussed and illustrated. As a result of these investigations, considerable increase of computational efficiency has been identified and demonstrated.

Advanced aircraft design is driven by accurate analysis tools which are combined to an optimization process. Modern aerodynamic analysis for complex aircraft configurations are mostly dependent on flowfield simulation results with small uncertainty levels. A software package which more and more fulfills those requirements is the TAU Code developed at the DLR over many years. The aim of this contribution is to demonstrate the ability of the TAU code to compute physically complex flow problems which occur in flow field simulation tasks for combat aircraft at high angle of attack. It is shown that the improvements developed during the Megadesign project have led to efficiency gains beyond a factor of 2 with respect to run time. This progress has contributed to two main milestones of the project and fulfilled the requirements posed.

A universal wall-function method for RANS turbulence modelling is presented which allows for a considerable solver accelaration and reduction of memory consumptions at only a small loss in accuracy even in flows with separation and reattachment. The range of validity of the approach in investigated by considering the near-wall RANS solutions in regions of strong adverse pressure gradient leading to separation and near stagnation points. The method is then applied successfully to aerodynamic flows with separation including transonic flows with shock induced separation and a subsonic highlift airfoil close to stall. From this some first best practice guidelines are suggested.

The present work reports on the further development of the Hanjalić–Jakirlić (1998) near-wall, second-moment closure (SMC) model in the RANS (Reynolds-Averaged Navier–Stokes) framework, updated to account for a wall-normal free, non-linear version of the pressure-strain term model, its implementation into the DLR-FLOWer code and its validation in computing some (compressible) transonic flow configurations. Furthermore, the wall boundary condition is based on the asymptotic behaviour of the Taylor microscale λ and its exact relationship to the dissipation rate ε in the immediate wall vicinity. In addition, the calculations were performed using the DLR-FLOWer’s default Reynolds stress transport model (Eisfeld, 2006), representing a numerically robust combination of the Launder–Reece–Rodi (1975) model resolving the near-wall layer with the Speziale–Sarkar–Gatski (1991) model being employed in the outer region. The flow geometries considered in this work include the transonic RAE 2822 profiles (cases 9 and 10), the ONERA M6 wing and the DLR-ALVAST wing-body configuration. The model results are analysed and discussed in conjunction with available experimental databases and the results of two widely used eddy-viscosity-based models, the one-equation Spalart–Allmaras model (1994) and the two-equation k-ω model of Wilcox (1988). The SMC predictions show encouraging results with respect to the shock position, shock-affected flow structure and the strength of the wing-tip vortex.

A computational method for automatic transition prediction for general three-dimensional configurations is presented. The method consists of a coupled program system including a 3D Navier–Stokes solver, a transition module, a boundary layer code and a stability code. The transition module has been adapted to be used with parallel computation to account for the high computational demand of predicting flows around three-dimensional configurations. A comprehensive investigation on general computational and parallel performance identifies the numerical effort for the transition prediction method. The procedure has been validated comparing numerical results with experiments for the flow around an inclined prolate spheroid. Feasibility studies on generic transport aircraft show the code’s capability to predict transition lines on general complex geometries.

A Reynolds-averaged Navier–Stokes solver, a laminar boundary-layer code and different transition prediction methods for the prediction of Tollmien–Schlichting and cross flow instabilities were coupled for the automatic prediction of laminar-turbulent transition on general three-dimensional aircraft configurations during the ongoing flow computation. In this article, the procedure is applied to a two-dimensional three-element high-lift airfoil configuration which is characterized by the existence of laminar separation bubbles. The automatic transition prediction procedure is applied using different operation modes and different transition prediction strategies.

The MEGADESIGN project was supposed to investigate the status that hybrid CFD prediction has on accuracy, in a relative as well as absolute sense. Different test cases have been analyzed and their results were compared to requirements specified by aerodynamic engineers. Produced solutions stayed significantly below requested error limits, however, the maximum lift area with flow separation still poses a challenging task. Proposals have been elaborated to improve the physical modeling of the tests, which seem to be the major source of remaining errors.

In this part of the MEGADESIGN project, aeroelastic effects are introduced into the aerodynamic analysis of aircrafts by coupling DLR’s flow solvers TAU and FLOWer to a Timoshenko-beam solver. The emerging aeroelastic solvers and a method for the automatic identification of Timoshenko-beam models for wing-box structures were integrated into a simulation environment enabling the combined optimisation of aerodynamic wing shape and structure.

Aeroelastic effects can play a significant role in wind-tunnel testing under high Reynolds number conditions, as shown for example within the European project HIRETT [12]. Due to the high static pressure in the wind tunnel the deformations can reach a magnitude which cannot be neglected, as shown for example in [3]. And also within the project EUROLIFT I a discrepancy has been found between the computed polar and the polar measured in the ETW wind tunnel. The discrepancy could be attributed to either model deformation, a non-uniform onflow due to the presence of wind tunnel walls or the influence of specific geometry installation effects.

In order to provide a more realistic aerodynamic simulation of the flying aircraft, a simplified structural method “WingDACC” was coupled with the DLR TAU code and is applied on simple wing/body as well as complex high-lift cases. The results obtained for the flexible wing are considerably improved when compared with rigid wing computations. Details of the technique and the testcases are addressed and the results are discussed.

Numerical Optimization requires a large amount of intermediate computations for the design data sets suggested by any optimization strategy. The results of these computations are necessary in order to find directions to the optimum. Nevertheless, most results are useless from the quality standpoint of view. Thus, it is desirable to avoid these, which would save a lot of time and money. It will be shown that the application of Artificial Neural Networks can serve in this sense and result in computational savings of about 77%. The problem in this context, i.e. the choice of an appropriate network topology, is discussed and solutions, resulting from extensive numerical investigations, are presented. Finally, the application to a challenging multimodal optimization problem, which serves as a surrogate for multidisciplinary optimization with comparable multimodal solution spaces, demonstrates the power of this approach.

Design optimization is a process aimed at selecting the best design, referred as objective function, satisfying certain requirements, called constraints, by modifying the input variables or design parameters. modeFRONTIER is a software that allows to use data coming from many different sources to evaluate several designs and, through the use of an optimization algorithm, can change the input parameters in order to find a configuration that performs better than the others. The preferred methodology to carry out the optimization is usually the use of numerical analysis, instead of experimental or analytical analysis, because it combines good accuracy of the predicted solution with fewer time and resources needed for the investigation.

The flow simulation tools in aerodynamics have reached a level of sophistication such that realistic flow simulations can be done in a reasonable computational time. Hence, the use of computational fluid dynamics (CFD) in the aerodynamic research as well as in the industry increases.When already using simulation tools, a natural extension is also to apply numerical shape optimization methods in the design process of an aircraft. With gradient free optimization methods, hundreds or even thousands of flow simulations are required during the optimization process. Even though the simulation tools are well developed and fast, this is still numerically very costly.

This chapter addresses the efficient computation of accurate sensitivity information in the aerodynamic design process. Mathematically, this sensitivity information is expressed by a derivative of a function that is defined via the numerical model of the aerodynamic system. This function links a number of independent variables to relevant target quantities such as lift, drag, or pitching moment.

Multi-disciplinary analysis is necessary to reach physically meaningful optimum designs. For aero-structural shape optimization this means coupling two disciplines – aerodynamics and structural mechanics. In this paper the sensitivity evaluation for aerodynamic shape optimization is considered, while taking into account the static aeroelastic effects introduced by the variations in the aerodynamic forces, which are associated with changes in the aerodynamic shape. Due to the high computational cost of a finite difference evaluation step for such a coupled problem, an extension of the adjoint approach to aeroelasticity is necessary for an efficient calculation of the sensitivities. The theory, implementation, validation and application of such a method in the multi-disciplinary design optimization (MDO) context described above, is presented.

Within the next few years, numerical shape optimization based on high fidelity methods is likely to play a strategic role in future aircraft design. In this context, suitable tools have to be developed for solving aerodynamic shape optimization problems, and the adjoint approach – which allows fast and accurate evaluations of the gradients with respect to the design parameters – is seen as a promising strategy. Based on the unstructured RANS solver TAU, a continuous as well as a discrete adjoint have been developed and applied to cruise and high-lift configuration optimization problems. This paper describes investigations of planform optimizations for a flying wing transport aircraft with an Euler continuous adjoint method, the wing design of the DLR-F6 wing-body aircraft configuration and the flap and slat settings of the DLR-F11 high-lift wing-body aircraft with a viscous discrete adjoint method.

Advanced aircraft design is characterized by multipoint, multidisciplinary requirements. Optimization techniques probe the aerodynamic, flight mechanical and structural design sensitivities for a balanced vehicle-system. Aircraft optimization exercises were performed in an universal optimization environment, which controls the CAD, robust mesh generation, RANS-flow simulation and the selection of multidisziplinary variables. Genetic algorithms, evolutionary strategies and simplex were used. The algorithms applied will be compared.

A multi-disciplinary optimisation (MDO) process chain for shape optimisation of a wing including the static deformation has been developed. The objective function, which should be minimized, is equivalent to the total aerodynamic drag force of an aircraft in stationary horizontal flight.

The CATIA V5 parametric model of the wing is controlled by the optimiser using an external CATIA-DesignTable. For four predefined wing sections there are parameters to control the thickness, camber, and twist independently giving in total 12 design parameters for the outer shape while maintaining a fixed wing planform. Additionally, two structure design parameters control the relative thickness change of the wing front and rear spars in combination with the upper and lower sheet thicknesses of the metallic wing box. The stiffness and the weight of the wing depend on these structural parameters.

An equivalent beam stick model is then automatically generated for any change of the wing box geometry. 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. Here, 12 coupling iterations are carried out using a Volume Spline technique to deform the CFD mesh according to the resulting equivalent beam stick deformation. TAU has been used as Navier-Stokes (RANS) solver on a structured mesh.

The optimiser chosen for this task is a Downhill Simplex method, which is very robust, does not require gradients, and performs well even with objective function evaluations that are subject to random noise (non-smooth). The reduction of the total drag force (objective function) and the total aircraft mass decreases due to wing box mass optimisation is shown while at the same time the aerodynamic performance L/D improves.

Multidisciplinary Design Optimization (MDO) is a challenging goal for designers. The aim is to use optimization strategies to integrate a number of different disciplines simultaneously in the optimization process. The optimum of the multidisciplinary problem is better than the design found by optimizing each discipline sequentially, since it can exploit the interactions between the disciplines. However, including all disciplines simultaneously significantly increases the complexity of the problem.

At EADS Military Air Systems (EADS-MAS) the analysis of the multidisciplinary design optimization of an aircraft was carried out. While the whole aircraft has been optimized with regards to the aerodynamic part, structural concerns were taken into account for the wing alone. The whole process is not a real multidisciplinary optimization, because the structural part has not been integrated in a closed loop with the aerodynamic analysis. Instead, the structural code has been used to size the wing weight already optimized with regard to aerodynamic. The aim of this study was to prove that it was possible to eventually carry out such a multidisciplinary optimization with the available tools for future analyses.