Coupling Parametric Aircraft Lofting to CFD & CSM Grid Generation for Conceptual Design

Coupling Parametric Aircraft Lofting to CFD & CSM Grid Generation for Conceptual Design Journal: Manuscript ID: Draft Conference: 49th AIAA Aerospace Sciences Meeting Date Submitted by the Author: n/a Contact Author: Moore, Mark

Page 1 of 18 Coupling Parametric Aircraft Lofting to CFD & CSM Grid Generation for Conceptual Design A. Rizzi, J. Oppelstrup M. Zhang and M. Tomac Royal Institute of Technology (KTH), Stockholm, 100 44, Sweden The CEASIOM software system for airctaft conceptual design generates tables of forces and moments for the rigid or elastic aircraft with control surfaces using a choice of numerical methods of varying fidelity. In order to obtain this data, the aircraft geometry must be defined and computational meshes for CFD and CSM analyses need to be created. This paper focuses on the generation of computational meshes, for analysis models of different fidelity, from the parameters that define the geometry. As an example to illustrate the procedure, the lofting geometry that was generated by the conceptual design system RDS Professional for a high-speed transport configuration termed Danbus is imported into CEASIOM from which a meshable model is created. Grids for CFD and CSM analyses are created, and results produced in order to study aspects of stability& control characteristics and structural weights along with flutter properties. I. Introduction The CEASIOM software system developed in the EU-funded collaborative research project SimSAC 1 generates stability and control data for preliminary aircraft design using a choice of numerical methods of varying fidelity. The system aims at automation of the computation of tables of forces and moments for the rigid or elastic aircraft with control surfaces. In order to obtain this data, the aircraft geometry must be defined, computational meshes for (Computational Fluid Dynamics) and Computational Structural Mechanics analyses need to be created and finally, the solver parameter settings must be adapted to reliably perform multiple solutions from which the tables are compiled. This paper focuses on the generation of computational meshes, for analysis models of different fidelity, from the parameters that define the geometry. The process is illustrated by examples used in SimSAC, from left to right in figure 1, the TCR-C15 TransCruiser design study and wind-tunnel model, the EADS Ranger 2000 jet, and the SACCON UCAV wind-tunnel model. The TCR is a 200 ton MTOW 200 pax airliner with design cruise speed M 0.97. Its wind-tunnel model is scaled 40:1, and the EADS Ranger 2000 is a tandem seat light jet trainer which flew but was not produced. SACCON (Stability And Control CONfiguration) is the DLR-F17 scale 1:8 low speed wind tunnel model with a wing planform with significant negative twist at the outer kink. A. CFD and CSM on Low Fidelity Geometry? SimSAC 1 aims to generate stability and control data for use in aircraft conceptual design using methods of varying fidelity. The principal design decisions taken during the conceptual design phase regarding aircraft configuration choice, shape and size are obtained via numerous statistical estimations, analytical predictions and numerical optimizations, as indicated in the top half of Fig. 2. The configuration must be specified in sufficient detail so that its aerodynamic, weight and volume characteristics can be determined. Among the parameters that describe the design are lofting quantities that produce an external wetted surface, and propulsion, payload, fuel, and structural member data, both geometry and mass. The outcome of this process Professor, Dept. of Aeronautical & Vehicle Engineering, Teknikringen 8, rizzi@kth.se, AIAA Associate Fellow Professor, Div. Numerical Analysis, CSC/KTH, jespero@kth.se, AIAA Member Research Assistant, Dept. of Aeronautical & Vehicle Engineering, Teknikringen 8, mzha@kth.se, AIAA Member Research Assistant, Dept. of Aeronautical & Vehicle Engineering, Teknikringen 8, maxtomac@kth.se, AIAA Member 1 of 18

Figure 1. The examples: Left, TCR-C15 W/T model; Center, : EADS Ranger 2000; Right, SACCON UCAV W/T model is a 3-view drawing of the aircraft with performance characteristics predicted by the low-fidelity analyses. A Boeing 737-200 drawing is shown in figure 3. Now we would like to increase the fidelity of the analyses by using CFD and CSM to advance the design (bottom half of Fig. 2), but these both require some type of computational grid, and the problem is that the 3-view drawing is not sufficient information to readily build a grid upon. Grid generation for such methods usually begins with a CAD-specified IGES file of the aircraft. This paper addresses the challenge of how to bring higher-fidelity CFD and CSM analysis methods into the conceptual design stage when the geometry description is still low fidelity, see ref.2 The crucial issue is the construction of a meshable model, i.e., a geometry definition which has the right level of detail as well as suitable mathematical properties such as smoothness, closure of surfaces, and topology. The different ways to approach this problem can be characterized by their degree of reliance on a CAD-system proper. The CAD-based approach is to base the entire design on a CAD system, e.g. CATIA, from the start. The parameter-based lofting procedure is defined by macros in the CAD system itself, see Oliveira2 and Amadori.4 Although theoretically highly attractive, this approach has significant disadvantages. The complex nature of both CAD systems and high-fidelity flow solvers, combined with the associated substantial licensing and training cost, limits the number of aerospace designers who also are experienced users of CAD, mesh generation, CFD and CSM software. Furthermore, the geometry description exported by standard CAD systems to the mesh generation software tends to be unnecessarily complex and usually requires extensive manual CAD repair simplification efforts, thereby severely encumbering automation. A second approach is CAD-free, and lofts the configuration using the parameters in the classical way, see Kulfan5 and Rodriguez.6 The result is a customized geometry modeler for aircraft design, as exemplified by the NASA VSP7 project, which supports export of meshable surface models and/or meshes in standard formats such as IGES, STEP, and CGNS. The third approach is a hybrid of these two in that it uses a vendor-neutral API, such as the CAPRI system [see www.cadnexus.com], to a number of CAD systems, relieving the problem of being tied to just one CAD system, see Fudge.8 We have worked with both the hybrid Berard9 and the CAD-free Eller10 approaches, and found the latter easier for us to reach our goal of automatic grid generation, Eller10 and Ito.11 This paper describes the CAD-free procedure that we have implemented to construct CFD and CSM grids for use in conceptual design. Application to the Danbus transonic transport design illustrates the modeling and grid generation. The Danbus design was studied in Raymer12 by the RDS conceptual design tool. Starting with the RDS model, the study is extended here by higher-fidelity CFD to elevator authority assessment in the transonic speed range and flutter and aileron reversal limitations on the flight envelope. B. Paths to Meshable Models It is clear that the conceptual design should be passed to the following design phases in a format which enables easy application of the relevant analysis methods: further down the line, the aircraft lives in a CAD system, such as CATIA V5. The are several paths in CEASIOM from the set of geometric parameters which describe a CEASIOM aircraft to a computational mesh. The sumo path, figure 4, employs the custom sumo surface modeler and -mesher and the TetGen tetrahedral mesh generator. The CADac path,13 uses the vendor-neutral geometric 2 of 18 Page 2 of 18

Page 3 of 18 Figure 2. Early Design Landscape: From lower to higher fidelity analysis - Two design loops in conceptual phase then downselect for further analysis in preliminary phase (adapted from Raymer 3 ) modeling API Capri 14 from CADNexus to build a meshable model in the CAD system. The inhouse sumo path has become preferred because of the complexities of developing for several different CAD systems, even if CAPRI does hide much of the idiosyncrasies. C. The CEASIOM Multi-Fidelity Framework CFD computations to estimate aerodynamic forces and moments early in the design stage can provide a head start on the controls design. For this strategy to succeed, so that changes in the aircraft configuration can be assessed at acceptable costs, the simulation methods must be: 1) fast, 2) reasonably accurate, and 3) easy to use. These three requirements can be addressed by adaptive fidelity CFD. Low order methods are used in the low speed linear region, and higher order solvers in the high speed and non-linear region. Another aspect that needs to be taken into account is the capabilities of the methods to represent the actual geometry, illustrated here by the Ranger 2000, figure 1(b). 1, 15 Figure 5 shows an overview of (part of) the CEASIOM framework. For further details see references. The framework integrates discipline-specific tools with main focus on aircraft conceptual design, and we show here only the geometry to computational mesh part. The green components are in-house developments 3 of 18

Page 4 of 18 Figure 3. 3-view drawing of Boeing 737-200 and the light blue and white modules are third party commercial or free products. As mentioned above, the focus of this paper is on streamlining the determination of aerodynamic data. Instead of adopting a fast potential-flow model enhanced with highly-refined empiricism, CEASIOM uses adaptive-fidelity CFD. The CFD module models range from DatCom, via vortex lattice modeling to CFD in inviscid Euler or full RANS mode. The choice of inviscid flow model depends on how much realism is needed to capture the flow physics at hand, with only secondary reliance on empirical modeling to enhance that realism further. At the high end a RANS solver completes the span in fidelity. Figure 6 shows the geometryand flow model-fidelity hierarchy. The simplest geometry could be a collection of global shape parameters with a level of accuracy suitable for Vortex Lattice (VLM) aerodynamics, Fig. 6 (lower left) and Fig. 8, left. The next higher level would take into account the actual surface of the aircraft, with some details excluded. This geometrical level would be suitable for Panel and Euler simulations, Fig. 6 (center) and Fig. 8, right, and at figure 6, right hand top, the full CAD and RANS type of simulation would be employed for, e.g., analysis of extreme portions of the flight envelope, high-lift systems, and control surface hinge moments. II. Coupling Lofting Geometry to Meshable Models Figure 2 indicates the increasing fidelity of the analysis. The progression from simple to more detailed analysis is generic, and the discussion is quite general. However, to make it more concrete here, we make reference to the specific computational tools used in CEASIOM, and to the concept-design & initial-sizing package RDS developed by Raymer, 16 without claim to uniqueness for features or functionality. The particular set of shape parameters adopted for the SimSAC project is here referred to as the CEASIOM geometry description and is in XML format. The simplest geometry representation is a collection of global shape parameters which can be used to describe a set of typical aircraft configurations with a level of detail suitable for use with DATCOM or Vortex Lattice (VLM) aerodynamics. At the next higher level of detail, the graphical surface modeling tool sumo can be used to define a more detailed geometry based on a moderate number (often less than 30) spline surfaces. This description is used to generate input for a 3D panel method and for CFD solutions based on the Euler equations. At a later stage, when Navier-Stokes solutions are of interest, the sumo geometry can be exported to CAD systems or commercial mesh generation software by means of IGES files. A. From 3-views to CFD and CSM Mesh A number of tools exist for the concept design stage, such as RDS Professional, 16 AAA,, 17 etc. They are streamlined for creating the initial configuration and sizings, systems, etc and are great for sizing, performance and rapid requirements trade studies because they allow the development, assessment, and optimisation of a new notional design concepts in as little as a day. The geometric fidelity offered can be quite refined, including items like external stores, landing gears extended and retracted, and interior 4 of 18

Page 5 of 18 Figure 4. sumo Path from CEASIOM TCR-C15 to Tet Mesh and ICEM CFD Hexa Mesh objects such as ejection seats. This makes it possible to render realistic images and 3-view drawings of the configuration from the start. But these tools cannot put a CFD or CSM mesh on the geometry. This is because the connectivity and quality of the individual surfaces are in some cases inadequate for CFD, and one is faced with the well-known CAD repair -task. However, when the configuration has been described by a custom tool, such as those mentioned, the problems can be mitigated because the tools know about aircraft components. As an example, figure 7, left, is the designer s impression of a low-wing, boom-tail, pusher-prop general aviation craft. The inverted-v tail keeps the lifting surfaces out of the prop stream. This model is not meshable. Typical problems have been indicated in figure 7, right. Inverted-V tail The tail is composed of a symmetric wing with unconventional dihedral and d:o root chord position, and a top portion which is a set of two parallel, curved surfaces; To be meshable, the whole tail should be modeled as a wing with strongly varying dihedral. Wing trailing root The wing root makes a very sharp spike with a very small angle with the fuselage bottom. To avoid extremely small elements, the wing root and the fuselage must be better matched, in this case it is enough to shift the whole wing up. Boom-tail junction The booms merge smoothly with the wing, but not with the tail roots. This requires more precise positioning to ensure that the tail intersects the boom surface properly. We claim that the CAD-Repair problem for the transition from initial sizing tool to higher fidelity analysis is manageable, and exemplify its solution by the RDS-CEASIOM tool chain. The (semi)automatic repair tools in sumo include only features to close wingtips, fuselage nose and tail and make rounded air intake lips. Components can be selected or de-selected, their positions and attitudes can be modified, and with more 5 of 18

Page 6 of 18 Figure 5. CEASIOM Framework work new components can be designed and added. To substantiate the claim, a simple interface from RDS to sumo has been developed. 1. RDS RDS 16 is an integrated software for aircraft design and analysis using Multidisciplinary Design Optimization (MDO) described in Raymer s Ph.D. thesis. 12 The geometry is built from components, each component described by a set of cross sections, in turn defined by a set of points, and can be stored as a DNS text file with sufficient syntactic information to be interpreted by other programs. The DNS format includes SAWE RP8A (Mil-Std-1374a) Group Weight Statement component categories codes for the components. In order to do further aerodynamic and flight analysis for the aircraft configurations creates in RDS, we have created a converter for RDS models into sumo models. CSM analysis are enabled by conversion to CEASIOM, which is at present only semi-automatic, requiring e.g. approximation of the fuselage described by cross-sections into the analytic model of CEASIOM. 2. RDS - sumo Interface The logic of geometry data stored in DSN files and SMX files is quite similar, that is the prerequisite for the idea of transformation. Both RDS and sumo geometry files describe the aircraft geometry by defining the coordinates of the sections for each component so contain essentially the same information, for instance, names of airfoil sections to build lifting surfaces. However, the file formats, and the exact definition of some quantities are different. In particular, symmetry flags and 3-D translations and rotations are different, rotations are applied in different orders, etc. The DNS file contains for each component General data, including 3D location and rotation for the component; Data for symmetry, surface type, axis orientation, etc., SAWE code ; Change date, author; Reference data matrix; Local coordinates, including Number of sections, planarity flag, Point coordinates for each section 6 of 18

Page 7 of 18 Figure 6. Adaptive Fidelity CFD map (a) RDS pusher-prop, front view (b) RDS pusher prop, geometric glitches. Figure 7. RDS pusher-prop: Left, front view, right, problem areas highlighted The sumo SMX file is an XML file defined standard XML technology. The converter is a Matlab program which builds a hierarchical structure (Matlab struct) containing all the information. The geometric information is translated, and finally transformed to the SMX format. B. CEASIOM Geometry The particular set of shape parameters adopted in SimSAC is referred to as the ceasiom15 geometry. The parameters describe a model built from components such as fuselage, wing1, wing2, horizontal tail, etc., and the parameters have immediate interpretation for the aerodynamicist. The geometric model describes lifting surface planforms (with twist and dihedral) and cross-sections from a library of airfoil shapes, control surfaces, engine configurations, and a simple fuselage model. The geometry is created and edited by the ACBuilder interactive module, and saved in XML format. Further processing steps, such as payload and fuel positioning, estimations of weights and balance, etc., record their results by augmenting the XML file. The ceasiom Ranger 2000 geometry is shown in figure 8, right. The CEASIOM fuselage model does not support a canopy, and the square air intakes and the nozzle have been approximated by two closely spaced cylindrical nacelles. CEASIOM geometry is easily interrogated for lifting surfaces such as wing, vertical tail, and stabilizer for VLM analysis. CEASIOM geometry may also be appropriate for Euler flow solutions, but as the Ranger example shows, it may be too inaccurate. Thus, at the next higher level of detail, sumo can be used to refine the CEASIOM model into a more detailed geometry based on a moderate number (often less than 30) spline surfaces, figure 9, top left. This description is used to generate a triangular mesh on the surface, for a 3D panel method and as surface mesh for CFD solutions based on the Euler equations, figure 9, 7 of 18

Figure 8. Left: VLM Geometry; Right: CEASIOM Ranger 2000 top right. When viscous flow models are called for, the sumo geometry can be exported to CAD systems and/or mesh generation software by means of IGES files. Figure 9, bottom, shows a RANS mesh made semi-automatically from the sumo mesh. Whereas the Euler mesh generation is quite robust, RANS meshing is much more challenging. The CEASIOM system has taken steps in this direction, as described below. Figure 9. Top-Left: Sumo geometry; Top-Right: Euler mesh; Bottom: RANS mesh III. CEASIOM Mesh Generation From the ceasiom geometry parameters, the VLM mesh can be generated directly within the Matlabbased ceasiom environment. For higher-fidelity geometry descriptions, sumo and TetGen are employed. Starting from the spline surface geometry, a closed unstructured triangular surface mesh is generated by a 8 of 18 Page 8 of 18

Page 9 of 18 heavily modified version of Chew s algorithm. 18 The mesh quality is controlled by local geometric quantities such as element dihedral angles and aspect ratio. Following the surface meshing procedure, the volume between aircraft and farfield boundary is filled by Hang Si s TetGen quality Delaunay tetrahedral mesh generator. 19 The resulting unstructured volume mesh is suitable for inviscid flow solutions as long as the quality of the underlying surface mesh suffices. Generation of volume meshes for Navier-Stokes computations can exploit the comparatively clean geometry representation of the sumo-exported IGES (CAD) and CGNS (Mesh) files. Using the icemcfd a mesh generator, a meshing template including boundary layer regions is carefully prepared for a baseline geometry. By means of a set of Python and Tcl/Tk scripts 20 and the mesh generator, this template can then be applied to geometrically similar configurations in order to create a sequence of computational grids. Figure 9 illustrates the typical level of detail which can be achieved. A. Surface Mesh Generation The accuracy of the numerical solution of the flow problem is affected by the quality of the surface discretization. The surface mesh must approximate the geometric surface sufficiently accurately; resolve inviscid flow features such as pressure peaks and shocks. While the first requirement is, in most cases, a necessary condition for satisfaction of the second, it is not always sufficient. As an example, consider the pressure recovery near the trailing edge of a lifting surface. Although the surface geometry usually is flat in this region, small elements are needed to capture the steep pressure gradient. The surface mesher automatically ensures that the first requirement above is fulfilled to given tolerances. A set of geometric heuristics described below are used to refine particular regions where large variations in pressure are typically observed. The mesh generation is performed in the parameter space (u, v) by means of an algorithm similar to that presented by Chew, 18 which produces a Chew-Delaunay mesh where the circumsphere of any (x, y, z) triangle does not enclose vertices of any other triangle. The iterative process of improving mesh quality, which dominates the whole meshing process, is robust only if it starts from a mesh which is Chew- Delaunay in this three-dimensional sense. 1. Mesh Initialization All surfaces are represented by bi-parametric patches of the form (x, y, z) = S(u, v), where at least the first derivatives with respect to the parameters u and v are continuous. To initialize the process with an (at least approximately) Delaunay surface mesh, a structured quadrilateral mesh is first created on each surface. Another pass is applied in which the initial, highly stretched quadrilateral mesh is converted into a geometrically adapted triangular mesh. 2. Mesh Refinement Mesh improvement for size and shape is performed in an iterative process by refinement: If all elements are small enough on a smooth surface, the criteria will be satisfied. In each pass, all triangles which violate one of the quality criteria are collected as candidates for refinement. The refinement is performed by splitting the longest edge of any offending triangle. The process is repeated as long as non-conforming elements are found, which usually requires between four and eight passes. The criteria can be set by the user, and the mesh generator attempts to define sensible default values which work in most cases. The following parameters can be used to control the mesh generation process: Dihedral angle - the angle between the normal vectors of two triangles sharing an edge. Edge length - a minimum and maximum length. The minimum avoids resolution of irrelevant geometric details. a http://www.ansys.com/products/icemcfd.asp 9 of 18

Page 10 of 18 The triangle stretch ratio - the ratio of the longest to the shortest edge in a triangle. Leading- and trailing-edge refinement factors are used to control resolution of regions which typically feature large pressure gradients. 3. Surface Intersections Surface intersections are computed based on the triangular meshes on each surface. The current implementation requires that no more than three surfaces intersect in a single point and uses a bounding volume hierarchy (BVH) for the polyhedral meshed surfaces to speed up the computation of element intersections. A postprocessing stage removes excessively short intersection line segments where possible, and the intersection polylines are then imposed as constrained edges for a subsequent mesh refinement pass. 4. Creation of Wetted Surface Mesh The meshes are merged and duplicate vertices eliminated. There remains to determine the wetted set of triangles. Starting from a small initial set of elements on the external (wetted) surface, the merged mesh is processed by a topological walk over element edges with geometric criteria to select next external elements. B. Euler Volume Mesh Generation When a high quality surface mesh has been created, an unstructured tetrahedral volume mesh for the solution of the Euler equations is generated by TetGen, developed by Hang Si at the Weierstrass Institute in Berlin. TetGen is a very efficient quality-constrained tetrahedral Delaunay mesh generator. Starting from an initial constrained Delaunay tetrahedrization of the domain, nodes are dynamically inserted until a given set of quality criteria is met. The domain is bounded by surface meshes of the aircraft configuration and the farfield boundary, which are created by sumo. Tetrahedral quality criteria available in TetGen, version 1.4.3, include maximum element volume; maximum ratio of circumsphere radius to tetrahedron edge; minimum dihedral angle between faces. Furthermore, the maximum permitted element volume can be defined to vary with (x, y, z) once a volume mesh exists. This feature is not yet exploited by the sumo calls to TetGen. In order to comply with imposed element quality requirements, TetGen will subdivide even boundary triangles if explicitly allowed to do so. Since it currently has no access to the original higher-order surface description, the subdivision is performed in the polyhedral surfaces. C. Towards Automatic RANS Mesh Generation RANS meshes must resolve not only large pressure gradients but also the boundary layer and other types of shear layers. Economical RANS meshing is often a time consuming, iterative process. The approach here is to create a boundary layer mesh of triangular prisms on the wetted surface meshes, and fill the volume between the upper surface of the prism layer and the farfield surface by a tetrahedral mesh. Mesh refinement for free shear layers and free vortices must be done by e.g. adaption in the flow solver. Since the surface grid cannot be modified at this stage, it must be of high quality to start with. The initial volume grid needs to be satisfactory as well, however the requirements on this grid are not as stringent since it will be smoothed by the RANS meshing schemes. The experiments use surface meshes from sumo as input to the ICEM CFD mesh generator to generate the unstructured volume mesh of prisms and tetrahedra, and if necessary, pyramids. A Python program and a template library of Tcl/Tk scripts for different aircraft configurations was developed for semi-automatic RANS meshing. The engineer sets up a prism parameter file, the Python program reads it and adjusts the scripts which control the mesh generator. When the mesh has been created, its quality is displayed. If unacceptable, the engineer must modify the prism file and run again, and/or run mesh improvement tools. The scripts export the mesh in e.g. CGNS format. This approach has been tested on a number of different aircraft configurations. For demonstration purpose the Ranger 2000 was fitted with a V-Tail instead of the conventional T-Tail configuration. The program was 10 of 18

Page 11 of 18 Figure 10. Example configurations where the automatic RANS meshing works. Top: left, V-tail Ranger, right, light executive jet; Bottom: Left, SACCON UCAV, right, TCR-C15 also applied to a light exclusive jet with winglets, the SACCON UCAV, and the TCR-C15TCR, all shown in Fig. 10. But the robustness comes at a price in terms of poor quality and stringent requirements on the surface grid. The engineer is hence still an important part of the loop in terms of monitoring the outcome of the meshing schemes. D. 1. CEASIOM Flow Solvers Edge Flow Solver The Edge CFD solver21 is an edge- and node-based Navier-Stokes flow solver applicable for unstructured grids. Edge is based on a finite volume formulation where a median dual grid forms the control volumes with the unknowns allocated in the centers. The governing equations are integrated to steady state, with a line-implicit approach in areas with highly stretched elements and explicitly elsewhere with a multistage Runge-Kutta scheme. The steady state convergence is accelerated by FAS agglomeration multigrid. 2. Tornado VLM Code The Tornado VLM code22, 23 is an open source Matlab implementation of a modified horse-shoe vortex singularity method for computing steady and low reduced frequency time-harmonic unsteady flows over wings. The lifting surfaces are created as unions of thin, not necessarily flat, quadrilateral surface segments. Effects of airfoil camber is modeled by surface normal rotation. Leading edge control surfaces are modeled likewise, and trailing edge devices by actual mesh deformation. The basic flow solver is wrapped by user interfaces to create tables of aerodynamic coefficients, and d:o derivatives, for export to flight simulators and flight control system design software. IV. Grid Generation for CSM The geometry is built upon the SimSAC ACBuilder (which produced Fig. 11) merged with the hierarchical geometry modeler sumo, which provides geometry design, IGES output, and high-quality unstructured 11 of 18

Page 12 of 18 surface meshing for panel and Euler flow modeling as described above. It approximates all major features of transport, general aviation, and military aircraft, conventional and un-conventional. It supports lifting surfaces, fuselages, tail booms, fairings, pylons, nacelles, and integration of engine intakes and nozzles and is aware of control surfaces such as ailerons, elevators, rudders, flaps, slats, and all-moving surfaces (Fig 11(a)). The internal shape representation supports placement of components such as payload, structural members, engines, control actuators, and landing gear (Fig 11(b)). The structural model (Fig 11(f)) is illustrated by (but not limited to!) a beam model with point masses. A higher-fidelity shell model can be produced from the CFD-meshed surfaces and information on locations of spars and other important structural members. (a) (b) (c) (d) (e) (f) Figure 11. Six sub-sets of the GeoSUMO geometry: a) external shape, b) cabin and payload, c) fuel, d) equivalent structures, e) vortex-lattice mesh, f) spar locations The complete Aero-Elastic model of the concept aircraft is represented in the CEASIOM XML schema, which includes properties of surfaces, masses and volumes,structures, fuel, control surfaces, propulsion, and control systems and performance evaluation models. 12 of 18

Page 13 of 18 A. Example Case: Shell Model for Canard Flier An example of the shell model created by sumo is shown in 12 for a canard general aviation flier from.24 The geometry is made from intersections of the external shapes of the surfaces with planes to make the wing box and the profiles. Since there is no need for modeling the intersections between fuselage and lifting surfaces geometrically accurately in the mechanical model it is easy to generate the quadrilateral meshes needed for good shell element accuracy. The leading and trailing edges of the wings have been left out because they contribute but little to strength and are not stressed by most sizing maneuvers. However, these structures are usually taken into account for deriving the equivalent beam model because of their added stiffness which is important for the flutter assessment. Figure 12. Left: Shell models of lifting surfaces; Right: Surface mesh for panel methods. V. Example Case Danbus Transport The procedure outlined above is applied to the design of the DanBus commercial transport reminiscent of the Airbus 321, designed and analyzed in Raymer s Ph.D. thesis.12 It is a twin-engine, single aisle, 188 pax design with takeoff gross weight of 97 metric tons and an empty weight of 52 tons. The airplane has a length of 44.5 m and a wing of span 40 m, with a wing aspect ratio of 10.13 and an area of 155 sqm. Figure 13 shows the DanBus model in sumo. The pylons and control surfaces are added in the repair step from RDS to sumo. (a) sumo TetGen volume mesh (b) sumo tail control surfaces Figure 13. The DanBus civil transport in sumo 13 of 18

Page 14 of 18 A. Stability & Control Analysis The control surfaces were selected similar to the A320, with two elevator segments for longitudinal control, two ailerons for lateral control, and a rudder. The surface grid of the tail is shown in Fig. 13(b). Control surface deflections are treated in the inviscid CFD solvers by smooth mesh deformation, or by transpiration conditions, i.e. a modification of the surface normals in expression of the flow tangency condition. Therefore, the mesh need not conform exactly to control surface edges, as is revealed by close scrutiny of the mesh pictures. For more accurate viscous modeling, e.g. to take the gap flows into account, re-meshing for each deflection angle is called for, as well as more detailed geometry modeling. Such high-fidelity modeling is necessary for hinge moment prediction. 1. High-speed trim The aerodynamic forces and moments are computed based on the DanBus geometry and the automatic Euler mesh generation, with 9 million cells. The surface grid resolution is quite fine, as seen in 13(b). The solution is computed for the flight condition M = 0.8 at 10km and α = 6, and Fig. 14 displays the surface pressure distribution that indicate the transonic shock waves. Figure 14. Computed surface pressure on Danbus, Euler simulation M = 0.8, α = 6 Lift, drag, and pitching moment coefficients as predicted by DATCOM, VLM with Prandtl-Glauert correction, and inviscid CFD are compared in Fig. 15. Fig. 15(a) shows that DATCOM lift-curve slope does not agree with Edge and Tornado. The Prandtl-Glauert correction is effective, but the linear flow models do not predict the stall - and the Euler model is not expected to be accurate, either, at such angles of attack. Fig. 15(c) demonstrates that both DATCOM and TORNADO seriously underestimate the drag. TORNADO only predicts the drag due to lift. The Euler prediction includes the wave drag, which is substantial and is expected to dominate over drag due to friction. The most interesting quantity for trim is the pitching moment, Fig. 15(b), which shows large differences between DATCOM, TORNADO and EDGE results. DATCOM predicts the slope correctly but not the zero-lift moment C m0. It is expected that the transonic shift of the neutral point from 21.9m at M = 0.15 to 22.9m at M = 0.80 (as given by the Euler simulation) is responsible for the failure of the linear flow 14 of 18

Page 15 of 18 models. It is also known that the Prandtl-Glauert correction is inaccurate for moment prediction, and neither DATCOM nor TORNADO predict the non-linearity of the moment curve. The trim condition given by the Euler simulation for straight and level flight at the cruise point M = 0.8, Alt = 10km (i.e. U = 239.6m/s) is α trim = 0.26 and δ trim = 1.15 using the NeoCASS-predicted center of gravity CoG = 21.54m from the nose. (a) Lift coefficient (b) Pitching moment coefficient, X Ref 17.5 m from the nose (c) Drag coefficient Figure 15. Force and moment coefficients for DanBus from Euler and VLM calculation, M = 0.8, Altitude = 10 km B. Weights and Structures Analysis To carry out the structural analysis for the Danbus aircraft, the NeoCASS suite is used. It includes two main modules, GUESS and SmartCAD. GUESS derives the structural equivalent beam stick model from aircraft geometry, sizing maneuvers, and mass distribution. The SmartCAD module runs the NeoCASS modal, flutter, and static aero-elastic analyses of the stick model. The flutter is calculated by a coupled structural beam - aerodynamic doublet lattice model. The interpolation of DLM mesh deformations from beam deflections is done by radial basis functions, and the transfer of aerodynamic loads to concentrated structural loads satisfies a Galerkin condition to ensure virtual work consistency, see Cavagna et al. 25 1. GUESS Estimated Structural Weight GUESS provides the stick model (equivalent beam) for the aircraft and the estimated structural weight, component weights on the left and stick model on right in Fig. 16. GUESS uses default sizing maneuvers and needs only the aircraft geometry as input and predicts a total structural weight of 25574 kg. Without fuel the weight is 22006 kg. Table 1 compares the weights predicted by Raymer and by NeoCASS. [kg] MDO(Raymer s thesis) NeoCASS total structure 26641.3 25873.9 Wing 9761.0 7423 HT 343.6 520 VT 1285.4 1297 Fuselage 9959.7 12766 Landing gear 3544.8 1733.15 Table 1. Weight estimation comparison 2. Modal Analysis - Inertial Relief of Wing Tank Fuel The CEASIOM model includes wing spar and fuel tank geometry, so the dynamics can be investigated for different payload and fuel levels. The flutter speed and flutter modes are computed for two cases, (1) MTOW and (2) MTOW minus fuel in wing tank. 15 of 18

Page 16 of 18 (a) Weight breakdown & location, CG = 21.5m (b) Equivalent beam model - MTOW Figure 16. Structural model DanBus civil transport in NeoCASS For case (1), modes 9 (2.48 Hz) and 17 (5.69 Hz) are marginally stable and prone to flutter. The lowest flutter speed is 294 m/s, see Fig. 17, which is well above the cruise speed, hence outside the envelope. (a) Vibration Mode 9 (b) Vibration Mode 17 Figure 17. Vibration Modes for the DanBus at MTOW In case (2), with empty wing tank, flutter speed decreases to 181 m/s which may restrict the envelope. The change is due to loss of inertia of the fuel in the wing tank. In this case, modes 10 (2.99) and 16 (6.10 Hz) become unstable, see Fig. 18. Eigen-systems for parametric families of matrices are continuous only as long as eigenvalues stay separate, and following the modes through a degenerate point is not always possible. Also, mode plots are scaled, and there is a sign ambiguity, so visual comparison must be careful. In this case we see that case (1) mode 9 and case (2) mode 10 are very similar. Case (1) mode 17 and case (2) mode 16 are both concentrated to wing tip bending, but mode 16 is symmetric whíle mode 17 is almost anti-symmetric. The ensuing roll-vibrations of the fuselage are transmitted to the horizontal tail which participates much more than for the symmetric mode. 16 of 18

Page 17 of 18 (a) Vibration Mode 10 (b) Vibration Mode 16 Figure 18. Vibration Modes for the DanBus - MTOW minus wing fuel VI. Conclusions & Outlook The example of the Danbus high-speed transport has demonstrated that it is possible in a nearly automated way to generate CFD and CSM grids upon the lofted surfaces that are typical for aircraft designed at the conceptual phase. The lofting geometry that was generated by the conceptual design system RDS Professional for the Danbus configuration was imported into CEASIOM and a meshable model was generated upon that geometry. The CEASIOM software system then could compute forces and moments for the rigid or elastic Danbus with control surfaces using a choice of numerical methods of varying fidelity. This paper described the sumo module that produces the computational meshes on the lofted surfaces. Aspects of stability& control characteristics and structural weights along with flutter properties of the Danbus were studied and discussed. Work continues on improving the meshing techniques, especially for RANS mesh generation. In addition to beam models other CSM meshes e,g, equivalent plate and particularly shell models are under investigation. Exporting the output of NeoCASS to e.g. NASTRAN will be explored. Also interfaces to other aircraft conceptual design programs are being considered to increase the generality of imports into CEASIOM. VII. Acknowledgments The authors gratefully acknowledge Dr. David Eller, KTH, the author of sumo, for support and technical discussions, and permission to reproduce figures 12(b) and 12(a), Dr. Dan Raymer for the RDS Professional DanBus and pusher prop models, and Dr. Stefan Hitzel of EADS-MAS, for providing the Ranger 2000 data in accessible form. References 1 The SimSAC Project homepage. Simsac. http://simsacdesign.org/, July 2009. 2 A. Martins G. Becker M. Reis N. Spogis R. Silva G. Oliveira, L. Santos. A Tool for Parametric Geometry and Grid Generation for Aircraft Configurations. 26TH ICAS Congress. 3 D. P. Raymer. Aircraft Design: A Conceptual Approach. AIAA, Berlin, 4th edition, 2006. 4 K. Amadori. On Aircraft Conceptual Design : A Framework for Knowledge Based Engineering and Design Optimization. University Dissertation from Dept. Economic and Industrial Development, Linkoping University of Technology, Linkoping, Sweden, 2008. 5 B. Kulfan and J. Bussoletti. Fundamental Parametric Geometry Representations for Aircraft Component Shapes. AIAA Paper 2006-6948, 2006. 6 D. Rodriguez and P. Sturdza. A Rapid Geometry Engine for Preliminary Aircraft Design. AIAA 2006-929, 2006. 7 A. S. Hahn. Vehicle Sketch Pad: A Parametric Geometry Modeler for Conceptual Aircraft Design. In 48th AIAA Aerospace Sciences Meeting, Orlando, Florida, January 2010. AIAA 2010-657. 8 D. Zingg D. Fudge and R. Haimes. A CAD-Free and a CAD-Based Geometry Control System for Aerodynamic Shape Optimization. AIAA Paper 2005-0451, 2005. 17 of 18

Page 18 of 18 9 A. Rizzi A. Berard and A. Isikveren. CADac - A New Geometry Construction Tool for Aerospace Vehicle Pre-Design and Conceptual Design. AIAA Paper 2008-6219, 2008. 10 D. Eller. Procedure to Vary/Control Geometric Parameters that Allows Automatic Unstructured Mesh Generation in Aeroelastic Analysis as well as in Design Optimization. SimSAC Project Report D2.3-5, 2009. 11 Y. Ito and K. Nakahashi. Unstructured Mesh generation for Viscous Flow Computations. Proc 11th Intl Meshing Roundtable, pages Ithaca, New York, 2002. 12 D.P. Raymer. Enhancing Aircraft Conceptual Design Using Multi-Disciplinary Optimization. Doctoral Thesis, 2002. 13 A. Rizzi A. Berard and A. T. Isikveren. CADac: a New Geometry Construction Tool for Aerospace Vehicle Pre-Design and Conceptual Design. In AIAA, 26th Applied Aerodynamics Conference, Honolulu, Hawaii, USA, August 18-21 2008. 14 R. Haimes. CAPRI: Computational Analysis Programming Interface, A Solid Modeling Based Infra-structure for Engineering Analysis and Design, CAPRI users guide Revision 1.00. MIT, December 2000. 15 A. Rizzi et al. CEASIOM Validation and Its Use in Design - Status of SimSAC Project. In SAAB Flygteknikseminarium, Kolmården, November 2008. 16 D.P. Raymer. RDS: A PC-Based Aircraft Design, Sizing, and Performance System. AIAA Paper 92-4226, 1992. 17 J. Roskam. Airplane Design I-VIII. Roskam Aviation & Engr Corp, Kansas, 2nd edition, 1990. 18 L. P. Chew. Guaranteed-Quality Mesh Generation for Curved Surfaces. In Proceedings of the Ninth Annual Symposium on Computational Geometry, San Diego, May 1993. Available from the ACM Digital Library, portal.acm.org/citation.cfm?id=161150. 19 H. Si. On Refinement of Constrained Delaunay Tetrahedralizations. In Proceedings of the 15th International Meshing Roundtable, September 2006. Software available from http://tetgen.berlios.de, July 2009. 20 Tomac, M. Creation of Aerodynamic Database for the X-31. 48th AIAA Aerospace Sciences Meeting, 2010. 21 Edge home page. http://www.foi.se/. 22 Tornado home page. http://redhammer.se/tornado/. 23 T. Melin. Using Internet Interactions in Developing Vortex Lattice Software for Conceptual Design. Licentiate Thesis, 2003. Software available from http://redhammer.se/tornado/, July 2009. 24 D. Eller. Simulation of Maneuvering Elastic Aircraft. In SAAB Flygteknikseminarium, Kolmården, November 2008. 25 S. Ricci L. Cavagna and L. Travaglini. NeoCASS: an integrated tool for structural sizing, aeroelastic analysis and MDO at conceptual design level. AIAA Paper 2010-8241, 2010. 18 of 18