Optimization of Waverider-Derived Crew Reentry Vehicles Using a Rapid Aerodynamics Analysis Approach

AIAA SciTech 5-9 January 2015, Kissimmee, Florida 53rd AIAA Aerospace Sciences Meeting AIAA 2015-0757 Optimization of Waverider-Derived Crew Reentry Vehicles Using a Rapid Aerodynamics Analysis Approach Marcus A. Lobbia 1 The Aerospace Corporation, El Segundo, California, 90245 h / A unique implementation of commonly used engineering methods for estimating supersonic and hypersonic aerodynamics of arbitrary shapes is discussed. A validation case comparing the aerodynamic coefficients generated by this technique with both wind tunnel and published computational results for the NASA HL-20 configuration from Mach 1.2 to Mach 10 is presented. These results indicate that the method can generate aerodynamic databases where the lift and drag coefficients are within 15% of experimental data in minutes on a typical workstation. Therefore, the benefits of this rapid aerodynamic analysis approach are further investigated by integrating it into a multidisciplinary design optimization framework for waverider-derived crew reentry vehicles. A shape design code, mass estimating model, and trajectory/aeroheating analysis are linked using a genetic algorithm optimization process to generate a pareto front comparing vehicle downrange versus landed mass. The results of this analysis indicate that a higher lift-to-drag ratio (L/D) typically results in a longer range, but also correlates with a larger vehicle mass. Finally, a comparison with the HL-20 is performed to highlight how the waverider-derived design approach can produce configurations exhibiting a 20% increase in maximum L/D with only a 4% increase in vehicle length. = speed of sound = pressure coefficient = altitude = Lift-to-drag ratio = Mach number = surface normal vector = pressure = radial coordinate in Taylor-Maccoll equation Nomenclature = waverider shock profile curve radius for cone region in base plane = Length-based Reynolds number = radial velocity in Taylor-Maccoll equation = angular velocity in Taylor-Maccoll equation = axial center of gravity location from vehicle nose (non-dimensionalized to vehicle length) = waverider lower surface base curve control point (vertical) in base plane = waverider lower surface base curve control point (horizontal) in base plane = velocity vector = shock wave angle = ratio of specific heats 1 Senior Project Engineer, Launch Strike and Range Department, 2310 E. El Segundo Blvd., Mail Stop M1-557; AIAA Member. 1 Copyright 2015 by The Aerospace Corporation. Published by the, Inc., with permission.

= Prandtl-Meyer function value = density = surface inclination angle; angular coordinate in Taylor-Maccoll equation Subscripts 02 = station number for stagnation point behind a normal shock wave = surface panel index = streamline segment index = freestream I. Introduction ITH the retirement of the Space Shuttle in 2010, there has been substantial work in recent years aimed at Wdeveloping new crew reentry vehicle (CRV) systems. Many of these efforts leverage heritage reentry system designs, such as the Apollo crew capsule and NASA HL-20 lifting body shapes, to potentially reduce risk and cost. However, techniques such as multidisciplinary design optimization (MDO) can be combined with specific design methodologies such as hypersonic waveriders to create configurations that may provide improved aerodynamic performance and expanded mission capabilities (e.g., increased range). Waveriders have been shown to provide a potentially higher lift-to-drag ratio (L/D) relative to conventional configurations at on-design conditions [1] due to the attached shock wave along their leading edge. Application to a reentry vehicle, however, will require assessment of the effects of operation at off-design conditions and blunted leading edges. One key component in an MDO model for hypersonic reentry vehicles will be the aerodynamics estimation. While the speed and accuracy of computational fluid dynamics (CFD) techniques have advanced significantly in the last decade, the relatively long grid generation process and analysis time required for a single condition still limits the use of CFD in MDO analysis models. On the other hand, there are several lower fidelity engineering-level analysis approaches that can be useful for applications such as the preliminary design and optimization of complex hypersonic configurations. These approaches require only a surface mesh and the specification of the flow conditions, and the rapid computation time means that a table of aerodynamic properties can be generated on the order of minutes on a typical workstation vs. the days to weeks typically required for higher-fidelity techniques. The first part of the present work describes the development of the Fast Aerodynamics Analysis Tool (FAAT), which is a unique implementation of commonly-used supersonic/hypersonic aerodynamics analysis techniques (i.e., Modified Newtonian, Tangent-Cone, and Shock-Expansion methods) to allow automated application to complex Computer Aided Design (CAD) generated geometries where only a surface mesh is required. Validation of the tool is performed by comparing results for the NASA HL-20 lifting body configuration to published wind tunnel and numerical results at Mach numbers ranging from 1.2 to 10. In the second part, the development of a waveriderderived CRV shape optimization model is discussed, demonstrating how FAAT can be integrated with a parametric shape generation code, a simple mass estimating model, and a trajectory optimization tool to create a CRV MDO framework. Multi-objective optimization is performed to generate vehicle designs that meet launch vehicle mass and reentry heating/acceleration constraints, and highlight the trade-offs between vehicle mass and downrange. Finally, several optimized designs are compared with the HL-20 configuration to highlight the pros/cons of using waveriders as a basis for CRV design. II. Aerodynamic Analysis Model Overview The current version of FAAT is coded in C/C++, and utilizes OpenMP to parallelize computations on shared memory computer systems. Fig. 1 presents an overview of the implementation the analysis is broken up into a pre-processing phase (e.g. preparing the mesh geometry and tabulating analytical cone flow solutions) and an iterative analysis phase (e.g. performing the aerodynamic estimation at each condition specified). 2

Pre-Processing Phase Analysis Phase Read Input Remove Zero Area Elements Geometry Pre-Processing Rotate Mesh Geometry Calculate Surface Normals/Tangents Iterate for each Alpha, Mach, Altitude Refine Mesh Flag Shadow and Base Panels Find Shared Nodes Integrate Streamlines Flag Mesh Interior/Exterior Tabulate Cone Flow Solutions Calculate Freestream Properties Calculate Panel Properties Integrate Forces and Moments All Cases Complete? Write Output Fig. 1 Overview of FAAT implementation and analysis process A description of each of the steps shown in Fig. 1 is as follows: Pre-Processing Phase several steps are conducted to prepare for the aerodynamic analysis. o Read Input: In order to perform an aerodynamic analysis of a given shape, two input files are required. These include the geometry surface mesh (i.e. a convex hull representing the outer mold line) and a text file with the analysis settings and flow/shape parameters to be analyzed (see Table 1 for a summary of the user-specified options). The surface mesh is assumed to consist of triangular surface elements/panels this type of surface representation can be easily created from a CAD model by exporting the geometry to a Stereo Lithography (STL) file, for example. The current implementation of FAAT assumes an ASCII format Tecplot file in the unstructured point format, as this facilitates the use of both STL files (easily converted to Tecplot format), as well as allows the analyst to directly view the surface mesh being analyzed in the Tecplot visualization software package. o Remove Zero Area Elements: As it is possible that the external surface mesh creation may result in some elements with zero area (i.e. the three triangular vertices degenerate to a line or single point), a check is performed to remove any elements that have a zero area. (For example, when dividing a structured quadrilateral mesh into unstructured triangular elements, it is possible that a zero area element may result when the original quadrilateral is near a singular point and has two vertices with the same coordinates.) o Refine Mesh: An optional routine is performed to check the length of the sides of each triangular surface element against a user-defined maximum, and subdivide the element into three triangular elements when the maximum edge length exceeds that specified for the analysis. As the subdivided elements share the same surface normal as the original element (i.e. overall surface curvature is not considered), Yes No 3

this technique only provides refinement in calculating the streamline length used to estimate viscous effects. o Find Shared Nodes: The mesh is analyzed to determine the neighboring elements for each surface element. This check is performed by identifying elements that share common vertices. For each element, a list of neighboring elements is stored in memory to facilitate streamline tracing later in the analysis phase. o Flag Mesh Interior/Exterior: Each of the surface element normal vectors is evaluated using an intersection test to determine which direction corresponds to outward pointing. This test proceeds by creating a ray from the calculated normal vector in a specified direction (e.g. y-direction), and counting the number of surface elements intersected by the projection. If the number of intersections is odd, this indicates that the calculated normal vector for the element is inward pointing, and a flag is set to reverse the surface normal direction in future calculations for that element. o Tabulate Cone Flow Solutions: For the Tangent-Cone aerodynamic analysis model, a cone flow is assumed at each surface element. Due to the variety of surface inclinations that will be encountered during analysis of a given shape, a table of analytical cone flow solutions is generated for a large set of cone half-angles (for each of the analysis Mach numbers and altitudes) and stored in memory. Analysis Phase an iteration over each analysis case (Mach number, altitude, angle of attack) is performed. o Rotate Mesh Geometry: The current version of FAAT assumes that the freestream flow is always in the positive x-direction. Therefore for each analysis case, the surface mesh is rotated to the appropriate pitch, roll, and yaw orientation. o Calculate Surface Normals/Tangents: Using the rotated mesh, the surface normal vectors and flow tangency vectors are calculated. The results from the Flag Mesh Interior/Exterior step in the preprocessing phase are used to ensure that the surface normals are consistently outward pointing. o Flag Shadow and Base Panels: The projected area of each surface element is checked to identify panels that are shadowed in the freestream direction by other elements; these shadow panels are flagged appropriately. During the aerodynamic analysis process, shadow panels that have a surface inclination greater than a specified value (e.g. 45 deg.) are flagged as base surfaces. o Integrate Streamlines: The streamline leading to each surface element is calculated by tracing the flow tangency vectors upstream until a leading edge or stagnation point is reached. (Section III provides more details on this tracing process.) o Calculate Freestream Properties: As the user defines a set of altitudes and Mach numbers to analyze, a standard atmosphere model [2] is incorporated to generate the freestream properties (e.g. temperature, density) at a given altitude. o Calculate Panel Properties: For each surface element, the aerodynamic properties (e.g. pressure, density, Mach number, temperature, local velocity) are calculated using the appropriate aerodynamic model. The streamline tracing results are then applied to calculate the viscous skin friction coefficient for each panel. o Integrate Forces and Moments: The aerodynamic forces and moments are calculated on each surface element, and then integrated over the entire shape to provide the overall aerodynamic performance. 4

Table 1 Summary of input settings/parameters that can be specified by the user Input Model Settings Comments Aerodynamic Model Setting Choose between Newtonian, Tangent Cone, Shock-Expansion, or Averaged 1 Viscous Model Settings Choose between Laminar, Turbulent, or Transitional 2 Wall Settings Flag to Calculate Stability Derivatives Flow Conditions Mach Number Angle of Attack Altitude Roll Angle Yaw Angle Shape Parameters Reference Area Reference Length Surface Refinement Parameter Center of Gravity Location Choose between Adiabatic or Isothermal (with specified wall temperature) Choose between True or False Specify min, max, and number of cases total (including min/max) Specify min, max, and number of cases total (including min/max) Specify min, max, and number of cases total (including min/max) Specify value Specify value Specify value corresponding to input geometry Specify value corresponding to input geometry Specify maximum allowed edge length of triangular elements Specify Cartesian coordinates corresponding to input geometry 1 "Averaged" indicates that the results of the Newtonian, Tangent Cone, and Shock-Expansion methods are averaged for each condition analyzed 2 "Transitional" evaluates an empirical relationship on a per-panel basis to determine whether to use a laminar or turbulent analysis method After the pre-processing phase, FAAT will iterate over each of the angles of attack and flow conditions specified and evaluate the aerodynamic forces and moments at each. This allows the user to specify a range of Mach numbers, altitudes, and angles of attack to perform analysis for the tool will then proceed through the analysis, and tabulate the data in several output files. At various points during the analysis, the surface data (e.g. pressure, density, Mach number contours) can be written to a Tecplot file to allow the user to see various surface properties on the vehicle. (While Table 1 notes the user input for the aerodynamic model, the analysis actually is performed using all models the user input only determines which model results will used to write the surface contour data.) If the user has selected the option to calculate the stability derivatives, these will be calculated numerically by offsetting the pitch, yaw, and roll angles with small deltas to determine the longitudinal and lateral-directional stability derivatives. A. Inviscid Flow Analysis There are several engineering-level analysis techniques frequently used in the preliminary estimation of the aerodynamic performance of supersonic or hypersonic vehicles. These include Modified Newtonian flow, Tangent- Cone, and Shock-Expansion theory. All of these techniques are inviscid approaches that can be used to approximate the pressure distribution on a body in hypersonic flow. While an engineer might typically apply one of these techniques to quickly get an estimate of the pressure forces acting on a vehicle, the present work combines all three techniques, along with the assumption of a perfect gas and isentropic flow, to obtain relevant flow properties on the vehicle surface. Modified Newtonian flow theory is a technique that assumes the flow impacts the surface and immediately changes direction to correspond to the surface inclination to the flow. In this case, the pressure coefficient on the surface can be estimated as [3] = sin (1) where represents the inclination angle of the surface to the freestream flow. The maximum value of the pressure coefficient ( ) can be evaluated assuming it lies at a stagnation point behind a normal shock wave = (2) 5

where represents the stagnation pressure behind a normal shock. In Eq. (2), is the freestream pressure, is the freestream density, and is the freestream velocity. For any panels that are shadowed by another (relative to the freestream velocity vector), Newtonian flow assumes these are at freestream conditions. The Tangent-Cone approach is another frequently used engineering analysis method. It assumes that the pressure on the surface is equivalent to a cone with a half-angle equal to the surface inclination. For this approach, the analytical solution for a cone in supersonic or hypersonic flow can be represented by the Taylor-Maccoll equation [4]: 1 +cot +2 =0 = (3a) (3b) Equation (3) is written in spherical coordinates (with origin at the nose of the cone), where and are the velocities in the and directions, respectively, and is the local speed of sound. As Eq. (3) is an ordinary differential equation, it can be numerically integrated to obtain the flow properties on the cone surface. In the present work, Eq. (3) is integrated using a 4 th -order Runge-Kutta technique, and the results tabulated, for a range of cone angles to use in applying the tangent cone flowfields to each surface panel of the geometry. Similar to Modified Newtonian theory, Tangent-Cone analysis assumes that shadowed panels are at freestream conditions. The Shock-Expansion technique is a third aerodynamic analysis method, where the analytical oblique shock and Prandtl-Meyer expansion wave relations [5] are used to determine the surface properties. This technique requires following a streamline on the vehicle surface, and the properties are updated depending on the inclination of each segment of the streamline. The process is implemented by starting at the beginning of a streamline leading to a given panel, and calculating the change in inclination angle ( ) between each segment based on the surface normals of the panels traversed by the streamline: = =sin ( ) where p represents the index of the current segment of the streamline, and is the outward-pointing surface normal vector for the surface element being traversed by the streamline segment. For an oblique shock wave (positive ), the properties on the panel are calculated by solving the --M relation implicitly for the shock angle based on the inclination angle change and Mach number ahead of the shock ( ): tan =2cot For this purpose, Brent s algorithm [6] is used to obtain based on a given. The remaining flow properties are calculated on the panel using the standard oblique shock relations for a perfect gas. For an expansion wave (negative ), the properties on the panel are calculated using Prandtl-Meyer expansion theory: (4a) (4b) (5) Δ = (6a) ()= tan ( 1) tan 1 (6b) As the expansion angle ( ) and Mach number ahead of the expansion ( ) are known, Eq. (6) is solved implicitly (again using Brent s algorithm) to obtain the Mach number after the expansion ( ). Perfect gas and isentropic flow relations are used to obtain the other properties after the expansion. 6

B. Streamline Tracing An important part of the analysis process is the tracing of streamlines. This is required for both the Shock- Expansion technique, as well as the reference temperature method employed to estimate viscous effects. For this purpose, the flow tangency vectors on each panel are calculated by = (7) where is the local velocity vector (i.e. flow tangency vector) on the panel with index, is the freestream velocity vector, and is the surface normal of the panel. (Note that the flow tangency vectors only approximate the streamline directions, as boundary layer effects are not accounted for.) Once the flow tangency vectors are obtained, the approximate streamlines are created for each panel by tracing the flow tangency vectors backwards from the panel center. This is done by calculating the intersection of the flow tangency vector with the edges of the current panel, and then using barycentric coordinates to determine which of the neighboring panels has an edge coincident with intersection point. This process is repeated to trace the streamline backwards to a leading edge or stagnation point of the vehicle. This point is identified when the tracing process results in the streamline x-direction component reversing such that it is in the same direction as the freestream velocity vector, or when two segments of the streamline have a change in angle greater than a specified threshold (e.g. 30 deg.). This process results in the determination of a streamline leading to each panel on the surface, and is stored in memory. It should be noted that the surface normals of the panels traversed by each streamline are also stored in memory these are used in determining whether to apply oblique shock or expansion wave analysis for the Shock- Expansion method as shown in Eq. (4). C. Skin Friction Estimation For slender vehicles, the skin friction can be a major component of the drag on the configuration; therefore Eckert s reference temperature approach is employed [7]. Equations for both laminar and turbulent flow are available to provide results depending on the specific vehicle configuration and freestream conditions. In FAAT, the local properties on each surface element (including the streamline length leading to each panel) are used to generate a local skin friction coefficient. As the reference temperature approach is based on two-dimensional flat plate assumptions, the Mangler fraction ( 3) is multiplied against the skin friction coefficient to account for the relieving effects present in three-dimensional flow. This skin friction coefficient is then applied to the local flow tangency vector to calculate the viscous force components on a given panel. D. Other Implementation Considerations The Modified Newtonian method can be applied to any geometry without restrictions, as the flow properties are either determined by Eq. (1), or are at freestream conditions for shadowed panels. However, the Tangent-Cone and Shock-Expansion methods do have limitations in their application based on the inclination of the panel being analyzed. For example, at large cone half-angles, the shock wave will detach from the tip of a cone, and the analytical theory described in Eq. (3) no longer holds. Similarly, a two-dimensional oblique shock will also detach at sufficiently large inclination angles. Therefore, the Tangent-Cone and Shock-Expansion aerodynamic models implemented in FAAT include switches to blend to Modified Newtonian flow when the inclination angle exceeds a specified maximum (e.g. 80% of the maximum possible surface angle for an oblique shock wave). During the analysis phase, a check is performed to identify base regions on the geometry. These base surfaces are determined as shadowed panels that have an inclination angle greater than a specified threshold (e.g. 45 deg.). The base regions are analyzed using Gabeaud s relationship for base pressure [8]: =.. () 7 1 (8) E. Comparison with Other Analytical Tools Other tools exist which employ similar analysis techniques for rapid estimation of hypersonic vehicle aerodynamic coefficients. In particular, the NASA-developed Configuration Based Aerodynamics (CBAERO) tool [9] is comparable in many ways to the FAAT program discussed in the present work. While both tools are capable of applying engineering-level techniques to estimate aerodynamics of high-speed vehicles, the implementation and usage framework are different between the two tools. Specifically, CBAERO is tailored around a Graphical User

Interface (GUI) used to set up the analysis, where the user can specify the base, body, and wing regions (for example), as well as select options to analyze TPS or control surfaces. On the other hand, FAAT is a relatively simple executable that relies on only two inputs: a short text file describing the analysis conditions and vehicle configuration, and a grid file (i.e. a Tecplot data file) describing the unstructured surface mesh. FAAT employs techniques to identify the base region (for example), which automates some of the work that a user might perform in setting up an analysis in the CBAERO GUI. While CBAERO provides many capabilities beyond the current version of FAAT (e.g. TPS analysis/optimization, control surface analysis, and subsonic aerodynamics analysis), the simple script-driven process used by FAAT lends itself to easy integration into an MDO environment, as will be demonstrated in Sections IV and V. III. Aerodynamic Analysis - Validation For any new analytical tool, validation of its accuracy is required. Therefore, the NASA HL-20 lifting body configuration was used in the present work to compare results generated by FAAT s implementation of different aerodynamic models to published wind tunnel and computational results. A surface mesh of this vehicle (see Fig. 2) is included as a sample case in the CBAERO distribution provided by NASA this mesh was converted to Tecplot format using a CBAERO-supplied utility, which provided the surface mesh input for the validation analyses conducted with FAAT. Fig. 2 Surface mesh for HL-20 configuration. In addition to the CBAERO results published in Ref. [9] for this configuration, NASA has conducted substantial subsonic through hypersonic wind tunnel testing of the HL-20 configuration at various facilities. In particular, a series of transonic and supersonic tests were performed in the NASA Langley Unitary Plan Wind Tunnel using a 7% scale model (Ref. [10] and [11]), and hypersonic tests were performed in the NASA Langley 20-inch (Mach 6) and 31-inch (Mach 10) tunnels using a 2% scale model (Ref. [12] and [13]). (Ref. [12] also includes inviscid CFD results for Mach 6 and 10 as well.) Using the data from these sources, comparison cases were selected at Mach numbers of 1.2, 4.5, 6, and 10. While FAAT typically uses Mach number and altitude to determine the freestream conditions in an analysis, modifications were made for these comparison cases to allow specification of the actual wind tunnel conditions for each test, as described in the references. Finally, based on the test descriptions in the references, turbulent flow was specified for all cases. A. Importance of Lee-Side Flow Estimation Initially, the Modified Newtonian and Tangent-Cone methods were run assuming freestream conditions on the lee-side shadowed panels, as discussed in Section II. However, upon inspecting the results, it was observed that while the Shock-Expansion method matched the wind tunnel data reasonably well, the Modified Newtonian and Tangent-Cone techniques predicted significant differences in the aerodynamic coefficients at lower Mach numbers in the transonic/supersonic regime (see Fig. 3 for a comparison of these initial results for the Mach 1.2 and Mach 10 drag coefficient). Based on this, FAAT was then run with the option enabled to analyze the shadowed panels for the Modified Newtonian and Tangent-Cone methods using the Shock-Expansion technique (rather than assuming freestream conditions). This resulted in a much better match with experimental results, and therefore subsequent 8

results were run using this option. This provides evidence that while the simple assumption of freestream conditions for shadowed regions is reasonable at hypersonic speeds, a more sophisticated approach capturing some of the flow physics in this region is required to improve the accuracy of these analytical techniques at low supersonic Mach numbers. Drag Coefficient 1.6 1.4 1.2 0.8 0.6 0.4 FAAT - Modified Newtonian FAAT - Tangent Cone FAAT - Shock/Expansion FAAT - Averaged CBAERO Wind Tunnel 0.2 Mach 1.2 Conditions Mach 10 Conditions Fig. 3 Drag coefficient comparison at Mach 1.2 and Mach 10, where Modified Newtonian and Tangent-Cone methods assumed freestream conditions for shadowed regions. B. Comparison of Results The results from running FAAT for the Mach 1.2, 4.5, 6, and 10 cases are shown in Fig. 4 through Fig. 7; it can be observed that in general, the various analysis methods described in Section II exhibit good agreement with the wind tunnel data for all four cases. While the results for the three analytical techniques (Modified Newtonian, Tangent-Cone, and Shock-Expansion) are shown in each figure, the following discussion will refer to the Averaged values as the baseline predicted by FAAT unless otherwise noted. For all four Mach numbers, the lift and drag coefficients calculated using FAAT appear to match the wind tunnel data with differences of approximately 15% at most. In general, FAAT appears to slightly overpredict the drag at lower angles of attack, and somewhat underpredict the lift at higher angles of attack. Due to these differences in the force coefficients relative to the wind tunnel data, the L/D predicted by FAAT is typically slightly under that based on the wind tunnel experiments. The Mach 4.5 results show the best agreement over the full range of angle of attack for the cases analyzed, whereas for the Mach 6 and Mach 10 cases FAAT underpredicts the maximum L/D of the HL-20 on the order of 10-15%. While this difference is not insignificant, FAAT demonstrated the ability to predict the aerodynamic force coefficients and L/D with an accuracy similar to that of CBAERO (Mach 1.2 and 4.5) and the inviscid CFD (Mach 6 and 10). Looking at the pitching moment results, it can be seen that for the Mach 1.2 transonic case, FAAT is able to provide a reasonable approximation up to 15 deg. angle of attack. Both FAAT and CBAERO are unable to predict the measured rise in pitching moment that occurs at moderate angles of attack in the wind tunnel data at this Mach number. For the Mach 4.5 case, the pitching moment coefficient estimated by FAAT follows a trend similar to the experimental data through the entire angle of attack range. The Mach 6 and 10 cases show good agreement of FAAT with the wind tunnel data at medium angles of attack (e.g. 15-40 deg.), but there is a significant underprediction of the pitching moment at lower angles of attack. It should be noted that the transonic and supersonic wind tunnel experiments plotted in the results were based on the HL-20 without the center fin (which corresponds to the mesh shown in Fig. 2). Based on the results presented in Ref. [10] and [11], it can be observed that while the lack of a center fin appears to have a small effect on the results, some of the largest differences do appear to be a slight decrease in pitching moment at lower angles of attack. As the Mach 6 and 10 wind tunnel model had the center fin for all tests, this could potentially be the source of some of the discrepancy in pitching moment at lower angles of attack for these two Mach number cases. Finally, comparing the analytical aerodynamic analysis approaches, it can be seen that in general the different models produce similar results, especially for the Mach 1.2 case. At higher Mach numbers (4.5, 6, and 10) the Shock-Expansion technique tends to underpredict the lift and drag coefficients, and slightly overestimate the pitching moment, relative to the other models. As was discussed above, both the Modified Newtonian and Tangent- Drag Coefficient 1.2 0.8 0.6 0.4 0.2 FAAT - Modified Newtonian FAAT - Tangent Cone FAAT - Shock/Expansion FAAT - Averaged CFD (Inviscid) Wind Tunnel (1990) Wind Tunnel (1999) 9

Cone methods are using Shock-Expansion analysis on the shadowed lee-side regions while this collapses all the results together for the Mach 1.2 case, the higher Mach numbers show a larger difference in results between the different methods, providing further evidence that the windward side flowfield has a greater influence (relative to the lee side) on the aerodynamic coefficients as Mach number increases. Overall, these results indicate that FAAT s implementation of the Modified Newtonian, Tangent-Cone, and Shock-Expansion methods is able to predict the aerodynamic coefficients of the complex HL-20 configuration with reasonable accuracy, for conditions ranging from transonic to hypersonic speeds. It is also interesting to see the good agreement that is obtained for the transonic case the types of analytical methods discussed in Section III are typically considered to increase in accuracy at higher Mach numbers, yet application of the Shock-Expansion technique to the lee-side flow region demonstrated the ability to provide a reasonable estimate of the force coefficients even at Mach 1.2. Considering that the agreement between FAAT and the wind tunnel data is approximately equivalent to other numerical techniques (CBAERO and inviscid CFD), the ability of FAAT to provide a rapid first-order estimate of the aerodynamics of a given complex shape was verified. L/D Lift Coefficient 1.2 0.8 0.6 0.4 0.2-0.2-0.4 2.5 2.0 1.5 0.5-0.5 - -1.5-2.0 Pitching Moment Coefficient Drag Coefficient 1.6 1.4 1.2 0.8 0.6 0.4 0.2 FAAT - Modified Newtonian FAAT - Tangent Cone FAAT - Shock/Expansion FAAT - Averaged CBAERO Wind Tunnel 6 4 2 0-2 -4-6 -8-0.10-0.12-0.14 Fig. 4 Aerodynamic analysis results for HL-20 at Mach 1.2 (ReL = 3.4 10 6 ). 10

Lift Coefficient 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1-0.1-0.2 Drag Coefficient 1.2 0.8 0.6 0.4 0.2 FAAT - Modified Newtonian FAAT - Tangent Cone FAAT - Shock/Expansion FAAT - Averaged CBAERO Wind Tunnel L/D 2.0 1.5 0.5-0.5 - -1.5 Pitching Moment Coefficient 2 1 0-1 -2-3 -4-5 -6-7 -8 Fig. 5 Aerodynamic analysis results for HL-20 at Mach 4.5 (ReL = 3.4 10 6 ). 11

Lift Coefficient 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1-0.1-0.2 2.0 Drag Coefficient 1.2 0.8 0.6 0.4 0.2 FAAT - Modified Newtonian FAAT - Tangent Cone FAAT - Shock/Expansion FAAT - Averaged CFD (Invscid) Wind Tunnel 2 L/D 1.5 0.5-0.5 - -1.5 Pitching Moment Coefficient 1 0-1 -2-3 -4-5 -6-7 Fig. 6 Aerodynamic analysis results for HL-20 at Mach 6 (ReL = 0.3 10 6 ). 12

Lift Coefficient 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1-0.1-0.2 Drag Coefficient 1.2 0.8 0.6 0.4 0.2 FAAT - Modified Newtonian FAAT - Tangent Cone FAAT - Shock/Expansion FAAT - Averaged CFD (Inviscid) Wind Tunnel (1990) Wind Tunnel (1999) L/D 2.0 1.5 0.5-0.5 - Pitching Moment Coefficient -7 Fig. 7 Aerodynamic analysis results for HL-20 at Mach 10 (ReL = 10 6 ). C. Computational Time Discussion The surface mesh shown in Fig. 2 contains 13,546 surface elements. For each Mach number case discussed above, analysis was conducted for an angle of attack range between -5 and 50 deg. in 2.5 deg. increments this resulted in 23 iterations during the analysis phase for each case. The analysis was conducted on a circa 2010 notebook computer with an Intel Core i7-620m processor (containing two cores running at 2.67 GHz each, along with Intel Hyper-Threading technology to allow four threads to operate simultaneously) OpenMP detected this as four processing units, and therefore parallelized the computations over four threads. On this specific system, each Mach number case took approximately six minutes to run, including all input, analysis, and output steps. It should be noted that one of the primary drivers in computational time is the geometry analysis at the start of each angle of attack iteration in the analysis phase. For the results discussed above, each Mach number case was run separately, as FAAT was modified to allow specification of the freestream conditions for each wind tunnel test. In a more typical analysis where a range of altitudes and Mach numbers are specified in the input file, FAAT would analyze the different altitude and Mach number conditions while iterating through each angle of attack. This would result in a moderate increase in compute time for additional freestream conditions. Additionally, it is recognized that replacing the mesh rotation and surface normal recalculation with rotations of the freestream velocity and surface normal vectors should decrease the computational time, and may be implemented in future versions of the tool. 3 2 1 0-1 -2-3 -4-5 -6 IV. CRV Shape Optimization Model Overview A. Objectives and Mission Definition As the ability of FAAT to rapidly generate reasonably accurate supersonic/hypersonic aerodynamic databases for arbitrary shapes was demonstrated above, an MDO framework was developed to highlight the utility of this type of analytical tool in preliminary aerodynamic design. For this purpose, the HL-20 was used to develop a baseline set of mission requirements, including the requirement to carry a cylindrical payload representing the pressurized crew 13

compartment (assumed to be a fixed 1.8 m diameter by 3 m length size, based on obtaining a habitable crew volume similar to the HL-20 conceptual design). [14] The HL-20 was developed as a lifting body concept for a CRV with moderate L/D. Therefore, it was decided that investigating the use of waveriders would provide an interesting contrast to see how their typically high design L/D would translate to a reentry vehicle where substantial shape modifications are required. B. Waverider-Based Design Approach Waveriders are typically designed using an inverse design approach, where the lower surface is created by tracing streamlines in the generating flowfield backwards until the shock wave is reached this determines the leading edge of the vehicle. Idealized waveriders are then created by tracing from the leading edge in the freestream direction to create the upper surface. Frequently hypersonic wedge or cone flowfields are used to create waveriders, as the two-dimensional or axisymmetric generating flowfields can be easily solved using analytical techniques. The osculating cone method [15] is another technique that is popular for waverider design it is a strip method that scales a single cone flow solution based on the local curvature of the shock wave in the base plane. (For regions of zero curvature, the osculating plane corresponds to a cone flow of infinite radius, and therefore is equivalent to the flow past a twodimensional wedge.) This allows some control over the shock wave shape, while at the same time requiring only a single cone flow solution. Fig. 8 indicates the variables used in the present work to specify both the shock profile curve (SPC) and lower base curve (LBC) used in the waverider design 4 control points are used to specify the LBC, and the SPC is created by dividing the shock wave into wedge and cone regions as shown. Fig. 8 Parameters used to specify waverider SPC and LBC (in waverider base plane). A CRV design tool was created that utilizes several parametric inputs to develop a payload-optimized waverider design for a given mission. While previous payload-optimized waverider design approaches have looked at fitting the payload inside an idealized waverider [16], the present work instead focuses on modifying the upper surface of the waverider to fit the payload. This is one of the advantages of using an inverse design approach for waveriders, as the lower surface is used to recreate the shock wave in the generating flowfield, allowing the upper surface to be tailored to mission requirements without affecting the lower surface s ability to reproduce the generating flowfield. The design tool divides the waverider into forebody and aftbody sections, where the aftbody section contains the cylindrical payload. Modifications are made to the upper surface to wrap the body around the payload for the aftbody section; a minimum drag Haack ogive is used to blend the upper surface of the forebody section to the modified aftbody section. Finally, the forebody upper surface near the nose was smoothed by blending an elliptical cross-section (near the nose tip) to the Haack ogive-generated cross section (near the end of the forebody section); this also has the benefit of providing additional volume near the front of the vehicle. One potential issue with the use of waveriders in realistic hypersonic missions is that the nearly-flat lower surface typically creates a pitching moment slope that has minor variation with respect to angle of attack this can make it difficult to design a vehicle that is statically stable without requiring a large trim force. Therefore, in the present work the aftbody section of the waverider lower surface is rotated such that it is parallel to the freestream direction at the vehicle centerline. This angle break means that the aftbody is no longer a waverider surface in the traditional sense; however, the reduced inclination of a substantial part of the lower surface can allow for axial CG 14

locations to be selected that will create a longitudinally stable vehicle with minimal trim requirements at maximum L/D, as shown in Fig. 9. (This example also shows that not including the angle break makes it difficult to create a vehicle with these characteristics, even if the axial center of gravity location is varied substantially.) Considering that the reentry vehicle may fly at most a small portion of the trajectory at the waverider design conditions, and that significant rounding of the nose and leading edges will be required to mitigate heating, the deviation from an idealized waverider surface was deemed acceptable. In fact, it is interesting to observe in Fig. 9 that the CRV design with the angle break (and modified lower surface) actually produces a slightly higher maximum L/D than the two designs that have an unmodified lower surface. Fig. 9 Example of L/D and pitching moment coefficients for CRV designs with and without modified lower surfaces (assumes M = 10, altitude = 50 km, laminar flow). Leading edge bluntness is applied to the CRV designs using several steps: 1. First, the upper and lower surfaces are separated by a user-specified vertical distance corresponding to the desired nose tip thickness. Recognizing that the heat flux will be less at the leading edges for a given bluntness due to sweep effects, the vertical thickness is reduced to a leading edge value after a defined percentage (e.g. 20%) of lengthwise distance from the nose tip. The upper surface region (between the leading edge and the junction of the wing/body regions) is then blended gradually to the reduced leading edge height. 2. Next, the rounded leading edge surface is initially created by generating a bi-arc curve between the upper and lower surfaces. The local sweep angle is used to rotate the bi-arc curve such that the correct slopes are used at the ends, providing a smooth blend between the upper/lower and leading edge surfaces. 3. Finally, as the bi-arc technique can lead to a jagged leading edge surface depending on local sweep angles and upper/lower surface slopes, the blunted leading edge is recreated using a cubic spline technique (again, considering the sweep angle in determining the slopes at the upper/lower surfaces). The stagnation point 15

location with respect to the upper/lower surfaces is taken from the bi-arc curve generated in step 2. To ensure a smooth blend along the CRV, the stagnation point location (with respect to the upper/lower surfaces) is interpolated between the nose tip and the end of the forebody section based on the local leading edge thickness. Finally, wingtip fins are integrated into the CRV design by specifying a sweep angle (which can potentially provide increased lateral stability), a dihedral angle, and the fin spanwise length in the base plane. The fins are created as triangular surfaces that extend from the start of the CRV aftbody section using the specified sweep and dihedral angles. C. Mass Estimation A simple mass model was created based on the mass breakdown table listed in Ref. [17] for the HL-20 configuration. The results for both contractor designs were averaged to calculate a baseline HL-20 landed mass of 10,337 kg, a thermal protection system (TPS) mass of 973 kg, and a total takeoff mass of 16,303 kg (which includes crew, consumables, a launch escape system, and launch vehicle adapter). These results were used to derive linear mass estimating relationships based on the HL-20 planform area of 469.3 kg/m 2 (landed mass, excluding TPS) and 765.0 kg/m 2 (takeoff mass, excluding TPS). These relationships are applied to determine the takeoff and landing mass for a given CRV design, excluding the TPS mass. The TPS mass is estimated assuming the same Shuttle-class materials used in the HL-20 design. This consists of Advanced Carbon-Carbon (ACC) with an oxidation inhibitor coating for the nose and leading edge regions, high temperature reusable surface insulation (HRSI) tiles for the lower surface, and flexible reusable surface insulation (FRSI) blanket material for the upper surface. For the CRV designs considered in the present work, the material densities and maximum reuse temperatures [18] as well as the assumed thicknesses for each region are shown in Table 2. The surface areas for each of the vehicle regions is output from the CRV shape design code discussed previously, and multiplied by the data in Table 2 to determine an overall vehicle TPS mass. Table 2 TPS materials, properties, and assumptions. Assumed Density, Maximum Reuse Vehicle Region Material Thickness, cm kg/m 3 Temperature, K Nose and leading edges ACC a 1.27 1600 1922 Lower surface acreage HRSI b 2.03 352 1644 Upper surface acreage FRSI 7.62 96 922 a Density for ACC only; assumes mass of oxidation inhibitor coating is negligible b Use of LI-2200 assumed D. Trajectory and Aeroheating Analysis In order to assess the mission performance of the CRV designs, a 3-degrees-of-freedom trajectory model was developed to integrate the equations of motion for a point mass around a spherical Earth [19] using a 4 th -order Runge-Kutta technique; two-dimensional flight only is considered in the current analysis. The initial conditions include a reentry interface at an altitude of 122 km and 7620 m/s based on the baseline HL-20 trajectory [20]; a reentry flight path angle of -1.5 deg. is also assumed. The trajectory control variables include the initial angle of attack at reentry (constrained to be between -5 and 40 deg.), as well as constant pitch rates (constrained between ± 2 deg./s) defined at 250 s intervals. The angle of attack was also constrained to be between -5 and 40 deg. throughout the trajectory. The desired trajectory end state (again based on the HL-20 baseline trajectory) is an altitude of 18.3 km and velocity of 365 m/s, which corresponds to approximately Mach 1.2. A hybrid optimization technique was implemented by combining a genetic algorithm with a hill-climbing technique as discussed in Ref. [21]; the control variables were varied to maximize the following objective function: F =( ) ( ) 16 (9) where is the maximum range at the end of the trajectory, is the max acceleration (in g s) experienced by the crew,,, and correspond to the maximum heat fluxes experienced in the nose tip, leading edge, and lower surface acreage regions, and and h correspond to the calculated end state Mach number and altitude (with the desired values of each are annotated with the appropriate subscript). The objective function

weights are set to =1 and ====10, as preliminary investigations indicated that these values were effective in maximizing range while meeting trajectory constraints. Note that and are set to zero when the acceleration and heat flux values are below the constraints specified (e.g., 5 g s for, and 90, 90, and 20 W/cm 2 for the stagnation, leading edge, and acreage heat fluxes, respectively). Finally, the weights and are set negative when the end state value exceeds that desired this provides a mechanism to force the trajectories to the specified end state. Approximate methods for estimating the heat flux on a stagnation point, swept leading edge, and flat plate are implemented based on the equations described in Ref. [22]. For the CRV designs considered in the present work, the average radius of curvature for the stagnation point and leading edge is calculated based on the cubic-splined surface. For the acreage region (assumed to be at the vehicle lower surface centerline), either fully laminar or fully turbulent flow is assumed based on the empirical relationship for a transition Reynolds number discussed in Ref. [1] and the freestream conditions during each point in the trajectory. The leading edge and acreage heating analyses are conducted at the 20% lengthwise distance location only to provide representative results. E. Aerodynamic Analysis The performance of idealized waveriders is typically estimated at cruise conditions by applying the generating flowfield properties to the lower surface, and freestream conditions to the upper surface. For the waverider-derived CRV designs in the present work, however, substantial modifications to the waverider shape, as well as the large variation in flow conditions experienced during a reentry trajectory, mean that an alternative approach is needed to assess the aerodynamic performance of the designs. FAAT is therefore used in the present work to generate a table of aerodynamic coefficients for each waverider-derived CRV design. To provide a reasonable compromise between computational time and database precision, the aerodynamics are assessed for an altitude of 15 to 125 km (in 5 evenly-spaced steps), Mach numbers of 1.2 to 25 (in 5 steps), and angles of attack of -5 to 40 deg. (in 5 deg. increments). This results in a total number of 250 conditions for each table; currently laminar conditions are assumed for all aerodynamic coefficients for the sake of simplicity. Trilinear interpolation is used to generate aerodynamic coefficients that lie between specific conditions used to create the tables. F. Optimization Framework Integration The CRV design tool, FAAT, the mass model, and the trajectory model are integrated into an optimization framework using the Phoenix ModelCenter environment as shown in Fig. 10. A multi-objective optimization was implemented using ModelCenter s Darwin genetic algorithm (GA), where the optimized trajectory range was maximized, and the CRV landed mass was minimized. (Note that as shown in Fig. 10, an inner loop trajectory optimization is run for each iteration of the main optimization loop.) The CRV design tool variables that are varied during the main optimization process, along with minimum and maximum ranges on each, are show in Table 3. Shape Design Tool Modify Design Variables Constrained Design? Yes No Aero Analysis (FAAT) Mass Model Main Optimization Loop Trajectory Optimization Inner Optimization Loop Mark as Constrained Design Yes Trajectory/Mass Constrained? No Mark as Feasible Design Fig. 10 Flowchart of CRV optimization framework. 17

Table 3 List of CRV design variables and min/max constraints for each. Variable Min Value Max Value Comments Length, m 6 10 Total length of vehicle Waverider Design Mach Number 5 15 Mach number used in waverider design Waverider Design Altitude, km 30 60 Altitude used in waverider design Waverider Shock Wave Angle, deg. 8 20 Shock wave angle for cone flow used in waverider design z SPC2 0.5 Non-dimensional coordinate of SPC used in waverider design z SPC3 0.5 Non-dimensional coordinate of SPC used in waverider design R SPC 0.8 4.0 Non-dimensional radius of SPC used in waverider design z LBC2 0.5 Non-dimensional coordinate of LBC used in waverider design z LBC3 0.5 Non-dimensional coordinate of LBC used in waverider design y LBC2-0.2 0.2 Non-dimensional coordinate of LBC used in waverider design y LBC3-0.2 0.2 Non-dimensional coordinate of LBC used in waverider design y LBC4-0.2 0.2 Non-dimensional coordinate of LBC used in waverider design Nose thickness, m 0.3 Vertical distance between upper/lower surfaces at nose tip Leading edge thickness, m 0.1 0.3 Vertical distance between upper/lower surfaces at leading edges Fin sweep, deg. 0 15 Sweep angle of wingtip fins Fin dihedral, deg. 30 80 Dihedral angle of wingtip fins Fin height, m 0.8 1.6 Spanwise length of wingtip fins (in base plane) While the design tool is robust, there are many combinations of design variables that can lead to infeasible designs (for example, the user-specified control points that determine the waverider shape could be located outside the generating flowfield for a given design iteration). The tool therefore has an output that indicates whether it succeeded in designing a given configuration. To aid integration into an optimization process, this output is further refined by outputting either a zero (if the design was feasible), or a value inversely proportional to how many steps the tool was able to complete before reaching an infeasible design state. This value is then included in the optimizer as a constraint, which helps drive the results to a feasible design space as the optimization progresses. Several additional constraints are implemented to generate feasible designs that meet mission requirements. The CRV designs are restricted to a maximum length and width of 10 m and 7 m, respectively, to allow generation of a configuration that could potentially be integrated onto a modern launch vehicle such as the Atlas V or Delta IV. Similarly, the takeoff mass of the vehicle is required to be no more than 26,000 kg, which approximates the maximum Low-Earth Orbit payload capability for a Delta IV Heavy class launch vehicle. Finally, the trajectory g- limit and heat flux constraints discussed previously are enforced as well. V. CRV Shape Optimization Results A. Optimization Results The GA was run for 122 generations (approximately 8600 function evaluations), and was terminated when 15 generations had passed with no change in the pareto front. The pareto front identifies the non-dominated designs found by the GA as shown in Fig. 11 (this figure also shows all feasible designs generated during the optimization). Interestingly, the results indicate that there is a trend for higher mass vehicles to have longer range this implies that effects such as aerodynamics or trajectory constraints may be more favorable for heavier vehicles in increasing range. This is reinforced by plotting the range and landed mass for the pareto front designs as a function of maximum L/D for each Fig. 12 indicates that in general there is a correlation between higher L/D and higher mass / longer glide range. 18

13,000 12,000 Pareto Front Mid-Mass/ Mid-Range Max Range Range, km 11,000 10,000 Min Mass 9,000 Dominated Designs 8,000 12,000 13,000 14,000 15,000 16,000 17,000 Landed Mass, kg Fig. 11 Pareto front generated by waverider-derived CRV optimization framework. Range, km 13,000 12,500 12,000 11,500 11,000 10,500 10,000 9,500 9,000 8,500 R² = 0.849 8,000 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 L/D R² = 0.8427 12,000 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 L/D Fig. 12 Range and Landed Mass vs. maximum L/D for pareto front designs. B. Comparison with the HL-20 Three-dimensional and plan views of the three configurations identified in Fig. 11 as corresponding to Min Mass, Max Range, and Mid-Mass/Mid-Range designs are shown in Fig. 13. It can be seen that the optimizer is able to find different configurations that meet each objective; for example, the Mid-Mass/Mid-Range design exhibits a much smaller dihedral angle relative to the other two waverider-derived CRV designs. Table 4 also provides a comparison of some of the key characteristics of these three designs with the HL-20. From a size perspective, it is interesting to note that the Min Mass design is smaller than the HL-20 configuration, while the Max Range design is only 0.3 m longer than the HL-20. Finally, Fig. 14 shows plots of the L/D for the three waverider-derived CRV designs as compared with the HL-20, where the performance was assessed for all four vehicles using FAAT. As can be seen from plots, the Min Mass and Mid-Mass/Mid-Range waverider-derived designs have a maximum L/D less than or equal to the HL-20; however the Max Range design demonstrates an approximately 20% increase in maximum L/D relative to the HL-20 design. Landed Mass, kg 17,000 16,500 16,000 15,500 15,000 14,500 14,000 13,500 13,000 12,500 19

L/D 2.0 1.5 0.5-0.5 - Fig. 13 Min Mass, Mid-Mass/Mid-Range, and Max Range designs from waverider-derived CRV optimization. Laminar -1.5 L/D 2.0 1.5 0.5-0.5 Min Mass Mid-Mass/Mid-Range - Max Range Turbulent HL-20-1.5 Fig. 14 Comparison of L/D for waverider-derived CRV designs and the HL-20 configuration (assumes M = 10, altitude = 50 km for laminar; M = 5 and altitude = 30 km for turbulent). 20