Geometric Model for a Parametric Study (cid:3) of the Blended-Wing-Body Airplane y C. Wayne Mastin z Robert E. Smith x Ideen Sadrehaghighi { Michael R. Wiese ABSTRACT NOMENCLATURE A parametric model is presented for the blended- D0;D1 Dirichlet boundary conditions wing-body airplane, one concept being proposed N0;N1 Neumann boundary conditions for the next generation of large subsonic trans- P Point on NURBS surface ports. The model is de(cid:12)ned in terms of a small P(cid:22) NURBS control points set of parameters which facilitates analysis and X Surface coordinate optimization during the conceptual design process. C Airfoil chord length The model is generated from a preliminary CAD F1;F2 Airfoil thickness parameter geometry. From this geometry, airfoil cross sections M1;M2 Airfoil camber parameters are cut at selected locations and (cid:12)tted with analytic N NURBS basis function curves. The airfoils are then used as boundaries for P1;P2 Airfoil camber parameters surfaces de(cid:12)ned as the solution of partial di(cid:11)erential T Airfoil thickness parameter equations. Both the airfoil curves and the surfaces h Wing section span are generated with free parameters selected to give u;v Surface parametric variables a good representation of the original geometry. The w NURBS weight original surface is compared with the parametric x;y;z Surface coordinates model, and solutions of the Euler equations for x(cid:22);y(cid:22) Airfoil coordinates compressible (cid:13)ow are computed for both geometries. xe;ye Airfoil trailing edge The parametric model is a good approximation of y^ CAD airfoil coordinate the CAD model and the computed solutions are y(cid:22)c Airfoil camber qualitatively similar. An optimal NURBS approxi- y(cid:22)t Airfoil thickness mation is constructed and can be used by a CAD (cid:18) Twist angle model for further re(cid:12)nement or modi(cid:12)cation of the (cid:24);(cid:17) Wing section parametric variables original geometry. (cid:27) PDE weighting factor 1. INTRODUCTION ManyofthebusiestairportsinAsiawillrunoutof spaceinthenearfuture. Onesolutiontotheproblem proposedby the aircraftindustry is the development of huge airliners carrying up to 800 passengers. Along with the tradition aircraft con(cid:12)gurations, 1 some manufacturers, including McDonald Douglas , (cid:3)ThispaperisdeclaredaworkoftheU.S.Governmentand have advanced a (cid:13)ying wing con(cid:12)guration similar to isnotsubjectedtocopyrightprotection intheUnitedStates. yNRCSeniorResearchAssociate,NASALangleyResearch the B2 bomber. This unconventional blended-wing- Center,Hampton,Virginia23681-0001. body (BWB) concept for future subsonic transports zSenior Research Engineer, NASA Langley Research Cen- is an aerodynamically e(cid:14)cient airplane with the ter,HamptonVirginia23681-0001,AssociateFellow,AIAA. xResearch Associate, Department ofMechanical Engineer- interior passenger volume distributed over a large ing,OldDominionUniversity,Norfolk,Virginia23529-0247. portion of the centerbody. {MemberoftheTechnicalSta(cid:11),ComputerSciencesCorpo- ration,3217N.ArmisteadAvenue,Hampton,VA23666. 1 New concepts in aircraft designs often begin with ure 2. Four of these cuts were on the main horizon- only a rough sketch on the designers note pad. tal section of the wing and two were on the vertical Theseconceptualdesignsalwayshaveto bemodi(cid:12)ed winglet. These cross sections were output as a dense due to structural and aerodynamic constraints. set of points. The airfoils in the parametric model For this reason, it is convenient to have a model were then generated with the parameters selected to which is de(cid:12)ned by a set of parameter values so the give the best (cid:12)t to the CAD airfoils. The paramet- the model can be easily modi(cid:12)ed as more design ricairfoilequationsde(cid:12)ningthecoordinatesx(cid:22)((cid:24))and variables are considered. This report will describe y(cid:22)((cid:24)) of the points on an airfoil are the application of a conceptual design tool that will facilitate aerodynamic analysis in the early stages x(cid:22)((cid:24))=Csin(cid:25)(cid:24); y(cid:22)((cid:24))=y(cid:22)t((cid:24))+y(cid:22)c((cid:24)); in the development of an aircraft. The parametric model which is generated in this manner now can T y(cid:22)t((cid:24))=(cid:0) (sin2(cid:25)(cid:24)+F1sin4(cid:25)(cid:24)+F2sin6(cid:25)(cid:24)); be optimized to improve the aerodynamic charac- 2 teristics of the aircraft by varying a small set of 8 M1 2 pexaarammineetderst,oodrettheermcoinmepalelltepposasriabmleetfeearssipbalecedceasingnbse. >>>>>< (PP122M(cid:0)(2P1P1)13r((cid:0)r(cid:0)rP)2)2if(2rr(cid:20)+PP12(cid:0)3P1) lBsetysasgpeleiskrefoolyfrmttoihneghadaveemvetoloroepmmcaoekmnetp,mlettaehjoearndadeleyvsseigilsonpaecthrtahwnegileelasrbilnye y(cid:22)c((cid:24))=>>>>>: (cid:0)(1(cid:0)(iMPfP22P2M(cid:0))12P2(1<1)3(cid:0)r(r<2(cid:0)PP2P2+1)22(P22rr+(cid:0)Pr12)(cid:0)3iPf2r)(cid:21)P2; the (cid:12)nal Computer-Aided Design (CAD) models to r=sin(cid:25)(cid:24) be used for Computational Fluid Dynamics (CFD) analysis and wind tunnel testing. 0(cid:20)P1 (cid:20)P2 (cid:20)1; 0(cid:20)(cid:24) (cid:20)1: (1) Thedesignparametersintheseequationsallowsuf- This report will describe an application of the (cid:12)cient freedom to give a good approximation to all conceptual design tool Rapid Airplane Parametric 2;3 the airfoil crosssections used for the model. The pa- Input Design (RAPID) in creating a parametric rameterC is the section chord length. The thickness model for the the BWB airplane. Since RAPID was developed for modeling conventional airplane curve y(cid:22)t((cid:24)) is de(cid:12)ned with the thickness parameter geometries, modi(cid:12)cations were necessary to apply T and Fourier coe(cid:14)cients F1 and F2. A general- ization of the piecewise quadratic camber curve in the techniques to the current application. The the RAPID system has been developed to model the input parameters for a geometry may come from wing sections of the BWB con(cid:12)guration. The cam- a variety of sources and range from crude sketches to detailed drawings or surfaces generated from a ber curve y(cid:22)c((cid:24)) is a piecewise cubic curve which can have both a local maximum and a local minimum. CAD system. For the BWB geometry a preliminary CAD model was available and was used to gener- The camber curve has relative extrema of M1 and atemostoftheparametersusedintheRAPIDmodel. M2 at locations P1 and P2, respectively. The cam- ber locations are the fractions of total chord length 2. AIRFOIL APPROXIMATION measured from the airfoil trailing edge. If P1 = P2 and M1 = M2 this expression reverts to the original The construction of the parametric model for the piecewisequadraticcambercurveinRAPID.Thead- BWB begins with the de(cid:12)nition of airfoil cross sec- ditional parameters are needed for the BWB model tions. The actual CAD data was used to de(cid:12)ne the to approximate the forward camber and aft re(cid:13)ex of airfoil parameters in RAPID. The number of airfoils the centerbody airfoil. de(cid:12)ning the model determines the total number of The selection of a set of design parameters to (cid:12)t a parameters in the model and also e(cid:11)ects the (cid:12)delity given airfoil can be accomplished by solving a non- of the model. If a large number of cross sections linear optimization problem. Assume that a set of are made, the CAD model can be more accurately approximated by the RAPID model. However, the values (cid:24)i; i = 1;(cid:1)(cid:1)(cid:1);m have been selected based on some desired criteria for distributing points around number of parameters increases with the number of the airfoil. From these values, a correspondingset of airfoils, and there is the ever present danger of os- cillations in the model that is inherent in this as values x(cid:22)i; i = 1;(cid:1)(cid:1)(cid:1);m can be calculated from the (cid:12)rst equation in (1). Now use the airfoils from the well as many other interpolation and approximation CADdataandcalculatecorrespondingy coordinates schemes. From the CAD surface given in Figure 1, six cuts were made at the positions indicated in Fig- y^i; i = 1;(cid:1)(cid:1)(cid:1);m, by interpolating the data if neces- sary. Observe that on the upper and lower surfaces 2 of the airfoil the value of y(cid:22) can be given as a single- example, if the airfoil generated in Section 2 corre- valued function of x(cid:22). The chord length C can be sponded to (cid:24) = 0, then the Dirichlet boundary con- measured and will not be included in the set of vari- dition would be ables for the optimization problem. Thus, the best 8 (cid:12)t to the given CAD data is the airfoil de(cid:12)ned by < xe+x(cid:22)((cid:24))cos((cid:18))+y(cid:22)((cid:24))sin((cid:18)) equation (1) with the remaining design parameters D0((cid:24))=: xe(cid:0)x(cid:22)((cid:24))sin((cid:18))+y(cid:22)((cid:24))cos((cid:18)) h T;F1;F2;M1;M2;P1;P2 chosen to minimize Xm The Neumann boundary conditions in(cid:13)uence the 2 [y(cid:22)i(T;F1;F2;M1;M2;P1;P2)(cid:0)y^i] shape of the wing surface and must be determined i=1 in part from the CAD data. From a geometric point ofview,theNeumannboundaryconditionscanbein- The solution of this nonlinear least-squares problem terpretedasatangentribbonwhichde(cid:12)nes theslope de(cid:12)nes the airfoil shapes for the parametric model. of the wing surface at the airfoil section. This tan- This optimization problem can be easily solved by gent ribbon, as depicted in Figure 3, is constructed a number of methods. The method which has been by o(cid:11)setting the cross sections in Figure 2. For each usedinthisworkisthetruncatedNewton method in 4 airfoil,ano(cid:11)setdistanceintheparametricvariable(cid:17) the TNBC package developed by S. G. Nash . The is selected. A new airfoil is constructed at the o(cid:11)set software is designed for minimization with bound locationusingtheRAPIDparameterswhichwillgive constraints and can be obtained from netlib. Other a good approximation of the actual CAD cross sec- dataisnecessarytopositiontheairfoilsintheproper tionat thesamelocation. Althoughtherearealarge locationinthreedimensionalspace. Fromtheairfoils numberofparameterswhichdeterminetheshapeand in Figure 2, the spanwise locations, h the locations location of an airfoil, for calculating the o(cid:11)set airfoil of the trailing edges, xe and ye, and the twists, (cid:18), it is su(cid:14)cient to consider only changes in the trail- can be determined. It is assumed that each airfoil ing edge, (xe;ye), chord length, C, and thickness, T. alongthefuselageandwingsectionsarecontainedin Changes in other airfoil parameters are neglected in a z =constant plane and that both airfoils de(cid:12)ning de(cid:12)ning the tangent ribbon. The o(cid:11)set airfoils now the winglet are in a y =constant plane. Once the determine the changes in x and y with respect to airfoils are de(cid:12)ned in space, the wing surfaces can z, that is, @x=@z and @y=@z at each point. Now if be constructed to span the airfoils and create the @z=@(cid:17) is given, then @x=@(cid:17) and @y=@(cid:17) can be cal- airplane surface. culatedfrom @xe=@z, @ye=@z, @C=@z,and @T=@z by constructing the o(cid:11)set airfoil and applying the chain 3. WING SURFACES rule. Thus, the procedure for generating the Neu- The RAPID methodology creates wing surfaces in mann boundary conditions can be described in the xyz-space in the form following steps: (1) calculate airfoil parameters for the o(cid:11)set airfoils with chord, thickness, and trailing X=(x((cid:24);(cid:17));y((cid:24);(cid:17));z((cid:24);(cid:17))); 0(cid:20)(cid:24) (cid:20)1; 0(cid:20)(cid:17) (cid:20)1 edge coordinates given by The coordinates of surface points are obtained by @C @z @T @z C+ ; T + ; solving the fourth order partial di(cid:11)erential equation @z @(cid:17) @z @(cid:17) (cid:20)(cid:27)2 @2 + @2 (cid:21)2X=0; (2) xe+ @xe @z; ye+ @ye @z; @(cid:24)2 @(cid:17)2 @z @(cid:17) @z @(cid:17) where(cid:27) isaconstantweightingparameter. Dirichlet resectively;(2)constructtheo(cid:11)setairfoilwithcoordi- and Neumann boundary conditions are imposed at nates(xo((cid:24));yo((cid:24)));and(3)settheNeumannbound- (cid:17) =0 and (cid:17) =1. The boundary conditions are: ary conditions 8 X((cid:24);0)=D0((cid:24)); X((cid:24);1)=D1((cid:24)); < xo((cid:24))(cid:0)x((cid:24);0) N0((cid:24))= yo((cid:24))(cid:0)y((cid:24);0) : @z X(cid:17)((cid:24);0)=N0((cid:24)); X(cid:17)((cid:24);1)=N1((cid:24)): @(cid:17) The solution is periodic in (cid:24). The spanwise changes in trailing edge coordinates, The Dirichlet boundary conditions, de(cid:12)ned by the chord length and thickness can be easily estimated functions D0((cid:24)) and D1((cid:24)), are determined from the fromthe CAD data. However,an estimate of @z=@(cid:17), airfoils which form the edges of the surface. For which gives the spanwise in(cid:13)uence of the Neumann 3 boundary condition, is not apparent from the data. 4. CFD ANALYSIS Thus some trial and error was necessary to obtain The RAPID system can output surface grids with the best (cid:12)t to the CAD surface. In most cases, speci(cid:12)ed grid point distributions in each coordinate setting @z=@(cid:17) equal to the span of the wing section direction. Thesesurfacegridscanbeuseddirectlyin is a good starting value. a potential (cid:13)ow analysis code or they can be used as boundary surfaces to generate a volume grid for an Once the boundary conditions are determined, Euler or Navier-Stokes calculation. Similar volume equation (2) is solved by generating a Fourier series gridswereconstructedfor theregionabout theCAD expansion. An approximation of the CAD surface and RAPID models. In order to guarantee com- using the RAPID technology appears in Figure 4. parable grids, a grid was (cid:12)rst generated about the RAPID surface and then this grid was distorted to generate the grid about the CAD surface. The grid The total number of parameters will now be sum- distortion process was carried out as a application marized. There are a total of six airfoils de(cid:12)ning the of the Coordinate and Sensitivity Multidisciplinary BWB geometry. Each airfoil is de(cid:12)ned by eight pa- 5 Design and Optimization (CSCMDO) system . rameters With this system, the di(cid:11)erences in the volume C; T; F1; F2; M1; M2; P1; and P2: grids is no greaterthan the di(cid:11)erences in the surface grids. The CSCMDO system could also be used The location and orientation of each airfoil in space to generate new volume grids after making small mustbede(cid:12)ned. Thisisdonebyspecifyingthespan- changesin designparameters. Solutionsto the Euler wise location of the airfoil, h, the coordinates of the equations for compressible (cid:13)ow were computed on trailingedge,xe andye,andthetwistangle,(cid:18). There grids for the RAPID and CAD geometries. The are additional parametersfor each airfoil used to de- following comparisons are for a free stream Mach (cid:12)ne the Neumann boundary conditions. These are number of 0.85 and an angle of attack of 2 degrees. Figure 8 contains comparisons of contour plots of @z @xe @ye @C @T the local Mach number on the upper surfaces of the ; ; ; ; and : @(cid:17) @z @z @z @z CAD and RAPID geometries. The similarity in the solutions is evident, especially the shock structure Thus, there are a total of 17 parameters for each along the wing segment. Further comparisons reveal of six airfoils. Now deleting the spanwise location the e(cid:11)ects of the di(cid:11)erences in geometry. Figures 9 of the centerbody airfoil which we assume to be at and 10 are pressure coe(cid:14)cient plots near the center z = 0, the BWB airplane, as de(cid:12)ned by RAPID, is of the fuselage cross section and on the wing. The generated with a total of 101 parameters. pressure coe(cid:14)cient plots are superimposed over plot of the airfoil cross sections, so that di(cid:11)erence Since considerable e(cid:11)ort has gone into attempting in solutions can be compared with the di(cid:11)erences to approximate a given CAD surface with a para- in geometries. It was interesting to note that the metricmodel, somecomparisonsoftheoriginalCAD RAPID surface geometry gave smoother plots and surface and the RAPID surface will be presented. a larger lift coe(cid:14)cient than the solution computed It is evident from comparing Figures 1 and 4 that with the CAD geometry. the gross geometry of the BWB airplane has been captured in the RAPID model. Figures 5 and 6 5. NURBS APPROXIMATION compare the models at two cross sections used in NonUniform Rational B-Splines (NURBS) have de(cid:12)ning the RAPID model. The approximation become a standard surface representation in the of the supercritical airfoil section in Figure 6 is 6 CAD industry . For this reason, a method for excellent, but there is noticeably more error in the accuratelyrepresentingtheRAPIDsurfacegeometry approximation of the centerbody airfoil in Figure 5. by a NURBS surface has been investigated. As with A comparison of the planforms appears in Figure 7. the RAPID methodology, one of the advantages in Whether this degree of (cid:12)delity in the approximation a NURBS representation of a surface is the fact is su(cid:14)cient would of course be a judgement call of that the surface is de(cid:12)ned by a set of parameters the designer. Further accuracy could be achieved which can be manipulated to change the shape of with the RAPID technology, but at the cost of the surface. However, there is a di(cid:11)erence in the increasing the complexity of the model and the two representations. The RAPID parameters are number of parameters. actual geometric lengths and distances, whereas the 4 NURBS parameters are the control points, weights, and knots is solved using the TNBC optimization and knots. routine. 6. SUMMARY AND CONCLUSIONS A NURBS surface of degree p in the u-direction anddegreeq inthev-directionisapiecewiserational The RAPID methodology has been used to function of the form generate a parametric model for the BWB subsonic PIi=0PJj=0Ni;p(u)Nj;q(v)wi;jP(cid:22)i;j tisraanlsipmoirttt.oWthitehaocnculyraacsyminaltlhseetaopfpproaxraimmaettieorns,otfhtehree P(u;v)= PIi=0PJj=0Ni;p(u)Nj;q(v)wi;j original CAD model. In examining the di(cid:11)erences in geometry and CFD solutions with the CAD and where P(cid:22)i;j are the points of the control net, wi;j Rapid models, it should always be noted that this are the weights, and Ni;p(u) and Nj;q(v) are the B- methodology is intended for aerodynamic analysis spline basis functions de(cid:12)ned on the knot vectors at the conceptual stage of the design process. At u(cid:22)0;(cid:1)(cid:1)(cid:1);u(cid:22)I+p+1 and v(cid:22)0;(cid:1)(cid:1)(cid:1);v(cid:22)J+q+1 with this stage, the CAD geometry may only be a crude approximation of the (cid:12)nal con(cid:12)guration. The ob- u(cid:22)0 =(cid:1)(cid:1)(cid:1)=u(cid:22)p = 0 =v(cid:22)0 =(cid:1)(cid:1)(cid:1)=v(cid:22)q+1 jective in constructing the RAPID model is to have u(cid:22)I+1 =(cid:1)(cid:1)(cid:1)=u(cid:22)I+p+1 = 1 =v(cid:22)J+1 =(cid:1)(cid:1)(cid:1)=v(cid:22)J+q+1: a parametrization of the BWB con(cid:12)guration. The RAPIDmodelcanthenbeusedinaparameterstudy to predict the e(cid:11)ects of changesin the geometryand Let Xm;n; m = 1;(cid:1)(cid:1)(cid:1);M; n = 1;(cid:1)(cid:1)(cid:1);N be a sur- improve upon the original design. face grid generated by RAPID. Surfaces as NURBS are de(cid:12)ned on the unit square and can be evaluated 7. ACKNOWLEDGEMENTS at an array of points (um;vn); m = 1;(cid:1)(cid:1)(cid:1);M; n = 1;(cid:1)(cid:1)(cid:1);N in the uv-plane. The optimal NURBS ap- This work was performed while the (cid:12)rst author proximation is obtained by selecting the control net, held a National Research Council-NASA Langley weights, and knots which yields the minimum value Research Center Research Associateship. of XM XN 8. References [Xm;n(cid:0)P(um;vn)]2: 1 Liebeck, R. H., Page, M. A., and Rawdon, B. K., m=1n=1 \Evolution of the Revolutionary Blended-Wing-Body," Transportation Beyond 2000: Technologies Needed for The optimal NURBS will depend on the selection of Engineering Design, NASACP 10184, pp431-459, 1996. the arrayof parametric variables u and v. Both uni- 2 Smith, R. E., Bloor, M. G. I., Wilson, M. J., and form and arclength parameters have worked well in Thomas, A. M., \Rapid Airplane Parametric Input practice. Design (RAPID),"AIAAPaper 95-1687, June,1995. 3 Smith, R. E., Cordero, Y., and Jones, W. T., \Au- An example of a NURBS approximation of the tomated Airplane Surface Generation," Numerical BWB airplane and the control net is plotted in Grid Generation in Computational Field Simulations Figure 11. This surface is de(cid:12)ned using piecewise pp. 293-301, NSF Engineering Research Center for Computational Field Simulation, Mississippi State, MS, cubic basisfunctions in both directionsand an 8(cid:2)8 1996. controlnet. The NURBS surface is therefore de(cid:12)ned 4 Nash, S. G., User’s guide for TN/TNBC, Technical by 64 values for each coordinate of the control net, Report 397, Department of Mathematical Sciences, The 64 weights, and four knot values for each direction. Johns HopkinsUniversity,Baltimore, MD, 1984. 5 Jones, W., and Samareh-Abolhassani, J., \A Grid Thus the NURBS surface is de(cid:12)ned with a total of GenerationSystemforMulti-disciplinaryDesign," AIAA 264 parameters. Paper 95-1689, June, 1995. 6 Tiller, W., \Rational B-Splines for Curve and Surface Representation," IEEE Computer Graphics and Applica- The NURBS approximation requires the solution tions, 3, pp. 61-69, 1983. of a nonlinear optimization problem. The solution is computed using a two-stage process. First, the optimal control net is constructed using unit weights and uniform knots. This is a linear least-squares problem that is solved directly using a QR matrix factorization. Theresultingsetofcontrolpointswith unit weights and uniform knots are initial values for calculation of the optimal NURBS. Next, the full nonlinearproblemwithvariablecontrolnet,weights, 5 Fig. 1 Blended-wing-bodyCADgeometry Fig. 3 Tangent ribbonsfor RAPIDmodel Fig. 2 Airfoil cross sections for RAPIDmodel Fig. 4 Blended-wing-bodyRAPIDgeometry 6 +RAPID (cid:0)CAD : RAPID (cid:0) CAD Fig. 5 Comparison of CAD andRAPIDmodel for centerbody airfoil +RAPID (cid:0)CAD Fig. 6 Comparison of CAD andRAPIDmodel for wing airfoil Fig. 7 Comparison of CAD andRAPIDplanforms 7 Pressure Coefficient Comparison at Z = -1 1.0 400 300 Surface RAPID Cp RAPID 200 Surface CAD Cp CAD 0.5 100 0 RAPID Y Cp -100 0.0 -200 -300 -400 -0.5 -500 -1500 -1000 -500 0 X Fig. 9 Pressure coe(cid:14)cient at fuselage cross section Pressure Coefficient Comparison at Z = -1016 0.5 100 Surface RAPID Cp RAPID Surface CAD Cp CAD 50 0.0 Y Cp CAD 0 -0.5 Fig. 8 Mach contours of CADand RAPIDmodels -50 -1.0 -500 -400 -300 -200 X Fig. 10 Pressure coe(cid:14)cient at wing cross section 8 Fig. 11 NURBS approximationof RAPIDgeometry 9